# FLEXIBILITY IN THE MIGRATION STRATEGIES OF ANIMALS

EDITED BY : Nathan R. Senner, Yolanda E. Morbey and Brett K. Sandercock PUBLISHED IN : Frontiers in Ecology and Evolution

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88963-811-6 DOI 10.3389/978-2-88963-811-6

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# FLEXIBILITY IN THE MIGRATION STRATEGIES OF ANIMALS

Topic Editors: Nathan R. Senner, University of South Carolina, United States Yolanda E. Morbey, University of Western Ontario, Canada Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway

Citation: Senner, N. R., Morbey, Y. E., Sandercock, B. K., eds. (2020). Flexibility in the Migration Strategies of Animals. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-811-6

# Table of Contents


*182 Automated VHF Radiotelemetry Revealed Site-Specific Differences in Fall Migration Strategies of Semipalmated Sandpipers on Stopover in the Gulf of Maine*

Rebecca L. Holberton, Philip D. Taylor, Lindsay M. Tudor, Kathleen M. O'Brien, Glen H. Mittelhauser and Ana Breit

*196 Why are Whimbrels Not Advancing Their Arrival Dates Into Iceland? Exploring Seasonal and Sex-Specific Variation in Consistency of Individual Timing During the Annual Cycle*

Camilo Carneiro, Tómas G. Gunnarsson and José A. Alves

*204 Effects of Spring Migration Distance on Tree Swallow Reproductive Success Within and Among Flyways*

Elizabeth A. Gow, Samantha M. Knight, David W. Bradley, Robert G. Clark, David W. Winkler, Marc Bélisle, Lisha L. Berzins, Tricia Blake, Eli S. Bridge, Lauren Burke, Russell D. Dawson, Peter O. Dunn, Dany Garant, Geoff Holroyd, Andrew G. Horn, David J. T. Hussell, Olga Lansdorp, Andrew J. Laughlin, Marty L. Leonard, Fanie Pelletier, Dave Shutler, Lynn Siefferman, Caz M. Taylor, Helen Trefry, Carol M. Vleck, David Vleck, Linda A. Whittingham and D. Ryan Norris

*214 Spatial and Temporal Variability in Migration of a Soaring Raptor Across Three Continents*

W. Louis Phipps, Pascual López-López, Evan R. Buechley, Steffen Oppel, Ernesto Álvarez, Volen Arkumarev, Rinur Bekmansurov, Oded Berger-Tal, Ana Bermejo, Anastasios Bounas, Isidoro Carbonell Alanís, Javier de la Puente, Vladimir Dobrev, Olivier Duriez, Ron Efrat, Guillaume Fréchet, Javier García, Manuel Galán, Clara García-Ripollés, Alberto Gil, Juan José Iglesias-Lebrija, José Jambas, Igor V. Karyakin, Erick Kobierzycki, Elzbieta Kret, Franziska Loercher, Antonio Monteiro, Jon Morant Etxebarria, Stoyan C. Nikolov, José Pereira, Lubomír Peške, Cecile Ponchon, Eduardo Realinho, Victoria Saravia, Cağan H. Sekercioğlu, Theodora Skartsi, José Tavares, Joaquim Teodósio, Vicente Urios and Núria Vallverdú

#### *228 One Hundred Pressing Questions on the Future of Global Fish Migration Science, Conservation, and Policy*

Robert J. Lennox, Craig P. Paukert, Kim Aarestrup, Marie Auger-Méthé, Lee Baumgartner, Kim Birnie-Gauvin, Kristin Bøe, Kerry Brink, Jacob W. Brownscombe, Yushun Chen, Jan G. Davidsen, Erika J. Eliason, Alexander Filous, Bronwyn M. Gillanders, Ingeborg Palm Helland, Andrij Z. Horodysky, Stephanie R. Januchowski-Hartley, Susan K. Lowerre-Barbieri, Martyn C. Lucas, Eduardo G. Martins, Karen J. Murchie, Paulo S. Pompeu, Michael Power, Rajeev Raghavan, Frank J. Rahel, David Secor, Jason D. Thiem, Eva B. Thorstad, Hiroshi Ueda, Frederick G. Whoriskey and Steven J. Cooke

#### *244 Prevalence and Mechanisms of Partial Migration in Ungulates* Jodi E. Berg, Mark Hebblewhite, Colleen C. St. Clair and Evelyn H. Merrill

# *261 Energetic Status Modulates Facultative Migration in Brown Trout (*Salmo trutta*) Differentially by Age and Spatial Scale*

Samuel J. Shry, Erin S. McCallum, Anders Alanärä, Lo Persson and Gustav Hellström


Devin R. de Zwaan, Scott Wilson, Elizabeth A. Gow and Kathy Martin


Carl Tamario, Johanna Sunde, Erik Petersson, Petter Tibblin and Anders Forsman

*372 A Migratory Divide Among Red-Necked Phalaropes in the Western Palearctic Reveals Contrasting Migration and Wintering Movement Strategies*

Rob S. A. van Bemmelen, Yann Kolbeinsson, Raül Ramos, Olivier Gilg, José A. Alves, Malcolm Smith, Hans Schekkerman, Aleksi Lehikoinen, Ib Krag Petersen, Böðvar Þórisson, Aleksandr A. Sokolov, Kaisa Välimäki, Tim van der Meer, J. David Okill, Mark Bolton, Børge Moe, Sveinn Are Hanssen, Loïc Bollache, Aevar Petersen, Sverrir Thorstensen, Jacob González-Solís, Raymond H. G. Klaassen and Ingrid Tulp

*389 Variation From an Unknown Source: Large Inter-individual Differences in Migrating Black-Tailed Godwits*

Mo A. Verhoeven, A. H. Jelle Loonstra, Nathan R. Senner, Alice D. McBride, Christiaan Both and Theunis Piersma

#### *398 High Migratory Survival and Highly Variable Migratory Behavior in Black-Tailed Godwits*

Nathan R. Senner, Mo A. Verhoeven, José M. Abad-Gómez, José A. Alves, Jos C. E. W. Hooijmeijer, Ruth A. Howison, Rosemarie Kentie, A. H. Jelle Loonstra, José A. Masero, Afonso Rocha, Maria Stager and Theunis Piersma

#### *409 Interspecific Variation in Seasonal Migration and Brumation Behavior in Two Closely Related Species of Treefrogs*

Amaël Borzée, Yoojin Choi, Ye Eun Kim, Piotr G. Jablonski and Yikweon Jang

# Editorial: Flexibility in the Migration Strategies of Animals

Nathan R. Senner <sup>1</sup> \*, Yolanda E. Morbey <sup>2</sup> and Brett K. Sandercock <sup>3</sup>

<sup>1</sup> Department of Biological Sciences, University of South Carolina, Columbia, SC, United States, <sup>2</sup> Department of Biology, Western University, London, ON, Canada, <sup>3</sup> Department of Terrestrial Ecology, Norwegian Institute for Nature Research, Trondheim, Norway

Keywords: conservation, developmental plasticity, environmental change, phenotypic flexibility, phenotypic plasticity

**Editorial on the Research Topic**

**Flexibility in the Migration Strategies of Animals**

# INTRODUCTION

Climatic and environmental changes are global phenomena, altering every biome, and affecting nearly every species. At a population level, significant effort has been devoted to identifying demographic "winners" and "losers" in the face of rapid environmental change (Wiens, 2016). Armed with information on population status, a major focus in evolutionary ecology has been to attribute organismal responses to behavioral or physiological processes (i.e., phenotypic plasticity and flexibility), genotypic change, or some combination thereof (Gienapp et al., 2008). Migratory species may be especially vulnerable to environmental change because they often have life-history strategies characterized by low fecundity and high survival, because long distance movement exposes them to many different types of risk, and because they require patches of habitat separated by vast distances (Wilcove and Wikelski, 2008). Accordingly, the dramatic environmental changes that have occurred during the Anthropocene have led to rapid population declines for many migrants (Lascelles et al., 2014; Pearce-Higgins et al., 2017; Tucker et al., 2018). Nevertheless, some migratory species have maintained stable population sizes and displayed surprising levels of phenotypic flexibility (Pedler et al., 2018), phenotypic plasticity (Eichhorn et al., 2009; Verhoeven et al., 2018), and even evolutionary adaptability (Kovach et al., 2012; Helm et al., 2019). In light of these organismal responses, significant questions remain about the degree to which migratory species can adapt to change, both in the short term and across generations (Hebblewhite and Haydon, 2010; Piersma, 2011).

For our Research Topic on Flexibility in the Migration Strategies of Animals, we invited a wide array of conceptual, theoretical, and empirical papers. Our intention was to develop a more complete understanding of the degree of variation in migratory behaviors exhibited by individuals and populations, so that we could further our ability to project how future environmental change might affect the population dynamics of migratory species. To conceptually organize our topic and evaluate the timescales over which individuals and populations can respond to environmental change, we adopted an ontogenetic approach to the study of migration. An ontogenetic approach recognizes that traits can have a genetic basis, but argues that different phenotypic traits can have varying degrees of lability over the course of an individual's lifespan. For instance, traits can fall anywhere along a continuum of lability, from traits that are canalized and immutable, to traits that are plastic but become fixed during specific windows of development (e.g., developmental plasticity), and those that remain flexible and can be reversibly changed at any life stage (e.g., phenotypic flexibility; Piersma and Drent, 2003). As a result, environmental change that

#### Edited and reviewed by:

Jordi Figuerola, Estación Biológica de Doñana (EBD), Spain

> \*Correspondence: Nathan R. Senner senner@mailbox.sc.edu

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 24 March 2020 Accepted: 06 April 2020 Published: 05 May 2020

#### Citation:

Senner NR, Morbey YE and Sandercock BK (2020) Editorial: Flexibility in the Migration Strategies of Animals. Front. Ecol. Evol. 8:111. doi: 10.3389/fevo.2020.00111 is encountered at different points during an individual's life and affecting different traits will potentially have different consequences for individuals and populations (Senner et al., 2015). Such an ontogenetic approach is particularly important when considering migratory species with annual movements that traverse entire hemispheres and make it difficult to determine where and when environmental change is having the most dramatic effect on a population (Norris and Taylor, 2006).

Our Research Topic resulted in a collection of 30 peerreviewed articles that consider a broad range of taxa and a variety of migratory behaviors, from partial and differential migration to some of the champions of extreme long-distance migration. The articles also span a range of topics, from the development of new theory to the documentation of intra- and interspecific differences in migratory behaviors; the physiological, ecological, and evolutionary drivers of those patterns; and the implications of flexible migration for the development of improved management and conservation actions. Here, we review the contributions of the articles to four major topics in migration ecology: Theory, Pattern, Process, and Synthesis and Applications. By organizing the articles into these four categories, we highlight how the collection provides an important framework for the study of animal migration and furthers our understanding of the potential responses of migratory species to a changing world.

### THEORY

The study of animal migration has a long history in the theoretical literature of ecology and evolutionary biology, dating back to early attempts to identify the important physical and environmental factors that affect the migratory behavior of birds (Alerstam, 1979). Theoretical investigations remain an active field and have been used to refine hypotheses about observed patterns, as well as drive new empirical work. For example, population dynamic models have been useful for understanding the cues used by individuals to make their migratory decisions (Budaev et al., 2019), the degree to which events occurring during one part of the year may have ramifications for events occurring in entirely different locations and at different times of the year (Taylor and Stutchbury, 2016; Taylor, 2017), and the configuration of events that leads to an optimal organization of the migratory annual cycle (Schmaljohann and Both, 2017; Lindström et al., 2019; Pirotta et al., 2019).

New theoretical papers in our Research Topic build upon these themes and make important contributions to our understanding of migratory strategies in seasonal environments. Morbey and Hedenström, for instance, constructed an optimization model to investigate whether migratory species should alter their departure timing from non-breeding sites or the speed of their migration as a means to optimize their arrival timing at their breeding areas. In general, they found that earlier departure should be the primary mechanism underlying earlier arrival timing (e.g., in males vs. females), although as migration distances become longer, both an earlier departure and faster migration should be beneficial.

Whereas the models in Morbey and Hedenström treated migration-related traits as locally adapted features of a migratory system, Oudman et al. explored finer scale aspects of migratory decision making. To do so, they used individual-based simulation models to explore the decision rules that account for Barnacle Geese (Branta leucopsis) switching among stopover sites in consecutive years while migrating from non-breeding sites in continental Europe to arctic breeding sites at Svalbard. They found that social interactions, combined with flexible responses to the densities of other geese encountered at stopover sites, determined an individual's decision-making process, thus enabling birds to maximize their fueling rates and expedite their northward migrations (Tombre et al., 2019).

The density-dependent responses of Barnacle Geese during migration suggest intriguing flexibility in stopover site use and indicate that even a population's migratory route itself may frequently be in flux. Links between competition and patterns of migration were also supported by an innovative network population model developed by Taylor, which showed that the strength of density-dependent population regulation during the breeding and non-breeding seasons, along with natal dispersal, can drive variation in patterns of migratory connectivity across populations.

This suite of theoretical studies suggests that a population's current migratory patterns are a product of a complex array of factors, and that as ecological drivers of migration undergo change, so too will the patterns of animal migration. Identifying such processes in action, though, requires a better understanding of the amount of inter- and intraspecific variation that exists in the patterns and behavior of migratory species.

# PATTERNS

The development of miniaturized tracking devices has led to many remarkable discoveries in migration ecology (McKinnon and Love, 2018). Relatively few studies, however, have been able to track enough individuals for long enough periods of time to characterize the full range of migratory behaviors exhibited across the lifespan of an individual, among different demographic groups in a population, or across the entire geographic range of a species (Both et al., 2016; Finch et al., 2017). Indeed, while empirical support for the environmental responsiveness predicted by the modeling studies in our collection is growing, one of the major contributions of our research topic is to provide detailed studies of the patterns of migration in a diverse array of migratory species, including insects (Menz et al.), cartilaginous and bony fishes in freshwater and marine environments (Eldøy et al.; Lennox et al.; Tamario et al.; Thorburn et al.), treefrogs (Borzée et al.), a wide range of birds (Carneiro et al.; Fraser et al.), and large-bodied ungulates (Berg et al.; Brown and Bolger; Found and St. Clair).

Our diversity of study subjects makes clear the overwhelming degree to which migratory behaviors can vary within individuals, as well as among populations and species. For instance, it is perhaps not surprising that ecologically and phylogenetically disparate species might differ in their migratory patterns: Egyptian Vultures (Neophron percnopterus) are soaring birds that use terrestrial habitats and avoid overwater crossings (Phipps et al.), whereas extreme migrants such as Upland Sandpipers (Bartramia longicauda) can make non-stop flights of up to 7,600 km in 7 days over the ocean and across mountain ranges (Hill et al.). It is far more intriguing, however, that two populations of Red-necked Phalaropes (Phalaropus lobatus) breeding immediately adjacent to each other in northern Europe migrate not only to different geographic regions, but different oceans and hemispheres altogether (van Bemmelen et al.). Intra-specific variation is not limited to the routes taken by individuals either, but can also include significant differences in timing between the sexes (Carneiro et al.) or among different nonbreeding sites or parts of a species' range (Phipps et al.; Battley et al.). Similarly, while age-related differences in migratory behavior are not unexpected, we are learning more about how juveniles differ from adults in their migration routes, timing, diet, and physiology (McCabe and Guglielmo; Thorburn et al.). Growing evidence also indicates that dramatic changes in migratory timing and space use can occur during adulthood, even though many species were previously thought to exhibit limited flexibility in their migratory behaviors (Fraser et al.; Senner et al.). Most intriguing, though, is the indication that individuals within populations can vary in the degree to which they show consistent migratory behaviors from year-to-year (Grecian et al.; Verhoeven et al.).

These broad-scale patterns of migratory movements therefore provide additional support for the overarching importance of ecological context in determining migratory behaviors. What thus remains is the identification of those factors most strongly influencing how populations respond to their current ecological context.

# PROCESS

Identifying the specific factors that either constrain or enable individuals and populations to respond to environmental change can be exceedingly difficult given the potential for carry-over effects to connect different life-history stages and geographic regions (Senner et al., 2015). For example, linkages between the quality of the non-breeding and breeding habitats used by an individual can exacerbate the consequences of events in early life (van de Pol et al., 2006). In addition, the same habitat may provide different resources for different groups of individuals, making it difficult to determine the direct connections between individual performance and the apparent quality of a site (Masero et al., 2017). As a result, the study of how migratory patterns are affected by environmental conditions is still in its relative infancy (Piersma, 2011).

A number of studies in our Research Topic investigated the complicated relationships between individual ontogeny and contemporary ecological conditions in the development of migratory behavior. The general pattern that begins to emerge from these studies is one where physiological constraints first interact with inexperience to influence the migratory patterns of juveniles. Two experimental studies of migratory brown trout (Salmo trutta), for example, found that natal growth conditions influence the probability of seaward migration in juveniles, although the effect differed between studies, possibly because food limitations were imposed during different developmental periods (Archer et al.; Shry et al.). Similarly, in migratory songbirds, the slow development of digestive physiology results in sub-optimal physiological performance during an individual's first southbound journey and may underlie well-documented age-related differences in migration speed (McCabe and Guglielmo). Finally, in a migratory shorebird, the Bar-tailed Godwit (Limosa lapponica), once juveniles have made their first southward migrations, individuals may explore widely prior to choosing a non-breeding site to which they will remain largely faithful over the course of the rest of their lives (Battley et al.).

The culmination of an individual's early life experiences therefore appears to be the development of a specific annual routine (Campioni et al., 2020). The development of a regular annual routine often results in surprisingly high repeatability of migratory behaviors (Carneiro et al.; Eldøy et al.; Grecian et al.; Verhoeven et al.). Some species, though, do retain significant flexibility in their migratory behaviors into adulthood, likely in response to variation in environmental conditions, including food, weather, predation risk, and competition (Fraser et al.; Senner et al.). The relative degree to which that flexibility is then employed to respond to current conditions, as opposed to environments experienced during the past, appears to differ depending on the relative severity of the conditions experienced over those two periods. For instance, individuals may alter not only the timing of their subsequent migratory movements (Anderson et al.), but also the length and direction of their movements in response to current food availability and their energetic condition (Brown and Bolger; Holberton et al.). Weather conditions can also play a direct role and drive movements both during migration (de Zwaan et al.) and the non-breeding season (McKinnon et al.). Reversible state effects that result from conditions encountered during previous stages of an individual's annual cycle, on the other hand, may be rarer than once thought and are only likely to occur under specific circumstances (Gow et al.).

The framework that has emerged from our collection of empirical studies, then, is that migratory behaviors are likely determined by a loosely inherited "structure" that can then be honed during development by interactions between physiological constraints, social information, and individual experience, and then repeatedly modified by the environmental conditions that are experienced during adulthood. Taken together, results from the different study systems in our Research Topic suggest that migration may be a system that is more environmentally responsive and potentially less constrained than previously thought. Therefore, what steps can be taken to conserve migratory animals that are exposed to ongoing environmental change?

# SYNTHESIS AND APPLICATION

The studies included in our Research Topic indicate that many characteristics of migratory life-history strategies are shared across a broad range of taxa, including endogenous programs for photoperiodic control of migratory movements (Åkesson and Helm) and a role for food, climate, predation, and competition in driving variation in migratory behavior (Berg et al.; Menz et al.). Taken together, these different lines of evidence suggest the potential for a broadly shared migratory syndrome that results less from a shared evolutionary history and more from the common ecological context of breeding in habitats with seasonal pulses of resource availability (Piersma et al., 2005; Dingle and Drake, 2007; Winger et al., 2019).

In this context, the key challenge for the conservation of migrants is that the pace, magnitude, and number of environmental changes that migrants are facing may outstrip the natural variation in flexibility that exists within most species. For example, Tamario et al. reviewed the major threats to migratory fishes and found that populations can simultaneously face overharvesting, rising water temperatures, drying rivers, and increasingly frequent barriers to migratory movements. Concurrent changes can lead to synergistic interactions, which multiply the effects of separate threats and overwhelm potential flexible responses, ultimately impacting population viability and threatening biodiversity. Growing evidence suggests that the key to mitigating the consequences of multiple changes lies in harnessing the significant flexibility that exists in many migratory species. For example, Found and St. Clair examined transitions from migration to residency in wild populations of elk (Cervus canadensis) and found that it was the most flexible individuals that abandoned migration and created humanwildlife conflicts. The flexibility of individuals, however, was part of a more complex shy-bold behavioral syndrome that includes behaviors that can be manipulated through directed management techniques. Thus, individuals can be specifically targeted to encourage them to migrate, thereby reducing the potential for habituation to anthropogenic environments.

We therefore need improved plans for conservation and management that recognize that migration patterns may not be static—migration routes can shift, new stopover sites can be adopted, and the timing of migratory movements can be flexibly molded to environmental conditions as they are experienced. As a result, existing networks of protected sites may not be

#### REFERENCES


adequate under future scenarios of environmental change. One possible approach may be to develop dynamic conservation plans that provide incentives for private landowners to improve conditions for migratory animals along their migration route for short periods of time (Reynolds et al., 2017). Moreover, conservation plans also need to anticipate future changes in resource availability, weather, and predation risk that may be outside the range of environmental conditions for which a population's current migration strategies have evolved. Last, the development and refinement of plans needs horizon scanning that identifies and prioritizes knowledge gaps for different taxa of migratory species (Lennox et al.; Tamario et al.). We hope that the new ideas and discoveries presented in the collection of papers in our Research Topic on Flexibility in the Migration Strategies of Animals will stimulate innovative research and that an improved understanding of organismal flexibility will lead to effective conservation actions for migratory species in the future.

### AUTHOR CONTRIBUTIONS

All authors have made substantial contributions to the editorial and to organizing the Research Topic.

#### FUNDING

NS was funded by startup funds from the University of South Carolina. BS was supported by the Norwegian Institute for Nature Research.

#### ACKNOWLEDGMENTS

We thank all of the 280 contributing authors for submitting their manuscripts and for making our Research Topic on animal migration a great success. We are grateful to the referees who are acknowledged on the first page of each article, and who provided constructive feedback and thoughtful comments on drafts of the manuscripts. Thank you also to the editorial team at Frontiers in Ecology & Evolution for their support at all stages of our project.


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Senner, Morbey and Sandercock. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Leave Earlier or Travel Faster? Optimal Mechanisms for Managing Arrival Time in Migratory Songbirds

#### Yolanda E. Morbey <sup>1</sup> \* and Anders Hedenström<sup>2</sup>

<sup>1</sup> Department of Biology, Western University, London, ON, Canada, <sup>2</sup> Department of Biology, Lund University, Lund, Sweden

We develop an optimization model with two decision variables to explore optimal migration mechanisms to facilitate optimal breeding timing in migratory songbirds. In the model, fitness is a function of date-dependent mortality, speed-dependent predation risk, and phenological match at arrival. The model determines the optimal combination of departure date for spring migration and migration speed, which can be mediated either by the power requirement for flight (P) or foraging effort at stopover sites (k). Our model predicts that earlier departure for spring migration should be the primary mechanism underlying earlier breeding timing, with a lesser role for faster migration via lower P or higher k. In contrast, longer migration to breeding areas selects for both earlier departure and faster migration. Empirical data on sex-specific migration traits largely conform to model predictions, since males generally migrate earlier than females but not faster than females. In contrast, empirical data on age-specific migration traits show some disagreement with model predictions, thus implicating additional tradeoffs. In partial agreement with the model, a comparative analysis of 25 songbird species showed that populations with longer migrations migrate more quickly, but do not initiate migration earlier. Our model proves to be a useful framework for interpreting migration strategies in animals making costly seasonal migrations.

#### Edited by:

Alexei B. Ryabov, University of Oldenburg, Germany

#### Reviewed by:

José F. Fontanari, University of São Paulo, Brazil Heiko Schmaljohann, University of Oldenburg, Germany

> \*Correspondence: Yolanda E. Morbey ymorbey@uwo.ca

#### Specialty section:

This article was submitted to Population and Evolutionary Dynamics, a section of the journal Frontiers in Ecology and Evolution

Received: 05 September 2019 Accepted: 03 December 2019 Published: 08 January 2020

#### Citation:

Morbey YE and Hedenström A (2020) Leave Earlier or Travel Faster? Optimal Mechanisms for Managing Arrival Time in Migratory Songbirds. Front. Ecol. Evol. 7:492. doi: 10.3389/fevo.2019.00492 Keywords: optimization model, migration timing, migration speed, phenological adaptation, Passeriformes

# INTRODUCTION

Costly migratory journeys between non-breeding and breeding areas pose a fundamental challenge: how to ensure optimal arrival timing to breeding areas with respect to seasonally-variable biotic and abiotic conditions (i.e., phenological match), while minimizing the costs of migration (e.g., energy expenditure, time, and mortality)? This challenge is particular acute for animals making long, energetically costly, seasonal migrations, such as birds and anadromous salmonid fishes (e.g., Oncorhynchus and Salmo spp.). Intuitively, these selection regimes should influence not only when animals depart for migration, but also how fast to travel. For example, late spawning populations of Oncorhynchus nerka generally migrate later (Hodgson and Quinn, 2002), and anadromous fish populations with longer migration distances migrate faster (Bernatchez and Dodson, 1987). In the bar-tailed godwit (Limosa lapponica baueri), populations that breed later at higher latitudes migrate later (Conklin et al., 2010), and in North American bird species using powered flight, those that migrate longer distances migrate faster through North America (La Sorte et al., 2013). Less is known about the relative importance of departure date vs. travel speed for facilitating optimal arrival timing, and theoretical models are lacking. Here we develop optimization models with two decision variables (departure date and travel speed) to inform optimal migration schedules in migratory songbirds (Order Passeriformes).

Migratory passerines (and near passerines) typically show several patterns of intraspecific co-variation in their breeding timing. These include protandry, which refers to the earlier onset of breeding activities in males than females (Newton, 2008; Morbey et al., 2012), the earlier arrival of adults than first-time breeders to breeding areas (Stewart et al., 2002), and later breeding timing at higher latitudes (Both and te Marvelde, 2007; Gow et al., 2019). Given sex-, age-, and latitude-specific breeding timing, differential migration schedules are expected, because migration is the life history phase immediately preceding breeding. In the context of sex-specific timing in birds, multiple aspects of the spatio-temporal organization of migration have been identified as having the potential to differ between males and females in order to facilitate protandry, with departure timing, migration speed (rate of fueling and rate of travel), and nonbreeding latitude receiving the most attention by empiricists (Coppack and Pulido, 2009). Our objective is to provide a theoretical basis to better understand key aspects of differential migration behavior in songbirds.

Our models of the co-evolution of departure time and migration speed are intended to be simple and general. Whenever possible, we aim to use realistic functions and parameters, but recognize there is considerable uncertainty regarding these choices. Moreover, because the models sacrifice realism for generality, the predictions that emerge are intended to be qualitative rather than quantitative. Two related problems are modeled. The first is the daily commute for people traveling to work each day. This is a simpler and more familiar problem than avian migration, and demonstrates the main tradeoffs affecting choice of departure time and travel speed. The second is latitudinal migration of songbirds to their breeding grounds, and is the primary problem of interest. This model is more complex because it must account for the fact that migration comprises stationary fueling and movement phases. Recent studies using advanced tracking technology are now providing an unprecedented amount of individual-based data on migration traits in songbirds (McKinnon et al., 2014; Briedis et al., 2017, 2019; Ouwehand and Both, 2017). Thus, we end by reviewing evidence for differential departure date and migration speed by sex, age, and migration distance.

#### MODELS

Model 1 was conceptualized as a daily commute to work, with two decision variables: departure time (t0) and vehicle speed (v). The general problem was to determine the optimal combination of t<sup>0</sup> and v {t ∗ 0 , v ∗ } which maximizes fitness. The model was formulated and solved in R 3.5.1 (R Core Team, 2018) by simulation. Model 2 was an adaptation of this model for a shorthop, overland songbird migrant, with t<sup>0</sup> being departure date and v being migration speed, which incorporates flight speed and the fueling rate to cover the power requirements for flight. In both models, we were particularly interested in the effect of varying target arrival time and travel distance on optimal departure time and travel speed. In the context of spring migration of songbirds, we addressed three questions: (1) to achieve a target degree of protandry, should males depart for migration earlier, or migrate faster than females? (2) for adult birds to arrive at breeding areas earlier than first-time breeders, should they depart for migration earlier or migrate faster? (3) should longer-distance migrants depart for migration earlier or migrate faster than shorter-distance migrants?

#### Model 1—The Daily Commute to Work Decision Variables and Assumptions

We let the decision variable t<sup>0</sup> be the time of departure (in hours) and the decision variable v be travel speed (km·h −1 ). Both t<sup>0</sup> and v were considered to be behavioral decisions. The variable t<sup>0</sup> was constrained to be in the range {0, 10}, where 0 is midnight and 10 is 10:00 am. The variable v was constrained to be in the range {30, 150}. Travel conditions were assumed to improve with departure time, which could be due to the combination of higher light levels (better visibility) and higher temperatures (less ice or snow). Commuters were assumed to have a target arrival time of τ , which would permit enough time to park and get to work on time.

#### Fitness Functions

Our approach closely followed Abrams et al. (1996). We specified fitness (W) to be the product of three fitness components:

$$\mathcal{W} = \mathcal{S}\_1 \mathcal{S}\_2 f$$

so that fitness (W) is the probability of arriving at the destination at the target time τ . In this function, fitness combines the minimization of delay events (i.e., weather-related accidents or speed traps) and the benefits of time matching to τ . Without any delay events, arrival time (t1) depends only on t<sup>0</sup> and v. To keep the problem simple, we ignored any effects of traffic congestion or priority effects at arrival.

The functions for the three fitness components were based on previous applications and produce intermediate optima. The first fitness component (S1) is the probability of avoiding weatherrelated delay events, which depends on the rate of delay events per kilometer traveled (Cs) and commuting distance (d).

$$\mathcal{S}\_1 = \exp(-C\_s d)$$

Thus, traveling a longer distance was assumed to be more costly (cf. Bell, 1997). We further let C<sup>s</sup> be a function of travel conditions at departure time t0. Two formulations of C<sup>s</sup> were considered. The first assumes that travel costs decrease linearly with departure time, as in Bell (1997):

$$S\_{1.1} = \exp(-\left[C\_0 - \alpha\_{1.1}t\_0\right]d)$$

where C<sup>0</sup> is the maximum travel cost (rate of delay events) when t<sup>0</sup> = 0, and α1.1 is the decrease in travel cost for each unit increase in t<sup>0</sup> (subject to the constraint α1.1τ < C0). This formulation accounts for the increased probability of a delay event, such as an accident, when departing earlier under poorer conditions. The fact that conditions might improve over the course of an individual's commute was not considered here, but could be considered in more complex formulations.

An alternative formulation assumes that C<sup>s</sup> declines exponentially with t<sup>0</sup> (see Appendix in Jonzén et al., 2007):

$$\begin{aligned} \mathcal{C}\_{\mathfrak{s}} &= \mathcal{C}\_{0} \exp \left( -\alpha\_{1,2} t\_{0} \right) \text{ such that} \\ \mathcal{S}\_{1,2} &= \exp \left( -\left[ \mathcal{C}\_{0} \exp \left( -\alpha\_{1,2} t\_{0} \right) \right] d \right) \end{aligned}$$

where α1.2 determines the rate of decline in C<sup>s</sup> . Functions S1.1 and S1.2 are shown for comparison (**Figures 1A,B**).

For the second fitness component (S2), we assumed that travel speed deviating from a speed limit of vlim (km·h −1 ) is costly in terms of a higher rate of speed-related delay events (being pulled over) per km traveled, such that:

$$S\_2 = \exp(-\alpha\_2 \left[ \left( \nu - \nu\_{lim} \right)^2 \right] d)$$

where v is travel speed and α<sup>2</sup> is a constant which determines how the per km travel costs change as v deviates from vlim (**Figure 1C**). S<sup>2</sup> is essentially the probability of avoiding a delay event across the entire commute.

For the third fitness component, we assumed a penalty for arriving earlier or later than optimal arrival time (τ ). For example, arriving too early means a longer wait time until τ , and arriving too late could increase the risk of disciplinary action by an employer. Following Abrams et al. (1996), the equation for this fitness component was assumed to be:

$$f = \exp(-\alpha\_3 \left[t\_1 - \tau\right]^2)$$

where t<sup>1</sup> = t<sup>0</sup> + d/v, and α<sup>3</sup> determines the cost of mismatched timing (**Figure 1D**).

#### Optimal t<sup>0</sup> and v

Simulation was used to find the optimal combination of t<sup>0</sup> and v {t ∗ 0 , v ∗ } which maximizes W. A baseline set of parameters was used to model a commute of d = 200 km at a speed limit vlim = 100 km·h −1 , with a target arrival time of 8:00 am (τ = 8 h). Other parameters were chosen to impose some costs (**Figure 1**): C<sup>0</sup> = 0.002, α1.1 = 0.0001, α1.2 = 0.1, α<sup>2</sup> = 0.0000001, α<sup>3</sup> = 0.01. Fitness surfaces were plotted using the filled.contour3 and filled.legend functions in R (http://wiki.cbr.washington.edu/ qerm/index.php/R/Contour\_Plots). Allowing for either a linear or exponential decrease in delay events and using the baseline set of parameters, the model produced a fitness surface with intermediate optima (**Figure 2**). This suggests that the chosen functions and parameter values were reasonable. The choice of S1.1 or S1.2 made little difference to {t ∗ 0 , v ∗ }.

#### Predictions

Simulations were run with randomized parameter combinations to explore the consequences of an earlier target arrival time (τ ) and a longer travel distance (d). Following Kokko et al. (2006), we used randomization because of uncertainty in the baseline parameter values, and to allow for a broad range of parameter combinations. In each simulation (n = 1,000), τ and d were drawn from uniform distributions between minimum and maximum values. Each remaining parameter was chosen from a normal truncated distribution, with mean = baseline value, sd = mean/3, and bounding values defined by the 2.5% and 97.5% quantiles. Separate simulations were run using S1.1 or S1.2.

Randomized simulations using S1.2 showed that departure time, but not travel speed, was sensitive to τ (**Figures 3A,B**). Using the correlation coefficient as an index of model sensitivity, the correlations with τ were r = 0.70 and r = −0.02, respectively. Thus, when aiming to arrive at a destination earlier, it is optimal to leave earlier but not travel faster. In contrast, both departure time and, to a lesser extent, travel speed were sensitive to d (**Figures 3C,D**). The correlations with d were r = −0.26 for departure time and r = 0.15 for travel speed. Thus, it is optimal to leave earlier and travel faster when commuting a longer distance.

Departure time was also sensitive to the speed limit (r = 0.35), C<sup>0</sup> (r = 0.36), and α<sup>3</sup> (r = −0.33). Commuters should leave later when the speed limit is higher, the maximum risk of delay is higher (e.g., poorer weather), or the penalty for mismatched arrival timing is lower. The remaining correlations with departure time were <0.07. Travel speed was also sensitive to the speed limit (r = 0.96) and α<sup>2</sup> (r = −0.14) but not to the other parameters (r's < 0.10). Commuters should drive faster when the speed limit is higher and when the penalty for mismatched travel speed (i.e., more enforcement) is lower. The majority of the results were similar when these scenarios were modeled under the assumption of a linear decrease in delay events (S1.1). The exception was that departure time was not sensitive to C<sup>0</sup> (r = 0.01) but instead was sensitive to α1.1 (r = 0.36), the rate of linear decrease in the travel cost. With a greater decrease in the travel cost (i.e., a more rapid improvement in driving conditions), commuters should leave later.

#### Model 2—Spring Migration in an Overland, Nocturnal Migrant

This model was parameterized for small songbirds (∼20 g) with overland, nocturnal migration with no major ecological barriers for stopover fueling. Functions and parameter values were chosen based on their general shapes, and whenever possible, were informed by empirical data. However, we recognize a general lack of data on mortality and reproductive success in wild songbirds to support our choice of parameters in the fitness functions.

#### Decision Variables and Assumptions

In the model we determined the optimal departure date (day of year, t0) and migration speed (km·h −1 ) for a bird leaving its final non-breeding (premigratory) site in the south and traveling north a distance (d) to its breeding site. The variable t<sup>0</sup> was considered to be a behavioral decision which determines the onset day of fueling in advance of the first migratory flight (Lindström et al., 2019). We assumed that variation in migration speed is determined either by the power requirement for flight (P), which is partly determined by wing morphology (model 2.1), or k (the fueling rate; model 2.2). In the former case, we considered wing morphology, and therefore P, to be a fixed, developmental decision (i.e., an evolved trait). In the latter case, we considered fueling rate (k) to be a behavioral decision about foraging intensity (cf. Weber et al., 1998). We also assumed a target arrival date (day of year, τ ). The variable t<sup>0</sup> was constrained to be in the range {0, τ }, where 0 is the earliest possible migration date.

FIGURE 3 | Predictions arising from randomized simulations of the commuting model. (A,B) Show the effect of target arrival time τ on departure time t<sup>0</sup> (hours since midnight) and travel speed v (km·h −1 ), respectively. (C,D) Show the effect of commuting distance d on t<sup>0</sup> and v, respectively.

#### Fitness Functions

We specified fitness (W) as the product of three fitness components:

W = S1S2f

so that fitness (W) is expected reproductive output. Fitness combines mortality minimization and the reproductive benefits of matching arrival time to τ . Prior theoretical work on optimal migration strategies have considered a variety of decision variables, fitness criteria, and model formulations including analytical models of single decision variables, dynamic optimization models, and optimal annual routine models (review: Alerstam, 2011). Compared to previous models, ours is a deterministic model which considers two decision variables, and fitness criteria that include the minimization of time spent on migration and the minimization of predation risk (Alerstam and Lindström, 1990) and phenological match (Weber et al., 1998; Jonzén et al., 2007).

The first fitness component (S1) is the cumulative probability of surviving extrinsic mortality events, which depends on the instantaneous mortality rate per kilometer traveled (Cs) and travel distance (d).

$$\mathcal{C}\_0 = \exp(-\mathcal{C}\_s d)$$

Thus, migration over a longer distance was assumed to be more costly (cf. Bell, 1997). Extrinsic mortality events may include predation or exposure to severe weather. We further allowed C<sup>s</sup> to be a function of departure time, t0, assuming that costs begin to accrue during the first (predeparture) fueling period. In the commuting model, model results were not sensitive to the choice of formulation for S1, thus we only consider the formulation in which extrinsic mortality declines exponentially with t<sup>0</sup> (Jonzén et al., 2007):

$$\begin{aligned} \mathcal{C}\_{\mathfrak{s}} &= \mathcal{C}\_{0} \exp \left( -\alpha\_{1} t\_{0} \right) \text{ such that} \\ \mathcal{S}\_{1} &= \exp(-\left[ C\_{0} \exp(-\alpha\_{1} t\_{0}) \right] d) \end{aligned}$$

where C<sup>0</sup> is the maximum mortality rate when t<sup>0</sup> = 0, and α<sup>1</sup> determines the rate of decline in C<sup>s</sup> . This formulation accounts for an increased probability of total migration mortality when departing earlier in the year, in presumably harsher environmental conditions. The possibility of variable hazards across the migration route were not considered here. For baseline parameter values, we let d = 5,000 km and assumed C<sup>0</sup> = 0.0001 and α<sup>1</sup> = 0.03. Thus, S<sup>1</sup> = 0.61 when t<sup>0</sup> = 0, and S<sup>1</sup> = 0.98 when t<sup>0</sup> = 100.

The second fitness component (S2) reflects the costs associated with migration speed. In model 2.1, we assumed that these costs are mediated by wing shape, which can be optimized for flight speed or maneuverability, but not both (e.g., Hedenström and Møller, 1992; Vágási et al., 2016). A rounded wing shape that facilitates maneuverability is known to be important for escaping predation (Swaddle and Lockwood, 1998; Fernández-Juricic et al., 2006). Thus, wings adapted for long-distance migratory flights might be traded off against the ability to escape from predators (cf. Lank et al., 2017; Anderson et al., 2019). For comparison, in model 2.2 we assumed that costs associated with migration speed are mediated by foraging effort at stopover sites. Higher effort leads to a higher fueling rate and shorter stopovers, but at a greater risk of predation.

In model 2.1, we assumed birds follow a policy of timeselected migration and maximize their speed of migration, which is migration distance divided by cumulative flight time plus cumulative fueling time to cover flight costs (Norberg, 1981; Alerstam and Lindström, 1990; Hedenström and Alerstam, 1995). Under this fitness criterion, optimal migration speed (Vmigr) can be calculated as a proportion of optimal flight speed (Vflight) given the power requirement for flight (P in Watts) at Vflight and the fueling rate (k in Watts):

$$V\_{m\text{ign}} = \frac{kV\_{fl\text{light}}}{k+P}$$

P is affected by the overall elevation of the U-shaped power curve for flight. Lowering P, for example due to increased wing pointedness (which is related to aspect ratio), is expected to decrease Vflight, but only to a small degree compared to its effect on the time required for fueling (see Figure 1 in Hedenström et al., 2007). The effect of lowering P is to reduce the time required for fueling, which in turn increases Vmigr. For simplification, we assumed Vflight to be invariant in the model, and let P be the decision variable. We assumed that k is a constant that characterizes the environmental conditions for fueling (note that this assumption was relaxed in model 2.2).

In the model, we constrained the choice of flight power input P to be in the range of 1–4 W. For reference, P was estimated to be 1.6 W for ∼12.6 g free-flying yellow-rumped warblers (Setophaga coronata) (Guglielmo et al., 2017), and 4.2 W for ∼33 g freeflying Swainson's thrushes (Catharus ustulatus) (Gerson and Guglielmo, 2011).

An estimate of k (rate of fuel gain while activity foraging) was determined from empirical estimates of fat deposition rate (daily gain in fat mass as a proportion of lean body mass), the caloric value of fat (36.3 kJ/g), and the proportion of the day actively foraging (Lindström, 1991). Fat deposition rates of wild songbirds can vary among species and ecological circumstance (Moore and Kerlinger, 1987; Alerstam and Lindström, 1990; Schmaljohann and Eikenaar, 2017); we chose the estimate of 2.4% of lean mass d−<sup>1</sup> , which was the median value extracted from 31 species- and/or population-specific values (Alerstam and Lindström, 1990). Assuming a model 20 g lean bird, this corresponds to 0.48 g fat gain per day, which is equivalent to 0.202 W. Assuming an active foraging period of 12 h and 0 W fat gain overnight, this translates to 0.404 W fat gain while foraging. Based on empirical studies, Vflight was assumed to be 12 m·s <sup>−</sup><sup>1</sup> or 43.2 km·h −1 (Bruderer and Boldt, 2001).

Birds with power requirements for flight (P) increasingly below Pmax (due to increased wing pointedness) were assumed to experience a higher predation rate following this functional form: αp1[αp2(Pmax-P) 2 ], where αp1 is a constant which determines the minimum predation rate and αp2 is a parameter which determines how predation rate increases with decreasing P. This cost was intended to reflect a reduced ability to escape predators due to having more pointed wings. Surviving predation is then:

$$\mathcal{S}\_2 = \mathcal{S}\_{\mathcal{P}} = \exp(-\alpha\_{\mathcal{P}1} \left[ \alpha\_{\mathcal{P}2} (P\_{\max} - P)^2 \right]),$$

This function was chosen to be similar to one used in the commuting model. Unlike in the commuting model, however, S<sup>p</sup> is not a function of d. This is because wing shape is expected to affect predation rate all year, both during migratory and nonmigratory periods. We let αp1 = 0.005, αp2 = 10, and Pmax = 4 W. Thus, S<sup>p</sup> = 0.64 when P = 1, and S<sup>p</sup> = 1 when P = 4.

In model 2.2, we let fueling rate (k) be the decision variable and constrained the choice of k to be in the range {kmin, kmax}. For the minimum value, we let kmin = 0.2 W. We calculated kmax assuming a maximum fat accumulation rate of 5.4% d−<sup>1</sup> (Lindström, 1991), the caloric value of fat, and a daily foraging period of 12 h. After conversion, kmax = 0.9 W. In contrast to model 2.1, we assumed P was invariant and let P = 2 W. A higher fueling rate was assumed to carry a predation cost due to increased exposure, such that:

$$\mathcal{S}\_2 = \mathcal{S}\_k = \exp(-\alpha\_k k)$$

where α<sup>k</sup> is a parameter which determines how predation rate increases with k. We let α<sup>k</sup> = 0.2 so that S<sup>k</sup> = 0.96 when k = 0.2 W, and S<sup>k</sup> = 0.84 when k = 0.9 W.

For the third fitness component (f) in models 2.1 and 2.2, we assumed an optimal arrival date (day of year, τ ) with a reproductive penalty (reduction in offspring production) for arriving earlier or later than this time. As in the commuting model, the equation for this fitness component was:

$$f = \exp(-\alpha\_3 \left[t\_1 - \tau\right]^2)$$

where t<sup>1</sup> = t<sup>0</sup> + tmigr. Allowing reproductive output to be maximal at τ is similar to assumptions in Weber et al. (1998) and is consistent with the commuting model. For baseline parameters, we let α<sup>3</sup> = 0.0001 and τ = 125. Thus, f = 1 when t<sup>1</sup> = 125, and f = 0.94 when t<sup>1</sup> = 100 or 150.

Total time spent on migration (tmigr in days) depends on time in flight, time spent fueling for those flights, and time spent inactive while at stopover sites. Letting total flight time be d/Vflight and total fueling time be d/Vflight·P/k, the ratio between flight time and fueling time is 1:P/k (Hedenström and Alerstam, 1997). For example, if P = 2 W and k = 0.404, this ratio would be 1:5. If we let hflight be the hours spent in flight per day (because birds fly for only part of the night) and hfuel be the hours spent fueling per day (because birds fuel only during the day), then

$$t\_{m\text{igr}} = \frac{d}{V\_{fl\text{ight}}} \left(\frac{1}{h\_{fl\text{ight}}} + \frac{P}{kh\_{fucl}}\right).$$

Recognizing uncertainty and individual variability, we assumed hfuel = 12 h and hflight = 6 h. As an example, the calculated value of tmigr was 69 days after substituting values of d = 5,000 km, Vflight = 42.3 km·h −1 , k = 0.404, W, P = 2 W, hfuel = 12 h, and hflight = 6 h.

#### Finding {t ∗ 0 , P ∗ } or {t ∗ 0 , k ∗ }

Using a baseline set of parameters (d = 5,000 km, τ = 125 d, P = 2 W, k = 0.404 W, C<sup>0</sup> = 0.0001, α<sup>1</sup> = 0.03, αp1 = 0.005, αp2 = 10, α<sup>k</sup> = 0.2, α<sup>3</sup> = 0.0001, Vflight = 42.3 km·h −1 , hflight = 6 h, and hfuel = 12 h), models 2.1 and 2.2 produced fitness surfaces with intermediate optima (**Figure 4**). This suggests that the chosen functions and baseline parameters were reasonable approximations.

#### Predictions

As we did in model 1, simulations were run with randomized parameter combinations to explore the consequences of an earlier target arrival time (τ ) or longer travel distance (d). In each simulation (n = 1,000), τ and d were drawn from uniform distributions between minimum and maximum values. Each remaining parameter was chosen from a normal truncated distribution, with mean = baseline value, sd = mean/3, and bounding values defined by the 2.5 and 97.5% quantiles. Separate simulations were run for models 2.1 and 2.2.

Randomized simulations of model 2.1 showed that departure day t0, and to a lesser extent P, was sensitive to τ (**Figures 5A,B**). The correlations with τ were r = 0.28 and r = 0.10, respectively (**Table 1**). Thus, it is advantageous for birds to leave earlier and have a slightly lower power requirement for flight (e.g., greater wing pointedness) when needing to arrive at a destination earlier. Departure day, and to a greater extent, P were also sensitive to d (**Figures 5C,D**). The correlations with d were r = −0.44 for departure day and r = −0.74 for P (**Table 1**). Thus, it is advantageous for birds to leave earlier and have a lower power requirement for flight (e.g., greater wing pointedness) when migrating a longer distance. Randomized simulations of model 2.2 gave similar results: it is advantageous to leave earlier and to fuel slightly faster when needing to arrive at a destination earlier, and to leave earlier and fuel faster when migrating a longer distance (**Table 1**, **Figure 6**).

Departure day and migration speed were also sensitive to other parameters in models 2.1 and 2.2, and predictions largely recapitulated predictions from the commuting model (**Table 1**). Higher extrinsic mortality (determined by C<sup>0</sup> and α1) and lower predation rates (determined by αp1, αp2, and αk) favored later departure and slower migration via adjustments to P or k. A greater penalty for phenological mismatch (α3) favored earlier departure. Several parameters directly contributing to faster migration (k, hfuel, Vflight) favored later departure and slower migration via adjustments to P (model 2.1) or k (model 2.2). Longer nocturnal flights favored later departure but no adjustment to P or k.

Compared to the commuting model, target arrival time τ had larger effects on optimal P and k in models 2.1 and 2.2, respectively. We also found that the effect of τ on migration speed (P or k) depended on other parameters in the model. For example, in model 2.1, the effect of τ on P was stronger when migration distance d was higher or fueling rate k was higher (**Figure 7**).

#### Empirical Evidence

To assess the empirical evidence regarding the alternative mechanisms used to achieve an earlier target arrival date by songbirds, we tested qualitative predictions regarding withinpopulation sex and age effects. In other words, assuming sex and age differences in target arrival time at breeding areas, do males or adult birds depart for migration earlier and/or migrate faster than females or young birds? Within-population comparisons provide a robust evaluation of model predictions, because most ecological covariates are expected to be similar between comparator groups. Published information was compiled on sex and age comparisons in the onset of spring migration and other traits related to spring migration speed in migratory songbirds. An important caveat is that some species may not conform exactly to our modeled songbird, especially with respect to their body size and presence of an ecological barrier for en route fueling. Traits related to migration speed included migration speed (km·d −1 ) across the whole migratory journey, total migration duration (d), wing shape (usually pointedness), flight speed, fueling rate (based on mass change or plasma metabolite analysis), and stopover duration (d) (**Tables S1, S2**). For each study, we present the methodology (e.g., radio-telemetry, geolocator, markrecapture/resighting), species, statistical evidence of significant sex (M < F, M = F, M > F) or age effects (A < J, A = J, A > J; where A refers to adult or after-second-year birds and J refers to juvenile or second-year birds), and reported effect sizes for timing and speed traits. These data were then tabulated by effect direction and study. No formal meta-analysis was done to evaluate overall sex and age effects due to duplication of species across studies, variation in the number of species per study, and inconsistences in how data were analyzed.

We used an among-species, phylogenetically-controlled, comparative approach to assess the effects of migration distance on migration speed and departure day, as migration distance commonly differs among but not within populations. Published information was compiled for spring migration traits for 25 songbird species (**Table S3**), where the majority of studies used geolocators. One exception was Kirtland's warbler where spring migration duration was estimated from observations of colorbanded individuals (Ewert et al., 2012). These species differed in body size, the presence of ecological barriers for en route fueling, and other ecological traits such as trophic guild, and thus did not exactly conform to our modeled songbird.

days, k

\* = 0.42 W, and t

\*

<sup>1</sup> = 134 days. The fitness contours are based on a loess fit of model output with span = 0.005, with values indicated in the legend.



To test qualitative predictions from the migration model, we extracted or derived information on spring migration distance (km), departure day of year, migration speed (km·d −1 ), body mass (g), and breeding latitude. We only included one set of data for each species, and in cases where the same species had been studied more than once, we included the one with the biggest sample size. We used lean body mass if available and otherwise used the reported body mass. Only studies of three or more spring tracks were included. In studies where information about migration distance were lacking or not explicitly given, we derived migration distance either from breeding and wintering locations or we extracted approximate locations from published maps. Migration distances were calculated as orthodromes when there were no or only minor detours, or as the sum of migratory distances between consecutive stopovers from wintering to breeding locations when the birds made detours. From these data we derived overall migration speed, Vmigr, which is total migration distance divided by total time of migration, which should include the time for fueling before the first migratory flight (Lindström et al., 2019). We note that we may be underestimating migration distance and therefore underestimating Vmigr. This is partly compensated for by the fact that in most cases the duration of migration excludes the time for fueling before the first migratory flight. This is a notorious problem related to tracking studies, where onset of migration is usually defined based on when the birds start moving. In long distance migrants having many stopovers, the influence of excluding the first fueling episode from the duration of migration will have a relatively small effect. If estimated Vmigr exceeds 300 km·d −1 for songbirds, one should take that as an indication that the duration of migration is likely underestimated, or birds flew with tailwind assistance (cf. Hedenström and Alerstam, 1998). This was the case for two species in our data (**Table S3**), and therefore our analysis should be considered as provisional. In cases where body mass was missing, we obtained body mass from other sources (Dunning, 1993; Conway and Eddleman, 1994).

To account for relatedness among species, migration speed was analyzed using a phylogenetic generalized least squares model using function pgls in package caper (Orme et al., 2018). The main explanatory variable of interest was migration distance. Body mass and departure day were also included to account for these potentially confounding effects. In the analysis, migration speed, and distance were log-transformed to reduce skew. Phylogenetic information (n = 1,000 trees; Hackett backbone) was obtained from BirdTree.org (Jetz et al., 2014). A consensus tree was built using function consensus.edges in package phytools (Revell, 2012). In the pgls analysis, we optimized branch length transformations in a sequential fashion by fitting the parameters λ, δ, and κ by maximum likelihood. Models with different combinations of the explanatory variables were compared by AIC, where a lower AIC indicates a better fit. Breeding latitude was excluded from the global model because of difficulties in estimating λ when it was included, but we tested for its inclusion by AIC. Ordinary least squares regression was also done on the selected model for visual comparison with the pgls model. Departure day was similarly analyzed using function pgls with log-transformed migration distance and body mass included as explanatory variables.

#### Sex and Age Effects

An earlier onset of spring migration by males than females was commonly reported in migratory songbirds (14/18 cases). Based on reported effect sizes (n = 13 cases), the sex difference in onset was variable (−38 to 3 days) with a median value of −7 days (i.e., protandry in departure day). In contrast, sex differences in traits related to faster travel speed (greater migration speed, flight speed, or fueling rate; shorter migration duration or stopover duration) were reported in only 5 of 23 cases (**Table 2**, **Table S1**). In these five cases, males had traits consistent with faster travel speeds than females. Estimates of sex-specific migration speed, flight speed, migration duration, and stopover duration were sparse. Despite minimal evidence that males migrated faster than female, males usually had more pointed wings than females (6/8 cases).

Regarding age effects in spring migration, empirical data was more sparse than for sex effects (**Table 3**, **Table S2**). Only three studies examined age differences in the onset of spring migration, and these found inconsistent effects. Age effects were frequently reported for wing shape, with adults having more pointed wings than juveniles in 5/6 cases, and stopover duration, with adults have similar or shorter stopovers than juveniles in 4/7 cases. In 5/5 cases, fueling rates were similar between age classes.

#### Distance Effects

The model of log-migration speed with the lowest AIC included log-migration distance, departure day, and body mass. Birds migrated more quickly when they had longer migration distances (β = 0.318 ± S.E. = 0.123; t<sup>21</sup> = 2.6, p = 0.017) and when they departed later (β = 0.003 ± 0.002; t<sup>21</sup> = 2.2, p = 0.037; **Figure 8**). In this model, body mass was not significant (p = 0.756). Optimized branch length transformations were λ = 0.435, δ = 1.952, and κ = 0.942. The model of departure day with

the lowest AIC only included the intercept, so none of the explanatory variables were important.

fueling rate k (in W), respectively. (C,D) Show the effect of migration distance d on t0, and k, respectively.

#### DISCUSSION

To achieve an earlier arrival time at a destination, the commuting and migration models predict adjustments to departure time as the dominant mechanism, with adjustments to migration speed (adaptation of P or plasticity in k) playing a lesser role. Extending these results to avian protandry in arrival timing, our models predict that males should depart non-breeding areas before females, and sex-specific departure timing should be the primary mechanism underlying protandry. Males may also have a lower power requirement for flight or higher fueling rate, but these effects should be more subtle than for departure day. The model predictions largely agree with empirical data (**Table 2** and additional citations in Newton, 2008; Coppack and Pulido, 2009). Sex-specific departure from non-breeding areas is commonly observed in migratory songbirds, with males typically departing for migration about 7 days before females (cf. Briedis et al., 2019). In contrast, traits related to travel speed usually do not differ between the sexes. Where sexes do differ, however, they are in the expected direction for facilitating faster migration by males than females.

Variable results regarding sex-specific migration speed may be related to unaccounted for ecological covariates that act as selection agents on migration traits. For example, our model predicts sex differences in migration speed to be greater for longer distance migrants or under better fueling conditions, but such information on these and selection agents are generally unavailable. Moreover, environmental variables such as temperature or weather conditions can serve as cues for, or directly affect, migration traits (Ahola et al., 2004; Both et al., 2005; Marra et al., 2005; Knudsen et al., 2011; Haest et al., 2018). Seasonal carry-over effects can also influence phenology (Marra et al., 1998; Gow et al., 2019). Regarding the empirical evidence, we note that traits underlying migration speed are difficult to measure or may be inherently variable, small differences in these traits may be difficult to detect without large sample sizes, and departure day is not the same as the onset of predeparture fueling. Also, stopover duration may not be the best indicator of migration speed, because stopover departures can be associated with landscape-level re-locations within a stopover region, rather than directional, migratory flights (Taylor et al.,

TABLE 2 | Summary of sex effects in departure date and traits related to spring migration speed from published studies of migratory songbirds (see Table S1).


Each count represents one study and effect direction. M, males; F, females.

TABLE 3 | Summary of age effects in departure date and traits related to spring migration speed from published studies of migratory songbirds (see Table S2).


Each count represents a study and effect direction.

A, adults; J, juveniles.

2011; Schmaljohann and Eikenaar, 2017). Fueling rates represent a better index of migration speed (Lindström et al., 2019), but measuring fueling rates remains a challenge for wild songbirds in natural environments (Schmaljohann and Eikenaar, 2017).

The migration model was less able to recapitulate age differences in spring migration traits of songbirds. Accordingly to a theoretical model, the earlier arrival timing of adult birds than juvenile birds is evolutionarily favored due to within-sex competition for breeding territories, with adults outcompeting juveniles (Kokko et al., 2006). Thus, adults should have an earlier target arrival date τ . In light of our model, we would predict differential departure date from non-breeding areas as the primary driver of differential spring migration by age, but not differential migration speed. However, evidence regarding age-specific departure date is mixed and does not strongly support model predictions. In contrast to the model, adults and juveniles seem to differ more often in migratory speed, not because of slower fueling by juveniles, but perhaps because of longer stopovers and less efficient migratory flight behavior due to their wing shape. Age differences in stopover duration but not k also suggest a decoupling between k and stopover decisions in young birds, for reasons which may be related to the importance of energy rather than time minimization in young birds (Hedenström and Alerstam, 1997).

Slower migration speeds in juveniles than in adult birds suggests additional constraints acting on juveniles, which could limit the evolution of factors controlling migration speed such as P or k. Juveniles are known to differ from adults in many behavioral, morphological, and physiological traits, some of which persist to their first spring migration. For example, juveniles can retain their shorter and generally more rounded wings than adults, owing to the retention of their first primary feathers (Pyle, 1997). Thus, wing shape may be optimized for post-fledging and juvenile survival rather than for migration efficiency (Alatalo et al., 1984). Juveniles also may continue to show inexperience with navigation and orientation during their first spring migration. For example, juveniles show less tailwind selectivity than adults when making departure decisions both in fall (Mitchell et al., 2015) and spring (Morbey et al., 2018). In several species, juveniles in the fall have smaller flight muscles, larger digestive organs, and higher basal metabolic rates than adults (McCabe and Guglielmo, 2019), although it is not known if these effects carry over to their first spring migration. Further research on age effects during spring migration seem warranted. For example, if juvenile wing shape constrains spring migration, such effects should not be apparent for species with two complete molts per year (e.g., willow warbler Phylloscopus trochilus or bobolink Dolichonyx oryzivorus).

The commuting and migration models predict that longer travel distances should be facilitated by adjustments to departure

time and travel speed. The prediction that migration distance selects for reduced power required to fly through adaptation of wing shape is generally supported by comparative studies of simple or composite wing shape indices (Kipp, 1958; Marchetti et al., 1995; Mönkkönen, 1995). Aerodynamic models of flight cost include wing span and aspect ratio as descriptors of wing morphology, but simple or composite wing indices are correlated with aspect ratio (Hedenström, 1989). Conforming to the predicted effect of travel distance on travel speed, we found that longer-distance migratory songbirds have faster migration speeds. Similar findings were also reported in a recent analysis of migration speed among and within species of songbirds that used some of the same sources of data as ours, but used a different statistical approach (Schmaljohann, 2019). While these results are encouraging, estimated migration speed does not include the fueling episode before the first migratory flight.

In contrast to the predicted effect of travel distance on departure time, longer-distance migratory songbirds did not depart for migration earlier but in fact departed later. This was the case even though we included the ecological covariates body size and breeding latitude, and accounted for phylogenetic relatedness among species. However, the 25 songbird species differ in many other respects, including their migratory routes, habitat preferences, trophic guilds, mating systems, and molt strategies. These factors likely select for differences in the seasonal phenology of migration and breeding, and possibly swamp any effect of migration distance on the onset of spring migration.

Our migration model has the potential to inform predictions regarding phenological and morphological adaptation to climate change. Climate change at mid- to high latitudes has advanced breeding phenology and extended species ranges. Similar to optimal annual routine models (Hedenström et al., 2007; Jonzén et al., 2007), our model predicts that an earlier optimal breeding date should select strongly for earlier departure from nonbreeding areas. To a lesser extent, we also predict faster migration (e.g., a lower power requirement for flight or a faster fueling rate). Desrochers (2010) observed increased wing pointedness in boreal forest songbirds which was attributed to selection for increased mobility due to deforestation. According to our model, this pattern in migratory species could also be explained as an adaptation to climate change, assuming an advancement of optimal breeding date in these northern forests. If breeding populations shift northward in response to climate change, increasing migration distance should strongly select for faster migration, with less clear-cut effects on departure timing. On the one hand, longer migration distances favor earlier departure, but if optimal breeding dates shift to later in the year at higher latitudes, this would select for later departure.

In conclusion, we developed a simple optimization model of the onset day of spring migration and migration speed. Many of the model predictions agreed with empirical data, although age effects presented a challenging problem for future consideration. Although we focused on migratory songbirds, the model could be parameterized or re-formulated for other migratory or commuting systems. For example, a logical next step would be to model and summarize sex differences in the onset of spring migration and migration speed in shorebirds or in anadromous salmonids. As more migration studies are published in the near future, a consistent reporting of migration traits and important ecological covariates will facilitate future meta-analyses and development of theory.

#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

#### AUTHOR CONTRIBUTIONS

YM and AH contributed to model formulation and interpretation. YM wrote the code and carried out simulations, collated the data for **Tables S1, S2** and drafted the manuscript with input from AH. AH collated the data for **Table S3**.

#### FUNDING

YM was supported by an NSERC Discovery Grant. AH was supported by a project grant (2016-03625) and a Linnaeus grant (349-2007-8690) from the Swedish Research Council.

#### ACKNOWLEDGMENTS

We acknowledge helpful advice on source information from C. G. Guglielmo.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00492/full#supplementary-material

#### REFERENCES


annual cycle in a migratory songbird. Proc. Royal Soc. B 286:20181916. doi: 10.1098/rspb.2018.1916


in migratory songbirds: a cross-continental analysis. Mov. Ecol. 7:25. doi: 10.1186/s40462-019-0169-1


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Morbey and Hedenström. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Interacting Roles of Breeding Geography and Early-Life Settlement in Godwit Migration Timing

Phil F. Battley<sup>1</sup> \*, Jesse R. Conklin<sup>2</sup> , Ángela M. Parody-Merino<sup>1</sup> , Peter A. Langlands<sup>3</sup> , Ian Southey<sup>4</sup> , Thomas Burns<sup>3</sup> , David S. Melville<sup>5</sup> , Rob Schuckard<sup>6</sup> , Adrian C. Riegen<sup>7</sup> and Murray A. Potter<sup>1</sup>

<sup>1</sup> Wildlife and Ecology Group, Massey University, Palmerston North, New Zealand, <sup>2</sup> Conservation Ecology Group, Groningen Institute for Evolutionary Life Sciences, University of Groningen, Groningen, Netherlands, <sup>3</sup> Independent Researcher, Christchurch, New Zealand, <sup>4</sup> Independent Researcher, Auckland, New Zealand, <sup>5</sup> Independent Researcher, Nelson, New Zealand, <sup>6</sup> Independent Researcher, French Pass, New Zealand, <sup>7</sup> Independent Researcher, Waitakere, New Zealand

#### Edited by:

Yolanda E. Morbey, University of Western Ontario, Canada

#### Reviewed by:

Kevin C. Fraser, University of Manitoba, Canada Javier Perez-Tris, Complutense University of Madrid, Spain

> \*Correspondence: Phil F. Battley p.battley@massey.ac.nz

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 31 August 2019 Accepted: 19 February 2020 Published: 17 March 2020

#### Citation:

Battley PF, Conklin JR, Parody-Merino ÁM, Langlands PA, Southey I, Burns T, Melville DS, Schuckard R, Riegen AC and Potter MA (2020) Interacting Roles of Breeding Geography and Early-Life Settlement in Godwit Migration Timing. Front. Ecol. Evol. 8:52. doi: 10.3389/fevo.2020.00052 While avian migration timing is clearly influenced by both breeding and non-breeding geography, it is challenging to identify the relative and interdependent roles of endogenous programs, early-life experience, and carry-over effects in the development of adult annual schedules. Bar-tailed godwits Limosa lapponica baueri migrate northward from New Zealand toward Asian stopover sites during the boreal spring, with differences in timing between individuals known to relate to their eventual breedingground geography in Alaska. Here, we studied the timing of northward migration of individual godwits at three sites spanning 1,100 km of New Zealand's 1,400-km length. A lack of morphological or genetic structure among sites indicates that the Alaskan breeding population mixes freely across all sites, and larger birds (southern breeders) tended to migrate earlier than smaller birds (northern breeders) at all sites. However, we unexpectedly found that migration timing varied between the sites, with birds from southern New Zealand departing on average 9.4–11 days earlier than birds from more northerly sites, a difference consistent across 4 years of monitoring. There is no obvious adaptive reason for migration timing differences of this magnitude, and it is likely that geographic variation in timing within New Zealand represents a direct response to latitudinal variation in photoperiod. Using resightings of marked birds, we show that immature godwits explore widely around New Zealand before embarking on their first northward migration at age 2–4 years. Thus, the process by which individual migration dates are established appears to involve: (1) settlement by sub-adult godwits at non-breeding sites, to which they are highly faithful as adults; (2) a consequent response to environmental cues (i.e., photoperiod) that sets the local population's migration window; and (3) endogenous mechanisms, driven by breeding geography, that establish and maintain the well-documented consistent differences between individuals. This implies that behavioral decisions by young godwits have long-lasting impacts on adult annual-cycle schedules, but the factors guiding non-breeding settlement are currently unknown.

Keywords: geolocation, migration timing, phenology, photoperiod, Scolopacidae

# INTRODUCTION

fevo-08-00052 March 13, 2020 Time: 19:6 # 2

In birds, breeding ground geography, or more specifically geographic variation in breeding phenology, can be a major determinant of migration timing (e.g., Both, 2010; Conklin et al., 2010; Emmenegger et al., 2014; Briedis et al., 2016; Ouwehand et al., 2016), and associated processes of molt (Conklin and Battley, 2011a) and migratory fueling (Fry et al., 1972; Scheiffarth et al., 2002). This likely occurs through a combination of inheritance (genetic and/or parental effects) and response to early-life conditions (Ciarleglio et al., 2010), resulting in individuals showing natal site-fidelity (at least at a regional scale) and having a circannual program that enables timely arrival for breeding. Although migration timing can be very consistent within individuals (reviewed in Both et al., 2016), it is also subject to annual variation based on environmental conditions (Duriez et al., 2009; Conklin and Battley, 2011b), and can be modified over time through social information and individual improvement (Mueller et al., 2013; Sergio et al., 2014).

Additionally, migration timing can vary by non-breeding site, particularly in populations with a large non-breeding range (e.g., Myers et al., 1985; Piersma et al., 2005; van Bemmelen et al., 2019). This is to be expected, as populations traveling farther ought to start migrating earlier if they require longer to reach the breeding grounds. Less clear are the mechanisms that generate such population-level differences in phenology. Migration timing in birds is thought to be controlled by an internal circannual clock that is entrained by photoperiod (Gwinner, 1996a). Population-level comparisons indicate that differences in schedules can result from differences in the underlying circannual cycles and their responses to photoperiod (Helm et al., 2009), so the timing of migration of individuals may reflect both inherited circannual cycles and the photoperiodic environment the birds experience (Helm and Gwinner, 2005; Bojarinova and Babushkina, 2015). Hence, differences in timing could simply reflect photoperiod cues that vary geographically, or they could also arise through local environmental conditions (Dawson, 2008) or differences in migration strategy (Alerstam and Lindström, 1990). This means that annual-cycle schedules are not simply a product of the natal site, but can be modified by experience and both biotic and abiotic conditions after the first southbound migration.

In many species, adults show extremely high fidelity to nonbreeding sites (e.g., Lourenço et al., 2016), but we generally know little about how these sites are chosen. Non-breeding settlement may occur non-randomly, through ecological selection for certain aspects of phenotype (e.g., size, feeding morphology; Myers, 1981; Nebel, 2005) or competitive occupation of highquality sites (Gunnarsson et al., 2005; Studds and Marra, 2005), but a large element of chance may determine where juveniles end up at the end of their first southward migration (Thorup et al., 2003; Cresswell, 2014). In short-lived species that migrate to breed in their first year of life, it may then be difficult to differentiate among endogenous programs, early-life experience, and potentially temporary carry-over effects of natal or migratory conditions (Senner et al., 2015) in the development of life-long adult migration timing patterns.

By contrast, many migratory species show delayed maturity and do not migrate to the breeding grounds for one or more years. During these immature years, birds may be highly mobile and 'sample' potential non-breeding sites before settling at a site to which they remain faithful as adults (Battley et al., 2011). Thus, individuals in such species have potentially several years in which to make settlement decisions that may affect their subsequent lifelong migration schedules. For these species, the window in which information relevant to settlement decisions and migration timing is assimilated may be prolonged. Examination of behavior in this 'pre-migratory' phase of life may shed light on how routines as adults are established, with implications for site choice and timing of migration.

Bar-tailed godwits Limosa lapponica baueri provide a clear example of the relative individual timing of migration being predominantly 'set' by breeding ground geography on the other side of the world. Conklin et al. (2010) showed that bar-tailed godwits from a single non-breeding site on the North Island of New Zealand bred across the entire Alaskan range, from the Yukon-Kuskokwim Delta in the south to the North Slope in the north. As there is a difference of about 3 weeks in the timing of the spring thaw across that range, breeding opportunities arise much earlier for southern breeders than for northern breeders. This difference in optimal arrival date was reflected in the timing of migration of individuals across the entire northward migration, with southern breeders migrating earlier than northern breeders both from New Zealand and after a 4- to 6-week stopover in Asia (Conklin et al., 2010; Battley et al., 2012). Additionally, as godwit size also varies across Alaska (larger in the south, smaller in the north), larger birds in New Zealand tend to embark on northward migration earlier than smaller birds (Conklin et al., 2011). Monitoring of departures of marked birds, and repeat tracking of individuals by geolocators, showed that individual godwits were highly consistent in their timing of initiation and later stages of migration (Battley, 2006; Conklin et al., 2013). What we know about migration timing in baueri, however, comes almost exclusively from latitudes 37–41◦ S in New Zealand, while the non-breeding range extends from 34.5◦ S to 46.5◦ S in New Zealand, and extends much further north into the Tropics in eastern Australia. If migration timing varies by nonbreeding latitude, then the juvenile settlement period might have important impacts on annual-cycle schedules, with earlylife decisions modifying or over-riding endogenous programs derived from natal areas.

Here, we document variation in northward migration timing of bar-tailed godwits among three sites covering 1,100 km of the non-breeding range in New Zealand. Given that regional differences in migration timing could arise through population structure on the non-breeding grounds (i.e., differential settlement of southern- versus northern-breeders, which show some genetic differentiation: Parody-Merino, 2018; J. R. Conklin, unpublished data), we also test for population structure via biometrics (culmen length as a size measure) and neutral genetic variation (microsatellites). Then, we explore the pre-migratory settlement period of sub-adult godwits using resightings of marked birds to describe how extensively immature birds range around New Zealand and at what age they first

migrate north. We discuss the relevance of our findings for understanding the role of early-life experience and the interaction of breeding and non-breeding geography for the development of individual annual schedules.

### MATERIALS AND METHODS

#### Study Sites and Individual Marking

We studied migration timing of bar-tailed godwits in detail at three sites across New Zealand—the Firth of Thames near Auckland in the northern North Island (37.17◦ S, 175.32◦E), Manawatu River Estuary in the southern North Island (40.47◦ S, 175.22◦E) and the estuary of the Owaka River in the southern South Island (46.48◦ S, 169.70◦E)—we refer to these as Auckland, Manawatu, and Otago hereafter (**Figure 1**). The distances between sites are approximately 365 km (3.3◦ latitude) between Auckland and Manawatu, and 800 km (6.0◦ latitude) between Manawatu and Otago. Godwits were caught by cannon-net or mist-net, aged on the basis of plumage characteristics (age 1 = juvenile, 2 or 3 = immature non-migrant, or 3+ = adult), measured (exposed culmen, mm) and weighed, and individually

marked with either color-bands or a leg-flag engraved with a unique three-letter code. Juveniles were aged based on retained juvenile plumage. Age 2 and 3 birds were aged by a combination of features: presence during the boreal breeding season, stage of primary molt (starting during the late austral winter, so are more advanced than adults in the spring), retained juvenile outer primary feathers or greater primary coverts for age 2, presence of breeding plumage (suggestive of age 3) and relative primary feather wear [including the presence of replaced (unworn) primaries]. Adults could be distinguished by primary feather wear, extent of breeding plumage and later primary molt than younger birds. As aging of year 2 and 3 birds can be difficult, we group them here as immatures. Birds were sexed by culmen length (males = 70–99 mm, females 89–130 mm; Conklin et al., 2011), but ca. 10% of birds cannot be sexed by this method, due to overlapping ranges; for large males and small females, sex was confirmed by the extent of breeding plumage before departure (Conklin and Battley, 2011a). Birds were caught from 2004 onward in Auckland, 2006 onward in Manawatu and 2009 onward in Otago. As part of a wider study of movements of northern hemisphere shorebirds in New Zealand (Battley et al., 2011), we also banded godwits at a number of other sites spanning the length of New Zealand. Resightings of these birds have been compiled and we used this larger dataset to explore movements of sub-adult godwits.

#### Non-breeding Population Structure

There is a latitudinal cline in godwit body size across Alaska (northern = smaller; Conklin et al., 2011) and evidence of slight genetic structure in the breeding range (Fst = 0.013 between northern and southern breeders, based on microsatellites; Parody-Merino, 2018) that could also be present in the nonbreeding range. We looked for evidence of population structure among non-breeding sites using both morphometrics and neutral genetic markers. For biometric comparisons we also used data from birds caught at other sites around the Auckland region (Manukau and Kaipara Harbours) and Otago–Southland (Warrington, Otago, and Awarua Bay and Invercargill Estuary, Southland; see Battley et al., 2011 for site details).

A subset of birds was blood-sampled at the time of capture (95 in Auckland, 109 in Manawatu, and 19 in Otago) for genetic analyses. We genotyped 223 godwits at 27 microsatellite loci (full methodological details are provided in the **Supplementary Material**). For comparison with structure detected within the Alaska breeding range using the same microsatellite loci [Fst = 0.013 between northern (>65◦N) and southern (<65◦N) breeders; Parody-Merino, 2018], we calculated pairwise Fst among the three non-breeding sites using Arlequin v.3.5.2.2 (Excoffier and Lischer, 2010). To further explore potential non-breeding structure, we used the PRIORLOC function in STRUCTURE v.2.3.4 (Hubisz et al., 2009) to test whether birds from the three study sites formed distinguishable genetic clusters. The PRIORLOC function uses an individual's nonbreeding location in New Zealand (Auckland, Manawatu, or Otago) to estimate the most likely number of clusters in the population. K-means clustering Bayesian information criterion (BIC) indicated a single population (K = 1) or two clusters. To further explore potential subtle structure by study site, we ran STRUCTURE with PRIORLOC again with an assumption of K = 3. STRUCTURE was run with the following parameters: length of burn-in period = 1,000,000; MCMC runs = 500,000; number of iterations per run = 15. Results were visualized using Genesis v.0.2.5 (Buchmann and Hazelhurst, 2015).

#### Migratory Departures

fevo-08-00052 March 13, 2020 Time: 19:6 # 4

At each of the key study sites, we undertook monitoring of migratory departure in late February–early April by visual observation of marked birds, recording the last day of observation or, where possible, confirming the exact day of departure when a bird was seen in a departing flock (Conklin and Battley, 2011b). At the Manawatu site (population ∼200 birds), >80% of individuals were directly observed migrating, and the remaining departures were deduced from intensive daily resighting and flock counts. Geolocator conductivity values confirm that observation-based dates at this site are exact (Battley and Conklin, 2017). In Otago (∼350 birds) migration dates were mostly deduced from daily resighting and flock counts; four dates were derived from geolocators. The number of daily records of birds ranged from 1–24 with a mean of 7.0. Low values were associated with very early migrants. In Auckland (population 3,000+), the last date of observation for individuals seen repeatedly during the observation period was taken as the migration date, although seven birds were visually confirmed departing. We restricted records to those with six or more resightings unless a departure was observed, or a record with <6 resightings was later than records in other years for that individual or was corroborated by the dates in other years. The number of daily records of birds ranged from 1 to 26 with a mean of 9.8. The distributions of resighting frequencies for Auckland and Otago are given in the **Supplementary Material**.

There are some subtle biases in the determination of migration timing at the three sites. Very early departures in Auckland and Otago are likely to be overlooked, as repeated sightings are necessary to evaluate a bird's likelihood of being resighted and some birds seen only once soon after fieldwork started in Otago were not included as February departures. Last dates of observation for Auckland birds usually represent minimum estimates of departure date (as birds could not be confirmed as being absent in a large population), and the true departure dates for many birds will be later than assumed. Auckland birds might migrate on average slightly later than documented, but our dataset may underrepresent the early-departing sector of the population. Intensive departure monitoring took place at Auckland in 2014–2016, at Manawatu from 2008–2017 and in Otago in 2013–2016, but we conducted analyses of migration timing on the period of greatest overlap in the datasets, 2013–2016 (4 years for Manawatu and Otago; 3 years for Auckland). This resulted in a sample of 409 birds for which we had a migration date in one or more years (range 1–4 years); for birds with multiple years of data, we used an individual's mean migration date for analysis.

We compared migration phenology between sites using ANOVA with sex and site as factors, followed by a Tukey test for differences between levels of any significant factors. To test whether the relationship between size (culmen length) and migration date was consistent across all sites, we ran a linear model of migration date with bill nested within sex within site; this tests whether departure dates of birds within each sex varied by size, allowing for differences in migration timing between sites. Trends for each site (each sex separately) were compared via their slope estimates and 95% confidence intervals. The size-structures of the populations at the study regions were compared in ANOVAs with site as a factor but with sexes analyzed separately, with Tukey tests for differences between sites.

#### Geolocator Tracking

A subset of godwits from Manawatu and Otago were also tracked with light-level geolocators in 2013 and 2014. We retrieved 27 loggers (23 Migrate Technology Intigeo-CK65K and four Biotrack MK4093) from Manawatu birds and four from Otago (one Migrate Technology and three Biotrack). Loggers of one Manawatu and two Otago birds recorded data only as far as Asia. While there can be uncertainty about precise migration timing derived from analysis of light data, in shorebirds such as godwits, wetting of the logger during foraging or bathing means that extended dry periods clearly delineate non-stop migratory flights. This pattern is easy to identify in the conductivity data and these have been shown in godwits to give exact correspondence with observed migration departure dates (Battley and Conklin, 2017). We therefore used conductivity data to determine the departure date from New Zealand, duration of flight to Asia, length of the subsequent staging period in Asia, and the migration date toward the Alaskan breeding grounds, and compared these measures between Manawatu and Otago birds. Positional data were analyzed using Geolight (Lisovski and Hahn, 2012) and confirmed that all birds had their stopovers in the Yellow Sea region of eastern Asia. The Migrate Technology loggers also recorded min/max temperatures across each 4-h block of recording, which we use to evaluate the relative climate before and after the flight to Asia.

# Non-migratory Movements and Age of First Migration

As the distribution of godwits around New Zealand reflects the non-breeding settlement decisions of young birds, we used our database of resightings of individually marked godwits to identify how extensively young birds explore New Zealand and over how long a period. Of 2,020 total godwits marked during 1993– 2018, 327 were aged as 1–3 years when captured, including birds that could not be aged precisely but could be confirmed by a combination of date, plumage and wing molt not to be migratory adults (i.e., 3+ years old). To detect long-distance movements by sub-adults, we extracted all records of these immature birds that were seen away from their banding region (see Battley et al., 2011 for region details) before adulthood (≤3 years old). We summarized by region how many immature birds banded elsewhere had been recorded in that region. A similar analysis of adult resightings confirmed that they have virtually complete non-breeding site-fidelity between years (Battley et al., 2011).

We also use resightings to identify the age of first migration and to evaluate the period (age in months) over which young godwits were potentially still exploring New Zealand. For this we restricted the analysis to 215 individuals confirmed as age 1 (juvenile; n = 159) or age 2 (n = 56) when captured (on the basis of retained juvenile plumage). To estimate the ages of first northward migration we used records of these known-age birds either overseas (Asia or Alaska, demonstrating migration), or in New Zealand during the boreal breeding season (demonstrating non-migration). For these 215 known-age birds, we also looked at the age (in months, assuming hatch in June) of the last known long-distance (between-region) movement either southward or northward within New Zealand in the first 36 months of life, as most birds migrate by that time (see the section "Results"). For this, we recorded the bird's age at the last known resighting of the bird before a subsequent record in a new region; this represents the minimum age at the time of the move, and may underestimate the age by weeks or months. We excluded records of birds from the migration departure and arrival periods (March–April and August–October) to exclude movements that may reflect migratory stopovers rather than true non-breeding location shifts.

#### RESULTS

There was little evidence of population structure across New Zealand. Comparisons of the size distributions between sites found only subtle differences between sites (males, ANOVA F2,<sup>660</sup> = 8.651, P < 0.001; females, F2,<sup>772</sup> = 4.218, P < 0.02; **Figure 2**, upper panels). The range of sizes was similar across sites for both sexes (**Table 1**), but Manawatu males were significantly smaller on average than in both Auckland and Otago (Tukey post hoc test, P < 0.001) and Otago females were slightly larger than in Auckland (Tukey post hoc test, P < 0.01; **Table 1**).

Analyses of neutral genetic variation in microsatellites revealed no evidence for population genetic structure among non-breeding sites. All values of among-site pairwise Fst were indistinguishable from zero: Otago vs. Manawatu, Fst = −0.0001, P = 0.50; Otago vs. Auckland, Fst = 0.0011, P = 0.41; Manawatu vs. Auckland, Fst = −0.0004, P = 0.62. Moreover, the STRUCTURE analysis indicated uniform representation of assumed genetic clusters among sites (see **Supplementary Material**).

Godwits departed New Zealand from late February to late March, but the timing of migration varied by site (ANOVA, F2,<sup>405</sup> = 202.9, P < 0.001). There was a small difference between the two North Island sites (1.7 days; Tukey post hoc test, adjusted P < 0.05) but large differences between the South Island site (Otago) and both North Island sites (Otago departures being 9.3 days earlier than Manawatu and 10.9 days earlier than Auckland; Tukey post hoc test, adjusted P < 0.001 for both; **Figure 3** and **Table 2**). There was a slight difference in the timing of migration of males and females (males were 1.1 days earlier on average; ANOVA, F1,<sup>405</sup> = 5.343, Tukey post hoc test, adjusted P < 0.05). At each site, larger birds within each sex tended to depart earlier than small birds (**Figure 4**); these trends were statistically significant for all slopes (**Table 3**). The slopes of size vs. departure date were steepest at Manawatu, being significantly so compared to Auckland and Otago for males and compared to Otago for females, based on non-overlapping 95% confidence intervals of slope estimates.

The difference in migration timing inferred from the monitoring of marked individuals is corroborated by flock counts at the two smaller sites where numbers could be monitored closely (Manawatu and Otago; **Figure 5**). Across all 4 years of study, godwit numbers at Otago dropped dramatically in early March (on 4–8 March). In contrast, major declines at the Manawatu Estuary occurred only in the second week of March (9–14 March). In 2017, weekly counts were also available for the Avon-Heathcote Estuary in Canterbury (see **Figure 1**), 400 km NE of the Otago site and 400 km SW of Manawatu (A. C. Crossland, personal communication). The migration phenology matched that of Otago (gray points in **Figure 5**).

While, we had only limited geolocator tracking available for the Otago birds, comparisons with birds tracked from the Manawatu in the same years (**Figure 6**) showed that Otago birds departed from New Zealand earlier and arrived in Asia earlier than Manawatu birds, but departed from Asia around the same time. Specifically, Otago birds left New Zealand 12 days earlier on average (Otago: day 63.5 ± SD 5.0 days, Manawatu: 77.6 ± 4.7 days; t = −5.3, P < 0.01), took a similar time to fly to Asia (8.8 ± 1.0 days vs. 7.9 ± 0.6 days; t = −0.18, n.s.) and arrived in Asia around 13 days earlier (72.3 ± 4.6 days vs. 85.4 ± 4.7 days; t = −5.4, P < 0.01, n = 4 and 27, respectively, for all comparisons). The two Otago birds tracked after staging departed Asia within the same period as Manawatu birds (days 119 and 137 vs. 131.2 ± 9.3 days, range 119–156; **Figure 6**) and arrived in Alaska within the same period (days 124 and 140 vs. 139.4 ± 8.9, range 121–159 days, n = 2 and 26, respectively, for both comparisons), suggesting similar timetables at this stage of the migration.

Of 327 godwits marked when 1–3 years of age, 113 individuals were recorded making 119 movements between regions (55– 1,200 km from the banding site) before adulthood, showing that young birds range widely across New Zealand (**Figure 7**). This is necessarily an underestimate of movements made by young godwits, as it does not include: (1) movements made prior to initial capture, (2) brief stops missed by observers, and (3) temporary stops or permanent settlement at unsurveyed sites. Of 215 godwits of known age (marked at age 1–2), resightings provided information regarding age of first northward migration for 92 individuals (**Figure 8**); a combination of resightings unambiguously identified the age of first migration for 24 individuals, and narrowed it down to one of 2 years for an additional 68. A small number of birds migrated north at age 2, but most migrated north for the first time at age 3 or 4 (**Figure 8**). They therefore have a period of 2–4 years in which to settle in a non-breeding site from which they will subsequently migrate. Some young birds were still moving in their third year of residence in New Zealand, and birds were as likely to move northward within New Zealand as southward over that period (**Figure 9**). Compared to birds banded as adults, young birds had a much higher rate of being recorded away from the banding region (75 of 193 immatures with resighting histories (38%)

versus 173 of 1,208 adults (14%); Fisher exact test, P < 0.001). The adult records include birds caught on migration and birds seen on migration in New Zealand; only 19 adults (1.5%) appear to have relocated outside their banding region (evidenced by multiple consecutive resightings at those sites).

# DISCUSSION

We show that the timing of migration of bar-tailed godwits in New Zealand is more complex than realized from earlier studies, in which the only recognized driver of differences in migration timing was geographical variation on the breeding grounds that leads to consistent differences between individual birds within a non-breeding site (Battley, 2006; Conklin et al., 2010; Conklin and Battley, 2011b). We found an unexpected population-level difference in migration timing associated with latitude, with southern New Zealand birds migrating earlier than northern birds, and show that immature birds explore widely around the country before settling at a non-breeding site. This implies that the settlement decisions made by young birds set the 'window' within which departures may take place, and thus have life-long consequences for migration timing of individuals of this site-faithful species.

TABLE 1 | Summary of bar-tailed godwit bill lengths (mm, mean ± SD, range and n) by sex and region.


Males from Manawatu were significantly smaller on average than those from Auckland and Otago, while Otago females were slightly larger than Auckland females.

#### Geographic Differences on the Non-breeding Grounds

We expanded previous monitoring of bar-tailed godwits departing from the North Island of New Zealand to include birds from the southernmost extent of the non-breeding range, so that our three study sites spanned 1,100 km of the 1,400-km latitudinal 'length' of New Zealand. Three lines of reasoning led us to expect that migration schedules would be similar in Otago to elsewhere in New Zealand: (1) similar migration timing had been documented previously at different sites across the northern half of New Zealand (Battley, 1997; Battley, 2006; Conklin et al., 2010; Conklin and Battley, 2011b); (2) biometric analyses indicated little or no population structure in the non-breeding season, with godwits from across the Alaska breeding range mixing freely in New Zealand (Conklin et al., 2011); and (3) the flight lengths to Asia from each of our study sites were relatively similar, so that southern birds do not have appreciably farther to fly than northern birds.

Despite expectations, we found that departures from southern New Zealand were much earlier overall than those from central and northern New Zealand, a pattern that was consistent across all 4 years, and we further detected a small difference between the two North Island sites. Although godwits have a departure span of over 3 weeks at each site, the 9–11 days earlier initiation of migration in Otago meant that in some years half of the southern birds had departed before the northern birds had even begun to migrate.

Previous work from the Manawatu Estuary established that the timing of migration from New Zealand relates to an individual's eventual breeding latitude in Alaska, with birds from the southern extent of the breeding range (the Yukon-Kuskokwim Delta) leaving in early and mid-March, and birds breeding on the Seward Peninsula and North Slope not migrating until late March (Conklin et al., 2010). Because body size also varies along this S–N axis in Alaska, a relationship between body size and migration timing exists, with larger birds migrating earlier (Battley, 2006; Conklin et al., 2011). Our larger samples reinforce this previously described pattern across all sites. There was a significant negative relationship between migration date and body size for both sexes at all sites. While the slopes of the relationship were steepest at Manawatu, for any given body size Otago godwits leave substantially earlier on migration than do northern birds. Across New Zealand it seems that within a site, individuals vary according to the same 'rule' that arises from breeding-ground variation, but there is additional variation at the population level between non-breeding sites varying in latitude.

The slight differences in body size distributions and a lack of genetic population structure among study sites indicate that the observed differences in migration timing are not driven by geographic structure within the non-breeding range. At Manawatu, there were relatively few males >85 mm in bill length, which might cause a slight skew toward smaller, later departing birds at that site. However, such subtle differences cannot explain the magnitude of disparity in migration times between Otago and the North Island, or the regional differences for birds of the same size.

Recent work indicates that geographic variation in body size across the Alaska breeding range is accompanied by some degree of genetic differentiation, in both microsatellites and genomewide markers (Parody-Merino, 2018; J. R. Conklin, unpublished data). Our genetic analysis is based on the expectation that any potential genetic structure would be detectable among nonbreeding sites, if it was strong enough to drive different migration timing. Compared to differences between northern and southern breeders in Alaska detected in the same microsatellite loci (Fst = 0.013; Parody-Merino, 2018), we found no differences among sites in New Zealand: pairwise Fst values were effectively zero (all P > 0.40) and STRUCTURE detected no unequal distribution of genetic clusters. This lack of structure implies that godwits from different breeding areas are distributed approximately equally among non-breeding sites. Therefore, we are confident that hidden population structure cannot explain our results.

#### Why Do Southern Birds Depart Earlier?

There is no clearly adaptive reason for godwits in Otago to depart more than a week earlier from New Zealand. Migration distance alone cannot explain this: the straight-line (great circle) distance to the primary stopover area in the Yellow Sea, the Yalu Jiang National Nature Reserve in China, is ca. 10,000 km from Auckland and ca. 10,600 km from Otago. These flights differ by only 6%, a distance easily traveled by a godwit in less than 12 h of non-stop flight.

Earlier departure could potentially confer benefits of early arrival in Asia or Alaska. Although we have only two geolocator tracks from Otago birds, the limited data suggest that they do not arrive in Alaska earlier than other godwits; despite departing New Zealand earlier than all godwits tracked from Manawatu, departures from Asia and arrivals in Alaska were in the same ranges as for Manawatu birds (**Figure 8**). Both groups flew non-stop to the Yellow Sea region, so the earlier New Zealand departures do not reflect an alternative migration strategy, in terms of route or number of stops, but did achieve a longer staging duration in Asia. Godwits spend ca. 4–6 weeks in intertidal areas of the Yellow Sea (Conklin et al., 2010; Battley et al., 2012), during which they recover from the non-stop flight from New Zealand, complete their molt into breeding plumage (Conklin and Battley, 2011a), and fuel for the subsequent flight to Alaska. Additional stopover time, or a competitively early arrival, could therefore have benefits for a bird's condition upon arrival



Values represent individual mean departure dates of marked birds (1–4 years per individual) during 2013–2016. Results given are mean ± SD, range and n. Males departed slightly earlier on average than females, and birds from Otago departed substantially earlier than those from the other sites.

in Alaska, especially if food depletion occurs during staging and early-arriving birds have access to higher food levels than later-arriving birds (Choi, 2015). However, it is not clear why early arrival would be particularly advantageous for birds from southern New Zealand.

Early arrival in Asia may also come with energetic costs, given the potentially severe conditions at latitudes 35–40◦N in early March. Tidal flats in north-east China can still have substantial ice cover when the first godwits arrive (Choi, 2015), and cold conditions on arrival were confirmed by geolocators (22

with different symbols: males (smaller) are the circles, while females (larger) are the triangles. The symbol shading represents the three study sites: black = Auckland, dark gray = Manawatu, and light gray = Otago. Trends are shown by the fitted lines (from a linear regression of size nested within sex within site).

Manawatu, 1 Otago). For these 23 birds, the lowest temperature experienced in the week after arrival in Asia averaged −0.8◦C (range −4.4 to 3.0◦C), while the coldest 4-h block (i.e., with the lowest maximum temperature) averaged 2.6◦C (range −1.9 to 8.8◦C). These were considerably lower than temperatures in the week before departure from New Zealand (lowest temperatures: mean 7.5◦C, range 3.5–10.8◦C; lowest maximum: mean 12.5◦C, range 8.5–16.5◦C), so godwits are flying to colder conditions than they left from.

Still, godwits in Otago might face different energetic tradeoffs (i.e., the relative advantages of being in New Zealand or Asia), if they face more steeply declining temperatures or prey resources in February–March than do northern birds. We have no data to address fine-scale temporal variation in prey availability in New Zealand, but we find this explanation unlikely, as any scenario based on deteriorating conditions in southern New Zealand would also have to explain how these birds manage to fuel sufficiently for a 10,000 km non-stop flight earlier than more northerly godwits. If Otago godwits face harsher or more unpredictable conditions in New Zealand and Asia, we would expect them to experience lower or more variable annual survival or breeding success; as we also lack data to address this question, the fitness consequences of these migration differences remain unknown.

# Photoperiod and the Regulation of Migration Timing

In general terms, annual routines in birds are believed to involve an endogenous circannual cycle, which is entrained by photoperiod (Gwinner, 2003). Given that migrants experience a range of photoperiods through the year, there are complex interactions between photoperiod and circannual cycles that make birds responsive to critical daylengths at seasonally appropriate times. Differences between populations in the response to photoperiod and therefore the timing of annual cycle events can be regarded as 'adaptive population-specific reaction norms' (Gwinner, 2003). The influence of photoperiod has been studied most extensively in relation to the timing of breeding, but some key insights from studies of photoperiodism and the annual cycle are relevant to the timing of migration. First, a given cue-response system will show conditional plasticity, in which birds with identical photoperiod response systems will produce different, and potentially appropriate, timing of annual cycle events under different photoperiods (Hahn and MacDougall-Shackleton, 2008). Second, plastic responses to novel photoperiod conditions need not result in adaptive change (Coppack and Pulido, 2004).

A limited number of experiments have simulated, in effect, a range shift in migratory birds similar to our situation with godwits in northern and southern New Zealand. Gwinner (1996b) studied the nocturnal activity of garden warblers Sylvia borin exposed to photoperiods simulating 0◦ and 20◦ S, and showed that birds with 20◦ S photoperiods (outside the usual range) exhibited zugunruhe about 2 months earlier than those with equatorial photoperiods. He interpreted this advancement as being advantageous if it would allow individuals to reach the breeding grounds on time, despite a longer migration. Coppack et al. (2008) simulated a northward shift by pied flycatchers Ficedula hypoleuca from 10◦N to 50◦N, and found that the onset of migration was advanced by 25–33 days in all treatments (20◦N to 50◦N) compared with 10◦N, suggesting the existence of a photoperiod threshold between 10◦N and 20◦N. These studies indicate that photoperiod can have a direct influence on the timing of migration in birds, and that longer photoperiods resulted in earlier migration.

TABLE 3 | Slopes of the relationships between northward migration date and bill length for bar-tailed godwits in New Zealand.


Slopes were treated as significantly different if the 95% confidence intervals were non-overlapping; sites that differ are specified in the slope difference column.

FIGURE 5 | Flock counts through the migratory periods at the Manawatu and Otago sites, 2013–2016. March 1 is represented by day 60 or 61 (the latter in a leap year). Gray circles represent the Avon-Heathcote Estuary, Canterbury, South Island, in 2017.

In this context, it seems likely that the earlier migration of godwits in southern New Zealand represents a direct response to the longer photoperiods experienced throughout the southern summer by those birds. If true, there need not be any selective advantage to migrating earlier. The pattern of larger birds migrating earlier than smaller birds was similar across all sites, suggesting that individuals from across the breeding range respond similarly to photoperiod, regardless of the actual photoperiod experienced. This implies that the inputs to the finer-scale control of timing, derived on the breeding grounds (genetic inheritance, parental effects, and entrainment by perinatal conditions), are strong and persistent, and individuals from different parts of the breeding range respond differently to a common photoperiod environment at any given non-breeding site.

Photoperiod responses presumably evolved as adaptive systems to conditions experienced by given populations. The current timing of migration of godwits in southern New Zealand seems excessively early relative to the timing of birds further north. It could be that a general system in which birds living further from the equator leave earlier on migration is adaptive if the migration is income-fueled, with birds making multiple short flights and fueling at each stop. In contrast, the longdistance flights of shorebirds are fueled by large tissue deposits accumulated before migration starts (Piersma and Gill, 1998; Battley and Piersma, 2005), resulting in quick travel between very distant sites, changing the relative balance between active traveling time and overall migration speed. It could also be that the mechanism and response evolved under more northerly photoperiods, and result in appropriate local timing at those latitudes. If the distribution of godwits has expanded further south within the East Asian-Australasian Flyway, birds may be experiencing longer photoperiods than previously. It is not known whether the trans-Pacific migration system of Alaskan bar-tailed godwits evolved through a shift in the wintering range (from Asia to Australia and New Zealand) or from a shift in the breeding range from Russia to Alaska (Hedenström, 2010); the

former would entail a shift toward increasingly long photoperiods on the wintering grounds.

If the timing of migration of godwits at the population level does respond to photoperiod, this should lead to predictable differences in migration timing across the entire non-breeding range of latitude, which extends northward into the Tropics in eastern Australia. For a preliminary look at this, we compiled all previous information about bar-tailed godwit migration timing from New Zealand and Australia (Battley, 1997; Wilson et al., 2007; this study; **Figure 10**). Morphometric and phenology data from eastern Australia (Wilson et al., 2007) suggest that these sites also contain individuals from across the entire Alaska breeding range. There is no published information about migration phenology of L. l. baueri north of 32◦ S, however, two godwits were recently tracked by satellite-telemetry from Moreton Bay (27.2◦ S) to breeding sites on the north slope of Alaska (Z. Ma, personal communication). Their departure dates are ca. 1– 2 weeks later than northern breeding birds from Manawatu, and more than 3 weeks later than the latest observed departures from Otago (**Figure 10**). Although these studies include a variety of methods and time periods, and therefore are not ideally comparable, it appears that migration timing in New Zealand can be viewed as part of a cline that extends for the entire non-breeding range, as might be expected if differences are photoperiod-driven. Again, migration distance can explain very little of this variation, as a non-stop flight from Moreton Bay to the Yellow Sea is ca. 2,600 km shorter than from Otago, a difference of less than 2 days of flight.

A second bar-tailed godwit subspecies (L. l. menzbieri) breeds in northeastern Russia and spends the non-breeding season in western and northern Australia. This population is known to migrate later than baueri, both on departure from northwest Australia (**Figure 10**; Wilson et al., 2007) and arrival at staging sites in the Yellow Sea (Choi et al., 2015), which is generally attributed to its later breeding phenology

FIGURE 9 | Minimum age at which young bar-tailed godwits (n = 65) made their last regional movements within New Zealand before becoming migratory adults.

FIGURE 10 | Variation in timing of northward departure by bar-tailed godwits across non-breeding latitudes in New Zealand (filled circles) and Australia (open circles). For each site, line indicates range of departure dates directly observed or inferred from flock counts, and circle indicates date when ca. 50% of local population had migrated. Data sources: Battley, 1997; Wilson et al., 2007; this study. Diamonds indicate departures from Moreton Bay, Australia by two PTT-tracked godwits in 2019; both were tracked to breeding sites in northernmost Alaska (Z. Ma, personal communication).

and shorter migration distance (Battley et al., 2012). However, this intuitive interpretation is subject to confounding effects of non-breeding latitude. At one site in northwest Australia,

Verhoeven et al. (2016) found a surprising lack of differences in timing of fueling and migratory departure in two subspecies of red knots, Calidris canutus rogersi and C. c. piersmai. Based on their disparate phenologies on the breeding grounds (Chukotka Peninsula and the New Siberian Islands in Russia, respectively) and perceived passage times through the Yellow Sea, these populations were expected to differ by 2–4 weeks in departure timing; their indistinguishable timing leaving Australia suggests that common non-breeding geography effectively over-rides circannual schedules conferred by breeding geography, at least for the first stage of northward migration. If this similarly applies to bar-tailed godwits, we may more correctly view the migration timing of menzbieri as part of the latitudinal cline seen in baueri (**Figure 10**).

Other components of the annual cycle are also known to be influenced by photoperiod, and an additional question is whether photoperiod-driven differences in departure are reflected in similar differences in timing of molt and fueling, or carry through to later stages of migration and even breeding. We require more individual data on these factors, and complete northward and southward migration timing, to determine the extent to which non-breeding latitude influences phenology of the entire annual cycle.

# When Are Adult Annual Routines Established?

Regardless of the physiological mechanisms involved, we have shown that some portion of between-individual variation in migration timing in bar-tailed godwits is governed by nonbreeding site, and this is independent from variation associated with the natal site. This demonstrates that adult annual schedules, while guided to some degree by an endogenous program conferred by direct inheritance combined with the pre-fledging environment, are further modified according to behavioral decisions of young birds after arrival in the nonbreeding range. We have also shown that, although some young godwits appear to settle at their ultimate non-breeding sites quite quickly after arrival, others do not settle until the age of 2– 3 years or possibly later, providing quite an extended period for extrinsic forces to shape the highly repeatable behavior of adults. Furthermore, some young godwits arrive in Australia and subsequently shift to New Zealand as juveniles or immatures (Australasian Wader Studies Group, unpublished data), so there may be additional variation in when birds reach New Zealand resulting from differences in their initial southward migration. Once in New Zealand, young birds may move widely around the country, both northward and southward, indicating that postbanding movements are not just extensions of the first southward migration but appear to represent large-scale 'sampling' of habitats around the country.

It is not clear whether this suggests a prolonged 'ontogenetic window' (sensu Senner et al., 2015) for godwits and other avian species showing delayed maturation, in the sense of having a longer period of 'developmental plasticity' (sensu Piersma and Drent, 2003). If the population departure time is set by a simple response to a local photoperiod, then birds might simply need to have settled at a site for a single summer before migrating, to match other local individuals. What is more interesting is whether the internal cues for relative migration time are reinforced by repeated exposure to local photoperiod (being stronger in early-settling birds), and whether these cues are reinforced with migration to the breeding grounds.

It is intriguing that young godwits vary substantially in both when they settle at a non-breeding site and when they make their first northbound migration. Currently, we lack the data to determine whether these timings are linked. If earlier-settling birds indeed also migrate at a younger age, the causality could plausibly operate in either direction: (1) birds are somehow predisposed to migrate at different ages and then settle accordingly to ensure timely preparation for the first northward migration, or (2) the act of settling effectively initiates the adult annual cycle, including molt and fueling, after which migration naturally ensues. In the latter scenario, age of first migration could be influenced by the specific time of year that a bird settles at its final non-breeding site. For example, if godwits use an environmental cue to begin migratory preparation (e.g., photoperiod in late December), perhaps birds that have not settled by this time are insensitive to the cue and thus delay migration until the following year. Alternatively, all birds are sensitive to the cue, but birds that have not yet settled simply cannot complete migratory preparation in time. So, it is possible that age of first migration is a pre-determined strategy that varies among individuals, or a carry-over effect of circumstances experienced after arrival in New Zealand.

With its demonstrated influence on migration schedules in adult bar-tailed godwits, the processes and circumstances promoting non-breeding settlement by subadult birds may have life-long effects on behavior of individuals. Observed premigratory movements in New Zealand suggest a variable period of 'sampling' before individuals 'choose' a non-breeding site, to which they are extremely faithful as adults. To understand this process, the first step is to quantify the between-individual variation in duration and extent of site-sampling, and to link this with adult migratory behavior; this requires tracking individual movements from first arrival in New Zealand until the adoption of adult routines. The next step is to understand the specific processes that promote an individual's movement or settlement, which likely include the interaction of intrinsic factors (e.g., personality, quality, circannual rhythm, condition) and extrinsic aspects of sampled sites, such as carrying capacity, prey types, and social environment.

# DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

All work was undertaken under appropriate Animal Ethics permits from the University of Otago and Massey University and Department of Conservation approvals.

#### AUTHOR CONTRIBUTIONS

fevo-08-00052 March 13, 2020 Time: 19:6 # 13

PB and JC conceived the study and wrote the manuscript. PB, JC, and ÁP-M analyzed the data. All authors participated in bird captures. DM, AR, and RS led substantial cannonnetting efforts as part of this. JC, PL, IS, and TB conducted the departure monitoring.

#### FUNDING

Much of this work was supported by a Marsden Grant (MAU1202) from the Royal Society of New Zealand to PB. PB was also supported by the Massey University Research Fund. JC was supported by a Massey University Doctoral Scholarship, the Manawatu Estuary Trust, the Dobberke Foundation for Comparative Psychology, and the Academy Ecology Fund of the Royal Netherlands Academy of Arts and Sciences (KNAW). Early banding work was funded by a Foundation for Science, Research and Technology Postdoctoral Fellowship to PB (UOOX0231), and by the New Zealand Department of Conservation (Investigation number 3739 to PB, DM, and RS).

#### REFERENCES


#### ACKNOWLEDGMENTS

Thanks to B. Helm and J. Karagicheva for discussions about photoperiodic control of migration. Assistance with godwit capturing, and resightings of marked birds, were provided by numerous people—thanks to them all! Thanks to A. Crossland for the counts from the Avon-Heathcote Estuary in Christchurch. Thanks to A. Fidler and Y. Verkuil for assistance with microsatellite analyses, S. Lisovski for assistance with geolocator analyses, and Z. Ma (Fudan University) for access to godwit tracking data. We would also like to thank the various Maori iwi around New Zealand who have given their support to our kuaka research, including Ngati Kur ¯ ¯ı and Te Aupouri Trust Boards, Ngati Paoa, Rangitaane o Manawat ¯ u, ¯ Ngati Raukawa and Ngai Tahu.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo.2020.00052/ full#supplementary-material



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Battley, Conklin, Parody-Merino, Langlands, Southey, Burns, Melville, Schuckard, Riegen and Potter. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Migration Patterns of Upland Sandpipers in the Western Hemisphere

Jason M. Hill <sup>1</sup> \*, Brett K. Sandercock <sup>2</sup> and Rosalind B. Renfrew<sup>1</sup>

*<sup>1</sup> Vermont Center for Ecostudies, Norwich, VT, United States, <sup>2</sup> Department of Terrestrial Ecology, Norwegian Institute for Nature Research, Trondheim, Norway*

Integrated models of the ecology of migratory species require tracking of individual migratory organisms throughout the annual cycle. Here, we report the first information on the movement patterns of nine Upland Sandpipers (*Bartramia longicauda*) that were captured at breeding sites in Kansas and Massachusetts, and tracked with GPS and PTT tags to non-breeding sites in South America. Upland Sandpipers were extreme migrants that regularly made non-stop flights that were >5,000 km in length and lasted up to 7 days. Sandpipers traveled up to 20,000 km per year in their annual movements. Our project resulted in a series of new discoveries. Sandpipers regularly crossed major ecological barriers during migration, which included long oceanic flights, high elevation mountains, and tropical forests. Migrating birds used known stopover sites in the central flyway of North America and eastern slope of the Andes in South America, and a subset of birds wintered in core non-breeding sites in the Pampas ecoregion of Uruguay and Argentina. We documented new staging sites at canefields in the mountain valleys of Colombia, grasslands in the Llanos of Venezuela, and at airports along the Atlantic Coast of the US. Unexpectedly, some sandpipers spent the non-breeding season on river islands in the Amazon basin, and pastures in the Cerrado ecoregion of Brazil; areas not previously known to host overwintering Upland Sandpipers. Like many other migratory birds in the Western Hemisphere, Upland Sandpipers had elliptical migration routes within the Southern Hemisphere, moved among separate activity areas during the non-breeding season, migrated faster during northbound than southbound migration, and spent more time at non-breeding than breeding sites. Collectively, the birds used sites across much of northern South America as a broad front migrant. Overall, the migratory patterns of Upland Sandpipers were more similar to migratory landbirds than to shorebirds that typically stage at wetlands and coastal estuaries. Upland Sandpipers should be buffered against habitat loss and degradation at local sites within their migratory range, but it may be difficult to protect specific sites or broad landscapes that would be needed to conserve a high percentage of the global population.

Keywords: Bartramia longicauda, elliptical migration, full annual cycle, long-distance migration, seasonal, space use, transoceanic flight

#### Edited by:

*Shannon J. McCauley, University of Toronto Mississauga, Canada*

#### Reviewed by:

*Celina Baines, University of Toronto Mississauga, Canada Kevin C. Fraser, University of Manitoba, Canada*

> \*Correspondence: *Jason M. Hill jhill@vtecostudies.org*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *01 August 2019* Accepted: *21 October 2019* Published: *19 November 2019*

#### Citation:

*Hill JM, Sandercock BK and Renfrew RB (2019) Migration Patterns of Upland Sandpipers in the Western Hemisphere. Front. Ecol. Evol. 7:426. doi: 10.3389/fevo.2019.00426*

# INTRODUCTION

Long distance migration between breeding and non-breeding areas is a common life history strategy for animals that live in seasonal environments, and understanding the biology of migratory movements is central to four core areas of ecology. In functional ecology, mechanistic questions include study of the morphological and physiological adaptations for long distance movement, the navigation systems used to travel through unfamiliar environments, and the timing of migration in relation to reproduction, feather molt, organ regulation, and other events in the annual cycle (Piersma and Gill, 1998; Hedenström, 2008; Alfaro et al., 2018). In population biology, the goals are to understand the role of food, predation, and climate as limiting factors, and how carryover effects and density-dependence may act to regulate population numbers (Newton, 2006; Rakhimberdiev et al., 2015; Watts et al., 2019). In evolutionary ecology, the adaptive significance of migration is considered in terms of tradeoffs between the demographic costs and benefits of movement, and the role of past events and phylogenetic history in shaping migratory routes (Ruegg and Smith, 2002; Alerstam et al., 2003). In conservation biology, increasing concerns about population declines among migratory species have increased interest in identifying stages of the annual cycle and sites where conservation actions can be targeted, and understanding patterns of migratory connectivity that link spatially structured populations at different stages of the annual cycle (Vickery et al., 1999; Fraser et al., 2012; Jahn et al., 2017; Pearce-Higgins et al., 2017; Cohen et al., 2018). Integration of knowledge across these four key areas has become the basis for development of full-annual-cycle (FAC) models (Hostetler et al., 2015; Marra et al., 2015).

Integrated models of the ecology of migratory species require tracking of individual small-bodied animals across continentalscale distances. New technologies including geolocators, GPS tags, satellite transmitters, and other miniaturized tracking devices have provided a wealth of new movement data (McKinnon and Love, 2018; Sergio et al., 2018; Hofman et al., 2019). The emerging information has shown that migration systems are shaped by species characteristics such as phylogeny, body size, and social systems, as well as environmental features such as the distribution of critical habitats and ecological barriers to migration.

Among migratory birds, shorebirds are remarkable examples of long-distance migrants that often travel up to 20,000– 40,000 km in the course of their annual cycle (Battley et al., 2012; Lanctot et al., 2016; Conklin et al., 2017). Migration strategies are often closely linked to body size due to physiological relationships between fuel stores and flight efficiency that determine maximum flight distance (Warnock, 2010). Among migratory shorebirds, small-bodied species can make short "hops" of 1,000–5,000 km (Western Sandpiper Calidris mauri; Iverson et al., 1996; Semipalmated Sandpiper C. pusilla; Brown et al., 2017), mid-sized species may "skip" up to 5,000–8,000 km (Great Knot C. tenuirostris Lisovski et al., 2016; Red Knot C. canutus; Niles et al., 2010), whereas large-bodied species can make longer "jumps" of up to 7,000–11,000 km (Ruddy Turnstones Arenaria interpres; Minton et al., 2011; Bar-tailed Godwits Limosa lapponica; Battley et al., 2012; Hudsonian Godwits L. haemastica, Senner et al., 2014). Space use and movements of migratory shorebirds are also linked to variation in mating systems and social behavior. Socially monogamous and polyandrous species often show strong fidelity to breeding sites (Weiser et al., 2017; van Bemmelen et al., 2019), whereas promiscuous species can be more vagile with low site fidelity and wide-ranging movements during the breeding season (Lanctot et al., 2016; Kempenaers and Valcu, 2017).

The migratory shorebirds that breed or stage in native grasslands of North America include both short-distance migrants that remain on the continent (Page et al., 2014; Pierce et al., 2017; Ruthrauff et al., 2019), and long-distance migrants that travel to South America (Blanco and López-Lanús, 2008; Penner et al., 2015; Jahn et al., 2017). In the Western Hemisphere, intercontinental shorebird migrants must cross major ecological barriers including the water barriers of the Gulf of Mexico and Caribbean Sea, high elevation terrain in the Andes mountains, and unsuitable habitats including the vast tropical forests of the Amazon Basin (Bayly et al., 2018). In this study, we used new tracking technologies to conduct one of the first investigations of the individual, year-round movements of Upland Sandpipers (Bartramia longicauda). Upland Sandpipers are long distance migrants that use temperate grasslands on both the breeding grounds in North America (Bowen and Kruse, 1993; Garvey et al., 2013; Sandercock et al., 2015) and non-breeding grounds in southeast South America (White, 1988; Blanco and López-Lanús, 2008; Alfaro et al., 2015, 2018). Little is known about migratory connectivity of this species because banding effort has been low and band recoveries are rare (Garber et al., 1997; Houston et al., 1999), and previous satellite tracking has provided information on southbound migration for a single bird (Grosselet et al., 2019). Based on specimen and natural history records, migratory routes are thought to include corridors through the Great Plains and Atlantic Coast, stopover sites in central America, and southern routes east of the Andes (Blanco and López-Lanús, 2008; Houston et al., 2011), as well as newly discovered sites along the Pacific coast of northern Chile (Medrano et al., 2018).

Population numbers of Upland Sandpipers are stable within their core range in the Great Plains of US and Canada, but are declining at the edge of their distribution in the Upper Midwest and New England (Osborne and Peterson, 1984; Houston, 1999; Vickery et al., 2010; Andres et al., 2012). The Upland Sandpiper is an area-sensitive species that requires large tracts of native grasslands with heterogeneous vegetative structure (Vickery et al., 1994; Sandercock et al., 2015). Threats on the breeding grounds include habitat loss due to expansion of rowcrop agriculture and afforestation in New England (Vickery et al., 1999; Foster et al., 2002), and habitat degradation due to changes in rangeland management in the Great Plains (Sandercock et al., 2015; Hill and Renfrew, 2019a). Threats on the non-breeding grounds include intensification of livestock grazing and loss of grasslands to rowcrops (Blanco and López-Lanús, 2008; Jahn et al., 2017). Threats during migration include exposure to agricultural chemicals and legal harvest (Strum et al., 2010; Pérez-Arteaga et al., 2019), and regular mortality events have been reported at high elevation lakes in the central Andes of Ecuador (Vickery et al., 2010). Conservation planning for Upland Sandpiper has been difficult because migratory strategies, key habitats, and migratory connectivity have been unknown (Vickery et al., 2010; Jahn et al., 2017).

We fitted adult Upland Sandpipers with satellite tags on their breeding grounds in Kansas and Massachusetts. Our objectives were to use new tracking technologies to collect basic data on the migratory routes, timing, and movement behavior of individual birds. A number of shorebirds in the Western Hemisphere migrate along north-south routes (Myers et al., 1990; Page et al., 2014), sometimes with westerly routes in spring and easterly routes in autumn (Senner et al., 2014; Brown et al., 2017). We predicted that Upland Sandpipers from Kansas and Massachusetts would migrate elliptically along separate routes in the midcontinent and Atlantic Coast but converge in the nonbreeding range in Uruguay and northern Argentina. Population studies have indicated that the breeding period in Kansas lasted 3.0 mos from late April to mid-July, and the non-breeding season in Uruguay lasted 3.5 mos from mid-November to late February (Sandercock et al., 2015; Alfaro et al., 2018). We predicted that sandpipers would be stationary during these two stages but would be in transit at other times of year. Upland Sandpipers are a mid-sized shorebird with a body mass averaging 140–160 g, and we predicted that they might be able to make long "skips" of 5,000–8,000 km. The species is socially monogamous with a mate defense mating system, semi-colonial nesting, and biparental incubation (Bowen and Kruse, 1993; Casey et al., 2011). Color-banded adults have annual return rates of 38.1% (n = 189 birds, Mong and Sandercock, 2007). We predicted that adult sandpipers would show site fidelity to breeding areas.

# METHODS

#### Study Sites and Field Methods

Our field sites included three US military installations and a natural preserve located in the western and eastern parts of the continental range of Upland Sandpipers. Our two field sites in Kansas included Fort Riley (39.207◦N, −96.824◦W) and the Konza Prairie Biological Station (Konza; 39.100◦N, −96.608◦W). Field sites in Massachusetts included Joint Base Cape Cod (Cape Cod; 41.658◦N, −70.521◦W) and Westover Air Reserve Base (Westover; 42.199◦N, −72.542◦W). The habitat at our Kansas field sites was tallgrass prairie dominated by warm-season grasses with a mixture of forbs. The field sites were used for military training or ecological research. The adjoining lands in Kansas were private rangelands managed with prescribed fires in spring and used for cattle grazing, and were also suitable habitat for sandpipers (Sandercock et al., 2015). Our field sites in Massachusetts were active airfields with air strips surrounded by open fields dominated by cool-season grasses with a short sward during our field work (<30 cm). At Cape Cod, we also captured sandpipers in grassland habitats at a covered landfill that was also on the base and 0.8 km from the airfield. Landscapes surrounding our field sites in Massachusetts had small patches of grasslands embedded in a matrix of suburban

FIGURE 1 | Upland Sandpipers with tracking tags attached with an elastic leg-loop harness; solar-powered Argos Platform Transmitter Terminals (5 g PTT, Microwave Telemetry, USA, Top) and battery-powered PinPoint Argos-GPS tags (4 g, Lotek Wireless, Canada, Bottom). The two birds were captured at Konza Prairie, Kansas on the night of 4 May 2016 (KO-PTT-69, Top; KO-GPS-02, Bottom).

housing, golf courses, and forested areas, and were less suitable for sandpipers.

We captured Upland Sandpipers during April and May 2016. We searched for roosting birds at night by driving slowly (∼5 km h −1 ) along dirt tracks in the prairie or on the edge of airport runways. Roosting sandpipers were located with handheld spotlights, and then approached on foot and captured with longhandled dip nets. We recorded body mass and morphometrics at capture and considered birds to be females if they were >160 g in body mass or if we could detect the presence of an egg by palpitating the abdomen. We attached tracking tags to sandpipers with leg loop harnesses constructed from elastic cord (1 mm, Stretch Magic, Pepperell Crafts, Massachusetts, US). Harnesses were individually fit to each bird by adjusting the leg loops around the upper thigh so that the tag was positioned above the pelvis with a whip antenna extending out over the tail (**Figure 1**). Harnesses were individually adjusted for a relaxed but secure fit and were secured with one double-overhand knot that was


TABLE 1 | Summary of movement data and status of Upland Sandpipers monitored with PinPoint GPS Argos tags (GPS tags), and solar-powered Argos Platform Transmitter Terminals (PTT tags) from 24 April 2016 until 1 May 2017.

*Table sections include information regarding each bird's capture (location and date of capture), monitoring (date of first and last location fix), tracking data (number of unique days with location fixes and the total number of location fixes received during the monitoring dates), and its last known location. BirdID included a 2-digit code for the breeding location, a 3-letter code for the tag type (PTT vs. GPS), and a unique 2-digit identifier. SD-GPS-90 was captured as transient migrant in Kansas but moved northward to a breeding site in South Dakota. Sex was based on body mass at capture (F, female; M, male; U, unknown).*

*<sup>a</sup>Known mortality.*

*<sup>b</sup>Possible mortality.*

coated with a thin film of Loctite superglue (Henkel Corporation, Connecticut, US). Our harness design has little effect on behavior or seasonal survival of sandpipers but may reduce annual return rates (Mong and Sandercock, 2007; Smith et al., 2017). Individual sandpipers were identified by a unique Bird ID: a two-digit code with the breeding location (e.g., FR = Fort Riley and WO = Westover), a three-letter code representing the tag type (PTT vs. GPS), and a unique two-digit identifier (**Table 1**).

#### Tracking Tags and Movement Data

We used two types of tracking tags to investigate the migratory movements of Upland Sandpipers: 4-g battery-powered PinPoint GPS Argos tags (GPS tags, Lotek Wireless, Canada), and 5 g solar-powered Argos Platform Transmitter Terminals (PTT tags, Microwave Telemetry, Inc., Maryland, US). The GPS tags were less expensive than the PTT tags (ca. US\$1,200 vs. \$4,500), but PTT tags can provide real time movement data with more locations. We opted to use a combination of GPS and PTT tags to obtain good quality movement data for a representative sample of birds, but our study was not designed to directly compare tag performance. Mass and dimensions were similar for both types of tags. GPS tags were 2.5 L × 1.4 W × 0.7 H cm and had an 18-cm whip antenna, which was reinforced at the base to guard against bird-inflicted damage. Solar PTT tags were 2.5 L × 1.5 L × 0.8 W cm wide with a 22 cm whip antenna. The average body mass of sandpipers that received GPS tags was 171.7 ± 34.3 (SD) grams (range = 135 to 229 g, n = 11), whereas the average body mass of birds that received PPTs was 175.5 ± 15.6 (SD) grams (range = 162 to 196 g, n = 4). The tracking devices with the elastic harness were <3% of the bird's body mass at the time of capture. All of the Upland Sandpiper movement data recorded by the GPS and PTT tags were archived at Movebank (www.movebank.org; Hill and Renfrew, 2019b), an open online database for animal tracking data (Wikelski and Kays, 2018).

# GPS Tags

We deployed PinPoint GPS Argos tags on 11 Upland Sandpipers between 24 April and 25 May 2016. Six birds were captured in Kansas (Fort Riley: n = 2; Konza: n = 4), and five birds were captured in Massachusetts (Cape Cod: n = 1; Westover: n = 4). The GPS tags had an expected battery life of just under 1 year, and a memory capacity for storage of 30 locations from GPS fixes with an expected accuracy of ∼10 m. Movements and habitat use of Upland Sandpipers have been studied on the breeding grounds (Mong and Sandercock, 2007; Sandercock et al., 2015). Here, we were mainly interested in collecting movement data outside of the breeding season, and we programmed the GPS tags to start collecting locations ∼2 months after deployment, on 15 July 2016, and to continue for a 9-month period until 15 April 2017. We set the check frequency for biweekly fixes during the expected stationary periods of July to August and December to January, and for weekly fixes during the expected migratory periods of September to November and February to April. All GPS location fixes were programmed to occur during midday at 12:00 UTC, and were sequentially added to the tag memory. To recover information from the GPS tags, we set a pre-programmed date of 15 April 2017 to automatically upload all stored data on the tags to the Argos satellite system. We anticipated that some sandpipers would still be migrating northward in April 2017, but we expected battery life to expire by 1 May 2017 (Lotek Wireless, pers. comm.). Movement data were then downloaded from the satellite system and sent to us via email. Successful recovery of movement data required the GPS tag to be functional, and its battery operational through 15 April 2017. Movement data could not be recovered if a tag malfunctioned or was damaged in the 12-month period before the scheduled upload date.

## PTT Tags

We deployed solar Argos Platform Transmitter Terminal tags on four Upland Sandpipers between 23 April and 25 May 2016. Two birds were captured in Kansas (Konza: n = 2), and two birds were captured in Massachusetts (one each at Cape Cod and Westover). We were able to monitor several birds for multiple years with PTT tags, but here we limit our analysis to the first year of tracking data collected from April 2016 through April 2017. PTT transmitters are monitored by the Argos satellite system, which is operated by Collecte Localisation Satellites (CLS). Transmitter locations are calculated via the Doppler effect based on the frequency shift in the signal from the transmitter that is received by the Argos satellites (Lopez et al., 2013). Thus, the solar PTT tags can produce multiple daily locations for multiple years but the accuracy and precision of locations can vary depending on environmental conditions. The Argos system assigns location estimates from PTT tags to seven different location quality classes based on a Kalman filtering algorithm (LC 3 to LC B). Some location estimates can be too imprecise to receive any error estimate, but the assigned error radii usually range from <0.25 to >1.5 km (Douglas et al., 2012). Validation trials with stationary PTT tags at known locations have suggested that the error estimates from CLS may be too optimistic, and in one study >75% of the true locations were not contained within the estimated error polygon (Douglas et al., 2012; Boyd and Brightsmith, 2013).

Location information from the PTT tags was passed directly from the Argos satellite system to the Movebank system. We used two integrated tools in Movebank to improve the quality of our data from the PTT tags and our subsequent data products. First, we applied the Douglas Argos-filter (DAF) algorithm to our data, which removed low-quality locations from our dataset. The DAF filter identifies problematic locations and movement rates by examining distances, velocities, and turning angles within clusters of sequential locations. Application of the DAF filter may reduce the number of location estimates for subsequent analysis, but the overall accuracy of the remaining data is increased by 50–90% (Douglas et al., 2012). We opted to use the "best hybrid" method of the DAF algorithm in Movebank which was developed for filtering avian movement data. We used the default settings with the exception of two parameters that we adjusted following the recommendations of Douglas et al. (2012). We set the maximum sustainable movement rate over several hours (MINRATE) to 145 km h−<sup>1</sup> , and we set the maximum redundant setting (MAXREDUN) for filtering data to 5 km during the stationary periods of breeding and non-breeding periods, and to 15 km for the two migratory periods (D. Douglas, pers. comm.). The MAXREDUN setting retains near-consecutive locations within those distance thresholds to ensure independent error estimates. Setting MAXREDUN to 15 km results in large error estimates which are acceptable when analyzing continental scale movements (Douglas et al., 2012). Second, we used less optimistic error estimates for locations rather than those provided by CLS with our PTT data (D. Douglas, pers. comm.). For each Argos Doppler location class (LC) we assumed the following error radii: LC 3 = 0.46 km, LC 2 = 0.91 km, LC 1 = 1.81 km, LC 0 = 6.66 km, LC A = 1.59 km, and LC B = 1.95 km. Data assigned to the location class of LC Z or "invalid location" were not used in our analyses.

# Statistical Analysis

The GPS and PTT tags differed in the quantity and quality of movement data: GPS tags recorded a single location every 1–2 weeks with high accuracy, whereas the solar PTT tags generally recorded multiple locations per day with relatively lower accuracy. Thus, the movement data from the GPS and PTT tags required different analytical methods. For both types of tags, we examined patterns of habitat use by examining bird locations in relation to geographic features identified in aerial photographs from Google Earth. Timing and speed of migratory movements were difficult to determine from GPS tags due to low frequency of location fixes. Thus, we used the median of dates that bounded the onset of a seasonal or behavioral change such as the initiation of southbound migration. It was easier to determine the onset and duration of seasonal movements with daily fixes from the PTT tags. For birds with PTT tags, we calculated travel rate (km day−<sup>1</sup> ) as the total migration distance divided by the number of days that a bird was moving. Also, we measured ground speed (km h−<sup>1</sup> ) for long flight segments over open water when birds were likely to be in continuous flight. We were able to detect only long stopover events for birds with GPS tags because locations were recorded every 1–2 weeks. For birds with PTT tags, we defined migratory stopover events as local movements within an area of a 50-km radius that lasted more than 24 h. To calculate total migration distance, we first calculated a centroid for stopover events for birds with PTT tags to make our estimates comparable to birds with GPS tags. We then used the distance function in the "move" package for Program R to estimate the length of flight segments and migration distances for migratory paths during northbound and southbound migration (Kranstauber et al., 2018; R Core Team, 2018).

The daily locations from the PTT tags allowed for additional seasonal analyses of movements and space use of migratory Upland Sandpipers. We used dynamic Brownian bridge movement models (dBBMM) to analyze the detailed information available from the PTT tags (Kranstauber et al., 2012). The dBBMM model has at least two advantages: it controls for temporal autocorrelation among sequential locations that are collected a short time apart, and it tests for behavioral changes within a movement path associated with turning radius and length of movements. For example, star-shaped movements from a central point might indicate a roosting site, short movements with a high rate of turning might indicate residence at a staging site, and long unidirectional movements are expected with migration. The dBBMM model is especially appropriate for analyzing location data that are collected at frequent (<1 h apart) but irregular intervals, which was the case for our data from the PTT tags (Kranstauber et al., 2012; Walter and Fischer, 2018).

We filtered the movement tracks for Upland Sandpipers with the DAF algorithm in Movebank to screen implausible locations (Douglas et al., 2012), and then analyzed the movement data with the brownian.bridge.dyn function in the "move" package of program R. For each individual sandpiper, we analyzed the movement data separately for the four stages of the annual cycle: breeding in North America, southbound movements during autumn migration, non-breeding in South America, and northbound movements during spring migration. Long gaps between consecutive locations can result in large movement variances and create problems for model convergence with dBBMMs. Thus, we excluded gaps of >3 days from the variance calculation using the burst function of the "move" package (B. Kranstauber; pers. comm.). We used a margin of 11 and a window size of 25 locations to obtain stable estimates of the Brownian motion variance. During the breeding and nonbreeding periods, we used dBBMMs to calculate 50% (core areas) using small grid cells of 0.001 km<sup>2</sup> , and calculated area of the utilization distributions using ArcGIS (ESRI, 2018). For the migratory periods, we calculated 99% utilization distributions to characterize movement paths with larger grid cells of 0.01 km<sup>2</sup> . The dBBMMs explicitly incorporate location uncertainty into the estimation of the utilization distributions, which is a better approach than treating the location estimates as if they were recorded without error (Kranstauber et al., 2012).

# RESULTS

# Performance of Tracking Tags

We deployed tracking tags on 15 Upland Sandpipers in Spring 2016 and recovered movement data from 5 of 11 GPS tags (45%) and all four solar PTT tags (100%, **Table 1**). Four of 5 successful GPS tags (80%) worked as programmed and recorded movement data for a 10-month period from 15 June 2016 until 15 April 2017. For unknown reasons, one GPS tag only recorded movement data for a 6.2-month period that started on the nonbreeding grounds on 1 Oct 2016 and continued until 8 April 2017 (WO-GPS-62). Six of 11 GPS tags (55%) provided no movement data. Two of 4 PTT tags provided a full year of movement data (KO-PTT-66 and JB-PTT-67). The other two tags (50%) provided data for 4.6–6.8 mos, and then stopped transmitting data during southbound migration in 2016 (KO-PTT-69 and WO-PTT-68). The PTT tags required constant recharging of the battery from the solar panels and it was not uncommon for the tags to suspend reporting, but then restart again after a hiatus of 2–3 days. Overall, our analyses of seasonal variation in migratory movements and space use were based on nine Upland Sandpipers, where the GPS tags provided 18–25 locations over 190–305 days of monitoring and the PTT tags recorded locations on 71 to 196 unique days during 140–373 days of monitoring (**Table 1**).

#### Mortality and Annual Survival

We documented one known mortality event among our 15 tagged birds. A sandpiper that was marked with a GPS tag on 6 May 2016 at Westover ARB was recovered dead a year later on 17 May 2017 on an airport runway and 0.89 km from the original capture location (WO-GPS-62). While cause of death was unknown, a collision with an aircraft is plausible, as the bird had successfully carried the tag and harness for an entire annual cycle. The GPS tag was recovered from the carcass but showed no signs of damage. The GPS tag apparently malfunctioned because the movement track was incomplete: the first locations were recorded during the non-breeding period when the bird was in Mato Grosso province in Brazil, but the last locations were recorded during northbound migration from Vichada province in Colombia. The GPS tag failed to record movements between North and South America during either migration period.

We were unable to determine the fate of tagged birds if the GPS tags failed to upload data as scheduled or if the PTT tags stopped transmitting in <12 months. In both situations, missing data could have been due to harness failure, damage to the tracking tag or the external antenna, tag malfunction, or death of the bird. Six of the 11 GPS tags provided no data, but we had expected some attrition since the tags were programmed to upload data 1 year after deployment on 15 April 2017. Two birds with PTT tags disappeared during southbound migration before or after long distance water crossings. One bird from Kansas was last recorded near Victoria, Texas on 30 Nov 2016 close to the Gulf of Mexico (KO-PTT-69), whereas a second bird from Massachusetts was last recorded south of Calabozo, Venezuela on 21 Sept 2016 after a successful flight across the Caribbean Sea (WO-PTT-68). If we assumed that all losses were due to mortality, the minimum annual survival rates were 36.4% for GPS tags (4 of 11) and 50% for PTT tags (50%, 2 of 4). A pooled survival rate of 40.0% (6 of 15) for birds with GPS and PTT tags in this study was not significantly different from annual return rates of Upland Sandpipers marked with VHF radio tags (20.9%, 18 of 86, Fisher's Exact test: P = 0.18) or with color bands only (38.1%, 72 of 189, P = 1) from our long-term population study at Konza Prairie Biological Station in Kansas (Mong and Sandercock, 2007).

# Breeding Season

We obtained information on breeding season movements for 8 of 9 sandpipers because one GPS tag did not start recording locations until the bird had reached the non-breeding grounds (WO-GPS-62). One sandpiper captured at Fort Riley, Kansas on 27 April 2016 turned out to be a migratory transient (SD-GPS-90). The first locations from the GPS tag on 15 June and 17 July indicated that this bird had continued moving north after it was tagged, and spent the breeding season near Hosmer, South Dakota. The other 7 of 8 sandpipers remained on breeding home ranges near their capture sites in Kansas and Massachusetts from late April until late summer. The minimum duration of the breeding season from first capture until southbound migration averaged 81 days (range = 57–121 days) or 22% of the annual cycle (range = 16–33%, n = 8 birds). Patterns of space use during the breeding season were estimated for the four birds with PTT tags (**Table 1**). Space use and home range size differed between birds in open tallgrass prairie in Kansas and the grassland remnants in Massachusetts (**Figure 2**). Two sandpipers in Kansas had multiple activity centers and large home ranges with a 50% core area of 49 and 64 km<sup>2</sup> (KO-PTT-66 and KO-PTT-69). In contrast, the two birds in Massachusetts had a single activity

panel indicates its location on the inset map, and all panels are shown at the same spatial scale.

center and smaller home ranges with a 50% core area of 22 and 23 km<sup>2</sup> (JB-PTT-67 and WO-PTT-68).

#### Southbound Migration

We obtained complete routes for southbound migration for six sandpipers: four birds with GPS tags and two birds with PTT tags (**Figure 3**). The three birds from Kansas and the lone sandpiper that spent the breeding season in South Dakota, followed a narrow inland corridor in the Great Plains to the Texas coast, and then made long distance flights across the Gulf of Mexico and Central America to staging sites in the Andes of Colombia and Ecuador. The two birds from Massachusetts made shorter regional movements along the Atlantic coast and then made long flights over the Caribbean Sea to inland sites in northeast Venezuela. Sandpipers followed one of two migration routes upon arrival in South America: three birds crossed the Amazon basin to reach non-breeding sites in central Brazil, and three birds continued down the eastern slopes of the Andes with stopover sites in Bolivia, Brazil (Mato Grosso do Sul), and Paraguay en route to non-breeding sites in Uruguay and northern Argentina.

Upland Sandpipers traveled long distances during southbound migration and the total length of routes averaged 8,793 km (range = 5,410–10,675 km, n = 6 birds, **Table 2**). Individual variation included a 2-fold difference in distance

and a 3-fold difference in duration. The shortest migration was a sandpiper from Massachusetts that wintered in northern Pará province, Brazil, and this bird traveled 5,410 km over 36 days (JB-PTT-67). One of the longest southbound migration routes recorded was a sandpiper tagged in Kansas that traveled 10,040 km over 123 days to a non-breeding site in northwest Uruguay (mean stopover duration = 8 days, range = 1 to 38 days, n = 13 stopover events; KO-PTT-66). The length of migration segments between consecutive stopover events averaged 1,057 km (range = 20–3,758 km, n = 20 segments). The longest non-stop flights occurred over water for two sandpipers from the breeding population in Massachusetts. One bird completed a long-distance flight from Cape Cod to a site west of El Tigre, Venezuela that lasted up to 5 days and included a 3,758 km non-stop flight (JB-PTT-67; **Figure 4**). A second bird flew 3,432 km from Baltimore, Maryland to a site north of Calabozo, Venezuela in the Llanos grasslands (WO-PTT-68). We also recorded two instances of reverse migration with relatively short flights back to the north (138 km and 290 km) that occurred during southbound migration of two birds through Oklahoma (KO-PTT-66 and KO-PTT-69). Both cases occurred at midday and at approximately the same time in early August.

We recorded the onset of migration at the end of the breeding season for seven sandpipers, and all birds initiated migration in the 2-month period between 1 July and 1 September. The total duration of southbound migration averaged 99 days (range = 36–146 days) or 26% of the annual cycle (range = 10–40%, n = 6 birds). Migration flights were usually initiated around dusk with ground speeds that ranged between 33 and 61 km h −1 , including an average flight speed of 40 km h−<sup>1</sup> during


TABLE 2 | Summary of migration distances throughout the annual cycle for six Upland Sandpipers from 24 April 2016 to 1 May 2017, including the southbound and northbound migration distances, and the difference (%) between the two migration routes.

*The direct distance was calculated as the orthrodrome between the northernmost and southernmost locations in a sandpiper's movement path. Three birds were on northbound migration at their last detection, and we calculated northbound and total annual migration distances (values in parentheses) by assuming these birds returned to the same breeding site via the most direct path of travel. Distance to breeding represents the orthrodrome between a bird's last received location fix and its breeding location in the previous year.*

FIGURE 4 | Two examples of migratory segments illustrating non-stop flights over long distances for multiple days by Upland Sandpipers with PTT tags, 2016–2017. Extreme migratory movements included a southbound flight of 3,785 km in a 6-day period over the Atlantic Ocean from Massachusetts to Venezuela (Left, JB-PTT-67), and a northbound flight of 6,166 km in a 7-day period across the Andes and along the Pacific coast from Argentina to Honduras (Right, KO-PTT-66). Colored lines connect consecutive locations and highlight the longest suspected non-stop segments of the migratory flights for the two birds. Annotated values for a subset of fixes in the figure include the date, time (in UTC), and the cumulative distance traveled (km). The exact times of departure and arrival were not known precisely, but we provide timestamps for fixes immediately preceding and following the non-stop flight segments. Note the difference in spatial scales between the two panels.

3,476 kilometers of overwater flight (n = 4 overwater migration segments). The average rate of travel was 556 km per day for the six sandpipers where tracking tags recorded their entire southbound migration route. Stopover events averaged 12 days (range = 1–54 days, n = 15 stopover events) among birds with PTT tags. The longest staging event in North America was for a sandpiper that flew 484 km from Westover, Massachusetts to Baltimore Washington Airport, Maryland, where it spent 54 days (20 July to 12 September) before continuing on a 4-day nonstop flight of 3,441 km to Venezuela (WO-PTT-68). Long staging visits after arriving in South America were also common. One sandpiper from Kansas staged for 41 days (6 September to 16 October) near Cali, Colombia before traveling >4,500 km over a 21-day period to a non-breeding site in Uruguay (KO-PTT-66). Similarly, a bird from Massachusetts staged for 28 days (27 August to 23 September) in northern Venezuela before moving another 1,614 km over a 5-day period to a site in northern Brazil (JB-PTT-67).

#### Non-breeding Season

We collected movement data from seven sandpipers at their non-breeding sites in South America. The three general areas where non-breeding birds were located included: northern Brazil, central Brazil, and Uruguay/Argentina. Of four sandpipers that wintered in Brazil, one bird overwintered on islands within the Amazon River at the mouth of the Tapajós River in the northern state of Pará (JB-PTT-67), and three birds overwintered further south in the Cerrado ecoregion at the southeastern edge of the Amazon Basin in the states of Mato Grosso and Bahia (FR-GPS-82, KO-PTT-69, WO-GPS-62). The remaining three birds overwintered in the Pampas ecoregion of Uruguay and a 3-province region of northern Argentina (Buenos Aires, La Pampa, and Córdoba, SD-GPS-90, KO-PTT-66), including one bird that also used overwintering sites in southern Brazil (KO-GPS-94). Two sandpipers that were captured at sites only 27 km apart in Kansas overwintered ca. 2,600 km apart from each other in central Brazil (KO-GPS-82) and Uruguay (KO-PTT-66, **Figure 3**). On the other hand, one sandpiper from Kansas overwintered at a site in central Brazil (KO-GPS-02) that was <100 km from a sandpiper from the Massachusetts breeding population (WO-GPS-62).

Migratory sandpipers completed southbound migration in the 2-month period between late September and late November, and then spent an average of 147 days (range = 116–167 days) or 39% (range = 32–46%, n = 7 birds) of their annual cycle at the non-breeding grounds. Upland Sandpipers were not stationary during the winter, and all seven birds used multiple discrete areas on the non-breeding grounds. In one case, a sandpiper tagged with a GPS tag in Kansas spent 35 days in Mato Grosso do Sul, Brazil (8 October to 15 November), and then moved 1,459 km to spend another 105 days at sites southwest of Rosario, Argentina (1 December to 15 March, KO-GPS-94). Local movements were observed among birds with GPS tags wintering in central Brazil. One bird from Kansas used four activity centers that were 20– 100 km apart over 169 days (15 October to 1 April, FR-GPS-82), and a bird from Massachusetts moved among five activity centers that were 30 to 400 km apart over a 152-day period (15 October to 15 March, WO-GPS-62). We were able to examine space use during the non-breeding season for two sandpipers tracked with PTT tags. A bird from Massachusetts that wintered in the Amazon had three activity centers with a 50% area of 47 km<sup>2</sup> (JB-PTT-67), and a bird from Kansas that overwintered in the Rio de la Plata Basin of Uruguay and Argentina had multiple activity centers with a 50% utilization distribution of 118 km<sup>2</sup> (KO-PTT-66, **Figure 5**). Non-breeding birds used a variety of open lands that included both natural and agricultural habitats. Wintering sites in northern Brazil were islands in the Amazon River with short vegetation affected by seasonal flooding. Stopover and nonbreeding sites in Venezuela and Brazil were open cropfields where forests had been cleared for agriculture. Non-breeding sites in Uruguay and Argentina were mainly open grassland habitats in native rangelands used for livestock grazing.

#### Northbound Migration

We obtained routes for northbound migration for six Upland Sandpipers, including four birds with GPS tags and two birds with PTT tags. The total length of estimated northbound migration routes averaged 8,906 km (range = 6,833–10,229 km, n = 6 birds, **Table 2**). All six birds had elliptical migration routes in South America and in five cases, the paths for northbound movements were west of their southbound migration routes (**Figure 3**). The westerly routes used during northbound migration were similar in length to the easterly routes used during southbound migration (mean percent difference = +3.1%, range = −4.3 to +23.2%, n = 6 birds). Three birds from Kansas that wintered in the Pampas ecoregion of Uruguay and Argentina moved northwest and crossed the Andes to the southern edge of the Atacama desert of Chile before continuing north over the Pacific Ocean on the west coast of South America (KO-GPS-94, SD-GPS-90, KO-PTT-66). Another bird from Kansas that wintered in Mato Grosso, Brazil flew west and crossed the Andes of central Peru and then turned north after reaching the Pacific coast (KO-GPS-02). All four birds followed the Pacific Coast and made continuous non-stop flights until making landfall in Central America or the southern Great Plains. One bird from Kansas that wintered in Bahia, Brazil had an elliptical migration pattern where the northbound migration route followed a more easterly path to stopover sites in Venezuela and Mexico (FR-GPS-82). Last, a bird from Massachusetts that wintered in Brazil retraced her migratory path to Venezuela, but then used a westerly route where she flew 2,591 km over a 4-day period (9 to 12 April) and stopped over in Cuba for 9 days, then flew 1,686 km and stopped over at an airfield near Blackstone, Virginia for 4 days (24 to 27 April), and then completed the final 791 km back to Cape Cod in 1 day (JB-PTT-67).

Long-distance movements of >5,000 km were a common feature of northbound migration for Upland Sandpipers (**Figures 3**, **4**). The longest recorded flight was for a bird from Kansas that departed the non-breeding grounds in eastern Argentina on 24 March, was detected in flight at night over Chile on 27 March, again off the coast of Ecuador on the morning of 29 March, and continued northward until it reached Honduras on 30 March; a non-stop flight of 7,581 km in 7 days (KO-PTT-66). The flight segment over northern Chile included a 5-h overnight

flight that passed 125 km southwest of Ojos del Salado where the mountainous terrain had an average elevation of 3,703 m (range = 2,070–4,792 m, n = 8 locations). Two other sandpipers from Kansas made long-distance movements during the same time period: one bird traveled 5,952 km from Argentina to El Salvador in the week of 19 March to 1 April (KO-GPS-94), and a second bird traveled 4,475 km from the coast of Peru to Texas in the week of 23 March to 1 April (KO-GPS-02).

the panels indicates their location on the inset map. Note the difference in scales between the two panels.

Upland Sandpipers started migrating northward over a 2 month period between 3 February and 4 April (mean = 17 March, n = 7 birds). The routes used during northbound and southbound migration were similar in length, but birds migrating north completed their movements in half the time, with an average duration of 47 days (range = 23–84 days) or 13% of the annual cycle (6–23%, n = 3 birds). Flight speeds for overwater flight segments averaged 62 km h−<sup>1</sup> (range = 60 to 64 km h−<sup>1</sup> over 875 km, n = 2 segments). Tagged sandpipers traveled 683 km per day with an average non-stop flight segment of 1,615 km (range = 84–7,581 km, n = 11 segments). Stopover events during northbound migration were similar to southbound migration and typically lasted 11 days (range = 2–28 days, n = 9 events).

### Annual Movements and Breeding Site Fidelity

We were able to evaluate the complete annual cycle for six sandpipers with tracking tags (**Table 2**). The four stages of the annual cycle differed in duration: northbound migration was shorter than southbound migration, and the non-breeding season was longer than the breeding season (**Figure 6**). Two birds from Kansas (KO-GPS-02, KO-PTT-66) and one bird from Massachusetts (JP-PTT-67) successfully completed migration and returned to the breeding grounds. All three birds showed strong breeding site fidelity and the locations of the tagged birds after 305–373 days of monitoring were <7 km from their previous breeding locations in 2016. A fourth bird with a malfunctioning GPS tag had an incomplete track but this individual also showed site fidelity because it was recovered dead 0.89 km from the original banding site (WO-GPS-62). The remaining three birds with GPS tags were still migrating north when their movement data was uploaded on 15 April; the last known locations on northbound migration were 1,519– 2,944 km south of the breeding grounds near San Vincent, El Salvador (KO-GPS-94), Ciudad Madero, Mexico (FR-GPS-82), and Dallas, Texas (SD-GPS-90). We assumed that these three birds returned to the same breeding site for calculation of the round-trip distances traveled during the annual cycle. Total migration distance averaged 17,700 km (n = 6 birds), including a round trip of 20,904 km for a sandpiper that bred in South Dakota, a median of 18,526 km for four birds from Kansas (range = 16,064–19,984 km), and 12,467 km for a bird from Massachusetts (**Table 2**).

#### DISCUSSION

In our field study, we collected the first complete migratory tracks of individual Upland Sandpipers during their entire annual cycle in the Western Hemisphere. The scope of our project was relatively limited, as a 1-year study with movement data from nine birds captured in two separate breeding populations. Nevertheless, our data provide new insights into the migratory strategies, routes and sites, and the phenological timing of Upland Sandpipers. Our project resulted in three new discoveries. First, Upland Sandpipers were extreme migrants that can travel long distances >20,000 km during their annual cycle. Individual birds used long non-stop flights that were >5,000 km and lasted 5–7 days to cross major ecological barriers. Second, birds from two disparate breeding populations wintered across large areas of South America. We confirmed use of known stopover and non-breeding sites in Uruguay and Argentina, but we also identified unexpected staging sites in Colombia and non-breeding sites in two different areas of Brazil. Last, the migration patterns included several phenomena that have been reported in other migratory birds but were not previously known for Upland Sandpipers, including staging events that lasted up to a month, frequent movements during the non-breeding season, elliptical migration within South America with different northbound vs. southbound routes, and strong fidelity to breeding sites.

## Tag Performance and Effects of Tracking Tags

Our project joins previous work in demonstrating that miniature 1-to 5-g GPS and PTT tags can be successfully used to track relatively small-bodied birds throughout their annual cycle (McKinnon and Love, 2018), including Purple Martins (52 g, Progne subis, Fraser et al., 2017), Common Nighthawks (>70 g, Chordeiles minor; Ng et al., 2018), Common Cuckoos (102 g, Cuculus canorus, Vega et al., 2016), and Upland Sandpipers (170 g, this study). The solar-powered Argos PTT tags were more expensive but provided higher resolution movement data with fewer tag malfunctions than the PinPoint GPS Argos tags. Successful completion of long distance migration and carrying tags for multiple years suggests that our tags and harness design had relatively little effect on the movements or demographic performance of Upland Sandpipers (Mong and Sandercock, 2007; this study). Recent analyses have shown that tracking tags attached with elastic leg harnesses usually have little effect on behavior or reproductive output, but may have weak effects on the annual survival rates of small-bodied birds (Weiser et al., 2016; Smith et al., 2017; Brlík et al., 2019). Long-distance flights seem to be a risky part of migration because two sandpipers with PTT tags disappeared around a transoceanic flight, similar to mortality patterns of Whimbrel (Numenius phaeopus) during flights across the Caribbean basin (Watts et al., 2019) and Black-tailed Godwits (Limosa limosa) crossing the Sahara Desert (Loonstra et al., 2019). Despite these losses, annual return rates of sandpipers tagged with GPS and PTT transmitters were comparable to birds marked with color bands only (Mong and Sandercock, 2007). Further tests of the potential impacts of tracking tags awaits future developments in tag miniaturization and improved attachment methods (Wikelski et al., 2007).

#### Extreme Migration

Upland Sandpipers were known to be long distance migrants based on the wide separation of their breeding and nonbreeding ranges in temperate grasslands in the Northern and Southern Hemisphere (Blanco and López-Lanús, 2008; Houston et al., 2011). With our field study, we provide the first data on how individual Upland Sandpipers complete their long migratory movements between temperate grasslands on different continents. Open water, forest habitats, and mountain ranges were not ecological barriers for migrating sandpipers because individual birds made non-stop flights across the Caribbean Sea, Amazon basin, and Andes mountains, and also moved long distances along the Pacific coast.

Our data revealed that Upland Sandpipers were capable of long non-stop flights of up to 5 days and 3,758 km during southbound migration, and up to 7 days and 7,581 km during northbound migration. Individual birds traveled long distances during their annual movements with total migration distances ranging from 12,467 to 20,904 km. While these migratory movements are remarkable, a growing body of evidence suggests that extreme flights are relatively common among migratory shorebirds. Conklin et al. (2017) reviewed evidence for longjump movements among migratory birds and compiled data

showing that 19 other species of shorebirds are capable of making extreme non-stop flights >5,000 km in length and round trip migrations of >20,000 km. The most extreme migrants tend to be large-bodied shorebirds that breed in the arctic or subarctic regions but make long oceanic flights of >10,000 km to reach non-breeding sites in the Southern Hemisphere for total migration distances >30,000 km, including Red Knots (Tomkovich et al., 2013), Bar-tailed Godwits (Battley et al., 2012), and Hudsonian Godwits (Senner et al., 2014).

In this study, we tracked sandpipers that were captured in two breeding populations in their core range in the continental USA. However, a separate disjunct population of Upland Sandpipers breeds in Alaska and the Yukon (Buss, 1951; Houston et al., 2011), and as far north as Ivvavik National Park (69.2◦N, Miller et al., 2015), which is roughly 4,400 km north of our field sites in Kansas. The migratory ecology of the boreal populations of Upland Sandpipers remains unknown, but has the potential to be among the longest routes used by migratory shorebirds.

#### Migratory Routes

Movement tracks of migrating Upland Sandpipers confirmed use of sites that have been identified as important, but also led to discovery of some previously unknown staging and nonbreeding sites. Birds that bred in Kansas and South Dakota used a relatively narrow corridor in the Great Plains during southbound migration (Houston et al., 2011), and birds from Massachusetts used grassland habitats at airfields along the Atlantic coast for both breeding and staging (Garber et al., 1997). Newly discovered staging sites used during southbound migration that were not previously known included canefields in mountain valleys of Colombia and the Llanos grasslands of central Venezuela. One of our tagged birds (SD-GPS-90) moved through the high elevation sites near Ozogoche Lagoon in Ecuador where mass mortality events have been reported for this species (Vickery et al., 2010). Grosselet et al. (2019) recently reported a similar path for an Upland Sandpiper tagged in Mexico which also crossed the Andes in northern Ecuador, and continued south along the eastern side of the Peruvian Andes until the signal was eventually lost. Three of the birds that we tracked wintered at sites in the expected non-breeding distribution in the Pampas ecoregion of Uruguay and Argentina (Blanco and López-Lanús, 2008; Alfaro et al., 2018). However, our tracking data showed that four other Upland Sandpipers wintered in two different areas of Brazil, far north of the main non-breeding range, including grassland habitats in the Cerrado ecoregion, and more unexpectedly, river islands in the Amazon basin. Small numbers of Upland Sandpipers were thought to spend the nonbreeding season in northern South America (Haverschmidt, 1966; Houston et al., 1999; Blanco and López-Lanús, 2008), and our tracking data have confirmed this prediction. Finally during northbound migration, three tagged birds crossed the Andes through northern Chile, and near sites where Upland Sandpipers have recently been reported as nocturnal migrants (Medrano et al., 2018).

Migratory routes of Upland Sandpipers generally followed a northwest-southeast axis with birds from Kansas using more westerly routes than birds from Massachusetts. Migration along a north-south axis is common among migratory shorebirds in the Western Hemisphere, including both short (Page et al., 2014) and long-distance migrants (Johnson et al., 2016; Brown et al., 2017). A majority of Upland Sandpipers also had elliptical or loop migration with a clockwise pattern within South America where northbound routes were more westerly than southbound routes. For example, the individual bird with the greatest difference in routes had a direct flight over the Caribbean Sea during southbound migration but then used a westerly route and stopped over in Cuba during northbound migration (JB-PTT-67). Elliptical migration has been previously reported for other shorebirds that use eastern flyways to travel to South America (Myers et al., 1990; Senner et al., 2014; Brown et al., 2017; Johnson et al., 2018). The movement pattern may be related to predictable seasonal dynamics of the atmospheric conditions over the Gulf of Mexico and Caribbean Sea, with migratory birds taking advantage of favorable tailwinds (La Sorte et al., 2014; Bayly et al., 2018).

Many shorebirds that migrate long distances show a high degree of structure in their migratory routes, with a majority of a population using key staging sites at inland wetlands along continental flyways (Myers et al., 1987; Senner et al., 2014), coastal estuaries such as Chesapeake and Delaware Bays on the Atlantic Coast (Baker et al., 2004; Johnson et al., 2016), or the Yellow Sea region of eastern China (Battley et al., 2012; Studds et al., 2017). In contrast, the diversity of migration tracks among our tagged birds suggests a pattern of weak migratory connectivity. Birds from two breeding populations were broad front migrants without shared staging sites, and individual movement tracks covered a large area of northern South America. Our analysis was based on a relatively small sample of birds and it is possible that adding more tracks and additional populations would allow identification of migratory network nodes for different breeding and non-breeding populations (Knight et al., 2018), and quantitative analyses of the patterns of migratory connectivity (Cohen et al., 2018). Overall, the migratory patterns of Upland Sandpipers appear to be more similar to migratory landbirds, where migratory connectivity is often fairly weak (Renfrew et al., 2013; Finch et al., 2017; Hill and Renfrew, 2019a).

Reliance on a relatively small number of staging sites increases population vulnerability for migratory shorebirds, but it offers opportunities to target conservation actions. Alternatively, weak migratory connectivity may buffer local breeding populations against loss or degradation of habitat elsewhere in their migratory range. Low densities over a wide distribution make it more difficult to implement conservation measures at specific sites (Vickery et al., 2010; Pearce-Higgins et al., 2017), with two possible exceptions. First, our field sites in Kansas are part of the Flint Hills ecoregion, which has been designated as a Landscape of Hemispheric Importance under the Western Hemisphere Shorebird Reserve Network (WHSRN) based on its importance to Buff-Breasted Sandpipers, Upland Sandpipers, and American Golden-Plovers (Penner et al., 2015). Second, space use and movement tracks indicated that airfields provide critical habitat for breeding and staging sites for migratory Upland Sandpipers (Osborne and Peterson, 1984; Garber et al., 1997; this study). Thus, conservation of eastern populations of Upland Sandpipers would benefit from protection of appropriate habitat within airfields and other remaining patches of native grasslands along the Atlantic coast.

# Time-Budgets During the Annual Cycle

Northbound migration (13%) was a shorter period than southbound migration (26%), and Upland Sandpipers spent less time at the breeding (22%) than the non-breeding grounds (39%). Thus, the duration of northbound migration was relatively short because the average ground speed and distances traveled were greater for Upland Sandpipers during northbound (ca. 62 km h −1 and 683 km per day) compared to southbound migration (40 km h−<sup>1</sup> and 556 km per day). Our estimates of movement rates were comparable to non-stop oceanic flights on southbound migration for an Upland Sandpiper tagged in Mexico (40 km h−<sup>1</sup> , Grosselet et al., 2019). Faster northbound migration and similar ground speeds have also been reported in Pacific Golden-Plovers Pluvialis fulva (northbound vs. southbound: 63 and 58 km h−<sup>1</sup> ; Johnson et al., 2011), two subspecies of Bar-tailed Godwits (L.l. baueri: 59–63 and 53 km h−<sup>1</sup> , L.l. menzbieri: 60–76 and 53– 58 km h−<sup>1</sup> , Battley et al., 2012), Ruddy Turnstones (48–79 and 30–40 km h−<sup>1</sup> , Minton et al., 2011), and Great Knots (24–92 and 13–74 km h −1 ; Lisovski et al., 2016). Seasonal differences in migration speed may be related to reproductive advantages of early arrival at the breeding grounds (Weiser et al., 2018; Morrison et al., 2019), or to predation risk during southbound migration (Ydenberg and Hope, 2019). The potential carry-over effects from linkages of events at different stages of the annual cycle have been studied in some shorebirds (Barshep et al., 2011; Senner et al., 2014; Carneiro et al., 2019), but await further investigation in Upland Sandpipers.

One of the main advantages of GPS and PTT tags is that they allow constant monitoring of individuals in space and time, and can record forays that cannot be detected with ground-based telemetry systems (McKinnon and Love, 2018). Our previous estimates of home range size for birds breeding in Kansas were based on VHF radio tags (8.4 km<sup>2</sup> , Sandercock et al., 2015) and our new estimates based on improved tracking technologies were about seven times larger (49–64 km<sup>2</sup> , this study). The main difference between estimates was due to our discovery that sandpipers in Kansas had multiple activity centers during the breeding season that were ca. 40–60 km apart. Multiple activity centers might have been due to renesting after clutch failure, foraging to prepare for migration, or flocking with other birds that had completed nesting. Estimates of home range size were smaller for the two birds breeding in Massachusetts, presumably because other available habitat was extremely limited in the surrounding landscapes. Large space requirements help to explain why Upland Sandpipers and other grassland birds are area-sensitive species that are less likely to occur in small grassland fragments (Vickery et al., 1994).

The non-breeding season is sometimes described as a stationary period, and migratory shorebirds that use coastal wetlands are often sedentary during the non-breeding season (Battley et al., 2012; Senner et al., 2014). In contrast, our tracking data revealed that within-season movements were common during the non-breeding season for Upland Sandpipers. Birds did not settle in a single location, but rather moved among consecutive activity centers that were 20–400 km apart before eventually departing on northbound migration. Movement among separate activity areas during the non-breeding season has also been reported for Buff-breasted Sandpipers and Bobolinks Dolichonyx oryzivorus wintering in grassland habitats in South America (Renfrew et al., 2013; Lanctot et al., 2016), Red-necked Phalaropes (Phalaropus lobatus) wintering in the Arabian Sea (van Bemmelen et al., 2019), and a diversity of tropical songbirds (Stutchbury et al., 2016; Bayly et al., 2018; McKinnon and Love, 2018). Migratory birds may be mobile during the non-breeding season because they are tracking ephemeral food resources (Jahn et al., 2010), which for Upland Sandpipers would primarily be grasshoppers and other arthropods (Alfaro et al., 2015).

Our new tracking data suggest that Upland Sandpipers are a highly vagile species because individual birds had multiple activity centers during both the breeding and non-breeding seasons, and the diversity of migratory tracks suggests that they are broad front migrants. Despite this suite of traits, individual sandpipers also demonstrated remarkable homing skills with strong fidelity to breeding sites. Four birds returned to breeding sites that were <6.5 km from their locations in the previous year, despite traveling up to 20,000 km during their annual migration. Upland Sandpipers nest in loose colonies among birds that are genetically related, and both females and males share incubation duties (Casey et al., 2011). Thus, strong breeding fidelity may enhance reproductive success for a long-distance migrant if an experienced sandpiper is able to breed near relatives, repair quickly with a former partner, or nest at a familiar site where they were successful in a previous year.

#### Future Research

Our project demonstrates that new tracking technologies can provide unexpected insights into the migratory ecology of small-bodied birds, and opens new lines of enquiry for future research. The migratory tracks from a small number of Upland Sandpipers were highly variable and more work is needed to clarify the importance of the new non-breeding areas that we have discovered in Brazil. More tracking data for birds from other breeding populations are needed to better understand migratory connectivity and the population structure of the species within in the continental range vs. boreal populations in Alaska and the Yukon (Buss, 1951; Miller et al., 2015). Our analyses were based on adults only, and tracking of juveniles is needed to understand the ontogeny of migration (Rotics et al., 2016; Vega et al., 2016). Moreover, we tracked birds for a single year and multiple tracks from the same set of individuals are needed to investigate repeatability in migratory behavior (Carneiro et al., 2019; Ruthrauff et al., 2019). Understanding potential carryover effects will require better integration of movement data with additional information on local environmental conditions, habitat use, and the timing of feather molt, fat deposition, and other costly activities (Barshep et al., 2011; Senner et al., 2014; McKinnon and Love, 2018). Similarly, parameterization of full-annual-cycle models requires better methods for determining causes of mortality from different types of tag failure (Sergio et al., 2018; Loonstra et al., 2019; Watts et al., 2019). Upland Sandpipers were capable of extreme migratory movements across oceans and mountainous terrain, and new types of tracking tags with accelerometers and altimeters will provide information on their physiological capacity during sustained flights. Finally, the sandpipers demonstrated a remarkable ability to return to the same breeding sites despite traveling long distances within the Western Hemisphere, and the sensory systems and environmental cues used for navigation will be an important area for future work.

# DATA AVAILABILITY STATEMENT

The datasets for migratory movements of tagged birds have been archived at Movebank (www.movebank.org) as a project of the Vermont Center for Ecostudies: Upland Sandpiper Annual Life Cycle Ecology (Movebank ID: 176770813; Hill and Renfrew, 2019b).

# ETHICS STATEMENT

Our research protocols followed recommended procedures in the Guidelines to the Use of Wild Birds in Research of the Ornithological Council. We conducted our project under government permits for scientific research, including federal and state permits for live capture and tagging of wild birds, and access permits for working at facilities of the US Department of Defense and Konza Prairie Biological Station.

# AUTHOR CONTRIBUTIONS

RR conceived and designed the project. JH and RR directed and executed the project. JH coordinated data management with Movebank. JH and BS conducted statistical analyses of the movement tracks. BS led the writing of the manuscript. All authors contributed to the fieldwork, edits, and approved the final version.

#### FUNDING

Funding for this research project was provided by the Legacy Resource Management Program of the US Department of Defense (Project 15-764).

#### ACKNOWLEDGMENTS

We thank the biological field technicians and crew leaders for their hard work, and the biologists and staff of the Fire Desks, Air Traffic Control Towers, and Range Controls that supported our research and kept field workers safe while working at Fort Riley (KS), Konza Prairie (KS), Camp Grafton (ND), Fort McCoy (WI), Camp Ripley (MN), Joint Base Cape Cod (MA), Westover Air Reserve Base (MA), and Naval Air Station Patuxent River (MD). Thanks to Asociación Calidris and Jim Giocomo for their assistance in locating sandpipers, and Sarah

#### REFERENCES


Davidson at Movebank and David Douglas at USGS for project guidance and generous use of their time. Thank you to Daniel Blanco, Mark Balman, and BirdLife International permission to reproduce the Upland Sandpiper range maps. Two referees provided helpful comments that improved the quality and content of our manuscript. The ESRI Conservation GIS Program provided software.

during the annual cycle. Front. Ecol. Evol. 7:248. doi: 10.3389/fevo.2019. 00248


longicauda) to the western slope of the Andes. Wilson J. Ornithol. 130, 805–809. doi: 10.1676/17-075.1


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Hill, Sandercock and Renfrew. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Endogenous Programs and Flexibility in Bird Migration

Susanne Åkesson<sup>1</sup> \* † and Barbara Helm<sup>2</sup> \* †

<sup>1</sup> Center for Animal Movement Research, Department of Biology, Lund University, Lund, Sweden, <sup>2</sup> Groningen Institute for Evolutionary Life Sciences, University of Groningen, Groningen, Netherlands

Endogenous programs that regulate annual cycles have been shown for many taxa, including protists, arthropods, fish, mammals and birds. In migration biology, these programs are best known in songbirds. The majority of songbirds rely on a genetic program inherited from their parents that will guide them during their first solo-migration. The phenotypic components of the program are crucial for their individual fitness and survival, and include time components, direction, and distance. This program is constructed to both guide behavior and to regulate flexible responses to the environment at different stages of the annual cycle. The migration program is driven by a circannual rhythm, allowing for, and resetting, carry-over effects. With experience, the migration decisions of individual migrants may be based on information learnt on breeding sites, wintering sites, and en route. At the population level, substantial variation in route choice and timing of migration may be explained by inherited variation of program components, by interactions with environmental and social factors, and by individual learning. In this review we will explore the components of endogenous migration programs and discuss in what ways they can lead to flexibility and variation in migration behavior.

#### Edited by:

Nathan R. Senner, University of South Carolina, United States

#### Reviewed by:

Elizabeth A. Gow, University of Guelph, Canada Franz Bairlein, Institute of Avian Research, Germany

#### \*Correspondence:

Susanne Åkesson susanne.akesson@biol.lu.se Barbara Helm b.helm@rug.nl †These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 30 July 2019 Accepted: 10 March 2020 Published: 26 March 2020

#### Citation:

Åkesson S and Helm B (2020) Endogenous Programs and Flexibility in Bird Migration. Front. Ecol. Evol. 8:78. doi: 10.3389/fevo.2020.00078 Keywords: circannual rhythm, endogenous programs, geomagnetic field, migration, navigation, songbirds

# INTRODUCTION: MIGRATIONS OVER TIME AND SPACE

Migrations, which we here define as regular periodic movements, have been linked to fluctuations of relevant environmental factors, such as nutrients (Newton, 2008; Häfker et al., 2017), breeding site or mate availability (Barlow et al., 1986; Hodgson and Quinn, 2002; Friesen, 2015), predation risk (McKinnon et al., 2010; Häfker et al., 2017), infection probability (O'Connor et al., 2018), and abiotic survival threats (e.g., severe weather or UV radiation; Hut and Beersma, 2011; Reid et al., 2018). Such environmental fluctuations, and the associated periodic movements of organisms, usually recur on time scales defined by geophysical cycles: annual cycles caused by Earth's orbit around the sun, diel cycles caused by Earth's rotation around its axis, lunar cycles caused by the orbit of Moon around Earth, and tidal cycles caused by the combined movements of Earth and Moon (DeCoursey, 2004; **Figure 1A**). Their common, planetary basis makes these cycles predictable, even if their downstream environmental consequences (e.g., temperature or food availability) may be more variable. The principle predictability of environmental cycles has provided the substrate for organisms to evolve time-keeping mechanisms that are fitted to their life-styles and specific environments.

Organisms use biological time-keeping to predict upcoming changes, to prepare for them, and to spatially relocate in anticipation (Åkesson et al., 2017; Helm et al., 2017). In a nutshell,

these mechanisms integrate internally generated (i.e., endogenous) rhythms and responses to environmental cues into timing programs. In migratory organisms, spatial features, for example direction of travel or magnetic field cues, are often included, so that even naive individuals can perform effective migrations (Kramer, 1957; Åkesson et al., 2017; Reppert and de Roode, 2018).

These intriguing spatio-temporal programs have been extensively studied from a full-cycle perspective in migratory organisms on all four predictable time-scales introduced above (**Figure 1A**; annual, diel, lunar and tidal; e.g., Barlow et al., 1986; Gwinner, 1996a,b; Åkesson et al., 2017; Häfker et al., 2017). Central questions in this research field are concerned with inheritance of migration programs, their integration of environmental cues, and their physiological and genetic mechanisms (van Noordwijk et al., 2006; Åkesson et al., 2017; Merlin and Liedvogel, 2019). Here, we review key insights, and apply them to address sources of variation in spatiotemporal migratory traits within and among individuals, as well as among populations. Our review is aimed at explaining how endogenous programs can generate, or counteract, variation in migration. Among the countless contributions to the field, we highlight those that in our view are most suitable to achieving this aim, rather than attempting to give a balanced or exhaustive record of this vast research field.

The review actively contributes to a collection of articles that constitute the Frontiers research topic Flexibility in the Migration Strategies of Animals (Senner et al., 2020). Coherently with this Frontiers research topic, we will focus on annual time scales, choosing birds as study subjects, but the principle considerations also hold for other time scales and other periodically migrating organisms. We also streamline the use of terminology in our article as laid out for the entire Frontiers research topic. To facilitate cross-fertilization of ideas, we will first clarify the terminology and conceptual background of migration programs, which are not widely known among field-based researchers. We will then explain in greater detail how migration programs work, how they interact with environmental information, and how they regulate aspects of variation. Thereafter we will discuss additional variation and its possible interpretation. We address these points from theoretical perspectives and illustrate them by selected examples. Due to both our own geographic location and geographic differences in research focus on timing programs, our article is greatly biased toward the northern hemisphere, with emphasis on European data, although we strove to give global examples.

# TERMINOLOGY AND CONCEPTS

# Flexibility, Plasticity, Variation

The central topic of this Frontiers research topic is flexibility, defined on an individual level as variation in traits that can be reversed in response to an individual's environment throughout their lives. Such trait reversals can for example be based on experience, can occur in response to inter-annual differences in environmental conditions, or can be a generic part of the annual cycle (i.e., life-cycle staging, sensu Piersma and Drent, 2003). Variation through flexibility is distinguished from polyphenisms (sensu Piersma and Drent, 2003), where individuals differ from each other irreversibly because of genetic differences. It is also distinguished from developmental plasticity, which denotes variation in traits that is irreversibly determined during ontogeny (sensu Piersma and Drent, 2003). We here adopt this terminology for coherence across the Frontiers research topic, diverging from our use elsewhere (Helm et al., 2017).

Additionally, we will introduce a distinction between two types of flexibility. Because migration programs by definition regulate responses to the environment, we consider a substantial proportion of flexibility to be programmed (i.e., resulting from inherited reaction norms; van Noordwijk et al., 2006; Visser et al., 2010; Helm et al., 2017). A remaining part of residual flexibility is not readily explained by our current knowledge of migration programs. On a population level, we will refer to differences in migration as variation because their mechanistic basis is not clear. Differences in individual traits, such as timing or route choice, and in strategies, such as partial or differential migration, could be based on genetic differences, developmental plasticity, or individual flexibility.

# Ontogenetic Perspective

fevo-08-00078 March 24, 2020 Time: 16:1 # 3

Additional emphasis in this Frontiers research topic is on ontogenetic variation in migration. From a perspective of migration programs, the main distinction is between an individual's first journey and its subsequent migrations. During their first migrations, juveniles are naive and depend on their programs, on social guidance, or on trial and error. In subsequent years, birds will have additional experience (Perdeck, 1958), which may override the initially expressed migration program. A further distinction will be made between open-ended learners and those that adhere to their first experience for their future migrations (Gill et al., 2014).

## Carry-Over Effects

A full annual-cycle perspective is at the heart of research on migration programs (Gwinner, 1996b; Briedis et al., 2016), and hence, carry-over effects have long been studied within this field. In adherence to the lay-out for the present Frontiers research topic, we adopt a broad definition of carry-over effects to include all instances when previous history explains current performance, as long as they are functionally important and separated in time (O'Connor et al., 2014). Hence, learning and developmental adjustments mostly fall also under this definition.

From a perspective of timing programs, a distinction is made between carry-over effects within an annual cycle, and those between cycles. In many species, the timing program allows for high variation during some phases of the annual cycle, whereas during others individuals resynchronize to environmental cues and reset their annual timing (e.g., Helm et al., 2005; Conklin et al., 2013; Karagicheva et al., 2016; Briedis et al., 2018; Gow et al., 2019). Thus, we consider carry-over effects within an annual cycle as flexibility that can be regulated by the migration program. In contrast, we view carry-over effects between annual cycles as modifications of the migration program, for example due to experience or to poor state. In either case, we view carry-over effects as an outcome of trade-offs, or, alternatively phrased, of different optimization criteria (Alerstam and Lindström, 1990). Animals may compromise optimal timing for other benefits, such as improved state or additional broods, but may pay costs for suboptimal timing at subsequent annual-cycle stages (see also Senner et al., 2015). Alternatively, they can adhere to timing, at potential costs to state (i.e., departing in poor condition) or to reproduction (e.g., skipping a breeding opportunity to depart in time).

#### Migratoriness

Species differ greatly in the level of variation of their migrations (Newton, 2008). Generally, spatio-temporal precision and consistency increase with migration distance, with proportion of population members migrating, and with the rigidity of its regulation (Tryjanowski et al., 2005; Newton, 2008). This trend can be captured in the term migratoriness. Migration programs are most useful if birds move between sites that are spatially too distinct to assess environmental conditions of goal areas, and if conditions at the goal areas are sufficiently predictable to facilitate evolution of migration programs. If these conditions are met, birds tend to score high on migratoriness. An example for clear differences in migratory precision of related species with similar ecology are waders on Iceland, where relatively short-distance migrants (e.g., Black-tailed Godwits, Limosa limosa islandica) have substantially advanced arrival time over the last decades (Gill et al., 2014), whereas a long-distance migrant (Whimbrel, Numenius phaeopus islandicus) has not (Carneiro et al., 2019).

#### MIGRATION PROGRAMS FOR TIME AND SPACE

As new data from avian migrations flood in during this golden age of bio-logging (McKinnon and Love, 2018), efforts to distill patterns have invigorated interest in migration programs. Consistent timing is perhaps the most commonly emerging pattern (Altshuler et al., 2013; Briedis et al., 2018; McKinnon and Love, 2018; Carneiro et al., 2019; Gow et al., 2019). Routes are often more variable (Stanley et al., 2012; Vardanis et al., 2016), although in some studies they were more consistent than timing (Vardanis et al., 2011; Sugasawa and Higuchi, 2019). When timing is consistent within individuals, there may be large variation within and between populations, for example in departure date of sympatrically overwintering individuals (e.g., Conklin et al., 2010, 2013; Briedis et al., 2016; cf. Gow et al., 2019).

Much of the new evidence fits well with our current understanding of migration programs. Researchers had long postulated the existence of innate programs to explain why migratory birds do not simply stay at the wintering grounds, and how they return for breeding at the right time of year. Support for innate programs first came from observations by bird fanciers. When wild conspecifics would migrate, caged birds, provided with ample food and shelter, also performed migrationlike behaviors (Birkhead, 2008) and directional movements (e.g., Kramer, 1957). This behavior is called migratory restlessness, or Zugunruhe. It is most readily observed in nocturnally migrating species which show bouts of migratory restlessness at night, but similar arguments have also been made for some diurnally migrating species (e.g., Bojarinova and Babushkina, 2015). It is important to note that the restless hopping and flying in cages does not directly represent migration, but rather a captive expression of motivation to migrate (Helm, 2006; Van Doren et al., 2016; Bäckman et al., 2017). However, this frustrated movement state (John Rappole, pers. comm.) often captures important aspects of migration and has been key to our understanding of bird migration. Migratory restlessness has been reported for many northern hemisphere-breeding migrants that breed in Europe, America and Asia. Species included mainly passerines and some other taxa, for example quail (e.g., Helms, 1963; King and Farner, 1963; Gwinner, 1996a,b; Budki et al., 2009; Bertin et al., 2007; Eikenaar et al., 2014; Watts et al., 2016). For tropical and southern hemisphere breeders, only a few Zugunruhe records exist, including intra-tropically migrating yellow–green vireos, Vireo flavoviridis; Styrsky et al., 2004) in the Americas, for stonechats in Africa (Saxicola torquata axillaris, Helm and Gwinner, 2006, which however are locally resident), and for Australian silvereyes (Zosterops lateralis; Chan, 1995).

Although migratory restlessness is not always easy to interpret, its regulation through robustly innate programs was confirmed in migratory birds that were kept under constant conditions of daylength, temperature and food (Gwinner, 1986, 1996b; Holberton and Able, 1992). Over many years Zugunruhe alternated with molt and with reactivation of the reproductive system approximately annually (hence, called circannual). Thereby, it became clear that life-cycle stages including migratory behaviors were driven by an endogenous (i.e., selfgenerated) circannual rhythm. Furthermore, when tested for their directional preference, birds shifted the orientation of their Zugunruhe activities as appropriate for the corresponding leg of migration (Gwinner and Wiltschko, 1980). A host of additional physiological changes that enhance migration also occurred (e.g., hyperphagia, fuel deposition; King and Farner, 1963; Gwinner, 1996a; Newton, 2008). Migratory activity within this endogenous migration program is encoded in individual birds in relation to migration distance, with short-distance migrants generally expressing shorter periods of migratory restlessness than long-distance migrants (e.g., Berthold, 1973; Berthold and Querner, 1981; Maggini and Bairlein, 2010; Bulte and Bairlein, 2013). Because the endogenous migration program also encodes migration direction, it can lead migratory naïve individuals along routes to population-specific wintering areas (e.g., Helbig, 1991).

#### Time

#### Annual Timing

Under constant conditions, without any changes in the captive environment, circannual cycles recur, but their period lengths are ca. 9–15 months, so that life-cycle stages usually drift to occur at earlier or later dates over progressive years (Gwinner, 1996b; Karagicheva et al., 2016). In nature, conversely, annual cycles do not drift, and life-cycle stages recur annually, usually at similar dates. Hence, it became clear that the circannual clock functioned in interaction with environmental cues that synchronize and modify its timing. Therefore, a spate of experimental studies investigated the synchronizing effects of environmental factors on migration programs across the annual cycle (Gwinner, 1996b; Helm et al., 2009).

Among the synchronizing cues, photoperiod, the annually changing light fraction of the 24 h day (**Figure 1B**), has the strongest effects, and can both, advance or delay the annual cycle. For example, in multi-brooded species, chicks hatch at widely different times of year. The correspondingly different daylengths experienced in early life can then synchronize chicks from consecutive broods. The contribution of the timing program to synchronizing these birds is illustrated by data from captive stonechats (Saxicola spec.) from three regions (Europe, East Africa, Kazakhstan) that bred under naturally changing daylengths (Helm et al., 2005; **Figure 2**). Chicks that grew up under shorter photoperiods, simulating late hatching at the end of summer, compensated by accelerated postjuvenile development in population-specific ways. Late-hatched of all populations largely caught up with earlier-born conspecifics by advancing autumn Zugunruhe by 0.9 days per day of later hatching. By this genetically programmed compensation mechanism, which

was confirmed by field observations from wild conspecifics, stonechats achieved a high level of within-population synchrony (Helm et al., 2005) that counteracted carry-over effects of late hatching. Similar advancement of Zugunruhe for later-hatched chicks was also observed in birds breeding at low latitudes, for example in yellow–green vireos (Styrsky et al., 2004).

hatch date on the x-axis (from left to right). Based on Helm et al. (2005). Inlay:

stonechat during postjuvenile molt.

Synchronization within populations that counteracts carryover effects also occurs at the end of overwintering when migrants initiate spring migration (Conklin et al., 2013; Senner et al., 2014; Briedis et al., 2018). For example, recent work on socially migrating tree swallows (Tachycineta bicolor) suggests that differential timing of migration may continue as a dominoeffect set by breeding latitude until resynchronization in the winter quarters (Gow et al., 2019). Such resetting effects can be so strong that the non-breeding period has been described as buffering the build-up of carry-over effects (Senner et al., 2014; Briedis et al., 2018). However, for low-latitude wintering grounds, where photoperiod undergoes little change, it is still unclear how such synchronization is achieved, although effective cues have been described for breeding in equatorial birds (e.g., Goymann et al., 2012; Shaw, 2017). A key property of timing programs is that the responses to environmental factors, which they encode, are specific to stage (phase) of the annual cycle (**Figure 3**). It is intuitively sensible that a migratory bird would respond to a long, warm day differently on the breeding grounds than at its winter quarters. The same environmental cues can thereby cause either advances, delays, or no changes to the annual cycle, depending on time of year. This time-dependence can be shown systematically

FIGURE 3 | Responses to environmental factors depend on the phase of the annual cycle. Graphs show advance and delay responses of seasonal events (y-axis) to cues experienced at different times of year (x-axis). (A) Alternating responses to temperature and precipitation, depending on the phase of the annual cycle, has been described as the dominant pattern of wild organisms based on 10,000 time series from the UK (Thackeray et al., 2016). Shown is the phase-dependent temperature response of egglaying in wild Barn Swallows (image kindly generated by Dario Massimino). (B) Phase-dependent response of follicle growth to photoperiod in captive Garden Warblers (based on Gwinner, 1996a). (C) Schematic difference between migrants and residents in the phase-specific response to spring-like cues. In late winter and spring, long days and high temperatures advance subsequent life cycle stages, such as spring migration and reproductive activation. Conversely, later in the year the response reverses, and long days and high temperatures delay life-cycle stages. Long-distance migrants (inlay: Garden Warbler) differ from residents (inlay: Great Tit) by a rigid phase during wintering, when long days and high temperatures have little or no effect on the annual cycle; image credit to commons.wikimedia.org: Garden Warbler by Kristjan Osbek, Great Tit by Biillyboy.

by plotting timing responses over the annual cycle. Such phaseresponse curves (**Figure 3A**) or sensitivity profiles (**Figure 3B**) have been described by studies of biological rhythms (DeCoursey, 1960; Gwinner, 1996b; Helm et al., 2009) and phenology time series (Thackeray et al., 2016), respectively.

**Figure 3A** shows the response of female reproductive timing (follicle growth) to long days in captive Garden Warblers (Sylvia borin) (Gwinner, 1996b). This usually single-brooded species responds to long days in summer with a shut-down of the reproductive system. Subsequently, reproductive responsiveness is low over winter, but in spring garden warblers respond to long days with reproductive activation and advance their annual cycle. A downregulated response is important for migratory species which in their winter quarters experience conditions that may induce breeding (Hamner and Stocking, 1970; Gwinner, 1996b; Helm et al., 2009). To be sure to return to breeding sites in spring, rather than breed in the winter quarters, it is important to ignore potentially misleading cues at some times of year, while paying close attention to these cues at other times.

A recent, grand-scale study of phenology of wild species in the United Kingdom showed wide-spread, time-dependent sensitivity to ambient temperature and precipitation (Thackeray et al., 2016). The most common pattern was that prior to a phenological event (e.g., breeding), high temperature or precipitation advanced its timing. At earlier dates, response profiles were typically flat, and even earlier, high temperature or precipitation delayed events. **Figure 3B** shows the phasedependent response of egglaying to ambient temperature in Barn Swallows (Hirundo rustica; Dario Massimino, pers. comm.; Thackeray et al., 2016). Barn Swallows show advance responses to high temperatures in spring and summer, followed by delay responses during autumn and winter. The broadly similar findings from captive and wild birds emphasize the relevance of phase-specific responses for seasonally appropriate behavior. Differences between the species, in turn, may be due to the more flexible annual behaviors of Barn Swallows. It might be no coincidence that in Barn Swallows, recent cross-hemispheric colonization was observed, associated with complete inversion of the annual cycle (Winkler et al., 2017). Such an inversion is easy to envision if some individuals become sensitive to long and warm days while still on the winter grounds.

Differences between species in response-profiles have important implications for the ability of birds to respond to climate change. Strongly migratory species typically differ from residents by lower flexibility in response to spring cues (Phillimore et al., 2016). **Figure 3C** shows schematically how flexibility of migrants is specifically reduced in winter compared to resident species. Because of this programming difference, which has likely been adaptive, migrants may now be constrained in their ability to flexibly adjust annual cycles, and instead require evolutionary change (Phillimore et al., 2016). That such change may be possible has recently been shown for Pied Flycatchers (Ficedula hypoleuca; Helm et al., 2019). In this study, a full-annual cycle experiment on captive birds was replicated after 21 years, over which period a wild population had been continuously monitored. Spring activities of both, captive and wild birds, advanced at similar rates (9 and 11 days in 21 years, respectively). In the captive birds, where the full annual cycle was monitored, this advancement occurred selectively during the late winter and early spring phases, suggesting evolutionary acceleration of the circannual clock during winter (Helm et al., 2019).

#### Diel Timing

Migratory flights in many species occur at night, implying a seasonal change in individuals from almost exclusively daytime activity to nocturnal flight. This shift to additional nocturnality is starkly detectable in the wild, for example on a continental scale using weather radars (Horton et al., 2020), and in captivity

as migratory restlessness as described above. Recent studies have indeed indicated that strong Zugunruhe associates with higher probability to migrate in the wild (Eikenaar et al., 2014; Mukhin et al., 2018). However, wild and captive birds also show differences in the extent of night activity during migration seasons. For example, whereas wild birds intersperse migration nights between several nights of rest, captive birds typically show restlessness on most nights of the migration seasons (Åkesson et al., 2017).

Shifts to nocturnality during migration seasons are puzzling, given that in birds and most other organisms, day-night rhythms are stably organized by circadian clocks (Helm et al., 2017; **Figure 4**). Experimental studies on several species of songbirds have shown that Zugunruhe is, however, organized as part of the avian circadian system, rather than supplanting it (Bartell and Gwinner, 2005; Kumar et al., 2006; Coppack et al., 2008; Coverdill et al., 2008; Mukhin et al., 2018). For example, bouts of Zugunruhe, recognizable by extensive flight behavior, recur rhythmically even under experimental conditions when birds are fully sheltered from environmental information (i.e., receiving continuous dim light, constant temperature and food availability). Several studies have suggested that the birds' circadian system contains at least two internal drivers of rhythms (i.e., oscillators), of which one produces diurnal and the other nocturnal activity (Bartell and Gwinner, 2005; Mukhin et al., 2018). For much of the year, output from the day-time driver dominates, but during migration seasons, the night-time driver's activity becomes discernible. The location and functional details of the oscillators driving daytime and nighttime activity are not yet resolved. The avian circadian system consists of several pacemakers (**Figure 4**) that are interconnected, and are in turn sensitive to multiple environmental sensory inputs as well as to signaling from within the body (Kumar et al., 2006; Cassone, 2014; Helm et al., 2017). The particular responsiveness of Zugunruhe (see Adjusting the Drive to Migrate and Fueling in Response to Geomagnetic Cues) to food availability suggests links of the Zugunruhe oscillator to metabolic signals, and perhaps to brain circuits that are part of the award system (Bartell and Gwinner, 2005; Horton et al., 2019).

#### Space

Birds tend to follow inherited species- and population-specific migration routes (e.g., Helbig, 1996; Willemoes et al., 2014), which will lead them to suitable stop-over and wintering areas (Fransson et al., 2005; Newton, 2008). The endogenous programs guiding young birds encode compass and space information in relation to the internal clock (Berthold, 1996; Gwinner, 1996a,b; Able, 1980; Åkesson et al., 2014). The three biological compasses used by migratory birds are based on information from the sun and the skylight polarization pattern, stars and the geomagnetic field (e.g., Wiltschko and Wiltschko, 1972; Emlen, 1975; Able, 1980; Schmidt-Koenig, 1990; Åkesson et al., 2014), and their use is tightly connected to the diel and circannual time sense. The sun compass has a time-compensation mechanism enabling compensation for the apparent movement of sun across the sky (Schmidt-Koenig, 1990; Schmidt-Koenig et al., 1991), while the stellar compass encodes direction toward geographic north based

FIGURE 4 | Mechanistic framework of spatio-temporal migration programs. Gray frame delineates the organism. It receives spatio-temporal information from the geophysical (yellow) and the biotic (green) environment, perceived through its sensory systems. This information is integrated with the animal's biological clock (blue) to generate internal clock time. Clock time modulates effector systems, which integrate additional modifying information (red) from within the body (e.g., fuel reserves) and from the environment (e.g., weather), to set the spatio-temporal migration behavior and physiology. Oscillator symbols indicate biological rhythms (central clock in inner circle, additional clock components peripheral). The clock system is itself modified by different factors (blue); for details see Helm et al. (2017).

on the rotation center of the sky independent of time of day (Emlen, 1967, 1970). The magnetic compass is expressed relative to the angle of inclination providing directions along a northsouth axis toward and away from the poles without direct diel time input for its functionality (Wiltschko and Wiltschko, 1972), but changes of courses are expressed at relevant times of year (e.g., Gwinner and Wiltschko, 1978; Wiltschko and Wiltschko, 1992). Perception of the magnetic field seems dependent on specialized photoreceptors activated by a limited range of wavelengths of light involving cryptochrome molecules (Ritz et al., 2009; Muheim et al., 2014). Compass interactions may further lead to recalibrations during migration (e.g., Cochran et al., 2004; Muheim et al., 2006; cf. Åkesson et al., 2015), while during ontogeny a combined experience of geomagnetic information and a rotating star pattern is crucial for birds to express a relevant population-specific migratory direction at the right time of year (Weindler et al., 1996).

Once migrants have started their journey, simple compass mechanisms can sometimes explain the routes they follow (e.g., Kiepenheuer, 1984; Alerstam and Pettersson, 1991; Muheim et al., 2003, 2018; Åkesson and Bianco, 2016; cf. Sokolovskis et al., 2018). For instance, long-distance bird migrants have been proposed to set a course at sunset or sunrise and follow it relative to the position of the sun as they cross longitudes using their

time-compensated sun compass without readjusting it for local time during flight (Alerstam and Pettersson, 1991; Alerstam et al., 2001). This mechanism has gained some support from radar observations of long-distance migrating arctic waders (Alerstam et al., 2001). Birds could also use their geomagnetic compass (i.e., inclination compass; Wiltschko and Wiltschko, 1972), and keep track of the apparent inclination angle during long continuous flights, i.e., following magnetoclinic routes (Kiepenheuer, 1984), in many cases leading birds along realistic migration routes as confirmed by tracking data (Åkesson and Bianco, 2016, 2017). Especially challenging situations are met by using any of the alternative compasses in high arctic regions (e.g., Åkesson et al., 2001a; Muheim et al., 2003, 2018; Åkesson and Bianco, 2016). Recently, a comparative study evaluating route simulations demonstrated potential use across the widest latitudinal range for the magnetic compass (i.e., magnetoclinic route; Åkesson and Bianco, 2017).

An increasing number of bird tracking studies have revealed complex course changes throughout the annual cycle (e.g., Sutherland, 1998; Berthold et al., 2004; Åkesson et al., 2012; Willemoes et al., 2014). Such complex routes, involving one or more shifts during migration (Helbig et al., 1989; Willemoes et al., 2014), raise the question of how course shifts are encoded relative to the circannual program in different species and populations of birds. We find experimental support for course shifts expressed at expected times under constant environmental conditions (Gwinner and Wiltschko, 1978). In turn, there is evidence that geomagnetic information also affects the migration program. For example, expression of a relevant course shift required in some species exposure to the geomagnetic information expected at specific latitudes along the migration route (Beck and Wiltschko, 1982, 1988).

Geomagnetic information has also been shown to prompt ecophysiological changes, leading to increased mass increase in response to magnetic parameters associated with sites just in front of a large barrier. These findings, first shown in juvenile Thrush Nightingales (Luscinia luscinia; Fransson et al., 2001), suggest that this response is inherited and encoded in the endogenous migration program. For course shifts and refueling, the endogenous circannual time program seems to be involved in controlling the timing of events and in determining a seasonally correct response to geomagnetic information (Kullberg et al., 2003, 2007; Henshaw et al., 2008). These findings strongly suggest that geomagnetic cues can trigger, advance or delay phases of the migration program, reminiscent of its responsiveness to daylength (**Figure 2**).

In research on spatial programs, consideration of ontogenetic effects, usually captured by distinguishing naive from experienced migrants, has provided important insights. Whereas the navigational abilities described above hold for all age groups, adherence to the inherited migration program is typically strong in first-time migrants, but may thereafter be supplanted by experience. This has been shown repeatedly in displacement experiments, in which naive migrants followed the blueprint of the inherited program, whereas adults navigated to goal areas they had previously visited. The documented ability to correct for longitudinal displacements of adult birds (Perdeck, 1958; Åkesson et al., 2005; Thorup et al., 2007; Kishkinev et al., 2015), remains to be further explored in juveniles (cf. Åkesson et al., 2005).

#### Variation in Migration Programs

Variation between individuals and populations in the timing program and its response profile to environmental factors (**Figures 2**, **3**) can take several forms. Individuals can differ from each other in the timing of some phases of the annual cycle, but then resynchronize during a specific phase (see section "Time"). Alternatively, individuals or populations may show shifted timing of the entire annual cycle, as documented for two populations of Collared Flycatchers (Ficedula albicollis) (Briedis et al., 2016). The perhaps most extreme example is the full inversion of the annual cycle in Barn Swallows that colonized the southern hemisphere (Winkler et al., 2017).

Several mechanisms underlie changes in the timing program. If timing is consistent within individuals (i.e., individuals displaying different chronotypes), differences may have a genetic basis. Genetic differences can selectively affect specific phases of the annual cycle. For example, heightened light sensitivity of the reproductive axis (Ramenofsky, 2011; Watts et al., 2018) may advance the timing of spring but not autumn migration, and advanced circannual timing under selection for early breeding occurred specifically in late winter and spring (Helm et al., 2019). Another possibility are epigenetic changes, e.g., methylation of genes involved in biological rhythms (Merlin and Liedvogel, 2019). Ontogenetic effects, such as daylength at birth, affect timing mechanisms in mammals (e.g., Ciarleglio et al., 2011), and may do so also in birds. Additional variation within and between populations may arise from flexible responses to environmental factors that do not change the timing program, but modify its output (see below).

Variation in spatial programs can arise from different sources. Classical studies, notably experiments with European Blackcaps (Sylvia atricapilla) have emphasized genetic determination of directional preference and duration of migratory restlessness (Berthold and Querner, 1981; Berthold et al., 1992; Helbig, 1996). Such polyphenisms are now used in comparative genomic studies aimed at revealing the genetic underpinnings of migration (e.g., Liedvogel et al., 2011; Lundberg et al., 2017). The studies will be important to identify genes involved in encoding variations in space, time and fueling, but also how these genes expressed during migration are regulated.

For individual variation in orientation capacity and compass route-following, the underlying reasons may be related to the perception of the celestial and geomagnetic cues themselves (Muheim et al., 2014), as well as how these cues are encoded in the endogenous migration program. However, we still need to understand exactly how the endogenous migration program interacts with external information, and how birds keep track of space during long migrations throughout the annual cycle. A successful research agenda may be to combine an experimental approach (e.g., Kishkinev et al., 2015; Willemoes et al., 2015; Wikelski et al., 2015; Ilieva et al., 2018) with advanced tracking in the wild (e.g., Willemoes et al., 2014; Bäckman et al., 2017; Sokolovskis et al., 2018; Norevik et al., 2019).

Variation within the population can reveal interesting characteristics, where interactions of the inherited migration phenotype with different environmental factors can lead to the evolution of diverse migration patterns. Phenotypic plasticity may for example lead to advancement of migration timing in response to environmental conditions, in particular in flock-migrating birds (Fraser et al., 2019). Populations may furthermore comprise migratory and resident fractions (i.e., partial migration), or migration may differ between sex and age classes (i.e., differential migration) (Terrill and Able, 1988; Newton, 2008). In partial migration, individual phenotypes may range from completely sedentary to completely migratory (Chapman et al., 2011). Partial and differential migration systems thus enable investigation of effects of selection pressures and fitness consequences of different migration strategies, in particular in long-lived species (Gaillard, 2013; Reid et al., 2018). In different species, this variation can be based on different mechanisms. For example, in European Blackcaps migratory phenotype appears to have a strongly genetic basis (i.e., polyphenism) (Pulido and Berthold, 2010), and in Northern Wheatears (Oenanthe oenanthe; Maggini and Bairlein, 2012) and Dark-eyed Juncos (Junco hyemalis; Holberton, 1993), differences between the sexes are part of the circannual program. In other species, for example Stonechats, differences between migrants and residents are partly environmentally induced (Van Doren et al., 2017). While for most species contributions of genes and environment are unknown, it is likely that both factors are involved.

# Mechanistic Integration

Physiological studies, which are beyond the scope of this review, have provided a general picture of the mechanisms of migration programs, although details are still largely unclear. **Figure 4** summarizes these findings with an emphasis on timing. It shows schematically how information from the environment affects components of the biological clock that drives the migration program. Spatial cues may also be integrated at this stage. This information is processed in the brain, affects the clock, and prompts a response that is specific to the phase of the annual cycle (**Figure 3**). Effector systems then fine-tune behavioral and physiological responses by integrating information from within the bird (e.g., its energetic or health state) and from the immediate environment (e.g., weather, food availability) via hormonal pathways (Ramenofsky, 2011; Goymann et al., 2017; Watts et al., 2018). Ultimately, synthetization of this information leads to spatio-temporal migration behavior and physiology. The migration program itself can be modified via genetic change, during ontogeny, and epigenetically.

## PROGRAMMED FLEXIBILITY IN RESPONSE TO ENVIRONMENTAL FACTORS

Within the time window set by the program for migration, decisions about its implementation are sensitive to a range of environmental factors that determine successful migration (**Figure 4**). Departure time may be adjusted in response to level of fuel reserves, and relative to the expected onward migration route, including distance of barrier crossings (Müller et al., 2018). Responses to these environmental factors are partly inherited, and therefore, we here expand on their effects. We consider some other aspects of flexibility, for example learning, social behavior and responses to weather, to represent residual flexibility.

# Adjusting the Drive to Migrate in Response to Food

An important feature of migration is the capacity to prepare for prolonged migratory flights by fueling at stopover sites (Åkesson and Hedenström, 2007). In fact, birds are predicted to spend 1:7 parts of migration time on flight and refueling at stopover, respectively (Hedenström and Alerstam, 1998). Timing and extent of fueling events may be encoded in the endogenous program in relation to expected flight distances (e.g., Fransson et al., 2001; Kullberg et al., 2007), but may also be modified in response to environmental conditions met during flight as well as at stopover sites.

The duration of stop-over, when migrants rest at night, can be predicted by a bird's body reserves, both in the field (Goymann et al., 2010) and in captivity (Gwinner et al., 1988; Gwinner, 1996a): with increasing reserves, birds are more likely to depart, or to show high levels of Zugunruhe (i.e., migratory drive), respectively. However, this relationship only holds when sufficient local food is available. When food is scarce or inaccessible, birds show the opposite behavior, departing, or showing particularly high Zugunruhe, on low fuel stores (Gwinner et al., 1988). **Figure 5** shows how this behavior interacts with the migration program in a series of experiments on garden warblers (Gwinner et al., 1988). In September, during main autumn migration, Zugunruhe was high. When food was temporarily removed, the birds responded by increased night activity (I in **Figure 5**). On return of ad libitum food, the birds paused Zurunruhe while refeeding (P in **Figure 5**), and resumed Zugunruhe after regaining body mass (R in **Figure 5**). In winter, after Zugunruhe had naturally stopped, it was reactivated by food reduction and immediately stopped after food return. These studies show that food availability affects movement decisions of birds during and outside migration seasons, presumably via endocrine pathways (Goymann et al., 2017). The food-induced changes had no effect on the overall migration program of the study birds (Gwinner, 1996a), but likely modify a bird's actual migration.

### Adjusting the Drive to Migrate and Fueling in Response to Geomagnetic Cues

Migratory restlessness can be modified by exposure to geomagnetic field parameters expected to be met en route or at destination areas. This insight, which was previously shown for nocturnally migrating Northern Wheatears (Bulte et al., 2017), was recently also experimentally revealed for diurnally migrating Dunnocks (Prunella modularis) (Ilieva et al., 2018). Migratory restlessness was recorded for individual Dunnocks

during a 2-week period, during which one group was kept in the local Swedish geomagnetic field, while two other groups were geomagnetically displaced north (away from wintering area) or south (toward the wintering area in southern France) (Ilieva et al., 2018). The birds showed two peaks of activity throughout the 24 h-cycle, with the longest peak in the morning, associated with migration, and a shorter evening peak associated primarily with feeding (Ilieva et al., 2018). The Dunnocks displaced south reduced the morning migratory restlessness as they were exposed to the geomagnetic parameters, i.e., inclination angle and total field intensity, at the wintering area, while the control birds instead increased the migratory restlessness over the study period (**Figure 6**; Ilieva et al., 2018). The northern displacement resulted in continued, but somewhat reduced migration activity, suggesting it was not only the magnetic change itself, but also the characteristics of the magnetic parameters (i.e., combination

from Billyboy, Sweden, wikimedia.org; based on Gwinner et al. (1988).

of inclination and total field intensity at expected destination area), that interacted with the endogenous program resulting in reduced migratory restlessness as the winter destination was geomagnetically reached (Ilieva et al., 2018). Like food availability, geomagnetic cues in this example appeared to modulate the output of the migration program, but at least within the study period, not the time course of it.

# Adjusting Directions in Response to Geomagnetic Cues

Geomagnetic information can also modify the directional output of the migration program, but its effect differs between species. It is not completely understood why some species express a directional shift encoded with time as the migration season progresses (Gwinner and Wiltschko, 1978), while orientation

shifts by other species are only expressed by exposure to changes of geomagnetic information (Beck and Wiltschko, 1988). To understand this we see a need for further studies of different bird species under controlled environmental conditions including magnetic displacements (e.g., Kishkinev et al., 2015).

Feed-back from geomagnetic information helps migratory birds meet a further challenge. Trans-hemispheric long-distance migrants are crossing the geomagnetic equator, which involves a 180◦ -shift of the angle of inclination, a key-feature of the birds' magnetic inclination compass (Wiltschko and Wiltschko, 1972). Cage experiments with two long-distance migrating songbirds, the European Garden Warbler and the North American Bobolink (Dolichonyx oryzivorus), have revealed that these birds possess inherited responses where they adaptively change their preferred orientation with respect to the inclination angle, as they are exposed to a horizontal magnetic field simulating a magnetic equator crossing (Beason, 1992; Wiltschko and Wiltschko, 1992). It would be interesting to investigate if this response to magnetic inclination by shifting courses is present in most avian migrants, or if it is characteristic for the long-distance migrants adapted to trans-hemisphere flights.

Once terrestrial birds have initiated migration they may cross landmasses, but also barriers such as seas, mountains and deserts, where they may be unable to land. The inhospitable terrain may challenge their migration performance and stopover use during migration, leading to special adaptations (Åkesson and Hedenström, 2007). At coastal sites, in particular, young birds may hesitate to continue on a sea-crossing and are grounded in large numbers. They may then search for foraging sites and shelter (Alerstam, 1978), and perform reverse migration to inland sites before they continue in the migration direction several days later (e.g., Åkesson et al., 1996b). Temporary reverse migration is typically expressed near coastal barriers leading to more suitable stop-over sites (Åkesson et al., 1996b; Zehnder et al., 2002; Buler and Moore, 2011), whereas at inland locations reverse migration is less common (Åkesson, 1999; Komenda-Zehnder et al., 2002). Temporary reverse migration and movements to nearby stopover sites are predominantly found in birds with low fuel reserves (Åkesson et al., 1996b; Sandberg, 2003; Covino et al., 2015), and its directions are expressed in relation to the geomagnetic field (Sandberg, 1994; Bäckman et al., 1997). Thus, as shown for effects of food shortage and magnetic cues on migratory drive, adaptive responses to barriers seem to be embedded in the migration program.

# FLEXIBILITY RESIDUAL TO THE MIGRATION PROGRAM

Some aspects of flexibility have no clear relationship with migration programs, or appear to contradict, override or supplant them. Major effects are exerted by weather and availability of favorable winds (e.g., Åkesson and Hedenström, 2000; Åkesson et al., 2002; Shamoun-Baranes et al., 2007; Eikenaar and Schmaljohann, 2015; Sjöberg et al., 2015; Van Doren and Horton, 2018). Availability of celestial compass cues (Åkesson et al., 1996a, 2001b) and locations of suitable stopover sites may furthermore have a strong effect on individual route choices when crossing large barriers (Åkesson et al., 2016). A further contributing factor are social effects on migration (Helm et al., 2006). This underrated factor is beginning to be addressed by exciting new data from tracking studies. For example, a recent study on European Bee-Eaters (Merops apiaster) presented migration data from 29 individuals moving

in different groups (Dhanjal-Adams et al., 2018). Timing within groups was closely coordinated, and once separated, groups rejoined each other within days, even after a transcontinental journey. These data indicate both strong social effects and a detailed migration program, which could be innate, learned, or resource driven.

Learning and development are forms of flexibility that are still poorly known. For most birds it is unknown how migration performance develops with time. There are examples where adult birds have been shown to migrate faster (Ellegren, 1993), cover longer flight paths per day (Weimerskirch et al., 2006), handle wind drift more efficiently (Thorup et al., 2003), follow shorter routes and have shorter stopover times (Crysler et al., 2016) than juvenile birds. A recent example from a long-term study documents all these parameters for individually tracked migratory Black Kites (Milvus migrans; **Figure 7A**) across a time span of 1-27 years (Sergio et al., 2014). This study shows that the development of migratory behavior in young birds follows a consistent trajectory, and that the development is more gradual and prolonged than previously assumed (**Figures 7B,C**). More efficient migration performance was further shown to be promoted by a combination of individual improvement across time and selective mortality occurring most frequently in early phases of life during the pre-breeding migration period (Sergio et al., 2014). Several migration components improved across time, including increased migration speed, shorter stopover time, and increased efficiency to handle cross-winds in adult birds as compared to juveniles (Sergio et al., 2014). The strongest selection occurred on the flanks of the distribution during the early stages of life. Individuals that were able to improve their ability to handle environmental conditions efficiently on migration and to depart progressively earlier than conspecifics, obtained higher breeding and survival rates, leading to a longer life span (**Figure 7C**).

Residual flexibility can either increase, or decrease, variation between individuals in time and space. Tracking data show that in some species delays in timing of breeding and migration departure may be difficult to correct for at later stages of the annual cycle. For instance, migratory Eurasian Nightjars (Caprimulgus caprimulgus) tracked by geolocators reveal that individuals being late in their departure timing in autumn, will continue to be late throughout the annual cycle, while early birds continue to be early (Norevik et al., 2017; **Figure 8**). It is currently unknown whether these effects persisted beyond the duration of the annual cycle, or whether birds were eventually synchronized, as discussed above (see section "Time"). If differences indeed persisted, they could partly reflect individual differences in chronotype (i.e., polyphenism).

#### ILLUSTRATING EXTREMES OF VARIATION IN INHERITED SPATIOTEMPORAL BEHAVIOR

The wide range of mechanisms reviewed above can produce highly divergent outcomes. Below, we use extremes of variation within populations, ranging from great consistency to massive differentiation, to illustrate the many axes of variation in migration programs.

## Low Within-Population Variation in Cuckoos and Willow Warblers

An example of limited inter-individual variation of a rather complex migration and stopover program comes from satellite tracking of the Common Cuckoo (Cuculus canorus). Cuckoos use a largely fixed sequence of stopover sites and route directions throughout the non-breeding period (Willemoes et al., 2014; Hewson et al., 2016; **Figure 9A**). Initially the adult birds tracked from northwestern Europe depart toward the southeast in autumn, and stopover in central northern and thereafter southeastern Europe, before they initiate the Sahara crossing on southerly courses (**Figure 9A**). After the barrier crossing they make a prolonged stopover in eastern Sahel (Willemoes et al., 2014; Hewson et al., 2016; **Figure 9A**). Later in the season they will proceed to wintering areas further to the south, from where they will initiate northerly movements, following a loop migration in spring via West Africa (Willemoes et al., 2014; Hewson et al., 2016; **Figure 9A**). The Cuckoo's somewhat stereotypic sequence of flight steps and stopovers during the non-breeding period is expressed timely with the availability of food resources in the areas visited, and encoded in the endogenous program (Willemoes et al., 2014). The phenotypic expression of this complex spatiotemporal program, and apparent use of goal areas throughout the non-breeding period, open up questions on what specific information is used to identify those areas throughout the annual cycle, and how potentially external information may interact with the migration decisions and route choices. In a displacement experiment with adult cuckoos during autumn migration Willemoes et al. (2015) investigated the capacity to return to the normal migration route, stopover and wintering areas. The displaced adult cuckoos showed some individual variation in the strategy they used to return, but also a capacity to navigate across areas potentially new to them (Willemoes et al., 2015). Juvenile cuckoos have further been shown to migrate later, show larger scatter in route choice and perform migration at slower pace than adults, but still they have been confirmed to reach the expected wintering areas by following their endogenous migration program (Vega et al., 2016).

Another species with highly conserved migration routes are Willow Warblers (Phylloscopus trochilus) (**Figure 10**). However, these routes are specific to subspecies (P. t. trochilus, acredula, yakutensis) within the species' vast breeding range in Northern Europe and Asia. Willow warblers, thus, represent an interesting example where the evolution of migration routes, speciation and range expansion may be investigated (Bensch et al., 2009; Lundberg et al., 2017). Thanks to miniaturization of tracking technology we are now able to document migration behavior in these smallest of songbirds from different parts of their range (Lerche-Jørgensen et al., 2017; Sokolovskis et al., 2018; **Figure 10**), which opens up the possibility to record route constancy and migration timing for birds with known, locally differing, genotypes (Lundberg et al., 2017). Special attention

is given to hybrid zones where individuals with different endogenous migration programs breed side by side and may cross-breed, possibly resulting in intermediate directions in hybrids (Helbig, 1991; Delmore and Irwin, 2014). The consistent, but locally differentiated routes of Willow Warblers, thus offer great opportunities for detailed studies of the genetic program encoding migration behavior, and the phenotypic expression of it (Bensch et al., 2009; Ruegg et al., 2014).

#### High Within-Population Variation in Plovers and Albatrosses

(2014), reprinted by permission from Springer, Nature.

Conversely, exceptional variation in migration directions, route choice and wintering areas is exemplified within one single study population of the Little Ringed Plover (Charadrius dubius) (Hedenström et al., 2013). In this population, birds breeding side by side within the same gravel pit may migrate to wintering areas from Nigeria in the west to India in the east (Hedenström et al., 2013; **Figure 9B**). The extreme variation in inherited direction and longitudinal range of wintering area selection call for further understanding of the genetics behind the phenotypic expression of the migration program in birds like the Little Ringed Plover.

Another example for exceptionally high variation comes from partially migratory Wandering Albatross (Diomedea exulans), whose behavior differs from conventional migration systems (**Figure 11**). The Wandering Albatross is known for its long lifespan, oceanic lifestyle and for breeding on isolated sub-Antarctic Islands. Between reproductive events it spends a sabbatical year at sea before returning to its previous breeding island (Tickell, 1968; Weimerskirch and Wilson, 2000). In this species, differential non-breeding movement strategies have evolved in populations with limited genetic differentiation (Milot et al., 2008). Birds from different breeding colonies may predominantly move to different ocean areas (Weimerskirch et al., 2015; **Figure 11**). Young wandering albatrosses have further been shown to follow similar routes as adults during their first migration and to

FIGURE 8 | Carry-over effects across the annual cycle for migratory adult European Nightjars. A graphical presentation of distributions of stops and movements in the annual cycle for European Nightjars tracked by microdataloggers, plotted with respect to starting date 1 July. Black bars represent stationary periods and the pale gaps show time of movement. Gray sections refer to periods for which occurrence of stops could not be resolved. The dots show the individual's timing of the six distinct annual events as shown in inset, and the color gradient shows the order of the birds in each event. Green illustrate the first bird and red the last bird, sorted by the date of arrival to the breeding area. From Norevik et al. (2017).

FIGURE 9 | Examples of contrasting variation in migration routes. (A) Common Cuckoo, (B) Little Ringed Plover. (A) Staging areas of eight satellite-tracked adult Common Cuckoos with vector directions between stopovers indicated by inserted orientation diagrams. Lines are connecting staging sites and do not necessarily represent the paths followed. From Willemoes et al. (2014). (B) Autumn migration tracks of Little Ringed Plovers as revealed by geolocators. Filled small circles show three-day means of positional data and filled large circles are mean location for winter positions. Open circles indicate the location of a stopover period. Broken lines indicate unknown movement around the autumn equinox. Note that one individual was tracked for two consecutive migrations using different geolocators (male B). From Hedenström et al. (2013). Maps in Mercator projection.

trochilus and southeast-migrating acredula Willow Warblers, and their approximate initial wintering areas in sub-Saharan Africa. Black bar indicate location of hybrid zone between acredula (north) and trochilus (south) in Scandinavia (Bensch et al., 2009; map courtesy Keith W. Larson, Sweden; Wikimedia.org). (B) Migration of southwest-migrating trochilus Willow Warblers from breeding to wintering grounds (individuals represented by different colors; from Lerche-Jørgensen et al., 2017). (C) Migration of P. t. yakutensis tracked by geolocation from breeding sites in Far East Russia to initial wintering areas in East Africa. From Sokolovskis et al. (2018).

spend their first year in sex-specific ocean areas overlapping with the adults, with males moving over twice the distance of females, demonstrating endogenous control of area use (Åkesson and Weimerskirch, 2014).

During the sabbatical year, adult wandering albatrosses follow three alternative movement strategies: sedentary, sedentary with excursions, or migratory (Weimerskirch et al., 2015; **Figure 11**). The proportion of birds adhering to the different strategies may differ between colonies. Wandering Albatrosses at Kergulean Island are all migratory, while the sedentary strategy is present in Crozet Island birds (Weimerskirch et al., 2015; **Figure 11**). The migratory strategy is furthermore more commonly used by males than by females (Weimerskirch et al., 2015), and differentially expressed

already in young birds (Åkesson and Weimerskirch, 2014). A sedentary strategy, predominantly used by females at Crozet Island, has increased in recent years (Weimerskirch et al., 2015), possibly as a consequence of climate change (Fryxell and Holt, 2013), revealing a potential to adapt to new environmental conditions by phenotypic flexibility or as a consequence of natural selection. A long-term population change in movement strategies including settling down, may in turn lead to bird diversification and speciation as suggested by Rolland et al. (2014).

# CONCLUSION AND OUTLOOK

In this review, we have shown how in bird migration, inherited programs and responses to the environment interact. As in other fields of biology, it is time to leave behind old dichotomies between genetics and physiology on the one hand, and ecology on the other. Evolution has brought about an impressive range of solutions to the problem that migration requires both predictive anticipation and flexibility (**Figures 9**, **11**). For some aspects of migration, we begin to understand the birds' flexibility on the basis of inherited reaction norms that provide solutions which were effective over evolutionary time. These insights refine views of environmental effects on migration, as being dependent on the phase of the annual cycle and a bird's migration program (**Figures 3**, **5**). They also refine views of carry-over effects as being at least in part permitted, or counter-acted, by migration programs (**Figure 2**). The combined developments of molecular tools and tracking technology, if applied to rewarding model systems (e.g., **Figure 10**), are set to greatly foster this understanding. For some other aspects, we currently do not know how decisions are made, and how migrants improve by learning. The study on Black Kites (**Figure 7**) is one of the most informative about the development of migration performance in individual birds, but calls for follow-up studies on other species (Campioni et al., 2020), in particular those relying strongly on an endogenous program for their first migration. Likewise, the cited study on European Bee-Eaters (Dhanjal-Adams et al., 2018) offers

exciting insights on migration of highly social species, which will hopefully be complemented by work on species with different social systems. In addition, recent findings from the European Nightjar call attention to yet another temporal domain that may have been largely overlooked, at least in landbirds. Nocturnally migratory nightjars time fueling and migration events to the different phases of the moon (Norevik et al., 2019), and thereby call for further studies of moon cycle effects on migratory birds, ideally with a circannual perspective (Chapin and Wing, 1959; Cruz et al., 2013).

Ultimately, combined ecological and mechanistic studies may explain why some species strictly adhere to spatiotemporal programs whereas others are flexible, and how

#### REFERENCES


selection pressures, from competitive temporal niche segregation (Heim et al., 2018) to climate change, may shape migrations in the future. We hope that our overview will encourage future cross-over between approaches that focus on annualcycle programs and those that focus on the environment, as jointly these components enable migratory birds to carry out their great journeys.

#### AUTHOR CONTRIBUTIONS

Both authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.


gradual divergence in a ring. Mol. Ecol. 18, 3087–3096. doi: 10.1111/j.1365- 294X.2009.04210.x


individuals are not? Proc. R. Soc. Lond. B 281:20132161. doi: 10.1098/rspb.2013. 2161


perspectives of timing in the wild. Philos. Trans. R. Soc. B 372:20160246. doi: 10.1098/rstb.2016.0246


few and cluster in large haplotype blocks. Evol. Lett. 1, 155–168. doi: 10.1002/ evl3.15


Newton, I. (2008). The Migration Ecology of Birds. London: Academic Press.


(London: Academic Press), 205–237. doi: 10.1016/b978-0-12-374932-1.00 046-9



**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Åkesson and Helm. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Male-Biased Partial Migration in a Giraffe Population

Michael B. Brown<sup>1</sup> \* and Douglas T. Bolger <sup>2</sup>

*<sup>1</sup> Department of Biological Sciences Graduate Program in Ecology, Evolution, Ecosystems, and Society, Dartmouth College, Hanover, NH, United States, <sup>2</sup> Environmental Studies Program, Dartmouth College, Hanover, NH, United States*

#### Edited by:

*Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway*

#### Reviewed by:

*Derek Benjamin Spitz, University of California, Santa Cruz, United States Colleen Cassady St. Clair, University of Alberta, Canada*

\*Correspondence: *Michael B. Brown michael.b.brown.gr@dartmouth.edu*

#### Specialty section:

*This article was submitted to Population and Evolutionary Dynamics, a section of the journal Frontiers in Ecology and Evolution*

Received: *02 July 2019* Accepted: *31 December 2019* Published: *21 April 2020*

#### Citation:

*Brown MB and Bolger DT (2020) Male-Biased Partial Migration in a Giraffe Population. Front. Ecol. Evol. 7:524. doi: 10.3389/fevo.2019.00524*

Partial migration is a common movement phenomenon in ungulates, wherein part of the population remains resident while another portion of the population transitions to spatially or ecologically distinct seasonal ranges. Although widely documented, the causes of variation in movement strategies and their potential demographic consequences are not well-understood. Here, we used GPS telemetry data and individual-based photographic surveys to describe evidence for the partial migration of giraffe (*Giraffa camelopardalis*) in the tropical savanna habitat of Murchison Falls National Park, Uganda. Seasonal movements in giraffe have been described but have not been systematically investigated within the framework of partial migration. We characterized movement behaviors of eight female GPS tracked giraffe across one full year using a model-driven approach of net-squared displacement metrics. To further evaluate these space use patterns at the population-level, we used closed robust design multi-state capture recapture models derived from individually based photographic surveys collected seasonally over three years. We also characterized environmental conditions associated with seasonal space use by conducting ground-based vegetation surveys and analyzing remotely sensed phenology data. Our results from both individually based telemetry models and population-level multi-state models suggest intra-population variation in seasonal space use strategies with three dominant movement classes: (1) Residents in deciduous savanna characterized by *Acacia sieberiana, Acacia senegal, Harrisonia abyssinica,* and *Crateva adansonii* in the far western end of the park. (2) Residents in the broadleaf savannas characterized by *Pseudocedrela kotschyi, Stereospermum kunthianum, Termalia* spp., and *Combretum* spp. in the central sector of the park (3) Male-biased migrants that transitioned seasonally between the acacia savanna in the wet seasons and the broadleaf savanna in the dry seasons. Our results offer insights into how giraffe navigate spatiotemporally dynamic environments at both individual and population levels, providing ecological mechanisms for the emergent population dynamics of these large-bodied topical browsers.

Keywords: partial migration, giraffe, mark recapture, multi-state model, net squared displacement, GPS telemetry

# INTRODUCTION

Understanding the ecological interactions that influence an organism's movement decisions and the subsequent fitness consequences of movement remains a major theme in ecology. Movement allows organisms to utilize resources that are distributed heterogeneously over space and time (Dingle and Drake, 2007). Landscape level movement patterns vary considerably across species, within species across geographic regions, and even among individuals within a single population (Mueller et al., 2011; Naidoo et al., 2012b; Singh et al., 2012). As such, studying the causes and consequences of variation in movement strategies can cast light on how organisms' life history characteristics influence space use and population dynamics in spatiotemporally varying environments.

Migration is a common movement strategy wherein organisms consistently move to spatially distinct stable ranges to track resource distribution or avoid predation risk in temporally and spatially varying environments (Fryxell and Sinclair, 1988; Dingle and Drake, 2007). Increasingly, however, studies suggest that many populations are only partially migratory (Chapman et al., 2011). In partial migration, some members of the population are resident while others exhibit migratory behaviors (Dingle and Drake, 2007; Chapman et al., 2011). Identified in a diverse suite of taxa (Chapman et al., 2011; Ohms et al., 2019) and observed on a wide range of spatial scales, from several kilometers (Mysterud, 1999; Gaidet and Lecomte, 2013) to hemispheres (Shaffer et al., 2006), partial migration, because of the inherent variation of movement strategies with a population, provides a useful process to evaluate the causes of intraspecific variation in movement behaviors and the fitness consequences of different space-use strategies (Chapman et al., 2011). Although these variations in movement may have a genetic basis in some systems (Berthold and Helbig, 1992; Bensch et al., 2011; Hess et al., 2016) researchers are increasingly identifying scenarios in which these alternative movement strategies are conditional on the state of an individual and may be plastic over the life of an individual (Sutherland, 1998; Found and St. Clair, 2017). Studies examining conditional migration have suggested that these movement behaviors may be contingent upon asymmetries in sex or social dominance, or the ability of individuals to assess resource conditions and respond to social cues (Chapman et al., 2011).

Despite a growing body of research on the ecological mechanisms for the emergence and maintenance of individual variation in movement strategies, there is a lack of studies evaluating this variation at the population level (Ohms et al., 2019). Many studies examining varying movement strategies use tracking devices (GPS, VHF, PIT tags) to monitor the movements of focal individuals and extrapolate these processes to the population level (Struve et al., 2010; Mysterud et al., 2011; Cagnacci et al., 2015). Although useful in characterizing movement behaviors, quantifying seasonal ranges, and identifying the timing of seasonal movements with precision, many of these telemetry/tracking studies are limited in their inference by smaller sample size and relatively short study durations (Hebblewhite and Haydon, 2010). Additionally, few studies account for variation in movement strategies across different age and sex classes because of logistical constraints associated with collaring multiple individuals across these different categories. Because of these limitations, studies connecting varying movement behaviors among different age/sex classes to population level processes over longer time periods are rare and as result, researchers lack the ability to evaluate causes and consequences of movement across multiple scales (Torney et al., 2018). This shortage of empirical inquiry limits the understanding of the associations among varying movement strategies, population dynamics and landscape-level processes and can potentially result in misinformed conservation strategies that do not properly account for the demographic effects of movement processes over larger timescales (Bolger et al., 2008).

In this study we employ multiple complementary approaches to evaluate partial migration at both the individual and population levels across multiple seasons over three years for a large-bodied tropical browser, the giraffe (Giraffa camelopardalis). The unique foraging behaviors and life history characteristics of the giraffe make it a suitable study species for examining variation of movement strategies. Once widely distributed across much of sub-Saharan Africa, giraffe have recently undergone substantial population declines and range restrictions (Muller et al., 2016). Despite this continent scale population decline, the current giraffe distribution encompasses a wide range of habitats and climates, from the hyper-arid Hoanib desert of Namibia to more mesic savannas in Uganda, Tanzania, and Democratic Republic of Congo (van der Jeugd and Prins, 2000; Fennessy, 2009; Flanagan et al., 2016). Giraffe exhibit a wide range of space-use behaviors across these habitat types with larger home ranges reported in more arid environments and smaller home ranges in more mesic savannas (van der Jeugd and Prins, 2000; Fennessy, 2009; Flanagan et al., 2016; Knüsel et al., 2019). Giraffe are large-bodied tropical browsers and forage almost exclusively upon leaves, flowers, and seeds of woody vegetation (Pellew, 1984a). The quality and quantity of forage resources varies considerably in the seasonal tropical savannas that are characteristic of much of their range and as a result, giraffe have been shown to exhibit seasonal variation in diet composition and habitat selection (Field and Ross, 1976; Pellew, 1984a; Bercovitch and Berry, 2018). Unlike many other ungulates, giraffe are aseasonal, asynchronous breeders and consequently do not have defined breeding or birthing seasons that are often characteristic of partially migratory ungulates in temperate systems (Leuthold and Leuthold, 1975), although some studies do report minor increases in calf abundance during dry seasons (Sinclair et al., 2000). Additionally, giraffe are capable of simultaneous gestation and lactation throughout all seasons (Deacon et al., 2015). Giraffe social structure is generally characterized as a fission-fusion system in which loosely associated herds often change membership, with associations potentially influenced by kinship and individual preferences (Carter et al., 2013; Dagg, 2014). As a result of female reproductive asynchrony and fluid social associations, males are thought to adopt a roaming reproductive strategy in which they search for sexually receptive females (Bercovitch et al., 2006). Prior research also suggests sexual variation in resource selection (Pellew, 1984b; Young and Isbell, 1991; Ginnet and Demment, 1997), providing potential mechanisms for sexual variation in space use. Although there is empirical support for seasonal long-distance movements in giraffe (Le Pendu and Ciofolo, 1999) other studies describe giraffe populations as non-migratory (Pellew, 1984a). Despite these foundational studies and the unique opportunities presented by giraffe's natural history, a systematic investigation of intra-population variation of giraffe movement behaviors has not been conducted.

In this study, we analyzed a full year of GPS telemetry data with net squared displacement models to characterize variations in giraffe movement patterns and identify the temporal and spatial extent of landscape-level movements of eight focal female giraffe. We also used over three years of seasonal population surveys and closed robust design multistate mark recapture models to evaluate population level movement behavior at the seasonal scale. Using these models, we tested hypotheses that giraffe adopt space-use strategies to in response to seasonal variation in resource distribution, and that sexual variation in resource use or reproductive tactics influence space use strategies. We evaluated observed movement strategies with regard to measured variation in the woody vegetation community composition to explore the causes and consequences of partial migration of a large-bodied browser in a tropical environment.

# METHODS

#### Study Site

We conducted this study in Murchison Falls National Park (MFNP), Uganda. MFNP is located in northwestern Uganda (02◦ 15′ N, 31◦ 48′E), and encompasses an area of 3,840 km<sup>2</sup> , making it Uganda's largest national park. The park is bisected by the Victoria Nile River, with the southern portion dominated by dense forest and the northern portion characterized by savanna, Borassus palm woodland, and riverine woodland. The average annual rainfall for MFNP ranges from ∼1,100–1,500 mm and is bimodally distributed with the short rains occurring from mid-March to June and August to December, and with the long dry season occurring from late-December to mid-March (Fuda et al., 2016).The current natural distribution of giraffe is limited to the northern portion of the park. Northern MFNP is divided into a series of management sectors, which roughly correspond with drainages and habitat type. We restricted our study to the western half of this area, comprised largely of the Delta and Wankwar sectors since small rivers in the central area of the park limit the potential for giraffe movement across this east/west gradient and our own mark-recapture data suggest that the majority of giraffe occur in these two sectors with little interchange with giraffe further east (**Figure 1**).

MFNP currently supports the largest population of giraffe in Uganda with recent surveys estimating a population size of 1,318 adults/subadults (Brown et al., 2019). Over the past 60 years, periods of civil unrest led to large scale defaunation of the park, including a substantial reduction in the giraffe population. However, following the cessation of conflict in Uganda in the mid-1990's, the giraffe population has increased rapidly such that it is now larger than at any point in recorded history. Recent population estimates derived from mark-recapture methods suggest annual population growth rate from 2014 to 2017 of λ = 1.14 (Brown et al., 2019).

#### Characterizing Spatiotemporal Dynamics of Vegetation

Because spatiotemporal variation of resources is a requisite condition for the emergence of migratory behavior, we first characterized potential bottom-up effects by quantifying the composition of vegetation communities in the study site with a series of woody vegetation surveys using a modified plotless, k-tree sampling method (Kleinn and Vilcko, 2006; M ˇ agnussen et al., 2012) across the entire extent of northern MFNP. During concurrent giraffe surveys (see methods below), we conducted vegetation surveys at the location of each giraffe herd, such that vegetation surveys represent plant community composition associated with giraffe positions in the heterogeneous savanna. In each survey, we measured the distance to the nearest ∼15 tree (>1 m height) using a laser rangefinder and identified every tree within that radius to species. We then calculated the density of each tree species by dividing the number of trees counted by the area of a circle with radius equal to the distance to the fifteenth tree. We conducted 259 woody vegetation surveys in the Delta and 139 surveys in the Wankwar sector. To validate plant identification, we collected and pressed voucher specimens of each unique species which were then independently identified by botanists at the Makerere University herbarium. To characterize woody vegetation composition in each sector within the park, we combined all the surveys and calculated the proportional species composition in each sector. We compared raw counts of surveyed woody vegetation across the sectors using a Pearson's chi-squared test to evaluate the prediction that the different sectors were comprised of different communities of woody vegetation.

To quantify temporal variation in primary productivity, we used MOD13Q1 MODIS 16-day (250-m) Enhanced Vegetation Index (EVI) data for the period July 2014 to May 2019. EVI is remotely sensed vegetation index that measures greenness and has been shown to be effective for monitoring primary productivity in African savannas (Sjöström et al., 2011) and has been used as an indicator of vegetation quantity and quality for herbivore spatial ecology studies (Naidoo et al., 2012a; Villamuelas et al., 2016). We accessed these data through the NASA Application for Extracting and Exploring Analysis Ready Samples (AppEEARS) platform, extracting all available values for EVI at each of our earlier woody vegetation sampling points. Although the spatial resolution of this product renders it inappropriate for isolating the phenology of woody vegetation surrounding grassy vegetation likely contributes to the spectral signature at this grain- the overall time series is a useful indicator of the timing of seasonal transitions and potential phenological responses of deciduous woody vegetation to rainfall. We plotted the timeseries of EVI data to evaluate seasonality of productivity across both sectors. If EVI is an effective measure of plant phenology, we expected pronounced declines in EVI values during the dry seasons and subsequent increases in EVI during the wet seasons.

#### Characterization of Giraffe Movement Behaviors GPS Telemetry

To track the location of individual giraffe over time, we deployed solar-charged, ossicone-mounted GPS units (Savannah Tracking) on 20 giraffes in MFNP during April 2018 (and an additional GPS unit in August 2018). We deployed the units at the beginning of the wet season, such that all giraffe were expected to be in their wet season ranges when tracking commenced. Giraffe were immobilized from a Landcruiser using a mixture of etorphine and azaperone and tracking units were attached under the supervision of a local government wildlife veterinarian. We selected both male (n = 5) and female (n = 15) focal individuals across both the Delta (n = 12) and Wankwar (n = 8) sectors to monitor movement strategies across sexes and habitat type. Before immobilizing giraffe, we photographed the right side of the candidate focal individuals and compared their spot patterns to a database of previously observed giraffe encounters (see survey methods below; **Supplementary Table 1**). We then selected individuals to ensure relatively even representation among three possible prior space-use patterns: (1) Individuals that were previously observed only in the Delta (n = 6); (2) Individuals that were previously observed only in Wankwar (n = 6); and (3) Individuals that were previously observed in both the Delta and in Wankwar (n = 8). We programmed all GPS tracking units to record coordinate fixes at hourly intervals and transmit location data to an off-site server twice daily via satellite link. We excluded data from units that lost function before the seasonal transition to the long dry season, resulting in eight functional units deployed on females for subsequent analyses (**Table 1**).

#### Analyses: Net Squared Displacement Models

To classify movement behaviors of individual giraffe, we employed a model driven approach based on the net squared displacement calculated from each giraffe's movement trajectory (Bunnefeld et al., 2011). Net squared displacement (NSD) is the squared value of the Euclidean distance between the starting location of a trajectory and every subsequent coordinate fix (Turchin, 1998). To categorize the movement behavior of each giraffe, we fit individual NSD time series data with a set of a-priori non-linear models, each representing the theoretical NSD signature of different movement strategies (residence, migration, mixed migration, dispersal, and nomadism) (Bunnefeld et al., 2011; Singh and Leonardsson, 2014; Spitz et al., 2017). In addition to categorizing movement behaviors, these models have ecologically interpretable parameters, allowing for direct estimates of migration departure date, the rate of movement between seasonal ranges, the distance between seasonal ranges, and the duration of residence on the seasonal ranges. We used Akaike information criterion (AIC) to identify the best fitting model (Burnham and Anderson, 2002; Bunnefeld et al., 2011). Before model fitting, we designed several a priori decision rules to limit the possibility of movement strategy misclassification.

#### TABLE 1 | Summary of individual giraffe movement categories.


*Among the eight collared individuals monitored over one wet season to dry season transition, we found three emergent space used patterns: (1) delta residents, (2) Wankwar residents, and (3) seasonal migrants.*

To reduce the potential for small scale intra-seasonal movements being misclassified as migration, we only considered migration, mixed migration, and dispersal models in which the estimated parameter value for the squared migration/dispersal distance exceeded <sup>√</sup> 150km<sup>2</sup> = 12.25 km. This spatial threshold for migratory behavior was set to exclude most previously reported values for total daily movements for giraffe, such that movements on the scale of reported daily displacement would not be misidentified as migration (McQualter et al., 2015). Additionally, we restricted migrant/partial migrant models to those in which had a minimum time of occupancy in the seasonal range of 21 days, effectively restricting intra-seasonal exploratory behavior being categorized as migration (Spitz et al., 2017). In scenarios where a priori decision criteria disqualified the top model, we used the next supported model according to AIC. We conducted NSD model fitting and model selection with the MigrateR package (Spitz et al., 2017) in R (R Core Team, 2019).

#### Population-Level Movement

To examine population level patterns of seasonal transitions across all age and sex classes and to evaluate potential differences in survival across geographic sectors over time, we conducted photographic surveys of the entire population and analyzed encounter data with multi-state mark recapture methods.

#### Individual-Based Photographic Surveys

We conducted photographic surveys of the study area at 4-month intervals between December 2014 and December 2017. We scheduled these surveys to correspond with periods of seasonal transitions (December: end of the long rains, March/April: end of the dry season, and July/August: end of the short rains). In accordance with a robust survey design (Pollock, 1982; Kendall et al., 1995; Pollock et al., 2002), each primary sampling event consisted of two secondary sampling events during which we drove a series of fixed routes comprising the road network over the entire study area. Secondary sampling occasions were separated by a time of <1 week, during which we assumed that the system was closed (no births, deaths, immigration, emigration, or substantial movement). Along these routes, we photographed the right side of each individual and identified every individual giraffe using its unique, unchanging coat pattern in association with WILD-ID, a pattern recognition software program (Foster, 1966; Bolger et al., 2012). We also recorded the spatial coordinates of each observation, the age class and sex of each giraffe, and any visible signs of disease or injury. We estimated the age class of each giraffe (calf: 0–12 months; subadult female: 1–3 years; subadult male: 1–6 years; adult female: >3 years; adult male >6 years) using physical characteristics in association with estimated axial and appendicular body proportions (Strauss et al., 2015). During photographic surveys, we also conducted opportunistic observations of foraging to quantify giraffe diet composition across the different habitat types. While photographing each individual giraffe, if it was foraging, we identified the species of woody vegetation being consumed.

We completed 10 primary events, each comprised of two secondary events, resulting in 20 surveys over 3 years (consisting of 80 total days of field surveys). After filtering the data to exclude individuals observed outside of our defined study area, and individuals with insufficient location data, our photographic database consisted of records for 1,453 unique giraffe over 9,374 individual encounter records.

#### Analysis: Closed Robust Design Multi-State Capture Recapture Models

From the seasonal robust surveys, we developed encounter histories for every individual giraffe. We assigned a geographic state (Delta or Wankwar sector) for each encounter based on the location of the observation. We then used a closed robust design multi-state (CRDMS) modeling framework to estimate associated parameters: capture probability (**p**), survival (**S**), transition probabilities (9) between sectors, and a derived parameter of population size (**N**) (Lebreton et al., 2009; Chabanne et al., 2017). CRDMS models assume that at there are no sector transitions within each primary sampling event, an assumption that our raw data only infrequently violated (<1% of encounters) (Arnason, 1972, 1973). To correct for this, if we observed an individual in different sector within the same primary sampling event, we assigned both encounters to the sector where it was first encountered during the primary period.

We used multi-model comparisons to test a series of hypotheses relating movement and variation of demographic parameters to sex, location and time of year. To test for the effect of temporal variation of resources on demographic parameters, we developed four different schemes for temporal parameterization of models: constant values, variation by primary session, variation by a three season classification (post short rain, post long rain, and post dry season) and variation by a two season classification (post short rain/post long rain, and post dry season). If there was no temporal variation in movement, survival, or capture probability, we expected best supported models to have constant parameter values across all primary sampling events. If there was a consistent seasonal signature in demographic parameters in which giraffe responded differently during the three seasons, then we expected the three season model to be best supported. Conversely, if there was seasonal variation in demographic parameters but no difference in giraffe response between the short rainy season and the long rainy season, we expected best support for models with the twoseason classification. Lastly if demographic parameters varied over time but did not consistently vary in magnitude or direction with our a priori seasonal classification schemes across multiple years, we expected best supported models to have demographic parameters vary across primary sampling events. Similarly, we incorporated sex of the individual as a classification factor to test for sex biased responses of demographic parameters. If sex affected movement or survival, we expected models with sex as a classification to be better supported by the data than models that do not incorporate variation due to differences in sex. Lastly, to test for differences in survival across the sectors, we incorporated location as a potential covariate for the estimate of survival.

We then developed a suite of candidate models in which we allowed most model parameters to vary by state (sector location), sex, and primary session/2-season/3-season. We constrained the capture probability and within session secondary resight probability to better estimate capture probability across primary sampling events. Because we had similar secondary resight rates across all primary sampling events, we also constrained the capture probability so that it remained constant over primary periods. We then ran all possible combinations of session, season, sex, and sector varying parameter estimates and ranked the output models using AIC to identify the model that was best supported by the data (Burnham and Anderson, 2002). We performed these analyses in MARK (White and Burnham, 1999) called from R (R Core Team, 2019) with the RMARK package (Laake, 2013).

#### Analysis: Assessing Spatial and Sexual Variation of Diet Composition

To test for differences in diet composition across both sex and geographic sectors (Delta/Wankwar), we partitioned opportunistic foraging observations by sex and sector. We then compared the relative proportions of woody plant species in each diet using a series of pairwise Pearson's chi-squared tests.

#### RESULTS

#### Spatiotemporal Variation in Resource Distribution

We found a significant difference in the species composition of woody vegetation across the two sectors of the park (χ <sup>2</sup> =4583.8, p ≤ 0.01, df = 10). The available woody vegetation in the Delta sector was comprised largely of the deciduous/semi-deciduous acacia species (primarily of A. senegal, A. sieberiana, and A. drepanolobium), the semideciduous leafy Crateva adansonii, and the evergreen shrub Harrisonia abyssinica whereas the central Wankwar sector was comprised predominantly of the broad-leaf semi-deciduous Pseudocedrela kotschyi, Stereospermum kunthianum, Piliostigma thonningii, and Terminalia spp. (**Figure 2**).

We found strong seasonal signals in EVI measurements that corresponded with our a priori understanding of seasonal rainfall patterns. During the dry season, which typically commenced mid/late December, EVI values rapidly dropped until the onsets of the rainy season in March/April, after which there was rapid green-up of vegetation. These productivity trends are consistent in timing and magnitude across years and habitat types, demonstrating the strong effects of seasonality on vegetation dynamics in this savanna system (**Appendix 2**).

# Individual-Level Telemetry and NSD Models

The NSD models classified three major space use categories among the eight individually tracked female giraffe: (1) yearround Delta residents (n = 2); (2) year-round Wankwar residents (n = 2); and (3) Individuals that migrated from the Delta to Wankwar seasonally (n = 4) (**Table 1**).

For seasonal migrants, the transitional period typically occurred rapidly at the onset of the dry season in mid-December/early January (**Figure 3**). We found individuals' seasonal migration distances varied from 13 to 30.4 km (**Table 1**), although all migrants moved from the a priori defined boundaries of the Delta to Wankwar during the dry season. Notably, NSD models classify movements independent of underlying environmental covariates, so this finding is independent of a priori definitions of sectors. Individuals categorized as migrants returned to the wet season range in mid/late April. These movement behaviors were mostly consistent with our classification of movement behaviors from previous survey encounters of the tracked individuals (**Table 1**).

Seasonal migration was characterized by rapid, directed movement between Wankwar and the Delta. Conversely, both Wankwar residents and the Delta resident exhibited relatively tortuous movement trajectories within their respective sectors throughout the study period (**Figure 4**).

### Population-Level Surveys and CRDMS Models

Our best fitting CRDMS model based on AIC score was a model where survival (**S**) varied by sex and time, capture probability (**p**) varied by sex and sector, and transition probability (9) varied by sector, sex, and the three season temporal categorization scheme (**Table 2**). Notably, the models with sex varying transition probabilities (9) outperformed models in which sex was not a factor. Males consistently exhibited the highest transition probabilities during both dry and wet season transitions. Capture probability (**p**) for both sexes varied across sectors with **p** being higher in the Delta sector (male: 0.485, SE 0.005; female: 0.496, SE 0.006) than in Wankwar sector (male: 0.267, SE 0.006; female: 0.275, SE 0.005). The best fitting model's estimates for transition probability (9) indicated a strong seasonal variation in the direction of transitions. These parameters effectively represented the probability of migration across these distinct sectors. For both males and females, the transition probability from the Delta to Wankwar (9 <sup>D</sup>→W) was consistently highest between December and March (seasonal shift to the dry season) (**Figure 5**). Transition probabilities from Wankwar to the Delta (9 <sup>W</sup>→D) for both males and females during this same period were consistently low. Similarly, during the seasonal shift from the dry season to the wet season (March to July) and between the wet seasons (July to December), seasonal sector transition probabilities from Wankwar to the Delta (9 <sup>W</sup>→D) were highest.

The best supported model yielded apparent survival parameter estimates (**S**) that varied over sex and primary sampling event. For both male and female, these apparent seasonal survival estimates were consistently high with no

apparent seasonal pattern in survival estimates for these adult/subadult giraffe (**Figure 6**).

# Spatial and Sexual Variation of Diet Composition

There were significant differences in diet composition across geographic sector and sex (**Figure 7**). Females in the Delta had a different diet profile than females in Wankwar (χ <sup>2</sup> = 176.82, p ≤ 0.01, df = 13) with the latter group characterized predominantly by broadleaf S. kunthianum, and P. kotschyi and the former group characterized by Acacia sp., C. adansonii, and H. abyssinica. Males exhibited a similar significant difference in diet composition across the two sectors (χ <sup>2</sup> = 291.41, p ≤ 0.01, df = 13). We also found a significant difference in diet composition between females and males in the Delta (χ <sup>2</sup> = 65.35, p ≤ 0.01, df = 10) with females being observed consuming relatively more Acacia sp. and males consuming proportionally more C.


adansonii. We found no difference in diet composition between males and females in Wankwar (χ <sup>2</sup> =5.97, p ≤ 0.54, df = 7).

#### DISCUSSION

Our study suggests that in the spatiotemporally dynamic savannas of western MFNP, giraffe exhibit intra-population variation in space-use strategies, with some individuals transitioning between wet season ranges dominated by deciduous Acacia sp., Harrisonia abyssinca, and C. adansonii savannas in the Delta, to spatially distinct dry season ranges dominated by semi-deciduous broadleaf P. kotschyi, and S. kunthianum in the Wankwar sector. Given the complementary evidence suggesting variation in seasonal movements in which a portion of the population inhabits geographically and ecologically distinct seasonal ranges, we propose that this population exhibits partial migration. In migratory individuals, NSD models indicate a rapid departure from the wet season ranges at the end of December, characterized by directed movement to the dry season range and a subsequent synchronous return to the wet season range in mid-April at the onset of the short rains. These tracked giraffe exhibited range fidelity within the seasons and typically only exhibited long distance transitions across sectors between seasons. These seasonal transitions were also detected by population-level CRDMS models derived from photographic surveys, which demonstrated consistently higher transition

probabilities from the Delta to Wankwar between December and March (dry season), and from Wankwar to Delta in both March to July and July to December (wet seasons). Additionally, these models also suggest sex-biased partial migration, with males having greater seasonal transition probabilities, both to Wankwar in the dry season and to the Delta in the wet seasons. Despite the spatiotemporal variation in habitat quality metrics and the temporal variation in giraffe density across the two sites, we found only a marginal difference in adult/subadult survival over time.

Partial migration is a common movement phenomenon in ungulates, providing unique ecological contexts to explore the causes and consequences of intrapopulation variation in movement strategies (White et al., 2007; Hebblewhite and Merrill, 2009; Mysterud et al., 2011; Singh et al., 2012). Although partial migration has been widely documented in ungulates in temperate systems, which are characterized by strong seasonal variations in resource distribution between summer and winter months, researchers are increasingly identifying evidence for partial migration in tropical ungulate systems, which experience different seasonal resource patterns for both grazing and browsing ungulates (Naidoo et al., 2012a; Gaidet and Lecomte, 2013). Since giraffe are large-bodied browsers in tropical systems, identifying intrapopulation variation in space use strategies of giraffe provides a unique system to evaluate hypotheses on the ecological mechanisms that give rise to partial migration. Although seasonal movements of giraffes have previously been described (Pellew, 1984a; Le Pendu and Ciofolo, 1999; Fennessy et al., 2003) this is the first time that migration between distinct seasonal ranges has been rigorously demonstrated. Our findings also represent the first systematic description of partial migration in giraffe. Furthermore, identifying this pattern of male-biased partial migration in a rapidly growing giraffe population highlights the need to better understand the role of spatiotemporal resource dynamics in driving movement

and female diets in Wankwar sector.

decisions and potentially population dynamics. Studies of temperate ungulates have suggested that not all seasonal variations in space use qualify as migration and that migrants must demonstrate stabilization of seasonal ranges (Gaudry et al., 2015). In our study, however, we noted consistent seasonal ranges for migratory individuals, with relatively few exploratory movements beyond these stable core areas. Furthermore, in migratory or partially migratory populations, migratory individuals move across both geographical space and ecological niche space (Peters et al., 2017). In this population of giraffe, in both collared individuals and population-level surveys, we observed seasonal shifts in space use between phytosociologically distinct sectors. Several hypotheses have been proposed to explain the emergence and maintenance of partial migration (Hebblewhite and Merrill, 2007; Chapman et al., 2011; Mysterud et al., 2011). A prominent hypothesis is the "competitive release hypothesis" wherein individuals move to reduce intraspecific competition in seasonally dynamic environments (Chapman et al., 2011). For this hypothesis to explain partial migration, some individuals in the population must be more vulnerable to competition and thus more likely to migrate to escape it. At high population densities, optimizing resource use by minimizing intraspecific competition may be a dominant driver of movement behaviors (Fryxell and Sinclair, 1988). The MFNP giraffe population is currently at its highest density relative to any point over the past 100 years (Brown et al., 2019), with foraging herds of giraffe exceeding 120 individuals in the Delta during the wet season (M. Brown pers. obs). Given the seasonal dynamics of vegetation in MFNP and the reduction in quantity of forage resources in the dry season, increased resource competition may instigate migratory behaviors. During this period, deciduous trees often lose their leaves and the foliar nutritional and phytochemical properties change, thereby altering the quality and quantity of forage available to browsers (Owen-Smith, 1994). In MFNP, however, plant species vary in their expressed degree of deciduousness with deciduous trees such as Acacia senegal losing a large percentage of their leaves (Omondi et al., 2016), semi-deciduous trees such as C. adansonii and Acacia sieberiana losing only some of their leaves (Shorrocks and Bates, 2015) and evergreen species such as H. abyssinica retaining leaves throughout the dry season. This phenology of leafing, coupled with relatively high browsing pressure in the Delta during the wet seasons, may result in a relative overall decrease in the availability of suitable forage during the dry season. Under these conditions, seasonal partial migration in spatiotemporally heterogenous environments may be viewed as a dynamic realization of ideal free distribution (McPeek and Holt, 1992; Cressman and Krivan, 2006). Lundberg (1987) describes a similar ecological scenario in which the persistence of partial migration in a population results from frequency dependent selection arising from individual decisions conditioned upon resource availability and density of conspecifics. In this way, migration may be perceived as a context dependent tactic in which giraffe respond to shifting cues of relative habitat quality as a function of conspecific density. In this giraffe population, we describe the uncommon scenario of two distinct resident types (Wankwar resident and Delta resident) and a migratory type that seasonally moves between the residential sectors, thereby providing a mechanism for achieving this shifting distribution in a temporally dynamic environment. Under theoretical ideal free distribution, fitness of residents and migrants must be equal over time. Although a more systematic demographic study will be required to assess the relative fitness consequences of various space use strategies in the MFNP giraffe population, the consistently high survival rate across both Delta and Wankwar sector, despite the spatiotemporal changes in giraffe density, lends support to the possibility of dynamic ideal distribution.

For the competitive release hypothesis to explain male-biased seasonal transitions, males must be impacted by competition differently than females or compete for different resources (Dobson, 1982). Among giraffe, this possibility is possible since giraffe exhibit sexually divergent foraging strategies and sexual niche partitioning of forage resources (Young and Isbell, 1991; Ginnet and Demment, 1997; O'Connor et al., 2015). Females typically forage on woody vegetation at lower heights than the larger males (Young and Isbell, 1991; O'Connor et al., 2015). We observed similar trends in the Delta sector of MFNP, with male giraffes consuming proportionally more C. adansonii and female giraffe consuming proportionally more Acacia sp. As a result of the differences in diet composition, the sexes may demonstrate different responses to the availability of the suite of woody vegetation species and the seasonally dynamic competition for forage resources. Furthermore, giraffe exhibit marked sexual size dimorphism with the larger males consequently having different energetic/nutritional requirements. Because of these asymmetries in energetic and nutritional requirements male giraffe may be more affected by the loss of forage quantity due to greater dry season deciduousness in the Delta (Main et al., 1996). Studies examining resource partitioning on the basis of body size in African browsers and grazers suggest that with increased body size, individuals may expand their diets to favor greater quantities of forage species at the expense of consuming forage species of lower nutritional quality (McNaughton and Georgiadis, 1986). In the MFNP ecosystem, this body-size effect may explain the male-biased dry season shift from the diverse, highly nutritious Acacia sp., C. adansonii, and H. abyssinica savannas of the Delta to the lower quality but more abundant forage of the broad-leaf P. kotschyi, S. kunthianum, and Terminalia spp. savanna/woodlands of Wankwar.

Since forage resource quality and quantity are linked to giraffe space use strategies and population dynamics, it is important to develop a deeper understanding of seasonal dynamics of forage availability. Giraffe seasonal migrations are associated with plant phenology in MFNP, with giraffe rapidly returning to the Delta sector following the onset of the rains and the subsequent green-up during the beginning of the short wet season. This seasonal space use pattern is consistent with the forage maturation hypothesis, which predicts that ungulate movement is influenced by selection for high quality forage resources (Hebblewhite et al., 2008). The quality of forage is typically greatest in newly developed plant tissue because of its high cell soluble content and relative lack of structural carbohydrates (Van Soest, 1982). In this way, as the plant resources respond to the commencement of the wet season with new growth, they are of high nutritional value to giraffe. The forage maturation hypothesis may also partially explain how the Delta sector attracts and sustains such large numbers of giraffe throughout the duration of the wet seasons (Fryxell, 1991; Hebblewhite et al., 2008). Other studies suggest that sustained browsing can keep woody vegetation in a chronic state of regrowth, maintaining high forage quality shoots and young leaves (Du Toit et al., 1990; Fornara and du Toit, 2007). For these "browsing lawns" to persist, however, there must be sufficient resources for plants to maintain regrowth processes (Cromsigt and Kuijper, 2011). Thus, in the wet seasons, the Acacia sp., H. abyssinica, and C. adansonii characteristic giraffe diets in the Delta sector, may be able to sustain intense browsing and still provide high quality forage for giraffe but in the dry season, these plants may lack sufficient resources for regrowth, shifting the distribution of giraffe to favor forage quantity (Fryxell, 1991).

In addition to competition for spatiotemporally varying forage resources, competition for mating opportunities may be a potential factor contributing to male giraffe movement strategies. Interestingly, available literature on ungulate migration describes largely female biased migration (Ohms et al., 2019), however the unique reproductive strategies of giraffe among ungulates may help explain this discrepancy. Since female giraffes are as seasonal, asynchronous breeders, and since they are only sexually receptive for a few days during a biweekly estrous cycling, male giraffe allocate much of their time in all seasons moving among scattered herds of females to assess their sexual receptivity (Pratt and Anderson, 1985; Bercovitch et al., 2006).

Male access to sexually receptive females is largely mediated through a dominance hierarchy wherein the largest, oldest bulls may displace subordinate male giraffe to monopolize breeding access to females in estrous (Pratt and Anderson, 1985). Previous studies suggest that competition for mates is a primary driver for subordinate juvenile male dispersal across a wide range of mammalian taxa (Dobson, 1982). Under this premise, younger subordinate adult male giraffe in MFNP may track seasonal shifts in female density to ranges where resident dominant bulls may not be able to monopolize access to the seasonal increases in female abundance. In seasonal environments where resource distribution varies predictably over space and time, these constant temporal shifts in both forage resources and mating resources, coupled with asymmetrical sexual competition among males of different sizes, may lead to male biased partial migration. It is important to note, however, that in MFNP, the adult sex ratio is skewed toward females in Wankwar in all seasons. Thus, males moving in any season to Wankwar would seem to be favored by this mechanism—not just movement there in the dry season.

The combination of individual-level GPS telemetry and population-level CRDMS models employed here can allow for key insights into ecological processes that give rise to changes in these movement strategies and the resulting consequences for population dynamics. For instance, researchers can parameterize CRDMS models with temporarily varying sector-specific population size and transition probability between geographic sectors and thereby test for changes through time in movement behavior. The MFNP giraffe population is growing rapidly (Brown et al., 2019), thus as the population continues to grow, these models provide a technique to evaluate density-dependent effects on transition probabilities and survival parameters across space and time. As the population density in the Delta grows larger, we might expect to see an increasing trend in transition probability to Wankwar during the resourcelimited dry season. Individual- based GPS telemetry can provide complementary fine-scaled information on changes in the timing of movement, duration of time spent in each range, and specific resources used in each seasonal range, with population-level surveys providing insights on the resulting demographic impacts. The combination of these two approaches has great potential for increasing our understanding of the ecological drivers of partial migration and better understanding the ecological mechanisms giving rise to intraspecific variation of movement strategies and the effects on population dynamics.

# DATA AVAILABILITY STATEMENT

The datasets generated for this study will not be made publicly available. The datasets are part of ongoing research which may generate additional publications.

# ETHICS STATEMENT

Research protocols were approved under Dartmouth Institutional Animal Care and Use Committee (IACUC) guidelines with all fieldwork conducted under Uganda Wildlife Authority (UWA) approved research permit and Uganda National Council of Science and Technology UNCST permit.

# AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

# FUNDING

The majority of funding for fieldwork was provided through a partnership with the Giraffe Conservation Foundation (GCF) a conservation non-governmental organization based out of Windhoek, Namibia. Individual donors that contributed to GCF in support of the Uganda programme are listed in the acknowledgments. GCF supported the project logistically as well, providing access to vehicles and partners in country. Dartmouth College provided graduate fellowship stipends and discretionary research funds through the Department of Biological Sciences. Dartmouth College Library provided additional funds for open access publication fees.

## ACKNOWLEDGMENTS

We extend a special thanks to the Giraffe Conservation Foundation, Julian and Steph Fennessy for their incredible support and partnership. Funding and support for all fieldwork and collaring was provided through the Giraffe Conservation Foundation and partners Cleveland Metroparks, Dallas Zoo, Fort Wayne Childrens Zoo, Hogle Zoo, Maltz Family Foundation, Total Uganda, Wildlife Conservation Alliance and WoodTiger Fund. We would also like to thank the dedicated ranger, wildlife veterinarians, research and monitoring crew and administrative staff of the Ugandan Wildlife Authority for facilitating this research. We also thank Jim Nichols and Jim Hines for their invaluable advice and guidance in developing multi-state MARK models for the analysis of population level surveys. S. Mutebi and the staff at Makerere University Herbarium were instrumental in providing validation for plant identification.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00524/full#supplementary-material

#### REFERENCES


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Brown and Bolger. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Ontogenetic Variation in Movements and Depth Use, and Evidence of Partial Migration in a Benthopelagic Elasmobranch

James Thorburn<sup>1</sup> \*, Francis Neat <sup>2</sup> , Ian Burrett <sup>3</sup> , Lea-Anne Henry <sup>4</sup> , David M. Bailey <sup>5</sup> , Cath S. Jones 6† and Les R. Noble7†

<sup>1</sup> Coastal Resource Management Group, Scottish Oceans Institute, University of St. Andrews, St. Andrews, United Kingdom, <sup>2</sup> World Maritime University, Malmö, Sweden, <sup>3</sup> Scottish Sea Angling Conservation Network, Stranraer, United Kingdom, <sup>4</sup> School of GeoSciences, University of Edinburgh, Edinburgh, United Kingdom, <sup>5</sup> Institute of Biodiversity Animal Health and Comparative Medicine, University of Glasgow, Glasgow, United Kingdom, <sup>6</sup> School of Biological Sciences, College of Life Sciences and Medicine, University of Aberdeen, Aberdeen, United Kingdom, <sup>7</sup> Faculty of Biosciences and Aquaculture, Nord University, Bodø, Norway

#### Edited by:

Yolanda E. Morbey, University of Western Ontario, Canada

#### Reviewed by: Paddy Walker,

Dutch Elasmobranch Society, Netherlands Jose A. Masero, University of Extremadura, Spain

> \*Correspondence: James Thorburn jat21@st-andrews.ac.uk

> > †Joint last authors

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 30 May 2019 Accepted: 06 September 2019 Published: 25 September 2019

#### Citation:

Thorburn J, Neat F, Burrett I, Henry L-A, Bailey DM, Jones CS and Noble LR (2019) Ontogenetic Variation in Movements and Depth Use, and Evidence of Partial Migration in a Benthopelagic Elasmobranch. Front. Ecol. Evol. 7:353. doi: 10.3389/fevo.2019.00353

Tope (Galeorhinus galeus) is a highly mobile elasmobranch in the temperate to subtropical northeast Atlantic. It is highly migratory and has been shown to display complex movement patterns, such as partial migration, in the southern hemisphere. In the northeast Atlantic, previous mark-recapture studies have struggled to identify movement patterns and the species behavior is poorly described, yet identification of migratory behaviors and habitats of importance for the species is of paramount importance for effective management. Here, we combined fisheries independent survey data with markrecapture (MR) data to investigate the distribution of different age classes of tope across the northeast Atlantic. We further investigated depth use in detail with archival electronic tags and a pop-up satellite archival tag (PSAT). We suggest previous studies struggling to find consistent movement patterns using MR data were confounded by a combination of site fidelity, partial migration by females, and increasing depth and home range of juveniles. Survey and MR data showed immature tope <40 cm were caught exclusively in continental shelf waters <45 m deep, showing a significant relationship between habitat depth and total length. Immature individuals seemed to remain on the continental shelf, while mature tope of both genders were caught in both shelf and offshore waters. This use of deeper water habitats by mature tope was further supported by archival tags, which indicated individuals use both shallow (<200 m depth) and deep-water habitats, diving to depths of 826 m; the deepest record for this species. The PSAT tag tracked the horizontal movements of an adult male, which confirmed utilization of both shallow inshore and deep offshore habitats. Most tope remained within 500 km of their tagging site, although some mature females had a larger, more southerly range, including connectivity with the Mediterranean. This study clearly demonstrates the highly migratory habits of tope, and suggests larger individuals divide their time between shallow and deep-water habitats. It shows the northeast Atlantic tope population should benefit from consistent management throughout its range.

Keywords: tope, school shark, depth range, archival tags, migration, site fidelity

# INTRODUCTION

Many elasmobranchs (sharks, skates, and rays) are highly mobile, ranging across regional seas (Mucientes et al., 2009) with some species undertaking trans-ocean movements (Templeman, 1976; Gore et al., 2008). They occupy a wide depth range, from surface waters to 3,000 m (Priede et al., 2006), with species displaying daily, yearly and life-long variations (e.g., Bres, 1993; West and Stevens, 2001; Sims et al., 2003; Andrews et al., 2009; Neat et al., 2014; Thorburn et al., 2015). These factors, together with the long lifespans of many elasmobranchs, present an opportunity for an individual to use many different geographic locations and habitats during its lifetime (Bres, 1993). Consequently, spatial management plans to conserve elasmobranchs across different habitats and sociopolitical regimes, including areas beyond national jurisdiction (the High Seas), is challenging. Hence, effective spatial management for mobile species may only be possible in situations where they display site fidelity to habitats critical for life history events, e.g., reproduction, and where such aggregations are at risk to anthropogenic pressures. It is crucial, therefore, to better understand and define behaviors that determine elasmobranch spatial ecology and the habitats and regions with which they are associated.

There is substantial evidence that elasmobranchs often remain within a home range (Carrier et al., 2012), i.e., a behaviorally confined geographic area, which may justify an element of spatial management in species conservation. Within this home range, they may exhibit daily, seasonal, and ontogenetic changes in depth use (Grubbs, 2010; Afonso and Hazin, 2015; Thorburn et al., 2015) and populations can segregate into sub-units based on age and sex (Bres, 1993; Wearmouth and Sims, 2008; Thorburn et al., 2018). This can give rise to complex patterns of spatial ecology at a population level, for example, juveniles will often remain within specific pupping/nursery grounds within the population's home range (Gruber et al., 1988). From here, they gradually increase their spatial use as they grow (Gruber et al., 1988) including an increase in their depth range (Grubbs, 2010; Afonso and Hazin, 2015). Many species also exhibit seasonal variation in depth, often in the form of a move to deeper, offshore waters during colder months, returning to shallower waters in the spring and summer (Andrews et al., 2009; Queiroz et al., 2010; Thorburn et al., 2015). These seasonal movements are mostly attributed to the reproductive cycle (Hurst et al., 1999), temperature requirements, or dietary needs (Bres, 1993; Wearmouth and Sims, 2010). Some species also display a dietary shift as they move over the shelf edge to deeper waters, changing foraging strategies to presumably utilize the most locally abundant prey species (Queiroz et al., 2010).

Tope, or soupfin shark, (Galeorhinus galeus), classified as Vulnerable by the IUCN, are generally considered a benthopelagic species, meaning that they occupy most of the water column and are therefore potentially vulnerable to multiple pressures from human activities from the surface to the seafloor. They occupy a wide temperature range of 8.1◦C−27◦C (West and Stevens, 2001; Cuevas et al., 2014; Rogers et al., 2017) and are distributed throughout the northeast and southeast Pacific, the northeast and southern Atlantic, Mediterranean Sea, off southern Australian and New Zealand waters (Compagno, 1984; Walker, 1999). Parturition occurs in shallow coastal waters (Hurst et al., 1999; McAllister et al., 2015) where juveniles remain within the confines of their nursery grounds for up to 2 years before expanding their home range (McAllister et al., 2015). They are capable of longdistance migratory behavior, with the greatest migration distance estimated to be 4,940 km (Hurst et al., 1999). Tope populations segregate by age and sex class sub-units (Lucifora et al., 2004), which display different movements and habitat choices (Hurst et al., 1999; Walker, 1999; Lucifora et al., 2004).

In the northeast Atlantic, tope occur from Iceland to the Azores and Canary Islands and are also found in the Mediterranean (Capapé et al., 2005). Conventional markrecapture (MR) data have shown some transboundary movement between these areas including some movement across oceanic waters (Holden and Horrod, 1979; Sutcliffe, 1994; Little, 1995, 1998; Fitzmaurice et al., 2003; Capapé et al., 2005), but have been unable to identify clear seasonal patterns (Stevens, 1990). Broadly they suggest that mature females display an annual southerly migration to pupping areas in the south (Holden and Horrod, 1979; Stevens, 1990; Sutcliffe, 1994; Little, 1998; Fitzmaurice et al., 2003), utilizing different grounds off Portugal and the Canary Islands (Munoz-Chapuli, 1984). However, there does appear to be variation in migratory behavior, as some females appear to remain within the proximity of their northern European tagging sites for most of the year (Holden and Horrod, 1979; Sutcliffe, 1994; Little, 1995) while others undertake offshore movements over the winter including movements to these more southerly areas (Wheeler, 1969; Stevens, 1990). It has been suggested that mature female tope in the Mediterranean display partial migrations (Capapé et al., 2005), some females remaining within a limited home range to pup, while others undertake longer migrations to pupping sites outside the home range (Capapé et al., 2005); this would be similar to behavior observed in Australasia (McMillan et al., 2018b) linked to discrete pupping grounds (McMillan et al., 2018a,b). It was further suggested that females displaying residential behavior may reproduce every year because they can allocate more energy to reproduction, whereas those following a more demanding migratory strategy are only able to reproduce every second year (Capapé et al., 2005).

On the basis of these studies, tope clearly has a complex spatial ecology, but there is no clear consensus on movement behaviors. The aim of this study was to consolidate data from a variety of sources in an effort to better define the movements of tope within the northeast Atlantic, and to assess evidence for site fidelity, propensity to aggregate, preferential habitat use, and investigate possible relationships between size, sex, and life history stage. A combination of survey data and mark and recapture data were pooled to understand the species' movements in this area for which there is currently limited data. We also deployed multiple archival electronic tags and 1 pop off satellite archival tag (PSAT) on a tope captured off southern Scotland.

# MATERIALS AND METHODS

#### Data Collection

#### Mark-Recapture and Survey Data

Mark and recapture (MR) data were combined from the Scottish Shark Tagging Program (SSTP), the Glasgow Museum Tagging Program (GMTP), the UK Shark Tagging Program (UKSTP), and Holden and Horrod (1979). All MR data consisted of conventional ID mark and recapture data. In most cases, but not all, data consisted of date of capture, location of capture in either longitude and latitude format or verbal description, stretched total length (TL), weight either measured or estimated from TL, and gender. Recaptures were reported by anglers and commercial vessels. Fisheries-independent survey data for tope up to and including 2014 were downloaded from the International Council for the Exploration for the Sea (ICES) Database of Trawl Surveys (DATRAS) portal (https://datras.ices.dk/), with catch location assigned as trawl retrieval latitude and longitude. Data were filtered so only records with length measurements (TL) were used (NS-IBTS, BTS, and EVHOE). These data were added to the MR data to create a Presence Dataset. For all data, individuals were assigned a maturity status, either mature or immature, based on TL at time of capture. Males were deemed to be mature at TL ≥ 126 cm (Capapé et al., 2005) and females with TL ≥ 130 cm (Lucifora et al., 2004). In instances where no TL was recorded for the recapture, length was estimated from weight at recapture using length-weight charts developed by both the UK Shark Tagging Programme (www.ukstp.co.uk) and the Scottish Shark Tagging Program (www.tagsharks.com). If no weight or length was recorded at recapture, TL was either measured or estimated from weight at time of tagging if the recapture was within 1 year. Records of recapture were used to form a separate Recapture Dataset. Records where the recapture did not match the tagging record (i.e., gender change, or unrealistic length differences) were removed. The straight-line distance (Distance) and days at liberty (Freedom) between tagging and recapture event were calculated for each tag number, and dates were assigned a day of year (1–365). The exceptions to this were recaptures in the Mediterranean, for these, minimum wet distance (avoiding land) was calculated rather than straight-line. In total, 2,043 records, both tag and recapture, were collated. Of these, 138 recapture records were useable [53 males: 13 immature (84–126 cm), 40 mature (126.1–168 cm); 85 females: 40 immature (86–130 cm), 45 mature (130.1–180 cm)], having both a location of capture and recapture and at least one TL record per individual. Filtered DATRAS data produced 457 records between 1984 and 2011.

#### Archival and PSAT Tagging

Tope were caught using individual baited hook and line in Luce Bay, southwest Scotland (54.7◦N, 4.7◦W; **Figure 4**). Tags were deployed over three periods; June 2012 (Archival tags: Star Oddi centi-TD, n = 5), September 2014 (Archival tags: Lotek LAT2900- XW, n = 10), and October 2015 (PSAT tag: Wildlife computer MiniPAT, n = 1). Total length (TL) and gender were recorded. All tags were pre-started and fitted externally. Star Oddi tags were mounted on a silicone pad and anchored through the base of the first dorsal fin using two stainless steel pins and Peterson disks. Lotek tags, fitted with an external float jacket, were attached via 200lb nylon with a rubber tube casing and inserted through the dorsal spine at the base of the first dorsal fin using a sterilized stainless-steel needle at a minimum of 3 cm from the trailing edge. Once through, the nylon was crimped to itself, creating a loop. The PSAT tag was attached intramuscularly using a titanium plate inserted into the dorsal musculature next to the first dorsal fin. The plate was pushed in place using a sterilized stainless-steel applicator. A 200lb monofilament leader of 5 cm was attached the tag to the plate. The PSAT tag was preprogrammed to release after 180 days. All tags were marked with a specific ID number. Star Oddi tags were set to record depth and temperature every 5 min. Lotek tags recorded depth and temperature every 10 min, their wet/dry state every 40 min, and light levels every 2 min, while the PSAT was set to record depth every 5 min with temperature summarized every 24 h (**Table 2**). Tagged animals were released at their capture site within Luce Bay. Tags were marked with contact details and notice of a cash reward for their return.

#### Data Analysis

#### Mark-Recapture and Survey Data **Site associations**

A Generalized Additive Model (GAM) was used to investigate the relationship between Distance, day of year, gender and maturity status in R (R Core Team, 2013) using the MGCV package (Wood, 2001). Distance was log-transformed to reduce the impact of outliers. Day of year was smoothed using a cyclic penalized cubic regression spline. Model choice was based on the Akaike's Information Criterion (AIC).

#### **Ontogenetic and sex difference in ranges**

Using the Recapture Dataset, the relationship between Distance traveled, gender, and size was investigated using linear models in R 3.1.3 (R Core Team, 2013) with distance modeled as a function of TL and gender. As we were interested in the furthest Distance traveled per age class, length data from MR recaptures for each sex were divided into 5 cm length classes. Distance was taken to be the maximum distance traveled by an individual from each length class.

#### **Immature tope spatial use**

To provide a geographic representation on the size distribution of tope, the Presence Dataset was interpolated using kriging methods with a spherical model in ArcGIS 10.2. Kriging interpolation is better suited to clustered data than other methods as it helps to compensate for the effect of non-uniform effort on the data. Kriging was based on the minimum TL value recorded at locations where there were multiple individuals caught in order to identify areas important to smaller tope. Data were grouped into 10 cm length classes for visualization, starting at 26 cm based on the record for the smallest TL recorded. To investigate the maximum water depth use by immature tope, water depth for each presence data point was extrapolated from GEBCO bathymetric data extracted at a 1-min arc cell size in ArcGIS 10.2 based on the latitude and longitude of that point. Data were split into TL classes of 5 cm increments up to 130 cm, and the

FIGURE 1 | Distribution of immature and mature tope with gender symbolized from presence data, combining mark-recapture, and International Bottom Trawl Survey (IBTS) data. (Left) Distribution of immature tope (m < 126 cm, f < 130 cm). (Right) Distribution of mature tope (m ≥ 126 cm, f ≥ 130 cm).

maximum water depth for each length class was recorded from the deepest water record within each size range. This method was undertaken to look at maximum water depth occupied by each size class. Relationships between TL and maximum depth were explored using linear models (LM) with depth as a function of TL and gender in R 3.1.3 (R Core Team, 2013).

#### Archival and PSAT Tag

Depth was smoothed to 30-min averages and wavelet transformation analysis (Rösch and Schmidbauer, 2014) used to look for cyclic patterns. Wavelet analysis was undertaken in the R package WaveletComp (Rösch and Schmidbauer, 2014) on the smoothed depth data obtained from all archival and PSAT tags using the following parameters: loess span = 0.1, dt = 0.5, dj = 1/250, lowerPeriod = 8, upperPeriod = 256 (30 min average data were used, so to define the range of periods in time steps the analysis searches 8 steps = 4 h and 256 = 128 h), n.sim (number of simulations) = 100. See Rösch and Schmidbauer (2014) for a full description of these parameters. Geolocation was not undertaken on archival tags due to large amounts of uncertainty around the geographical position of the end of the tag record and the lack of temperature data in two of the tags. A Maximum Likelihood Path was recreated for the PSAT tag with geographic positions being estimated using Wildlife Computers' own state-space GPE3 model, which produces maximum likelihood positions with 50, 95, and 99% confidence estimates, while latitude and longitude are estimated using light levels (dawn, dusk, and noon) that are further refined using sea surface temperature (SST) and bathymetry data. The GPE3 model was run using a swimming speed of 1 ms-1 based on previous PSAT tagging research on G. galeus in the southern hemisphere (McMillan et al., 2018b).

### RESULTS

#### Presence Data

Male and Female immature tope were found throughout the north-eastern Atlantic continental shelf (**Figure 1**, left panel). Besides shelf environments, mature individuals of both genders were also found in oceanic waters, with mature females inhabiting more southerly waters around the Azores and the Canary Islands (**Figure 1**, right panel), while mature males were more regularly found in northerly waters off northwest Scotland. The length distribution map (**Figure 2**) suggests that tope < 46 cm long were found in coastal waters. Juvenile tope (males < 126 cm, females < 130 cm) were absent in oceanic waters and from more southerly latitudes and tope < 40 cm were found in coastal waters in the southern North Sea and off the west coast of England and Wales. Most tope were captured in ≤50 m (**Figure 3**, left panel), however, there was a significant linear relationship between maximum environmental water depth and total length, with larger tope being found in increasing water depths (**Figure 3**, right panel). Gender did not have a significant effect on maximum depth use.

## Mark-Recapture

Recaptures of immature tope occurred on average 314.85 km away from the original tagging site, with a difference of 103 km between the average distances traveled by males and females (F = 366.4, M = 263.3, **Table 1**). The average distance traveled by mature individuals showed greater variation, with mature males recaptured on average 342.8 km (maximum 2,168 km over 2,000 days) from their tagging site, similar to average distances seen in immature males; females were captured on average 799.1 km (maximum 3,900 km over 1,960 days) away, over double that shown by immature females (**Table 1**). Most

represents smallest sized (based on total length) animal predicted to occur in that area. Smoothing was performed using kriging methods on minimum total length at each site using ArcGIS 10.2.

males, both immature and mature, were recaptured within the confines of the continental shelf, except for one mature male recaptured of the coast of Iceland after being tagged in Scotland and one being recaptured in the Azores after being tagged in Scotland (**Figure 4**, left panel). Immature females were, similarly to males, recaptured on the continental shelf. However, there were several examples of mature females, after being tagged around the coast of the UK, being recaptured in the Azores, the Canary Islands and in the Mediterranean Sea (**Figure 4**, right panel). Maximum range traveled by female tope significantly increased with body length (**Figure 5**, **Table 1**), but females < 95 cm did not have ranges larger than 500 km (**Figure 5**). There was no significant relationship between TL and maximum range traveled by males, and maximum distance traveled varied considerably as TL increased (**Figure 5**), however, the minimum male recapture length of 84 cm prevented investigation of a range of smaller animals. General Additive Modeling (GAM) of recapture data showed that the day of year of recapture had the most significant influence on distance from tagging site, accounting for 30% of the deviation seen in distance traveled; gender and TL did not improve the model (**Figure 6**). The GAM shows a general population trend of all size and sex classes moving away from their tagging site over winter and spring, returning to an area near their original tagging site (<50 km) during summer and autumn (**Figure 6**).

TABLE 1 | Summary of distance traveled by tope of each sex and maturity state from all tagging data showing median, mean, and maximum distance traveled in kilometers as well as number of records.


FIGURE 4 | Recaptures of tope tagged around the UK from the Scottish Shark, Glasgow Museums and UK Shark Tagging Programmes with a straight-line connector. Left, male; Right, female.

0.2809. Gray ribbon represents standard error.



\*Data received: the data types that the tag was programmed to record; T, Temperature; D, Depth; L, Light. Time: length of recording interval (minutes) for each parameter in the Data received column. For Tag 153233, only depth was recorded as a series, temperatures were binned over 24 h and light levels were used for on-board geo-location estimates but not recorded as a series.

#### Archival and PSAT

Overall, 5 archival tags were recovered, 1 Star oddi and 4 Lotek. Depth data between 150 and 161 days were recovered from 3 of these (1 Star oddi and 2 Lotek), temperature records were recoverable from only the star oddi tag (**Table 2**). Data were also recovered from PSAT tag via satellite transmission after the full 180 days deployment time (**Table 2**). The Star oddi tag was recaptured off the Portuguese coast (un-disclosed location) in March 2013, 129 days after the end of the tag record. The 4 lotek tags were all found on the western coast of the UK by beachcombers (minimum 585 days after the end of tag record). Tag 2089 remained in water shallower than 200 m until the 12th October, when it started occupying waters up to 300 m deep during the day. During this time, there was some evidence of standard and crepuscular diurnal migrations (**Figure 7**). However, diel patterns were generally weak compared to other tags. On the 20th January, 2089 moved to deeper water (max depth 542.5), upon which a strong diel pattern was initiated; occupying waters <100 m, including surface waters (<5 m deep), during hours of darkness, moving to depths between 400 and 664.5 (max depth) during daylight (**Figure 7**). Tag 2089 was found on the Isle of Arran off western Scotland 936 days after the last data record. Tag 153233 detached after the programmed 180 days, on the 11th April 2016 at latitude 51.59◦N, longitude 11.86◦W (**Figure 7**). It remained between surface waters and 300 m depth until January. Wavelet analysis did not detect

strong cyclic patterns during this time, but this may be due to incomplete data transmission creating gaps in the time series. On the 3rd January, 153233 increased its vertical range and occupied the water column from the surface to 644 m over the course of 4 days. From the 7th to 13th January, it moved back to a shallower depth range, between the surface and 175 m. On the 14th January, 153233 occupied a wider depth range again, from surface waters to 654.5 (max depth), until the end of the tag record. While the time series was broken, enough data were recovered for a strong 24 h cyclic pattern to be detected during the times when 153233 moved to deeper water (**Figure 7**). Both standard and crepuscular diurnal migration were observed during these periods (**Figure 7**). The most likely pathway recreation for tag 153233 shows the tope leaving southwest Scotland, traveling west passing Northern Ireland toward the shelf edge (**Figure 8**). In December, it moved over the edge of the shelf, continuing to head west toward Rockall Bank (**Figure 8**). In January, its westward trajectory turned southwards, passing between the continental shelf and Porcupine Bank off of Ireland (**Figure 8**). The distance traveled was reduced in March, with the tope remaining in the channel to the east of Porcupine Bank (**Figure 8**). The tag detached after 180 days north of the deep Porcupine Seabight (**Figure 8**).

## DISCUSSION

Tope in the northeast Atlantic are currently considered one large population dispersing throughout the region (Holden and Horrod, 1979; Stevens, 1990; Sutcliffe, 1994; Little, 1998; Fitzmaurice et al., 2003). We found no evidence to contradict this; the tope in this study showed high migration potential across the northeast Atlantic, even into the Mediterranean as far as Sicily (first reported in Colloca et al., 2019). The extent of migration appears to relate to age and gender, with adults having more latitudinal variation, as seen in other elasmobranch species (Olsen, 1954; Gruber et al., 1988; Speed et al., 2010), and females being found in more southerly waters and males in more northerly waters. The MR data also suggested that mature tope may have a greater latitudinal range than immature tope. Physical differences, dietary and habitat requirements,

FIGURE 8 | Main image: most likely pathway for 153233 recreated using Wildlife Computers GPE3 model. 50, 95, and 99% confidence limits are also shown. Insert: depth data for 153233. In both panels, months are color coded as per the legend.

reduction in resource competition, females avoiding mating, and the reduction of pup mortality can often result in different age and sex classes utilizing different geographic areas and depths (Klimley, 1987; Economakis and Lobel, 1998; Pratt and Carrier, 2001; Wearmouth and Sims, 2010). The variation in latitudinal extremes may be a product of females using warmer waters (Hurst et al., 1999; Robbins, 2007) to decrease embryonic development time (Economakis and Lobel, 1998) and reduce pup mortality (Hanchet, 1991), while males may use cooler water to optimize sperm production (Wearmouth and Sims, 2008). However, the presence of female tope around the Shetland Isles, and a male being recaptured in the Azores suggests temperature alone may not impact gender distribution greatly.

Globally, there is evidence that tope use shallow coastal waters, such as bays and estuaries, as nursery grounds (Hurst et al., 1999; McAllister et al., 2015; Bovcon et al., 2018). This was supported by analysis of the Presence data, where tope < 40 cm were only found in coastal regions and then seemed to expand their range, including depth, when they reached 50 cm. This is reminiscent of juvenile behavior in the southern hemisphere, where tope under 2 years old remain within coastal nursery areas before expanding their home range (McAllister et al., 2015).

It appears that the majority of the tope population display similar movements based around a cyclic annual migration. It should be noted though that this is based on the Recapture data which was only available for tope 84 cm or larger. Due to the apparent depth limitations of juvenile tope shown by the Presence data, it is likely smaller tope have more spatially restricted home ranges, as observed in other elasmobranch species (Kinney and Simpfendorfer, 2009). The extent of this annual migration could be the basis for determining the population's core home range. This is further strengthened by the average distance mature males are recaptured from their tagging location being similar to that of immature tope of both genders. A caveat here is that data were only available for 13 immature male tope, so the movements of immature males could not be fully explored.

While most of the recaptured males, both mature and immature, were in shelf waters, the maximum recapture distance for mature males is significantly higher than immature tope of either gender. This is caused by a recapture in the Azores, one of two recapture records which show mature males do make wider movements and cross oceanic water. This is congruent with the findings from the archival data that mature males move into deeper water over the shelf edge. Mature female tope appear to display differing intra-gender migration strategies. While some mature females were recaptured at similar ranges to mature males and immature individuals, on the continental shelf within 500 km of the tagging site. Others were recaptured in southerly areas around the Azores and the Canary Islands, as has been observed previously (Sutcliffe, 1994; Little, 1995), producing a high average distance between capture and recapture sites for mature females.

This variation in movement distance displayed by females may be caused by gravid and non-gravid individuals undertaking discrete movements, as has been observed in other elasmobranch species in these two states (Howey-Jordan et al., 2013; Papastamatiou et al., 2013). That non-gravid females remain within the population's core home range while gravid females migrate to southerly nursery grounds has been suggested before (Sutcliffe, 1994; Little, 1995; Capapé et al., 2005). Small tope would then leave nursery grounds and migrate northwards as they grow (Holden and Horrod, 1979; Sutcliffe, 1994; Little, 1995). However, if this were the case, we would not expect to find evidence of nursery areas in more northerly regions as we did in this study, with tope < 40 cm being caught in coastal waters in the southern North Sea and elsewhere around the UK. This suggests that pupping occurs in both southerly and more northerly areas, and that gravidness is not responsible for the difference observed in mature female movements.

Mature females of other species of elasmobranch have been shown to display partial migration in relation to nursery sites, whereby some individuals make use of nursery grounds within the populations home range, while others make extended movements to more distant nursery grounds (Mourier and Planes, 2013; Papastamatiou et al., 2013). A similar behavior has been observed in southern hemisphere tope, where some females used pupping grounds within core home ranges, while others undertook wider movements to discrete pupping areas further afield (McMillan et al., 2018a,b).

Given the occurrence of northerly pupping grounds within the apparent core home range of the northeast Atlantic, it is likely that this strategy is undertaken by female tope in this region. We suggest concurrent use of pupping grounds in both the southerly and more northerly areas causes partial migrations in northeast Atlantic female tope, with some undertaking longer migrations to southern pupping grounds, while others remain within the populations apparent core home range, using more local pupping sites. There is some previous evidence of northeast Atlantic female tope displaying partial migration in the Mediterranean (Capapé et al., 2005), where two groups of gravid females were observed; one residential, the other migratory. The residential females were able to undergo vitellogenesis and gestation concurrently, shortening the reproductive cycle to 1 year, while females in the other group were only able to undertake one of these processes at a time due to a decreased energy budget for reproduction, doubling the length of the reproductive cycle. This variation in reproductive strategies may explain conflicting reports on the length of the reproductive cycle in female tope with annual (Ripley, 1946; Capapé et al., 2005), biennial (Olsen, 1954), and triennial (Lucifora et al., 2004) cycles all reported. The determination of which nursery grounds, and therefore reproductive strategy, are used by the female may be determined by philopatric behavior (well-documented in elasmobranchs Pratt and Carrier, 2001; Feldheim et al., 2002; DiBattista et al., 2008; Jorgensen et al., 2010), which has been shown to cause dispersal to multiple nursery sites (Hueter et al., 2005).

Movement in relation to nursery grounds does not appear to be the only migratory driver in northeast Atlantic tope, as there appears an annual cycle; with all age and sex classes being recaptured closer to their tagging sites during summer months and further away over winter. Site fidelity, which would cause an annual movement similar to that observed, has also been shown in elasmobranchs (Carrier et al., 2004). Mating has been shown to occur concurrently at different sites around the UK in June, including Luce Bay in Scotland to the Channel Islands (personal observation). Fidelity to these different mating grounds would cause an annual cyclic pattern, similar to that observed in this study, with adults being captured closer to a mating site over mating months (summer) and further away during winter. As all mating grounds are not in one location this suggests that tope annual migrations would not display a "north–south" pattern, explaining why previous studies using mark and recapture data struggled to find consistent patterns, as seasonal movements are in relation to tagging site rather than latitudinal position. The recapture of a single male in the Azores is not enough to infer partial migration in mature male tope, it is possible that after pupping, mature females are ready to mate and there are southerly mating grounds near pupping sites to which mature males travel, this requires further evidence to substantiate but would not explain the movement of the male tope to Iceland.

Movements may, as is common in elasmobranchs, be associated with prey migrations (Hussey et al., 2009; Papastamatiou et al., 2013), which would impact the movement of both mature and immature tope. The diet of tope undergoes ontogenetic shifts, with immature tope consuming more benthic invertebrates than adults, and adults having a higher proportion of cephalopods in their diet (Lucifora et al., 2006). This difference would allow immature tope to maintain smaller home ranges, explaining the shallow depth range of small tope, while mature individuals may have to undertake wider migrations to follow prey such as Atlantic mackerel, Scomber scombrus. The mackerel stock in the northeast Atlantic comprises three spawning units: southern, western and North Sea (Jansen et al., 2013). The western and southern components undertake seasonal north/south movements, with the western component traveling west off Ireland during the spring and summer and the southern spawning component moving up from the Bay of Biscay, along the English Channel and Irish Sea. Both components return to the Bay of Biscay over winter (Lockwood, 1978; Jansen et al., 2013). This may at least partly account for the movement observed by tag 153233, the mature male moving in relation to the western mackerel movements off the shelf edge. The archival depth data from the electronic tags also suggest a seasonal change in diet in three of the four tagged tope. Prey species migrations may also be accountable for the recaptures of mature male tope observed in Iceland and the Azores.

All tope tagged with electronic tags displayed diurnal migrations to some degree while in shallower waters, suggesting exploitation of similar resources. This was mostly in the form of standard or crepuscular vertical migration, consistent with vertical behaviors observed in other areas (West and Stevens, 2001; Cuevas et al., 2014; Rogers et al., 2017). The strength of the cyclic pattern varied temporally between individuals, which may reflect experience of different thermal regimes. In other species of elasmobranch, patterns of diel behavior in coastal areas has been linked to thermal stratification of the water column, with well-mixed waters suppressing regular cyclic movements and stratified waters promoting diurnal movements (Queiroz et al., 2010). The changes in depth behavior observed in the archival and PSAT tags suggest that tope undertake at least three different foraging strategies in deeper waters in the northeast Atlantic determined by the local abundance of prey species. The move to deep water was in winter months, either November (5159) or January (2089 and 153233). Once in the deep-water habitat, tope 5159 remained at depth (below 500 m) and diurnal movement broke down, suggesting that 5159 may have been foraging in the deep scattering layer. Tope 2089 and 153233 however, undertook large vertical movements, traveling from bathyal waters as deep as 600 m to surface waters in just over a 24-h cycle. This suggests that tracking of vertically migrating prey species, such as squid, similar to behavior observed in tope in the southern hemisphere (West and Stevens, 2001). It appears that male tope, at least, may change their foraging behaviors even in oceanic waters. Tope 153223 only appeared to utilize the deep-water habitat in January, 2 weeks after it crossed the shelf edge to a deeper water environment. This suggests that for the first 2 weeks of occupying oceanic waters, it remained foraging in waters shallower than 300 m before switching to a different foraging strategy, probably involving a different species in deeper water, which given the geographic variation observed in tope diet compositions is highly probable (Ellis et al., 1996; West and Stevens, 2001; Morato et al., 2003; Lucifora et al., 2006; Torres et al., 2014). This variation in oceanic foraging strategies is similar to that observed in blue sharks in oceanic waters (Queiroz et al., 2010). There was also proof that some mature tope are able to maintain foraging in shallow waters over winter, as tope 2036 remained in waters <100 m over winter. Without geolocational estimates, it is not possible to say whether this tope moved over the shelf edge but remained in shallow waters, or whether it remained in shelf waters.

The results presented here provide the most comprehensive overview of tope movements and distributions in the northeast Atlantic, an area from which such data have been lacking. The effectiveness of area-based management strategies such as Marine Protected Areas (MPAs) for mobile species is not always clear (Bonfil, 1999; Hilborn et al., 2004). For elasmobranchs, many of which are strong k-strategists (Stevens et al., 2000; Ellis et al., 2005; Camhi et al., 2009), it has been suggested that MPAs are most effective for younger age classes of mobile species with limited home ranges (Heupel and Simpfendorfer, 2005; Kinney and Simpfendorfer, 2009). However, reliance on management of tope nursery areas in Australia in the 1950s failed; one of the earliest attempts to use spatial management of an elasmobranch (Stevens, 2002). Despite spatial management of nursery areas, tope stocks continued to decline severely as juveniles moving out of the nursery sites, which they are now known to do after 2 years (McAllister et al., 2015), were immediately vulnerable to exploitation before they had a chance to reproduce (Kinney and Simpfendorfer, 2009). This attests that management cannot depend on spatial protection of nursery areas alone. Other areas where spatial management may be considered are those that meet two criteria: (1) The population exhibits site fidelity to the area, ensuring the long-term effectiveness of any protection. (2) There are conflicts that pose a disproportionality high risk to the population in that area. As demonstrated in this study, the coastal environment provides many important habitats for tope which they utilize for critical life history events such as nursery grounds and mating, and tope do display site fidelity to some of these areas. Coastal development projects, such as renewable energy, have the potential to create ongoing impacts at these sites. This potential for conflict and impact should be thoroughly investigated to ascertain the potential effect they could have on northeast Atlantic tope. Spatial management at such sites could be used as part of a wider management strategy for the species to benefit multiple life stages. If spatial management is to be used, it is essential that it is based on good science to prevent the creation of "paper parks" and this approach should be used in conjunction with wider management measures. Due to the limited depth and geographic range of juvenile tope, nursery grounds in UK waters do already, to some extent, naturally benefit from a degree of de facto fisheries protection. However, due to the evidence from Australia (Kinney and Simpfendorfer, 2009), it is apparent that the protection of all age classes is vital to ensure the population's continuity (Kinney and Simpfendorfer, 2009). Mature tope in UK waters are protected by either a prohibition on landing (Scotland) or a low Total Allowable Catch (rest of UK), however, their capacity for wide ranging movements take them beyond these protections into other countries or international jurisdiction, placing them at risk of fisheries interactions. This highlights the need for species management to be conducted via an international unified management plan, using tools such as the Convention for Migratory Species, with pressure on all countries with waters in the populations range to adopt appropriate management strategies. With many migratory species displaying similarly complicated movements, the need for unified management plans at an international level should be a priority. Furthermore, full understanding of the drivers behind a population's movements are essential in order to allow for effective management to be implemented, this should involve simultaneous tracking of both predator and prey species throughout the target species distribution in order to pick up temporal variation and response to prey movements. This should help separate movements based

on diet, which will affect the entire population, from movements based on reproduction which may only effect mature females. For the northeast Atlantic tope, we recommend further PSAT tagging of mature individuals of both gender, with focus on the latitudinal extremes (both north and south) to help identify migratory strategies. Determining the movements of small juvenile tope would also be advantageous and could be undertaken using data loggers and acoustic telemetry.

#### DATA AVAILABILITY STATEMENT

All data from the mark and recapture studies from the UK, Glasgow Museums and Scottish Shark tagging programmes are available upon request from the Scottish Sea Anglers Conservation Network, but applicants should be aware the data may be subject to restrictions due to sensitivities around exact locations of angler marks. Filtered DATRAS data is available from the authors upon request or from https:// datras.ices.dk/. Archival data is available upon request from the authors.

#### ETHICS STATEMENT

The animal study was reviewed and approved by Marine Scotland Science.

#### AUTHOR CONTRIBUTIONS

JT led on the conception and design of this work, undertook all field work, made a substantial contribution to analysis and created and assembled comments on all drafts of this manuscript. FN provided substantial contributions to the conception or design of the work, subsequent analysis, also advised on tagging methodology, provided training for archival tagging work and provided comment during drafting and revising of the manuscript. IB provided substantial contributions to the acquisition and interpretation of the Mark and Recapture data, assisted all field work and provided comment on the manuscript. L-AH gained funding via SIORC, assisted on field work for the deployment of the archival tags in 2014 and provided comment during drafting and revising of the manuscript. DB provided substantial contributions to the analysis of the data and provided comment during drafting and revising of the manuscript. LN and CJ provided substantial contributions to the conception and design of the work, subsequent analysis, and provided comment during drafting, and revising of the manuscript.

#### FUNDING

The Marine Alliance for Science and Technology for Scotland (MASTS) provided funding via a Ph.D. studentship and through the community project SIORC (Sharks, skate, and rays in the offshore and coastal regions of Scotland). MASTS was funded by the Scottish Funding Council (grant reference HR09011) and contributing institutions. This work was supported by the Fisheries Society for the British Isles and Scottish Natural Heritage.

#### ACKNOWLEDGMENTS

Thank you to Ken Collins, Richard Sutcliffe, and the Scottish Shark Tagging Programme for the mark and recapture data from the UK, Glasgow Museums and Scottish shark Tagging programmes for the data and all the anglers that have contributed data over the years. Thank you to the members of the public who kindly returned the Lotek tags found washed up on beaches. Thank you Barbara Pereira whose help in retrieving the recaptured Star-Oddi tag from Portugal was invaluable as well as the unnamed fishermen who returned the tag in the first place. Thank you to the funders, MASTS, SNH, and the FSBI and to the Scottish Sea Angling Conservation Network for the purchase of the PSAT tag.

#### REFERENCES


Specialist Group Pelagic Shark Red List Workshop. IUCN Species Survival Commission's Shark Specialist Group.


reserve of Bahia San Blas, northern Patagonia. Anim. Biotel. 2:11. doi: 10.1186/2050-3385-2-11


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Thorburn, Neat, Burrett, Henry, Bailey, Jones and Noble. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Young Birds Switch but Old Birds Lead: How Barnacle Geese Adjust Migratory Habits to Environmental Change

Thomas Oudman1,2 \*, Kevin Laland<sup>1</sup> , Graeme Ruxton<sup>1</sup> , Ingunn Tombre<sup>3</sup> , Paul Shimmings <sup>4</sup> and Jouke Prop<sup>5</sup>

*<sup>1</sup> School of Biology, University of St. Andrews, St. Andrews, United Kingdom, <sup>2</sup> Department of Coastal Systems, NIOZ Royal Netherlands Institute for Sea Research and Utrecht University, Den Burg, Netherlands, <sup>3</sup> Department of Arctic Ecology, Norwegian Institute for Nature Research, Tromso, Norway, <sup>4</sup> BirdLife Norway, Trondheim, Norway, <sup>5</sup> Arctic Centre, University of Groningen, Groningen, Netherlands*

#### Edited by:

*Nathan R. Senner, University of South Carolina, United States*

#### Reviewed by:

*Mitch D. Weegman, University of Missouri, United States Andrew Laughlin, University of North Carolina at Asheville, United States*

> \*Correspondence: *Thomas Oudman thomas.oudman@gmail.com*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *30 May 2019* Accepted: *09 December 2019* Published: *09 January 2020*

#### Citation:

*Oudman T, Laland K, Ruxton G, Tombre I, Shimmings P and Prop J (2020) Young Birds Switch but Old Birds Lead: How Barnacle Geese Adjust Migratory Habits to Environmental Change. Front. Ecol. Evol. 7:502. doi: 10.3389/fevo.2019.00502* Long-distance migratory animals must contend with global climate change, but they differ greatly in whether and how they adjust. Species that socially learn their migration routes may have an advantage in this process compared to other species, as learned changes that are passed on to the next generation can speed up adjustment. However, evidence from the wild that social learning helps migrants adjust to environmental change is absent. Here, we study the behavioral processes by which barnacle geese (*Branta leucopsis*) adjust spring-staging site choice along the Norwegian coast, which appears to be a response to climate change and population growth. We compared individual-based models to an empirical description of geese colonizing a new staging site in the 1990s. The data included 43 years of estimated annual food conditions and goose numbers at both staging sites (1975–2017), as well as annual age-dependent switching events between the two staging sites from one year to the next (2000–2017). Using Approximate Bayesian Computation, we assessed the relative likelihood of models with different "decision rules", which define how individuals choose a staging site. In the best performing model, individuals traveled in groups and staging site choice was made by the oldest group member. Groups normally returned to the same staging site each year, but exhibited a higher probability of switching staging site in years with larger numbers of geese at the staging site. The decision did not depend on food availability in the current year. Switching rates between staging sites decreased with age, which was best explained by a higher probability of switching between groups by younger geese, and not by young geese being more responsive to current conditions. We found no evidence that the experienced foraging conditions in previous years affected staging site choice. Our findings demonstrate that copying behavior and density-dependent group decisions explain how geese adjust their migratory habits rapidly in response to changes in food availability and competition. We conclude that considering social processes can be essential to understand how migratory animals respond to changing environments.

Keywords: Branta leucopsis, climate change, decision-making, explorative behavior, group decision, memory, migration, social learning

# INTRODUCTION

The choices that animals make in response to their environment have typically been shaped by evolution, and are therefore expected to maximize the animal's survival and reproduction. However, environments can change in ways that are hard to predict (Dall et al., 2005). In those cases, animals must deal with uncertainty in the consequences of their decisions. To understand those decisions, it is necessary to know which environmental factors individuals use to inform their decision, and how they integrate those factors to make the decision (i.e., their "decision rules"; Bauer et al., 2011; Budaev et al., 2019). This is particularly true for long-distance migrants, which must make decisions in anticipation of future and distant conditions (Kölzsch et al., 2015).

Animals use current environmental conditions on which to base their decisions, but also previous experiences may affect decisions (Berbert and Fagan, 2012). Memories allow animals to predict habitat quality by deducing temporal trends in stochastic seasonal environments (Abrahms et al., 2019). Furthermore, exploration of the environment can extend such experiences and thereby contribute to making better decisions in the future (Mettke-Hofmann et al., 2002; Tebbich et al., 2009), for instance, by informing the animal about the spatial distribution of resources. Another mechanism that can help the animal to make better decisions is social learning, which allows animals to exploit the experiences of others (Danchin et al., 2004; Couzin et al., 2005; Guttal and Couzin, 2010). Social learning can be an effective means to solve complex problems (Hoppitt and Laland, 2013), especially when combined with learning from previous individual experiences (Rendell et al., 2010). Recent semi-natural experiments suggest that animal populations can indeed accumulate improvements of migratory routes over several generations by combining individual learning with social learning (Sasaki and Biro, 2017; Jesmer et al., 2018), but evidence from natural populations is lacking. It remains largely unknown how migratory animals combine current and previous individual experiences with social learning to make decisions, and whether this combination helps them to adjust their migrations to environmental change.

A good candidate for further investigation is the barnacle goose, which is a social migratory species that has shown striking changes in migratory behavior in response to population growth and climate change (Eichhorn et al., 2009; Jonker et al., 2013). Barnacle geese follow the green wave of grass growth in spring (van der Graaf et al., 2006), but the sites where they stop along the way to accumulate crucial fat reserves for breeding (Drent et al., 2007) seem to be largely determined by tradition. For example, the barnacle goose population that migrates north along the Norwegian coast to breed on the Svalbard archipelago traditionally staged exclusively in Helgeland (**Figure 1A**; Black et al., 2014). Recently, a striking change has occurred in this tradition (Tombre et al., 2019). After a small group of birds in the 1990s colonized a new staging site 250 km further north, Vesterålen, the majority of the population has moved to the new site within a few generations (**Figure 1C**). The increasing number of birds in Vesterålen coincided with a strong increase in population size, which increased competition for resources at the traditional staging site. The shift in distribution also fits with an increase over the years of suitable habitat in Vesterålen due to climate change. Spring has advanced at both staging sites by 3 weeks since 1975. Grass growth simulations indicated that this advance has led to a higher grass production during the staging period at both sites, and simultaneously to a strong decrease in grass digestibility in Helgeland, but not in Vesterålen where spring starts ∼4 weeks later. As a result, the total production of digestible biomass per square meter of grass during the staging period has more than doubled in Vesterålen, but remained constant in Helgeland (**Figure 1B**).

Tombre et al. estimated from ring resightings of individually marked birds that ∼62% of the increasing use of Vesterålen can be attributed to birds that switched from the traditional to the new staging site in subsequent years, suggesting that the choice of staging site might be partly determined by geese responding to the changes in resource availability. However, in a year-toyear comparison switching rates did not correlate with foraging conditions, neither in the current nor in the previous year. This suggests a lack of direct response to changes in food availability, and implies that optimal foraging models (e.g., Bauer et al., 2006; Klaassen et al., 2006) are unlikely to explain the observed dynamics in staging site choice. Furthermore, young birds exhibited higher switching rates than older birds (**Figure 1D**). This implies that age-dependent changes in decision-making, which may (partly) result from social processes, affected the observed changes in migratory behavior.

We reason that the observed dynamics in staging site choice may be better understood when explicitly taking into account the ecological and social information that is available to individual animals, and the "decision rules" by which they integrate this information. To this end, we designed a set of simulation models, in which we implemented different potential sets of decision rules by which each individual in a simulated population of barnacle geese decides whether it stages in Helgeland or in Vesterålen. We used individual-based models, which are particularly suitable when the decisions by individuals and interactions between individuals are expected to affect the dynamics of the population (Bauer and Klaassen, 2013). Specifically, we analyzed which set of decision rules best explains the observed changes in stagingsite use, by comparing the performance of different models. Using Approximate Bayesian Computation (Beaumont, 2010), we simultaneously test which model is the most plausible given the empirical data, and estimate the values of the parameters in the selected model(s). Each model contains a different combination of the following five components: (i) adjusting choice to the expected quality of the current staging site, obtained by memorized individual experiences in previous year(s), (ii) comparing expected quality of the current staging site with expected quality of the alternative staging site, which is obtained through explorative behavior in previous year(s), (iii) leaving the choice to others by traveling in a group, (iv) reconsidering staging site choice at arrival in Helgeland, dependent on the current number of geese and/or grass cover, and (v) impact of age on any of the previous four processes.

FIGURE 1 | Barnacle goose spring-staging sites. All panels are reproduced from Tombre et al. (2019). (A) is a map of the migratory route (green arrows), and the two staging sites in red and blue. The geese winter at the Solway Firth, and breed on Svalbard. (B) shows the annual estimated staging site quality at both staging sites, estimated as the sum of the daily digestible biomass growth of grass leaves during the staging period. The lines are linear regressions and the shaded areas delineate the 95% confidence interval of local regressions. Panel (C) shows the number of spring staging geese at the two sites as found by the same study. Lines are the trends estimated by local regression, the colored areas depict confidence intervals. (D) shows the probability of geese of particular ages (y-axis) in each year (x-axis) to switch from staging in Helgeland to staging in Vesterålen in the subsequent year, as obtained from resightings of individually marked geese.

# METHODS

# Individual-Based Models

We simulated barnacle goose population dynamics in individualbased population models with discrete time steps of one year (see **Figure 2** for a visual description). In each model, the simulation runs started in 1970 with a population of 3,000 individuals with randomly assigned sex (50% chance of either male or female) and age (the initial age distribution was derived from a pilot simulation). Each individual was also assigned an age at which to become available as a partner, determined by drawing randomly from the Poisson distribution + 1 and λ = 1.5. This specific distribution with a mean of 2.5 and a standard deviation of 1.2 matches the empirically observed distribution (mean = 2.5, SD = 1.1; Choudhury et al., 1996). At the start of each time step, partnerships were determined, with pairs randomly assigned between available individuals (i.e., at or above the age of first reproduction and unpaired) of the opposite sex. Individuals remained with the same partner in subsequent years, only becoming available again as a partner when the partner died (Black et al., 2014; in reality the annual chance of a pair to separate is 2%, which we chose to ignore). All unpaired birds and a randomly assigned bird within each pair then chose a staging site: either Helgeland or Vesterålen. During the first time step, all individuals were set to choose Helgeland. In later time steps, individuals could instead decide to visit Vesterålen (see section Staging Site Decision Rules). Subsequently, each paired female reproduced with probability bs,<sup>t</sup> , where s is the visited staging site and t is the calendar year.

Previous simulation studies of goose behavior have focused on energetics (Bauer et al., 2006; Klaassen et al., 2006). While explicitly modeling density-dependent energy gain at staging sites and the consequences for reproductive success, they simplified the process of decision-making by assuming optimal behavior. We focused on the decision-making process and instead simplified the energetic part of the model. We

assumed that bs,<sup>t</sup> depends linearly on the annually estimated grass production at the staging site that she visited, and also decreases

linearly with an increasing number of bricks at that stage site, depending on the surface of foraging area:

$$b\_{s,t} = r\_{s,t} \left( 1 - \frac{N\_{s,t}}{K\_s} \right),\tag{1}$$

staging site, see Equation 1) that the female has visited, being either Helgeland (H) or Vesterålen (V).

where Ns,<sup>t</sup> is the abundance of birds at the visited staging site and K<sup>s</sup> is the total surface area of suitable foraging habitat at that staging site (m<sup>2</sup> ). The probability of reproduction in absence of competition, rs,<sup>t</sup> , is a linear function of the digestible grass production per m<sup>2</sup> during the staging period in year t at staging site s, qs,<sup>t</sup> (measured in g/m<sup>2</sup> , see next section):

$$r\_{s,t} = \ 0.1 + \ a \cdot q\_{s,t},\tag{2}$$

where a is a conversion factor (m<sup>2</sup> /g). The lower boundary of 0.1 reflects the low probability of reproduction observed for geese with very low body condition before departing Helgeland (Prop et al., 2003). Instead of deriving K<sup>H</sup> (carrying capacity in Helgeland) and conversion factor a mechanistically, we fitted them by performing model simulations without staging site choice, assuming all individuals to stage in Helgeland. The simulated population sizes were compared to the population count data between 1970 and 1997, when virtually all individuals visited Helgeland (see **Figure 1**). K<sup>H</sup> and a were estimated as 44,000 and 0.0082, respectively, by selecting the values that minimized the distance between the simulated population sizes and the empirically derived values (see section Calculating the Distance of Each Simulation to Empirical Data). Based on the ratio of agricultural land in the two areas (summed surface of agricultural land in 2017 was estimated at 27.6 and 88.5 km<sup>2</sup> for the main goose areas in Helgeland and Vesterålen; data downloaded from www.ssb.no), and given that barnacle geese in Helgeland also make use of natural salt marshes and that barnacle geese in Vesterålen face competition for food with pink-footed geese (Tombre et al., 2019), we estimated conservatively that K<sup>V</sup> was two times KH. A higher value of K<sup>V</sup> had no strong effect on the model selection results, as the population in Vesterålen remained far below carrying capacity in all simulations (see **Appendix I** and **Table S1**).

The number of offspring produced by a reproducing female was drawn from a Poisson distribution + 1 with λ = 1, resulting in a mean of two offspring, which equals the distribution in the number of juveniles associated with successful breeders in the wintering area (Black et al., 2014). At the end of each time step, individuals had a probability of dying, d, estimated at 0.11 (Black et al., 2014). Each simulation consisted of 48 time steps, representing the period from 1970 to 2017.

#### Grass Production at the Staging Sites

The digestible grass production per m<sup>2</sup> during each spring staging period t at staging site s, qs,<sup>t</sup> (g/m<sup>2</sup> ), was taken from Tombre et al. (2019). It was estimated as the sum of the daily digestible biomass growth of grass leaves from 30 April to 20 May (Prop and Black, 1998). The daily values were calculated by means of the simulation model CATIMO (Canadian Timothy Model; Bonesmo and Bélanger, 2002a,b). CATIMO simulates the daily growth of cell walls and cell contents in the leaves of timothy, Phleum pratense. Timothy is one of the main agricultural grass species and an important food source for barnacle geese in Norway (Black et al., 1991). Daily grass growth (g/m<sup>2</sup> ) was converted to digestible daily grass growth (g/m<sup>2</sup> ) by taking into account that the digestible proportion for barnacle geese is 0.16 and 0.64 for cell wall and cell content respectively (Prop and Vulink, 1992). The simulations were based on daily local temperature and radiation values. See Tombre et al. (2019) for a full explanation.

#### Staging Site Decision Rules

We compared 22 models, all with different decision rules determining the choice of staging site. Each set of decision rules is a combination of five components. The first component is memory, which we incorporated as an effect of staging site quality that the focal individual experienced in previous years. The second component is exploration, which we modeled as an effect of staging site quality at the alternative staging site in previous years when the individual was alive. The third component is traveling in groups. This is an effect of the staging site choice of others, in most cases the group leader, and hence a consequence of social learning. The fourth component is to reconsider staging site choice at arrival in Helgeland, with each individual continuing migration to Vesterålen with a probability that depends on the number of geese in Helgeland and/or the grass cover in Helgeland. As the fifth component, we included age-dependent differences between individuals in any of the four previous components (see also **Figure 2**).

In all models, paired birds stay together and normally return to the staging site of the previous year. In case newly paired birds did not visit the same staging site in the previous year, they make a random choice between both staging sites. Analysis of ring resightings before and after pair formation does not suggest a sex bias (TO and JP, unpublished data). Unpaired birds normally visit the staging site of the previous year. We assumed that each individual has an 18% probability of remaining with its parents during the first spring migration (Black et al., 2014), thereby copying the staging site choice of the parents. The first-year birds that do not stay with their parents follow others, based on one of the following criteria (denoted by parameter cjuv, for all parameters see **Table 1**): (1) follow a random non-first-year bird, (2) follow a parent (i.e., an individual that has produced offspring in the previous year), or (3) follow an individual of at least 10 years old, which is approximately the top 30% of the age-distribution (Black et al., 2014).

On top of this basic scheme, each individual can decide to switch staging site relative to the previous year. In the first model, each individual has a fixed annual probability of switching staging site (parameter n). Subsequent models incorporate different combinations of the five components as described below.

#### Memory

In each year, the expected probability of reproducing when returning to the current staging site (as opposed to switching to the other staging site), E bc , is given by a weighted average of its past experiences at that site. The weight of each of those experiences is given by the decay function e − y <sup>m</sup> , where y is the 'age' of the experience (in years) and parameter m determines the rate at which memories fade. We assumed that individuals start switching to the other staging site when E(bc) falls below a threshold that is given by parameter xa. Below this threshold, the probability of switching increases with decreasing E(bc)with a rate that is determined by parameter x<sup>r</sup> , where:

$$P\left(switch\right) = \chi\_r\left(max\left(0, \chi\_a - E\left(b\_c\right)\right)\right). \tag{3a}$$

#### Exploration

Individuals explore the alternative staging site at the end of the staging period with probability (v), enabling them to inform their expectation of the reproduction probability when visiting the alternative staging site, E ba . If the difference between E ba and E bc is larger than x<sup>b</sup> , then the probability of switching staging site in the next year is given by:

$$P\left(switch\right) = \chi\_r\left(max\left(0, \chi\_b - E\left(b\_c\right) + E\left(b\_a\right)\right)\right),\tag{3b}$$

where parameter x<sup>r</sup> determines how fast the probability of switching increases as the difference between E ba and E bc increases. This component only affects the model results when memory is also implemented, with equation 3b replacing 3a.

#### Groups

Instead of individually deciding where to go, birds may also choose to follow another individual, thereby copying its choice of staging site. We modeled this by assigning each bird to a group, and determining staging site choice per group instead of per individual. In this case, juveniles do not join an individual, but a group. We assumed that 18% of the juveniles joins the group of their parents (Black et al., 2014), and the rest joins a randomly chosen group. Group decisions may be made in different ways, denoted by parameter cgroup. We assumed that individuals either (1) form groups with a single leader, which may be (1) a random bird, (2) a randomly chosen parent (i.e., an individual that has produced offspring in the previous year), or (3) a randomly chosen bird from among the oldest ones. Alternatively, each group member first makes an individual choice as explained above, after which the group reaches consensus by adopting the "majority vote" (4). Note that simple and plausible behavioral mechanisms allow individuals to follow any of these rules, without having an overview of the process (Couzin et al., 2005). We further assumed that individuals join the same group as in the previous year (but see component v, Aging). Maximum group size is determined by parameter g, with groups splitting into two equally sized groups when larger than g, and merging with a random other group when smaller than 0.25g.

#### Reconsidering Staging-Site Choice

At arrival in Helgeland, individuals have the possibility to reconsider their choice, and continue to Vesterålen. The probability to continue is either linearly dependent on the number of geese, N<sup>H</sup> (Reconsidergeese), or on the grass cover at


arrival in Helgeland (Reconsidergrass). Grass cover is calculated for each day in CATIMO as the leaf area index, LAI, measured in m<sup>2</sup> of grass leaves per m<sup>2</sup> of ground. Both functions depend on three parameters: the number of geese or the grass cover at which the probability to switch starts to increase (ge<sup>0</sup> and gr0), the linear rate at which the probability increases (ge<sup>r</sup> and grr), and the maximum switching probability (ge<sup>m</sup> and grm):

$$P\left(\text{Recondier}\_{\text{gecse}}\right) = \left(\max\left(0, \min\left(\text{gec}\_m, \text{gec}\_r\*(N\_H - \text{gec}\_0)\right)\right)\right) \tag{4a}$$

$$P\left(\text{Recondier}\_{\text{gans}}\right) = \left(\max\left(0, \min\left(\text{gcr}\_m, \text{gcr}\_r\*(LAI\_H - \text{gcr}\_0)\right)\right)\right) \tag{4b}$$

#### Aging

We explored four different potential effects of age. The first assumed that the influence of previous experiences on the current decision decreases with the age of the individual. We modeled this by multiplying the probability of switching (see Equations 3a and 3b) with an age-factor a that changes with age according to the function

$$a = e^{\frac{1-a\rho\epsilon}{a\_r}},\tag{5}$$

where age is measured in years. Parameter a<sup>r</sup> (also in years) determines the strength of the age-effect. A second possibility is that the probability of exploring (v) decreases with the individual's age, which is modeled by multiplying v by age-factor a. Thirdly, if the animals make migratory decisions in groups (see component i), then there may be an age-effect in the probability of changing to a randomly chosen new group, w0, which is then multiplied by the age-factor a. Fourthly, there could be an ageeffect in the tendency of individual geese to reconsider their staging site choice upon arrival in Helgeland. This is modeled by multiplying the probability to reconsider (Equations 4a and 4b) with the age-factor a.

### Empirical Data

To determine which model is most plausible, we compared the simulations to two different sets of empirical data, both published by Tombre et al. (2019). The first set consists of the annual number of spring staging barnacle geese in Helgeland and in Vesterålen. This set contains 86 data points, being the estimated numbers of birds at each site in each year from 1975 to 2017 (**Figure 1C**). They were derived from annual counts during the staging period in Helgeland and Vesterålen, and annual counts of the total population size in the wintering area. The second set of data points consists of the probabilities of individual birds switching from staging in Helgeland to staging in Vesterålen in subsequent years (from here on referred to as "switching probabilities"). Each data point is the switching probability for an individual of a given age (age 1 to age 20) in a particular calendar year between 2000 and 2016 (**Figure 1D**). These data points were derived from resightings of marked individuals at both staging sites, as well as the wintering and the breeding area. Further details can be found in Tombre et al. (2019). As hardly any geese were observed staging in Vesterålen from 1975 to 1995, we infer that switching probabilities from Helgeland were zero from 1975 to 1995 for all ages. The years 1996 to 1999 were not part of the analysis. This resulted in a total of (21 + 17) × 20 = 760 data points. We did not compare the switching probabilities in the other direction (from Vesterålen to Helgeland), because these could not be estimated in years when the simulated bird numbers were zero in Vesterålen.

### Model Selection: Approximate Bayesian Computation

We evaluated the relative strength of the different models by comparing simulations to the empirical data using Approximate Bayesian Computation (ABC; Beaumont, 2010) in R (R Core Team, 2018). This statistical tool has been developed to quantify the fit of different individual-based models to different sets of empirical data simultaneously. The ABC-method allows the fit of different models to be compared (e.g., models with and without memory), as well as comparing the fit of different parameter values within each model (e.g., values of a parameter determining the rate of memory loss). The method is called "Bayesian" because the method updates the degree of belief in each model given the empirical data. It is "Approximate" because it is not an analytical method, which is generally not an option for individual-based models, but instead relies on simulations (van der Vaart et al., 2015). We used rejection-ABC, the simplest and most accessible type of ABC that can be used for ecological models with multiple parameters (van der Vaart et al., 2015, 2016). Calculations were performed as in the R-package "abc" (Csilléry et al., 2012), except for indicated differences. Below, we explain the method step by step.

First, parameter values are defined. Where possible, parameters were estimated from the literature (see **Table 1**). For the other parameters, distributions were defined such that all possible values are included (see **Table 1**). These distributions are referred to as "prior distributions." Then, 10,000 simulation runs were performed for each model. For each simulation, the values of all parameters in the model were drawn at random from the prior distributions. After all simulation runs were performed, we calculated the distance between each run and the empirical data (see next section). To give equal weight to both used datasets (bird numbers and switching probabilities), we calculated the distance of each simulation run to the empirical data separately for each dataset, and then took the mean of the two to arrive at a single distance estimate for each simulation run. Finally, the 100 runs with the smallest distance were selected. The evidence for model x relative to model y is expressed by the Bayes factor (Bx,y), which in this context is defined as the ratio of simulations from each model among the selected runs (van der Vaart et al., 2016).

To test whether the result would change with more simulations, we ran a bootstrapping test of the model selection accuracy by repeating the procedure 100 times, each time with a randomly chosen half of all simulation runs. To evaluate the ability of the ABC-method to distinguish between different models, we carried out cross-validation as implemented in the function "cv4postpr" in the "abc" R-package, and described in Csilléry et al. (2012). First, 100 simulation runs are randomly selected from each model. Then, for each of these runs, the complete model selection procedure is repeated after removing this run from the simulation data and replacing the empirical data with this run. The result is a "confusion matrix", where each row represents the number of simulations under each model, and each column represents the number of simulations assigned to that model by the model selection procedure.

The distribution of parameter values among the selected simulation runs ("posterior distributions") can be regarded as a probability distribution for each parameter, and acts as a sensitivity analysis. To test whether the posterior distributions were significantly different from the prior distributions (distribution of parameter values among all runs), we performed a Chi-square test after dividing the data into 10 equally-sized bins with the function "bin" in R-package OneR (von Jouanne-Diedrich, 2017). To correct for multiple testing, we applied a Bonferroni correction to the standard significance level of 0.05.

#### Calculating the Distance of Each Simulation to Empirical Data

Distance (ρ) is defined as the standardized Euclidian distance between all data points j in simulation i (Mi) and the same data points in the empirically derived data (D):

$$\rho\left(M\_{\rm i},D\right) = \sqrt{\sum\_{j} \left(\frac{M\_{\rm ij} - D\_{j}}{sd\left(M\_{\rm j}\right)}\right)^{2}},\tag{6}$$

where Mi,<sup>j</sup> is the output of run i for datapoint j, D<sup>j</sup> is the empirically derived value of data point j, and sd(Mj) is the standard deviation of data point t in all simulation runs. As in van der Vaart et al. (2015), we used standard deviation instead of median absolute deviation (as is done in the "abc" package; Csilléry et al., 2012), because the median was zero for several datapoints and this led to undefined distances. To TABLE 2 | Model selection results, showing for each model the number of runs among the best 100 simulation runs.


*Selected models are in bold.*

\**Model selection was performed in two steps: first only with models without "Age." The best model and competitive models were tested in a second step, together with a new set of models based on those models that included "Age".*

avoid overfitting, we chose to compare the simulations to the statistically estimated trends (**Figures 1B,C**), rather than to the raw empirical data. We made this decision because an unknown part of the inter-annual variation in the empirical data is caused by non-modeled processes, such as annual conditions in the breeding area and observation errors.

#### Reducing the Number of Simulations

To reduce the required number of simulation runs, we adopted a two-step model selection procedure. First, we performed a model selection of scenarios without the "age" component (models 1 to 15 in **Table 2**), and executing 10,000 simulation runs per model. We then composed seven additional models based on the selected models, but also including an age-effect (models 16 to 24 in **Table 2**), and executed 10,000 simulations per model. We did not consider models with an age-effect on more than one component, to further limit the number of models to be tested. These additional models were tested in a new model selection procedure, also including the selected models from the first model selection. For parameters in the first selection where the posterior distribution was significantly different from the prior distribution (**Figure S1**, **Table 3**), we updated the prior distributions for the simulations in the second model selection procedure (**Table 3**).

#### RESULTS

#### Model Selection

The simulation most similar to the empirical data was produced by model 16, which includes "reconsidergeese", "groups", and an age-effect on "groups." The pattern resulting from this simulation corresponded to the observed annual bird numbers at both staging sites (**Figures 3D,E**). Moreover, it showed a decrease in switching probability with age (**Figure 3F**), which was similar to the pattern in the empirical data (**Figure 1D**). This model was also the best represented model among the 100 best simulation runs (60 out of 100 runs, **Table 2**). The same model but with "memory" (model 21) was represented with 34 runs. With a Bayes factor of B16,21 = 1.8 there is no evidence that memory does not play a role, but it does not improve the performance of the model in explaining the empirical data. Roughly, a Bayes factor of 3 to 10 is regarded as "substantial evidence" and above 10 as "strong evidence" (Kass and Raftery, 1995; van der Vaart et al., 2016). Apart from models 16 and 21, only model 18 (model 21 but without "reconsidergeese") occurred among the 100 best models,


\**Significant after Bonferroni correction (significance level* = *0.05/20* = *0.0025).*

*†No p-value is given when not enough simulations with this parameter were among the selected runs to perform statistics.*

with 6 runs (B16,18 = 10 and B21,18 = 5.7), meaning that there is substantial evidence for models 16 and 21 over model 18, and strong evidence over all other models. Hence, the results suggest that staging site choice is made in groups, with a decrease over age in the probability that individuals change groups, and that groups switch to another staging site based on the current number of geese at the staging site. The results are indefinite regarding the role of previous experiences at the alternative staging site. There is no evidence that exploration of the other staging site in previous years plays a role, nor that there is an effect of current food conditions at the staging site.

#### Model Validation

Because models 16 and 21 both came out as likely to underlie the empirical data, we focused on these models in the model validation. When repeating the model selection analysis 1,000 times with a randomly chosen half of the data, models 16 and 21 together always made up the majority of the selected runs (range 91–100 out of 100 selected runs, mean 96, **Figure 4**). Hence, the evidence for models 16 and 21 relative to the others is robust. The only other model that appeared among the selected simulations was model 18 ("memory", "groups" and an age-effect on groups, mean 4, range 0–9). The cross-validation procedure suggested that the model selection performs badly in estimating the underlying model of randomly drawn simulations: of the runs that were produced by model 16 or 21, only 67% were also estimated as such (**Figure S2**). This result was to be expected, because simulations were similar between models for a large proportion of parameter combinations. For example, switching did not occur at all in many simulation runs of all models with "groups" (between 6 and 40%). When performing the cross-validation procedure with the 100 best fitting runs of each model instead of randomly drawn runs, then 98.5% of the runs produced by model 16 or 21 were also estimated as such (**Figure S2**). Hence, when the data was close to the observed trends, the model selection performed well.

#### Parameter Estimation

In the 100 selected simulations runs of the first step in the model selection (see **Figures 3A–C** for simulation results), the distribution of values (posterior distributions) of 10 out of 15 parameters were significantly different from the defined prior distributions, of which five were in models that were represented among the best simulations (**Table 3**, **Figure S1**). For those parameters, we defined new prior distributions for use in the simulation runs for the main model selection (**Table 3**). In the selected simulations after the second step in the model selection, the posterior distributions of five out of ten parameters were significantly different from the defined prior distributions (cgroup, g, ge0, w<sup>0</sup> and a<sup>r</sup> , **Table 3**, **Figure 5**).

In all of the selected runs the birds traveled in groups. Smaller groups occurred more often among the selected runs than larger groups (see **Figure 5B**). In most of the selected runs, the oldest individuals led the group (78 out of 100 runs, **Figure 5A**). Simulations where group decisions were made by a majority vote always performed badly (see **Table 3** and **Figure S1**; it did not occur in the selected runs in the first step, and was therefore removed from the prior distribution of the main selection procedure). Individuals switched between groups in all selected simulations, with most of the runs having an initial switching probability below 0.4, and a relatively slow decrease with increasing age (**Figure 5F**). In the selected runs where the

probability for a group to switch increased with goose numbers at Helgeland (94 out of 100), birds started to switch when numbers were between 10,000 and 15,000 geese (parameter ge0). The selected runs including "memory" and "reconsidergeese" responded less strongly to density (parameter ger) than the runs with "memory" but without "reconsidergeese" (**Figure 5D**). There was no pattern in the maximum probability to reconsider staging site (parameter gem; **Figure 5D**). The selected runs with memory (40 out of 100) showed no clear pattern in the rate of memory loss (parameter m; **Figure 5F**), suggesting that the rate of memory loss is not importantly affecting the dynamics. The same was the case for x<sup>r</sup> , the rate at which the probability of switching increases when the expected probability of reproducing declines (the slopes in **Figure 5C**).

# DISCUSSION

Simulations resembled the empirical data best when geese were assumed to travel in small groups that are led by the oldest individuals, and when young geese switched more between groups in subsequent years than did older individuals (**Table 2**, **Figure 5**). Further, the results suggest that the current food conditions are of minor importance to staging site choice, but that the abundance of geese in Helgeland does increase the probability for groups to reconsider their choice and continue to Vesterålen. The model results are indecisive about whether experiences acquired by the group leaders in previous years, i.e., the "memory" component, influence the decision to switch staging site. We found no evidence that experiences at the alternative staging site in previous years contributes to the decision (**Table 2**). Below we discuss the implications of these results in more detail.

#### Grouping

The well-known fact that geese operate in groups need not inherently imply that each individual's choice of staging site is influenced by other members the group. For example, group-foraging pink-footed geese during spring staging decided individually on their specific daily foraging locations

(Chudzinska et al., 2016). Our results are the first to suggest that group decisions do play a role in the choice of staging site. In all selected simulations (i.e., best fitting with the empirical data), staging site choice was made in groups.

The results further suggest that these decisions are not arrived at by a majority vote. The gradual increase in numbers in Vesterålen in the 1990s is not compatible with this decision rule, which requires a high proportion of all individuals to prefer switching, before the first geese start to switch. This aligns with the idea that strong conformity is generally not a good strategy in changing environments, because innovative behavior is unlikely to spread even when highly adaptive (Eriksson et al., 2007; Kandler and Laland, 2009). The most likely group decision rule was to follow the oldest, and therefore most experienced, bird of the group. This rule performed better than following parents (Chi-squared test, χ 2 <sup>1</sup> = 36.6, p < 0.0001), which in turn performed better than following a random individual (Chi-squared test, χ 2 <sup>1</sup> = 7.7, p = 0.006).

Following experienced birds might be adaptive because the annual food conditions at the staging site vary stochastically (**Figure 1B**), and longer experience will provide a better prediction of next year's staging site conditions. In contrast, following an individual that produced offspring in the previous year is hardly predictive of the chances to reproduce in this year if annual stochasticity is high (Baldini, 2012). This may explain why the model results indicated that following an individual that raised offspring was less likely than following an experienced leader. That following experienced birds is better than following successful breeders also could explain why in reality most firstyear barnacle geese choose not to follow their parents on their first spring migration; on average, it would provide a higher pay-off to follow old and experienced individuals than to follow the parents. However, inclusive fitness arguments predict that unrelated group members may be more hostile than parents or other related individuals. Indeed, this also holds for barnacle geese (Black et al., 2014). Nonetheless, there are more examples of animals that are more likely to copy old (Amlacher and Dugatkin, 2005) and knowledgeable (Kendal et al., 2015) individuals, and to copy experienced others rather than the parents (Agostini et al., 2017). In bird flocks, leaders have been shown to be the more experienced individuals (Flack et al., 2012; Mueller et al., 2013). Our results imply that following experienced birds is especially advantageous when recent success needs not be a good predictor of subsequent success, but multiple-year averages of success are.

# Reconsideration of Staging Site Choice at Arrival in Helgeland

The component "reconsidergeese" featured in all selected simulation runs. In models with this component, group leaders are more likely to reconsider their staging site choice after arrival in Helgeland in years when the number of birds in Helgeland is high. Simulations with this density-dependent effect corresponded better to the empirical data, because this effect keeps individuals from switching to Vesterålen before 1990. This also explains why simulation runs with "reconsidergrass" do not perform well, not even when combined with "reconsidergeese". In models with "reconsidergrass", the probability of reconsidering staging site choice increases as the grass phenology is more advanced at arrival in Helgeland. In those simulation runs, individuals do often colonize Vesterålen before 1990 because years with an early spring also occurred before 1990 (**Figure 1B**). Hence, these results suggest that the choice between Helgeland and Vesterålen is not a direct response to the "green wave" of spring phenology (van der Graaf et al., 2006). Instead, the growing preference for Vesterålen follows from a response to other geese, both positive (grouping) and negative (densitydependent switching).

#### Memory and Exploration

From an optimal foraging perspective, it is expected that any knowledge about the conditions at the current or alternative staging site should play a strong role in the decision whether or not to return to the current site in the following year (Stephens and Krebs, 1986, Abrahms et al., 2019). This influence was captured in the "memory" and "exploration" components of the model. The "memory" component was part of 40 out of 100 of the selected simulations (models 18 and 21; see **Table 2**), Although this is not evidence against memory playing a role, we conclude that there is no need to assume that geese memorized foraging conditions at the staging site in the previous year(s). Note that this only concerns memory of foraging conditions. In all models, individuals (or at least group leaders) are assumed to have spatial memory, and remember the migration route and staging site of the previous year (Mettke-Hofmann and Gwinner, 2003).

Adding the "exploration" component also did not improve the fit of simulations to the data, as the best model in the first step of the model selection with exploration (model 12) was less well represented than the same model without exploration (model 11). Hence, the current results are also indecisive with regard to the importance of exploration for decision-making. Geese have only rarely been observed to spend a significant amount of time at both staging sites in one spring, but they

simulation of the models with "groups" (parameter *cgroup*). (B) is the frequency distribution of the maximum group size in each simulation run (parameter *g*). The lines in (C) define how the annual switching probability depends on the individual's expected probability of reproducing at the current staging site, E(bc). They are determined by parameters *x<sup>a</sup>* (threshold value below which the probability becomes non-zero) and *x<sup>r</sup>* (the slope of the line below *xa*) in the models with memory but without exploration. In (D), the lines are determined by parameters *ge*0, *ge<sup>r</sup>* and *ge<sup>m</sup>* in the models with "reconsidergeese", which determine how the probability to switch preference after arrival at the staging site, depending on the number of geese there. (E) shows how, resulting from differences in *m*, the weight of each memory declines over the years. The lines in (F) depend on parameters *w*<sup>0</sup> and *a<sup>r</sup>* , and define how the probability of switching between groups decreases with age in each simulation (only models 16 and 21).

occasionally made a short stop in Helgeland before continuing to Vesterålen (PS, IT and JP, unpublished data). Less frequently, geese staging in Helgeland were also observed in Vesterålen at the end of the staging period, although most geese fly directly north after staging in Helgeland (PS and Larry Griffin, unpublished visual observations of departing geese and satellite tracks). A potential way forward is to add a third set of empirical data to the comparison, for example containing information on individuals that were (or were not) observed at multiple staging sites, in relation to their switching behavior. However, exploring individuals may be easily missed by observers if they land only shortly or not at all, making it hard to determine the rate of occurrence by ring resightings. More information on the rate of exploration and age-dependent changes in exploration could be derived by tracking individuals with gps-tags. Another possibility is to model the effect of exploration in more detail, which might lead to a better fit with the current empirical data. For example, new simulations could allow the probability of exploring Helgeland when staging in Vesterålen to be different from the probability of exploring Vesterålen when staging in Helgeland.

#### Aging

The finding that migratory decisions are age-dependent confirms a general trend that young birds become more consistent in their migratory decisions as they grow older (Lok et al., 2011; Oppel et al., 2015; Vansteelant et al., 2017). In Eurasian spoonbills (Lok et al., 2011) as well as in pink-footed geese (Clausen et al., 2018), a higher probability for young individuals to switch wintering site between years was attributed to young birds being more explorative. This has also been the main hypothesis to explain the higher probability of staging site switching by barnacle geese (Tombre et al., 2019). However, our results suggest that juveniles do not explore new staging sites deliberately. Instead, they are more likely to travel with different groups in subsequent years, which results in a higher probability of ending up at different staging sites. Also this groupswitching behavior might be understood as being "explorative", but it is social exploration rather than spatial exploration. This is an important distinction because it implies that migratory innovation needs not start with young and naïve individuals, as was suggested before. The modeling exercise indicates that the colonization of Vesterålen is more likely to have been initiated by old and experienced individuals, which were being followed by young animals.

## Suggestions for Future Research

By comparing our simulations to the statistical trends in the empirical data (instead of the raw empirical data) uncertainty in the empirical trends were not conveyed to the statistics of our model selection procedure. We think that this method is to be preferred over using the raw data in this case, because the empirical trends were derived from a resighting analysis of individual bird observations. That analysis takes into account the probability of either not observing a bird when it is actually there (resighting probability) or not observing a birds because it is dead (mortality probability; for details see Tombre et al., 2019). To take these considerations into account when comparing individualbased models to the raw data, it would be necessary to also simulate the process of data collection within these models. We propose that incorporating data collection in the simulation exercise could be an interesting venue for future research.

In individual-based models, each decision must be modeled explicitly (Bauer and Klaassen, 2013). The advantage is that all of the underlying assumptions, many of which remain implicit and are often ignored in other types of modeling, become explicit. A disadvantage is that it remains unknown how much the way a process is modeled affects the results. For example, we modeled the process of pair formation and group formation in a basic way, with young individuals choosing a partner or a group at random. There are indications that individuals will be more likely to group with others that they grew up with (Choudhury and Black, 1994; van der Jeugd et al., 2002). Other studies have shown that social structure within groups can have strong effects on group dynamics (e.g., Bateman et al., 2013). Modeling these aspects more precisely could produce further insights into the causes and consequences of group formation by barnacle geese.

We stress that we only investigated the tip of the iceberg when it comes to individual differences. There may well be differences in decision-rules between individuals other than those mediated by age. Research on individual differences (Dingemanse et al., 2010), including those in barnacle geese (Kurvers et al., 2009), has shown that animals within the same population and of the same age can differ greatly in personality characteristics such as dominance, aggression, and exploration. Although beyond the scope of this study, such individual variation could be incorporated as an extension of the current study by assuming that individuals within the same population can act according to different sets of decision rules.

#### Cultural Evolution of Migratory Behavior

Social learning is an essential part of migratory inheritance and development for many migratory bird species (Sutherland, 1998; Helm et al., 2006; Németh and Moore, 2014), and for barnacle geese in particular (e.g., Eichhorn et al., 2009; Jonker et al., 2013). This study is the first attempt to infer the details of the learning processes in migratory decision-making from empirical data. The results indicate that geese travel in groups led by the oldest individual whose decisions are density-dependent, and the modeling explains how barnacle geese are able to respond so rapidly to long-term trends in competition and climate change at the staging sites. This is in line with the long-held conviction that cultural evolution allows for faster adaptation than genetic evolution (Boyd and Richerson, 1985; Sutherland, 1998). However, copying the behavior of conspecifics can also inhibit behavioral adjustment, and cause sub-optimal traditions to be maintained (Warner, 1988; Day et al., 2001; Németh and Moore, 2014). In order for social learning to lead to rapid adaptation, it typically needs to be combined with some low amount of individual learning, or other processes that introduce variation (Rendell et al., 2010). Intriguingly, the decision process that we identified here as being the most likely for migratory decisions by barnacle geese, does exactly this. Most geese follow others, but some of the experienced geese that lead the groups alter their decisions in response to current conditions.

We have discussed how the adaptive value of the observed decision rules is expected to depend on the amount and nature of the environmental variation. Although we expect that our results will also apply to other decisions, both by barnacle geese and by other social species, we stress that care should be taken when generalizing the results. An interesting venue for future research will be to apply the methods presented here to other published studies of migratory behavior across taxa and across situations. Finding general patterns between decision rules and environmental and social context will help to understand why some populations are more vulnerable to environmental change than others, and allow for better predictions of the ecological consequences of climate change. Currently, most studies of population dynamics do not consider the specific processes by which animals make their decisions. While arguably in some cases this may be a legitimate simplification, in cases like the present one, social and developmental aspects of decisionmaking turn out to be essential for understanding the populationscale response to environmental change.

#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

# ETHICS STATEMENT

Ethical review and approval was not required for the animal study because all the data for this study has been published elsewhere, and where necessary were approved by an Animal Ethics Committee.

# AUTHOR CONTRIBUTIONS

TO, JP, KL, and GR conceived the study design. JP, IT, and PS provided the data. TO performed the analyses and wrote the manuscript with contributions from all authors.

#### FUNDING

This research was funded by a grant from the Netherlands Organization for Scientific Research awarded to TO (ref 019.172EN.011).

#### ACKNOWLEDGMENTS

The manuscript benefitted from comments on an earlier draft by Magda Chudzinska, and from advice by Luke Rendell. We thank the two reviewers for their helpful comments,

#### REFERENCES


and Dick Visser for editing the figures. Although this is a theoretical study, it fully depends on the data published by Tombre et al. (2019). This data was painstakingly gathered by a large international team of observers to which we are indebted.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00502/full#supplementary-material


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Oudman, Laland, Ruxton, Tombre, Shimmings and Prop. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Migration Takes Extra Guts for Juvenile Songbirds: Energetics and Digestive Physiology During the First Journey

#### Brendan J. McCabe and Christopher G. Guglielmo\*

*Department of Biology, Advanced Facility for Avian Research, University of Western Ontario, London, ON, Canada*

Many birds undertake long migrations when they are only a few months of age. Although they are typically of adult body size, their performance, and survival are often poor compared to adults. This differential performance could be due to lack of experience, selection against poor-performing cohort members, or inherent constraints of continuing physiological and morphological maturation of juveniles. Limited evidence suggests that digestive and muscle physiology of juveniles during their first migration may differ from that of adults. We compared body composition, metabolic rate, and digestive physiology between juvenile and adult passerines during fall migration. First, we measured fat and lean masses by quantitative magnetic resonance, and organ and muscle masses of salvaged carcasses of fall migrants from four passerine species. In general, juveniles had more lean mass and heavier digestive organs (especially liver) than adults in hermit thrushes (*Catharus guttatus*), Swainson's thrushes (*Catharus ustulatus*), ovenbirds (*Seiurus aurocapilla*), and white-throated sparrows (*Zonotrichia albicollis*). Principal components analysis of all organs and muscles revealed that juveniles for three of four species had overall larger digestive components and smaller flight muscles than adults. We then used open-flow respirometry to measure basal metabolic rates (BMRs) of juvenile and adult Swainson's thrushes and white-throated sparrows captured in fall at a migratory stopover site. Controlling for a significant effect of body mass, juveniles had 6% higher BMRs than adults in both species. We then conducted total collection mass balance feeding trials with fall migratory Swainson's thrushes and white-throated sparrows. Juvenile thrushes had greater metabolizable energy intake than adults, which was achieved through higher food intake rather than greater utilization efficiency. Age classes of white-throated sparrows did not differ in these measures of digestive performance, although juveniles had greater food intake capacity at low lean body masses. We propose that age-related differences in foraging ecology, diet composition, and energy requirements may be responsible for larger digestive organs of juvenile migrants. Larger guts may allow juveniles to consume more food or a more dilute diet, but may contribute to higher BMRs.

Keywords: age, metabolic rate, body composition, digestive efficiency, migration, stopover

#### Edited by:

*Nathan R. Senner, University of South Carolina, United States*

#### Reviewed by:

*Scott McWilliams, University of Rhode Island, United States Maria Stager, University of Montana, United States*

> \*Correspondence: *Christopher G. Guglielmo cguglie2@uwo.ca*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *29 May 2019* Accepted: *23 September 2019* Published: *16 October 2019*

#### Citation:

*McCabe BJ and Guglielmo CG (2019) Migration Takes Extra Guts for Juvenile Songbirds: Energetics and Digestive Physiology During the First Journey. Front. Ecol. Evol. 7:381. doi: 10.3389/fevo.2019.00381*

# INTRODUCTION

Piersma and van Gils (2011) described the morphology and physiology of animal bodies as expressions of their ecologies. Animal bodies are not static; changes during development transform phenotype and performance capacity, and some animals maintain an ability to reversibly alter phenotype even when fully mature to cope with changes in workload associated with environmental variation or life history stages (Secor and Diamond, 1995; Piersma and Lindström, 1997; Piersma and Drent, 2003). During migration, animals such as birds face greatly elevated physiological demands and may experience a variety of environmental conditions as they move between breeding and wintering areas. Migratory birds express a migratory phenotype, which includes hyperphagia, fattening, changes in muscle and organ sizes, increased liver fatty acid synthesis, and increased muscle aerobic and fatty acid oxidation capacity (Guglielmo, 2018). Individuals that do not achieve an appropriate migratory phenotype may be at risk of failure or mortality because they lack sufficient functional capacity, and in particular, this, along with potential behavioral deficits in navigation or foraging, could explain the poor migratory ability of juvenile birds.

Migrating between breeding and distant wintering areas is both energetically expensive (Wikelski et al., 2003) and dangerous (Sillett and Holmes, 2002) for birds. For most migratory passerines, migration consists of multiple flights interspersed with stopover periods where birds rest and refuel to prepare for subsequent migratory flights. Migrant birds spend more time (Dolnik and Blyumental, 1967; Fransson, 1986) and energy (Wikelski et al., 2003), at these stopover locations than during actual flight. Thus, the amount of time and energy spent at stopover sites can limit overall migration speed (Alerstam and Lindström, 1990).

Altricial migratory songbirds develop quickly from a helpless nestling with functionally immature muscles and poor motor and thermoregulatory abilities (Starck and Ricklefs, 1998) to a migratory phenotype capable of traveling thousands of kilometers. During their first migration, juvenile birds are often the same physical size and mass as adults, although they can sometimes be distinguished by plumage markings, amount of feather wear, or evidence of incompletely pneumatized skulls (Pyle, 1997). Juveniles generally tend to underperform adults during migration. For example, juveniles migrate slower (Ellegren, 1990; Fransson, 1995; Susanna et al., 2008), and may spend more time at stopovers (Veiga, 1986; Ellegren, 1991; Morris et al., 1996; Yong et al., 1998; Rguibi-Idrissi et al., 2003; Mills et al., 2011) than adults. These observations suggest that juveniles refuel differently from adults during fall migration. Indeed, for some species adults often arrive at stopover locations in better body condition, while juveniles may arrive lighter and with less fat (Woodrey, 2000; Jones et al., 2002), but these differences are not always observed (Kennedy, 2012). Before leaving breeding areas on fall migration, juveniles forage less efficiently than adults (Weathers and Sullivan, 1991; Heise and Moore, 2003; Vanderhoff and Eason, 2007, 2008), but may improve prior to departure (Heise and Moore, 2003; Wheelwright and Templeton, 2003). If juveniles remain less skilled foragers during migration, they may be able to compensate by increasing foraging time to achieve similar overall energy intake to adults. Several studies found refueling differences between juvenile and adult passerines during stopover on fall migration (Veiga, 1986; Morris et al., 1996; Woodrey and Moore, 1997; Yong et al., 1998; Jones et al., 2002; Leist, 2007; Arizaga et al., 2008; but for different findings see Heise and Moore, 2003; Seewagen et al., 2013).

Further evidence suggests juveniles are physiologically different from adults during fall migration. For example, juvenile western sandpipers (Calidris mauri) had lower concentrations of heart-type fatty acid binding protein in flight muscle and lower activities of several digestive enzymes than adults (Guglielmo et al., 2002; Stein et al., 2005). Activities of several aerobic and glycolytic enzymes in pectoralis muscle are much lower in juvenile than in adult barnacle geese (Branta leucopsis), but rapidly increase prior to fall migration departure (Bishop et al., 1995). Juvenile migrants had heavier livers and digestive tracts than adults in several studies (Graber and Graber, 1962; Hume and Biebach, 1996; Guglielmo and Williams, 2003; Stein et al., 2005). Non-migratory house sparrows (Passer domesticus) of similar age to migrant juveniles also had larger digestive tracts and livers than adults (Chappell et al., 1999).

Generally, larger organ size indicates a greater functional capability for that organ (Hammond and Diamond, 1992; McWilliams et al., 1999; Secor and Diamond, 2000). Larger alimentary tracts and livers of juveniles may indicate greater capacity for digestion of food, absorption of nutrients, and postabsorptive processing of nutrients (Klasing, 1998). Additionally, larger alimentary tracts could increase storage capacity for ingested food, and thereby facilitate higher levels of food intake (Dykstra and Karasov, 1992; Hammond and Diamond, 1992; McWilliams et al., 1999; Starck, 1999; Starck and Rahmaan, 2003). Furthermore, bigger intestines might provide greater surface area and more capability to absorb nutrients from ingested food (Klasing, 1998). If digesta moves more rapidly through larger and presumably longer digestive tracts, then digestive efficiency could remain equivalent to adult-sized digestive tracts, with comparable or shorter digesta retention times, while enabling higher rates of food intake. However, if digesta flow rates are similar through larger and longer digestive tracts, overall time contents are retained within the gut should be longer as well, which should enhance digestive efficiency, but not permit notably increased food intake (Penry and Jumars, 1986; Martinez del Rio and Karasov, 1990; Karasov and Martinez del Rio, 2007).

Little consideration has been given to the possibility that differential energy expenditure contributes to differences between age classes in stopover duration. In non-migratory house sparrows (Passer domesticus), Chappell et al. (1999) found that 4-month-old juveniles had higher basal metabolic rates (BMR) than adults. Juvenile non-migratory yellow-eyed juncos (Junco phaeonotus) had higher daily energy expenditures, but not higher BMR than breeding adults (Weathers and Sullivan, 1989). In several shorebird species, Lindström (1997) found juveniles had higher BMR during migration compared to after migration. If juveniles migrating in fall have physiological attributes that increase BMR, such as expensive digestive organs (Martin and Fuhrman, 1955; Piersma et al., 1999), then juveniles may be at an energy disadvantage compared to adults at stopovers. Juveniles might require more foraging time to consume enough food to meet elevated maintenance costs, rebuild tissues and accumulate sufficient energy stores to resume migration. To our knowledge, no one has yet tested for BMR differences between age classes of migrant songbirds at stopover.

Our objective was to investigate whether adult and juvenile birds differ in body composition (particularly digestive system size), BMR, and digestive performance during fall migration. We conducted three related studies. First we obtained carcasses of birds killed accidentally from collisions during fall migration through the city of Toronto, Ontario, Canada. We measured body composition of juveniles and adults of four passerine migrants (white-throated sparrow, Zonotrichia albicollis; hermit thrush, Catharus guttatus; Swainson's thrush, Catharus ustulatus; and ovenbird, Seiurus aurocapilla) to test the hypothesis that juvenile birds generally have enlarged digestive systems. All species differed in at least one aspect of primary diet (i.e., frugivorous, insectivorous, or granivorous) or migration distance (i.e., short or long; Van Horn and Donovan, 1994; Jones and Donovan, 1996; Mack and Yong, 2000; Falls and Kopachena, 2010). Second, we measured BMRs of adult and juvenile whitethroated sparrows and Swainson's thrushes captured during fall migration at a stopover site. We hypothesized that due to continued physiological maturation and potentially larger digestive systems, that juveniles would have greater BMRs than adults. Third, we conducted total collection mass balance feeding trials to measure food intake, diet utilization efficiencies, and total assimilated energy of juvenile and adult Swainson's thrushes and white-throated sparrows captured and kept in short-term captivity at the same stopover site. Measuring these parameters of digestive physiology allowed us to determine whether and how juvenile migratory birds benefit from larger digestive organs. We hypothesized that juveniles would assimilate more total energy and have either higher food intake or digestive efficiency than adults, but not both.

#### METHODS

All animal procedures complied with guidelines of the Canadian Council on Animal Care and were approved by the University of Western Ontario Animal Use Sub-committee (Protocol # 2010-020). Live birds were captured under a scientific collection permit from the Canadian Wildlife Service (CA-0255) and carcasses were obtained under a Canadian Wildlife Service salvage permit (SA 0208).

#### Body Composition Analysis of Window Strike Carcasses

Carcasses of hermit thrushes (Catharus guttatus), Swainson's thrushes (Catharus ustulatus), ovenbirds (Seiurus aurocapilla), and white-throated sparrows (Zonotrichia albicollis) were collected by volunteer participants of the Fatal Light Awareness Program (FLAP; www.flap.org) who routinely search sidewalks of downtown Toronto during spring (mid-March to early June) and fall (early August to mid-November) migration seasons for birds injured or killed by collisions with buildings or other structures. Volunteers conduct searches during mornings and record species, location, and collection date of each bird. Bird carcasses for our study were salvaged between 24 August and 11 November during four fall migration seasons (2008–2011), with 95% found during September and October of each year. During our study 88 to 94% of the birds were salvaged during mornings, which followed the nocturnal flights or morning feeding when birds likely perished, so carcasses were in a good condition for analysis. Carcasses were stored in freezers (−20◦C) at the Royal Ontario Museum, Toronto, Ontario, Canada until they were transported on ice to the Advanced Facility for Avian Research, London, Ontario, Canada and stored in a freezer (−30◦C) until dissection.

We thawed carcasses overnight in a refrigerator, brought them to room temperature, and measured fat and wet lean mass using a quantitative magnetic resonance (QMR) body composition analyzer (Echo-MRI-B, Echo Medical Systems, Houston, TX, USA; Guglielmo et al., 2011). QMR of thawed carcasses was previously validated against chemical analysis using bats (McGuire and Guglielmo, 2010), which showed relative accuracies (±3% for wet lean mass and ±11.6% for fat mass) very close to those for live birds (Guglielmo et al., 2011). Seventeen birds prepared for dissection showed obvious signs of decomposition and were discarded. We determined age class by the degree of skull pneumatization (Miller, 1946; Pyle, 1997). Since the sample of carcasses consisted of many more juveniles than adults, in later years we first removed and thawed heads of carcasses for careful examination of the skull so we could preferentially identify adults for dissection and prevent the sample from becoming excessively skewed toward juveniles. We measured unflattened wing length to the nearest mm using a wing ruler and measured tarsus, keel length, and bill (nares to tip) to the nearest 0.01 mm using digital calipers (Pyle, 1997). We dissected and placed proventriculus, gizzard, small intestine, large intestine, pancreas, liver, kidneys, heart, and flight muscle (pectoralis and supracoracoideus) into pre-weighed aluminum dishes. Organ fat was trimmed and placed with the remaining carcass. Digestive tract organs (proventriculus, gizzard, small intestine, and large intestine) were washed in 0.9% NaCl, pressed on paper towel to remove digesta, and re-washed in saline before being placed on weighing tins. We could not reliably locate gonads in their immature or regressed conditions, so therefore we did not determine sex. Remaining carcass components (including excess fat and feathers) were also placed in a weighing tin. We dried samples at 60◦C and weighed to the nearest 0.001 g on a digital balance (Sartorius CP 423S) until mass was constant for 2-days. Total gastrointestinal tract dry mass was calculated by summing dry masses of proventriculus, gizzard, small intestine, and large intestine. Total dry mass was calculated by summing dry masses of all body components (carcass, organs, and muscle).

#### Basal Metabolic Rates

Swainson's thrushes and white-throated sparrows were captured using mist nets at Long Point Bird Observatory (LPBO), Long Point, Ontario, Canada (latitude: 42◦ 34' 58" N; longitude: 80◦ 23' 52" W), between 11 September and 31 October, 2010. Age class was assigned according to plumage and feather molt characteristics and degree of skull pneumatization (Pyle, 1997). Each day, up to four Swainson's thrushes or white-throated sparrows in good condition (signs of visible fat stores and no injuries or signs of molt) were individually held in cages (66 cm × 46 cm × 50 cm) within animal quarters of a specially equipped mobile laboratory. Tarsus length and unflattened wing length were measured as described above. The animal room was temperature controlled (mean temperature = 19.2 ± 0.3◦C) and the birds were exposed to a natural light cycle for the day they were captured. Birds had ad libitum access to food (live mealworms and millet seed for sparrows and live mealworms for thrushes) and water until we removed food 2 h before sunset.

At approximately sunset on the day of capture, up to four fasted birds were weighed to the nearest 0.001 g on an electronic balance (Acculab Vicon-123), and placed into air and lighttight stainless steel canisters with opaque plastic lids (12 cm diameter × 13 cm height, 1.5 L). All canisters were equipped with a perch and attached to an open flow respirometry system. Within a temperature cabinet (PTC-1, Sable Systems), canister temperatures were maintained at 30◦C by a Peltier effect device controller (Pelt 5, Sable Systems), which is within the thermoneutral zone of both species (Yarbrough, 1971; Holmes and Sawyer, 1975). While a colder temperature would more closely approximate natural conditions experienced by birds during fall migration, 30◦C was sufficiently above lower critical temperature to avoid potential confounding effects due to age differences in insulation. Furthermore, 30◦C allowed both species to be measured simultaneously. DrieriteTM (W.A. Hammond Drierite Company, Zenia, USA) removed water from outside air before it was pumped through the chambers at flow rates between 380 and 420 mL min−<sup>1</sup> , which was measured continuously upstream of each chamber by a mass flow meter (Flowbar-8, Sable Systems). Over the entire night (11–13 h), dried excurrent and baseline air were sub-sampled at 10 min intervals using a multiplexer which led to an infrared CO<sup>2</sup> analyzer (CA-2A, Sable Systems) and a fuel cell O<sup>2</sup> analyzer (FC-1B, Sable Systems), to measure carbon dioxide and oxygen, respectively. Between the CO<sup>2</sup> and O<sup>2</sup> analyzers, DrieriteTM and soda lime (EMD chemicals, Cincinnati, USA) scrubbed water and carbon dioxide, respectively, from excurrent air. Both gas analyzers were calibrated with a certified span gas (20.94% O2, 1.000% CO2, balance N2; Praxair, London, ON, Canada) at the beginning of the field season. Analyzers were checked daily for gas concentration readings of dried atmospheric air (dried CO2-free air for O<sup>2</sup> analyzer). Variation of CO<sup>2</sup> analyzer readings of dried atmospheric air was < ±0.003% throughout the field season. The O<sup>2</sup> analyzer was re-spanned using dried CO2-free atmospheric air as a reference when readings of dried atmospheric air differed by more than 0.02% from expected. Flow rates, and CO<sup>2</sup> and O<sup>2</sup> concentrations were recorded with Expedata software (Version 1.1.15, Sable Systems). Before releasing birds in the morning, we sampled 70–140 µL of blood by puncturing the brachial vein of the wing with a 26-gauge needle and collecting blood in heparinized microhematocrit tubes (Fisher Scientific, Pittsburgh, USA). Whole blood was stored 1.5 mL microcentrifuge tubes in a freezer (−20◦C) for molecular sexing.

To calculate metabolic rate, we used LabAnalyst software (Warthog Systems) to select the lowest 3–5 min of carbon dioxide production with coefficients of variation less than two percent. We excluded the first 2 min of each sampling interval to account for transition from the previous channel. To ensure birds were post-absorptive, we only included sampling intervals recorded more than 3 h after sunset, and therefore when birds had fasted for at least 5 h. Using immediately preceding baseline intervals and lag-corrected fractional O<sup>2</sup> and CO<sup>2</sup> concentrations, we calculated VO˙ <sup>2</sup>, VCO ˙ <sup>2</sup> using equations 10.1 and 10.7 in Lighton (2008). Following the recommendation of Lighton (2008), we calculated basal metabolic rates (BMR) by multiplying VO˙ <sup>2</sup> by an oxyjoule equivalent [16 + 5.164 (RQ)].

Following McCabe (2006), we amplified CHD-W and CHD-Z genes directly from whole blood (Bercovich et al., 1999) to identify sex (Griffiths et al., 1998). Briefly, final polymerase chain reaction (PCR) volumes were 25 µL and consisted of 5 µL of a 2% whole blood suspension (Tomasulo et al., 2002), 10 mM Tris-HCl (pH 8.3), 50 mM KCl, 2.5 mM MgCl2, 200µM of each dNTP, and 1µM of P2 and P8 primers described by Griffiths et al. (1998), and 1 U TaqTM DNA polymerase (Takara, Inc.). Using an Eppendorf Mastercycler thermocylcer, PCR conditions were as follows: (a) three cycles of 94◦C for 3 min then 48◦C for 3 min, (b) 94◦C for 4 min, (c) addition of 1 U Taq, (d) 35 cycles of 94◦C for 45 s, 48◦C for 45 s, then 72◦C for 45 s, (e) 72◦C for 5 min. We separated PCR products on 3% agarose gels stained with ethidium bromide (0.5µg/mL) for 60–90 min at 8.5 V/cm, and then observed resulting bands under UV light. One band indicated males, whereas two bands indicated females.

#### Digestive Performance

We obtained migrating adult and juvenile Swainson's thrushes and white-throated sparrows at the Old Cut station of the LPBO between 12 September and 28 October, 2011. We used individuals that had been assigned fat scores by LPBO personnel of 1 or higher on a 0 to 7 scale (Dunn, 2003), and that weighed at least 27 g for thrushes and 22 g for sparrows. Birds were weighed to the nearest 0.001 g on a digital balance (Acculab Vicon-123) and body fat and wet lean mass was measured by QMR. We placed birds in individual cages (66 cm × 46 cm × 50 cm) within an animal room in the research trailer at the field site. Temperature was controlled (20◦C) and the birds were exposed to a natural light cycle that tracked local sunrise and sunset throughout the season.

We provided birds with ad libitum water and a species-specific diet. Food and water cups were elevated to reduce spillage and mixing with excreta during mass balance feeding trials (below). Swainson's thrushes ate a frugivore-based synthetic banana-mash style diet cut into ∼5 mm<sup>3</sup> cubes. This diet was modified from Denslow et al. (1987) and each batch consisted of: one L water, 680 g fully ripe bananas, 37 g wheat germ (Kretschmer, Quaker Oats Co.), 25 g casein (high nitrogen, ultrapure, 12845, Affymetrix, Inc.), 22 g agar (ultrapure, bacteriological, type A, 10906, Affymetrix, Inc.), 7 g vegetable oil (soybean oil, Crisco, The J.M. Smucker Co.), 3.5 g Briggs salt mixture (902834, MP biomedicals, LLC.), and 1.5 g vitamin mix (AIN vitamin mixture 76, 905454, MP biomedicals, LLC.). To encourage Swainson's thrushes to eat the diet, we used a sufficient quantity of red and blue food color (McCormick, Canada) to visually change the diet to a deep-purple color (Boyle, 2009). To further entice Swainson's thrushes to sample the synthetic diet, we provided several thawed, frozen-blueberries on top of the synthetic diet for the first day birds stayed in the animal room. We fed whitethroated sparrows hulled sunflower seeds (200805, Wild Birds Unlimited, Inc.) in order to provide a seed diet that would not produce large quantities of husks.

Birds ate their respective diets for 2-days prior to the total collection mass balance feeding trials to allow birds time to recognize and consume the diet and to pass remnants of food consumed prior to capture. To keep birds as close to their wild condition as possible, we minimized time in captivity prior to feeding trials to limit changes in digestive tract morphology related to captivity and captive diet (Miller, 1975; Levey and Karasov, 1989; Moore and Battley, 2006). Provided that birds ate at least some captive diet, minimal changes in digestive tract morphology likely occurred, given that Pierce and McWilliams (2004) reported similar masses of digestive components among white-throated sparrows fed either ad libitum or restricted diets. We released birds that did not eat the diet or fell below a critical mass (Swainson's thrush < 25 g, 21 released of 41 captured; white-throated sparrow < 21 g, 13 released of 33 captured).

Cages for 2-day total collection mass balance feeding trials were the same as those used to house birds upon arrival at the animal housing room. The smooth-walled collection cages included a smooth half-wall frontal barrier, and a galvanized steel mesh (1.9 cm<sup>2</sup> ) floor, raised (1.9 cm) above a dropping pan lined with a clear plastic sheet. The plastic lining facilitated collections of excreta and uneaten food, which were separated and collected daily, and then frozen (−20◦C) for later analysis. Collections of the diet, excreta, and uneaten food were later dried in a convection oven at 70◦C to constant mass (0.001 g, Sartorius CP 4235). Afterward, dried samples were crushed into powder using a mortar and pestle.

We measured energy content of excreta and food using a Philipson microbomb calorimeter (Gentry Instruments) with benzoic acid standards. Total nitrogen content of excreta and food sub-samples was measured using flow injection analysis at the University of Wisconsin-Madison Soil and Plant Analysis Lab (Verona, WI, USA). Samples were run in duplicate for both analyses with coefficients of variation <4% for energy content and <5% for total nitrogen content.

We calculated the apparent assimilable mass coefficient (AMC<sup>∗</sup> ) and apparent metabolizable energy coefficient (MEC<sup>∗</sup> ), as follows:

$$\begin{aligned} \text{AMC}^\* &= \frac{(Q\_i - Q\_\varepsilon)}{Q\_i} \\ \text{MEC}^\* &= \frac{(GE\_iQ\_i - GE\_\varepsilon Q\_\varepsilon)}{GE\_iQ\_i} \end{aligned}$$

where Q<sup>i</sup> and Q<sup>e</sup> are quantities of dry food intake and excreta production, respectively, and GE<sup>i</sup> and GE<sup>e</sup> are gross energy contents of dry food and excreta, respectively (Karasov, 1990; Guglielmo and Karasov, 1993). We also corrected these apparent utilization efficiencies for nitrogen balance using the following equations:

$$\begin{aligned} \text{AMC}\_{N}^{\*} &= \frac{(Q\_{i} - Q\_{\text{e}} - 3.0 \, (N\_{i}Q\_{i} - N\_{\text{e}}Q\_{\text{e}}))}{Q\_{i}} \\ \text{MEC}\_{N}^{\*} &= \frac{(GE\_{i}Q\_{i} - GE\_{\text{e}}Q\_{\text{e}} - 34.5 \, (N\_{i}Q\_{i} - N\_{\text{e}}Q\_{\text{e}}))}{GE\_{i}Q\_{i}} \end{aligned}$$

where N<sup>i</sup> and N<sup>e</sup> are proportion nitrogen content of dry food and excreta, respectively (Guglielmo et al., 1996). Dry nitrogen intake for the second day of the feeding trial was calculated from:

$$\text{Dry nitrogen intake } = N\_i \times Q\_i$$

Energy deposition during day 2 of the feeding trial was calculated as the sum of 1 QMR wet lean mass × energy density of protein and 1 QMR fat mass x energy density of fat, where the energy densities of protein and fat are 5.3 kJ g−<sup>1</sup> wet mass and 39.6 kJ g <sup>−</sup><sup>1</sup> dry matter, respectively (Jenni and Jenni-Eiermann, 1999). We calculated metabolizable energy intake (MEI) following Guglielmo et al. (1996) as:

$$\text{MEI } = GE\_iQ\_i \times \text{MEC}\_N^\*$$

#### Statistical Analysis

#### Body Composition of Window Strike Carcasses

For each species we used the first principal component of a PCA analysis of tarsus, keel and bill measurements as a size metric (SizePC1). Eigenvalues for SizePC1 always exceeded 1.0. We excluded wing length since adults of many passerine and nearpasserine species may have longer wing lengths than juveniles (Alatalo et al., 1984; Francis and Wood, 1989; Pyle, 1997). We found low loadings (<0.10) for bill length in PC1 for hermit thrush, Swainson's thrush, and ovenbird. Therefore, we recalculated SizePC1 for these three species with only tarsus and keel.

Separately for each species, we tested for age class differences in morphometric length measurements (wing, tarsus, keel, and bill) and SizePC1 using Student's t-tests. We then performed analyses of covariance (ANCOVA) on each body composition component (QMR fat and lean masses, dry organ and muscle masses, gastrointestinal tract dry mass and total dry mass) using age class as the factor and SizePC1 as a covariate. We excluded pancreas from analysis as it was usually in a deteriorated state. Other studies suggest that the pancreas quickly deteriorates after death (Shimizu et al., 1990). Using PCA, we condensed organ and muscle dry mass variables into fewer principal components (OrganPC's) so that we could examine overall relationships among body components and test in a more general way the patterns revealed by univariate analyses. We selected the first two OrganPC axes as dependent variables and tested for differences between age classes while controlling for SizePC1 using multivariate analysis of covariance (MANCOVA).

All mean and least squares mean (LSM) values given are ± SE unless otherwise noted. We performed analyses using IBM SPSS statistics version 20.0.0. We used one-tailed tests where we had reason a priori to predict differences between age classes (i.e., larger juvenile dry masses of proventriculus, gizzard, small intestine, large intestine, liver, and total gastrointestinal tract, as well as larger adult wing lengths). Tests of all other differences, including those involving OrganPC's, used two-tailed tests with α = 0.05. All P-values are unadjusted for multiple comparisons, because we expected associations among some body components, such as among organs of the digestive tract. In doing so, we accepted a higher probability of type I error (false positives) in order to avoid increasing the likelihood of type II error (false negatives).

#### Basal Metabolic Rate

For each species we used ANCOVA to test for the effects of age class and sex on body mass, while including tarsus length as a covariate. We used ANCOVA to test for effects of age class and sex on BMR, while including body mass and minimum daily temperature of the testing day (from Environment Canada's National Climate Data and Information Archive for Long Point, Ontario station; http://climate.weatheroffice.gc.ca/climateData/ canada\_e.html) as covariates. Minimum daily temperature was used as a covariate because variation in previous exposure to temperature may influence BMR (Williams and Tieleman, 2000). Additionally, we used ANCOVA to test for effects of age class on overnight mass loss, while controlling for initial mass and total time spent in respirometry chambers. We tested all two-way interactions and used backward selection to remove non-significant (P > 0.05) terms from analysis. We visually confirmed ANCOVA assumptions for normality and homogeneity of variance/covariance. Body masses and basal metabolic rates were log<sup>10</sup> transformed to account for allometric scaling for all statistical tests associated with BMR. However, untransformed BMR values were used to calculate least-squares means. Unless otherwise stated, values reported are mean ± SE. A Grubb's test was used to check for outliers (Dunn and Clark, 1987). Differences were considered significant at α = 0.05 for two-tailed tests. PASW Statistics (v. 18.0.0) was used to perform statistical tests.

#### Digestive Performance

To ensure that initial body composition of birds retained for feeding trials was not a bias for inclusion, we used ANOVA's to separately test species for differences in body mass, QMR wet lean mass, and QMR fat mass, between birds that were selected for and excluded from feeding trials, while including age class as a factor. Among birds that participated in feeding trials, we tested for differences in arrival date, body mass, QMR wet lean mass, and QMR fat mass at capture between adults and juveniles using t-tests. We tested for age class differences in measurements generated from feeding trials (mass at start of trials, nitrogen balance, dry mass food intake, dry mass excretion, dry mass utilization, energy intake, energy excretion, MEI, 2 day mass change, AMC<sup>∗</sup> , MEC<sup>∗</sup> , AMC<sup>∗</sup> <sup>N</sup>, and MEC<sup>∗</sup> <sup>N</sup>) using t-tests. We used ANCOVA to test for age class differences in dry food intake while controlling for QMR wet lean mass, 2 day mass change while controlling for dry mass intake, energy deposition while controlling for MEI, and separately for QMR wet lean mass deposition while controlling for dry nitrogen intake and nitrogen balance. We performed all statistical tests using IBM SPSS (version 20.0.00) and considered differences significant when P < 0.05. Since we expected associations among measures of digestive function, all P-values were unadjusted for multiple comparisons.

#### RESULTS

#### Body Composition Analysis

We dissected carcasses of 71 white-throated sparrows, 25 hermit thrushes, 37 Swainson's thrushes, and 44 ovenbirds. There were no age class related differences in wing, tarsus, keel, or bill length for any species (**Table S1**). Tarsus, keel, and bill generally produced a single principal component axis representing body size (**Table S2**) and SizePC1 did not differ between adults and juveniles of any species (**Table S1**).

We controlled for body size using SizePC1, regardless of whether the covariate contributed significantly to the ANCOVA. While SizePC1 typically was not a significant covariate when testing for differences in body composition, QMR wet lean mass [F(1, 68) = 13.92, P < 0.001], total dry mass [F(1, 68) = 8.83, P = 0.004], and dry heart mass [F(1, 68) = 7.93, P = 0.006] of white-throated sparrows, and dry flight muscle of white-throated sparrows [F(1, 68) = 37.26, P < 0.001] and Swainson's thrushes [F(1, 34) = 24.21, P < 0.001] increased with SizePC1.

Controlling for SizePC1, there were no differences in total dry mass between adults and juveniles for any species (**Figure 1A**). However, juveniles had 8–9% greater size-corrected wet lean mass measured by QMR than adults in white-throated sparrows [F(1, 68) = 13.18, P = 0.001] and hermit thrushes [F(1, 22) = 10.64, P = 0.004], and there was a tendency for juveniles to have higher wet lean mass than adults among Swainson's thrushes [F(1, 34) = 3.45, P = 0.07, **Figure 1B**]. No significant age differences in QMR fat mass were detected (**Figure 1C**).

Juveniles had larger size-corrected livers (4 species), proventriculi (3 species), gizzards (3 species), small intestines (2 species), and large intestines (1 species) than adults (**Figure 2**, **Table 1**). Additionally, size-corrected hearts were larger in juveniles from two species. Total gastrointestinal tract dry masses (size-corrected) of juveniles were heavier than those of adults among three of four species examined (**Table 2**).

The first two principal components generated by PCA explained 48% or more of the variation in organ dry masses for each species (**Table S3**). In general, OrganPC1 was characterized by positive loadings of nearly all organs. Positive OrganPC2 loadings characterized larger masses of hearts and flight muscles and smaller digestive organs, except for ovenbirds where OrganPC2 represented heavier hearts and lighter flight muscles. Controlling for SizePC1, age classes differed for the combination of OrganPC1 and OrganPC2 in white-throated sparrows [Wilks' λ = 0.83, partial ε <sup>2</sup> = 0.17; F(2, 66) = 6.95, P = 0.002], Swainson's thrushes [Wilks' λ = 0.82, partial ε <sup>2</sup> = 0.18; F(2, 33) = 3.67, P = 0.04], and ovenbirds [Wilks' λ = 0.73, partial ε <sup>2</sup> = 0.27; F(2, 40) = 7.48, P = 0.002], but not in hermit thrushes [Wilks' λ = 0.82, partial ε <sup>2</sup> = 0.19; F(2, 20) = 2.28, P = 0.13; **Figure 3**].

Compared with adults, OrganPC1 scores of juvenile whitethroated sparrows [F(1, 67) = 10.50, P = 0.002], Swainson's thrushes [F(1, 34) = 7.55, P = 0.01], and ovenbirds [F(1, 41) = 8.94, P = 0.005] were higher. OrganPC2 scores of juvenile white-throated sparrows approached being significantly lower

0.01 (\*\*\*). Comparisons of Pro, Giz, SI, LI, and Liv used one-tailed tests predicting larger dry masses among juveniles, whereas comparisons of Kid, Hrt, and F Mus used two-tailed tests.

than adults [F(1, 67) = 3.84, P = 0.054], and OrganPC2 scores of juvenile ovenbirds were very close to being significantly higher than adults [F(1, 41) = 3.954, P = 0.053]. The directions

TABLE 1 | Results of ANCOVA comparing dry organ masses of adult and juvenile migrants from four species.


*Covariate is SizePC1, a principal component containing tarsus and keel measurements for hermit thrush, Swainson's thrush, and ovenbird, and tarsus, keel, and bill measurements for white-throated sparrow. Organs and muscles are: Pro, proventriculus; Giz, gizzard; SI, small intestine; LI, large intestine; Liv, liver; Kid, kidney; Hrt, heart; and F Mus, flight muscle (pectoralis and supracoracoideus). Comparisons of Pro, Giz, SI, LI, and Liv are one-tailed tests predicting larger dry masses among juveniles, whereas comparisons of Kid, Hrt, and F Mus are two-tailed tests.*

TABLE 2 | Least square means (±SE) dry masses of total gastrointestinal tracts from adults (Ad) and juveniles (Juv) of four migrant bird species collected during fall migration (2008–2011).


*One-tailed ANCOVA using SizePC1 as a covariate tested for larger total gastrointestinal dry masses of juveniles. For hermit thrushes, Swainson's thrushes, and ovenbirds, SizePC1 is a principal component containing tarsus and keel measurements. For whitethroated sparrows SizePC1 is a principal component containing tarsus, keel, and bill measurements.*

of principal component loadings imply that overall, juvenile white-throated sparrows, Swainson's thrushes, and ovenbirds had heavier digestive organs and hearts, but lighter flight muscles than adults.

#### Basal Metabolic Rates

Morphometric measurements and associated statistical analyses of birds used for BMR measurements are presented in **Table S4**. Generally, juvenile male Swainson's thrushes were heavier than adults and juvenile females, and male white-throated sparrows were heavier than females, regardless of age class.

One adult Swainson's thrush and one juvenile white-throated sparrow did not satisfy the BMR selection criteria of at least 3 min of CO<sup>2</sup> production with a coefficient of variation below two percent, and were excluded. Furthermore, we detected one outlier BMR value for one adult Swainson's thrush [T(51) = 3.32, P < 0.05], so it was removed. In ANCOVA of the effects of age class, sex, minimum daily temperature, and body mass on BMR of both species, all interaction terms were not significant. For Swainson's thrushes, minimum daily temperature [F(1, 44) = 0.85, P = 0.36] and sex [F(1, 46) = 2.62, P = 0.11] did not influence BMR, and were sequentially removed from the ANCOVA model. In Swainson's thrushes, after controlling for body mass [F(1, 47) = 8.19, P = 0.006], juveniles had higher BMR than adults [least squares means untransformed BMR: juvenile = 0.38 ± 0.005 W; adult = 0.36 ± 0.007 W; F(1, 47) = 6.72, P = 0.013, **Figure 4A**]. For white-throated sparrows, sex [F(1, 51) = 1.22, P = 0.27] and minimum daily temperature [F(1, 52) = 1.69, P = 0.20] were removed from the ANCOVA model. After controlling for body mass [F(1, 53) = 105.90, P < 0.001] juvenile whitethroated sparrows had higher BMR than adults [least squares means untransfomed BMR: juvenile = 0.40 ± 0.004 W; adult = 0.38 ± 0.005 W; F(1, 53) = 12.45, P = 0.001, **Figure 4B**].

In Swainson's thrushes, when controlling for initial mass [F(1, 46) = 4.06, P = 0.050] and total overnight time [F(1, 46) = 6.69, P = 0.013], there were no age class related differences in overnight mass loss [F(1, 46) = 0.07, P = 0.80]. Similarly, for white-throated sparrows, when controlling for initial mass [F(1, 52) = 6.27, P = 0.015] and total overnight time [F(1, 52) = 0.96, P = 0.33], there were no age class related differences in mass lost overnight [F(1, 52) = 0.26, P = 0.62].

#### Digestive Performance

Within each species, body mass and composition at capture did not differ between age classes or between birds that were selected for or excluded from feeding trials (**Table S5**). Twenty Swainson's thrushes (7 adults and 13 juveniles) and twenty white-throated sparrows (10 adults and 10 juveniles) completed 2-day total collection mass balance feeding trials. In these birds, adults and juveniles had similar arrival dates [thrushes: t(18) = −1.69, P = 0.11; sparrows: t(18) = 0.76, P = 0.46], body masses [thrushes: t(18) = 1.47, P = 0.16; sparrows: t(18) = −0.92, P = 0.37], QMR wet lean masses [thrushes: t(14) = −0.50, P = 0.63; sparrows: t(18) = −1.16, P = 0.26], and QMR fat masses [thrushes: t(14) = 1.30, P = 0.22; sparrows: t(18) = −0.16, P = 0.87] at capture (**Table 3**).

At the start of a feeding trial, body masses of juveniles and adults were similar for both Swainson's thrushes [t(18) = −0.12, P = 0.91] and white-throated sparrows [t(18) = −0.044, P = 0.97; **Table 3**]. All Swainson's thrushes and white-throated sparrows were in positive nitrogen balance during the 2-day feeding trial. Nitrogen balance did not differ between adults and juveniles [thrushes: t(18) = −1.12, P = 0.28; sparrows: t(18) = −0.55, P = 0.59; **Table 4**], and nitrogen balance increased with dry food intake for both species (**Figure 5**). At the end of the feeding trial, body mass [thrushes: t(18) = −0.58, P = 0.57; sparrows:

species that included dry masses of the proventriculus, gizzard, small intestine, large intestine, liver, kidney, heart, and flight muscle (pectoralis and supracoracoideus).

t(18) = −0.42, P = 0.68], QMR wet lean mass [thrushes: t(18) = −0.73, P = 0.47; sparrows: t(17) = −0.18, P = 0.86], and QMR fat mass [thrushes: t(18) = −0.45, P = 0.66; sparrows: t(17) = 0.12, P = 0.91] were similar for juveniles and adults of both species (**Table 3**).

Juvenile Swainson's thrushes consumed more food than adults [t(18) = −2.30, P = 0.033; **Table 4**]. When controlling for QMR wet lean mass during the feeding trial [F(1, 17) = 10.36, P = 0.005], juveniles still consumed more food (LSM dry food intake = 18.11 ± 0.91 g) than adults [LSM dry food intake = 14.75 ± 1.25 g; F(1, 17) = 4.61, P = 0.046, **Figure 6A**]. Juvenile thrushes also had greater dry matter excretion [t(18) = −2.21, P = 0.04], dry matter utilization [t(18) = −2.11, P = 0.049], energy intake [t(18) = −2.42, P = 0.026], energy excretion [t(18) = −2.32, P = 0.032], and metabolizable energy intake [t(18) = −2.22, P = 0.039] than adults (**Table 4**). Adult and juvenile thrushes had similar utilization efficiencies as measured by AMC<sup>∗</sup> [t(18) = −0.78, P = 0.45], AMC<sup>∗</sup> <sup>N</sup> [t(18) = −0.82, P = 0.43], MEC<sup>∗</sup> [t(18) = −0.38, P = 0.71], and MEC<sup>∗</sup> <sup>N</sup> [t(18) = −0.40, P = 0.70; **Table 4**].

There was no difference in 2-day body mass change between adult (0.31 ± 0.32 g) and juvenile (0.77 ± 0.27 g) Swainson's thrushes during feeding trials [t(18) = −1.08, P = 0.30]. However, when controlling for dry food intake [F(1, 17) = 135.21, P < 0.001], adult thrushes (LSM body mass change = 0.92 ± 0.13 g) gained more mass than juveniles [LSM body mass change = 0.44 ± 0.09 g; F(1, 17) = 7.86, P = 0.012; **Figure 7A**]. While 2-day body mass change increased with nitrogen balance [F(1, 17) = 13.29, P = 0.002], no difference existed between age classes [F(1, 17) = 0.18, P = 0.68]. Energy deposited (as fat mass plus wet lean mass) increased with metabolizable energy intake [MEI; F(1, 17) = 36.48, P < 0.001], and was similar for adults and juveniles [F(1, 17) = 0.53, P = 0.48; **Figure 8A**]. There was no relationship between dry nitrogen intake and deposition of QMR wet lean mass [F(1, 17) = 0.073, P = 0.79], or between nitrogen balance and deposition of QMR wet lean mass [F(1, 17) = 0.24, P = 0.63], with no effect of age class on either [F(1, 17) = 0.18, P = 0.68; F(1, 17) = 0.043, P = 0.84].

Adult and juvenile white-throated sparrows consumed similar amounts of food [t(18) = −0.60, P = 0.56; **Table 4**]. Controlling for QMR wet lean mass during the feeding trial [F(1, 16) = 1.05, P = 0.32] revealed an interaction between age class and QMR wet lean mass [F(1, 16) = 15.60, P = 0.001], which complicated analysis. Adults increased food intake with lean mass, whereas juveniles had high food intake regardless of lean mass [F(1, 16) = 15.87, P = 0.001; **Figure 6B**]. Dry matter excretion [t(18) = −0.59, P = 0.56], dry matter utilization [t(18) = −0.50, P = 0.63],

(±0.289) \* log<sup>10</sup> mass, *r* <sup>2</sup> = 0.234; sparrows log<sup>10</sup> BMR = −2.009 (±0.214) + 1.135 (±0.153) \* log<sup>10</sup> mass, *r* <sup>2</sup> = 0.744.

energy intake [t(18) = −0.60, P = 0.56], energy excretion [t(18) = −0.75, P = 0.46], and metabolizable energy intake [t(18) = −0.14, P = 0.89] were similar for adults and juvenile sparrows (**Table 4**). Adult and juvenile sparrows had similar AMC<sup>∗</sup> [t(18) = 0.14, P = 0.89], AMC<sup>∗</sup> <sup>N</sup> [t(18) = 0.20, P = 0.85], MEC<sup>∗</sup> [t(18) = 0.59, P = 0.56], and MEC<sup>∗</sup> <sup>N</sup> [t(18) = 0.63, P = 0.54; **Table 4**].

There was no difference in 2-day body mass change between adult (0.62 ± 0.37 g) and juvenile sparrows (0.88 ± 0.32 g) during feeding trials [t(18) = −0.53, P = 0.61]. Controlling for food intake [F(1, 17) = 28.43, P < 0.001] did not change this finding, with adults (LSM body mass change = 0.74 ± 0.22 g) and juveniles (LSM body mass change = 0.76 ± 0.22 g) gaining similar amounts of mass [F(1, 17) = 0.006, P = 0.94; **Figure 7B**]. Similarly, while 2-day body mass change increased with nitrogen balance [F(1, 17) = 22.30, P < 0.001], there was no difference between age classes [F(1, 17) = 0.026, P = 0.87]. Energy deposited (as fat mass and wet lean mass) increased with metabolizable energy intake [MEI; F(1, 16) = 40.36, P < 0.001],

TABLE 3 | Body mass, body composition, wing chord, and arrival date for migrating adult and juvenile Swainson's thrushes and white-throated sparrows that completed 2-day total collection mass balance feeding trials, which began 2-days after birds were captured on stopover at Long Point, Ontario, fall 2011.


*Body mass and body composition measured at time of capture and at dawn following an overnight fast at both the start and end of feeding trials. Lean and fat masses measured by Quantitative Magnetic Resonance (QMR). Unless indicated otherwise by letters (an* = *4; <sup>b</sup>n* = *12; <sup>c</sup>n* = *9), samples sizes for adult and juvenile Swainson's thrushes were n* = *7 and n* = *13, respectively and n* = *10 for both adult and juvenile white-throated sparrows. All values are means* ± *SE and did not differ between adults and juveniles of either species at P* < *0.05.*

and was similar for adult and juvenile sparrows [F(1, 16) = 0.23, P = 0.64; **Figure 8B**]. There were no relationships between either dry nitrogen intake and deposition of QMR wet lean mass [F(1, 16) = 0.15, P = 0.71], or nitrogen balance and deposition of QMR wet lean mass [F(1, 16) = 0.16, P = 0.69], and no effects of age class [F(1, 16) = 1.74, P = 0.21; F(1, 16) = 1.69, P = 0.21].

#### DISCUSSION

Our findings that juvenile birds have larger digestive organs, higher metabolic rates, and higher rates of food intake support the hypothesis that juvenile birds making their first migratory journey are physiologically different from adults in ways that influence their performance and success. It is often conjectured that young birds have poor survival during migration because they lack foraging, navigation, and predator-avoidance skills, which require cognitive development through experience and learning to improve. However, evidence is accumulating that physiological maturation of flight muscles, skeletal components, the immune system, and the digestive system may continue during migration, potentially affecting metabolic rates, energy budgets, diet selection, and ultimately migration behavior. The apparently sub-optimal physiological characteristics of juveniles may be viewed as constraints derived from the requirement to migrate before development is fully complete. However, some TABLE 4 | Measures of digestive efficiency and nitrogen balance from a 2-day total collection mass balance feeding trial for migrating adult and juvenile Swainson's thrushes and white-throated sparrows captured on stopover at Long Point, Ontario, fall 2011.


*All values are means, SE. Sample sizes for adult and juvenile Swainson's thrushes were n* = *7 and n* = *13, respectively and n* = *10 for both adult and juvenile white-throated sparrows. For intraspecific comparisons between adults and juveniles, an* \* *in the adult column indicates differences significant at P* < *0.05.*

of their physiological characteristics are undoubtedly adaptive responses by juveniles to maximize fitness given the added challenges they face. For example, enlarged guts may allow juveniles to exploit different diets, consume more food when it is available, or employ a different digestive strategy than adults. Here we show that enlarged guts may come at the cost of increased energy expenditure. If bodies express ecology as put forward by Piersma and van Gils (2011), then differences in body composition suggest that juveniles and adults have different ecologies during migration. Moreover, if "migration takes guts" (McWilliams and Karasov, 2005), then it appears to take extra guts for juvenile birds. The larger digestive organs and higher BMRs of juveniles imply age-related disparities in foraging ecology and energy budgets. More studies of adult and juvenile birds during migratory stopovers are needed to reveal how foraging ecology and energetics of these age classes differ.

#### Body Composition

Our analysis of carcasses from building collisions showed that migrating juvenile songbirds were the same structural size and total dry mass as adults, which is consistent with most passerines attaining adult structural size before or shortly after fledging (Alatalo and Lundberg, 1986; Richner, 1989; Kaiser and Lindell,

2007; Verspoor et al., 2007). In contrast to some previous reports (Alatalo et al., 1984; Francis and Wood, 1989; Pyle, 1997), adults had similar wing lengths as juveniles. Juveniles from all species examined had larger livers and at least one larger component of the gastrointestinal tract than adults. These larger digestive organs appeared to contribute to greater wet lean masses of juveniles of the two short-distance migrant species (whitethroated sparrows and hermit thrushes). On the other hand, while juvenile Swainson's thrushes and ovenbirds had heavier digestive organs than adults, QMR wet lean mass did not differ between age classes. This may be because adults of both of these long-distance migrant species tended to have larger flight muscles than juveniles.

and dashed lines represent juveniles: thrushes nitrogen balance = 0.0075

balance = 0.0251 (±0.0221) + 0.0087 (±0.0025) \* dry food intake, *r*

(±0.0556) + 0.0058 (±0.0029) \* dry food intake, *r*

<sup>2</sup> = 0.26; sparrows nitrogen

<sup>2</sup> = 0.60.

Previous studies have reported similar results to ours. Hume and Biebach (1996) found migrating juvenile garden warblers (Sylvia borin) had larger dry digestive tract mass than adults. Similarly, juvenile Western sandpipers had larger digestive organs, such as small intestine and liver, than adults during fall migration (Guglielmo and Williams, 2003; Stein et al., 2005). In an analysis of organ masses of nocturnal passerine migrants killed at a television tower during fall migration, Graber and Graber (1962)reported that in general, juveniles appeared to have larger livers and smaller pectoral muscles than adults. We found similar results for the four passerine species examined, suggesting that heavier livers and digestive tract organs in juveniles during fall migration is a widespread pattern among passerine species.

Larger digestive organs of juvenile migrants may be a remnant of their developmental past. That is, juvenile migrants may still be undergoing maturation or remodeling processes during fall migration, having not yet reached a fully adult condition. As nestlings, digestive organs account for a higher proportion of overall body mass than they do in adults, but the proportion declines as nestlings grow and develop (Bech and Østnes, 1999; Vézina et al., 2009). Alternatively, larger guts in juveniles may have an adaptive function to match the refueling conditions that juveniles face at stopovers, such as lower foraging success due to competition with adults or poor prey capture skills (favoring digestive efficiency maximization), a bulky less digestible diet such as fruits, or greater energy expenditure (for maintenance, thermoregulation or activity).

#### Basal Metabolic Rates

Our measurements of BMR for adult Swainson's thrushes (0.013 W g−<sup>1</sup> ) and adult white-throated sparrows (0.015 W

g −1 ) were very similar to BMRs previously measured resting metabolic rates (RMR) of adult Swainson's thrushes (0.017 W g-1 equivalent) during summer (Holmes and Sawyer, 1975), and BMR of adult white-throated sparrows (0.014 W g−<sup>1</sup> equivalent) during winter (Yarbrough, 1971). Unlike our study, Holmes and Sawyer (1975) measured RMR during the active phase and during summer, which likely explains their slightly greater values. In our study BMRs of juvenile whitethroated sparrows and Swainson's thrushes were consistently about 6% greater than adults. The species differ in diets, migration distance, and evolutionary lineages (Mack and Yong, 2000; Falls and Kopachena, 2010), suggesting that higher BMR in juveniles may be widespread amongst migrant passerines. Previous studies have found BMR or daily energy expenditure (DEE) to be higher for non-migratory juvenile songbirds during a timeframe comparable to pre-migration for

circles and dashed lines represent juveniles: thrushes energy deposition =

<sup>2</sup> = 0.72; sparrows energy

<sup>2</sup> = 0.76.

−29.36 (±9.23) + 0.471 (±0.088) \* MEI, *r*

deposition = −48.01 (±14.16) + 0.919 (±0.184) \* MEI, *r*

fall migrants (Weathers and Sullivan, 1989; Chappell et al., 1999).

There are many possible physiological mechanisms for higher BMR in juveniles, including differences in organ sizes, rates of protein turnover, and continuing maturation. Lean mass, and digestive organs in particular, are energetically expensive to maintain (Martin and Fuhrman, 1955; Piersma et al., 1999), and larger lean mass can contribute to higher basal metabolic rates (Daan et al., 1990; Piersma et al., 1996; Hammond and Diamond, 1997; Williams and Tieleman, 2000; Battley et al., 2001a). Metabolic rates of digestive organs are among the highest of all organs (Krebs, 1950). For example, in European starlings (Sturnus vulgaris), in vitro basal oxygen uptake of liver was 2.66 ml O<sup>2</sup> g <sup>−</sup><sup>1</sup> h −1 compared with 0.66 ml O<sup>2</sup> g <sup>−</sup><sup>1</sup> h −1 for pectoralis muscle (Scott and Evans, 1992). One reason organs are expensive is that they have high rates of carbon turnover (Bauchinger and McWilliams, 2009), which likely reflect high rates of protein turnover (Carleton and Martínez del Rio, 2005; Bauchinger et al., 2010). Both protein turnover (Waterlow, 1980; Hawkins, 1991) and carbon turnover (Tieszen et al., 1983) rates are linked to metabolic rates. In particular, liver accounts for higher rates of fractional protein synthesis (Murphy and Taruscio, 1995), nitrogen incorporation rates (Muñoz-Garcia et al., 2012), and oxygen consumption due to protein synthesis (Rolfe and Brown, 1997) compared to skeletal muscle. High energy costs of organs mean they can have disproportionate effects on BMR; liver mass alone (Bech and Østnes, 1999), or together with intestines, kidney, and heart or breast muscle can explain at least half of the variation in BMR (Konarzewski and Diamond, 1995; Chappell et al., 1999). However, in an interspecific study of birds, Daan et al. (1990) instead found kidney and heart together explained 50% of variation in BMR. Burness et al. (1998) only found larger kidneys, but not liver, intestine, or heart, to be related to higher BMR among breeding tree swallows (Tachycineta bicolor).

Tissues of juvenile birds may continue to mature during migration, adding energy costs. The skeleton is incompletely formed as shown from the second layer of bone that grows in the skull of passerines during fall migration (Hamel et al., 1983; Wiley and Piper, 1992), and perhaps other bones continue to grow and mature. Similarly the immune system may be maturing in juvenile migrants, as indicated by their large and active bursa of Fabricius (Warner and Szenberg, 1964; Glick, 2000; Ratcliffe, 2006). However, adult and juvenile passerines had similar levels of constitutive (Owen and Moore, 2006; Palacios et al., 2009; Girard et al., 2011) and induced (Palacios et al., 2009) immune function prior to or during fall migration. Juvenile migrants may have higher rates of total body protein turnover than adults. It is well-documented that young rats (Rattus norvegicus) have higher protein turnover rates than adults (Yousef and Johnson, 1970; Millward and Garlick, 1972). However, Hobson and Clark (1992) found no age difference in carbon turnover rates of Japanese quail (Coturnix japonica).

Regardless of the specific mechanism(s) involved, juvenile migrants will have greater inherent energy expenditure at stopover than adults due to higher BMR. Even if daytime refueling rates are similar between age classes (Woodrey and McCabe and Guglielmo Migration in Juvenile Birds

Moore, 1997; Yong et al., 1998; Jones et al., 2002; Leist, 2007; Seewagen et al., 2013), higher BMR in juveniles will have an effect during nighttime roosting at ambient temperatures both above and below (if heat produced by BMR does not replace costs of thermogenesis) the lower critical temperature. Although we did not find a difference in the overnight loss of mass between adults and juveniles, higher nocturnal energy expenditure (due to higher BMR) could cause juveniles to have lower total daily refueling rates, and consequently longer stopover durations than adults. Higher BMR has been associated with higher field metabolic rate (FMR; Nagy, 1987; Daan et al., 1990; Koteja, 1991), but the full extent of this relationship is not clear (Ricklefs et al., 1996; Meerlo et al., 1997; Nagy, 2005). If this is the case, then juveniles will also have higher FMR than adults. Juveniles could also have greater thermoregulatory costs attributable to plumage of lower insulative quality, higher activity costs due to more frequent antagonistic interactions (Weathers and Sullivan, 1989), or greater protein requirements due to higher rates of tissue remodeling (Fisher, 1972; Bairlein, 1987). Any or all of these factors could further increase food requirements and reduce refueling performance. Additional measurements of these factors in migrating birds is needed. However, at the least, our results show that juveniles may endure costs imposed by higher BMR, which could make fall migration even more challenging for these first time migrants.

#### Digestive Performance

Digestive physiology of juvenile Swainson's thrushes clearly differed from that of adults, with juveniles using higher food intake to assimilate more energy. Conversely, we found no age class related differences in digestive physiology among whitethroated sparrows. Larger thrushes consumed more food, but at the same lean mass, juveniles consumed more food than adults. Presumably, juveniles achieved higher food intake due to larger digestive organs, which we observed in carcasses of conspecifics sampled in Toronto about 170 km northeast of Long Point. However, juveniles did not have higher digestive efficiency, which agrees with predictions of van Gils et al. (2008). Juvenile thrushes ingested more energy and appeared to pursue a more rate maximizing digestive strategy relative to adults by using their larger guts to accommodate extra food while maintaining a similar digestive efficiency. Swainson's thrushes in our study had lower MEC<sup>∗</sup> (0.53) compared to other passerines consuming natural fruit diets (0.64; Karasov, 1990). Compared to thrushes from our study, American robins (Turdus migratorius) had similar MEC<sup>∗</sup> values consuming fruit (0.55), but not when consuming a comparable banana mash diet (0.77; Levey and Karasov, 1989). Juvenile thrushes converted food into body stores less efficiently than adults, suggesting that they expend more energy than adults either for maintenance (BMR, see above), thermoregulation or activity. Although we did not measure activity levels, it is possible that juveniles were more active in cages than adults.

Juvenile and adult white-throated sparrows consumed comparable amounts of food. However, juveniles consumed a similar amount of food regardless of lean mass, whereas among adults, heavier sparrows consumed more food. This suggests that juveniles may maintain a high digestive capacity under all conditions, whereas food intake in adults may be limited by low digestive capacity. The presumed larger digestive organs of juveniles may have permitted them to consume more food at low lean body masses. Juveniles did not appear to use their larger digestive organs to pursue a more efficiency-maximizing digestive strategy as juveniles and adults had similar digestive efficiencies. White-throated sparrows that consumed sunflower seeds in our experiment had lower MEC<sup>∗</sup> values (0.62) than several passerines that consumed either cultivated (0.80) or wild (0.75) seeds (Karasov, 1990). Both juvenile and adult white-throated sparrows converted food into body stores with similar efficiencies. Thus, the age-related difference in BMR we measured did not translate into difference in how efficiently juveniles and adults converted food into body mass. However, as we did not monitor activity levels of birds while in captivity, it might be possible that adults were more active compared to juveniles. Among recently captured blackcaps (Sylvia atricapilla), adults displayed greater nocturnal restlessness than juveniles (Berthold, 1996), but Ketterson and Nolan (1985) noted no difference in nightly zugunrhue activity among adult and juvenile dark-eyed juncos (Junco hyemalis).

Why might juvenile Swainson's thrushes show increased food intake relative to adults, whereas juvenile whitethroated sparrows did not? During fall migration, Swainson's thrushes typically consume a predominantly fruit-based diet, supplemented with insects (Jones and Donovan, 1996; Mack and Yong, 2000; Parrish, 2000), whereas white-throated sparrows tend to consume a more varied diet that includes a combination of mostly seeds, fruits, and some insects (Falls and Kopachena, 2010). Generally, birds consuming fruits have lower digestive efficiencies than when consuming insects or seeds (Castro et al., 1989; Karasov, 1990). Additionally, due to the lower energy density and protein content of fruits relative to insects and seeds (Johnson et al., 1985; Moermond and Denslow, 1985; Karasov, 1990), frugivorous species need to consume greater quantities of food to satisfy daily energy and protein requirements (Aamidor et al., 2011). Accordingly, in our experiment, Swainson's thrushes consumed a synthetic fruit-based diet that was about 85% water, whereas sparrows were fed shelled sunflower seeds that were about 5% water. Over the 2-day feeding trials a typical 28.2 g adult thrush consumed 94.0 ± 10.4 g wet food, whereas a 23.7 g adult white-throated sparrow ate 8.46 ± 0.66 g wet food. Perhaps bulk of the synthetic fruit-based diet pushed digestive systems of thrushes nearer to the limits of food intake capacity, while digestive systems of sparrows might have had sufficient capacity to readily accommodate the energy-dense diet of sunflower seeds. Furthermore, the sunflower seeds were shelled, which require less handling time and have more metabolizable energy than intact, wild equivalents (Karasov, 1990). Consequently, juvenile sparrows may have met their energy and protein requirements from consuming a smaller volume of food than they otherwise would in a natural setting with a more varied diet. Along with many other passerines migrating during fall, white-throated sparrows include fruits as part of their diet (Parrish, 1997, 2000). During fall migration stopover on Block Island, Rhode Island, 92% of fecal samples from white-throated sparrows contained fruit (Parrish, 1997). Larger digestive organs of juvenile white-throated sparrows suggest they may be capable of accommodating the additional bulk from a diet that consists of more fruit. We propose that juvenile white-throated sparrows may rely more heavily on fruit to meet dietary needs at stopover than do adults, and under those conditions they may show similar patterns to thrushes.

Whereas, diets of passerines during migration are understood in general terms, whether and how diet compositions of juveniles and adults differ is not. Wheelwright (1986) found higher proportions of fruits in stomachs of juvenile than of adult American robins (Turdus migratorius), but it was not clear if the samples were taken during migration. Future studies of diet compositions of juvenile and adult passerines at migratory stopover sites may reveal how diets of juveniles and adults compare, and whether diets of juveniles consist of relatively more fruit. Use of DNA barcoding to identify species origins of diet remains within fecal samples may facilitate analysis (Hebert et al., 2003; CBOL Plant Working Group, 2009). Isotopic analysis of breath or tissues of migrants may also provide information on dietary differences (Hobson, 2011). Clearly, foraging ecologies of juvenile and adult migrant passerines during stopover require further study.

# CONCLUSION

Larger digestive organs of juvenile migrants can both result from and enable higher food intake, while larger organs can contribute to higher maintenance metabolic rates. The collective findings of this study imply that disparities in foraging ecology, physiology, or both prompt juveniles to consume greater quantities of food, which subsequently causes hypertrophy of digestive organs. Specifically, the diets of juvenile songbirds could be more dilute, consisting of relatively more fruit in comparison to adults. Indeed, Fruit is voluminous, high in water, and has lower energy densities and protein contents compared to insects and seeds (Moermond and Denslow, 1985; Karasov, 1990). Juveniles that consume relatively more fruit may consequently consume larger volumes to meet protein and energy requirements, which may cause juveniles to enlarge their digestive organs in response. However, the fact that livers were consistently larger in juveniles of all species suggests that the story may not just be about diet quality, and that more total energy and nutrients may need to be processed by juveniles following digestion and absorption (Starck, 1999; Starck and Rahmaan, 2003; Battley and Piersma, 2005). Thus, an alternative explanation is that high cellular activities and protein turnover rates associated with the development and maturation of the skeletal, immune, muscular, and other systems of juveniles could contribute to higher metabolic rates. To meet higher demands for energy juveniles would need to consume more food, which could cause hypertrophy of digestive organs. In this manner, it would be possible for physiology to influence ecology, as foraging of juveniles in the wild would likely be affected. Longer-term captive feeding studies could differentiate between ecology and physiology as drivers of large digestive systems in juvenile migrants.

### DATA AVAILABILITY STATEMENT

The data used in the analyses are available at Guglielmo, Christopher; McCabe, Brendan (2019), Migration takes extra guts for juvenile songbirds: energetics and digestive physiology during the first journey, Dryad, Dataset, https://doi.org/10.5061/ dryad.p5hqbzkkh.

# ETHICS STATEMENT

The animal study was reviewed and approved by Animal Care Committee, University of Western Ontario.

# AUTHOR CONTRIBUTIONS

BM participated in experimental design, conducted experiments and laboratory analyses, collected data, analyzed data, and cowrote the manuscript. CG participated in experimental design, obtained animal care protocols and permits, assisted with experiments, laboratory analyses and data analysis, and co-wrote the manuscript.

# FUNDING

Funding for this study provided by grants to CG from the Natural Sciences and Engineering Research Council of Canada (05245-2015 RGPIN), the Canada Foundation for Innovation (11743), an Ontario Early Researcher Award, and the Ontario Research Fund.

#### ACKNOWLEDGMENTS

We thank Michael Mesure and Paloma Plant of the Fatal Light Awareness Program and Mark Peck of the Royal Ontario Museum for coordinating transfer of frozen carcasses previously salvaged by volunteers. We thank the volunteers and staff of the Long Point Bird Observatory for assistance with field studies and Mike Burrell for his expertise on aging birds. We thank Alexander Gerson and Liam McGuire for advice on respirometry and statistics. We are very grateful to Lisa Kennedy for help with the feeding trials. We extend a special thanks to all volunteers (Brieann Adair, Dillon Chung, Lisa Cohen, Patrick David, Melissa Giamou, Alexander Macmillan, Angela Paric, Anna Pauer, Sara Rea, and Dustin Yen) that helped perform dissections.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00381/full#supplementary-material

## REFERENCES


in a long-distance migrating shorebird. Physiol. Biochem. Zool. 72, 405–415. doi: 10.1086/316680


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 McCabe and Guglielmo. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Mechanisms and Consequences of Partial Migration in Insects

Myles H. M. Menz 1,2,3,4 \*, Don R. Reynolds 5,6, Boya Gao7,8, Gao Hu<sup>8</sup> , Jason W. Chapman7,8,9 and Karl R. Wotton<sup>7</sup>

*<sup>1</sup> Department of Migration, Max Planck Institute of Animal Behavior, Radolfzell, Germany, <sup>2</sup> Centre for the Advanced Study of Collective Behaviour, University of Konstanz, Konstanz, Germany, <sup>3</sup> Department of Biology, University of Konstanz, Konstanz, Germany, <sup>4</sup> School of Biological Sciences, The University of Western Australia, Crawley, WA, Australia, <sup>5</sup> Natural Resources Institute, University of Greenwich, Chatham, United Kingdom, <sup>6</sup> Rothamsted Research, Harpenden, United Kingdom, <sup>7</sup> Centre for Ecology and Conservation, University of Exeter, Cornwall Campus, Penryn, United Kingdom, <sup>8</sup> College of Plant Protection, Nanjing Agricultural University, Nanjing, China, <sup>9</sup> Environment and Sustainability Institute, University of Exeter, Cornwall Campus, Penryn, United Kingdom*

Partial migration, where a proportion of a population migrates, while other individuals remain resident, is widespread across most migratory lineages. However, the mechanisms driving individual differences in migratory tendency are still relatively poorly understood in most taxa, but may be influenced by morphological, physiological, and behavioral traits, controlled by phenotypic plasticity and the underlying genetic complex. Insects differ from vertebrates in that partial migration is often associated with pronounced morphological differences between migratory and resident phenotypes, such as wing presence or length. In contrast, the mechanisms influencing migratory tendency in wing-monomorphic insects is less clear. Insects are the most abundant and diverse group of terrestrial migrants, with trillions of animals moving across the globe annually, and understanding the drivers and extent of partial migration across populations will have considerable implications for ecosystem services, such as the management of pests and the conservation of threatened or beneficial species. Here, we present an overview of our current but incomplete knowledge of partial migration in insects. We discuss the factors that lead to the maintenance of partial migration within populations, and the conditions that may influence individual decision making, particularly in the context of individual fitness and reproductive tradeoffs. Finally, we highlight current gaps in knowledge and areas of future research that should prove fruitful in understanding the ecological and evolutionary drivers, and consequences of partial migration in insects.

Keywords: animal migration, flight capacity, intraspecific variation, insect migration, migratory potential, movement ecology, wing polymorphism

# INTRODUCTION

Vast numbers of animals migrate seasonally across large geographic scales, usually due to shifts in resource availability—indeed, the importance of habitat ephemerality as a primary driver of insect migration has long been recognized (Southwood, 1962; Denno et al., 1991; Dingle, 2014)—and also in response to increased predation, parasitism and pathogen pressure (Altizer et al., 2011; Chapman et al., 2015). Migrants connect habitats and populations through their annual movements, but also have profound effects on ecosystem processes such as nutrient fluxes and the provision of ecosystem

#### Edited by:

*Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway*

#### Reviewed by:

*Stephen Baillie Malcolm, Western Michigan University, United States Michael T. Hallworth, Northeast Climate Adaptation Science Center, United States*

\*Correspondence:

*Myles H. M. Menz mmenz@ab.mpg.de*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *29 May 2019* Accepted: *08 October 2019* Published: *24 October 2019*

#### Citation:

*Menz MHM, Reynolds DR, Gao B, Hu G, Chapman JW and Wotton KR (2019) Mechanisms and Consequences of Partial Migration in Insects. Front. Ecol. Evol. 7:403. doi: 10.3389/fevo.2019.00403* services (Bauer and Hoye, 2014; Bauer et al., 2017; Wotton et al., 2019). There is no universally accepted definition of migration, and many authors take a restricted "vertebratecentric" view and define migration as round-trip movements between discrete "breeding" and "non-breeding" locations, which inevitably excludes most insect examples from this definition. In our review, we adopt a broader view of migration, based on the behavioral definition of Kennedy and Dingle, defined as any movements which are persistent and straightened-out, and characterized by some (temporary) inhibition of behaviors associated with feeding or reproduction (Dingle, 1996, 2014; Dingle and Drake, 2007; Chapman and Drake, 2019). The function of migratory movements is, of course, spatial relocation, but this shift to new habitats is best viewed as a population-level outcome of the individual behaviors. In other words, migration is defined as a behavioral process, with the consequences explained at the ecological or evolutionary level. Other movement ecology researchers might categorize some of the examples we provide in our review as dispersal instead of migration, but we adopt this broad view in order to discuss insect examples in the context of the established framework for partial migration.

"Partial migration," whereby part of a population remains resident while the rest migrates, is a common phenomenon among migratory species (Lack, 1943; Lundberg, 1988; Dingle, 1996, 2014; Chapman et al., 2011; Kokko, 2011; Shaw and Levin, 2011), and has been reported from a wide range of taxa such as fish (Chapman et al., 2012), birds (Nilsson et al., 2011), and mammals (Mysterud et al., 2011; Berg et al., 2019). However, the term has been little used in studies of insects and other invertebrates (but see Hansson and Hylander, 2009; Attisano et al., 2013; Slager and Malcolm, 2015; Dällenbach et al., 2018; Ruiz Vargas et al., 2018; Vander Zanden et al., 2018). Partial migration arises through intra-population variation in migratory tendency, may be driven by physiological, morphological, or behavioral variation (Chapman et al., 2011), and has been proposed to be an early evolutionary stage in the transition to full migration (Berthold, 2001) but, in insects, it could also mark a reversion to residency. Frequency distributions of insect flight duration are often sharply skewed, with short flights significantly more common than long flights (Davis, 1980). Therefore, if short migratory flights become adaptive because overwintering in situ in temperate areas becomes favorable due to warming conditions, short fliers could swiftly replace longdistance migrants in the population. Changes in the frequency of morphs indicates that there must be strong selection for longdistance insect migration to be maintained in the face of the higher mortality rates, physiological costs, and delays to breeding associated with migration (Roff and Fairbairn, 1991; Zera and Denno, 1997; Fox and Dennis, 2010; Bonte et al., 2012; Chapman et al., 2015).

The mechanisms influencing the incidence of partial migration within populations are not well-understood. Three types of partial migration are often recognized in the literature, "breeding," where a population remains together during the non-breeding season, but migrants and residents breed separately, "non-breeding" where a population breeds in the same habitat, but migrants and residents spend the non-breeding season separately and "skipped-breeding" where a population spends the non-breeding season in one location, but part of the population remains and does not breed, while the other migrates to breed (Chapman et al., 2011; Shaw and Levin, 2011; Dingle, 2014). However, these definitions are based on organisms with separate breeding and non-breeding areas, which is often inapplicable to migratory insects, many of which continuously breed year-round with several generations required to complete the migratory cycle (Flockhart et al., 2013; Stefanescu et al., 2013; Chapman et al., 2015). Furthermore, in contrast to vertebrates, migratory insects can show extreme morphological variation between generations, with the production of macropterous morphs, which are long-winged and can undertake migratory flights, brachypterous or micropterous morphs which are short-winged and sedentary, and apterous morphs which are wingless. Short-winged and wingless morphs are unable to migrate and are hereafter referred to collectively as short-winged forms (Johnson, 1969; Roff and Fairbairn, 1991, 2007; Gatehouse and Zhang, 1995; Zera and Denno, 1997; Dingle, 2014). In other cases, the ability to migrate may depend on traits other than wing-length, such as size of the flight muscles or fuel reserves. Thus, whether an individual is migratory or not may come from a "decision" based upon the context in which it finds itself or be pre-determined, for example maternally, as can occur in Hemiptera (Gatehouse, 1994; Vellichirammal et al., 2017).

Here we present an overview of what is known about the incidence and maintenance of partial migration, which is widespread in insects. We contrast the phenomenon in insects and vertebrates, and examine the current terminology used to define the types of partial migration. Knowledge gaps, and fruitful areas for future research, are highlighted. Finally, we argue that insects, with their developmental plasticity and short generation times, provide excellent subjects for investigating the mechanisms that influence migratory decisions.

# PARTIAL MIGRATION IN INSECTS

Insect immature stages (eggs, larvae, nymphs, and pupae) are typically comparatively sedentary compared to adults, so inter-individual differences in migration propensity are generally a feature of the adult stage. Partial migration has been described in a number of insect species from a broad range of orders, such as Hemiptera, Orthoptera, Lepidoptera, Diptera, and Odonata (**Figure 1**), but much of the work on variation in migratory potential has focused on wingdimorphic hemipterans (Johnson, 1969; Gatehouse and Zhang, 1995; Zera and Denno, 1997; Roff and Fairbairn, 2007; Dingle, 2014). In all cases, it is assumed that an individual will either migrate or remain more-or-less sedentary in one or another life stage in order to increase its overall fitness.

In contrast to most vertebrates, migrant insects are relatively short-lived and usually undergo multiple generations within a year (Chapman et al., 2015). Consequently, defining partial migration into the three main types developed primarily for vertebrates (Chapman et al., 2011; Shaw and Levin, 2011) is inappropriate for insects, particularly due to their short generation times. Some authors have adapted the existing


FIGURE 1 | Examples of insect species where partial migration has been studied. In all cases presented here, the migratory cycle consists of a number of generations annually and the proportion of migrants and non-migrants may change between generations. Images: *N. lugens*, Y. He; *O. fasciatus*, J. Gallagher (CC BY 2.0); *Gryllus firmus*, D. Roff; *D. erippus*, G. Ruellan (CC BY 3.0); *E. balteatus*, W. Hawkes; *A. junius*, M. Ostrowski (CC BY-SA 2.0).

FIGURE 2 | Partial migration in insects. (A) Migrants (colored purple) enter a breeding ground from a previous breeding, overwintering, or aestivation ground. In a northern temperate system this ground may be at lower latitudes. The migrants oviposit creating generation 1 that consists of a varying degree of migrants (purple) and non-migrants (green) depending on the conditions encountered (photoperiod, temperature, resources, population density). Non-migrants act as resident breeders (for example summer generations of monarch butterflies and migratory hoverflies or various flightless morphs of polymorphic species), producing additional generations in the breeding ground that may also consist of varying amounts of migrants or non-migrants. In contrast, the migrants move away from the breeding area becoming *temporally* separated from the non-migrants, a situation termed *sequential partial migration* (Ruiz Vargas et al., 2018). (B) *Migration to breeding grounds* (Class I; Johnson, 1969). Migrants may enter a second breeding ground and the process depicted in (A) continues (and may do so over multiple additional areas). Separate breeding grounds may vary through latitudinal or altitudinal and seasonal clines, and the relative fitness of each morph may vary between successive areas depending on conditions. Migration with multiple phases typically consists of relatively short-lived insects, or morphs of a particular species. Continuously breeding species such as the painted lady butterfly may cycle through this system, while other species may only undertake part of it, for example, aphids, planthoppers and spring migrations of hoverflies (see text for other examples). (C) In some cases, a species may switch to *migration to overwintering or aestivation grounds* (Class III; Johnson, 1969). Insects with long-distance migration are often relatively long lived, examples include autumn morphs of migrant hoverflies, monarch butterflies and bogong moths. However, overwintering may also take place *within* the breeding grounds without migration (colored black), such as for migratory hoverflies and the green darner dragonfly. Typically, migration or breeding continues again in the spring.

definitions to suit insects, coining terms such as "sequential partial migration," where migratory and non-migratory animals are separated temporally, rather than spatially (Ruiz Vargas et al., 2018), or "alternate migration," to reflect that some migratory individuals switch from a migratory to a nonmigratory strategy upon encountering a resident population (Vander Zanden et al., 2018). In many cases, sequential partial migration appears apt, as the proportion of migratory and non-migratory individuals change between generations and this definition reflects the multi-generational aspect of insect migration (**Figure 2**). Broad definitions, such as that a population with 1–99% migrants can be considered as partially migratory (Chapman et al., 2011), will obviously promote the inclusion of insect taxa. A number of hypotheses have been raised to understand the mechanisms driving individual variability in migratory tendency, and these are discussed further below.

### Morphological Variation Between Migrants and Non-migrants

In comparison to vertebrates, insects can show extreme wing polymorphisms between migratory and non-migratory phenotypes. Consequently, partial migration in insects needs to be considered in terms of the contrast between wing-monomorphic and wing-polymorphic species, as there are likely to be different mechanisms and selection pressures acting on these two fundamentally different types. As most work on the trade-offs between migration and residency has been conducted on wing-polymorphic species, comparing migratory and sedentary phenotypes in wing-monomorphic insects may prove useful for elucidating the underlying mechanisms, but such studies are rare (Tigreros and Davidowitz, 2019).

In birds, there are many examples of differences in body size between migrants and residents, with the latter often being larger, possibly due to larger-bodied individuals having a greater physiological tolerance to overwintering (Ketterson and Nolan, 1976; Belthoff and Gauthreaux, 1991) or the ability to endure periods of low resource availability (Boyle, 2008; Jahn et al., 2010; Chapman et al., 2011). In insects, migrants are often larger than non-migrants (Roff and Fairbairn, 2007), a pattern that has been demonstrated for wing-dimorphic species such as the milkweed bug (Oncopeltus fasciatus) (Hegmann and Dingle, 1982), and gerrid (water-strider) bugs (Fairbairn, 1992), as well as wingmonomorphic species (Altizer and Davis, 2010). Differences in wing loading and morphology have also been reported between migratory and non-migratory monarch (Danaus plexippus) and southern monarch (D. erippus) butterflies, with migrants having larger, more pointed wings and higher wing loads than residents (Dockx, 2007; Altizer and Davis, 2010; Slager and Malcolm, 2015; Vander Zanden et al., 2018), which should result in more fuelefficient flight (Roff and Fairbairn, 1991; Rankin and Burchsted, 1992). Interestingly, no differences in wing morphology were reported between overwintering adults and migrants of the marmalade hoverfly (Episyrphus balteatus; Raymond et al., 2014b). There was also no difference in resting metabolic rate between sexes in E. balteatus, but the smaller females were shown to have higher evaporative water loss than the larger males (Tomlinson and Menz, 2015).

### Reproduction or Migration?

The costs of migration in relation to reproductive fitness differ between the sexes such that some authors consider that males and females should be considered separately (Johnson, 1969; Gatehouse and Zhang, 1995); here we primarily discuss the relationship as it relates to females. Insect migration is often considered in the context of the "oogenesis-flight syndrome," which posits a trade-off between migration and reproduction (Johnson, 1969; Gatehouse and Zhang, 1995; Dingle, 1996). Development of flight muscles, and migratory flight itself, are energetically costly (Dudley, 1995; Dingle, 2014) and, whereas non-migrants can immediately allocate resources to breeding, migrating individuals will often spend time in reproductive diapause (Johnson, 1969; Rankin and Burchsted, 1992). Migration often occurs pre-reproductively (Gatehouse, 1994; Gatehouse and Zhang, 1995), with reproductive maturity being linked to the cessation of migration, or even the termination of diapause following a period of aestivation or overwintering (Johnson, 1969). However, there is sometimes a more nuanced relationship between reproduction and development of the flight apparatus in wing-monomorphic insects (Rankin et al., 1986; Sappington and Showers, 1992), with some species even migrating with fully-developed oocytes (May et al., 2017; Tigreros and Davidowitz, 2019).

The trade-off between migration and reproduction can be modulated by resource availability in both wing-monomorphic and dimorphic species (Roff and Fairbairn, 2007; Ruiz Vargas et al., 2018). In wing-dimorphic species, the production of macropterous individuals is often determined in early developmental stages or even maternally (Gatehouse, 1994; Wilson, 1995; Ogawa and Miura, 2014; Vellichirammal et al., 2017). Host quality strongly influences wing-morph in brown planthoppers (Nilaparvata lugens); upon colonization of a new resource patch, there is an increased proportion of short-winged individuals, which are unable to migrate but have a greater reproductive potential than the macropterous morph (Lin et al., 2018). As the rice crop matures there is an increase in the proportion of the macropterous form, which can migrate to colonize new rice fields, but the proportion of long-winged individuals within a population can vary between seasons and years (Hu et al., 2017). In aphids, the production of winged morphs may be influenced by environmental conditions such as crowding, decreasing food quality, or the presence of predators (Müller et al., 2001). In wing-monomorphic species, or in long-winged individuals of dimorphic species, the ability to respond to changes in resource availability and switch between a migratory and non-migratory state or vice versa may be driven by differences in physiology, such as the ability to reallocate nutrients from flight to reproduction. Indeed, Attisano et al. (2013) demonstrated that resident female milkweed bugs showed a higher level of oosorption (where females resorb nutrients from developing oocytes thus favoring survival over current reproduction) than did migrants.

#### Density Dependence

It has been predicted that an increased proportion of migrants should occur in populations at higher densities (Chapman et al., 2011). In insects, partial migration may allow individuals that move to breed to avoid the negative consequences of resource competition (Taylor and Taylor, 1983; Dingle, 1996). For example, in the planthoppers N. lugens and Sogatella furcifera, an increased proportion of long-winged individuals may be produced at high densities (Matsumura, 1996; Lin et al., 2018). Similarly, crowding can promote the production of winged offspring in aphids (Johnson, 1969; Müller et al., 2001). The lower fecundity typically found in winged forms typically is an example of the tradeoff between the colonization of new habitats and reproductive output.

### Predation and Parasitism Risk

Partial migration may confer some reduction in the risk of predation or parasitism, by movement into an enemy free space, resulting in improved survival for migrants. However, the role of trophic interactions has received relatively little attention in the partial migration literature (Chapman et al., 2011) and has rarely been studied in migratory insects (Altizer et al., 2011; Chapman et al., 2015). Nonetheless, there is evidence that migration can reduce the prevalence of infection from the protozoan parasite, Ophryocystis elektroscirrha in monarchs (Bartel et al., 2011; Altizer et al., 2015; Flockhart et al., 2018), with resident populations having higher infection rates than migrant populations (Satterfield et al., 2015, 2016, 2018), providing evidence of "migratory escape" (Altizer et al., 2011) from contaminated environments.

# The Evolution, Expression, and Maintenance of Partial Migration

Migratory flight tendency has been shown to be heritable in a broad range of insect species, indicating a strong genetic component to migratory behavior (Wilson, 1995; Dingle, 1996, 2014; May et al., 2017; Dällenbach et al., 2018). The capacity of insects to form migrants or non-migrants from within the same population could potentially be determined by genetic polymorphisms, for example alleles that influence flight or timing (Niitepõld et al., 2009; Hut et al., 2013; Zhan et al., 2014) and/or the expression of environmentally-induced phenotypic plasticity. While evidence for a solely genetically determined difference is lacking for partial migration, phenotypically plastic pathways are a widespread feature of insect life histories (Nijhout, 1999) and are likely to provide the predominant mechanisms allowing migrants to switch forms, an idea strengthened by the low level of genetic differentiation and phylogeographic structuring found within many partial migrant populations (Mun et al., 1999; Freeland et al., 2003; Raymond et al., 2013; Zhan et al., 2014).

How discrete migratory states within a population are maintained is unclear, but two hypotheses have been proposed (Chapman et al., 2011). One possibility is the attainment of an evolutionary stable state, where the fitness of each form is balanced by frequency-dependent selection. For example, in wing dimorphic insects where the more fecund flightless form is balanced by the colonizing abilities of the migrant morph (Roff, 1994; Zera and Denno, 1997). Alternatively, the fitness benefits of either morph may occur as a result of conditional strategies, were the decision to migrate is based upon gaining the highest fitness possible under certain circumstances and a balancing of fitness is not necessary (Chapman et al., 2011). The generally short life span of insect migrants and their higher reliance on favorable meteorological conditions for migration (Alerstam et al., 2011; Hu et al., 2016) highlights the importance for selecting the optimal strategy in any given situation. Migratory hoverflies, such as E. balteatus, for example, may migrate south to warmer climes (Wotton et al., 2019) but are also capable of sedentary overwintering behavior as adults, larvae, or pupae (Raymond et al., 2014a), an adaptation that presumably increases their fitness over attempting to migrate in unfavorable conditions (also see Vander Zanden et al., 2018).

The inheritance and phenotypic expression of migratory states has been investigated in both wing polymorphic (Fairbairn and Yadlowski, 1997; Roff et al., 1997) and monomorphic (Kent and Rankin, 2001) insects and interpreted in the context of the "threshold model": a quantitative genetic model for the evolution of polygenic, dichotomous traits (Roff, 1996). Under this model, a normally distributed trait, called the liability, underlies the expression of the migratory dimorphism and a threshold determines the developmental trajectory—in this case migrant or non-migrant. If the liability exceeds the threshold then the individual takes one path, say migration, if not it becomes sedentary. In the case of wing polymorphism, it is hypothesized that the liability for wing production may be governed by hormone profiles at a particular larval stage: in larvae where levels exceed the threshold (conceivably controlled by levels of hormone receptors among other factors) the flightless morph is formed (Oostra et al., 2011; Roff, 2011). An additional consideration is that threshold traits also vary with environmental factors such as temperature, photoperiod, and density (Hondelmann and Poehling, 2007; Guerra and Reppert, 2013). A more realistic model—the environmental threshold model—allows for both genetic variation, and for individual or environmental conditions to modify the threshold and the liability (Roff, 1994; Wikelski et al., 2006; Hallworth et al., 2018; see Pulido, 2011 for a full consideration of the model and its implications for partial migration) and therefore has the potential to provide a comprehensive framework for a deeper understanding of partial migration in insects.

# Ecological Implications of Partial Migration in Insects

Insects are the most abundant and speciose terrestrial migrants, with trillions of individuals undertaking movements annually (Holland et al., 2006; Chapman et al., 2015; Hu et al., 2016). Additionally, many migratory insect species are important agricultural pests (Drake and Gatehouse, 1995), or are beneficial—as pollinators or natural enemies (Wotton et al., 2019) or as food for other animals (Krauel et al., 2015; Warrant et al., 2016). Consequently, understanding the incidence and mechanisms involved in the regulation of partial migration in insect populations has significant implications for ecosystem functioning and species management. Models based on predatorprey dynamics and interactions with environmental conditions have been developed to study the ecosystem effects of partial migration in fish (Brodersen et al., 2008, 2011), and similar approaches may be considered for insects, particularly in the context of nutrient transfer between trophic levels and across landscapes. Furthermore, understanding the factors influencing the level of migration within populations may allow for the implementation of more realistic species management strategies.

# Future Directions and Gaps in Knowledge

Despite the deficiency of research investigating the mechanisms driving partial migration in insects, the phenomenon evidently occurs in numerous species, and there are exciting opportunities for research into the evolution and ecology of the phenomenon. Insects are excellent model systems; they are relatively small, easily maintained, and can be manipulated in a laboratory environment. The opportunity for identifying new partial migration study systems will be facilitated by the huge diversity of migratory insect species and their broad range of life histories.

Little is known about the influence of anthropogenic landscape change on partial migration in insect populations. There is evidence that landscape alterations can readily lead to an increase in the propensity for residency in migratory insects, usually in response to favorable conditions, such as the availability of food resources. For example, increased planting of tropical milkweed (Asclepias curassavica) in Florida has led to an increase in residency in monarchs, but residents suffer from increased parasitism compared to migrants (Satterfield et al., 2015). Urbanization can also increase the propensity for residency or overwintering through the provision of winter refugia or foraging resources, such as garden flowers. Luder et al. (2018) demonstrated that migratory hoverflies appeared earlier in the season in urban areas compared to agricultural areas, indicating that cities may provide favorable conditions for overwintering. Warming temperatures have also led to an increase in overwintering of migratory species in the UK, such as the red admiral butterfly (Vanessa atalanta), although much of the population still immigrates to the UK each spring (Sparks et al., 2005; Fox and Dennis, 2010). Fairly simple laboratory experiments could be used to shed light on whether warming or constant temperatures, or increased food constancy, influences the migratory propensity in wingmonomorphic insects.

Tracking the migratory behavior of insects in the field is difficult, primarily due to their small size and sheer numbers. Individual tracking of insects to determine migratory decisions has been hindered because the majority of species fall well below the body size required to carry active transmitters (Wikelski et al., 2006; Kissling et al., 2014; Knight et al., 2019). Consequently, many studies investigating insect migratory behavior which may be relevant to partial migration have been conducted in the laboratory, using proxy measures for migratory potential, such as flight duration and activity (Minter et al., 2018). Tethered flight experiments have proven useful for determining migratory tendency in a range of insect species (Dällenbach et al., 2018; Minter et al., 2018; Naranjo, 2019). However, the further miniaturization of individual tracking technology will provide exciting opportunities to understand the drivers of partial migration and the mechanisms that influence individual decision-making. The use of intrinsic markers, such as stable isotopes, has proven useful for elucidating the origin of migratory insects (Hobson et al., 2012; Flockhart et al., 2013; Hallworth et al., 2018) and is applicable to a range of species. Recent advances in molecular techniques, including metabarcoding of pollen carried on the bodies of insects also shows great promise (Suchan et al., 2019). Techniques using intrinsic markers, where the utility is not limited by the size of the insect, will likely prove key in understanding patterns of partial migration in many insect taxa.

#### AUTHOR CONTRIBUTIONS

MM and KW wrote the first draft of the manuscript. All authors contributed to the development of ideas and writing of the manuscript.

#### REFERENCES


#### FUNDING

This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie Grant Agreement No. 795568 (awarded to MM). Funding to KW was provided by the Royal Society through a University Research Fellowship (UF150126). Rothamsted Research receives grant-aided support from the United Kingdom Biotechnology and Biological Sciences Research Council (BBSRC). Funding to GH and BG were provided by the National Natural Science Foundation of China (31822043) and the Natural Science Foundation of Jiangsu Province (BK20170026).

trophic dynamics. Oikos 120, 1838–1846. doi: 10.1111/j.1600-0706.2011. 19433.x


juncos (Junco hyemalis hyemalis). Ecology 57, 679–693. doi: 10.2307/19 36182


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Menz, Reynolds, Gao, Hu, Chapman and Wotton. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Influences of Personality on Ungulate Migration and Management

Robert Found† and Colleen Cassady St. Clair\*

*Department of Biological Sciences, University of Alberta, Edmonton, AB, Canada*

#### Edited by:

*Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway*

#### Reviewed by:

*Paul Cross, United States Geological Survey (USGS), United States Ivar Herfindal, Norwegian University of Science and Technology, Norway*

\*Correspondence:

*Colleen Cassady St. Clair cstclair@ualberta.ca*

†Present address: *Robert Found, Parks Canada, Elk Island National Park, Fort Saskatchewan, AB, Canada*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *01 May 2019* Accepted: *25 October 2019* Published: *27 November 2019*

#### Citation:

*Found R and St. Clair CC (2019) Influences of Personality on Ungulate Migration and Management. Front. Ecol. Evol. 7:438. doi: 10.3389/fevo.2019.00438* Loss of migratory behavior in ungulates often occurs with habituation to people to cause several challenges for wildlife managers, particularly in protected and urban areas. Aversive conditioning to increase ungulate wariness toward people could be an important tool for managing wildlife conflicts, but it is frequently thwarted by variation in responsiveness among individuals, an aspect of personality that is currently little understood by managers. In our paper, we describe the potential role of personality in the ecological progression associated with habituation, loss of migration, and human-wildlife conflict in ungulates. We do so by (a) synthesizing our prior work on two populations of wild elk (*Cervus canadensis*) living in national parks in the Canadian Rocky Mountains, (b) using it to articulate a conceptual model to explain how anthropogenic changes in landscapes favor bolder individuals, and (c) showing how targeted use of aversive conditioning could limit the advantages to bold individuals that promote residency. Our review showed how bolder elk, defined by a combination of seven separate personality metrics on a bold-shy continuum, are three times more likely to forego migration, but are also quicker to learn by association, whether via the provision or cessation of aversive conditioning. Differences in personality may relate to cognitive flexibility, which we measured with limb use preferences, to imbue bolder elk with more rapid responses to changing environments. In our conceptual model, we show how four ecological drivers comprised by interactions with humans, predators and conspecifics, in addition to changes in forage, favor bolder elk that are more likely to adopt a resident migratory tactic. We also explain how bold personalities could result from behavioral flexibility, genetic differences, or gene-environment interactions, each of which could be moderated by frequency-dependent payoffs to individuals. We suggest that managers could limit the prevalence of bold, resident ungulates by targeting bolder individuals with active and specific aversive conditioning, while minimizing anthropogenic food sources in predator refugia. A better understanding of personality in wildlife could support more proactive strategies to limit habituation and encourage migration and other keystone behaviors in changing landscapes.

Keywords: behavioral flexibility, habituation, human-wildlife conflict, personality, ungulates

Diverse species of ungulates exhibit seasonal migration with potential benefits of increased foraging opportunities (Albon and Langvatn, 1992), reduced predation (Hebblewhite and Merrill, 2009) and avoidance of parasites (Altizer et al., 2011). Recently, ungulate populations around the world demonstrate reduced tendencies to migrate with profound effects on associated ecosystems (Bolger et al., 2008; Tucker et al., 2018). Such declines have been documented for wildebeest (Conochaetes spp.) in Africa (Morrison and Bolger, 2012), Mongolian gazelles (Procapra gutturosa) in Asia (Ito et al., 2005), moose (Alces alces) in Europe (Singh et al., 2012), and elk (Cervus canadensis) in North America (Hebblewhite et al., 2006). As ungulates become more sedentary, they often become hyperabundant, overconsume vegetation, disrupt historic predator-prey relationships, and otherwise alter ecosystem functions (White and Ward, 2010). Problems with overabundance are often greatest in protected areas, where humans present little risk, but ample reward in the form of anthropogenic food (Thompson and Henderson, 1998) and protection from predators (Berger, 2004). The combination of sedentary behavior and the absence of predators has been recognized as a threat to the health and sustainability of ungulate populations for almost 100 years (Murie, 1934), but it increasingly characterizes populations of wild ungulates around the world (Bolger et al., 2008; Polfus and Krausman, 2012; Tucker et al., 2018).

In the sequence of ecosystem change that occurs when ungulate prey are decoupled from their predators, habituation by ungulates to people appears to play a pivotal role (Thompson and Henderson, 1998; Whittaker and Knight, 1998). The process of habituation is described simply as a learned behavioral response or waning physiological response to stimuli that lack fitness consequences (Blumstein, 2016). Habituation was extensively studied by learning theorists over the past century and is a well-known contributor to human-wildlife conflict in diverse settings that range from crop damage by birds to urbanizing carnivores and ungulates (Conover, 2001). Habituation by ungulates frequently results in conflict, particularly in protected areas where it can compromise human safety (Thompson and Henderson, 1998; Kloppers et al., 2005). In those areas, ungulates can rapidly lose their historical wariness to people, allowing them to reduce energetically costly responses, such as escape behavior (Gates and Hudson, 1978) and vigilance (Shannon et al., 2014). Increasing habituation by deer species is occurring around the world in urban or urbanizing areas that exclude human hunting and predators (Honda et al., 2018).

The capacity to distinguish between aversive and benign forms of similar stimuli appears to be a critical component of the habituation process (Bejder et al., 2009) with fitness consequences suggested by its occurrence in examples as diverse as carnivores (Ohta et al., 2012), birds (Mackay et al., 2014) and insects (Davis and Heslop, 2004). Although habituated behavior occurs in diverse taxa, prey seem to be more prone to habituation than predators, which generally show a greater innate avoidance of people (Thompson and Henderson, 1998; Berger, 2004), perhaps because habituation helps prey species escape predation. In many ungulate species, proclivity to habituate often results in a positive feedback loop whereby predator refugia formed in human-disturbed areas select for individuals that habituate more readily, increasing, in turn, their capacity to exploit the benefits of the refugium (Polfus and Krausman, 2012). Even in songbirds, there appears to be a positive feedback between the tendency to habituate to people and the tendency to exploit urban areas (Atwell et al., 2012).

Exploitation of urban areas by wildlife is often addressed with the management tools of hazing, which is short-term deterrence, and aversive conditioning, which relies upon associative learning (Hopkins et al., 2014). Aversive conditioning is a systematic method of modifying an individual's behavior by imposing evolutionarily-relevant negative consequences on individuals exhibiting undesirable behaviors (Domjan, 2014), such as a willingness to tolerate close approaches by people or use of human-dominated spaces. Aversive conditioning may consist of chasing animals, firing projectiles, or emitting loud noises and has been used to restore wariness in black bears (Ursus americanus; Mazur, 2010), elk (Kloppers et al., 2005), wolves (Canis lupus; Hawley et al., 2009), and coyotes (Canis latrans; Bonnell and Breck, 2017). Human hunting may not lead to avoidance behavior, especially when it is highly targeted and immediately lethal, if it limits negative consequences experienced by other animals. However, the efficacy of aversive conditioning can also be limited by substantial variation among individuals in responsiveness to the aversive stimuli and duration of learned responses. This variation in responses along with subtle differences in context can determine whether individuals sensitize or habituate to stimuli that are intended by managers to be aversive (Blumstein, 2016).

Similar inter-individual variation in behavioral tendencies across contexts has been described for hundreds of species and is defined as personality (Gosling, 2001), behavioral syndromes (Sih et al., 2004), or temperament (Réale et al., 2007), all of which relate to the degree of plasticity in behavior that stems from a combination of genes, development, environment, and experience (Stamps, 2016). Individual variation in the tendency to habituate to people has high relevance to the management of migratory ungulates, but also to wildlife management more generally. For example, habituated individuals are often associated with hyper-abundance, ecological damage, and human-wildlife conflict (Polfus and Krausman, 2012), but they may also be more likely to spread zoonotic diseases (Murray et al., 2015) or lead conspecifics into adopting similar behavior (Modlmeier et al., 2014).

Relationships between the tendency to habituate and more general responsiveness to environmental stimuli suggest an underlying difference in behavioral flexibility. Such variation might stem partly from the degree of cerebral lateralization, defined as specialization of neural tasks to different brain hemispheres to speed neural processing and reaction times (Rogers, 2000; Vallortigara, 2006). Lateralization is familiar to humans as handedness; the strength of hand preference correlates with cognitive speed and efficiency in domains ranging from athletics to academics (Bisazza et al., 1998). In many animals, greater lateralization translates into greater speed in detecting and responding to predators (Brown et al., 2007). However, the greater speed of cognitive processing for familiar behavioral routines comes at a price, because strongly lateralized animals appear to have less flexibility in adjusting behavior to changing environments (Porac and Searleman, 2006; Carlier et al., 2011). Associations with response times potentially make the measurement of laterality a powerful complement to studies of personality, particularly in animals that express routine behaviors with which it can be easily quantified.

Behavioral variation consistent with definitions of personality have been described in both wild ungulates (Réale et al., 2000) and domesticated species (Wesley et al., 2012), but there has been no generalized effort to explore how personality traits affect ungulate migration and management. That application is overdue considering that livestock owners have recognized and used this variation for centuries to select for more docile animals. In the first section below, we synthesize our past work showing that behavioral syndromes can be quantified in wild, habituated elk (Found and St. Clair, 2016) that reside in the protected areas and mountain townsites of Banff and Jasper, Canada. Next, we develop a conceptual model to show how environmental changes wrought by four ecological drivers increase the benefits of bolder behavior, which favors the individuals that are more likely to use the resident migratory tactic. We explain the mechanisms by which directional selection for bolder individuals could occur and how it is limited by frequency dependence, which may also limit the correlation between personality and migratory tactic. We conclude by showing how greater acknowledgment of behavioral variation and explicit targeting of bold behaviors could increase the efficacy of aversive conditioning to manage both habituation and migration in wild ungulates.

#### REVIEW OF PERSONALITY AND MIGRATION IN WILD ELK

Between 2010 and 2013, we studied the impacts of individual behavioral variation on habituation and migratory choices in wild, adult, female elk near the townsites of Banff 51◦ 10′N 115◦ 34′W pop. est. 7,850) and Jasper (52◦ 52′N, 118◦ 04′W, pop. est. 4,500) AB, Canada, each contained in a national park of the same name. Both townsite areas exhibit high levels of human disturbance that reduce predatory activity in their vicinity (Paquet et al., 1996; Goldberg et al., 2014), while providing anthropogenic and natural foraging opportunities (McKenzie, 2001). Elk in Jasper are partially separated into three herds with only one making extensive use of the townsite each winter (Found, 2015). The elk in Banff annually form a single large over-wintering herd comprised of both migrant and resident elk that is centered on the townsite. It forms a predator refugium that is readily apparent when comparing the locations of radiocollared elk with the snow-tracked paths of wolves (Canis lupus) and their known kill sites of elk (Found, 2015). Wolves are the main predator of adult elk in Jasper in winter (Dekker et al., 1995) although cougar (Puma concolor) also hunt near the townsite of Banff (Kortello et al., 2007) and grizzly bears (Ursus arctos) are an additional important predator of elk fawns in late spring (Hamer and Herrero, 1991).

Populations of wild elk in both Banff and Jasper exhibited behavioral syndromes that we quantified with a composite metric of personality based on seven different traits that were intercorrelated and consistently expressed within individuals and among years (Found and St. Clair, 2016). We used multivariate statistical techniques to delineate these behavioral types along intra-population gradients we defined as "shy" to "bold." We use these terms as labels to connote broad suites of traits (Wilson et al., 1994), but acknowledge there is high variation in their use and interpretation by others (Carter et al., 2013). In our system, bolder elk were characterized by lower flight response distances, reduced responsiveness to sounds, occupancy of more peripheral positions within groups, greater exploration of novel objects, increased vigilance, social dominance over shyer conspecifics, and a greater frequency of leading other elk to new habitats (Found and St. Clair, 2016). We used several of these metrics with captive elk of known birthdates to show that personality was not influenced by age (Found and St. Clair, 2016).

Using these metrics to define behavioral types made it possible to determine that elk with bolder personalities were also more likely to adopt non-migratory, resident strategies. Specifically, resident elk were more exploratory, had lower flight response distances, and higher mean dominance rankings (Found and St. Clair, 2016). In both years and both study populations, individual personality scores from a multi-variate gradient measured in winter were a significant predictor of migratory status in the following summer. In fact, after dividing the elk population at median values for our composite metric into bold and shy halves, bold residents outnumbered shy residents with a 3:1 ratio, whereas bold migrants were outnumbered by shy migrants with a ratio of 1:3. The difference in ratios remained stable throughout the 3 years of our study (Found and St. Clair, 2016) and similar between parks, despite the occurrence of sub-populations in Jasper.

There is no comparable literature with which to compare ungulate personality to migratory patterns, but personality appears to contribute to the choice of migratory tactics in moose (Alces alces; Rolandsen et al., 2017). Personality is likely involved in the variable expression of migration, also known as partial migration, that occurs in all migratory taxa, but the vast movements involved with migration make personality very difficult to study in this context (Nilsson et al., 2014). Moreover, the few studies that directly link personality and migration are difficult to generalize. For example, in a freshwater fish (roach, Rutilus rutilus), boldness was positively correlated with migratory movement away from predators (Chapman et al., 2011b). But in blue tits (Cyanistes caeruleus), migrants were usually subdominant birds that expressed boldness via neophilia, partly to overcome exclusion from better habitat by dominant birds (Chapman et al., 2011b).

In our study populations of elk, the apparent importance of predation risk to the occurrence of behavioral types made it surprising that we found a similar gradient of shy through bold individuals in a captive elk population, where predators were effectively absent and forage was uniformly available (Found and St. Clair, 2016). Evidence of personality is similarly apparent for several domestic species (Finkemeier et al., 2018), many of which are also protected from natural predators. More generally, behavioral types appear to result from complex geneenvironment interactions that involve multiple, pleiotropic genes (Bouchard and Loehlin, 2001; Krueger et al., 2008) that persist in all populations owing to variable or changing environments (Dingemanse et al., 2007) and both frequency and densitydependent effects (Aplin et al., 2014; Nicolaus et al., 2016). Later, we explore how those factors might contribute to the maintenance of bold and shy elk and their relevance to migration and habituation.

We interpret our results from elk to suggest that the relative advantages of migration for an individual depend on its inherent personality, which interacts with development, learning, and environmental context. The environment includes interactions with people, predators, and conspecifics, as well as forage type and availability. Despite so many sources of variation, average personality-based pay-offs are likely to emerge. In our study system, a bold, neophilic individual, with high tolerance to sound disturbance, a tendency to occur on the periphery of the herd, and low vigilance is presumably more likely to discover novel food sources in a human-dominated area, more likely to be able to exploit them quickly, and more likely to dominate conspecifics. Contrastingly, a shy individual with higher vigilance that seeks natural forage that is abundant, but widely distributed, may have lower likelihood of predation owing to neophobia and greater vigilance. Those shy individuals may also be better able to escape competition via seasonal migration or through reduced population density (Hebblewhite et al., 2002).

Similar interactions between individuals and their environments correspond to predictable advantages for wellknown behavioral types in many other species, such as producers and scroungers (Giraldeau and Beauchamp, 1999), fast vs. slow explorers (Dingemanse et al., 2003), or proactive vs. reactive coping styles (Coppens et al., 2010). Each of these dichotomies may also extend to pace-of-life strategies that correlate with physiological and life history traits (Réale et al., 2010; Careau and Garland, 2012). All of these contrasting types potentially impose evolutionary and strategic constraints on individuals via fitness costs that occur as interacting effects of behavioral types and environmental context (Smith and Blumstein, 2008). Based on our observations of behavioral types in elk, we speculated that several anthropogenic changes to our study landscapes increase selection for bolder personalities, largely through habituation to people. In turn, those tendencies increase the fitness benefits of a resident tactic, but those benefits might be curtailed with management actions, especially aversive conditioning.

# PERSONALITY-DEPENDENT RESPONSES TO AVERSIVE CONDITIONING

A core purpose of our work to identify behavioral syndromes in habituated, town-dwelling elk was to determine whether that information could be used to increase the efficacy of aversive conditioning as a management technique. Aversive conditioning has been used primarily to increase human safety, but previously targeted town-dwelling animals that are disproportionately likely to be involved in human-wildlife conflict (Kloppers et al., 2005). We now know these animals have consistently bolder personality types and are less likely to migrate (Found and St. Clair, 2016). However, aversive condition has also been used in an effort to increase migratory tendency (Spaedtke, 2009). Accordingly, we compared elk of different behavioral types in Jasper before, during and after being exposed to aversive conditioning consisting of active chases by people and benign stimuli consisting of slow, non-targeted walking (Found and St. Clair, 2017). Our aversive conditioning consisted of high-speed foot pursuits of targeted elk with 10-min durations to create an energetic consequence of being pursued (Found, 2015) that might mimic pursuit by coursing predators like wolves (Kloppers et al., 2005). Human hunting can also change prey behavior, but the lacking impact of pursuit may influence space use more than wariness (Bateson and Bradshaw, 1997).

Somewhat counter intuitively, we discovered that bolder elk in Jasper responded more strongly to aversive stimuli with increases in their average flight response distances that were up to five times greater than those expressed by the shyest elk Found and St. Clair, 2018). However, bolder animals also returned to their (originally lower) baseline measures of flight response distances when the aversive stimuli ceased. In combination, bolder elk appeared to be more responsive to approaches by humans whether they were negative or neutral. One year after aversive conditioning treatment, migrants had retained about half of their conditioned increases in wariness, whereas residents had lost all conditioned gains (Found and St. Clair, 2018). The more rapid loss of conditioned responses suggest that aversive conditioning programs need to be targeted and consistent to achieve their desired outcomes, a topic we return to below.

We used the metric of flight response distance to explore individual variation in responsiveness to the frequency of aversive conditioning events for an older dataset collected from elk in Banff (Found et al., 2018). There, elk subjected to more frequent aversive conditioning exhibited greater increases in their flight response distances, but those elk also exhibited more rapid returns to baseline flight response distances when conditioning ceased. As for the Jasper population, the Banff elk with the lowest flight response distances at the beginning of the study (i.e., those exhibiting greater habituation) exhibited the greatest changes in flight response distances during both the conditioning and extinction periods. Additional work with a captive population demonstrated that elk can habituate rapidly with either of foodbased conditioning or benign approaches by people, which was also more rapid for the bolder individuals (Found, 2019).

Other study systems have revealed similar evidence that more habituated individuals exhibit greater responsiveness to human activity (Bejder et al., 2009). For example, house sparrows (Passer domesticus) demonstrated considerable individual variation in neophobia, measured as a latency to explore a novel object, while exhibiting consistent tendencies within individuals to habituate to human disturbance (Ensminger and Westneat, 2012). Similarly, when yellow baboons (Papio cynocephalus) were introduced to an area with accessible human food, only some individuals were bold enough to became crop raiders (Strum, 2010). Juncos (Junco hyemalis) with greater behavioral flexibility appear to be pre-adapted to thrive in urban areas (Atwell et al., 2012). We know of only one other study that attempted to relate existing habituation behavior to responses to aversive conditioning. In black bears (Ursus americanus), more habituated animals were more responsive to aversive conditioning, although it was the less habituated animals that exhibited greater recidivism (Mazur, 2010).

To better understand the sources of behavioral variation among individuals, we studied lateralization in elk by quantifying the proportion of times each marked individual used its left vs. right forelimb to dig in the snow to expose edible vegetation (Found and St. Clair, 2017). In both the Jasper and Banff study populations resident elk were more ambidextrous (less lateralized) than migrants in their use of forelimbs, which we interpreted as equating to increased cognitive flexibility (Found and St. Clair, 2017). Further evidence that laterality reflects cognitive flexibility stems from the congruence in the Jasper population between the conditioning experiments (above) and laterality. It was the bolder elk, comprised mostly by the ambidextrous residents, that expressed the most rapid increases in wariness during aversive conditioning and the most rapid losses in that response when it was removed. Together, these results suggest that behavioral flexibility manifested in weakly lateralized animals contributed to their ability to rapidly identify benign interactions with humans and habituate accordingly.

Taken together, our studies of elk in the mountain parks of Banff and Jasper, plus a nearby captive population, firmly establish the presence of a definable personality gradient in each population that correlates with migratory tendency (Found and St. Clair, 2016), responsiveness to aversive conditioning (Found and St. Clair, 2018; Found et al., 2018), and the process of habituation (Found, 2019). A further source of this variation appears to relate to cerebral lateralization (Found and St. Clair, 2017), which may be especially relevant to predation risk (Found, 2019). Relationships among different components of behavior have undeniable management implications (that we explore below), but they do not reveal the causative agents that maintain this behavioral variation in populations. Many others suggest that personality persists as a frequency-dependent function of changing environments (Dingemanse et al., 2004; Smith and Blumstein, 2008; Réale et al., 2010). A dependency on environmental variation makes personality an especially likely contributor to the dynamics of ungulate migration, which are also known to respond to environmental change and anthropogenic habitat (Bolger et al., 2008; Tucker et al., 2018).

The conceptual model we develop in the next section stems partly from what we know about changes in our study landscapes over the past century related to elk, human use, and predator distribution. In Jasper, elk were absent from the park when it was founded in 1907, introduced from Yellowstone National Park (US) in 1920, had become hyper-abundant by the 1940's, were extensively culled by wardens until 1970, and stabilized at about 1,000 animals in the 1990's (Dekker et al., 1995). Meanwhile, wolves that were abundant in the 1800's were also effectively absent when Jasper was established, but gradually increased until the 1940's when wolf control began, rebounded when it ceased (1966), and exhibited stable populations in the 1990's (Dekker et al., 1995). During their era of high population abundance, elk migrated extensively throughout the park, and the tactic for year-round residency at low elevations first appeared only in about 1980, on the heels of increasing elk mortality from wolves (Dekker et al., 1995; Beschta and Ripple, 2007). Banff's history is similar, but occurred a little later, with rebounding wolf populations in the 1980's and 90's gradually increasing the tendency for elk to congregate near town sites (Hebblewhite et al., 2002), which may have intensified with the cessation of lethal control of grizzly bears by about 2000 (St. Clair et al., 2019).

A history of predator control in our study areas amply demonstrate the capacity elk have to adjust their migratory tactics to changing environmental circumstances. Indeed, migration does not appear to be a genetically-fixed strategy in any ungulate species (reviewed by Berg et al., 2019) although the degree of plasticity may vary even within species (Cagnacci et al., 2011). In our Banff study population, between 2010 and 2011, 16% (8 of 50 marked animals) switched tactics and a similar rate occurred annually in a population adjacent to Banff National Park, most often by migrants switching to residency (Eggeman et al., 2016). Rates of switching migratory tactics appear to be 10–20% in many other ungulate populations (Berg et al., 2019). A capacity to switch tactics begs a question: how do individuals determine which migratory tactic optimizes fitness for their own personality and environment? We develop a model below to show how (a) how environments undergoing anthropogenic change might increase selection for bolder behavior via (b) four interacting ecological drivers, to (c) increase the benefits and, consequently prevalence, of a resident tactic. Later we describe how aversive conditioning might be used to increase selection for migration by increasing the costs of residency. We propose that selection for boldness results secondarily in the resident migratory tactic, but with frequency-dependent limits.

### A CONCEPTUAL MODEL RELATING BEHAVIORAL TYPES, ECOLOGICAL DRIVERS, AND MIGRATORY TACTICS IN ELK

Most biologists are familiar with the competitive and frequencydependent dynamic between hypothetical hawks and doves in the classic game theory (Maynard Smith and Price, 1973). As behavioral types, aggressive hawks and docile doves persist in a population as a mixed evolutionarily stable strategy via specific proportions that equalize the costs and benefits of the two strategies. Stable equilibria in mixed strategies of this sort require that individuals play the tactic suited to their morphological type (Gross, 1984), but strategies can be flexible within individuals according to ecological context (Maynard Smith and Price, 1973). Our system lacks information on the fitness pay-offs that would support development of a formal ESS, but similar concepts apply to the conceptual model we develop below and present schematically (**Figure 1**). In it, we show how behavioral types on a bold-shy continuum respond differently, on average, to several ecological drivers to change the relative fitness of migrant vs. resident strategies. We follow Chapman et al. (2011a) by

identifying ecological drivers in the environment that promote different migratory tactics, but differ in the specific drivers we name.

The first of our ecological drivers is human activity, which begins the cascade of landscape changes that favor bolder individuals. In protected areas, including town sites, where hunting is not permitted, ungulates have little need to fear or avoid people, as they do outside of protected areas where humans sometimes act as predators. As human activity and density increased in our study areas, the frequency of close, but benign encounters with humans necessarily also increased. In this and similar urban contexts, habituation by ungulates to people occurs rapidly (Honda et al., 2018). Captive elk with bolder personalities showed a greater tendency to habituate that began when they were still calves (Found, 2019).

Habituation to people increased opportunity for selection on personality by the second of our ecological drivers; forage availability. In the mountain parks of North America, elk historically foraged on natural vegetation that was widely distributed and migrated to higher elevations each spring to follow greening vegetation (Boyce, 1991). Others have suggested that finding such forage and avoiding competition for it have been core contributors to migration in ungulates (Fryxell and Sinclair, 1988; Chapman et al., 2011a). As anthropogenic development proceeded, novel forage sources in our study area included mowed parks, school yards, playing fields, golf courses, palatable ornamental vegetation, and spilled grain from train cars. Bolder elk are more likely to explore these novel food sources and more likely to tolerate people or infrastructure near them.

The third driver is predator distribution and activity. Historically, predators were well-dispersed in the landscape and difficult to avoid completely. Migration by small groups of ungulates lessens aggregation to reduce both detectability and attractiveness for predators (Hebblewhite et al., 2002). Predator refugia occur in areas where wildlife have high rates of benign encounters with people because carnivores are typically more wary around people than their ungulate prey (Muhly et al., 2011), and because they are actively excluded by managers to support human safety (Lennox et al., 2018). Such refugia would be expected to favor bolder individuals that can tolerate closer proximity to people and their infrastructure, such as townsites both inside and outside the boundaries of protected areas and, more generally, in landscapes where predators are consistently persecuted. Outside of protected areas, human hunters exhibit similar selection on ungulates (Bateson and Bradshaw, 1997), but the effect of personality may differ. There, bolder animals may be more likely to be shot by human hunters (Ciuti et al., 2012), even if they are also more likely to learn how to avoid hunters over time (Thurfjell et al., 2017).

The final driver is interactions with conspecifics, which would favor bolder animals because they can outcompete shyer animals that are more submissive. Intraspecific competition necessarily also interacts with each of the other drivers to provide a competitive advantage to bolder elk that are also able to share close proximity with people, physically dominate shyer elk at concentrated novel food sources, and better compete to exploit the best opportunities to avoid predators while accessing food.

Many other authors have shown how these four ecological drivers are among the factors associated with the loss of migration in ungulates (Berger, 2004, 2007; Bolger et al., 2008; Tucker et al., 2018), but our model differs by positing that these changes may result secondarily from selection on personality to favor bolder individuals. Two important parts of our explanation are that (a) directional selection for boldness is limited by frequency-dependent pay-offs and (b) bolder individuals are described by phenotypes, with little knowledge of the genetic basis of this behavioral type. The limitations imposed by frequency dependence are familiar enough from hypothetical example of hawks and doves, wherein the advantages of bold behavior are greater amid shy individuals. But another purpose of the hawk-dove game was to demonstrate the multiple mechanisms that could produce alternative behavioral phenotypes. Individuals might exhibit bolder behavior (hawks) as a fixed genetic strategy, as a context-dependent tactic with ongoing flexibility, or as a mix of the two via gene-environment interactions during development. By any of these mechanisms, higher proportions of bold-acting individuals within a group necessarily increase per capita competition for food, costs of agonistic behavior, poorer detection (via vigilance) but higher attraction (via aggregation) of predators, and higher susceptibility to parasites and disease. Despite these frequencydependent limits, it is easy to imagine that anthropogenic changes to environments increase the stable proportion of bold individuals in a population.

The consequences of such directional selection on boldness could explain why a majority of bold individuals are residents and a majority of shy ones are migrants, while frequencydependent limits to that selection could explain why some animals exhibit the opposing relationship as bold migrants or shy residents. As a minority, shy residents might exploit the foodfinding ability of the bolder animals, while increasing herd-level security near humans. Shy elk might also benefit from associating with large groups that attract cleaner birds into mutualistic interactions to remove ectoparasites that favor shyer elk almost exclusively (Found, 2017a). Conversely, bold migrants would be at a competitive advantage in groups of shyer individuals when they encounter concentrated, but limited, sources of natural foods. For example, in winter it is quite common for shy and submissive animals to dig through the snow to access underlying forage, only to then have bolder animals physically displace them and eat the forage without the effort of finding or exposing it (R. Found, personal observation). The mapping of personality to migratory tactic might be further blurred by the ongoing interactions between types. For example, habituation by shyer elk might be accelerated via imitation in the presence of bolder elk (Found, 2019). Despite the frequency-dependent limits to directional selection and diversity of interactions among conspecifics, environmental change has often favored bolder animals to increase the proportion of residents and create several challenges for wildlife managers.

# MANAGEMENT IMPLICATIONS OF PERSONALITY IN UNGULATES

Our conceptual model (**Figure 1**) reveals how the best way to limit the positive feedback between resident migratory strategies and boldness behavior that is otherwise favored by anthropogenic landscape change is to exert contrary effects on the drivers themselves. Of the four drivers we present, eliminating predation refugia through predator redistribution is presumably off the table; managers can educate the public to avoid worsening human-carnivore conflict in a variety of ways (Baruch-Mordo et al., 2011), but they cannot invite predators into town sites and other areas with concentrated use by people. Similarly, anthropogenic forage is already recognized as a contributor to human-wildlife conflict (Newsome et al., 2015) and it is increasingly managed to limit associated ecological problems (Nyhus, 2016). Its availability to elk is limited in Banff and Jasper via ungulate fencing (Shepherd and Whittington, 2006), exclosures around palatable vegetation, and efforts to reduce grain spills on railways near townsites (St. Clair et al., 2019), although more could be done. Predator-resembling aversive conditioning is a tool that can manipulate the first driver by making encounters with people less benign, but also influences the third driver by decreasing the benefits afforded to animals using refugia. However, refinement is needed in the way it is practiced and its goals need to be articulated clearly.

To be effective at restoring wariness to increase human safety, aversive conditioning would need to increase the costs of proximity to people via association with aversive stimuli that are immediate, initially intense, consistently applied, evolutionarily relevant, and unpredictable in space or time (after Conover, 2001; Domjan, 2014). To limit adoption of the resident tactic that results from boldness, aversive conditioning needs to be even more specific. It was possible to shift the short-term distribution of elk by targeting individual animals repeatedly with evolutionarily-relevant chases (Kloppers et al., 2005) and we expanded that approach by identifying bold animals (Found and St. Clair, 2016), using isolation to increase the costs of being targeted (Found and St. Clair, 2018), and determining the frequency of conditioning that minimizes extinction of learned wariness (Found et al., 2018). Others have shown that learned wariness by habituated animals will gradually disappear if the aversive stimulus is removed (Lattal and Lattal, 2012) and that frequent, low-intensity conditioning can generate habituation (Powell et al., 2016). Synthesis of laboratory studies predict that the products of aversive conditioning can be maintained with lesser effort than is required to initiate a conditioned response (Domjan, 2014), but more work will be needed to know what types and frequency of aversive conditioning are needed to achieve long term management objectives for public safety, migratory behavior, and ecological goals.

Ideally and more generally, aversive conditioning opposes directional selection on bold personality types, reducing the benefits of a resident tactic in ungulates by restoring a "landscape of fear," which is especially likely if perceived risks apply to calves (Laundre et al., 2001). By increasing the need for vigilance and other energetically costly anti-predator behaviors, aversive conditioning could reduce the benefits of a predator refugia provided by people in urban areas. Effective aversive conditioning could have a similar effect to the presence of wolves which, even days previously, can cause profound increases in the wariness of elk (Creel et al., 2005; Found and St. Clair, 2016). By contrast, culling animals via shooting removes the target animal, but does not seem to alter the distribution or behavior of any of the surviving animals, even if they witnessed the death of a conspecific (R. Found, personal observation). Aversive conditioning is also widely viewed as more ethical than either of lethal management (Koval and Mertig, 2004) or translocation (Whitwell et al., 2012).

Despite the uncertain effect and frequent public opposition to lethal management of urban ungulates (Dandy et al., 2012), it has been promoted as a necessary consequence of selection for bold, habituated deer that inhabit urban areas around the world (Honda et al., 2018). These authors acknowledged that inadvertent selection by humans for bolder behavior by ungulates can occur rapidly via behavioral flexibility to cause human-wildlife conflict and other management problems. Many managers see the situation similarly, partly because human injury by wildlife is both grave and potentially litigious. In Banff, managers culled 10–20 resident individuals annually for most of the last two decades and a similar practice was common in Jasper in the 1960's and 70's. In effect, they were attempting to oppose the directional selection on bold behavior that supports the switch by migrants, typically shyer, to a resident tactic. Beyond the problem of public opposition, our conceptual model suggests that culling cannot solve the problem of selection for bolder personalities because boldness itself is relative phenotype in this (and perhaps any) population that results from a combination of behavioral flexibility, genetic differences in temperament, and gene-environment interactions, such as developmental plasticity. Culled individuals are easily and rapidly replaced by the next boldest individuals in the population. These predictions are supported by the fact that the ratio of residents:migrants in Banff, along with mean metrics of boldness, did not change during our study despite ongoing culling of resident elk, which were almost exclusively the bolder individuals.

#### CONCLUSIONS

In our paper, we reviewed our past work addressing the contribution of animal personality to the problem of non-migratory, habituated elk in mountain town sites. We showed that personality metrics can be developed for wild ungulates (Found and St. Clair, 2016) and used to interpret responses to management actions that include aversive conditioning (Found and St. Clair, 2018; Found et al., 2018). In elk, personality appears to relate to expressions of behavioral laterality, which may signal cognitive flexibility (Found and St. Clair, 2017). A suite of personality metrics influence social dynamics in elk (Found and St. Clair, 2016), herd-level behaviors (Found, 2017b), and may also extend to inter-specific interactions (Found, 2017a). Similar personality gradients have been found in diverse species and ecological contexts (Sih et al., 2012) and certainly occur in other ungulates that could be similarly studied in the wild.

We synthesized our past work on habituated elk to develop a conceptual model to show how changes in human behavior, forage availability, predator distribution, and conspecific interactions have favored bolder individuals to contribute to the loss of migration. Specifically, individuals with bolder personalities are more likely to exploit human-dominated areas because they are quicker to discover novel food resources and learn that predators avoid these areas. The same personality and laterality characteristics make these animals more likely to habituate to people, further amplifying their benefits and reducing their costs of co-occurrence. We described how managers might limit this selection on boldness with aversive conditioning that consistently imposes costs on targeted, bold individuals. We also suggested that culling the bolder individuals is unlikely to solve management challenges stemming from boldness because it is partially caused by behavioral flexibility that responds rapidly to changing circumstances that include the distribution of personalities in a population.

Similar implications of inadvertent selection on ungulate personality may apply to other populations of migratory ungulates and extend well-beyond the problems caused by habituation and residency. For example, selection of animals to support captive breeding or enhance the genetic diversity of declining populations could also inadvertently target the bolder, neophilic individuals that are easier to catch or maintain in captivity (McDougall et al., 2006). The subset of bolder individuals and their descendants might be less likely to survive if they are translocated or reintroduced to landscapes containing predators or hunters (Smith and Blumstein, 2008; Ciuti et al., 2012). Similarly, the prevalence of conservation-relevant research based on GPS-collared animals may impose a systematic bias toward individuals with personalities, not just age and sex distributions, that are more likely to be captured in the first place (Merrick and Koprowski, 2017).

A major limitation of using personality in wildlife management is the paucity of studies that combine those contexts, despite rapid increases in personality research on many species (Dingemanse et al., 2012). Behavioral studies of animal personality are urgently needed for ungulates, whose large size and gregarious tendency can cause rapid changes to habitat (Polfus and Krausman, 2012). Personality-mediated choices of migratory tactics in our study populations may offer some general insights for the loss of migratory behavior in ungulates around the world (Berger, 2004; Bolger et al., 2008; Tucker et al., 2018). Similar selection for bold and behaviourally flexible individuals is likely occurring for hundreds of other synanthropic species owing to the rapid rate of human population growth and urbanization (Walter et al., 2010). For ungulates, additional interacting effects include climate change (Tucker et al., 2018), urbanization and habituation of predators (Bateman and Fleming, 2012), and declines in predators overall (Ripple et al., 2014). Wildlife managers of the Anthropocene urgently need more tools, which should include better understanding and use of animal personality (Sih et al., 2012; Merrick and Koprowski, 2017).

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

All authors are equal lead authors, reflecting equal contributions to this manuscript.

#### FUNDING

This work was supported by Natural Sciences and Engineering Research Council, via a Discovery Grant (RGPIN-2017-05915) to CC.


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Found and St. Clair. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Individual Repeatability in Marine Migratory Behavior: A Multi-Population Assessment of Anadromous Brown Trout Tracked Through Consecutive Feeding Migrations

#### Sindre H. Eldøy <sup>1</sup> \*, Xavier Bordeleau<sup>2</sup> , Glenn T. Crossin<sup>2</sup> and Jan G. Davidsen<sup>1</sup>

*<sup>1</sup> NTNU University Museum, Norwegian University of Science and Technology, Trondheim, Norway, <sup>2</sup> Department of Biology, Dalhousie University, Halifax, NS, Canada*

#### Edited by:

*Yolanda E. Morbey, University of Western Ontario, Canada*

#### Reviewed by:

*Thomas Reed, University College Cork, Ireland Nathan R. Senner, University of South Carolina, United States*

> \*Correspondence: *Sindre H. Eldøy sindre.eldoy@ntnu.no*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *01 May 2019* Accepted: *18 October 2019* Published: *01 November 2019*

#### Citation:

*Eldøy SH, Bordeleau X, Crossin GT and Davidsen JG (2019) Individual Repeatability in Marine Migratory Behavior: A Multi-Population Assessment of Anadromous Brown Trout Tracked Through Consecutive Feeding Migrations. Front. Ecol. Evol. 7:420. doi: 10.3389/fevo.2019.00420* Despite that the study of individual repeatability is a common topic in behavioral ecology, virtually nothing is known about inter-annual variability in the marine migratory behavior of iteroparous salmonids that can complete multiple feeding migrations in their lifespan. Behavioral data from 38 anadromous brown trout *(Salmo trutta)*, tracked by acoustic telemetry in 2–3 consecutive marine feeding migrations in two Norwegian fjord systems, were analyzed for intra-individual repeatability in key aspects of their marine migration. Individual brown trout displayed significant inter-annual consistency in marine area use and in the timing of marine exit (i.e. when they returned to spawning rivers), but not in the timing of marine entry or the time spent in the marine environment each year. Our study raises new questions about how anadromous brown trout respond to changing conditions and anthropogenic factors in the marine environment. Intra-individual repeatability of brown trout linked to changing environmental conditions should therefore be a focus for future studies.

Keywords: behavioral repeatability, habitat use, marine migration, migratory timing, Salmo trutta, sea trout, migratory continuum

#### INTRODUCTION

The post-spawning feeding migrations of iteroparous fish species have evolved to allow nutritionally depleted individuals the opportunity to exploit richer feeding habitats in an effort to recondition for future reproductive events. Needless to say, the mechanisms and patterns of migration can vary widely both within and among populations, as may the degree of individual flexibility and/or repeatability of migratory behavior in response to environmental fluctuations. Behavioral repeatability has been documented in various taxa (Bell et al., 2009), including species and populations of birds, mammals, and fish (e.g., Dias et al., 2010; Lea et al., 2015; Müller et al., 2015; Leclerc et al., 2016). For predatory fish feeding in the marine habitat, the availability, and distribution of resources in the marine environment can vary between years (Dragesund et al., 1997; Rikardsen and Amundsen, 2005), which should favor flexibility in traits like migration timing, distance, and duration of residency in various habitats if the organisms have reliable cues from the environment to adjust their behavior in response to the environmental changes (Reed et al., 2010).

The Salmonidae is a family of freshwater spawning fishes, where several of its species initiate feeding migrations to the marine environment (Pavlov and Savvaitova, 2008). Among these, brown trout Salmo trutta is a widely distributed, facultatively anadromous species known to display a continuum of migratory strategies ranging from freshwater residency and potamodromy to estuarine, short and long-distance marine migrations, both among and within populations (Cucherousset et al., 2005; Boel et al., 2014; del Villar-Guerra et al., 2014; Eldøy et al., 2015; Flaten et al., 2016; Bordeleau et al., 2018). As an iteroparous species, anadromous brown trout can undertake multiple annual marine feeding migrations during its lifetime (L'Abee-Lund et al., 1989; Thorstad et al., 2016), where the freshwater residency between marine migratory seasons is usually characterized by spawning and overwintering with opportunistic feeding (Davidsen et al., 2017) that have limited importance for somatic growth (Knutsen et al., 2001). While the drivers of the brown trout migratory continuum have remained somewhat mysterious, growing scientific evidence indicates a role of individual physiological and nutritional state, metabolic rate, and food availability (Olsson et al., 2006; Wysujack et al., 2009; Davidsen et al., 2014; Eldøy et al., 2015; Bordeleau et al., 2018). Despite high inter-individual variability in migratory behavior, the degree of intra-individual behavioral flexibility to changing environments and its consequences in terms of growth, survival, and ultimate fitness remain obscure. Beyond the role of environmental variability, the migratory behavior of anadromous brown trout can be influenced by anthropogenic impacts on coastal waterways, such as marine traffic, harbors and other nearshore infrastructure, renewable energy production, fisheries, and aquaculture (Thorstad et al., 2016; Aldvén and Davidsen, 2017). Importantly, recent work using acoustic telemetry has documented inter-annual shifts in the marine habitat use of different groups of anadromous brown trout in response to aquaculture-associated salmon lice abundances (Halttunen et al., 2018). However, due to logistical constraints imposed by battery life of acoustic transmitters, and relatively high mortality between spawning events (Fleming and Reynolds, 2004), no previous studies have yet assessed the inter-annual flexibility in the marine migrations of brown trout tracked through multiple years.

In order to investigate the degree of variation in behavior of brown trout individuals between consecutive marine feeding seasons, we extracted behavioral (movement) data from trout tagged in acoustic telemetry studies in two Norwegian fjord systems between 2012 and 2017 (e.g., Eldøy et al., 2015; Bordeleau et al., 2018). Studies of migratory species in various taxa have shown that individuals can exhibit both consistency and repeatability in behavior (Bell et al., 2009). Given the lack of previous studies on intra-individual repeatability in annual marine migratory behavior for salmonids, we chose not to make a priori predictions from specific hypotheses. Instead, we explored this unique dataset to investigate whether key behavioral aspects of the intra-individual marine behavior of anadromous brown trout was repeated among years. Specifically, we analyzed the degree of annual intra-individual behavioral repeatability in terms of (i) spatial dispersal, (ii) migratory timing, and (iii) duration of marine residency.

# MATERIALS AND METHODS

#### Study Area

The study was conducted in two fjord systems in central and northern parts of Norway (**Figure 1**). The Hemnfjord system consists of two interconnected fjords with more than 60 km<sup>2</sup> surface area and about 65 km of shoreline and is connected to the open sea by a 36 km long strait (**Figure 1**, Eldøy et al., 2015). The Tosenfjord system consist of two interconnected fjords with about 150 km<sup>2</sup> surface area and more than 270 km of shore line, connected to the open sea by a 15 km long strait (**Figure 1**). Several watercourses with partially anadromous populations of brown trout drain into both fjord systems. The Hemnfjord study area is described in detail by Eldøy et al. (2015, 2017) and Flaten et al. (2016), while the Tosenfjord study area is described by Bordeleau et al. (2018).

### Environmental Variables

Both fjord systems had aquaculture facilities with farmed salmon in open net pens during the study periods. Sea temperature and salmon lice count data from the salmon farms was downloaded from the Norwegian Fish Health Database (www. barentswatch.no), and all available recordings from marine aquaculture locations in the two fjord systems were combined. Data on sea temperatures and salmon lice counts (here shown as counts of all life stages combined) in the farms located within each fjord system revealed seasonal and annual variations in both temperature (**Figure 2**) and salmon lice infestation levels (**Figure 3**).

#### Acoustic Tracking

In the Søa watercourse in Hemnfjord, a total of 100 brown trout were tagged in freshwater or in the estuary and tracked with acoustic receiver arrays in the fjord system in 2012–2014 (**Figure 1**). In Tosenfjord, a total of 274 brown trout were tagged in freshwater and estuaries of River Åbjøra and Urvold watercourse and tracked with acoustic receiver arrays in the fjord during 2015–2017 (**Figure 1**). In general, anadromous brown trout in the two fjord systems migrate to sea each summer for feeding and return again to freshwater for spawning and/or overwintering during late summer (Eldøy et al., 2015; Bordeleau et al., 2018). The fish were either tagged during spring after spawning and prior to their marine migration, or in the autumn prior to potential spawning. For fish tagged during autumn, the tracking started at their outwards migration during the following spring. All fish included in this study were tagged following the same protocol. The fish were captured by rod fishing or gillnets that were continuously monitored and kept in holding nets for up to 4 h prior to tagging. The fish were sedated using 2-phenoxy ethanol for 4 min prior to making a 2 cm incision in the body cavity and inserting the sterilized acoustic tag. The incision was closed by 2–3 sutures, before the fish were placed in a recovery tank for up to 15 min and subsequently released at the site of tagging. The expected battery lifetime of the acoustic tags ranged from 15 to 24 months (**Table 1**). See Eldøy et al. (2015) and Bordeleau et al. (2018) for further details. Arrays of acoustic receivers (Vemco Inc., Canada models VR2, VR2W, and

FIGURE 2 | Sea temperatures reported by the salmon farms located within the study area in Hemnfjord (A) and Tosenfjord (B). Colored lines indicate the smoothed conditional mean and its pointwise 95% confidence interval (shaded bands).

lines indicate the smoothed conditional mean and its pointwise 95% confidence interval (shaded bands).

VR2-AR) were deployed at various locations in both freshwater, estuaries and saltwater in the two fjord systems to map the movements of the tagged brown trout (**Figure 1**). In Hemnfjord, receiver ID 34 and 35 were deployed in the estuary of River Søa and represented the transition zone between freshwater and saltwater. In Tosenfjord, receiver ID 44 was deployed in the estuary of River Urvold and receiver ID 63–68 were deployed in the estuarine parts of River Åbjøra. See Eldøy et al. (2015) and Bordeleau et al. (2018) for further details.

Prior to statistical analyses, the tracking data was filtered for false registrations resulting from noise in the surroundings of the receivers and/or code collision of simultaneously transmission from multiple transmitters (Pincock, 2012). Receivers found to contain frequently false detections were filtered by adding a filter that required at least two registrations within a 10-min time span to accept the registrations. The data was further visually inspected for false detections, and obviously false detections were removed prior to the statistical analyses. See Eldøy et al. (2015) and Bordeleau et al. (2018) for further details.

#### Data Analyses

All statistical analyses were conducted using R version 3.5.3 (R Core Team, 2019) and RStudio version 1.2.1335 (RStudio Team, 2019). Spearman's rank correlation was used to test the intra-individual correlation in migratory behavior between first and second year of tracking (Hanson et al., 2010; Taylor and TABLE 1 | Summary of capture timing and location, body size, and tag information of tracked fish.


Cooke, 2014; Nelson et al., 2015). Data on sea temperature and salmon lice prevalence were plotted using r-package ggplot2 (Wickham, 2016), using the "geom\_smooth" function to produce smoothed trend lines. Marine migratory tactics were classified as short- medium- and long distance migration in Hemnfjord for each tracking season, based on how far out in the fjord system the fish was detected (**Figure 1**, Eldøy et al., 2015). Similarly, marine migratory tactics were defined as short, longinner, and long-outer distance migrants based on migratory distance in Tosenfjord (**Figure 1**, Bordeleau et al., 2018), where fish remaining resident in the estuary of River Åbjøra were considered as short distance migrants. Only fish observed returning to freshwater the second season, being detected after 1 July in the second tracking, season or qualifying for the longest distance migratory tactic in the second tracking season were included in this analysis. The categorical variables of migration distance were transformed to ordinal structure according to relative migration distance (short = 1, medium/long-inner = 2, and long/long-outer = 3) prior to testing the intra-individual consistency using Spearman's correlation test.

Network analysis and bipartite graphs were made using the r-package igraph (Csardi and Nepusz, 2006), as previously demonstrated on acoustic telemetry data by Finn et al. (2014). Here, the individual's total yearly count of detections for each marine receiver that was operative through all the study years (numbered receivers in **Figure 1**) was used to compare the individual's marine area use among years. Only fish observed returning to freshwater the second season, being detected after 1 July in the second tracking season, or qualifying for the longest distance migratory tactic in the second tracking season were included in the network analysis. The number of detections was used as weights for the link (edges) between tracked fish and associated receivers. Receivers and tracked fish were grouped using the igraph "cluster\_walktrap" function (using 6 steps for Hemnfjord and 11 steps for Tosenfjord), which uses a randomwalk algorithm to try to find densely connected subgraphs (communities) within the network (Csardi and Nepusz, 2006; Finn et al., 2014). The results of the grouping within the network of each fjord system were organized as an ordinal variable according to distance in the network plot and geographic location of the receivers (**Figure 1**) prior to evaluating inter-individual consistency among years using the Spearman's correlation test. Timing of marine entry was defined as the time of the first detection on a receiver deployed in estuarine or marine waters preceding detection on a receiver deployed in freshwater. Timing of marine exit was defined as the time of the last detection on a receiver deployed in estuarine or marine waters prior to detection on a receiver deployed in freshwater. An exception was made for river Åbjøra, where the two outermost receivers in the river mouth (station 67 and 68, **Figure 1**) were defined as estuarine and receivers deployed further upstream in the large parts of the watercourse influenced by tidal water (station 63– 66, **Figure 1**) were defined as "freshwater" in the marine timing and duration analyses. Marine duration was calculated as the yearly accumulated time spent in the marine environment, where periods of freshwater residency between the first marine entry and the last marine exit were excluded. Individuals that were only residing in tidally influenced parts of river Åbjøra were excluded from analyses of timing and duration of marine migration.

# RESULTS

Of the 374 tagged individuals from the two fjord systems, we could extract data from 36 individuals (**Table 1**) to explore the intra-individual repeatability in area use, timing of marine entry, timing of marine exit, and/or marine residence time among consecutive marine feeding seasons. Of these, 10 individuals from Hemnfjord and eight individuals from Tosenfjord were tracked throughout two full marine seasons (i.e., they migrated back to freshwater after the second feeding migration). For two individuals from Tosenfjorden, tracking data could be extracted from three consecutive marine feeding migrations. Generally, the tagged individuals displayed a relative consistent marine behavior on the evaluated aspects of their marine behavior between the two or three consecutive marine feeding seasons of tracking (**Table 2**). However, the degree of consistency varied among individuals, with some individuals displaying large variations in some of the measured behavioral aspects among years (**Table 2**).

# Marine Area Use of Tagged Anadromous Brown Trout

A wide range maximum migratory distances were observed, ranging from remaining resident in the estuarine areas of the watercourse where the fish were tagged—to utilizing large parts of the fjord system and spending a significant amount of their total marine residence time in areas outside the outermost receiver arrays. There was a strong and significant intra-individual correlation between the observed migratory tactic during the first and second year of tracking in both Hemnfjord (Spearman's rank-correlation; r<sup>s</sup> = 0.84, n = 12, P < 0.001) and Tosenfjord (r<sup>s</sup> = 0.96, n = 18, P < 0.001) when marine migratory behavior was classified as defined by Eldøy et al. (2015) and Bordeleau et al. (2018) for Tosenfjord (**Figure 1**). For Hemnfjord, 9 of 12 tracked fish were assigned to the same migratory tactic in both marine seasons. For Tosenfjord, 16 of 18 individuals were assigned to the same migratory tactic in both years. The annual consistency in marine area use of tagged individuals during the 2 consecutive years of tracking was further investigated by network analyses (**Figures 4**, **5**; **Table 2**). There was a strong and significant intra-individual correlation between the assigned network group during the first and second year of tracking in both Hemnfjord (Spearman's rank-correlation; r<sup>s</sup> = 0.77, n = 12, P = 0.003) and Tosenfjord (r<sup>s</sup> = 0.89, n = 18, P < 0.001). The grouping analysis of network structure using a cluster walk trap algorithm on the network in Hemnfjorden resulted in two different groups; one containing receiver locations in the inner and central part of Hemnfjorden and one containing receiver locations in outer and eastern parts of the fjord system (**Figure 4**). In this fjord system, 11 of 12 of the tagged fish were assigned to the same community unit both years (**Table 2**). For Tosenfjord, the grouping analysis of network structure using a cluster walk trap algorithm resulted in seven different community with associated tagged fish and receiver locations (**Figure 5**). Here, 12 of 18 tagged fish that were followed for 2 years were assigned to the same community both years (**Table 2**). One of the two fish that was tracked and analyzed for three consecutive seasons was assigned to the same community all years. However, two of the fish tracked during two seasons, and the one fish tracked for three seasons that changed community, transitioned between communities that all had associated receivers in the estuarine areas of the Åbjøra watercourse.

#### Timing of Start and End of Marine Feeding Migration

Average day of year for marine entry was 123.8 (n = 27, SD = 15.8, range 103.0–151.0) for the first year of tracking and 120.2 (n = 27, SD = 24.0, range 64.4–155.7) for the second year of tracking. There was a weak and non-significant intra-individual correlation of the timing of marine entry between the first and second year of tracking (**Figure 6**, Spearman rank-correlation; r<sup>s</sup> = 0.32, n = 27, P = 0.10). Average individual difference in the timing of sea entry between the consecutive years was 18.0 days (n = 27, SD = 15.0 days, range 0.2–50.6 days).

Average day of year for marine exit was 207.1 (n = 17, SD = 57.5, range 151.8–332.8) for the first year of tracking and 190.5 (n = 17, SD = 43.9, range 140.0–296.4) for the second year of tracking. There was a strong and significant intra-individual correlation of marine exit timing between the first and second year of tracking (**Figure 6**, Spearman rank-correlation; r<sup>s</sup> = 0.81, n = 17, P < 0.001). Average difference in the timing of exit 


between the consecutive years was 21.1 days (n = 17, SD = 30.8 days, range 2.1–135.8 days).

#### Migratory Duration

Average marine residency was 91.6 days (n = 18, SD = 66.4, range 30.3–262.8) for the first year of tracking and 74.2 (n = 18, SD = 56.7, range 13.1–263.4) for the second year of tracking. There was no significant correlation of marine residence time between the first and second year of tracking (**Figure 7**, Spearman's rankcorrelation; r<sup>s</sup> = 0.31, n = 18, P = 0.20). The difference in the duration of the marine migration between the two seasons varied greatly among the individuals, ranging from 0.7 days to 152.6 days (n = 18, mean = 30.3 days, SD = 44.4 days).

#### DISCUSSION

This study revealed that some key aspects of the annual marine feeding migration of anadromous brown trout tend to be repeatable between years. Repeatable behavior is a common phenomenon in nature, but this is to our knowledge the first study to illustrate repeatable behavior by anadromous brown trout, and is among the few to evaluate behavioral repeatability in salmonid fishes more generally (Taylor and Cooke, 2014). Although large phenotypic and behavioral variability has been observed among brown trout in previous studies (Klemetsen

et al., 2003; Thorstad et al., 2016; Halttunen et al., 2018), and previous studies suggest pre-migratory nutritional state as a driver for the migratory continuum of brown trout (Bordeleau et al., 2018), this study suggests that the intraindividual behavioral flexibility during the marine migration is low. However, despite the general repeatability of marine migration behavior, the degree of repeatability varied greatly among individuals, with some individuals displaying large intraindividual variance among the 2 years of tracking.

Variation in behavioral traits among years can be divided into an individual effect and a residual effect, where the individual effect is thought to be determined by genetics and previous experiences (Bell et al., 2009; Biro and Stamps, 2010; Conrad et al., 2011). This study was not designed in a way that allowed us to evaluate the importance of environmental experience to the observed trends of behavioral repeatability. Due to low number of tracking years (only 2 years for most individuals, **Table 2**), low number of different environments, and the relatively low sample size, it was not possible to evaluate how yearly variation in environmental factors affected the behavior of anadromous brown trout. Sea temperature and salmon lice abundances reported by salmon farms in the two fjord systems suggest some degree of environmental fluctuation throughout the season and among years, and previous studies have in fact reported that the abundance of certain prey in Norwegian fjords can vary greatly among years (Dragesund

FIGURE 6 | Timing (day of year) of tagged individual's marine entry (A) and marine exit (B) in first year of tracking (x-axis) and second year of tracking (y-axis). Stippled line indicates the 1:1 line. Summary statistics (r<sup>s</sup> and *P*-values) are calculated by Spearman's correlation test.

et al., 1997; Rikardsen and Amundsen, 2005). It is therefore likely that the anadromous brown trout in our study experienced somewhat varying conditions interannually during their marine migrations, despite the observed individual consistency in their marine behavior. Alternatively, the observed differences in environmental conditions experienced by the tagged fish might not have been drastic enough to trigger intra-individual changes in marine habitat use patterns.

Inter-annual variation in individual marine area use was evaluated with two different approaches; subjectively defined lines by the distance from watercourse of tagging, and by performing network analyses to investigate the relationship between every tagged fish and receivers deployed in the fjord and grouping them by using a random-walk-algorithm. Using both methods, the fish from both fjord systems showed a strong and significant individual consistency in marine area use between the two tracking seasons. Previous studies have revealed that the area use of individuals in anadromous brown trout populations can vary greatly (Thorstad et al., 2016). Migratory tactics within brown trout populations have previously been linked with nutritional status in spring prior to seasonal marine feeding migrations, where fish in low body condition Eldøy et al. (2015) and low nutritional physiological state Bordeleau et al. (2018) were more likely to migrate further out in the fjord system. Bordeleau et al. (2018) suggested that individuals with poor body condition in spring may be more prone toward feeding in the distant, outer areas, where potentially better feeding conditions occur, in order to regain their energy reserves. The observed consistency in marine area use raises questions about whether there could be causal factors that act over longer time-frames, which might cause individual pre-migratory nutritional status during spring to be maintained across years. For example, it could be speculated that marine habitat or prey preferences have the potential to affect energy storage, energy investment into reproduction, and post-spawning nutritional state prior to the next feeding season (Bordeleau, 2019). However, the links between marine migratory behavior and prey choice, growth, reconditioning of body condition and subsequent spawning investment is poorly understood.

The individual migratory behavior of the tagged individuals was relative consistent among years despite some observed yearly variation in sea water temperature and salmon lice prevalence in salmon farms. Previous studies have linked both horizontal and vertical marine responses of anadromous brown trout to variation in seawater temperature (Rikardsen et al., 2007; Jensen et al., 2014; Eldøy et al., 2017; Kristensen et al., 2018). It has been thoroughly documented that open cage salmon farming can lead to the unnaturally high infestation of wild salmonids and alter their marine behavior (Thorstad et al., 2015; Finstad et al., 2017). Halttunen et al. (2018) documented shifts in the marine area use of different groups of anadromous brown trout in response to salmon lice abundance in Hardangerfjord, Southern Norway, and observed that brown trout utilized outer areas less in years when the risk of salmon lice infestation was high, compared with years with lower infestation risk. However, the variation in salmon lice infestation levels in the study by Halttunen et al. (2018) was probably greater than in our study, as they investigated the behavior of brown trout in years when salmon production cycles were active vs. the behavior of brown trout in years when all salmon farms in the inner fjord were fallow. There was a marginally non-significant, intra-individual correlation in timing of marine entry between the 2 years of tracking. Previous studies have documented that previous life history, morphology and physiology affect the timing of seaward migrations in salmonid populations (Halttunen et al., 2013; Thorstad et al., 2016), and so intra-individual consistency in timing of marine entry is thus expected. However, environmental conditions such as the timing of ice melting and increased water temperature and discharge has also been found to influence the timing of migration (Thorstad et al., 2016). Inter-annual variation in environmental conditions in the freshwater habitat is therefore likely to have influenced the annual timing of marine entry in the present study.

In contrast to timing of marine entry, a strong and significant intra-individual correlation was found for the timing of marine exit between the 2 years of tracking. This suggests that life history, physiological state and/or individual effects have some influence on the timing of when individuals end their marine feeding season. Marine exit was in the present study defined as the last detection at a marine or estuarine receiver prior to detection at a receiver in freshwater. Because marine exit timing varied so little between years, brown trout were probably able to migrate into freshwater under most water discharge conditions, and so inter-annual environmental conditions in the rivers probably had little influence on the timing of their movement into freshwater. More likely, the timing of marine exit and freshwater entry was probably more influenced by life history, stage of maturity, and sex, as previously shown (Thorstad et al., 2016). Regarding the inter-annual marine residence time of brown trout, we found no significant intra-individual relationship. If the timing of marine entry is mainly influenced by the environmental conditions in the freshwater habitat, and there is strong consistency in marine exit timing, the varying environmental conditions in freshwater prior to marine entry are likely the main determinants of the marine residence time of brown trout. Alternatively, marine residence time in brown trout has been inversely correlated with individuals nutritional state (i.e., plasma triglycerides) prior to migration, such that depleted fish spend more time at sea reconditioning (Bordeleau et al., 2018).

In summary, this study revealed a strong tendency for individual, inter-annual repeatability in anadromous brown trout with respect to migratory decisions and marine habitat use patterns. While the causes remain obscure, this is the first study assessing the intra-individual behavioral repeatability of a salmonid fish species in relation to aspects of their spatiotemporal marine habitat use during consecutive annual feeding migrations. The findings of this study may have strong potential implications for management purposes. The role of intra-individual repeatability and the inter-annual behavioral response of anadromous brown trout to changing environmental conditions should therefore be a focus for future studies.

# DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

# ETHICS STATEMENT

The experimental procedures followed national ethical requirements and were approved by the Norwegian National Animal Research Authority (permit No. 2012/22965 & 2015/8518).

# AUTHOR CONTRIBUTIONS

SE was as the lead author responsible for the main part of the data analyses and on writing the paper. SE has further contributed considerable in field work and obtaining the dataset. XB was central in the initial idea of the paper and contributed significantly in terms of study design and contributed on feedback and input in the process of writing the paper. XB has further contributed in field work in the Tosenfjord study site and have during his work in this fjord system also contributed in the analyses these data. GC has contributed significantly in planning, study design, and the writing of the paper. JD was the PI of the two tracking studies that this paper is based on and has through this role significantly contributed to study design, field work, and project management. Further, he has contributed to this specific study with planning, study design, and writing of the paper.

# FUNDING

This study was part of the CHASES project funded by the Research Council of Norway (ref: 255110/E50). The tracking study in Tosenfjord was financed or supported by contributions from Sinkaberg-Hansen AS, the County Governor of Nordland, Nordland County Authority, The Norwegian Environment Agency, The river Åbjøra landowners' association, Phlates Eiendommer, The Natural Sciences and Engineering Research Council of Canada, Ocean Tracking Network, and the NTNU University Museum. The tracking study in Hemnfjord was financed or supported by contributions from the Hemne municipality, the County Governor of Sør-Trøndelag, Sør-Trøndelag County Authority, The Norwegian Environment Agency, AquaGen AS, the Norwegian institute for Nature Research, the Lake Rovatnet landowner's association, TrønderEnergi AS, DTU aqua, UiT the Arctic University of Norway, Ocean Tracking Network, and the NTNU University Museum.

### ACKNOWLEDGMENTS

The crew of RV Gunnerus, Lars Rønning, Jan Ivar Koksvik, Aslak Darre Sjursen, Martin Georg Hansen, Ola Magne Taftø, Hans Erlandsen, Vegard Pedersen Sollien, Paul Skarsvåg, Stein Hugo Hemmingsen, Kristian Lian, Embla Østebrøt, Hilde Dørum,

#### REFERENCES


Kristina Johansen and Ashley Ann, Ole Johan Hornenes, Torjus Haukvik, and Charlotte Hallerud are all thanked for their extensive help during fieldwork. The experimental procedures were approved by the Norwegian National Animal Research Authority (permission number 2012/22965 & 2015/8518).


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The handling editor and reviewer, NS, declared their involvement as co-editors in the Research Topic, and confirm the absence of any other collaboration.

Copyright © 2019 Eldøy, Bordeleau, Crossin and Davidsen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Tracking Landscape-Scale Movements of Snow Buntings and Weather-Driven Changes in Flock Composition During the Temperate Winter

#### Emily A. Mckinnon<sup>1</sup> \*, Marie-Pier Laplante<sup>2</sup> , Oliver P. Love<sup>3</sup> , Kevin C. Fraser <sup>4</sup> , Stuart Mackenzie<sup>5</sup> and François Vézina<sup>2</sup>

<sup>1</sup> Access & Aboriginal Focus Programs, University of Manitoba, Winnipeg, MB, Canada, <sup>2</sup> Groupe de Recherche sur les Environnements Nordiques BORÉAS, Département de Biologie, Chimie et Géographie, Centre d'Études Nordiques, Centre de la Science de la Biodiversité du Québec, Université du Québec à Rimouski, Rimouski, QC, Canada, <sup>3</sup> Department of Biological Sciences, University of Windsor, Windsor, ON, Canada, <sup>4</sup> Department of Biological Sciences, University of Manitoba, Winnipeg, MB, Canada, <sup>5</sup> Bird Studies Canada, Port Rowan, ON, Canada

#### Edited by:

Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway

#### Reviewed by:

Davide Scridel, Museo delle Scienze, Italy Sissel Sjöberg, Lund University, Sweden David Douglas, Royal Society for the Protection of Birds (RSPB), United Kingdom

\*Correspondence:

Emily A. Mckinnon emily.mckinnon@umanitoba.ca

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 30 April 2019 Accepted: 16 August 2019 Published: 03 September 2019

#### Citation:

Mckinnon EA, Laplante M-P, Love OP, Fraser KC, Mackenzie S and Vézina F (2019) Tracking Landscape-Scale Movements of Snow Buntings and Weather-Driven Changes in Flock Composition During the Temperate Winter. Front. Ecol. Evol. 7:329. doi: 10.3389/fevo.2019.00329

Nomadic movements of migratory birds are difficult to study, as the scale is beyond the capabilities of hand-held telemetry (10 s of kms) but too fine-scale for long-range tracking devices like geolocators (50–100 km accuracy). Recent widespread installation of automated telemetry receiving stations allowed us, for the first time, to quantify and test predictions about within-winter movements of a presumed nomadic species, the Snow Bunting (Pletrophenax nivalis). We deployed coded radio-transmitters on 40 individual Snow Buntings during two winters (2015-16 and 2016-17) in southern Ontario, Canada, and tracked movements over a 300 by 300 km area with 69–77 active radio-receiving stations (Motus Wildlife Tracking Network). To complement our tracking data, we also examined the influence of weather on the demographics of winter flocks at a single wintering site over 6 consecutive years (n = 9312 tagged birds). We recorded movements of 25 Snow Buntings from the deployment sites to 1–6 different radio recievers (mean 2.68 locations/bird). Birds traveled a minimum average distance of 49 km between detections (range: 3 to 490 km) in the core wintering period of Dec-Feb, and cumulative total movements ranged from 3 to 740 km (average 121 ± 46 km). In March distances between detections increased to an average of 110 km, suggesting an extended early-migration period. Overall, older birds (after-second year or older) tended to move more (higher cumulative distances traveled) than younger (first winter) birds, even during the Dec-Feb period. The long-term banding data revealed that larger, male birds were more likely to be captured in colder and snowier weather, relative to female and smaller birds, suggesting that they can withstand these conditions more easily owing to their body size. We have provided the first direct-tracking data on nomadic winter movements of Snow Buntings, and tested the hypothesis that winter weather drives flock composition at a single site. Site-specific banding data suggest that weather-related changes in flock composition could explain the nomadic, landscape-scale movements of Snow Buntings we observed by using automated telemetry. Future work should explore the importance of resource availability, competition, and predation risk as drivers of winter movements in Snow Buntings.

Keywords: nomadic migration, irruptive migration, differential migration, telemetry, songbird, movement ecology, weather

#### INTRODUCTION

Many avian species escape cold winters at breeding sites by migrating to lower latitudes (Newton, 2008). Birds that remain at high latitudes during winter must cope with energetically challenging conditions (Belda et al., 2007). Some individuals, especially when facing highly unpredictable environmental conditions, adopt a nomadic strategy (Stearns, 1976; Senar et al., 1992; Watts et al., 2018). The variability in nomadic movements, both within and across populations and species, remains poorly understood and difficult to study (McKinnon and Love, 2018). In a time of rapid environmental change, it is particularly important to determine factors that drive the within-winter movements of northern animals (Leblond et al., 2016) and the extent of landscape they use.

Variability in resources or access to resources across space and time generally correlates with higher rates of landscapescale movements of animals (Newton, 2006a,b). At its extreme, unpredictable and patchily-distributed resources can lead migratory birds to exhibit nomadism throughout their life cycle, stopping to breed only when and where resources are most abundant (Stojanovic et al., 2015). Irruptive species exhibit nomadic movements during the non-breeding season in order to track irregular booms in resource abundance outside of their breeding range (Bock and Lepthien, 1976; Smith, 1986; Senar et al., 1992). However, irruptions of northern-breeding species may also be simply a consequence of high productivity in the previous breeding season, and not necessarily that individuals are flexibly tracking resources (Curk et al., 2018).

Weather can also directly impact food availability (e.g., rain/drought affecting arthropod abundance, or covering of seeds by snow or ice) triggering landscape-scale movements (Mittelhauser et al., 2012). Some migratory songbirds may hedge their bets by remaining as far north as possible (in otherwise marginal habitats), and only moving south in the case of extreme disruption of resources (i.e., "fugitive" migration patterns) (Terrill and Ohmart, 1984; Watts et al., 2018). Unfortunately, little is known about exactly how and why variability in weather parameters (i.e., temperature, snow cover, humidity, wind) trigger movement in wintering songbirds (Macdonald et al., 2016; Laplante et al., 2019).

A number of non-resource-based hypotheses have also been proposed to explain species-, population- and individual-based variation in winter movement strategies (Catry et al., 2003; Nebel and Ydenberg, 2005; Macdonald et al., 2016). For example, since social dominance hierarchies can affect access to resources among individuals (Ketterson and Nolan, 1979), higher-ranked individuals in flocking species have less to gain by moving since they can maintain priority over resources at a given site. Highranking individuals may thus adopt a resident strategy relative to transient subordinates, even within the same population or species (Campos et al., 2011; Fudickar et al., 2013). However, these individual-level strategies have been difficult to explore, in part due to the inherent challenges associated with tracking species during the winter (McKinnon and Love, 2018).

Here, we investigate the within-winter, spatio-temporal dynamics of Snow Buntings (Plectrophenax nivalis), an Arcticbreeding migratory songbird, using automated radio-telemetry and a long-term banding dataset. Snow Buntings are longdistance migrants, traveling between breeding sites in the Arctic and wintering sites south of the boreal forest (Macdonald et al., 2012). Overall Snow Buntings show a complex migration pattern, with a migratory divide in North America at Hudson Bay (**Figure 1A**; Macdonald et al., 2012), and another in Greenland, where western and southeastern breeding birds migrate to eastern North America, and northeastern-breeding birds migrate to the Russian steppes (Lyngs, 2003). There are subspecies of Snow Buntings that are year-round residents in Alaska, USA, and Iceland, respectively (Montgomerie and Lyon, 2011).

During the winter, Snow Buntings form flocks and forage in agricultural fields, natural prairie, rocky coasts, and shores of the great lakes in southern Canada and the northern US (**Figure 1A**; Montgomerie and Lyon, 2011). Vincent and Bédard (1976) described a flock of about 100 individual Snow Buntings occupying a 4 km<sup>2</sup> home range from Dec-Mar; although they did not individually mark these birds. In contrast, banded Snow Buntings have been recaptured within the winter over 150 km from initial banding sites, and geolocator tracking suggested large scale movements by individual birds (Macdonald et al., 2012). Overall there is little quantitative information on the scale and drivers of winter movements in Snow Buntings. Here, we combined data from a collaborative network of automated radiotelemetry stations (Motus Wildlife Tracking Network) with a large, single-site winter banding dataset (n = 9,312 individuals captured over 6 years) collected by a participant in a citizen scientist program, to test several hypotheses regarding drivers of within-winter movements in Snow Buntings.

Snow Buntings are granivorous ground-feeders during the winter, thus variability in snow cover results in dynamic availability of food over time and space. Early snowfalls or late spring snow-storms directly affect how much seed is available for foraging buntings. Bird-banders that rely on baited traps to catch Snow Buntings are well-aware that a significant amount of snow is necessary before the birds are attracted to the bait, indicating that the snow prevents birds from accessing "wild" foods (CSBN, pers. comm.).

FIGURE 1 | North American range of Snow Bunting and study site. (A) Breeding grounds, wintering sites and year-round resident sites are indicated by color (range based on Montgomerie and Lyon, 2011); most birds in our study population likely breed in Greenland (based on band recoveries; summarized in Macdonald et al., 2012); however, it is possible that some individuals breed in the Canadian Arctic, thus three potential spring flyways are shown. (B,C) Show the ∼300 × 300 km region of our study, with active Motus receiving stations as red points, tags deployment sites as blue triangles and stations that detected Snow Buntings as yellow dots.

Snow Buntings also exhibit male-biased sexual size dimorphism and previous work has suggested that larger individuals may have a thermoregulatory advantage in colder weather, losing less heat per unit mass than small individuals (Macdonald et al., 2016; Laplante et al., 2019). Harsh winter weather could thus have a different impact on the energy budgets and movement decisions of males and females, due to differential thermoregulatory constraints among the sexes, potentially also explaining why females carry larger energy reserves per unit size than males (Macdonald et al., 2016; Laplante et al., 2019). Furthermore, since buntings are highly gregarious in winter and exhibit dominance hierarchies within flocks (Smith and Metcalfe, 1997a,b), social hierarchies could also generate differential pressure to move in interaction with weather (Ketterson and Nolan, 1979).

We hypothesized that if winter movements are a flexible response to maximize access to food, then deterioration in weather conditions at a single site should result in changes in flock composition as birds move around the landscape. Differences in structural size and social dominance by sexand age-classes should result in: (1) larger-sized individuals, (2) males, and (3) younger individuals, forming the main constituents of flocks on days of severe weather (i.e., colder, snowier days). We predicted that birds with these three phenotypes should be able to remain at the site and feed during difficult weather conditions because (1) larger-sized individuals should have better capacity to withstand adverse weather, and because (2) males and younger birds (within a sex) are generally dominant over food resources (Smith and Metcalfe, 1997b). Smaller and subordinate individuals (female, older birds) are also predicted to engage more frequently in winter movements, to compensate for a relatively higher daily energy cost of thermoregulation and reduced access to food.

#### METHODS

#### Animal Care Statement

All methods followed the Canadian Council for Animal Care recommendations, as reviewed by Environment and Climate Change Canada. Bird banding and handling permission was obtained from the Canadian Bird Banding Office. Tracking and tag-deployment protocols were reviewed and approved by the University of Windsor's Animal Utilization Committee (protocol AUPP # 9-14).

### Field Methods—Tracking

We used commercially available cracked corn as bait to capture 40 individual Snow Buntings in custom-made walkin traps (Love et al., 2012), at two long-term study sites in southern Ontario, Canada in February 2016 (Fergus; 43.838◦ N, −80.40806◦ W) and December 2016 (Long Point; 42.5792◦ N, 80.4309◦ W). Birds <1 year old were classed as "secondyear" (SY) and birds >1 year old as "after-second-year" (ASY) by examining plumage characteristics. We deployed 20 coded radio Nanotags (NTQB-1, Lotek, Inc., attached with a polypropylene string or elastic cord leg-loop harness) each year, and aimed to evenly tag birds from all representative demographic groups (Winter 1: n = 2 ASY males, 8 ASY females, 5 SY males, 5 SY females; Winter 2: n = 8 ASY males, 4 ASY females, 4 SY males, 4 SY females). All 20 birds were tagged on the same day in each winter, and banded (including assessment of age-class and sex) by experienced members of the Canadian Snow Bunting Network (D. Okines and D. Lamble). We tagged birds within their wintering range (**Figure 1A**) in the center of the Ontario Motus Wildlife Tracking System (Motus, www.motus-wts.org, Taylor et al., 2017), a network of hundreds of receiving stations (**Figures 1B,C** for location of tagging in Winter 1 and Winter 2). Worldwide, there were 359 Motus stations active for at least part of the tracking period in Winter 1, and 522 stations active for at least part of the tracking period in Winter 2.

We expected that Snow Buntings would be traveling distances in the range of 10–100 km in the winter, based on coarsescale data obtained previously by tracking using geolocators (Macdonald et al., 2012). We focused on southern Ontario as our main tracking site, which had an area of approximately 300 × 300 km (**Figures 1B,C**), and contained 69 active stations in Winter 1 and 77 stations active in Winter 2. The Motus Wildlife Tracking System had never before been tested for winter tracking, nor for a ground-foraging bird such as Snow Buntings. We anticipated that the ground-foraging behavior might reduce the number of detections received by stations, for two reasons (Lotek Inc., staff, pers. comm.): (1) the antenna would be in regular contact with the ground or snow, and (2) birds on the ground would be underneath the typical detection range of mounted receiving antennae. Therefore, we expected that most detections would be from birds in flight. Despite the above limitations, the location of the Motus array was ideal: well within the Snow Bunting winter range, and most stations embedded directly in agricultural habitat that Snow Buntings prefer. Southern Ontario is also bordered on three sides by large bodies of water, and Snow Buntings also use lakeshore habitats. In the north, our focal study area is bordered by the boreal forest—which is functionally a migratory barrier to Snow Buntings (Macdonald et al., 2012).

#### Field Methods—Banding

From 2009 to 2014; Snow Buntings were captured (by D. Lamble) using the same methods at a banding site in Fergus, Ontario (same as deployment site for Nanotags in Year 2; 43.838◦ N, −80.40806◦ W, **Figure 1C**). Winter banding efforts at Fergus occur every day of the winter between 9 a.m. and 2 p.m., starting when birds are first sighted in the area and continuing until they depart. For example, in 2015, D. Lamble banded 7038 Snow Buntings between 2 January and 10 March (note that we only used birds with measurements in analysis, so our sample size was only a fraction of the birds banded). Duration of banding session (hours traps were open) and number of captured individuals varied dependent on weather, presence of predators, and other factors; thus, the total number of captured individuals could not be analyzed as a response variable and we focused instead of characteristics of the birds captured (i.e., sex, age-class, size). Each bird was banded with a uniquely-numbered aluminum band (issued by the Canadian Bird Banding Office), and sex and age-class were determined morphologically according to Canadian Snow Bunting Network guidelines (Love et al., 2012). Unflattened wing chord was measured (for every bird possible) to the nearest 1 mm as an approximation of structural body size; Snow Buntings carry varying fat levels and muscle mass varies seasonally and would confound using weight as a measure of overall size (Macdonald et al., 2016). All individuals were measured by the same bander (D. Lamble). Only individuals for which a complete set of information was available were kept in the dataset (wing chord, sex, age-class).

# Weather Data

All weather variables except snow depth were obtained from the Environment and Climate Change Canada (ECCC) weather station nearest to Fergus, Ontario (12.9 km away) for the day of capture for each individual bird. We obtained snow depth data from the National Snow and Ice Data Center (NSIDC) (Brasnett, 1999; Brown and Brasnett, 2010) using a grid cell (24 km x 24 km) that included the banding site in Fergus. In total, we extracted eight daily weather variables, which based on previous work on Snow Buntings and others, are predicted to have an influence on snow bunting behavior (Orlowski and Gebski, 2007; Macdonald et al., 2016; Laplante et al., 2019): minimum temperature (◦C), mean temperature ( ◦C), maximum temperature (◦C), snow depth (cm), total snowfall (cm), absolute humidity (g/m<sup>3</sup> ), maximum wind gust (km/h), and cloud cover (0–10).

#### Data Analyses—Tracking

For Motus tracking data, we removed potential false detections (e.g., detections outside the bunting winter range [latitude < 35 N, longitude > −50 W], detections in months outside the lifetime of the tags [May-Nov], run length of detections < 2, frequency standard deviation > 0.1, signal > −30) following protocols recommended by researchers in the Motus community and the Motus RBook (Crewe et al., 2018). We then calculated cumulative minimum distances traveled per individual by using the R package "geosphere" (Hijmans, 2017), summing the Haversine (great-circle) distance between consecutive detections at different receiver stations. We compared the cumulative distance traveled by sex and age-class classes by using a two-factor analysis of variance, with an interaction between sex and ageclass. We examined these sex-age-class differences in movements by using the full dataset as well as specifically during the core wintering period of Dec-Feb, as we noted some larger movements in March that could be considered early spring migration.

#### Data Analyses—Banding

We used banding data collected between November 1st and March 20th for analyses of weather correlates of flock composition. This time period covers the wintering period for Snow Bunting populations within eastern North America (Vincent and Bédard, 1976; Montgomerie and Lyon, 2011; McKinnon et al., 2016). To avoid conflicts with subsequent model selection, models included only banding entries for which a complete set of weather variables was available (i.e., no missing data; final sample size for day of capture dataset, n = 9,312). We tested for the effect of daily weather on 3 response variables: structural size (i.e., wing length), sex, and age-class of birds, by using linear and logistic mixed-effect models with the prediction that structurally large birds, males, and younger (juvenile) birds would be found in the flock on cold/harsh days.

For logistic mixed-effects models for sex and age-class, females and young birds were the reference (dummy) categories. Banding year was included as a random variable in all models. To bring all numerical variables on the same scale, each predictor was standardized as a z-score using the R package "arm" (Gelman et al., 2018). Standardizing predictors was essential to interpret the relative importance of parameter estimates for subsequent model averaging in our analyses (Grueber et al., 2011).

We checked for multicollinearity between weather variables by calculating the Variance Inflation Factor (VIF) for each predictor in the full models using a threshold value of 10 (Quinn and Keough, 2002). One variable (mean temperature) had to be removed due to multicollinearity. Additionally, we calculated the average VIF across predictors to ensure the value was not substantially >1 (Chatterjee and Price, 1991; average VIF = 2.44).

Using the full models, we derived all possible sub-models from each set of predictors. We calculated an Akaike information criterion value (AICc) for each model. Models differed in their AICc only by small amounts, therefore we opted for model averaging (Grueber et al., 2011) using the list of models that fell within 1 AICc < 2 (Burnham and Anderson, 2003), especially as one of our objectives was to predict the relative importance of each variable in explaining variation of the dependent variable. We assessed the importance of each predictor by calculating the sum of the Akaike weights over all the models in the subset of 1 AICc < 2 where each predictor occurred (Burnham et al., 2011). Variables were considered significant when their confidence intervals did not include zero.

Statistical analyses were conducted using R 3.2.1 (R Development Core Team, 2018). The functions "lmer" (linear models) and "glmer" (generalized linear models) (Bates et al., 2014) and lmerTest (Kuznetsova et al., 2014) were used to run mixed-effects models. The function "dredge" and the function

"model.avg" in the package "MuMIn" was used for model selection and averaging (Barton, 2019). All coefficients for parameter estimates are reported ± conditional SEM.

#### RESULTS

#### Tracking Results

We obtained detections of 25 of the 40 birds outfitted with Nanotags (12 males, 13 females; 10 SY, 15 ASY). Individuals were detected at up to 5 individual Motus stations in Winter 1 and up to 6 Motus stations in Winter 2 (**Figure 2**). Signals were received by stations only during the day time, with the exception of one bird detected at 21:00 h on 1 April 2016 (**Figure 3C**; clearly migration). The average distance between sequential detections at different stations was 75.87 ± 16 km (n = 66 detections across all individuals for the whole tracking period). In the core winter period (Dec-Feb), the average distance between sequential detections was 49.27 ± 13.19 km (n = n = 37 detections), while in March and April the distances between detections were on average 109.81 ± 32.36 km (n = 29).

Signals received by the Motus stations were mostly typical of a "fly-by" (Crewe et al., 2018), indicating the individual birds were not remaining within the immediate station area (with some exceptions; see **Appendix A**—Bird 268 was detected 17 times in 20 days at a landfill station). For example, SY male #274 was first detected on January 4th, with 6 detections from one station in a 1-min period between 9:57 and 9:58 h. This individual was next detected at another station on the 30th of January, with 18 detections between 10:10 and 10:13 h. In total we detected SY male #274 at 4 different stations, each for short bursts, and the bird returned to the vicinity of one station 3 separate times after being detected elsewhere. Since detections at a given station were usually only over a span of a few minutes, in further analysis, we summarized data into hourly detections. For most birds this meant 1 detection per day, for a total of 66 detections for 25 birds.

Throughout the entire tracking period, cumulative distances traveled were longer for males (mean for males: 308.18 km,

1 breeding season and "juvenile" to birds that have yet to return to their breeding grounds.

females: 100.71 km, F(1, 21) = 5.85, P = 0.02), but not significantly different by age-class (SY: 110.16 km, ASY: 260.39 km, F(1, 21) = 2.81, P = 0.10) and the interaction between sex and ageclass was not significant (F(1, 21) = 3.14, P = 0.09). We counted the total number of days in the period within which birds were detected (i.e., last detection minus deployment date). Here we found a significant age-class effect (F(1, 21) = 7.52, P = 0.01) and age-sex-interaction (F(1, 21) = 5.35, P = 0.03) that we explored further with a Tukey post-hoc test. Adult males were detected over a significantly longer period (mean for adult males: 75 days) compared to juvenile males (22 days, P = 0.008) and juvenile females (35 days, P = 0.05) but not compared to adult females (40, days P = 0.06). Tracking period was not significantly different by age-class for females (P = 1.00), between juveniles by sex (P = 0.85), or between juvenile males and adult females (P = 0.59).

Because of our smaller sample size within the core Dec-Feb winter period (n = 15 individual birds, only 6 females), we compared sexes by using a t-test. Males traveled a cumulative distance of 152.68 km ± 76 SEM and females 74.79 km ± 20 SEM but the difference was not significant (t = −0.99, df = 9.04, P = 0.35). In Dec-Feb, we documented birds moving as far as 738 km in total (one adult male, who was tracked over almost the entire core Dec-Feb period, and picked up at 5 receiving stations). The average cumulative distance traveled during Dec-Feb was 121 km ± 46 SEM (n = 15 individuals). By March and April, cumulative distances increased to an average of 183.91 km ± 35.64 SEM.

#### Banding Results: Structural Size

Snow depth, snowfall, cloud cover, and maximum temperatures were the variables that best predicted variation in structural size of birds composing the flock on a given day (**Table 1**). Wing length of captured birds was longer on days where snow depth was higher (z = 2.52, β = 0.35 ± 0.14), snowfall was greater (z = 3.35, β = 0.24 ± 0.07), and the sky was cloudier (z = 4.82, β = 0.37 ± 0.07). Captured birds were also larger (i.e. longer wings) when maximum ambient temperature was lower (z = −3.93, β = −0.33 ± 0.07).

#### Banding Results: Sex and Age

Snow depth, snowfall, humidity, and maximum temperatures predicted the occurrence of sexes forming the flock at a single wintering site over the long-term (2009–2014; **Table 1**). Males tended to be captured more often on days where weather was snowier (effect of snow depth on sex: z = 3.34, β = 0.38 ± 0.12; effect of snowfall on sex: z = 3.99, β = 0.31 ± 0.08), more humid (z = 2.29, β = 0.27 ± 0.12) and colder (effect of maximum temperature on sex : z = −4.42, β = −0.46 ± 0.12).

Cloud cover, humidity, snow depth and snowfall were significant predictors of the age-class of birds captured on a given day. ASY (older) birds were more likely to be captured on days when weather was cloudier (z = 3.38, β = 0.17 ± 0.05), snowier (z = 2.32, β = 0.10 ± 0.04), less humid (z = −3.68, β = −0.23 ± 0.05). However, the influence of weather on daily variation in the age-class of birds forming the flock was not as strong as for structural size or sex. In fact, maximum temperature was not present in the final age-class model because it did not appear in TABLE 1 | Summary of standardized model coefficients for top models (females and juveniles are the reference categories in the sex and age-class models, respectively).


the top model set, indicating that it is not a useful predictor of age-class. Furthermore, most variables that were retained in the top age models had relatively low β values (i.e., <0.22).

# DISCUSSION

The extent of winter movements has only recently been revealed for many species with the use of miniaturized tracking technology (McKinnon et al., 2013a; McKinnon and Love, 2018), and the ecological and behavioral drivers of these movements remain poorly understood. The data we present here on winter movements of Snow Buntings are an important first step in understanding space-use year-round, which is critical for effective conservation assessment and actions (Marra et al., 2015). We also analyzed weather and patterns of winter movements, by using long-term banding data from a single site. Climate change is affecting high-latitude habitats at a greater rate than other environments in North America (IPCC, 2014), which may pose new challenges to Arctic-breeding, temperate-wintering bird populations (Rodenhouse et al., 2009; Princé and Zuckerberg, 2015; Williams et al., 2015). Understanding flexibility in patterns of movements of Snow Buntings may provide insight into how they could respond to warming temperatures.

Because of the sparse "fly-by" nature of our Motus detections, we were not able to analyze habitat use of buntings, because we could not pinpoint the exact location of the bird. We also could not analyze weather data as a correlate of movements, for several reasons. First, as above, we could not pinpoint the final destination of the bird. Detections indicate that the bird traveled near the station and thus justifies measuring a minimum distance traveled; however, we could not ascertain whether the bird stayed near the Motus receiver in most cases. Further, weather stations are irregularly spaced around southern Ontario, and for many Motus receivers, the nearest source of weather information would be the same, even if the receivers were fairly distant from each other. Finally, we had no control group, i.e., information on birds that did not move, therefore we could not assess whether individual birds were tracking weather or moving for some other reason. Despite the limitations of these tracking data, we were able to quantify for the first time with direct tracking, the extent of Snow Bunting movements across a large area.

Our tracking data indicate that most Snow Buntings use a "nomadic" strategy in winter (**Figure 3**), moving around an area of several hundred square-kilometers. Space-use of buntings was much larger than winter home-ranges reported for other flocking migratory songbirds (Shizuka et al., 2014; Weinkam et al., 2017). A recent geolocator study of Snow Buntings from a population breeding on Svalbard, Norway, suggested wintering birds were either completely stationary, or traveled to 1–3 separate sites, remaining locally at each place (Snell et al., 2018). Our previous geolocator work also suggested extensive winter movements (Macdonald et al., 2012), although it was difficult to quantify given the error range of geolocators (McKinnon et al., 2013b). Here, we found that 25 of 40 tagged birds moved 20–50 km every few days (see **Figure 3** for examples), which fits more clearly with the idea of "nomadic" or "wandering" movements (Rappole et al., 1989) than with birds occupying multiple stationary sites within a large home range. With our banding dataset, we found support for the prediction that birds move in response to changes in daily weather conditions, which in turn correlates with landscape-level demographic patterns of Snow Buntings (Macdonald et al., 2016).

Our results showing non-breeding season movements in the temperate zone are comparable to movements of some songbirds overwintering at tropical latitudes. For example, recent direct-tracking using geolocators has revealed that species expected to be stationary in winter are in fact making relatively large, within-season movements during the nonbreeding season (McKinnon et al., 2013a; Stutchbury et al., 2016). In some cases, these movements have been connected to patterns of rainfall associated with resource availability (Heckscher et al., 2015; Thorup et al., 2017). At the local scale, we found that for Snow Buntings, snow depth and temperature correlated with flock composition, suggesting individual birds are "wandering" to new sites when conditions become energetically unfavorable. Eastern Bluebirds (Sialis sialis) wintering in temperate areas diet-switched (from arthropods to fruit) when weather deteriorated, and flock size increased (Weinkam et al., 2017). Snow Buntings do not have the option of diet-switching in winter, as they only consume seeds and grains; selection may instead have favored a strategy of moving on when conditions are poor. However, for Snow Buntings and other species that form large overwintering flocks, the role of predators on flock composition and movements (Nebel and Ydenberg, 2005), intraspecific competition, or other social factors should be considered in future as potential proximate drivers of movement patterns.

The long-term banding data analyses revealed maximum temperatures and snow depth were the most important weather variables correlated with changes in flock composition by sex and size. These results are consistent with our earlier study showing that these variables were also central in informing daily fattening patterns in Snow Buntings (Laplante et al., 2019). Snow depth may rank foremost, considering that snow buntings are granivorous ground-feeders which can have more difficulty accessing resources in areas of high snow depth, and record-high numbers of Snow Buntings have been observed during warmer (less snowy) winters in Central Europe (Orlowski and Gebski, 2007).

Structurally larger individuals and/or male snow buntings were predominantly found in a flock on days of severe weather conditions (i.e., cold, snowy, cloudy, humid). We found a similar pattern in a study of sex ratio in wintering flocks across a broad range of wintering sites, where larger birds of both sexes were found at colder, snowier sites (Macdonald et al., 2016). Large individuals have less body surface area relative to their volume and therefore should loose less heat per unit mass than small individuals, suggesting that on cold days, smaller birds may "wander" until they find a more energetically favorable foraging area.

Movements may also be influenced by male dominance over food resources, where females may be forced to move away when weather is severe because they are simply unable to access food and accumulate the required fat store to survive cold nights. Laplante et al. (2019) reported that for their body size, females were carrying more fat than males. In Snow Buntings wintering in Scotland, females were more likely to move when flock sizes were larger, presumably because they were excluded by dominant individuals when intraspecific competition increased (Smith and Metcalfe, 1997a).

In contrast, our tracking data seemed to suggest that males, and particularly adult males, were moving more often. We detected adult (ASY) male movements over a longer time period, and covering the longest cumulative distances, which suggests these individuals are moving around the landscape more than other age-class-sex groups. Given that stations were 30–40 km apart and the range of each station was up to 20 km, we assume that birds that moved more and over longer distances would be more likely to be detected. Tag detections are biased toward capturing birds in flight, as opposed to foraging on the ground. Fifteen of the 40 tagged birds were never detected after the tag deployment. These birds could have either remained locally at the tagging area until the tag batteries died in late spring, or moved in such a way as to avoid detection by any of the 69–77 active stations within the ∼300 × 300 km study area. Alternatively, batteries on the tags could have failed, or the birds been taken by predators or otherwise succumbed. Given the novelty of tracking using these methods during the winter, and on a ground-foraging bird, much remains to be learned about the effective range of the tags and capabilities of the tags and receivers.

Some adult males were clearly departing on spring migration during the tracking period (see individual maps in **Appendix**), similar to findings of an earlier study of migratory phenology in Snow Buntings that found older males left wintering sites prior to all other demographic groups (McKinnon et al., 2016). However, even within the Dec-Feb period, males (especially older males) tended to move more than females (cumulative distance in Dec-Feb: 190 km for adult males, 84 km adult females, 13 km juvenile males, 37 km juvenile females), although the differences were not significant. If a cold/snowy weather system moves in across a large enough area, it may be that males move into sites previously occupied by more females, which would account for the correlation between weather and sex-ratio (Macdonald et al., 2016), as well as the increased movements we observed from males by direct tracking. Our tracking site in southern Ontario experiences winters that are in general, milder than in other areas of the Snow Bunting wintering range, such as the northern Prairies in Manitoba, Canada, where winter temperatures are regularly below −30◦C and where mostly adult (i.e., ASY) males are found (B. Maciejko, pers. comm.). Further direct-tracking or even experimental work in more extreme wintering sites might shed additional insight into flock dynamics and their association with weather.

There was a weak tendency for older (ASY) birds to be found in the flock on days with more snow cover and snowfall (**Table 1**). Although younger birds have been shown to be dominant in flocks, older birds have improved feeding efficiency (higher peck rate) (Smith and Metcalfe, 1994, 1997b). We also found that younger birds tended to be detected less than adult birds. Firstwinter birds might have higher mortality rates than experienced adults, which could be an important driver of overall population demographics (Robinson et al., 2004). It could also be that younger birds with less experience tend to remain longer at a given site, thus resulting in few detections at stations away from the initial deployment site. Confidence intervals for most weather variables included zero in the age-class model and the size of variable effects (Beta values) were relatively low, thus we cannot be confident that age-class is an important factor influencing movements in Snow Buntings until further work is conducted.

# CONCLUSION

Factors that drive the nomadic movements of temperate wintering species, as well as the characteristics of those movements, have been difficult to study. We combined local scale flock dynamics with regional tracking to show that weather influences site tenure by sex and structural size, and that movements in the non-breeding season vary by sex and, in some cases, age-class. We showed that Snow Buntings are weather-sensitive and that winter conditions can influence their behavior and distribution. While nomadism seems to be one of the strategies used by birds to respond and adjust to winter conditions, it is unknown whether it could be sustained if winter weather becomes more extreme, as predicted by climate models. Further, experimental work is required to strengthen understanding of the energetic challenges faced by male and female buntings. Improving our understanding in general of the nomadic movements of small species that travel widely remains a frontier of movement ecology studies, one which will continue to

# REFERENCES


be aided by the miniaturization of tracking technology that does not require retrieval for data recovery.

# DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

#### ETHICS STATEMENT

We followed animal care guidelines of the Canadian Council for Animal Care, the Canadian Bird Banding Office regulations (Environment and Climate Change Canada), and the Univ. of Windsor Animal Utilization Project Proposal (AUPP#9-14).

### AUTHOR CONTRIBUTIONS

EM and MP-L: ideas, fieldwork, analysis, interpretation, and writing. OL: ideas, writing, field support, analysis, and interpretation. KF: ideas, writing, analysis, and interpretation. SM: fieldwork, analysis, and interpretation. FV: ideas, field support, writing, analysis, and interpretation.

# FUNDING

Funding and other support was provided by the University of Windsor, the Université du Québec à Rimouski, NSERC, FRQNT, the Canada Research Chairs Program, Bird Studies Canada and the Mitacs Postdoctoral Fellowship Program.

#### ACKNOWLEDGMENTS

For field assistance and logistical support, we thank David Lamble, David Okines, the Canadian Snow Bunting Network, Chris Harris, and Matilda Fraser. We also thank Alain Caron and Nicolas Casajus for their help with statistical analyses of banding data.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00329/full#supplementary-material


Behav. Ecol. Sociobiol. 65, 23–35. doi: 10.1007/s00265-010- 1029-6


pay-off for the migratory swift parrot. J. Anim. Ecol. 84, 1194–1201. doi: 10.1111/1365-2656.12375


a mechanistic approach. Gen. Comp. Endocrinol. 256, 112–122. doi: 10.1016/j.ygcen.2017.07.027


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Mckinnon, Laplante, Love, Fraser, Mackenzie and Vézina. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Automated VHF Radiotelemetry Revealed Site-Specific Differences in Fall Migration Strategies of Semipalmated Sandpipers on Stopover in the Gulf of Maine

Rebecca L. Holberton<sup>1</sup> \*, Philip D. Taylor 2,3, Lindsay M. Tudor <sup>4</sup> , Kathleen M. O'Brien<sup>5</sup> , Glen H. Mittelhauser <sup>6</sup> and Ana Breit <sup>7</sup>

*<sup>1</sup> Laboratory of Avian Biology, School of Biology and Ecology, University of Maine, Orono, ME, United States, <sup>2</sup> Department of Biology, Acadia University, Wolfville, NS, Canada, <sup>3</sup> Bird Studies Canada, Port Rowan, ON, Canada, <sup>4</sup> Bird Group, Maine Department of Inland Fisheries and Wildlife, Bangor, ME, United States, <sup>5</sup> United States Fish and Wildlife Service–Rachel Carson National Wildlife Refuge, Wells, ME, United States, <sup>6</sup> Maine Natural History Observatory, Gouldsboro, ME, United States, <sup>7</sup> School of Biology and Ecology, University of Maine, Orono, ME, United States*

#### Edited by:

*Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway*

#### Reviewed by:

*Jonathan B. Cohen, State University of New York College of Environmental Science and Forestry, United States Jaime A. Collazo, North Carolina State University, United States*

\*Correspondence:

*Rebecca L. Holberton rebecca.holberton@maine.edu*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *29 April 2019* Accepted: *14 August 2019* Published: *13 September 2019*

#### Citation:

*Holberton RL, Taylor PD, Tudor LM, O'Brien KM, Mittelhauser GH and Breit A (2019) Automated VHF Radiotelemetry Revealed Site-Specific Differences in Fall Migration Strategies of Semipalmated Sandpipers on Stopover in the Gulf of Maine. Front. Ecol. Evol. 7:327. doi: 10.3389/fevo.2019.00327* The Gulf of Maine has long been recognized as a major stopover area for shorebirds in fall. Knowing how birds move within and beyond the region will be paramount to protecting threatened shorebird habitat. To determine stopover behavior during fall migration (2013–2017) in Maine, 180 (104 AHY, 76 HY) Semipalmated Sandpipers, *Calidris pusilla*, were tracked using VHF radiotelemetry and an extensive array of automated receivers (Motus Wildlife Tracking System). Birds tagged at three locations along the Maine coastline showed no effect of age class or stopover site on body condition (body mass, estimated fat mass) or stopover length (post-capture detection period). However, movement after departure varied greatly among sites. Few birds captured at the northern-most site ("Downeast," *n* = 71), which had the greatest amount of mudflats and offshore roost sites and the least amount of human disturbance, were detected beyond the initial tagging location, suggesting that they, like birds in the Bay of Fundy just to the north, initiated trans-oceanic flights from that location. At the Downeast site, leaner birds remained significantly longer than fatter birds, suggesting that time of departure there depended on energy reserves, which would be critical for making extensive flights. In contrast, over half of the birds tagged further south (Popham Beach, *n* = 59; Rachel Carson NWR, *n* = 50) were later detected at coastal locations to the north (few) or to the south (most). Stopover period at these sites was independent of fat, suggesting that other factors (e.g., feeding/roosting site availability, human activity) influenced departure decisions. In Maine, Semipalmated Sandpipers, regardless of age, may move north (Downeast) or south (e.g., Cape Cod, Rhode Island, Long Island Sound) where the local topography, habitat characteristics (feeding/roosting sites), and/or lower human activity, may best enable them to initiate trans-oceanic flights to the wintering grounds. Future study should determine if variation in stopover behavior is population-specific and if population-segregation occurs in Maine. Use of automated VHF radiotelemetry has led to a greater understanding of stopover behavior and the degree of connectivity among stopover sites, which should be taken into account for conserving migratory bird habitat across broad spatial scales.

Keywords: Semipalmated Sandpipers, Calidris pusilla, migration, stopover, radiotelemetry, shorebirds, habitat

## INTRODUCTION

Many Arctic-breeding shorebird populations have declined rapidly over the past few decades, with some populations decreasing by as much as 60% (for review, see 2012 North American Bird Conservation Initiative Report; U.S. Shorebird Conservation Plan Partnership, U.S. Shorebirds of Conservation Concern, 2015). Hunting on the wintering grounds and habitat loss experienced throughout the annual cycle are believed to be the primary factors underlying overall species declines (Brown et al., 2017). Within a species, trends for individual breeding populations may differ due to disparate factors that individual breeding populations experience on their respective breeding or wintering areas as well as along populationspecific migratory routes. For example, although Semipalmated Sandpipers, Calidris pusilla, breed across the North American Arctic, different breeding populations show a high degree of geographic segregation in their respective wintering areas and migration routes (Andres et al., 2012; Gratto-Trevor et al., 2012a; Brown et al., 2017). Western and central breeding populations appear to be stable or slightly increasing over the past few decades, but eastern Semipalmated Sandpiper populations have shown little to no increase, with some indications that they continue to decline (Andres et al., 2012; Gratto-Trevor et al., 2012b; Morrison et al., 2012; Smith et al., 2012; Brown et al., 2017). This pattern prompted Andres et al. (2012) to propose that eastern breeding populations of this species be considered "of high conservation concern." Understanding the behavior and ecology of different Semipalmated Sandpiper populations throughout the annual cycle will be key to determining how each breeding population is ultimately regulated and, consequently, how to manage resources to support them.

Recently, using light-level geolocators, Brown et al. (2017) confirmed the degree of geographic segregation of different Semipalmated Sandpiper breeding populations during migration as well as during the stationary periods of breeding and wintering. In spring, individuals from western populations of this species retrace much of their southward journey through the interior of North America as they move north to return to their respective breeding areas. In contrast, eastern-breeding Semipalmated Sandpipers exhibit an "elliptical migration" pattern between the wintering and breeding areas: in spring, these birds depart the wintering grounds in the Caribbean and along the east coast of South America and move northward along the U.S. Atlantic coastline. When most birds reach New Jersey's Delaware Bay, they stage there for as much as several weeks to acquire sufficient energy reserves needed to initiate extended flights that bypass New England on the way to their respective breeding grounds (Gratto-Trevor and Dickson, 1994; Gratto-Trevor et al., 2012a; Figures 2, 3 in Brown et al., 2017). In autumn, while up to 500,000 Semipalmated Sandpipers stage each year in the Bay of Fundy region, as many as 100,000 more arrive along the Gulf of Maine coastline, from the Bay of Fundy to Cape Cod (L. Tudor, unpubl. data, McNeil and Burton, 1977; Fefer and Schettig, 1980; Lank, 1983; Hicklin, 1987; Dunn et al., 1988; Tudor, 2002; Maine Department of Inland Fisheries Wildlife, 2015). Most of these birds rest and refuel in coastal estuaries before they initiate non-stop flights over the North Atlantic to reach the wintering grounds (Gratto-Trevor et al., 2012a; Brown et al., 2017).

Birds preparing for extended non-stop flight require stopover sites that offer abundant high quality food, low predation pressure, and roost sites where they can efficiently rest as they refuel before departing. The Gulf of Maine region (primarily the upper Bay of Fundy and coastal Maine) offers a wide diversity of features that support feeding and roosting, including numerous tidal mudflats, marshes, and beaches, as well as many islands and rocky ledge outcroppings offshore. Unfortunately, suitable shorebird habitat along the entire Atlantic coast continues to be threatened by increased human activities and rapid sea level rise. Understanding shorebird movement within and between stopover areas along the coast, therefore, is key to identifying and managing resources needed by birds during this critical stage of the annual cycle.

Much of what is known about stopover behavior and ecology of Semipalmated Sandpipers during fall migration in eastern North America comes from studies in the Bay of Fundy (McNeil and Burton, 1977; Hicklin, 1987; Mawhinney et al., 1993; Hicklin and Chardine, 2012; Mann et al., 2017); less is known about their movements in the Gulf of Maine region to the south (Dunn et al., 1988). Maine's coastal habitats, in particular, are under increasing pressure from residential and commercial development, coastal engineering, aquaculture, rockweed harvest, threats of sea level rise due to climate change, as well as disturbance issues associated with recreationalists and pet owners, which have recently escalated in Maine (Tyrrell, 2005; Maine Department of Inland Fisheries Wildlife, 2015). Semipalmated Sandpipers are currently listed as a Species of Special Concern and a Priority Species for conservation in Maine (Maine Department of Inland Fisheries Wildlife, 2015). Documenting shorebird behavior during migration in Maine has become a priority for effective conservation of the region's coastal habitats.

To document stopover behavior of individual Semipalmated Sandpipers in Maine, we capitalized on the international network of automated VHF radiotelemetry stations recently established as the Motus Wildlife Tracking System (described in Taylor et al., 2017). We first initiated a study at one location that, through an extensive collaboration among federal and state resource management agencies, several NGOs, and private landowners, subsequently led to an opportunity to add two more sites that varied in local topography as well as in the intensity of human activities (e.g., shellfish harvesting, beach recreation) during fall (July–October) migration (see **Figure 1**). The locations included a relatively undisturbed site in northern Maine (Pleasant Bay in Downeast Maine), a state-owned beach with extensive recreational use in mid-coast Maine (Popham Beach State Park), and a site situated within the National Wildlife Refuge system in southern Maine (Rachel Carson National Wildlife Refuge) that has limited human recreational use of shorebird habitat during fall migration (**Figure 1**).

Our objectives were to determine if groups of individuals using these sites differed in morphological characteristics, which might indicate spatial segregation by different breeding populations, as this could be used to understand changes in bird abundance in Maine possibly reflecting the disparate trends reported for different breeding populations (Andres et al., 2012; Gratto-Trevor et al., 2012b; Morrison et al., 2012; Smith et al., 2012; Brown et al., 2017). During the entire study, multiple

automated receiver towers within the Motus array extended along the Atlantic coastline from Atlantic Canada to as far south as the Carolinas, a flyway that includes major shorebird staging areas during fall migration (**Figure 1**). The extensive array of automated towers allowed us to examine stopover length (as estimated by the post-capture length of stay at the tagging location) and subsequent regional-scale movements (beyond the initial capture locations) of individuals on stopover at these different sites.

We tested several hypotheses made before and after data were collected from all three locations. We first examined agerelated differences in body condition and stopover length. As young birds on their first migration may not be as efficient in foraging or in predator detection (Cresswell, 1994; Fernandez and Lank, 2006; Stillman et al., 2007; van den Hout et al., 2017), we predicted, a priori, that young birds would have lower energy reserves at the time of capture and/or longer stopover periods to rest and refuel, compared to adults. We also predicted, a priori, that, regardless of age, birds with greater energy reserves at the time of capture would be more ready to depart (have shorter postcapture detection period) than birds with less fat (c.f. Williams et al., 2007). Finally, the ability to opportunistically compare bird movements at three sites in Maine allowed us to examine, a posteriori, potential differences in stopover behavior related to site-based differences in key topographical features (geographic location with respect to a goal, tidal mudflats vs. sandy beach, human disturbance level). We assumed throughout the analyses that age, body condition, and stage of stopover did not influence the likelihood of a bird being captured.

### MATERIALS AND METHODS

#### Tagging Locations, Automated Radiotelemetry System, and Tower Deployment

Bird capture and tagging activities, and automated receiver tower deployments were undertaken at three locations along the Maine coast designated as: (1) "Downeast" (2013–14, **Figure 1**), a relatively undisturbed site in Pleasant Bay, Downeast Maine that offers extensive feeding areas during low tide and many roost sites on rocky outcroppings offshore, (2) "Popham" (2015– 16, **Figure 1**), a public recreation area at Popham Beach State Park in mid-coast Maine that is used extensively by recreational beach-goers and hikers throughout the tidal cycle, and (3) "Rachel Carson" (2014–15, **Figure 1**), situated within the Rachel Carson National Wildlife Refuge in southern Maine that offers limited recreational use near where shorebirds feed on exposed mudflats during low tide but experiences extensive recreational use along sandy beaches where shorebirds are seen throughout the tidal cycle.

At each tagging location, we deployed two or more automated receiver stations that contributed to the Motus array. Motus towers designated as "local" for each of the three tagging locations, as well as additional towers that were active along the coastline from the Bay of Fundy to South Carolina during the study periods for each year of the studies, are shown in **Figure 1**. All components of the telemetry system, including descriptions of the VHF radiotags and automated receivers ("Sensorgnomes") are described by Taylor et al. (2017), and detailed maps of all tower locations for each year are available on the Motus web page (www.motus.org). The towers and associated antennas at each tagging location were positioned with respect to local topography to maximize detection of tagged birds moving within and beyond the sites. Each VHF tag was confirmed to be operating when it was deployed. In addition to the tag deployment days, towers were checked every 2–4 weeks during and for several months after the region's fall migration period (approx. early July through early November) to confirm operation and to collect detection data for processing (described below). Data from our tagged birds detected beyond our local tagging sites were processed by Motus (www.motus.org), downloaded using the "tagme" function in the Motus R package, and post-processed as described in Crewe et al. (2018).

**"Downeast," Maine** (Washington County, 44.55◦ N, 67.80◦ W)—From July through November, 2013 and 2014, we deployed two automated radiotelemetry receiver towers in the Pleasant Bay region within 5 km of the capture locations (**Figure 1**). This area includes the estuaries of the Mill, Harrington, and Pleasant Rivers and contains important shorebird habitat during fall migration. Extensive tidal mudflats associated with the numerous creeks and river outflows provide valuable feeding areas during low tide. In addition to vegetated saltpans, numerous rocky ledges offshore provide roosting sites during high tide. The area surrounding the study site is rural and not heavily developed. Human disturbance is low as there is limited sandy beach and, during the fall migration period, the exposed tidal mudflats are used for limited shellfish harvesting by hand. In both years, one telemetry station (Pineo Point, Washington County: "PINEO": 44.5645<sup>o</sup> N, 67.8066<sup>o</sup> W) was set up immediately adjacent to extensive mudflats exposed during low tides at the southern tip of a peninsula located between the Mill and Harrington Rivers. Only one antenna (174<sup>o</sup> ) was deployed on the PINEO tower in 2013. Three antennas (195, 151, 243<sup>o</sup> ) were deployed on the same tower in 2014. In both years, a second tower was set up at Seal Cove ("SECO": 44.5431<sup>o</sup> N, 67.7528<sup>o</sup> W) south and east from Pineo Point, on the eastern side of Pleasant Bay. Only one antenna was deployed on the SECO tower in 2013 (239<sup>o</sup> ) and three were deployed in 2014 (225, 263, 333<sup>o</sup> ). Several additional towers deployed and operated by USFWS-Maine Coastal Islands National Wildlife Refuge and other Motus participants in 2013 and 2014 were included as part of the array designated for this study as "local". These were located at the southern opening of Pleasant Bay: Nash Island ("NASH" 2013–14: 44.4648<sup>o</sup> N, 67.746<sup>o</sup> W; 2013–14) and Jordan's Delight ("JORDL" 2013: 44.44269<sup>o</sup> N, 67.8241<sup>o</sup> W), on the nearby Petit Manan Peninsula just to the east (PMP 2013 = PMP1 2014: 44.4131<sup>o</sup> N, 67.9058<sup>o</sup> W; PMP\_WL 2013: 44.4015<sup>o</sup> N, 67.8965<sup>o</sup> W; PMP2 2014: 44.4085<sup>o</sup> N, 67.9050<sup>o</sup> W), and on Petit Manan Island about 10 km to the southeast of Pleasant Bay (PMI 2013: 44.5385<sup>o</sup> N, 67.8805<sup>o</sup> W). Additional towers, considered outside the "local" tagging location, were deployed in 2013–14 (by multiple Motus participants) along the coast to the north and east and at locations on the mainland and on offshore islands to the southwest (**Figure 1**).

**"Popham," Popham Beach State Park** (Sagadahoc County, 43.4417<sup>o</sup> N, 69.4759<sup>o</sup> W)—This public beach area is bordered by the mouth of the Kennebec River to the north and the Morse River to the south. Only one telemetry tower was deployed in 2015 ("POPH": 43.736<sup>o</sup> N, 69.7997<sup>o</sup> W; 2015 antenna directions = 128, 164, 224<sup>o</sup> ). This tower was redeployed, at the same location, in 2016 (2016 antenna directions = 84, 159, 244<sup>o</sup> ) along with a new tower ("SEGUIN": 43.7099 <sup>o</sup> N, 69.7596<sup>o</sup> W; antenna directions = 224, 278, 344<sup>o</sup> ) set up on Seguin Island, one of several small rocky offshore islands within 1–2 kilometers east of the beach. These two towers, designated as "local" for this study, collectively provided extensive detection coverage in the immediate area, including the sandy beach and the small tidal lagoon behind it at the mouth of the Morse River, Seawall Beach just to the south, and the few rocky ledge outcroppings just offshore from these areas. Coverage also extended to the local area's tidal marsh inlets. The Park provides activities for the public such as swimming, kayaking, fishing, picnicking, and hiking and is considered the most visited public state park in Maine during the summer and fall. Dogs are not allowed on the beach 1 April−30 September to protect nesting seabirds and shorebirds as well as birds on migration. Additional towers outside the "local" tagging location were deployed in 2015–16 (by multiple Motus participants) along the coast to the north and east and along the coast to the south (**Figure 1**).

**"Rachel Carson," Rachel Carson National Wildlife Refuge** (RCNWR, 43.2100◦ N, 70.3228◦ W, York and Cumberland Counties)–This study area includes the Webhannet, Little, and Mousam Rivers, with estuarine areas dominated by tidal mudflats and saltmarsh. The sandy beaches are heavily visited by people, particularly during the late summer months that overlap with shorebird migration (Aug.–Sept.). Pedestrian access to the estuarine areas (where birds were captured during low tide) is limited and human disturbance in those areas is generally low. Unlike the Downeast and Popham Beach locations, the Refuge area lacks offshore roosting sites.

Two telemetry towers, considered "local" for this study, were each deployed in 2014 and 2015: "FURBISH" (43.2819◦ N, 70.5817◦ W, 2014 antenna directions = 17, 197, 85, 260◦ ; 2015 antenna directions = 18, 85, 197◦ , and "WNERR" (43.3351◦ N, 70.5491◦ W, 2014 antenna directions = 223◦ ; 2015 antenna directions = 145, 240, 281◦ ). Additional towers, considered outside the "local" tagging location, were deployed during the 2014–15 season (by multiple Motus participants) along the coast to the north and east and along the coast to the south (**Figure 1**).

#### Capture and Handling

All birds were captured during daylight hours and most birds were captured by mist net while feeding on exposed mudflats. The 2–3 h period of rising tide leading to peak high tide provided the greatest capture rates as birds concentrated in large numbers on exposed feeding areas as these were being gradually submerged by the rising water. During the 6 h surrounding the period of peak high tide, such feeding areas were entirely unavailable and birds either moved to feeding areas further upstream in tidal marshes or moved to available roost sites on exposed rocky ledges offshore or on sandy beaches along the mainland (as confirmed with hand-held telemetry receivers). Roosting birds were captured by mist nets on the beach during the high tide period at Popham Beach and at Rachel Carson NWR and by a net gun (rifle cartridge or CO<sup>2</sup> powered) deployed from the bow of a small boat as birds roosted on exposed rocky ledges offshore at the Downeast site.

Birds were handled by crews that worked at more than one site and an attempt was made to standardize measuring techniques among different crew members. Regardless of capture method, birds were immediately placed in small cloth bags until processed for body mass (with a hanging Pesola spring scale or table balance, to nearest 0.5 g), wing length (flattened wing to nearest 0.5 mm), and culmen length (to nearest 0.5 mm) and to an age class (adult: After Hatching Year = AHY; juvenile: Hatching Year = HY; Unknown = U) based on plumage characteristics (Pyle, 1997). We also estimated the amount of lipid energy stores independent of body size, using a formula developed by Dunn et al. (1988) based on Semipalmated Sandpipers collected in Maine: estimated fat mass = total body mass – fat free mass; fat free mass = (−9.0513 + [0.3134 X wing length]). Estimated fat mass was not available for three birds because either wing length or body mass was not recorded. After processing, each bird was banded with a uniquely numbered aluminum USGS band, a pair of color bands that indicated they were banded in Maine, and a green 3-character plastic flag band for individual identification that could be read from afar.

We attached coded VHF radiotags (Lotek model NTQB-2), using a small drop of quick drying "super glue" to a few clipped body feathers along the back just above the "rump." Tags were applied only to birds weighing at least 20 g but no more than 34 g so as to increase the opportunity to monitor bird movements during the period leading up to departure from the immediate area. Although this may have produced a sample population not truly representative of the groups of birds on stopover at these areas, birds weighing <20 g may not have been doing well (and thus may have been influenced by tagging), and those above 34 g may have been more likely to depart the site too soon after tagging to reveal much information about movement within the stopover area. As with many shorebird species, adult Semipalmated Sandpipers pass through the region in fall earlier than young of the year and capture efforts were targeted to yield an equal number of each age class at each site each year if weather conditions allowed.

All activities related to bird capture and handling were reviewed and approved by UMaine IACUC (#A2013-01-02 to RLH) and were performed under federal and state permits to U.S. Fish & Wildlife Service (to KMO) and Maine Department of Inland Fisheries and Wildlife (to LMT).

#### Data Handling and Analyses

Detection data collection, storage, and extraction are described in Taylor et al. (2017) and Crewe et al. (2018). Detection data (initial and final timestamp of individual VHF coded signals) for each bird were compiled into a spreadsheet containing information as to age class, body mass, and body size measurements. As our proxy for length of stopover at each of the three study locations, we determined the period of time between the release of the bird at the time of tag attachment and its final detection by any one of the antennas mounted on towers designated as "local" at each of the three sites. For the model of the nanotags used throughout the study, each tower had an estimated detection radius of up to 10 km under ideal conditions.

Statistical analyses were done using R, version 3.3.0 (2016- 03-10), Copyright © 2016, The R Foundation for Statistical Computing, on a Windows 10 PC. For all analyses, we fit generalized linear models (using the glm function in base R) with "site" and "age" as categorical predictors. For culmen, wing length, mass, and estimated fat mass (EFM), we fit models with an identity link and Gaussian errors. For the amount of time spent at the local site after being tagged and released (post-capture detection period, PCDP) we fit a model with an identity link and Gaussian errors, but we transformed the response using a square root transformation (to improve fit by reducing the influence of long-staying individuals on the parameter estimates). For the model assessing whether a bird remained at the initial tagging area, or disappeared completely from the initial tagging area (suggesting a likely direct departure over water to the wintering grounds), we fit a binomial model (logit link and binomial errors).

To assess the validity of assumptions regarding detection at local sites, and thus our ability to infer different patterns of departure at the three sites, we fit simple multi-state mark-release-recaptures models using the RMark interface in R to program Mark (White and Burnham, 1999; www. phidot.org). We fit a single model with two states (local and non-local; defined as above) with time-varying survivorship for each state and location and with detection probability allowed to vary between the two states (but common among locations). Capture periods were defined as the 24 h day beginning at midnight. For all models, we assessed overall model fit by examining standard errors of coefficients and using residual plots. We report the results of an analysis of deviance table (F-tests) and interpret interactions using plots.

#### RESULTS

Summary information for each site and age class, including morphometrics and body condition at the time of capture, stopover length (PCDP), and the proportion of birds detected beyond the initial tagging location can be found in **Table 1**.

#### Across Site Summaries

We found a significant effect of site on wing length [F(2, 173) = 20.6, p < 0.001] and on culmen length [F(2, 174) = 3.3, p = 0.04], but no evidence for differences in morphological features with age [culmen: F(1, 174) = 0.4, p = 0.5; wing: F(2, 173) = 0.04, p = 0.8]. Birds captured at the Downeast site had the smallest culmens (∼19 vs. ∼19.5 mm; **Table 1**), but larger wings than those captured at Popham Beach and Rachel Carson (∼97 vs. ∼95 mm; **Table 1**; **Figure 2**). We found no evidence for an effect of age [F(1, 172) = 1.34, p = 0.25] or site [F(2, 172) = 0.85, p = 0.43] [or their interaction: F(2, 172) = 0.49, p = 0.62)] on mass or on size-corrected estimated fat mass (EFM) [age: F(1, 173) = 1.2, p = 0.30; site: F(2, 173) = 0.87, p = 0.40], suggesting that all individuals, regardless of location or age, were in similar energetic condition at the time of capture (**Table 1**).

We found no effect of age [F(1, 143) = 0.04, p = 0.8] on PCDP, but evidence for a significant interaction between site and EFM [F(2,143) = 5.1, p = 0.007) on PCDP. Individuals from Popham and Rachel Carson remained in the local area irrespective of their original energy stores (Popham Beach 2015–16: R <sup>2</sup> = 0.0407, P = 0.1325, df = 56; Rachel Carson NWR 2014–15: R <sup>2</sup> = 0.0114, P = 0.4899, df = 43, **Figure 3**). In contrast, the amount of time that birds on stopover remained at the Downeast site declined linearly with increasing fat mass (Downeast 2013–14: R <sup>2</sup> = 0.2664, P < 0.0001, df = 63, **Figure 3**).

Daily probability of detection (Pd) was high when birds were in the designated local tagging area (P<sup>d</sup> = 0.96; 0.950–0.974; estimate plus lower and upper CL) but was considerably lower when birds moved out of the local area (the non-local state: P<sup>d</sup> = 0.22; 0.18–0.25). After accounting for detection probability, daily "survivorship" (remaining in either state, Ps) was initially high and similar for birds at all three sites (Downeast: P<sup>s</sup> = 0.99; 0.976– 0.994; Popham: P<sup>s</sup> = 0.99; 0.972–0.995; Rachael Carson: P<sup>s</sup> = 0.98; 0.96–0.96) but declined much more sharply at the Downeast



*EFM, estimated fat mass; PCDP, post-capture detection period (N in parentheses* = *# of birds confirmed detected after release); LDL, last detection local (%* = *proportion of individuals confirmed detected after release).* \* *Indicates one missing value;* \*\**indicates two missing values.*

site than the other two (final daily survivorship, Pds: Downeast: Pds = 0.45; 0.25–0.68; Popham: Pds = 0.72; 0.50–0.87; Rachael Carson: Pds = 0.78; 0.61–0.89).

The probability of a bird being detected at a foreign (nonlocal) tower differed significantly among sites (Chi = 46.7, df = 2, p < 0.001); no Downeast birds had their final detections at foreign towers, whereas more than 50% of the birds tagged at Rachel Carson (59%, ages pooled) and Popham Beach (61%, ages pooled) were detected beyond the local area (**Figure 4**). That Downeast birds were not likely to be detected at any other location to the north or south, and that their daily survivorship in the simple multi-state model declined much more quickly than the other sites, suggests that, similar to birds staging in the Bay of Fundy, they departed directly from the Pleasant Bay area on their way to the wintering grounds in eastern South America.

# Individual Sites

# Downeast (2013–2014)

The five towers deployed in the Pleasant Bay area provided an approximate local detection space of 30 × 50 km = 1,500 km<sup>2</sup> during the 2013–14 migration periods (**Figure 1**). Numerous towers beyond the area to the north (Bay of Fundy, NS) and to the south, in particular, the array of four towers, oriented NW-SE (perpendicular to the coastline) ∼40 km beyond the Downeast tagging area, provided opportunities to detect birds moving along the coast in either direction beyond the denoted local stopover area (**Figure 1**).

A total of 71 birds (32 AHY, 39 HY, years pooled) were tagged and released at the Downeast site in Pleasant Bay; all of these were later detected at towers within or beyond the tagging location. The post-capture detection period (PCDP) for Semipalmated Sandpipers tagged at this site (and meeting the criterion for inclusion in PCDP) ranged from 0.62 to 27.7 days for adults and from 0.50 to 37.2 days for juveniles (**Table 1**).

Most post-capture movements of birds captured and tagged in Pleasant Bay remained within the immediate area, with daily movements commonly occurring between exposed mudflats at the mouth of the rivers during low tide and offshore ledge and islands to roost during peak high tide (not shown). All of the 71 birds detected after release had final detections at towers within the local tagging area (**Figure 4**). The majority

of these final detections (39/71 = 55%) were made by towers along the southern boundary of Pleasant Bay, with most of these (26/39 = 67%) made by the tower on Nash Island (**Figures 1**, **4**).

#### Popham (2015–2016)

The location of the two towers (2015 POP only, 2015 and 2016 POP and SEGUIN, **Figure 1**) designated as "local" at the Popham Beach tagging site resulted in a local approximate detection space of 20 × 20 km = 400 km<sup>2</sup> for 2015 and 20 × 40 km = 800 km<sup>2</sup> for 2016 (**Figure 1**). Numerous towers were active along the coast to the northeast and to the southwest beyond the tagging area during June through November in both years (**Figure 1**).

A total of 59 birds (33 AHY, 26 HY) were tagged and released at the Popham Beach site but only 41 birds (16 AHY, 25 HY) were confirmed to have been subsequently detected at towers within or beyond the local tagging location (for unknown reasons, 18 tags confirmed to be active at the time of deployment were not later detected by any tower; these birds are excluded from tracking data analyses).

More than half (25/41 = 61%) of the birds tagged at the Popham Beach were last detected at the initial tagging location (**Table 1**, **Figure 4**). The post-capture detection period (PCDP) for Semipalmated Sandpipers tagged at this site (and meeting the criterion for inclusion in PCDP) ranged from 0.0 to 54.8 days for adults and from 0.02 to 30.3 days for juveniles (**Table 1**).

Three individuals tagged at Popham were later detected at sites to the NE, including one HY bird (Motus #20769, **Figures 5A,B**) that remained at the local site from 21 August through 4 September, after which it moved directly to the upper Bay of Fundy where it remained for at least another 12 days. Another bird (HY, Motus #15515, **Figures 5A,B**) tagged at Popham Beach was later detected to the north at Cutler, Maine, then flying by the western coast of Nova Scotia 1.6 h later. Another (HY; Motus #20771, **Figures 5A,B**) was detected at Grand Manan Island at the mouth of the Bay of Fundy.

FIGURE 4 | Plots showing spatial as well as temporal post capture movements of individual Semipalmated Sandpipers (ages, years pooled) tagged at each of the three tagging locations: "Downeast" (A,B), "Popham" (C,D), and "Rachel Carson" (E,F). Crosses represent locations of Motus towers present during each of the deployment periods. White dots represent towers where Semipalmated Sandpipers were detected. Colored lines represent tracks of different individual birds, and are consistent within a given site. Sample sizes and detailed information about each site's activities are provided in the text. Note that no birds tagged at the Downeast stopover site in Pleasant Bay were detected beyond the local tagging area, suggesting that these birds initiated trans-oceanic flights from this area. In contrast, more than 50% of the birds tagged at Popham Beach and Rachel Carson NWR were detected to the north (few) and to the south (many), with birds concentrating in coastal southern New England (Cape Cod and the islands, Rhode Island, Long Island Sound) before their final detection, suggesting that this site is also a major departure area for birds making trans-oceanic flights to wintering grounds in eastern South America. A few birds were detected as far south as Delaware Bay, the Carolinas and Virginia.

Most birds (8 AHY, 5 HY) whose last detections were at towers beyond the immediate area, however, were detected south of it, with final detections distributed around the southern New England coast (Southern Maine, Cape Cod and the islands of Martha's Vineyard and Nantucket, Block Island, RI, and Long Island Sound) with some individuals being last detected as far away as Cape May, NJ, Delaware, and Chesapeake Bays, Virginia, and North and South Carolinas (**Figure 4**).

Island. Birds tagged at Rachael Carson: Two birds, Motus #10956 (blue) and #10946 (green) made forays south to the Cape Cod area before returning to Rachel Carson NWR where they were both last detected; bird Motus#10941 (red) moved northward and was last detected at the mouth of the Bay of Fundy. These patterns

#### Rachel Carson NWR (2014–2015)

The estimated local detection area, based on the two local towers was 20 × 40 km = 800 km<sup>2</sup> , oriented northeast-southwest parallel to the coastline (**Figure 1**). Multiple towers further north and south were active in both years along the coast (**Figure 1**).

collectively reveal the connectivity among areas in the Bay of Fundy, the Gulf of Maine, and southern New England.

Due to weather constraints, the 2015 capture period was limited to the early half of the migration season for southern Maine, which biased captures to only adults. In 2014–15, 50 birds were captured and tagged: 44 (35 AHY, 9 HY) of these birds had confirmed detections after being released. The remaining six birds were excluded from tracking analyses.

Slightly more than half (59%) of the birds (ages pooled) tagged at Rachel Carson were never detected beyond the local tagging site (**Table 1**). Of the 18 individuals (16 AHY, 2 HY) that were last detected beyond the local area, 14 went as far south as Cape Cod (or further) with four of those individuals moving these distances within 24 h of capture (**Figure 4**). The rest of the birds remained at the local tagging site for between 7 and 12 days (**Figure 4**). Two individuals (Motus #10956, #10946, **Figures 5C,D**) traveled south to Cape Cod for a short period (<48 h) and returned to the local site, and two others traveled < 30 km to the SW. One individual (Motus #10941, **Figures 5C,D**) traveled ∼100 km to the NE.

#### DISCUSSION

Stopover behavior of individual Semipalmated Sandpipers varied not only in the amount of time they spent at stopover sites along the coast of Maine, but also in subsequent movements within and beyond the region. This variation did not appear to be agerelated, but may be in response to ecological, topographical, and environmental factors that varied among the sites, providing support for some of our predictions but not all. Counter to our predictions, we failed to find any support for age-related differences in stopover behavior: during their first migration, young birds on stopover in the Gulf of Maine do not appear to remain longer at stopover sites than adults even though they may be less efficient at foraging and detecting predators (Cresswell, 1994; Fernandez and Lank, 2006; Stillman et al., 2007; van den Hout et al., 2017).

We found partial support for our prediction that the amount of energy reserves birds had at the time of capture would influence the amount of time remained after tagging, but only for birds captured at the Downeast site. And, in contrast to birds tagged further south, few Semipalmated Sandpipers tagged in Pleasant Bay at the Downeast site were detected making forays out of the local tagging area before departing from it altogether. Finally, unlike birds tagged at the other two locations further south, none of the Downeast birds had final detections beyond the initial tagging location.

Site-based difference in final detection patterns suggests that birds arriving on stopover in the Downeast area have a different migratory strategy or route compared to most birds stopping over at sites to the south, which could indicate some level of population or sex class segregation occurring along the Maine coast. Our data provide little support, however, for sex-related segregation. While birds at the Downeast site had smaller bills compared to birds on stopover further south, potentially representing a sexbased bias with males moving to the north and females arriving further south along the Maine coast, this is counterintuitive. Within breeding populations, females of this species have longer bills than males (Gratto-Trevor et al., 2012a; Hicklin and Chardine, 2012;), and birds with longer bills (i.e., females) may be better at foraging in deeper mud substrates (Harrington, 1982). If the availability of optimal feeding substrates (mudflats), which is greater in the northern Gulf of Maine compared to the southern areas, influences sex-biased movements, we would have expected to see the opposite pattern, with larger billed birds (females) more robustly represented at the Downeast site compared to Popham Beach and Rachel Carson.

We cannot rule out the possibility, however, that birds arriving at the Downeast site are from the same population as those staging in the Bay of Fundy, and that birds from other populations stopover further south. Flattened wing and bill (culmen) lengths for both age classes of birds captured at the Downeast site in this study are consistent with those observed in Semipalmated Sandpipers on stopover in the Bay of Fundy during the same time period (wing: ∼97 mm, bill: ∼19 mm; see **Figure 2**, Anderson et al., 2019). As both of these characters are significantly different from birds captured at Popham Beach and Rachel Carson, the data suggest that more than one breeding population is moving through the Gulf of Maine region during fall migration and these populations may show some spatial segregation. However, it is unclear as to which populations birds arriving further south represent. The fact that birds captured in southern Maine had larger bill size than those on stopover at the Downeast site suggests a bias toward central rather than eastern populations arriving in northern Maine, but this is counter to previous populationlevel morphometric analyses and recent tracking data: bill size increases across an east to west cline, birds staging in the Bay of Fundy are predominantly from eastern breeding populations (Gratto-Trevor et al., 2012a; Hicklin and Chardine, 2012; Miller et al., 2013), and central and western populations do not move as far east as the Gulf of Maine/Bay of Fundy region during fall migration (Brown et al., 2017). Finally, we cannot rule out potential site-based differences in measurement error: in spite of attempts to standardize measurements across the three studies, only one of the researchers took all of the morphological measurements at the first of the three sites (Downeast) while several others collected the data at the other two sites. It would be useful, however, to confirm, in future studies, which Semipalmated Sandpiper populations move through the Gulf of Maine in order to better link events occurring on the breeding grounds with changes in bird numbers at different stopover locations in the region.

Although the underlying mechanisms influencing variation in stopover behavior are unknown, our results clearly show sitebased differences in Semipalmated Sandpiper stopover behavior in Maine. The fact that few birds tagged in at the Downeast site (Pleasant Bay) made forays out of the area before resuming migration, and the fact that none were detected anywhere else along the Atlantic coastline strongly suggest that the Pleasant Bay area, like the Bay of Fundy, provides an opportunity to rest and refuel before initiating an extensive trans-oceanic flight to South America. The fact that the timing of departure from the site after tagging was significantly influenced by the amount of fat reserves at the time of capture further reinforces this idea.

Unlike the Downeast site, departure decisions, which were not influenced by energetic condition, at the two southern sites were likely influenced by factors affecting the birds' ability to efficiently rest and refuel. Both Popham Beach State Park and Rachel Carson Wildlife Refuge areas have much greater human activity compared to the Downeast region, and such activity has been shown to directly and negatively influence the amount of time shorebirds spend feeding and resting (Burger, 1993; Thomas et al., 2003; Schlacher et al., 2013; Mayo et al., 2015). In particular, Popham Beach has extremely high public use during peak shorebird migration period (L. Tudor, unpublished data). While the tidal pools at Popham Beach offer areas for shorebirds to feed, these pools are also heavily visited by the public for recreation during both high and low tides. Shorebirds feeding on mudflats located outside the park are exposed to additional human and pet-related activities associated with the surrounding residential development areas (such as free-roaming dogs observed chasing shorebirds on the mudflats, M. Fahay, pers. obs., L. Tudor, pers. obs.). Roosting sites away from human activity are limited at Popham Beach. Similarly, although birds at Rachel Carson NWR were captured while feeding on tidal mudflats, which were within the marsh away from the beach and public hiking areas, these areas are much smaller in area than those in Pleasant Bay and are entirely unavailable for roosting during high tide. Undisturbed roosting areas are limited at Rachel Carson NWR and birds may actively seek them out: many Semipalmated Sandpipers were observed roosting during high tide in the more remote beach areas fenced off earlier in the season to protect nesting shorebirds from public disturbance (K. O'Brien, pers. obs.).

We failed to find age-related patterns in movement behavior at all three stopover sites in spite of the fact that earlier studies found that young birds may not be as efficient in foraging or predator detection as adults (Stillman et al., 2007; van den Hout et al., 2017). Although predation pressure can influence migration strategies in shorebirds (Lank et al., 2003; Sprague et al., 2008), we were not able to systematically collect data on predators. We did, however, note aerial predators when detected, and frequently observed Peregrine falcons (Falco peregrinus) and merlins (F. columbarius) actively attacking shorebird flocks during high and low tides at all three sites.

The Pleasant Bay area (Downeast site) is located along the northeast-southwest oriented coastline contiguous with the Bay of Fundy (**Figure 1**). Not only is it strategically situated as a jumping off site to fly non-stop over the North Atlantic, it offers expansive tidal mudflats for feeding and numerous offshore rocky ledge outcroppings for roosting, critical resources for shorebirds on migration. These offshore roosting sites are not only safer, away from terrestrial predators and human activity, they are only a few kilometers from the feeding areas. Similar to those in the Bay of Fundy, these tidal mudflats contain greater densities of macroinvertebrates than sand flats and sandy beaches that dominate southern Maine coast (Napolitano and Ackman, 1990; Napolitano et al., 1992). In the northern Gulf of Maine, such mudflats offer higher densities of the nutritionally rich amphipod, Corophium volutator, considered a high quality food for shorebirds, making up 86% of the diet of Semipalmated Sandpipers in the Bay of Fundy and influencing the spatial distribution of the birds feeding there (Napolitano and Ackman, 1990; Napolitano et al., 1992; Hamilton et al., 2003). Corophium volutator is found in greater abundance in the northern Gulf of Maine than in the south (Commito, 1982; Larsen and Doggett, 1991), making stopover sites there the ideal combination of lower human activity and greater abundance of high quality food in close proximity to relatively isolated roosting sites away from terrestrial predators.

In contrast, birds arriving at sites further south in the Gulf of Maine may experience not only greater human activity, but also fewer mudflats with lower quality and less abundant food, as well as fewer safe roost sites nearby. Thus, these birds may opt to do a "hop"/"skip" strategy (c.f. Warnock, 2010) to move either north (to the Downeast area or Bay of Fundy) or south along the coastline to reach a more strategic and more suitable site to prepare for and initiate a transoceanic flight. Indeed, we detected birds making such northward movements, but most birds moving beyond their initial tagging area moved southward, with many birds concentrating in and apparently departing from the southern New England coast (e.g., Cape Cod, Rhode Island, Long Island Sound, **Figures 4**, **5**). The fact that at least two individuals returned to the Cape Cod area after making forays further south suggests that Cape Cod and the islands, which, like the Downeast/Bay of Fundy region, serve as a strategic "jumping off " point for birds initiating a trans-oceanic flight to the east coast of South America. It is not known if the birds detected in Virginia or the Carolinas (**Figures 4**, **5**) arrived there by remaining along the coastline or if they were unable to continue an overwater flight initiated further north due to storms or other reasons.

Ocean acidity, which has been shown to severely affect calcium metabolism and thus, productivity of shelled invertebrates (Jacobson et al., 2009), is increasing. For Semipalmated Sandpipers stopping over in the northern Gulf of Maine/Bay of Fundy region, a significant decline in C. volutator, with its high concentration of energy-rich long chain fatty acids, could not only affect the birds' preparation for successful transoceanic flight (Maillet and Weber, 2006), but could also have long-term consequences on individual health through reduced immune function and increased tissue damage (Buehler et al., 2010; Eikenaar et al., 2019). Because more than 75% of the world's population of Semipalmated Sandpipers are believed to collectively stage in the Downeast and Bay of Fundy regions in fall, declines in available habitats and energy-rich food sources in this area could seriously impact the species stability altogether (Maillet and Weber, 2006).

Not only are areas along the Atlantic coastline experiencing increased pressure from development, the current rate of sea level rise in the Gulf of Maine is apparently accelerating, with some models predicting more than a half meter rise in sea level by 2100 (Jacobson et al., 2009). Such a rise will significantly and permanently reduce currently available feeding and roosting habitats, perhaps more rapidly than new ones can develop. This realization should prompt resource managers to identify potential feeding and roosting areas now in order to sustain shorebird populations in the future.

In summary, the use of automated VHF radiotelemetry to track individual movements of Semipalmated Sandpipers revealed important site-based, and not age-based, differences in migratory behavior and movements within and beyond the Gulf of Maine region. Impending sea level rise, ocean acidification, and increased human activity will have profound impacts on future shorebird populations worldwide (Galbraith et al., 2005; Iwamura et al., 2013; Fraser et al., 2018). Future research and resource management should be directed toward factors that affect individual behavior, including local as well as regional scale topography, habitat characteristics, human activity, and the degree of connectivity among different sites at the regional and continental scale.

# DATA AVAILABILITY

The telemetry data for this study can be found under Projects #8 (Downeast), #25 (Rachel Carson), and #110 (Popham Beach) on the Motus Wildlife Tracking System Motus (www.motus.org).

# ETHICS STATEMENT

All activities related to bird capture and handling were reviewed and approved by UMaine IACUC (#A2013-01-02 to RH) and were performed under federal and state permits to U.S. Fish and Wildlife Service (to KO'B) and Maine Department of Inland Fisheries and Wildlife (to LT).

## AUTHOR CONTRIBUTIONS

RH, LT, and KO'B developed the project and secured funding for it. RH, LT, KO'B, and GM conducted and supervised all aspects of the fieldwork, including tower assembly and deployment, and capturing and tagging birds. Data analyses were conducted by PT, AB, and RH were reviewed and approved by GM, KO'B, and LT. All authors contributed to the writing and editing of the manuscript.

#### FUNDING

Funding was awarded, collaboratively or as individuals, by the Maine Outdoor Heritage Fund (to RH and LT), US Fish & Wildlife Service Region 5 Division of Natural Resources - National Wildlife Refuge System (to KO'B), the federally funded State Wildlife Grant Program (to LT), and the Eastern Maine Conservation Initiative (to RH). This project was supported by the USDA National Institute of Food and Agriculture, Hatch Project #ME0-21609 through the Maine Agricultural and Forestry Experimental Station (to RH); Maine Agricultural and Forest Experiment Station Publication Number 3684.

#### REFERENCES


#### ACKNOWLEDGMENTS

We thank the numerous personnel and volunteers who helped capture and tag birds, set up and operate towers within and beyond our three focal sites, and/or provided valuable assistance with data handling: University of Maine (Sean Rune, Wesley Wright), Maine Department of Inland Fisheries and Wildlife (Brad Allen, Jason Czapiga, MaryEllen Wickett, Lisa Bates, Matthew O'Neal), Biodiversity Research Institute (Patrick Keenan, Kevin Reagan), Maine Bird Observatory (Donna Kausen), US Fish and Wildlife Service National Wildlife Refuges (Maine Coastal Islands National Wildlife Refuge, Rachel Carson National Wildlife Refuge), Popham Beach State Park managers Meagan Hennessey and Brian Murray, Friends of Seguin Island Lighthouse, and Seguin Island Ferry Service (Capt. Ethan DeBery), and Motus Wildlife Tracking Network (John Brzustowski, Stu MacKenzie). We are also grateful for the support of private landowners (the Mudge and Marshall families) who allowed towers to be deployed on their properties at the Downeast location. Michael Fahey and Ken Janes provided numerous observations of sandpipers, including resightings of tagged birds. All activities related to bird capture and handling were reviewed and approved by UMaine IACUC (#A2013-01-02 to RH) and were performed under federal and state permits to US Fish and Wildlife Service and Maine Department of Inland Fisheries and Wildlife (KO'B and LT).


loss from sea-level rise for shorebird populations. Proc. R. Soc. B 280:20130325. doi: 10.1098/rspb.2013.0325


evidence from the Arctic. Waterbirds 35, 106–118. doi: 10.1675/063. 035.0111


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

As per U.S. Fish and Wildlife policy (http://www.fws.gov/policy/117fw1.html), "The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the U.S. Fish and Wildlife Service."

Copyright © 2019 Holberton, Taylor, Tudor, O'Brien, Mittelhauser and Breit. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Why Are Whimbrels Not Advancing Their Arrival Dates Into Iceland? Exploring Seasonal and Sex-Specific Variation in Consistency of Individual Timing During the Annual Cycle

#### Camilo Carneiro1,2 \*, Tómas G. Gunnarsson<sup>2</sup> and José A. Alves 1,2

*<sup>1</sup> Department Biology and CESAM, University of Aveiro, Aveiro, Portugal, <sup>2</sup> South Iceland Research Centre, University of Iceland, Laugarvatn, Iceland*

#### Edited by:

*Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway*

#### Reviewed by:

*Daniel Robert Ruthrauff, U.S. Geological Survey, Alaska, United States Erica Nol, Trent University, Canada*

> \*Correspondence: *Camilo Carneiro camilofcarneiro@gmail.com*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *26 April 2019* Accepted: *14 June 2019* Published: *02 July 2019*

#### Citation:

*Carneiro C, Gunnarsson TG and Alves JA (2019) Why Are Whimbrels Not Advancing Their Arrival Dates Into Iceland? Exploring Seasonal and Sex-Specific Variation in Consistency of Individual Timing During the Annual Cycle. Front. Ecol. Evol. 7:248. doi: 10.3389/fevo.2019.00248* The timing of annual events is key for organisms that exploit seasonal resources, as deviations from optimal timing might result in considerable fitness costs. Under strong time selection, individuals likely have fewer suitable strategies available than when selection is more relaxed, hence both consistency and flexibility might be advantageous depending on the life history or annual cycle stage. For migrants using both the arctic and the tropics during their annual cycle, the faster warming at higher latitudes than elsewhere in the range may lead to mismatches with local environmental conditions. Additionally, while individuals might already be limited in responding to changes at each stage, the potential degree of a given response will likely also be limited by responses at previous stages of the annual cycle. Contrary to other migratory waders breeding in Iceland, Icelandic whimbrels *Numenius phaeopus islandicus* have not changed arrival dates during the past 30 years, suggesting high individual consistency in spring arrival timing and a potential limitation in responding to a changing environment. After repeatedly tracking 12 individual Icelandic whimbrels at least twice throughout their annual cycle between 2012 and 2018, we investigated individual consistency of spring arrival date and other annual stages and migration strategy, and explored differences between sexes and seasons. Individuals were more consistent on timing of spring than autumn migration, and the most consistent stage was departure from the wintering sites. Timing of laying was the stage that varied the most, and no overall significant difference between sexes was observed, except on spring stopover duration. While lower consistency in laying dates might allow individuals to track the advancement of spring, consistency at departure from the wintering sites, stopover duration, and arrival into Iceland might limit the degree of advancement. Transgenerational changes in the migratory behavior of other wader species allows population level responses to a changing phenology, but seems unlikely for Icelandic whimbrels, given the stable dates of spring arrival in this population. Under continuing advancement of spring onset, it is thus important to acquire information on the timing of spring arrival of recruits and on the ontogeny of migration to understand how migratory schedules are defined and might influence responses of long-distance migrants to environmental change.

Keywords: flexibility, consistency, annual cycle, whimbrel, individual, environmental change, timing, phenology

### INTRODUCTION

The annual cycle of many animals comprises migratory periods during which individuals travel to exploit seasonally available resources (Newton, 2007). Timing of specific events, such as breeding, is therefore fundamental (Alerstam and Lindström, 1990), but selection for optimal timing of events might not be equally strong between different stages of the annual cycle. Due to the influence that timing of breeding can have on breeding success and thus individual fitness (Perrins, 1970; Drent, 2006), selection on timing during the preceding spring migration is expected to be stronger than in autumn (McNamara et al., 1998). Under strong time selection, the available strategies (e.g., schedules) for individuals are likely to be fewer than when selection is more relaxed and individuals can perform a given task (e.g., migration) over a wider time window (Madsen, 2001; Warnock et al., 2004). Therefore, both consistency and flexibility can be advantageous depending on the life history of a given species (Vardanis et al., 2016).

For migrants breeding in arctic and sub-arctic environments, selection on timing of breeding is expected to be strong, given the relatively short time window with favorable environmental conditions for reproduction. However, climate is changing at a faster pace than some species are able to adjust to (Both and Visser, 2001), and the situation can become more complicated in the arctic, where temperatures are increasing more rapidly than at lower latitudes (Høye et al., 2007; Serreze et al., 2009; Cohen et al., 2014). Whereas, individual consistency can differ among distinct annual events (e.g., Conklin et al., 2013, Verhoeven et al., 2019), the performance of individuals at a given stage may also depend on the conditions experienced in previous stages (Harrison et al., 2011; O'Connor et al., 2014). Hence, consistent behavior at one stage might thus limit the degree of change in the subsequent one. Detailed data throughout the annual cycle (e.g., Senner et al., 2014) is therefore required to understand how individuals and populations might be limited in their capacity to respond to a changing environment (Marra et al., 2015).

Previous research has indicated that most species of waders (Charadrii) breeding in Iceland have been advancing their spring arrival dates (e.g., golden plover Pluvialis apricaria, common snipe Gallinago gallinago, black-tailed godwit Limosa limosa), accompanying the trend of temperature change, but Icelandic whimbrels Numenius phaeopus islandicus were an exception (Gunnarsson and Tómasson, 2011, Gill et al., 2014). We update this information by adding 9 years of data of first arrival for birds recorded in South Iceland (see Gunnarsson and Tómasson, 2011 for details) and confirm that arrival dates show no significant trends since 1988 for this population (**Figure 1**). Whimbrels are long-lived birds (typical lifespan: 11 years; Robinson, 2018) that are long-distance migrants. The Icelandic subspecies breeds predominantly in Iceland, with an estimated population of 256,000 pairs (Skarphéðinsson et al., 2016), and winters in West Africa (Gunnarsson and Guðmundsson, 2016; Carneiro et al., 2019). In spring, their departure from the wintering sites occurs on the second fortnight of April and arrival into Iceland from late April to early May, with individuals performing one of two migratory strategies: a non-stop flight or two flights divided by a relatively short stopover; in autumn, migratory movements occur from late June to late August and only the direct flight strategy has been recorded in this season (Alves et al., 2016, Carneiro et al., 2019). Icelandic whimbrels are site faithful to the breeding territory, monogamous, and most reproduce with the same partner from year to year (BWPi, 2006), with males arriving earlier and departing later from breeding sites than females (Carneiro et al., 2019).

The low variation of Icelandic whimbrels arrival dates into Iceland and the lack of a population advancement (**Figure 1**) in response to increasing temperatures may suggest consistency of individual spring arrival dates. Individual consistency in this population is currently unknown, but, at the same time, important to unravel in order to understand potential limitations to population-level responses. Hence, we quantified (1) individual timing consistency in arrival dates to Iceland, and at five other annual stages (departure and arrival during autumn migration, departure of spring migration, stopover duration in spring, and laying date) in order to investigate possible

FIGURE 1 | Arrival dates of whimbrels into South Iceland in the 31-year period from 1988 to 2018, updated from Gunnarsson and Tómasson (2011), showing no significant trend (day of the year = −0.027 \* year + 178.11, *n* = 31, *R* 2 = 0.004, *p* = 0.734).

constraints throughout the year; (2) individual consistency in migratory strategy; and tested for (3) seasonal and sex differences in timing consistency. Along with a lack of population variation in spring arrival date, and a higher pressure for breeding timing than for arrival into the wintering sites, we expected higher consistency during spring than autumn migration, but lower consistency in laying dates than during other stages, as this also depends on the schedule of the partner and on their return. For spring migration, consistency might be higher at departure because conditions are likely to be more stable at the wintering sites than at breeding areas, where weather conditions are more variable (as stochastic weather events can reduce food availability and lead to mortality; Vepsäläinen, 1969, Marcström et al., 1979) and the occurrence of a stopover during migration might also influence arrival. However, during autumn, consistency should not vary from departure to arrival due to the direct strategy. Because males arrive earlier than females to defend territories and also attend broods longer (BWPi, 2006), we anticipated that males should show higher consistency on spring migration and less so at autumn departure than females.

#### METHODS

Fieldwork was carried out on whimbrels at breeding grounds in the southern lowlands of Iceland (63.8◦N; 20.2◦W), between 2012 and 2018. Nests were searched for and upon finding, the incubation stage estimated through egg floatation (Liebezeit et al., 2007), and the laying date was back calculated from stage of incubation. Nests were monitored until hatching and for those that hatched, laying date was back calculated from the hatching date considering an incubation period of 25 days after the last egg laid (mean ± se: 24.8 ± 0.2 days, n = 24 nests found when laying, and hatch recorded). Because whimbrels were individually marked (see below), we were able to identify replacement clutches which were not included in the analyses.

Two hundred and twenty-six adult birds were caught on the nest, using a nest trap (Moudry TR60; www.moudry.cz). Birds were individually marked with a unique metal ring, issued by the Icelandic ringing scheme, and a combination of color rings. Geolocators were fitted on a leg flag to a subgroup of 86 individuals (number of geolocators deployed per year-2012: 10; 2013: 3; 2014: 10; 2015: 30; 2016: 40; 2017: 40). The device was replaced whenever possible each breeding season. Sixtytwo devices were retrieved one or more years later (number of geolocators retrieved per year-2013: 5; 2014: 4; 2015: 5; 2016: 14; 2017: 20; 2018: 14). For tags retrieved two or more years later, data on two autumn migrations were recorded. We used the Intigeo-W65A9RJ model from 2012 to 2014 and Intigeo-C65 in the following years (Migrate Technology Ltd.). One device stopped logging in mid-winter, another shortly after departure from Iceland, and a third one was damaged and contained no data. Sixty five individuals were sexed using biometrics following Katrínardóttir et al. (2013), 22 molecularly (as in Katrínardóttir et al., 2013), eight through behavioral observation (copulating position, assuming males on top) and two remained undetermined. Geolocator data analysis and determination of individual departure and arrival timings were performed using light, and temperature, conductivity and wet contacts as described in Carneiro et al. (2019). Due to the accuracy of geolocators (Phillips et al., 2004), we considered arrival and departures to/from the general area. For example, spring arrival was arrival into Iceland instead of arrival into the breeding territory (although some individuals have been observed on the breeding territory on the day of arrival into Iceland). We consider stopover any stop during travel between breeding and wintering locations, irrespective of length of stay, site quality or previous or future flight distance and duration (i.e., we do not discriminate from staging; Warnock, 2010). Although stops of few hours may be undetected with geolocator data, stopovers of Icelandic whimbrels are usually of several days (Carneiro et al., 2019), with the minimum stopover duration recorded during this study being 6 days.

Icelandic whimbrels show two migratory strategies in spring: a direct non-stop flight or two flights with a stopover in between (henceforward: "direct" and "stopover"; Carneiro et al., 2019). However, and adding to previous information, during this study one individual was recorded undertaking a stopover during autumn migration. To understand individual consistency in migratory strategy we calculated the percentage of individuals that changed strategy during the tracking period and the direction of change (i.e., from direct to stopover, from stopover to direct, or both).

Repeatability (R) was estimated in a mixed effects model framework, using 1,000 bootstrap iterations to estimate the confidence intervals, with R package rptR (Stoffel et al., 2017). Given that R takes into account both within- and between-individual variances, it does not translate into absolute consistency (see Conklin et al., 2013), and therefore we also calculated the mean individual range (difference between the latest and earliest record for each individual for each stage, in days) and the absolute interannual difference (absolute difference between consecutive years for each individual within stage, in days) in order to better evaluate individual consistency. To test for differences in consistency between stages and sexes, we fit a generalized linear model with absolute interannual difference as the dependent variable and stage, sex, and their interaction as explanatory variables (with family = Poisson due to the positively skewed dependent variable). In this analysis we did not account for the dependency of member of the same pair, because we could have only used those individuals that nested with the same partner in consecutive years, which would result in a reduced sample size. Data were analyzed in R Core Team (2018) and results are shown as mean ± se.

#### RESULTS

We recorded individual level data spanning 2–7 years, with a median of 3 years for autumn migration (n = 16 individuals), and 2 years for spring migration (n = 12 individuals), and for laying date (n = 70 individuals). Hence, autumn migration, spring migration, and laying date were recorded during a median of 27, 18, and 18%, and up to 54, 46 and 64%, of whimbrels typical

TABLE 1 | Direction and proportion of individuals that changed migratory strategy on each season (Autumn and Spring), from a stopover to direct, direct to stopover or on both directions; *n* = number of individuals.


lifespan (Robinson, 2018), respectively. During the present study, one individual was recorded making a stopover during autumn for the first time. In that season, one individual (out of 16) switched strategy between years, from a stopover to a direct one (**Table 1**). A change in the opposite direction, from direct to stopover in the following year, was observed in a higher proportion during spring migration (three out of 12 individuals; **Table 1**). No individual was observed changing strategy in both directions (**Table 1**).

Along the annual cycle, Icelandic whimbrels showed the highest consistency of timing at spring departure (**Figures 2**, **3** and **Table 2**). Despite low R values for spring arrival, individuals showed relatively smaller mean individual ranges (**Table 2**) and small absolute interannual differences (**Figure 3** and **Table 3**) than at autumn migration stages. Such low R values result from a relatively low variance among individuals compared to the variance within (**Figure 2**), which arises from the change in migratory strategies. Spring stopover duration was relatively consistent, with low mean individual range and low absolute interannual differences (**Table 1** and **Figure 3**). During autumn migration, both departure, and arrival timings showed similar consistency (**Figures 2**, **3** and **Table 2**), as in this season individuals seldom change strategy and almost always perform a direct flight (**Table 1**). However, contrary to autumn departure, autumn arrival absolute interannual difference was not statistically different from laying dates (**Table 3**). Laying date was the least consistent stage of the annual cycle (**Figures 2**, **3** and **Table 2**). When considering the absolute interannual difference, we found no overall difference between sexes (**Table 3**), but males showed a lower absolute interannual difference of stopover duration (**Figure 3** and **Table 2**), and a mean individual range in autumn ca. 2.5 days longer than females (**Table 2**).

#### DISCUSSION

By tracking individuals over multiple years and over a considerable part of their lifespan, it is possible to quantify relevant levels of individual consistency or flexibility regarding the phenology of important events during the annual cycle. Both consistency and flexibility can be advantageous depending on

the life history of each species (Vardanis et al., 2016), and can be essential to understand the capacity and rate of population responses to changing environments (Gill et al., 2014). The arrival dates of Icelandic whimbrels in spring have been stable over the past 30 years (**Figure 1**; Gunnarsson and Tómasson, 2011), despite a spring advancement of temperatures that drives resources for waders locally (Alves et al., 2019). At the individual level, we show that Icelandic whimbrels were more consistent in timing of spring than autumn migration, and most consistent at departure from the wintering sites. Timing of laying was the stage of the annual cycle that varied the most and no overall significant difference between sexes was observed, except for males lower absolute interannual difference of stopover duration.

number above each boxplot shows the sample size.

During autumn migration, a lower consistency in timing was observed compared to spring (**Tables 2**, **3** and **Figures 2**, **3**) and the values at arrival mirror the ones at departure because in nearly all occasions individuals flew directly from Iceland to the wintering sites (**Table 1**; Carneiro et al., 2019). On the other side of the Atlantic, Hudsonian whimbrels showed the same general pattern, as inter-individual variation was greater at autumn departure and arrival dates, than at spring departure and arrival dates (Johnson et al., 2016). Given the expected stable conditions in the wintering area, one could anticipate repeatability to be high during autumn migration (Nussey et al., 2005). In fact, despite a lower consistency in relation to spring, Icelandic whimbrels are still reasonably consistent in autumn migration timings, with a median individual departure range of 9 days and median absolute interannual difference of 5 days, which are in line, or even lower than those observed on other long distance migratory birds (median range of departure of ca. 15 days in individual continental black-tailed godwits L. l. limosa, Verhoeven et al., 2019; median absolute interannual difference of ca. 4, 5, and 6 days in bar-tailed godwits L. lapponica, TABLE 2 | Repeatability (*R*) with 95% confidence intervals (CI) of timing of annual events and spring stopover duration, for all individuals (a) and by sex (b and c).


*Sample size is the number of individuals (n). Also shown is the mean individual range (*± *se), in days. "Spring arrival (stopover)" represent the arrivals excluding direct flights.*

marbled godwits L. fedoa and red-backed shrikes Lanius collurio, respectively, Conklin et al., 2013, Pedersen et al., 2018, Ruthrauff et al., 2019; mean absolute interannual difference of 12.9 in great reed warblers Acrocephalus arundinaceus, Hasselquist et al., 2017). The observed variation in departure dates from Iceland is likely explained by the prior variation in laying dates and breeding success, since successful breeders tend to depart later (pers. obs.). The mean individual range recorded for males (ca. 2.5 days larger than females; **Table 2**) is likely due to their longer attendance of broods.

The relative low repeatability of laying date (**Table 2** and **Figures 2**, **3**) may be partially explained by the variation in arrival dates into the breeding sites and partner arrival timing and return. Nevertheless, under a scenario of arctic amplification and spring advancement (Høye et al., 2007; Serreze et al., 2009; Gill et al., 2014; Alves et al., 2019), flexibility on laying dates might be beneficial, allowing individuals to track the local conditions and breed successfully. But the potential advancement of laying might be constrained by previous annual events. While spring arrival dates showed some variability (mean individual range at arrival: 5.3 ± 1.4 days), it was mostly due to variation in migratory strategy (**Table 1**), with the occurrence of a stopover

TABLE 3 | Results of the general linear model testing the effects of stage and sex on absolute interannual differences; estimates for stages are in relation to laying and for sex it is of male in relation to female.


*P-values* < *0.05 are highlighted in bold.*

augmenting variation on arrival date after a consistent departure from the wintering sites (mean individual range at departure: 3.6 ± 0.7 days). When considering only the spring arrival of individuals that had a stopover (which is the common strategy), we find higher consistency of arrival dates (mean individual range at arrival: 3.0 ± 0.7 days; **Table 2**). Furthermore, stopover duration also shows considerable consistency. Hence, individuals tend to be consistent throughout spring migration, starting at departure, which might limit how much laying dates can vary after arrival. If individuals would advance the departure date from the wintering sites, the capacity of tracking the advancement of resource availability in the breeding sites would be higher. However, climatic conditions in the wintering areas seem to be more stable than at the breeding sites (Høye et al., 2007; Serreze et al., 2009; Cohen et al., 2014), and thus unlikely to trigger individual responses at a sufficient rate that allows individuals to track the changes at the breeding sites. In fact, over the last 30 years the population of Icelandic whimbrels showed a stable spring arrival date (**Figure 1**), despite the increasing temperatures in the breeding grounds (Gunnarsson and Tómasson, 2011; Alves et al., 2019).

Populations can change migration timing through transgenerational variation in phenology (Gill et al., 2014). Such a mechanism was identified in Icelandic black-tailed godwits (L. l. islandica), that have advanced their arrival date into Iceland and tracked the advancement of spring onset (Gill et al., 2014). Similar to whimbrels, individual godwits are consistent in their timing of spring arrival, but recruits tend to migrate earlier and drive the population timing (Gill et al., 2014). The lack of an advancement in whimbrels population arrival dates into Iceland, together with individual consistency, suggests none, or little, transgenerational changes of migration timing. While black-tailed godwits spend the winter in the temperate region, at a maximum of ca. 3,000 km from Iceland (Alves et al., 2012), Icelandic whimbrels migrate longer distances to winter in the tropical or subtropical region, ca. 6,000 km from the breeding sites (Gunnarsson and Guðmundsson, 2016; Carneiro et al., 2019). By wintering closer to Iceland (Alves et al., 2013), godwits might adjust arrival dates to the local environment (Alves et al., 2012), lay as soon as conditions are adequate and produce young early in the season, that are more likely to recruit and ultimately drive population changes (Alves et al., 2019). Whimbrels, on the other hand, by wintering further and likely with no cue of the environmental conditions in Iceland, might have narrowed the variation in timing of departure to a later date than godwits and other waders breeding in Iceland, which in turn reduces the variation at spring arrival, the time between arrival and laying and, consequently, laying dates, limiting the possibilities for transgenerational changes. The arrival date into the breeding sites can vary considerably with spring migratory strategy, with individuals arriving earlier after a direct strategy consequently having a longer period in Iceland before breeding and be more likely to track the advancement of spring onset. However, a direct strategy in spring is uncommon, and although our data on its variation is limited, no individual changed from a stopover to a direct strategy, suggesting that individuals might not track the advancement of spring onset through a change in migratory strategy.

Whimbrels show no advancement of arrival dates into Iceland while spring onset is advancing (**Figure 1**; Gunnarsson and Tómasson, 2011; Alves et al., 2019), but there is no indication of a population decline (Skarphéðinsson et al., 2016). In Iceland, whimbrels are one of the latest waders to arrive and have a relatively short breeding period when compared to other species breeding in the same area (Gunnarsson, unpublished data). The period length of available minimum resources for breeding is unknown, but if it is wider than that required for successful breeding by whimbrels, it could be that they have been reproducing within that window even while the environmental conditions are advancing. However, under continuous advancement of spring conditions, consistency might prevent individuals from responding to changes when breeding and resource time windows mismatch. At such a hypothetical point one might observe the recruits performing with a different phenology and allow the population to respond to environmental changes, similarly to Icelandic black-tailed godwits. While monitoring population size, it is thus important to acquire knowledge on resource dynamics at the breeding areas (e.g., does the peak abundance of a given resource influence chick growth and survival?), and on the ontogeny of migration and associated timing (e.g., does hatching date affect the migration timing of recruits?). Links between resources and ontogeny will allow understanding how migratory schedules are defined and forecast population-level responses of long-distance migrants to environmental change.

#### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of Icelandic Institute of Natural History. The protocol was approved by the Icelandic Institute of Natural History.

#### AUTHOR CONTRIBUTIONS

All authors designed and carried the study. CC performed the analysis and lead the writing with substantial discussion and inputs from TG and JA. All authors read and approved the final version.

#### FUNDING

This work was funded by RANNIS (Grants: 130412-052 and 152470-052), the University of Iceland Research Fund,

### REFERENCES


and by FCT/MCTES to CESAM (UID/AMB/50017/2019), and individual grants (PD/BD/113534/2015 and SFRH/BPD/91527/2012), through National Funds, and ProPolar.

#### ACKNOWLEDGMENTS

We are thankful for the logistic support of the Icelandic Soil Conservation Service, particularly to Anne Bau and Jóna Maria; Snæbjörn Pálsson for support with genetic sex determination of birds; Verónica Méndez, Borgný Katrínardóttir, and Edna Correia for fieldwork support, our group members in Iceland and Portugal for fruitful discussions, Kristinn Jónsson for kindly allowing us to work on his land and Gunnar Tómasson for sharing arrival dates for whimbrels. Last, we are thankful to Daniel Ruthrauff, Brett Sandercock, and a reviewer for their suggestions, which helped to improve our manuscript.


Newton, I. (2007). The Migration Ecology of Birds. London: Academic Press.


models. Methods Ecol. Evol. 8, 1639–1644. doi: 10.1111/2041-210X. 12797


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Carneiro, Gunnarsson and Alves. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Effects of Spring Migration Distance on Tree Swallow Reproductive Success Within and Among Flyways

#### Edited by:

Nathan R. Senner, University of South Carolina, United States

#### Reviewed by:

Emily Anne McKinnon, University of Manitoba, Canada Rose J. Swift, Northern Prairie Wildlife Research Center (United States Geological Survey), United States Michael T. Hallworth, Northeast Climate Adaptation Science Center, United States

\*Correspondence:

Elizabeth A. Gow egow@uoguelph.ca

†Deceased

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 25 April 2019 Accepted: 23 September 2019 Published: 11 October 2019

#### Citation:

Gow EA, Knight SM, Bradley DW, Clark RG, Winkler DW, Bélisle M, Berzins LL, Blake T, Bridge ES, Burke L, Dawson RD, Dunn PO, Garant D, Holroyd G, Horn AG, Hussell DJT, Lansdorp O, Laughlin AJ, Leonard ML, Pelletier F, Shutler D, Siefferman L, Taylor CM, Trefry H, Vleck CM, Vleck D, Whittingham LA and Norris DR (2019) Effects of Spring Migration Distance on Tree Swallow Reproductive Success Within and Among Flyways. Front. Ecol. Evol. 7:380. doi: 10.3389/fevo.2019.00380 Elizabeth A. Gow<sup>1</sup> \*, Samantha M. Knight <sup>1</sup> , David W. Bradley <sup>2</sup> , Robert G. Clark <sup>3</sup> , David W. Winkler <sup>4</sup> , Marc Bélisle<sup>5</sup> , Lisha L. Berzins <sup>6</sup> , Tricia Blake<sup>7</sup> , Eli S. Bridge<sup>8</sup> , Lauren Burke<sup>9</sup> , Russell D. Dawson<sup>6</sup> , Peter O. Dunn<sup>10</sup>, Dany Garant <sup>5</sup> , Geoff Holroyd<sup>11</sup> , Andrew G. Horn<sup>9</sup> , David J. T. Hussell 12†, Olga Lansdorp<sup>13</sup>, Andrew J. Laughlin<sup>14</sup> , Marty L. Leonard<sup>9</sup> , Fanie Pelletier <sup>5</sup> , Dave Shutler <sup>15</sup>, Lynn Siefferman<sup>16</sup>, Caz M. Taylor <sup>17</sup> , Helen Trefry <sup>11</sup>, Carol M. Vleck <sup>18</sup>, David Vleck <sup>18</sup>, Linda A. Whittingham<sup>10</sup> and D. Ryan Norris <sup>1</sup>

<sup>1</sup> Department of Integrative Biology, University of Guelph, Guelph, ON, Canada, <sup>2</sup> Bird Studies Canada, Delta, BC, Canada, <sup>3</sup> Environment and Climate Change Canada, Saskatoon, SK, Canada, <sup>4</sup> Department of Ecology and Evolutionary Biology, Museum of Vertebrates, Cornell University, Ithaca, NY, United States, <sup>5</sup> Département de Biologie, Université de Sherbrooke, Sherbrooke, QC, Canada, <sup>6</sup> Ecosystem Science and Management Program, University of Northern British Columbia, Prince George, BC, Canada, <sup>7</sup> Alaska Songbird Institute, Fairbanks, AK, United States, <sup>8</sup> Oklahoma Biological Survey, University of Oklahoma, Norman, OK, United States, <sup>9</sup> Department of Biology, Dalhousie University, Halifax, NS, Canada, <sup>10</sup> Behavioral and Molecular Ecology Group, Department of Biological Sciences, University of Wisconsin-Milwaukee, Milwaukee, WI, United States, <sup>11</sup> Beaverhill Bird Observatory, Edmonton, AB, Canada, <sup>12</sup> Ontario Ministry of Natural Resources, Peterborough, ON, Canada, <sup>13</sup> Department of Biological Sciences, Simon Fraser University, Burnaby, BC, Canada, <sup>14</sup> Department of Environmental Studies, University of North Carolina Asheville, Asheville, NC, United States, <sup>15</sup> Department of Biology, Acadia University, Wolfville, NS, Canada, <sup>16</sup> Biology Department, Appalachian State University, Boone, NC, United States, <sup>17</sup> Department of Ecology and Evolutionary Biology, Tulane University, New Orleans, LA, United States, <sup>18</sup> Department of Ecology, Evolution and Organismal Biology, Iowa State University, Ames, IA, United States

During migration, animals may experience high rates of mortality, but costs of migration could also be manifested through non-lethal carry-over effects that influence individual success in subsequent periods of the annual cycle. Using tracking data collected from light-level geolocators, we estimated total spring migration distance (from the last wintering sites to breeding sites) of tree swallows (Tachycineta bicolor) within three major North American flyways. Using path analysis, we then assessed direct and indirect effects of spring migration distance on reproductive performance of individuals of both sexes. When these data were standardized by flyway, females fledged 1.3 fewer young for every 1,017 km they traveled, whereas there was no effect of migration distance on reproductive success in males. In comparison, when these data were standardized across all individuals and not by flyway, longer migrations were associated with 0.74 more young fledged for every 1,017 km traveled by females and 0.26 more young fledged for every 1,186 km migrated by males. Our results suggest that migration distance carries over to negatively influence female reproductive success within flyways but the overall positive effect of migration distance across flyways likely reflects broader life-history differences that occur among breeding populations across the tree swallow range.

Keywords: tree swallow, migration, geolocation, migration distance, path analysis, young fledged

# INTRODUCTION

Migration is widespread throughout the animal kingdom (reviewed in Newton, 2008) and likely evolved as an adaptation to optimize resource use (Alerstam et al., 2003; Alerstam, 2011). However, traveling between locations, many of which are thousands of kilometers apart, is also considered to be costly (Wikelski et al., 2003). Such costs are primarily thought to be "direct" in the form of higher mortality when compared to non-migratory periods of the annual cycle (Lok et al., 2014), though the migratory period may not be the only period with the highest mortality rates (Leyrer et al., 2013; Rakhimberdiev et al., 2015a; Senner et al., 2019). Among those individuals that survive migration, the cost of traveling such long distances may also carry over to influence reproductive success the following season (Harrison et al., 2011). Determining the existence and strength of these carry-over effects and how they may vary within and among populations will contribute to our understanding of long-term population dynamics and how life-history trade-offs shape broad-scale migration patterns (Norris and Marra, 2007; Harrison et al., 2011; Betini et al., 2013).

Many species of birds migrate different distances, even within a single breeding population (Fraser et al., 2012; McKinnon et al., 2013; Knight et al., 2018; McKinnon and Love, 2018). Only a few avian studies have examined carry-over effects of migration distance on individual reproductive success following a breeding season, and most have focused on whether migration distance is related to timing of arrival at a breeding site (Hötker, 2002; Bregnballe et al., 2006; Gunnarsson et al., 2006; Alves et al., 2012; Briedis et al., 2019), the start of breeding (Lok et al., 2016; Kentie et al., 2017), and breeding productivity (Bearhop et al., 2005; Bregnballe et al., 2006; Lok et al., 2016; Kentie et al., 2017). For example, in great cormorants, Phalacrocorax carbo (Bregnballe et al., 2006), and pied avocets, Recurvirostra avosetta (Hötker, 2002), birds wintering farther south arrived later at breeding sites. For pied avocets, early arrival led to higher breeding success (Hötker, 2002), whereas this relationship was not observed in great cormorants (Bregnballe et al., 2006). More southerly wintering black-tailed godwits, Limosa limosa that crossed the Sahara, started breeding earlier than those wintering farther north that did not cross the Sahara, but there was little effect of migration distance on reproductive success (Kentie et al., 2017). Male Eurasian spoonbills, Platalea leucorodia, that migrated longer distances began breeding later and subsequently produced fewer and lower quality chicks, and recruited fewer young (Lok et al., 2016). Similarly, in a study on songbirds, European blackcaps, Sylvia atricapilla, wintering farther north, as estimated from stable isotopes, produced larger clutches and fledged more young compared to those wintering farther south (Bearhop et al., 2005). Collectively these studies suggest that non-lethal effects of migration distance on reproduction might depend on the species or ecological context, and strongly emphasize that further study is needed across a wider range of taxa and among multiple populations.

Recently, we described a migratory network based on year-round movements of tree swallows (Tachycineta bicolor) originating from 12 breeding populations across their range (Knight et al., 2018). In addition to linking breeding populations with stopover and wintering sites, we also identified three distinct migratory flyways (Knight et al., 2018). In the Western flyway, tree swallows breeding west of the Rockies migrated primarily to western Mexico, those in the Central flyway bred in central Canada or the U.S and either crossed the Gulf of Mexico to wintering sites in eastern Mexico, or wintered in Louisiana, Mississippi, or Florida (Central flyway). The Eastern flyway consisted of tree swallows that bred in eastern North America and primarily used wintering sites in Florida, the Caribbean Islands, or Cuba (Eastern flyway; **Figure 1**; Knight et al., 2018). In a subsequent study, the timing of arrival on the breeding grounds appeared to be most strongly influenced by both the latitude from which the birds departed and the latitude of the breeding site (Gow et al., 2019), suggesting that the distance an individual traveled could influence the timing of breeding. Whether and how spring migration distance could carry over to influence subsequent reproductive success is not known.

There are several factors that influence reproductive performance (i.e., number of young fledged), and ultimately population dynamics (Cox et al., 2018). These include reproductive traits, such as clutch size (Dunn et al., 2000; Millet et al., 2015), timing of events such as arrival at breeding locations (e.g., Norris et al., 2004), or start of breeding (i.e., first egg dates; Hochachka, 1990; Verhulst and Nilsson, 2008; Millet et al., 2015). Understanding the relationships among these factors and how they directly or indirectly affect reproductive success could provide mechanisms through which spring migration distance and the duration of spring migration may influence productivity. Here, we evaluated (1) the potential for spring migration distance to affect reproductive performance among populations, (2) how this may vary between major flyways, and (3) whether migration distance had different effects on each sex. We used path analysis (Shipley, 2009) to first quantify the direct effects of spring migration distance on timing of arrival to breeding sites, first egg date, clutch size, and the number of young fledged, and then estimated possible indirect effects of spring migration distance on the number of young fledged via the other reproductive metrics. We evaluated these effects by sex controlling for flyway (identified in Knight et al., 2018), and then compared these results to analyses when data from all flyways were combined. From the direct and indirect effects estimates from the path models, we then generated predictions of the total effect of migration distance on the number of young fledged.

#### MATERIALS AND METHODS

#### Study Sites and Data Collection

Between 2010 and 2014, we equipped 561 adult tree swallows with an archival light-level geolocator (hereafter referred to as "geolocators") at the 12 breeding sites (Fairbanks, Alaska, 65.90◦N, 147.70◦W; Vancouver, British Columbia, 49.21◦N, 123.18◦W; Prince George, British Columbia, 53.85◦N, 123.02◦W; Beaverhill, Alberta, 53.40◦N, 112.50◦W, Saskatoon, Saskatchewan, 52.17◦N, 106.10◦W; Ames, Iowa, 42.11◦N, 93.59◦W; Saukville, Wisconsin, 43.40◦N, 88.00◦W; Boone, North Carolina, 36.21◦N, 81.67◦W; Long Point, Ontario, 42.62◦N,

FIGURE 1 | The breeding (triangles) and wintering (females: circles; males: diamonds) locations of tree swallows from populations in (A) Western and Central flyways (B) Eastern flyway. Each circle (female) or diamond (male) is an individual tree swallow on its primary wintering grounds and the colors indicate breeding origin. The arrows in both panels depict the general spring migration routes from wintering to breeding locations (based on Bradley et al., 2014; Knight et al., 2018). These routes were used to calculate migration distance. Map is drawn using the default latitude/longitude projection. The size of each individual circle or diamond represents the approximate error associated with geolocators in this study. Sample sizes are: N = 33 Eastern flyway females; N = 20 Eastern flyway males; N = 13 Central flyway females; and N = 23 Central flyway males.

80.46◦W; Ithaca, New York, 42.50◦N, 76.50◦W; Sherbrooke, Québec, 45.55◦N, 72.60◦W; Wolfville, Nova Scotia, 45.10◦N, 64.39◦W). Overall, we retrieved 161 geolocators, 133 of which were free from malfunctions. Each bird was tracked for a 1 year period. Of these 133, we obtained reproductive history from 105 individuals. Birds from Vancouver, Alaska, Iowa, and North Carolina were not included in this study, either because of small sample size or because individuals at these sites were outliers based on migration distances within their respective flyway. Eighty-nine individuals (46 females and 43 males) were therefore used in this study.

## Geolocator Analysis and Definition of the Last Wintering Site

Light data from geolocators were analyzed using the BAStag package 0.1.3 (Wotherspoon et al., 2013) and FlightR package version 0.3.6 (Rakhimberdiev et al., 2015b) with R version 3.2.3 (R Core Development Team, 2016). The FlightR package works well for open area birds, and it uses a state-space hidden Markov model to estimate daily locations. For step-by-step details about geolocator deployment and analysis see Knight et al. (2018). Given the well-defined light transitions, geolocator error was minimal (46 ± 90 km in latitude and 52 ± 90 km in longitude; Gow et al., 2019). Geolocator error was calculated based on averaged location estimates from the breeding site (Gow et al., 2019). We followed the definitions in Gow et al. (2019) for determining wintering site, and identifying departure and arrival at sites. Briefly, locations were determined by calculating the mean location of all daily locations from a stationary period. We considered birds to have departed from the wintering site if they made a large (≥ 250 km) northward movement, lasting for at least 2 d, away from a stationary position. Given that tree swallows may use more than one wintering site (Knight et al., 2018), the last wintering site was defined as the last location that a tree swallow spent at least 28 d. Breeding arrival date was defined as the first day tree swallows had location estimates consistently matching those of the known breeding site.

# Definitions of Reproductive Metrics and Flyways

Tree swallows breed in natural tree cavities or nest boxes throughout the southern half of Canada and the north/central United States; they generally have clutches of 4–7 eggs (Winkler et al., 2011). All tree swallows in our study were single-brooded, although some females attempted a second clutch if their first nest was depredated. Tree swallow clutch sizes increase with breeding latitude (Dunn et al., 2000). At each study site, nest boxes were checked every 1–7 days (with most sites checking nests every 1–3 days) to obtain the following breeding information from individuals in the year following geolocator deployment: first egg date (date on which the first egg of the first clutch of the season was laid), clutch size (number of eggs laid in the first clutch of the season), and number of young fledged (number of young estimated to have survived to fledging). All failed nests were counted as zero in our analyses. We determined breeding arrival date (date bird first arrived at the breeding site), and spring migration distance for each tree swallow carrying a geolocator. We calculated spring migration distance as the great circle distance between the last overwintering site and the breeding site (i.e., spring migration distance; **Figure 1**; Bradley et al., 2014). The arrows in **Figure 1** show the general pathways used to calculate migration distance for each population. Most tree swallows migrated during the spring equinox making it difficult to estimate the true travel route. Thus, for consistency among individuals we calculated migration distance using several points along the migration pathways for tree swallows defined by Bradley et al. (2014; **Figure 1**). The connections between wintering and breeding sites are available in Knight et al. (2018).

We interpreted clutch size and the number of young fledged differently between sexes. For females, clutch size and the number of young fledged are direct measures of reproductive success. However, because there is rampant extra-pair paternity in tree swallows (e.g., 50–89%; Lifjeld et al., 1993; Barber et al., 1996; Kempenaers et al., 2001; Whittingham and Dunn, 2001; O'Brien and Dawson, 2007), the number of young fledged likely does not represent realized (i.e., genetic) reproductive success of males. Realized reproductive success would account for both a male's potential paternity lost within his nest and potential paternity gained via extra-pair fertilizations outside the social pair bond. Given that we did not have genetic paternity data from our sites, we could not include realized reproductive success of males in our study. Thus, for males, the number of young fledged within the social pair bond (the variable we measured and included in our study) more accurately reflects the quality of the social partner rather than true reproductive success in a given season. First egg dates and clutch size also mainly reflect the quality of the social partner rather than true reproductive parameters for males, meaning that the only reproductive variable that is solely influenced by males was breeding arrival date. Breeding arrival date may have subsequent carry-over effects that may influence the quality of the social mates that a male is able to acquire.

A network analysis by Knight et al. (2018) showed how breeding populations segmented into three migratory flyways: Western, Central, and Eastern. We chose to combine the Western and Central flyways because of the small sample of birds in the Western flyway (n = 11) and because these two flyways had similar means and standard deviations (s.d.) of migration distance (Western: 3,833 ± 795 km; Central: 4,217 ± 1,082 km; **Table 1**). We eliminated 4 populations from our analyses that had a small sample size (Vancouver, BC and Ames, IA) and/or represented extreme southern and northern regions of the tree swallows' range (Fairbanks, AK and Boone, SC). The populations we included from the Western and Central flyways (hereafter referred to as the Central flyway for simplicity) were Prince George, BC, Beaverhill, AB, and Saskatoon, SK (N = 36; female = 13; male = 23; **Figure 1A**), and those in the Eastern flyway were Saukville, WI, Long Point, ON, Ithaca, NY, Sherbrooke, QC, and Wolfville, NS (N = 53; female = 33; male = 20; **Figure 1B**).

#### Path Analysis

Prior to executing the path analysis, we undertook two types of data standardizations to help separate the potential effects of breeding location on life-history variation from the effects on individuals migrating farther than other individuals within their flyway. First, we standardized each variable by flyway (sexes pooled) by subtracting the mean then dividing by the standard

TABLE 1 | Summary of migration distances among breeding sites across sexes and separated by females and males.


The minimum (min), maximum (max), mean, median, standard deviation (s.d.), flyways and samples sizes (N) are indicated. All distances are in km. Flyways were identified by Knight et al. (2018).

deviation (i.e., z-transformation). This allowed us to control for flyway effects. Second, we z-transformed these data across all individuals independent of flyway. If there was a positive effect of migration distance on first egg date, clutch size, or young fledged when the dataset was standardized across all individuals, then this may indicate the relationship was due to life-history factors rather than migration distance per se. In contrast, when standardizing by flyway, if there is a negative relationship between migration distance on first egg date, clutch size or young fledged, then this may suggest a potential carry-over effect of migration distance.

Following this standardization, we evaluated the direct and indirect effects of migration duration, migration distance, breeding arrival date, first egg date, and clutch size on the number of young fledged using a multi-level path modeling framework (Shipley, 2000, 2009). We included a random effect of "breeding site" to account for local-level effects across the 8 breeding populations. All mixed effects models were fitted with a Gaussian distribution, as the response variables best fit this distribution, using the nlme package (Pinheiro et al., 2018) in R 3.5.2 (R Core Development Team, 2018). We identified the most parsimonious path model based on Akaike's Information Criterion corrected for sample size (AICc; Shipley, 2000, 2009, 2013). We evaluated four different sets of path models, separated by sex and standardization method (i.e., standardized by flyway or standardized across individuals).

We structured the path models based on previous knowledge of tree swallow ecology. For each set of path models, we started by fitting a global model, which included direct effects of spring migration duration, spring migration distance, breeding arrival date, first egg date, and clutch size on the number of young fledged. We removed terms associated with uninformative estimates for young fledged first, followed by those with uninformative estimates for clutch size, first egg date, breeding arrival date, and migration distance. We determined the order of deletion using AICc to assess the terms with the least support in each submodel using maximum likelihood estimation (MuMIn package; Barton, 2016; see **Figure S1** for path analysis submodels). We removed terms from the path model if their deletion did not increase the AIC by at least two units (**Table S1**). Models were not averaged because top models were all nested within preceding models (Arnold, 2010).

From the path analysis, we calculated direct effects of one variable on another as well as indirect effects (Mitchell, 1993). Indirect effects were calculated by taking the product of all possible pathways (path coefficients) from one variable to another. Direct effects occur between variables and are generated by path coefficients (regression beta coefficients). Because we standardized the dataset prior to conducting the path analysis we did not need to standardize the path coefficients. The total effect was calculated as the sum of all indirect and direct effects from one variable to another.

We compared migration distances between sex and standardization type using a linear mixed effects model (LMM). We included breeding site as a random effect. We used post-hoc pairwise differences to compare migration distances between the sexes standardized by flyway or across individuals.

### Predicted Effects of Migration Distance on Young Fledged

Because total effect values summarize all direct and indirect pathways between migration distance to the number of young fledged, we also produced a predictive model. This model involved first summarizing the effect of migration distance on the number of young fledged using the total effect values and standard deviations (s.d.) for each variable (Bart and Earnst, 1999). For example, the effect of migration distance on number of young fledged was expressed as: s.d. (migration distance) = s.d. (young fledged<sup>∗</sup> total effect). We took the mean migration distance and used it to predict differences in the number of young fledged across different migration distances. We only developed predictive models for the sex and standardizations in which there was at least one effect of migration distance on a reproductive metric.

This study was carried out in accordance with the principles of Animal Utilization Protocols or Animal Care Protocols and was approved by each University of the primary researcher for each field site.

# RESULTS

# Variation of Spring Migration Distance Between Flyways and Sexes

Tree swallows migrated on average 2,930 ± 1,110 km (range: 1,496–4,879 km). While males migrated farther and showed more variation in migration distances than females (males: 3,110 ± 1,187 km; females: 2,761 ± 1,017 km), there was no evidence the sexes arose from different distributions (LMM: β = −61.89 ± 82.44, t = −0.75, p = 0.45; **Table 1**). Females migrated significantly farther and had different distributions in their migration distances in the Central flyway than the Eastern flyway (LMM: β = 2,129 ± 261, t = 8.14, p < 0.001; **Table 1**), and a similar pattern was observed for males (LMM: β = 2,201 ± 161, t = 13.66, p < 0.001; **Table 1**). Males spent more time on average migrating than females but this difference was not significant (LMM: β = 5.58 ± 4.67, t = 1.19, p = 0.24; **Table 2**). In the Central flyway, females spent significantly more time on spring migration compared to the Eastern flyway (LMM: β = 15.58 ± 6.0, t = 2.60, p < 0.01; **Table 2**). Although males in the Central flyway spent more days migrating in the spring than in the Eastern flyway, this was not significantly different (LMM: β = 13.13 ± 7.24, t = 1.81, p = 0.14; **Table 2**).

### Effect of Migration Distance on Reproductive Performance: Variables Standardized by Flyway

For females, migration distance negatively affected the number of young fledged (**Figure 2**). Females migrating the shortest distances within their flyway fledged more young compared to those migrating the farthest distances, suggesting a direct carry-over effect of migration distance on young fledged. In contrast, for males there was no effect of migration distance on clutch size or young fledged, but males that spent longer migrating in the spring arrived later to the breeding site. Breeding arrival date positively influenced first egg dates in females, but males that arrived later had social mates that laid their first clutches earlier. For females, first egg dates did not influence clutch size. Earlier breeding males had social mates that produced larger clutches than later breeding males. For females, clutch size positively affected the number of young fledged, and males mated to females that laid large clutches also fledged more young. Overall, the total effect of migration distance on number of young fledged for females was −0.55 ± 0.33, resulting in a predicted 1.33 fewer young for every 1,017 km they migrated (**Figure 3**).

#### Effect of Migration Distance on Reproductive Performance: Variables Standardized Across All Individuals

Migration distance in females directly and positively affected migration duration, first egg dates, and young fledged (**Figure 2**). Given the later first egg dates and higher number of young fledged in the Central flyway than the Eastern flyway (**Table 2**), results from the path analysis showed that females migrating the farthest distances (most birds in the Central flyway) fledged


TABLE 2 | Summary of the means, standard deviations (s.d.) and range (minimum and maximum) of migration distance and duration, timing (breeding arrival date, first egg date) events, and reproductive variables (clutch size and number of young fledged) for females and males within each flyway.

Sample sizes are: N = 33 Eastern flyway females; N = 20 Eastern flyway males; N = 13 Central flyway females; and N = 23 Central flyway males.

more young and also began breeding later than females migrating shorter distances (most birds in the Eastern flyway). For males, migration distance had a positive direct effect on migration duration, and the male's social mate's first egg date and clutch size. Furthermore, males traveling the farthest distances (typically in the Central flyway) began breeding later and their social mate had larger clutches than those migrating shorter distances (typically the Eastern flyway). But the negative relationship between first egg date and clutch size suggested that males with social mates that began breeding later had smaller clutches. Overall, the total effect of migration distance on young fledged for females was 0.31, which implied that females fledged 0.74 more young for every 1,017 km farther they migrated. For males, the total effect of migration distance on young fledged was 0.16, which implied

that males fledged 0.26 more young for every 1,186 km farther they migrated (**Figure 3**).

# DISCUSSION

Collectively, previous research suggests that non-lethal effects of migration distance on reproduction might be species or context specific (Hötker, 2002; Bregnballe et al., 2006; Gunnarsson et al., 2006; Alves et al., 2012; Lok et al., 2016; Kentie et al., 2017), but if migration distance is energetically costly then individuals that migrate farther distances may experience carry-over effects into stationary periods of the annual cycle (Harrison et al., 2011 but see Conklin et al., 2017). Our results suggest that variation in migration distance within a flyway with individuals breeding at similar latitudes, had a negative effect on the number of young fledged for females. For males, there were no effects of migration distance when data were standardized by flyway, but males that migrated faster arrived earlier, and were mated to females that began breeding later. This suggests males migrating at faster paces may experience a cost in their ability to acquire an early breeding mate, even though they arrived early to the breeding site. In contrast, when data were standardized across all individuals, we show that overall migration distance is positively associated with fledging success, and thus, may be representing broader life-history differences.

It is possible that the negative relationship between migration distance on young fledged in females was the result of locationspecific life history variation rather than a carry-over effect. However, we argue that this is unlikely for several reasons. First, breeding sites within the Central and Eastern Flyways varied little in latitude within their flyways (i.e., 1.68◦N in the Central flyway and 3.05◦N in Eastern flyway), suggesting any such effects of migration distance were likely unrelated to latitudinal variation in the breeding location. Second, within breeding sites there was often large variation in migration distance (**Table 1**). This was especially true of birds in the Central flyway as individuals migrated to Mexico, the Gulf of Mexico, and in some cases, Florida resulting in migration distances within sites that varied by over 1,000 km (**Figure 1**). Thus, by standardizing within flyways we were able to separate the potential effects of breeding location from migration on reproductive variables. However, future research that includes additional sites and individuals within a migratory network would likely provide a clearer picture of the effect of migration distance on reproduction.

In contrast, when standardizing across all individuals, males and females that migrated the farthest distances fledged the most young. We argue that this result was likely driven by broader life-history differences related to breeding latitude (Ricklefs, 1980; Dunn et al., 2000; Jetz et al., 2008) rather than migration distance, per se. Tree swallows wintered within a narrow band ranging ∼11◦ of latitude whereas their breeding range covers ∼34◦ of latitude (Winkler et al., 2011), leading to spring migration distance, when examined across the entire network, being positively correlated with breeding latitude (see also Gow et al., 2019). Previously, we provided evidence that tree swallows breeding at higher latitudes arrived later and began breeding later (Gow et al., 2019), similar to the positive effects of migration distance we observed on first egg date of males and females in this study when we standardized across individuals.

One reason we observed a negative effect of migration distance on young fledged in females but not males (when data were standardized within flyways) may be related to how the sexes differ in their sensitivity to the energetic or physiological costs of migration distance. Under the reproductive stress hypothesis, the sex with the higher reproductive demands will be more sensitive to their energetic or physiological state when investing in reproduction (Nagy et al., 2007; Gow et al., 2013; Gow and Wiebe, 2014). In tree swallows, females invest in reproduction through nest building, egg laying, incubation, feeding young and intense female-female competition. Males only engage in feeding young, but also invest in male-male competition, pursuit of extrapair fertilizations, and securing and defending nest sites (Winkler et al., 2011). Thus, female tree swallows are the sex investing more heavily in the production of young. Experimental studies manipulating female quality or timing of breeding (Winkler and Allen, 1995; Dawson, 2008; Harriman et al., 2016) suggest a female's quality may affect her ability to produce and care for offspring. In this way, it is possible that migration distance influenced the number of young fledged by affecting female condition, but did not affect whether a male was capable of mating with a female that produced more young.

Male tree swallows that migrated faster relative to other individuals in their flyway arrived earlier to the breeding site. However, arriving earlier to the breeding grounds may not necessarily be beneficial if it leaves an individual in a poorer condition (e.g., González-Prieto and Hobson, 2013), unless they acquired larger reserves prior to migration (Bayly et al., 2016). Arriving to the breeding grounds in a poor physiological state may be particularly detrimental for male tree swallows given their reliance on aerial insects. Interestingly, male tree swallows that arrived early to the breeding grounds (when standardizing by flyway) seemed to be mated to females that began breeding relatively late. One reason for this negative relationship may be related to how breeding tree swallows are impacted by poor weather (Weegman et al., 2017; Cox et al., 2019) and low insect abundances (McCarty and Winkler, 1999; Imlay et al., 2017). Both poor weather and low insect abundances may impair reproductive performance via their effects on the timing of breeding (Dawson, 2008; Harriman et al., 2016), and female body condition (Winkler and Allen, 1995; Paquette et al., 2014). These factors may also differ among flyways. Populations in the Central and Western flyways appear to be more strongly affected by timing of breeding and insect abundances during egg laying (Dawson, 2008; Harriman et al., 2016), while those in the Eastern flyway may experience more negative effects from poor weather conditions (Weegman et al., 2017; Cox et al., 2019) rather than variation in insect abundance (McCarty and Winkler, 1999; Imlay et al., 2017). This difference may explain why when we did not standardize by flyway and instead standardized across all individuals faster arriving males arrived earlier (similar to within flyway standardization). These early arriving males mated with females whom also began breeding earlier, demonstrating an overall positive benefit to migrating faster and arriving early across the range.

Our findings provide valuable insight into how migration distance may influence current and future population declines of tree swallows, as well as other species. Many tree swallow populations in the northeastern parts of their range have experienced declines over the past couple of decades (Shutler et al., 2012). The cause(s) of these declines are unclear, but one possibility is that deterioration of overwintering habitat quality influences survival and carries over to influence reproductive performance. Another mechanism for population declines in this species, and potentially others, may occur if individuals are forced to migrate farther. Individuals may migrate farther distances if there is a reduction in habitat quality, which may reduce the carry-capacity of those sites (e.g., Stutchbury et al., 2016), forcing individuals to seek alternative roosting sites farther south. For swallows breeding in the Western or Central flyways, this may mean either crossing the Gulf of Mexico or moving to areas farther south in Central America, whereas those breeding in the Eastern flyway may be forced to seek habitat on Caribbean Islands or travel even farther distances to Mexico. Alternatively, with the globally rising temperatures, suitable habitat for tree swallows may be available farther north. The geographic differences among flyways may affect the potential distances some individual migrate, which in turn may affect the number of young fledged, and contribute to population declines.

# DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the manuscript/**Supplementary Files**.

# ETHICS STATEMENT

This study was carried out in accordance with the principles of Animal Utilization Protocols or Animal Care Protocols and was approved by each University of the primary researcher for each field site.

# AUTHOR CONTRIBUTIONS

EG and DN designed the research and wrote the manuscript. EG analyzed these data. SK processed the geolocator data. All other authors conducted fieldwork and helped with manuscript revisions.

# FUNDING

Funding was provided by Leaders Opportunity Fund Grants from the Canadian Foundation for Innovation (DN and RD), Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grants (DN, RC, RD, MB, DG, FP, and ML), an NSERC Research Tools and Instruments Grant (DN, MB, RD, DG, ML, FP, and DS), an NSERC Industrial Research and Development Fellowship (DB), an NSERC Alexander Graham Bell Canada Graduate Scholarship (LLB), the NSERC Canada Research Chairs Program (MB and FP), the University of Guelph (DN), Environment and Climate Change Canada (RC and OL), Bird Studies Canada (DB and DH), the University of Northern British Columbia (RD), the British Columbia Knowledge Development Fund (RD), the Skaggs Foundation (TB), a National Science Foundation Grant DEB-0933602 (CT), a National Science Foundation Grant IOS-0745156 (CV and DV), National Science Foundation Grants DEB-0717021 and DEB-1242573 (DW), Fonds de Recherche du Québec – Nature et Technologies (MB, DG, and FP), the James S. McDonnell foundation (CT), the Alberta Conservation Association (GH and HT), TD Friends of the Environment (GH and HT), the Shell Environmental Fund (GH and HT), and Nature Canada's Charles Labatiuk Nature Endowment Fund (GH and HT). The development and analysis of some of the geolocators were supported by the National Science Foundation Grants Nos. IDBR 1152356 and DEB 0946685 (EB), IDBR 1152131 (DW).

# ACKNOWLEDGMENTS

We thank the numerous field assistants, graduate students, undergraduate students, and volunteers who assisted with fieldwork at all breeding sites used in this study.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00380/full#supplementary-material

#### Gow et al. Migration Distance and Reproductive Success

#### REFERENCES


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer, MH, declared past collaborations with several of the authors, DN, DW, CT, and EB, to the handling editor.

Copyright © 2019 Gow, Knight, Bradley, Clark, Winkler, Bélisle, Berzins, Blake, Bridge, Burke, Dawson, Dunn, Garant, Holroyd, Horn, Hussell, Lansdorp, Laughlin, Leonard, Pelletier, Shutler, Siefferman, Taylor, Trefry, Vleck, Vleck, Whittingham and Norris. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Spatial and Temporal Variability in Migration of a Soaring Raptor Across Three Continents

#### Edited by:

Nathan R. Senner, University of South Carolina, United States

#### Reviewed by:

Wouter Marc Gerard Vansteelant, Estación Biológica de Doñana (EBD), Spain Tom Finch, Royal Society for the Protection of Birds (RSPB), United Kingdom

\*Correspondence:

W. Louis Phipps l.phipps@4vultures.org

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 18 April 2019 Accepted: 13 August 2019 Published: 10 September 2019

#### Citation:

Phipps WL, López-López P, Buechley ER, Oppel S, Álvarez E, Arkumarev V, Bekmansurov R, Berger-Tal O, Bermejo A, Bounas A, Alanís IC, de la Puente J, Dobrev V, Duriez O, Efrat R, Fréchet G, García J, Galán M, García-Ripollés C, Gil A, Iglesias-Lebrija JJ, Jambas J, Karyakin IV, Kobierzycki E, Kret E, Loercher F, Monteiro A, Morant Etxebarria J, Nikolov SC, Pereira J, Peške L, Ponchon C, Realinho E, Saravia V, ¸Sekercioglu ÇH, Skartsi T, ˘ Tavares J, Teodósio J, Urios V and Vallverdú N (2019) Spatial and Temporal Variability in Migration of a Soaring Raptor Across Three Continents. Front. Ecol. Evol. 7:323. doi: 10.3389/fevo.2019.00323 W. Louis Phipps <sup>1</sup> \* † , Pascual López-López 2†, Evan R. Buechley 3,4†, Steffen Oppel 5† , Ernesto Álvarez <sup>6</sup> , Volen Arkumarev <sup>7</sup> , Rinur Bekmansurov <sup>8</sup> , Oded Berger-Tal <sup>9</sup> , Ana Bermejo<sup>10</sup>, Anastasios Bounas 11,12, Isidoro Carbonell Alanís <sup>13</sup>, Javier de la Puente<sup>10</sup> , Vladimir Dobrev <sup>7</sup> , Olivier Duriez <sup>14</sup>, Ron Efrat <sup>9</sup> , Guillaume Fréchet <sup>15</sup>, Javier García<sup>16</sup> , Manuel Galán<sup>6</sup> , Clara García-Ripollés <sup>17</sup>, Alberto Gil <sup>6</sup> , Juan José Iglesias-Lebrija<sup>6</sup> , José Jambas <sup>18</sup>, Igor V. Karyakin<sup>19</sup>, Erick Kobierzycki <sup>20</sup>, Elzbieta Kret <sup>21</sup> , Franziska Loercher <sup>1</sup> , Antonio Monteiro<sup>22</sup>, Jon Morant Etxebarria<sup>23</sup>, Stoyan C. Nikolov <sup>7</sup> , José Pereira<sup>24</sup>, Lubomír Peške<sup>25</sup>, Cecile Ponchon<sup>26</sup>, Eduardo Realinho<sup>27</sup> , Victoria Saravia<sup>12</sup>, Çagan H. ¸Sekercio ˘ glu˘ 28,29, Theodora Skartsi <sup>21</sup>, José Tavares <sup>1</sup> , Joaquim Teodósio<sup>30</sup>, Vicente Urios <sup>31</sup> and Núria Vallverdú<sup>27</sup>

<sup>1</sup> Vulture Conservation Foundation, Zurich, Switzerland, <sup>2</sup> Terrestrial Vertebrates Group, Cavanilles Institute of Biodiversity and Evolutionary Biology, University of Valencia, Valencia, Spain, <sup>3</sup> Smithsonian Migratory Bird Center, Washington, DC, United States, <sup>4</sup> HawkWatch International, Salt Lake City, UT, United States, <sup>5</sup> Royal Society for the Protection of Birds, RSPB Centre for Conservation Science, Cambridge, United Kingdom, <sup>6</sup> Grupo de Rehabilitación de la Fauna Autóctona y su Hábitat, Madrid, Spain, <sup>7</sup> Bulgarian Society for Protection of Birds/BirdLife Bulgaria, Sofia, Bulgaria, <sup>8</sup> Educational and Scientific Laboratory - Monitoring and Protection of Birds, Elabuga Institute, Kazan Federal University, Elabuga, Russia, <sup>9</sup> Mitrani Department of Desert Ecology, Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Midreshet Ben-Gurion, Israel, <sup>10</sup> Bird Monitoring Unit, SEO/BirdLife, Madrid, Spain, <sup>11</sup> Department of Biological Applications and Technologies, University of Ioannina, Ioannina, Greece, <sup>12</sup> Hellenic Ornithological Society, BirdLife Greece, Athens, Greece, <sup>13</sup> SALORO, Salamanca, Spain, <sup>14</sup> UMR 5175, Centre d'Ecologie Fonctionnelle et Evolutive, CNRS - Université de Montpellier, Université Paul-Valéry Montpellier - EPHE, Montpellier, France, <sup>15</sup> Syndicat Mixte des Gorges du Gardon, Gard, France, <sup>16</sup> Department of Biodiversity and Environmental Management, University of León, León, Spain, <sup>17</sup> Environment Science and Solutions, Valencia, Spain, <sup>18</sup> Oriolus Ambiente e Ecoturismo, Atenor, Picote, Portugal, <sup>19</sup> Sibecocentar LLC, Novosibirsk, Russia, <sup>20</sup> Nature en Occitanie, Coordination Technique Plan National d'Actions Vautour Percnoptère, Bruges, France, <sup>21</sup> WWF Greece, Athens, Greece, <sup>22</sup> Instituto da Conservação da Natureza e das Florestas, Lisbon, Portugal, <sup>23</sup> Department of Ornithology, Aranzadi Sciences Society, Donostia-San Sebastián, Spain, <sup>24</sup> Palombar, Associação da Conservação da Natureza e do Património Rural, Antiga Escola Primária de Uva, Uva, Portugal, <sup>25</sup> Independent Researcher, Prague, Czechia, <sup>26</sup> Conservatoire Espaces Naturels Provence-Alpes-Côte d'Azur, Aix-en-Provence, France, <sup>27</sup> Transumância e Natureza – Associação (ATN), Figueira de Castelo Rodrigo, Portugal, <sup>28</sup> Department of Biology, University of Utah, Salt Lake City, UT, United States, <sup>29</sup> College of Sciences, Koç University, Istanbul, Turkey, <sup>30</sup> Sociedade Portuguesa para o Estudo das Aves (SPEA), Lisbon, Portugal, <sup>31</sup> Vertebrate Zoology Research Group, University of Alicante, Alicante, Spain

Disentangling individual- and population-level variation in migratory movements is necessary for understanding migration at the species level. However, very few studies have analyzed these patterns across large portions of species' distributions. We compiled a large telemetry dataset on the globally endangered Egyptian Vulture Neophron percnopterus (94 individuals, 188 completed migratory journeys), tracked across ∼70% of the species' global range, to analyze spatial and temporal variability of migratory movements within and among individuals and populations. We found high migratory connectivity at large spatial scales (i.e., different subpopulations showed little overlap in wintering areas), but very diffuse migratory connectivity within subpopulations, with wintering ranges up to 4,000 km apart for birds breeding in the same region and each subpopulation visiting up to 28 countries (44 in total). Additionally, Egyptian Vultures exhibited a high level of variability at the subpopulation level and flexibility at the individual level in basic migration parameters. Subpopulations differed significantly in travel distance and straightness of migratory movements, while differences in migration speed and duration differed as much between seasons and among individuals within subpopulations as between subpopulations. The total distances of the migrations completed by individuals from the Balkans and Caucasus were up to twice as long and less direct than those in Western Europe, and consequently were longer in duration, despite faster migration speeds. These differences appear to be largely attributable to more numerous and wider geographic barriers (water bodies) along the eastern flyway. We also found that adult spring migrations to Western Europe and the Balkans were longer and slower than fall migrations. We encourage further research to assess the underlying mechanisms for these differences and the extent to which environmental change could affect Egyptian Vulture movement ecology and population trends.

Keywords: migration connectivity, Neophron percnopterus, conservation biology, movement ecology, satellite tracking, GPS, phenotypic plasticity

### INTRODUCTION

Many migratory bird populations have undergone substantial declines, mainly as a consequence of widespread expansion of human infrastructures and activities, habitat alteration, direct persecution, and climate change (Bauer et al., 2018). Migration routes vary in habitat, landscape, and atmospheric characteristics, as well as resource availability and the prevalence of threats, all of which influence migratory behavior (Mandel et al., 2011; Trierweiler et al., 2014) and survival rates (Strandberg et al., 2010; Klaassen et al., 2014). Migratory birds provide many valuable ecosystem services (Whelan et al., 2015), and population reductions that are caused by detrimental effects that occur anywhere along the flyway may have ecosystem consequences across continents (Buechley et al., 2018b). Disentangling individual- and population-level variation in migratory movements is therefore essential to understand what factors influence migrations and to predict how different species and subpopulations might respond to environmental changes (Trierweiler et al., 2014; Bauer et al., 2018).

To gain a more complete understanding of migratory systems it is valuable to evaluate variation in migratory patterns within and among individuals and subpopulations and to produce continental-scale maps of flyways and migratory networks (Trierweiler et al., 2014; Bauer et al., 2018). Two types of migratory patterns are frequently used to compare migratory populations: we refer to "migratory performance" as all metrics relating to the timing, duration, distance and straightness of all migratory journeys, and to "migratory connectivity" as the metric that quantifies the spatial proximity at which individuals migrating from the same breeding origin spend the non-breeding season (Webster et al., 2002; Cohen et al., 2018). The study of long-distance migration has benefitted from a rapid growth in individual-based tracking data with increasing spatial and temporal resolution, enabling more detailed investigation of variability and flexibility of migratory movements (López-López, 2016). Large soaring birds have been the subjects of many tracking studies, partly because their large size enabled the attachment of transmitters since early technical development (Shamoun-Baranes et al., 2003; Alarcón and Lambertucci, 2018; Sergio et al., 2019). However, while migratory patterns have been assessed for individuals from the same or proximate populations (Sergio et al., 2014; Vardanis et al., 2016; Schlaich et al., 2017; Vansteelant et al., 2017), and migratory connectivity has been evaluated for some raptor species (Martell et al., 2014; Trierweiler et al., 2014; Finch et al., 2017), relatively few studies have analyzed these patterns across large portions of a species' distribution (Mandel et al., 2011; Dodge et al., 2014; Monti et al., 2018).

We compiled a large telemetry dataset on the Egyptian Vulture Neophron percnopterus to analyze spatial and temporal variability of migratory movements within and among individuals and subpopulations. The Egyptian Vulture is distributed across southern Europe, central and southern Asia, the Middle East, and Africa, and is listed as globally Endangered due to significant population declines caused by multiple anthropogenic threats such as poisoning, and power infrastructure collisions and electrocutions (Botha et al., 2017; BirdLife International, 2019; Safford et al., 2019). The majority of individuals from northern breeding populations are long-distance migrants that overwinter in sub-Saharan Africa and the Arabian Peninsula, with juveniles often remaining in the winter range for more than a year after their first migration (López-López et al., 2014; Oppel et al., 2015; Buechley et al., 2018b). Migratory Egyptian Vultures also overwinter in regions where non-migratory breeding populations occur (BirdLife International, 2019). As soaring migrants, Egyptian Vultures rely on thermal and orographic lift (Agostini et al., 2015), and so their migratory routes are largely shaped by geographic features, particularly the avoidance of water crossings (García-Ripollés et al., 2010; Buechley et al., 2018b). As a result, individuals in Eastern Europe and Western Asia that migrate along the Red Sea Flyway pass through up to four bottlenecks in order to avoid water bodies (Buechley et al., 2018b), compared to individuals from Western Europe (e.g., France and the Iberian Peninsula) which only cross one major bottleneck at the Strait of Gibraltar (López-López et al., 2014). Although the migratory performance from eastern (Buechley et al., 2018b) and western populations (Meyburg et al., 2004; Vidal-Mateo et al., 2016) have been investigated separately, there has never been a comparative analysis of the migratory performance of Egyptian Vultures across the majority of the species' geographical range.

In this paper, we investigate migratory connectivity and variation in individual migratory performance within and among subpopulations from Western Europe, the Balkans, the Middle East, and the Caucasus Region of Western Asia. We examined (1) whether Egyptian Vultures exhibit strong or weak migratory connectivity within and among subpopulations; and (2) whether migratory performance and variability differ between subpopulations and season, while accounting for ontogenetic improvements by comparing performance among age classes (Sergio et al., 2014). Based on geography and the generalist habitat preferences of Egyptian Vultures, we predicted that migratory connectivity would be relatively low because while Egyptian Vulture movements are constrained by bottlenecks during migration, their winter distribution in the Sahel region of Africa is less constrained by geographic barriers (Finch et al., 2017). We further predicted that the distance and duration of migrations would be greatest for the Balkans subpopulation which has to negotiate several large water bodies, while it would be shorter for the Middle Eastern, Caucasian and Western European subpopulations, which only have to negotiate the Red Sea or the Strait of Gibraltar, respectively. We predicted better migratory performance and earlier spring departure in adults compared to younger individuals, due to individual improvements with increasing age (Sergio et al., 2017). Our findings contribute to general migration ecology theory and provide valuable comparisons for further investigations into why Egyptian Vulture populations have declined more rapidly in eastern Europe (Velevski et al., 2015) than in western Europe (Garcia-Ripolles and Lopez-Lopez, 2006).

# MATERIALS AND METHODS

#### Origins and Acquisition of Tracking Data

Between 2007 and 2018, 94 Egyptian Vultures were fitted with transmitters in 11 different research projects (López-López et al., 2014; Buechley et al., 2018b; Caucanas et al., 2018), with deployments in Albania (2); Armenia (3); Bulgaria (23); Djibouti (1); Ethiopia (2); France (3); Greece (9); Israel (3); North Macedonia (3); Portugal (5); Russia (4); Spain (26); and Turkey (10). The ranges of tagged individuals extended across >4,000 longitudinal kilometers from the Iberian Peninsula in western Europe to the Caucasus Region in western Asia, and ∼4,000 latitudinal kilometers from southern Europe to sub-Saharan Africa. This range covers ∼70% of the species' current global distribution and almost the entire wintering range in Sub-Saharan Africa. We classified four distinct subpopulations based on geographically distinct breeding ranges separated by long distances that inhibit demographic exchanges between subpopulations (Lieury et al., 2015): Western Europe (including birds summering in Portugal, Spain, and southern France); Balkans (Albania, Bulgaria, Greece and North Macedonia); Middle East (Israel); and Caucasus (northeastern Turkey, Armenia, Georgia, Azerbaijan, northwestern Iran, and Dagestan Province of Russia). All transmitters weighed 24–45 g, <3% of body mass, which is below the recommended limits to avoid adverse effects; (Bodey et al., 2018) and were attached using backpack or leg-loop harness systems (Mallory and Gilbert, 2008; Sergio et al., 2015). GPS fixes and associated data were acquired at temporal resolutions ranging from one location per minute to one location every 2 h with dormancy periods during night, and with GPS positional accuracy of ±18 m. Individuals' age at deployment and age at the start of each separate migration were estimated in calendar years, based on known plumage traits of different age classes, with juveniles classed as in the first calendar year, immatures as second to fifth calendar year and adults classed as sixth calendar year or older (Clark and Schmitt, 1998). Four wild-origin adults from the Balkans (n = 2), Western Europe (n = 1), and Middle East (n = 1) subpopulations were released with transmitters after short periods of rehabilitation (**Supplementary Table 2**), but these individuals did not behave unusually compared to other individuals in their subpopulations. Capture and tagging procedures were carried out in accordance with the recommendations and regulations of each country of deployment.

# Data Processing and Delineation of Migration Periods

Tracking data from each project were uploaded to the online repository movebank.org (Wikelski and Kays, 2019). Erroneous GPS fixes were removed using general purpose data filters (Douglas et al., 2012), with maximum plausible average speed set to 25 ms−<sup>1</sup> and the error radius set to 30 m. To standardize the temporal resolution of the data, we censored the data to include only the first location from each individual every day. For each individual and season (fall, spring), net squared displacement (NSD) was calculated using the adehabitatLT package (Calenge, 2006) in R statistical software (R Core Team, 2018). Plots of NSD over time for each individual and season were visually inspected to censor data from any seasons where an individual did not migrate (i.e., juveniles and immatures that stayed in Africa) or where a migration was not completed (Bunnefeld et al., 2011). We then calculated start and end dates of each individual migration by fitting non-linear models to plots of NSD over time, using the "disperser" model in the migrateR package (Bunnefeld et al., 2011; Spitz et al., 2017; Buechley et al., 2018b). These estimates were visually verified and manually refined. We identified the point at which an individual first initiated a migration as the first point at which NSD continuously increased away from the summer or winter range (**Figure S1**). We defined the end of migration as the first point at which NSD values plateaued upon reaching the winter or summer range (López-López et al., 2014; Buechley et al., 2018b). For all further analyses, only data from completed migration trajectories were used. The final dataset after processing consisted of 188 complete migrations (71 spring, 117 fall; **Figure 1**) by 60 individuals (24 tagged as

juveniles, 8 tagged as immatures and 28 tagged as adults). Of the completed migrations, 24 were completed by juveniles, 36 by immatures and 128 by adults, with the age at subsequent migrations adjusted according to the age at deployment (**Table 1**; **Supplementary Tables 1, 2**).

# Migratory Connectivity

Migratory connectivity was quantified following methods described by Trierweiler et al. (2014), whereby the summer and winter range longitudes, identified as the first and last point of the first migration trajectory of each individual, were tested for correlations to assess for connectivity between and among subpopulations. Because the wintering range of Egyptian Vultures extends across most of the African Sahel, we did not use estimates of migratory connectivity that require the a priori definition of discrete geographic areas (Cohen et al., 2018). Instead, the strength of migratory connectivity was assessed using Mantel tests as described by Ambrosini et al. (2009) using the ade4 package (Dray and Dufour, 2007) in R statistical software (R Core Team, 2018), in which the statistical significance of the Mantel correlation coefficient was determined by 9,999 random permutations (Trierweiler et al., 2014). Mantel correlation coefficients correspond to simple Pearson product moment correlation coefficients between pairwise interindividual distance matrices constructed between start and end points of individual migrations (Ambrosini et al., 2009). Values range from −1 to 1, with higher values indicating higher migratory connectivity (i.e., low levels of overlap in winter ranges of individuals from different subpopulations). These analyses were performed separately for spring and fall migrations, across all individuals and within each subpopulation. Fall and spring connectivity were analyzed separately because, in contrast to most small landbird species for which connectivity tends to be analyzed using single winter range locations (McKinnon and Love, 2018), Egyptian Vultures often move extensively in winter (López-López et al., 2014). In addition, many of the birds tracked in this study were young birds that dispersed widely in breeding and non-breeding areas, and migrations therefore did not originate from the same location where the previous migration of the same individual terminated.

# Individual-Level Migration Parameters

Migration parameters were extracted for all complete migration trajectories using the amt package (Signer et al., 2019), following procedures previously described by Abrahms et al. (2017) and Buechley et al. (2018b): start and end dates (calendar and Julian days); start and end latitudes and longitudes; migration duration (days); direct distance (Euclidean) between start and end points (km); cumulative distance (Euclidean) between start and end points, calculated as the sum of distances between each location in a migration (km); migration straightness (direct distance/cumulative distance); and migration speed (cumulative distance/migration duration). We further summarized the above migrations by subpopulation (Western Europe, Balkans, Middle East, Caucasus), age class (juvenile, immature, adult), and season (spring, fall).

# Migratory Flexibility and Repeatability

A Generalized Linear Mixed Model (GLMM) approach was used to examine which factors accounted for the most variability in migration parameters. We used the migration parameters described above as dependent variables, and entered subpopulation (Western Europe; Balkans; or Caucasus), season TABLE 1 | Median and range of migration parameters by season, age class at start of migration (Juv. =hatch year; Imm.=2nd−5th calendar year; Ad.=6th calendar year and older) and subpopulation.


Migration start and end are the days on which migration initiated and concluded. Direct distance (km) is the Euclidean distance between summer and winter ranges, while cumulative distance is the sum of distances between each successive point in the migration trajectory. Migration duration (days) is the number of days spent on migration, and migration speed (km <sup>d</sup>−<sup>1</sup>) is the cumulative migration distance divided by the migration duration. Straightness is the ratio between the direct and cumulative distance. Only parameters from complete migration trajectories were included.

(spring or fall) and age at migration (in calendar years as a sixlevel factor) as fixed effects, and individual nested within year of migration as random intercept to account for non-independence in migration parameters within years and individuals. Overall, we evaluated eight different candidate models for each migration parameter: a null model and seven models including the three fixed effects and all potential additive combinations as independent predictors. We did not include the three individuals from the Middle East subpopulation in the comparative analysis of migration parameters among subpopulations because the small sample size prevented meaningful comparisons.

GLMMs were fitted with Gaussian distribution and identity link using the lme4 package in R (Bates et al., 2015), and we considered the model with the lowest Akaike Information Criterion (AIC) as the most parsimonious and present parameter estimates from that model. Two of the migration parameters (direct distance and cumulative distance) were log transformed to meet the assumptions of GLMMs (Zuur et al., 2009). Models were compared using the maximum likelihood estimation and were validated by checking for homoscedasticity and normality of the residuals. To that end, relevant model diagnostic graphs were computed (residuals against fitted values, residuals against each explanatory variable, histogram of residuals and normality Q-Q plots). We computed marginal and conditional R<sup>2</sup> following Nakagawa and Schielzeth (2013) using the piecewiseSEM R package (Lefcheck, 2016) to assess the overall explanatory power of the model (i.e., for fixed and random effects separately). All computations were performed in R version 3.5.1 (R Core Team, 2018).

To quantify the variation in migration parameters among populations, we estimated repeatability of migration parameters as an estimate of the fraction of total variance (sum of betweenand within-population variation) that scales from 0 to 1, with 0 indicating that all the variance is within a population, and 1 indicating that all the variance is between populations (Bell et al., 2009; Nakagawa and Schielzeth, 2010). Repeatability was estimated with the R package rptR (Stoffel et al., 2017), using the fixed factors supported by the above GLMM analysis to account for seasonal or age variation in the data. We concluded that there was significant repeatability of migration parameters within subpopulations if 95% confidence intervals of repeatability estimates did not overlap zero.

Additionally, we calculated the width of the migration corridor for each subpopulation in order to provide a measure of route flexibility (López-López et al., 2014). The width of the migration corridor was measured by computing the linear distance of the maximum longitudinal separation (i.e., East-West) of individual tracks at 5◦ latitude intervals from 15◦N to 40◦N, encompassing the full latitudinal range cover by both fall and spring migrations. The width of migration corridors was computed for the complete dataset of migration tracks and also for spring and fall seasons, separately. Computations were done in ArcMap 10.0 (ESRI, 2014) using the World Latitude and Longitude 1 × 1 degree Grid (available at https://www.arcgis.com/home/item. html?id=f11bcdc5d484400fa926dcce68de3df7). We compared the width of the migration corridors between seasons and subpopulations using a Mann Whitney and a Kruskal-Wallis test, respectively.

## RESULTS

#### Migratory Connectivity

There was high correlation between summer and winter longitudes and high Mantel test scores (>0.88) across all individuals, indicating very high migratory connectivity at the species level (**Table 2**; **Figure 2**). However, within each subpopulation, insignificant correlations and negative Mantel scores indicated very low migratory connectivity. The Balkans subpopulation exhibited the widest range of winter longitudes, overlapping with Middle East and Caucasus subpopulations. Migration routes entered 44 different countries, with the Balkan subpopulation entering the most (28 countries), followed by Caucasus (20), Western Europe (12) and Middle East (6; **Figure 1**). Complete fall migrations finished in four countries for the Balkans (Chad = 16/30; Ethiopia = 7/30; Sudan = 5/30; Yemen = 2/30 migrations) and Western Europe (Mauritania = 39/54; Mali = 11/54; Senegal = 3/54; The Gambia = 1/54 migrations), with Caucasus fall migrations mainly finishing in Ethiopia (23/30 migrations) and the three fall migrations from the Middle East ending in Chad (1) and Sudan (2), with spring departures following similar patterns (**Figure 2**).

### Individual-Level Migration Parameters

The Balkan subpopulation migrations were the least straight and longest in terms of duration and total cumulative distance, whereas those completed by individuals from the Caucasus subpopulation were longest in terms of direct distance (**Table 1**; **Figure 3**). Migrations completed by individuals from Western Europe were the straightest and shortest (**Table 1**; **Figure 3**). Spring migrations were longest in duration for adults from the Balkans subpopulation, and started later for both the Balkans (median start date for adults = 8th March; median duration for adults = 31 days) and Caucasus (median start date for adults = 22nd March; median duration for adults = 18 days) subpopulations, compared to Western Europe (median start date for adults = 26th February; median duration for adults = 21 days **Table 1**; **Figure 3**). Fall migrations started on similar dates (between 20th July and 9th October) among subpopulations, but lasted, on average, 6 and 8 days longer for individuals from the Balkans compared to those from the Caucasus and Western Europe, respectively. Migration speeds were fastest for the Caucasus, then Balkans, with individuals from Western Europe migrating more slowly. Adults from all subpopulations migrated slower in spring than fall, covering on average 71, 13, or 56 km less per day when migrating to the Balkans, Caucasus, or Western Europe, respectively (**Table 1**; **Figure 3**). Exploratory analyses confirmed that multi-day stopovers occurred very rarely among all subpopulations (López-López et al., 2014; Buechley et al., 2018b).

At the subpopulation level, adults from Western Europe demonstrated higher migration efficiency than juveniles, traveling faster along straighter and shorter (in distance and duration) migration routes. In the Balkans, straightness of


TABLE 2 | Mantel correlation coefficients of migratory connectivity and R-squared correlation coefficients between longitudes of migration start and end locations from the first completed migration of each individual in each season.

migration routes did not differ among age classes, but adults migrated more quickly (**Table 1**; **Figure 3**). Adults from the Caucasus subpopulation appeared to travel faster and longer distances than juveniles in fall, but this comparison lacked power because of the small sample size of complete juvenile migrations (**Table 1**; **Figure 3**). Immature individuals from all subpopulations started spring migration later than adults, with the greatest difference observed in the Balkan subpopulation (difference between adult and immature median start date = 53 days; **Table 1**; **Figure 3**). Although fall departure dates were similar among age classes in Western Europe and the Caucasus, immatures departed earlier than adults and juveniles in the Balkans. Overall, adults and immatures tended to complete fall migration earlier than juveniles across all subpopulations.

The two fall migrations completed by juveniles from the Middle East subpopulation were comparable to juvenile fall migrations in Western Europe in terms of cumulative distance [median (min-max) = 3297 (2,987–3,607) km] and straightness [0.809 (0.731–0.887)], but were faster [speed = 200 (150– 249) km d−<sup>1</sup> ] and shorter in duration [18 (12–24) days)] and direct distance (2,642 (2,635–2,649) km]. The fall migration parameters for the single adult individual from the Middle East were similar to the two juveniles (cumulative distance = 3,600 km; direct distance = 2,999 km; straightness = 0.83). The fall departure dates for the three Middle East migrations were similar to the other subpopulations (30th August (29th August−19th September).

#### Migratory Flexibility and Repeatability

The most parsimonious models for all migratory parameters included subpopulation as a factor, indicating that there were differences among these geographic subpopulations in migration distance, straightness, duration, start and end dates, and migration speed (**Supplementary Table 3**). Start and end dates, duration and speed also varied among seasons and age groups (**Table 3**). Cumulative distance and straightness values only

FIGURE 3 | Boxplots showing median and inter-quartile range of Egyptian Vulture migration parameters by season, age class (Juv. = hatch year; Imm. = 2nd−5th calendar year; Ad. = 6th calendar year and older), and subpopulation. (A) Direct distance is the distance between summer and winter ranges; (B) cumulative distance is the summed distances between each successive point in the migration trajectory; (C) migration speed (km d−<sup>1</sup> ) is the cumulative migration distance divided by the migration duration; (D) straightness is the ratio between the direct and cumulative distance; migration start (E,G) and end (F,H) are the days on which migration initiated and concluded; and (I) migration duration is the number of days spent on migration. Orange and green bars indicate fall and spring migrations, respectively. Only parameters from complete migration trajectories were included (refer to Supplementary Table 1 for sample sizes).

varied among subpopulations and seasons. While the models for start and end dates, cumulative distance and direct distance explained most (>80%) of the variability in the data, variation in speed and travel duration was poorly captured (<60%) by our three predictor variables (**Table 3**).

There was high repeatability (r > 0.5; **Table 3**) within subpopulations in the three route-related migration parameters, cumulative travel distance, direct distance, and straightness. This confirmed that there was more variation in these parameters between than within each subpopulation. We found no significant repeatability for duration (r = 0.120; 95% CI = 0–0.356; **Table 3**) or speed (r = 0.270; 95% CI = 0–0.603), indicating that there is large variability within each subpopulation.

The mean width of seasonal migration corridors for each subpopulation between 15◦N and 40◦N ranged from 802 ± 598 km at 35◦N to 1,429 ± 1,041 km at 20◦N. The maximum East-West separation of individual routes was recorded between 15◦ and 25◦N, both in autumn and spring migrations (**Table 4**), which approximately coincides with the latitudes of the Sahara and Arabian Deserts. Significant differences in the width of migration corridors were observed among subpopulations (Kruskal-Wallis test: H2,36 = 13.84, p < 0.001), with the Balkan subpopulation exhibiting an average migration corridor width 2.59 and 4.39 times larger than Western Europe and Caucasus subpopulations, respectively (mean ± SD corridor width for Balkans = 1,970 ± 859 km; Western Europe = 818 ± 446 km; Caucasus = 611 ± 287 km). No significant differences in the width of migration corridors were observed between seasons when data from different latitudes were pooled together (Mann-Whitney test: U = 164.00, Z = 0.363, p = 0.732). However, for the Western Europe subpopulation, the fall migration corridors at 15◦N and 20◦N (i.e., the Sahel and southern Sahara) were 1.68 and 2.02 times wider, respectively, than the spring migration corridors at the same latitudes, whereas the opposite was observed for the Caucasus subpopulation, with the spring migration corridors being >4 times larger than the fall migration corridors at those latitudes (**Table 4**). For the Balkans subpopulation, the much wider fall migration corridors at 35◦N and 40◦N were due to the single journeys of extreme easterly and westerly routes by a juvenile and an immature. Similarly, the wide spring migration corridor for the Western Europe subpopulation at 35◦N (i.e., south of the Mediterranean) was due to eastwards movements of a single immature individual, with similar widths recorded in fall (104 km) compared to spring (154 km) when that outlier was removed.

#### DISCUSSION

Several studies have described the migration of Egyptian Vultures along the western European-West African flyway (García-Ripollés et al., 2010; López-López et al., 2014) and along the Eurasian-East African flyway (Oppel et al., 2015; Buechley et al., 2018a,b). Our synthesis highlights that there is very little overlap in the wintering destinations between the western and eastern subpopulations of Egyptian Vultures in Europe, but that individuals from the Balkans, the Middle East and Central Asia often converge around the Horn of Africa, where major concentrations occur during migration (Welch and Welch, 1988; Buechley et al., 2018b) and in winter (Arkumarev et al., 2014). The different destinations and routes of the subpopulations also result in substantial differences in distance and duration of migrations, with birds from the Balkans performing the most convoluted and longest migrations, which can be twice as long as the relatively straight migratory routes of birds from Western Europe.

The results suggest that the key reason for the different migration distances is the presence of water barriers, which soaring raptors are generally reluctant to cross due to limited thermal uplift (Panuccio et al., 2012; Agostini et al., 2015). While the western population can cross the relatively short Strait of Gibraltar (Martín et al., 2016), birds from the Balkans tend to detour around the Mediterranean and the Red Sea (Oppel et al., 2015; Buechley et al., 2018b). Once these barriers have been negotiated, individuals may spread out, or travel along coastlines, depending on the geography and direction of travel. However, a more comprehensive assessment of the effects of other environmental conditions, such as wind (Vansteelant et al., 2017), thermal uplift (Duriez et al., 2018; Rotics et al., 2018) and human development (Tucker et al., 2018) is required to investigate the relative importance of these different factors in shaping migration routes for the different subpopulations.

#### Migratory Connectivity

We found a weak and insignificant Mantel correlation within subpopulations, indicating weak migratory connectivity at the subpopulation scale, as reported by Finch et al. (2017). However, we show that connectivity is relatively strong at the continental scale, with no overlap between the western and two eastern subpopulations during winter in Africa, and only moderate overlap in the Horn of Africa between the Balkan and Caucasian subpopulations (Trierweiler et al., 2014). High connectivity is uncommon for species with large non-breeding range spread (Finch et al., 2017), but our results indicate that even very widespread species such as the Egyptian Vulture can have reasonably strong migratory connectivity at large spatial scales (Trierweiler et al., 2014). Therefore, our results highlight that migratory connectivity is dependent on the spatial scale of analysis and that caution is required when assessing and interpreting connectivity for widespread species if comparisons are based on individuals from a relatively small or spatially biased portion of the species' range (Trierweiler et al., 2014; Finch et al., 2017; Cohen et al., 2018).

The population spread of wintering areas was greatest for the Balkan subpopulation, despite being the smallest of the subpopulations studied here (Velevski et al., 2015). The larger non-breeding range spread of the Balkan population may be caused by vultures bypassing the Mediterranean Sea on the eastern border and then bifurcating around the Red Sea, with some individuals continuing south through the Arabian Peninsula, while others traveled southwest via Egypt and across the Sahara. Conversely, both the Western Europe and Caucasus subpopulations are only constrained by bottlenecks at the


TABLE 3 | Generalized Linear Mixed Model (GLMMs) results of the most parsimonious models fitted for migration parameters of Egyptian Vultures tracked by GPS telemetry across three continents. AICw indicates the weight of evidence for this model among eight candidate models.

Repeatability of migration parameters is an estimate of the fraction of total variance (sum of between- and within-population variation) that scales from 0–1, with 0 indicating that all the variance is within a population, and 1 indicating that all the variance is between populations.

TABLE 4 | Width (km) of the migration corridor at 5◦ latitude intervals for 60 Egyptian Vultures tracked by GPS telemetry across three continents.


Asterisks (\*) indicate where spring and fall migration routes differed by at least a factor of two, indicating that either spring or fall migration is much more constrained.

Strait of Gibraltar and the Bab-el-Mandeb Strait, near their breeding and wintering ranges, respectively, which may reduce the range spread of these subpopulations. Longer migration distance and greater migratory spread have been associated with population declines of species that migrate using the Afro-Palearctic flyway, possibly as a result of uneven distribution of anthropogenic threats associated with uneven human population growth and development (Patchett et al., 2018). Although further work is required to assess variability in mortality patterns and demographic effects among the different subpopulations, the longer migrations and greater migratory spread for the Balkans subpopulation could partially explain faster declining populations compared to the other subpopulations (Velevski et al., 2015).

#### Migratory Flexibility

We found relatively high repeatability within subpopulations for distance and for straightness of travel, but much lower repeatability for duration and speed. The variation in duration and speed may be the result of varying environmental conditions and stopover use during each migratory journey (Vansteelant et al., 2015; Kölzsch et al., 2016; Vardanis et al., 2016; Monti et al., 2018) and, although multi-day stopovers are rare in Egyptian Vultures (López-López et al., 2014; Buechley et al., 2018b), further detailed investigation of both aspects is required. Greater speed during spring migration than fall migration has been hypothesized to be the result of a heightened drive to arrive on breeding grounds, and has been recorded in many species of soaring migrants (Alerstam, 2003; Nilsson et al., 2013). Greater migration speed in spring can also be a consequence of greater wind assistance (Bauchinger and Klaassen, 2005; Kemp et al., 2010), although fitness costs of early arrival due to less favorable atmospheric conditions during migration have been recorded in some species (Rotics et al., 2018). However, just as for other species (Schmaljohann, 2018), our data suggest that adult Egyptian Vultures migrate faster in fall than spring, although this effect was less pronounced for the Caucasus subpopulation. However, the spring migration of birds from the Caucasus subpopulation was slightly longer in both duration and distance, as most birds migrated to the west of the Red Sea in spring and therefore traveled farther compared to fall migration east of the Red Sea (Buechley et al., 2018b), explaining the wider spring migration corridor at those latitudes and emphasizing the importance of water barriers in shaping the migratory movements for the species. In contrast, the spring migration corridor between the Sahel and Sahara for the Western Europe subpopulation was half the width of the fall corridor at the same latitudes, likely due to the selection of more westerly migration routes in response to wind conditions (Vidal-Mateo et al., 2016). There is ongoing debate about the relative importance of innate motivation and external factors (e.g., wind) in causing seasonal differences in migration speed (Lindström et al., 2019). Although we found marked differences in route choice between seasons, but inconsistent differences in performance, much more work is needed to determine how various innate and external factors contribute to the development of seasonal and population specific migration patterns, not only for Egyptian Vultures but all migratory species (Schmaljohann, 2018). Similar to other raptors, we also found some age-related differences in migration distance, duration, speed and timing (Sergio et al., 2014), with adults traveling faster along shorter routes, and departing earlier in spring, than younger birds (Monti et al., 2018), although the patterns were not consistent in all subpopulations. While our dataset did not allow a full assessment of changes in individual migratory performance with age (sensu Sergio et al., 2014), our findings are consistent with expectations that individual raptors must improve their migratory performance in early life to eventually be recruited into the breeding population (Sergio et al., 2017).

Although further work is required to assess the effects of environmental factors on migratory movements of Egyptian Vultures, the variability within and among each subpopulation indicates that they could potentially respond to short term changes in environmental conditions along their flyway which could eventually affect migration phenology (Both, 2010; Klaassen et al., 2014). However, the migration corridor for all subpopulations was widest over the Sahara desert, where conditions for soaring migrants may be harsh during extreme weather conditions (Strandberg et al., 2010; Vansteelant et al., 2017). Although juveniles and immature individuals may be particularly vulnerable during Sahara crossings, adults also demonstrate aberrant behaviors there, sometimes resulting in carry-over effects on breeding success (Strandberg et al., 2010). Our results show that the spring and fall migration corridors for the Balkans subpopulation are 1.81 and 1.46 times wider over the Sahara than for the Western Europe subpopulation, respectively (**Table 4**). Suboptimal route selection, possibly due to limited conspecific guidance because of recent rapid population declines (Velevski et al., 2015), may result in higher mortality rates of juvenile Egyptian Vultures from the Balkans during fall migration when they attempt fatal sea crossings (Oppel et al., 2015). Although the effects of different migration strategies and route selection on Egyptian Vulture survival require further investigation at the subpopulation level, our results suggest that individuals from the Balkans use migration routes that may expose them to a broader range of different threats and migration conditions than individuals from Western Europe or the Caucasus (Patchett et al., 2018).

# FUTURE DIRECTIONS

This study provides the foundations for further investigation into the underlying causes of variation in migration strategies of Egyptian Vultures and the potential effects on individual survival and fitness, and ultimately population dynamics. We encourage future research to investigate the effects of environmental factors on migratory movements and to evaluate whether the different levels of anthropogenic threats encountered along the flyways used by different subpopulations could explain differences in population trends in breeding regions. A potential approach to resolve such differences would be more intensive study of resident populations of Egyptian Vultures in sub-Saharan Africa and the Middle East, and quantification of the trade-offs and benefits of migratory vs. resident lifestyles (Sanz-Aguilar et al., 2015). With recent tagging of Egyptian Vultures within wintering ranges (Buechley et al., 2018a; McGrady et al., 2018) this may soon be possible to explore in more detail, enabling a comprehensive comparison of movement strategies in relation to human activity (Tucker et al., 2018). Furthermore, although our dataset did not enable the investigation of the ontogeny of migration in Egyptian Vultures [e.g., Scott et al. (2014)], future analysis of movement data derived from individuals tracked from juvenile to breeding adult status will provide a clearer understanding of the development of migration strategies and the variation within and among individuals as they age. Finally, this study illustrates that broad-scale collaboration can contribute to overcoming one of the grand challenges of migration research by enabling the mapping of flyways at a continental scale (Bauer et al., 2018), with the ultimate aim of informing strategies to protect threatened species based on a sound understanding of their movement ecology (Fraser et al., 2018; Choi et al., 2019).

# DATA AVAILABILITY

The data analyzed in this study were obtained from tracking projects listed in the Movebank database (https://www. movebank.org/). Requests to access these datasets should be directed to the contact persons and principal investigators listed for each individual Movebank project, with contact details available from the corresponding author.

# ETHICS STATEMENT

All procedures were carried out in accordance with the recommendations and regulations of each country of deployment.

# AUTHOR CONTRIBUTIONS

EB, PL-L, WP and SO conceived and designed the study, collected and collated data, performed the statistical analyses and led the writing of the manuscript. All other co-authors contributed to the conception of the individual tracking projects, acquisition of the data and writing of the manuscript.

# FUNDING

Balkans and Caucasus data: This work was financially supported by the LIFE+ projects LIFE10 NAT/BG/000152 and LIFE 16 NAT/BG/000874 funded by the European Union and cofunded by the AG Leventis Foundation and MAVA, the US National Science Foundation, the Christensen Fund, National Geographic Society, the Whitley Fund for Nature, Faruk Yalçin Zoo and Kuzey Doga's donors (in particular Bilge Bahar, Devrim Celal, Seha I¸smen, Lin Lougheed, Burak Över, and Batubay Özkan). We are grateful to Turkey's Ministry of Forestry and Water Affairs General Directorate of Nature Conservation and National Parks and NorthStar Science and Technology for donating three transmitters each. State Nature Reserve Dagestanskiy and Russian Raptor Research and Conservation Network supported work in Dagestan. Western Europe data: deployments of transmitters in Portugal were funded by the EUfunded LIFE Rupis project (LIFE14 NAT/PT/00855); SALORO S.L.U. funded the deployment of transmitters in the Duero region of Spain; DREAL Nouvelle-Aquitaine—Fondation d'entreprises Barjane funded deployments in France; JE was supported by Basque government predoctoral grant (grant number: 569382696); GREFA (Grupo para la Rehabilitación de la Fauna Autóctona y su habitat)-Endangered Species Monitoring Project together with Poison Sentinels Project of WWF/Spain. The Migra Program of SEO/BirdLife (www.migraciondeaves.org/en/) deployed transmitters in collaboration with Fundación Iberdrola España, and were funded by La Rioja Regional Government in La Rioja, and Fundación Hazi and Diputación Foral de Gipuzkoa within the Interreg POCTEFA-ECOGYP project in Gipuzkoa.

#### ACKNOWLEDGMENTS

We dedicate the article to the memory of Michele Panuccio and his passion for and expertise in migration ecology. We also thank Michele and the other reviewers, Wouter Vansteelant and Tom Finch, for their positive and constructive comments which improved the article.

Balkans and Caucasia data: We thank our collaborators including Kuzey Doga Society (Turkey), Igdir Directorate of Nature Conservation and National Parks (Turkey), American University of Armenia, Ethiopia Wildlife Conservation Authority, Ethiopia Wildlife and Natural History Society, and our colleagues who assisted with Egyptian vulture trapping, including Emrah Çoban, Lale Aktay, Kayahan Agirkaya, Berkan Demir, Mete Türkoglu (Turkey); Karen Aghababyan, Anush Khachatrian, Garo Kurginyan (Armenia); Sisay Seyfu, Alazar Daka Rufo, Yilma Dellelegn Abebe, Girma Ayalew (Ethiopia); Svetoslav Spasov, Ivaylo Angelov, Saniye Mumun, Dobromir Dobrev, Tsvetomira Angelova, Stoycho Stoychev, (Bulgaria); Ewan and Jenny Weston, Emil Yordanov who assisted with Egyptian vulture trapping in Bulgaria Vasilis Sideris, Giannis Chondros, Christos Lambris, Antonis Vroikos, Dimitris Vavylis,

#### REFERENCES


Angelos Evangelidis (Greece), Mirjan Topi (Albania), Metodija Velevski and Zlatko Angeleski (North Macedonia). State Nature Reserve "Dagestanskiy" and Russian Raptor Research and Conservation Network supported work in Dagestan. Western Europe data: For deployments in Spain we thank the following people and organizations for assisting with field work, capture of the vultures and all other aspects of the study: L. Bolonio, J. de Lucas, V. Garcíia, R. Ibanez, M. Nieto, and A. Vela (Castellón and Guadalajara provinces, Spain); Luis Lopo, Ignacio Gámez, Francisco Javier Robres, Sandra Vela, Lidia Crespo, Diego García, Sergio Mikolta, Miguel Ángel Elvira, Carlos Fernández, Sara Josefa Herrero, Álvaro Alonso, Eduardo Miera, Miguel Ángel Marín and José Francisco Pedreño (La Rioja Goverment, Logroño province, Spain); José María Fernández, Iñigo Mendiola, Ibai Aizpuru, Fermín Ansorregi, Aitor Galdos, Aitor Lekuona, Mikel Olano, Jon Ugarte and Javier Vázquez (Fundación Hazi and Diputación Foral de Gipuzkoa, Guipúzcoa province, Spain); Saloro S.L.U. generously contributed data from the Duero region of Spain; the GREFA veterinary team and all who contributed to GREFA's project. For deployments in Portugal we thank EU-funded LIFE Rupis project partners (Sociedade Portuguesa para o Estudo da Aves; Associação Transumância e Natureza; Associação de Conservação da Natureza e do Património Rural; Guarda Nacional Republicana, Portugal; Fundación Patrimonio Natural de Castilla y León; EDP Distribuição—Energia SA; Instituto da Conservação da Natureza e Florestas, Portugal; Junta de Castilla y León, Spain; the MAVA Foundation); and the collaboration of Víctor García Matarranz (Ministerio de Agricultura y Pesca, Alimentación y Medio Ambiente, Spain). Parc National des Pyrénées—La salsepareille supported work in France. Middle East data: We thank the Israel Nature and Parks Authority (INPA) for their support of this project and specifically to Ohad Hatzofe and Ygal Miller for leading the conservation of vultures in Israel. We also thank Walter Nesser, for voluntarily assisting with field work.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00323/full#supplementary-material


Clark, W., and Schmitt, N. (1998). Ageing Egyptian vultures. Alula 4, 122–127.


Dray, S., and Dufour, A. (2007). The ade4 Package: implementing the duality diagram for ecologists. J. Stat. Softw. 22, 1–20. doi: 10.18637/jss.v022.i04

Duriez, O., Peron, G., Gremillet, D., Sforzi, A., and Monti, F. (2018). Migrating ospreys use thermal uplift over the open sea. Biol. Lett. 14:20180687. doi: 10.1098/rsbl.2018.0687

ESRI (2014). "ArcMap". 10.2.2 ed. Oaklands, CA: ESRI.


**Conflict of Interest Statement:** IK was employed by company Sibecocentar LLC, Russia; IA was employed by SALORO, Spain; and JJ was employed by Oriolus Ambiente e Eco Turismo LDA, Spain.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The reviewer, TF, declared a shared affiliation, with no collaboration, with one of the authors, SO, to the handling editor at the time of review.

Copyright © 2019 Phipps, López-López, Buechley, Oppel, Álvarez, Arkumarev, Bekmansurov, Berger-Tal, Bermejo, Bounas, Alanís, de la Puente, Dobrev, Duriez, Efrat, Fréchet, García, Galán, García-Ripollés, Gil, Iglesias-Lebrija, Jambas, Karyakin, Kobierzycki, Kret, Loercher, Monteiro, Morant Etxebarria, Nikolov, Pereira, Peške, Ponchon, Realinho, Saravia, ¸Sekercioglu, Skartsi, Tavares, Teodósio, ˘ Urios and Vallverdú. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# One Hundred Pressing Questions on the Future of Global Fish Migration Science, Conservation, and Policy

Robert J. Lennox <sup>1</sup> \*, Craig P. Paukert 2,3, Kim Aarestrup<sup>4</sup> , Marie Auger-Méthé<sup>5</sup> , Lee Baumgartner <sup>6</sup> , Kim Birnie-Gauvin<sup>4</sup> , Kristin Bøe<sup>7</sup> , Kerry Brink 8,9 , Jacob W. Brownscombe10,11, Yushun Chen12,13, Jan G. Davidsen<sup>14</sup>, Erika J. Eliason<sup>15</sup> , Alexander Filous <sup>16</sup>, Bronwyn M. Gillanders <sup>17</sup>, Ingeborg Palm Helland<sup>18</sup> , Andrij Z. Horodysky <sup>19</sup>, Stephanie R. Januchowski-Hartley <sup>20</sup> , Susan K. Lowerre-Barbieri 21,22, Martyn C. Lucas <sup>23</sup>, Eduardo G. Martins <sup>24</sup> , Karen J. Murchie<sup>25</sup>, Paulo S. Pompeu<sup>26</sup>, Michael Power <sup>27</sup>, Rajeev Raghavan28,29 , Frank J. Rahel <sup>30</sup>, David Secor <sup>31</sup>, Jason D. Thiem<sup>32</sup>, Eva B. Thorstad<sup>18</sup>, Hiroshi Ueda<sup>33</sup> , Frederick G. Whoriskey <sup>34</sup> and Steven J. Cooke<sup>11</sup>

<sup>1</sup> Laboratory for Freshwater Ecology and Inland Fisheries, NORCE Norwegian Research Centre, Bergen, Norway, <sup>2</sup> U.S. Geological Survey, Missouri Cooperative Fish and Wildlife Research Unit, Columbia, MA, United States, <sup>3</sup> The School of Natural Resources, University of Missouri, Columbia, SC, United States, <sup>4</sup> Section for Freshwater Fisheries and Ecology, Technical University of Denmark, Silkeborg, Denmark, <sup>5</sup> Department of Statistics, Institute for the Oceans and Fisheries, The University of British Columbia, Vancouver, BC, Canada, <sup>6</sup> Institute for Land, Water and Society, Charles Sturt University, Albury, NSW, Australia, <sup>7</sup> Department of Ocean Sciences, Memorial University, St. John's, NL, Canada, <sup>8</sup> World Fish Migration Foundation, Groningen, Netherlands, <sup>9</sup> School of Life Sciences, College of Agriculture, Engineering and Science, University of KwaZulu-Natal, Pietermaritzburg, South Africa, <sup>10</sup> Fish Ecology and Conservation Physiology Laboratory, Department of Biology, Carleton University, Ottawa, ON, Canada, <sup>11</sup> Department of Biology, Dalhousie University, Halifax, NS, Canada, <sup>12</sup> Institute of Hydrobiology and State Key Laboratory of Freshwater Ecology and Biotechnology, Chinese Academy of Sciences, Wuhan, China, <sup>13</sup> University of Chinese Academy of Sciences, Beijing, China, <sup>14</sup> Department of Natural History, NTNU University Museum, Trondheim, Norway, <sup>15</sup> Department of Ecology, Evolution and Marine Biology, University of California, Santa Barbara, Santa Barbara, CA, United States, <sup>16</sup> Department of Environmental Conservation, University of Massachusetts Amherst, Amherst, MA, United States, <sup>17</sup> Environment Institute and School of Biological Sciences, University of Adelaide, Adelaide, SA, Australia, <sup>18</sup> Norwegian Institute for Nature Research, Trondheim, Norway, <sup>19</sup> Department of Marine and Environmental Science, Hampton University, Hampton, VA, United States, <sup>20</sup> Department of Biosciences, Swansea University, Swansea, United Kingdom, <sup>21</sup> Fisheries and Aquatic Science Program, School of Forest Resources and Conservation, University of Florida, Gainesville, FL, United States, <sup>22</sup> Florida Fish and Wildlife Conservation Commission, Florida Fish and Wildlife Research Institute, St. Petersburg, FL, United States, <sup>23</sup> Department of Biosciences, University of Durham, Durham, United Kingdom, <sup>24</sup> Ecosystem Science and Management Program, University of Northern British Columbia, Prince George, BC, Canada, <sup>25</sup> Daniel P. Haerther Center for Conservation and Research, John G. Shedd Aquarium, Chicago, IL, United States, <sup>26</sup> Biology Department, University of Lavras, Lavras, Brazil, <sup>27</sup> Department of Biology, University of Waterloo, Waterloo, ON, Canada, <sup>28</sup> Department of Fisheries Resource Management, Kerala University of Fisheries and Ocean Studies (KUFOS), Kochi, India, <sup>29</sup> Mahseer Trust, Freshwater Biological Association, Wareham, United Kingdom, <sup>30</sup> Department of Zoology and Physiology, University of Wyoming, Laramie, WY, United States, <sup>31</sup> Chesapeake Biological Laboratory, University of Maryland Center for Environmental Science, Cambridge, MD, United States, <sup>32</sup> Department of Primary Industries, Narrandera Fisheries Centre, Narrandera, NSW, Australia, <sup>33</sup> Field Science Center for Northern Biosphere, Hokkaido University, Hokkaido Aquaculture Promotion Cooperation, Sapporo, Japan, <sup>34</sup> Ocean Tracking Network, Dalhousie University, Halifax, NS, Canada

Migration is a widespread but highly diverse component of many animal life histories. Fish migrate throughout the world's oceans, within lakes and rivers, and between the two realms, transporting matter, energy, and other species (e.g., microbes) across boundaries. Migration is therefore a process responsible for myriad ecosystem services. Many human populations depend on the presence of predictable migrations of fish for their subsistence and livelihoods. Although much research has focused on fish migration, many questions remain in our rapidly changing world. We assembled a

Edited by:

Nathan R. Senner, University of South Carolina, United States

#### Reviewed by:

Lucille Chapuis, University of Exeter, United Kingdom Daniel Cardoso Carvalho, Pontifícia Universidade Católica de Minas Gerais, Brazil

\*Correspondence:

Robert J. Lennox robertlennox9@gmail.com; role@norceresearch.no

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 29 March 2019 Accepted: 15 July 2019 Published: 19 August 2019

#### Citation:

Lennox RJ, Paukert CP, Aarestrup K, Auger-Méthé M, Baumgartner L, Birnie-Gauvin K, Bøe K, Brink K, Brownscombe JW, Chen Y, Davidsen JG, Eliason EJ, Filous A, Gillanders BM, Helland IP, Horodysky AZ, Januchowski-Hartley SR, Lowerre-Barbieri SK, Lucas MC, Martins EG, Murchie KJ, Pompeu PS, Power M, Raghavan R, Rahel FJ, Secor D, Thiem JD, Thorstad EB, Ueda H, Whoriskey FG and Cooke SJ (2019) One Hundred Pressing Questions on the Future of Global Fish Migration Science, Conservation, and Policy. Front. Ecol. Evol. 7:286. doi: 10.3389/fevo.2019.00286 diverse team of fundamental and applied scientists who study fish migrations in marine and freshwater environments to identify pressing unanswered questions. Our exercise revealed questions within themes related to understanding the migrating individual's internal state, navigational mechanisms, locomotor capabilities, external drivers of migration, the threats confronting migratory fish including climate change, and the role of migration. In addition, we identified key requirements for aquatic animal management, restoration, policy, and governance. Lessons revealed included the difficulties in generalizing among species and populations, and in understanding the levels of connectivity facilitated by migrating fishes. We conclude by identifying priority research needed for assuring a sustainable future for migratory fishes.

Keywords: ecosystem services, ichthyology, habitat connectivity, partial migration, conservation, ecology

#### INTRODUCTION

Migration is an adaptive and widely expressed behavior within the animal kingdom. Species' movements among habitats, whether by solitary individuals or as synchronized collective displacements by many animals, facilitate exploitation of patchy and seasonally variable resources, which is key to species' reproduction and persistence (Baker, 1978; Jørgensen et al., 2008). As the most speciose classes of vertebrates, fishes provide an excellent focal group for the study of the evolution and ecology of migration (Lucas and Baras, 2001). Among fishes, a taxonomy of migration types exist for species moving between, and within, marine and freshwater environments (e.g., diadromy, oceanodromy potamodromy; Myers, 1949).

Migration is ecologically important, but also a behavior that is under significant threat worldwide (Wilcove and Wikelski, 2008). Animal migrations connect ecosystems and transport matter and energy long distances—faster than would be conveyed by wind, currents, or tides. Carbon, nutrients, and pathogens carried in the bodies of migratory animals have been shown to make substantial contributions to recipient ecosystems (Naiman et al., 2002; Hall et al., 2012; Childress and McIntyre, 2015). Therefore, assuring secure pathways among habitats is essential to support migration; a key consideration for ecosystem management (Mumby, 2006; Fuller et al., 2015). Migratory species' reliance on multiple habitat types also increases their vulnerability to human disturbances such as fragmentation caused by dams, roads, or land use change, as well as climate change and other human-mediated global changes (Wilcove and Wikelski, 2008; Secor, 2015b). There have been an increasing number of studies to better our understanding of how different human disturbances influence and affect diverse migratory species. With increasing recognition of the importance of protecting migratory fish species, there have been rapid developments in our understanding of how these stressors operate and in the appropriate mitigations needed to limit their impacts on migratory fishes (Lucas and Baras, 2001; Brink et al., 2018; Lowerre-Barbieri et al., 2019).

Despite a growing number of studies, there remains a need to identify knowledge gaps as well as plan a research agenda based on focused questions related to understanding and conserving migratory species. The goal of our paper is to identify 100 outstanding questions about the mechanisms and processes of fish migration, and human disturbances that threaten migratory species' persistence (see **Figure 1** for an overview of related topics and considerations). Our questions encompass fish species across habitats, ecologies, and taxa. We identify timely and relevant questions that, if addressed, will advance our understanding of which fish species migrate, how, when, and why they do so, and the actions required to conserve these species and the habitats that they depend on. We present nine broad themes relevant to fish migration, beginning each theme with a brief description, followed by a series of related questions to be explored. We conclude with a proposed research agenda for migratory fishes, giving emphasis to the science that has the potential to inform management and policy actions.

#### ONE HUNDRED QUESTIONS

Each author of this paper independently derived a series of questions about fish migration and shared them with the first author. Questions were then sorted into themes using the movement ecology framework from Nathan et al. (2008) that integrates internal and external drivers, navigation, and motion capacity as fundamentals of animal movement. The questions we identified relate to animal internal state (energetics, drivers, endocrinology); navigation (orientation and timing); locomotion; external drivers of migration; threats to fish migration related and unrelated to climate change, as well as environmental conservation; policy and governance related to migratory fishes, and thematic questions on the role of migration (**Figure 1**). Breakout groups around each theme synthesized and refined related questions. We present the nine themes and related questions below.

#### Internal State

A fish's internal state and its maintenance of homeostasis are regulated by a combination of abiotic and biotic stimuli, as well as interactions with genetics, morphology, life history, cognition, and physiology (Uusi-Heikkilä et al., 2008). Individual and collective behavior ultimately feed back to influence internal state, meaning that migration itself can influence the internal

responsible for minimizing these threats and implementing effective management actions, including habitat restoration that maintains connectivity and ensures

state and that internal state can in turn influence migration. For example, recent evidence suggests that stress levels and nutritional status can both impact migration distance and success in salmonids (Bordeleau et al., 2018; Birnie-Gauvin et al., 2019a). Few such studies exist, but these can shed light on the mechanistic links between internal state and migratory behavior. These synergies are especially important given ecological consequences of recent human-mediated trait changes to fish populations (Jørgensen et al., 2007; Palkovacs et al., 2012; Rahel and McLaughlin, 2018). Understanding the effects of ecosystemorganism interactions on short-term movements and longerterm migrations of fishes requires approaches that unify: a) mechanistically-driven physiological studies, b) patternoriented behavioral studies, and c) quantitatively driven fisheries sciences (Horodysky et al., 2015). Quantifying the internal state of the animal from non-lethal biopsy of blood, gill, or other tissues can allow subsequent movement patterns to be ascertained through laboratory experiments or in the field using biotelemetry for remote monitoring. These methods can advance our understanding of mechanistic linkages between hormone levels, gene expression, or other internal variables and the movement patterns exhibited by the individual. Below, we provide a series of questions to guide future inquiry into the effects of fish internal state on movement and migration:

migratory fish will persist in the future. ALAN refers to artificial light at night.


# Navigation

Fish species migrate variable distances among habitats (Lohmann et al., 2008). Wayfinding mechanisms within fishes typically depend on the extent of their migration and vary with species, life stages, and environments (Ueda et al., 1998; Ueda, 2018). Magnetic senses likely play a role in the navigation of many species (Durif et al., 2013; Putman et al., 2014) along with olfactory and visual cues (Ueda et al., 1998) and learning (Dodson, 1988; Brown and Laland, 2003). Fish populations, and in some cases species, have evolved different spatial strategies through natural selection: some have high degrees of philopatry, whereas others exhibit substantial plasticity in their movement and migration behaviors (Secor, 2015a). This phenomenon needs further detailed exploration using comparative and experimental approaches. Timing, route efficiency, and accuracy of migration are critical for fish species to arrive at their destination at the right time and with sufficient energy reserves (Cooke et al., 2006), and dispersal to new areas is also critical for population resilience, gene flow (Klemetsen, 2010), and for recolonization (Perrier et al., 2009; Radinger and Wolter, 2014). Tracking technology has allowed us to elucidate the onset, periodicity, and progress of some migrations; although many fishes are too small to be tagged this way with the size of current technology. Despite an increasing number of studies on a broader diversity of fish species (i.e., beyond salmonids) and migration types, our understanding of how different species find their way and what can affect their navigation remains limited. These limitations in knowledge influence our ability to manage or conserve species and the habitats that they depend on. We propose that key research questions for future work in the navigation theme are:


### Locomotion

Migratory fish exhibit extreme variability in their modes of locomotion, from the highly maneuverable Anguilliformes to the streamlined Thunniformes that can sustain fast swimming speeds (Sfakiotakis et al., 1999). Within a population, individuals can also vary significantly in their locomotor abilities (Reidy et al., 2000). For migratory animals, swimming capacity is often an important factor influencing success as it determines an individual's ability to pass natural and human-made barriers or surmount other challenges (Hinch and Bratty, 2000; Cooke et al., 2006). Understanding locomotion can be critical for maintaining suitable conditions for fish to migrate in habitats where humans have some degree of control over the physical environment, for example in rivers where water levels or flows are regulated and effective fish passage infrastructure may or may not have been built. Migrations can be energetically costly; therefore, efficient and judicious use of energy stores can also affect migration success (Brownscombe et al., 2017). Human activities are rapidly changing the environmental conditions in ways that challenge the physical and metabolic capabilities that fish rely upon to power themselves to their destinations (Lucas and Baras, 2001). With the existence of such variability in fish locomotory modes, abilities, migratory types, and challenges, and alterations to the environmental conditions both on and off migration routes, human-induced environmental change will generate varied and in some cases unanticipated responses amongst migratory fishes. The questions posed in this theme address key knowledge gaps relevant to how various fish species and ecosystems will be affected.


selective fishing) affect the locomotor performance of migratory species?


#### External Drivers

External drivers, along with a fish's internal state, stimulate behaviors including movement and habitat shifts to important foraging or spawning grounds that maintain or restore individual homeostasis and satisfy life history demands. Fish migrations must be properly timed on short (e.g., diurnal) and long (e.g., seasonal) timescales to optimize the balance between costs (avoiding hostile conditions) and benefits (matching distributions to abundant feeding opportunities) to maximize fitness (Dingle and Drake, 2007). Decisions to migrate or not, and when to migrate, are often regulated by abiotic external drivers (e.g., day length; Bradshaw and Holzapfel, 2007). External cues can regulate long-term physiological and morphological developments that prepare fishes for arrival into new environments, and can synchronize groups of fish to migrate under favorable conditions (e.g., lunar phases). Human disturbances, such as noise, artificial light, and dam discharge can also influence decisions to migrate, and potentially also alter timing and route choice (Reid et al., 2019). By understanding how external drivers interact with fish internal states, and how these can regulate migration, we can develop more effective actions and policies to mitigate impacts on migratory species (Bowlin et al., 2010). Key questions about the role of external drivers upon fish migration include:


#### Threats (Excluding Climate Change)

Humans aggregate around water (Fang and Jawitz, 2019) and use rivers, lakes, and oceans for drinking water, producing food, wastewater treatment, transport and trade, and in many other ways that modify or threaten the ecological integrity of these systems (Halpern et al., 2008; Vörösmarty et al., 2010; Reid et al., 2019). Chemical pollution (Hellström et al., 2016), artificial light (Longcore and Rich, 2004), noise (Filous et al., 2017), water abstraction (Benstead et al., 1999), barrier installation (Silva et al., 2018), and fishing (Jørgensen et al., 2008) all affect or have the potential to affect fish migrations. Our understanding of impacts on migratory fishes by different human disturbances is often confined to shorter-term effects (e.g., one migration cycle) and single disturbances, but we must move to evaluate the effects of disturbances across longer time periods (e.g., intergenerational impacts), and to evaluate cumulative interactions among the many human disturbances that confront different migratory fish species. Important future research questions about threats are:


# Threats From Climate Change

Human mediated climate change is establishing a future that will be characterized by temperature extremes, evaporative water losses, and more variable timing and extent of precipitation, as well as surface water levels and flow, salinity, and temperature (Alexander et al., 2006; Hoegh-Guldberg and Bruno, 2010). These changes are altering community phenology, species dynamics, and distributions (Walther et al., 2002; Lynch et al., 2016). Climate change is also altering pathogen dynamics, the impacts of which are poorly understood with respect to fish migration (Miller et al., 2014; Vollset et al., 2016). Migratory species are disproportionately influenced by these ongoing global changes compared to resident species because of their reliance on multiple geographically separated habitats that are changing at different rates and in different ways (Both and Visser, 2001; Robinson et al., 2009). For example, multiple mismatches could develop in relation to food availability and migration, including temporal mismatch from fish using traditional drivers of migration to pursue movements, but where the drivers are no longer linked to favorable conditions along the migration route, and spatial mismatch if fish move to foraging grounds that are no longer productive (Free et al., 2019). On the same note, migratory species' abilities to move, especially long distances, could buffer such effects of change, because behavior is the fastest route available for species to cope with change (Lehodey et al., 2006; Chessman, 2013). Finally, climate change can interact with other disturbances such as fishing (Ottersen et al., 2010) and dams (Secor, 2015b), and there remains a need to explore these interactions and effects on different migratory species. Key questions related to the impact of climate change upon migratory fishes are:


# Conservation Management

The management of fish migrations developed primarily to restore the free movement of fishes in systems fragmented by dams and other in-stream infrastructure (McLaughin et al., 2013). Early efforts to provide passage at dams included royal decrees to remove weirs from salmon rivers in the Magna Carta (1215)<sup>1</sup> and the installation of fish ladders in Europe in the 1800s (Orsborn, 1987). Efforts to consider passage of all species are needed to ensure ecosystem-scale conservation of migratory fish species in impacted rivers, including successful downstream passage by the young of anadromous species and the adults of catadromous species such as freshwater eels (Anguillidae; Roscoe and Hinch, 2010). Selective fish passage systems that exploit species differences in physical ability, spawning behavior, and sensitivity to various sensory stimuli are the object of much current research (Birnie-Gauvin et al., 2018; Rahel and McLaughlin, 2018; Silva et al., 2018). Efforts also exist to re-connect rivers and processes such as sediment transport to riparian zones (Hauer et al., 2018; Hohensinner et al., 2018). Similar approaches could be explored for use in marine environments used by migratory species. For example, Murchie et al. (2015) found bonefish (Albula spp.) selected a manufactured canal as a migration route to access spawning grounds, in lieu of a historical natural corridor. Expanding beyond fishways as tools for assisting fishes over built infrastructure, there remains a need to further explore complementary management actions such as temporal protected zones (see Abell et al., 2007), designating protected species, regulating habitat loss and pollution, monitoring and managing exploitation, and gearuse restrictions for fisheries. Spatial planning efforts benefit from an understanding of the resource selection, distribution, and movements of migratory species that can be disturbed by human activity (Lennox et al., 2018a). We explore key questions related to conservation management below. Key conservation and management questions relevant to migratory fishes are:


<sup>1</sup> see section 33; originally published 1215. Available online at: https://www. constitution.org/eng/magnacar.htm (accessed July 27, 2019).

preserving spawning habitat compare in their performance as management objectives with regards to desirable population outcomes?


#### Policy and Governance

Policy instruments and governance structures are fundamental to the development, implementation, and enforcement of regulations of sustainable management action and protection of nature (Gunningham et al., 1998; Lange et al., 2013). This is particularly salient for migratory organisms. As noted at the start of this paper, migratory species routinely cross ecosystem boundaries, habitats, and jurisdictions at national and international scales (Shuter et al., 2010; Cooke et al., 2012; Runge et al., 2014), which can raise challenges for policy making and management (Link et al., 2011). Because many migratory species aggregate in migratory corridors or on their spawning grounds, they are also vulnerable to spatially explicit stressors including targeted fishing, red tides, and anthropogenic disturbances such as dams, weirs, and roads (Januchowski-Hartley et al., 2013; Lascelles et al., 2014). Consequently, effective policy and governance of migratory fishes often necessitates multiple government sectors working together, such as fisheries and those that deal with energy and water resource management (see Nieminen et al., 2017). It is also established that political will to enact policy that benefits the environment (including migratory fish) depends on an engaged and vocal public (i.e., the electorate; Chhatre and Saberwal, 2005). With regards to migratory fishes, these considerations raise the following questions:


# The Role of Migration

One of the great challenges that emerges when discussing fish migration is establishing effective definitions for a process that is highly flexible. We know that many fish species are migratory, but without agreed upon definitions of what is a migration it can be difficult to identify which ones are not. Unknowns related to the ecological function of migration, for example, are challenging to unravel and we are in the early stages of identifying what role fish movements play in connecting environments, conveying carbon and nutrients, and transferring pathogenic and parasitic species (Altizer et al., 2011; Hyndes et al., 2014). Moreover, this yields further questions about the genetic consequences of migration and how to define species/populations as management units and assign responsibility for fish that cross boundaries (Dionne et al., 2008; Riccioni et al., 2010; Zeng et al., 2019). Methodological advances and new technologies may emerge to assist in answering some of these overarching questions in fish migration that will assist with addressing other, finer scale questions dealing with mechanisms. Questions about the role of migration are:


#### SYNTHESIS

Failure to understand how, why, when, and where different fishes migrate, and the consequences of migration, limits our understanding of migratory fishes and their roles in aquatic and terrestrial ecosystems. Migration is a process occurring across different spatial and temporal scales, which has many implications for understanding how species, populations, communities, and ecosystems are structured and how they interact with one another. Understanding movements is critical to determining the resource requirements of species and identifying appropriate measures for protection (Lennox et al., 2018a). Here, we worked to identify outstanding questions about migratory fish species that could provide important knowledge about these species, and support guidance for the conservation of these species. Applying the movement ecology paradigm (Nathan et al., 2008) to engage scientists from fish ecology, physiology, evolution, behavior, and environmental conservation and management yielded a diverse set of questions that will better our understanding of migratory fishes, and provide evidence and knowledge needed to guide more effective conservation decisions for these species.

It can be challenging to evaluate how preservation, restoration, or degradation related to a migratory species' habitat can also affect the broader ecosystem, including human dependencies and economic activity (see **Box 1** for some examples). A complete understanding of the diversity and functional ecology of migratory fishes is essential to making effective conservation and management decisions (Lowerre-Barbieri et al., 2019). Migration research has expanded in recent decades with increased access and application of technologies such as electronic tags, chemical and molecular tracers, acoustic imaging, telecommunications, and bioinformatics (Secor, 2015a; Lowerre-Barbieri et al., 2019). The number of taxa investigated is also expanding (see **Box 1**) and movement ecology is increasingly integrated within hydro-ecology, oceanography, and fisheries and habitat management (Hidalgo et al., 2016; Birnie-Gauvin et al., 2019b) to begin addressing fundamental questions related to migratory fish ecology and conservation. Applying the movement ecology paradigm to engage scientists from fish ecology, physiology, evolution, behavior, and environmental conservation and management yielded a strong list of questions that if answered will transform our understanding of migratory fishes and lead to better management of these species.

Genetic studies focused on the evolution of migration are needed to understand the underlying architecture resulting in variation between migratory and non-migratory species, as well as within and among migratory species (Hendry et al., 2000; Kess et al., 2019). Many species are partially migratory, having the genetic disposition to express migration depending on the environmental conditions that they experience (Olsson et al., 2006). Migratory phenotypes can respond over generations to selective pressures of the environment (Bracken et al., 2015). Rainbow trout and steelhead (Oncorhynchus mykiss), for example, are genetically the same species but with different migratory life histories and the migratory and non-migratory forms frequently coexist in coastal streams (Hecht et al., 2015). Understanding partial migration is key to unlocking information about migratory species (Pulido, 2011) and migratory behavior of hybrids can reveal how genetics and the environment contribute to migration (Kovach et al., 2015). Within migratory species, there is variation in the spatial and temporal extents of migratory behavior exhibited by individuals; Prince et al. (2017), for example, recently isolated a gene in steelhead, associated with early arrival to freshwater. Protecting genetic diversity within migratory species must be a priority given that this diversity underlies behavioral and physiological diversity that confers resilience to species. It has been increasingly demonstrated that habitat fragmentation and migration obstacles significantly, and rapidly, negatively affect biodiversity (So et al., 2006). A better understanding of how genetic isolation of distinct spawning stocks, and the phenotypic adaptations arising as a result (e.g., body shape, metabolic capabilities), is central to directing conservation efforts (e.g., Eliason et al., 2011). In turn, this BOX 1 | Cartilaginous and bony shes from many different families exhibit migrations within, and between, lakes, rivers, and oceans. Migrations have signicant ecological importance but many are poorly understood, hindering our comprehension of ecosystem functioning and our ability to conserve sh species or any associated services or values. Here we highlight some migratory sh species from around the world and refer to how answering lingering questions about their migrations should be a priority for scientists and management groups.

Bonefish are culturally and economically important coastal species that undertake spawning migrations from their neritic foraging habitats offshore to pelagic waters (Adams and Cooke, 2015). Acoustic telemetry revealed that at Anaa Atoll, French Polynesia, nearly all spawning movements of shortjaw bonefish (Albula glossodonta) occurred in two passageways on the northern end of the island where most artisanal fish traps are located (Filous et al., Unpublished Data). Movements were synchronized with the lunar cycle, but at different phases than Atlantic Albula vulpes. Although the neritic movements were well-characterized by telemetry, offshore movements remain an enigma. Additional research is needed to identify critical spawning sites and ensure that they remain unimpacted by anthropogenic development and their fisheries can be managed. It is also crucial to the understanding of larval dispersal and metapopulation connectivity. Transfer rate among populations, navigational mechanisms, and interspecific differences are critical to better understand these migrations (photo: Filous).

The Japanese grenadier anchovy Coilia nasus migrates from the Yangtze River estuary up the Yangtze River and its adjacent lakes for spawning and growth (Dou et al., 2012). Dams and sluice gates have blocked key migration routes and other human stressors such as navigation and channel modifications and overfishing have caused dramatic habitat loss for this fish species (Xue et al., 2019). Understanding locomotor capabilities of this species may be necessary to determine if they can pass sluice gates to adjacent lakes, spawn and grow successfully. Investigating internal and external drivers will also assist in predicting migration and preparing to open gates to facilitate passage. Tracking studies are also needed in order to identify how habitat requirements change with ontogenetic stage and determine whether suitable habitat can be preserved or created (photo: Chen).

For many Neotropical fish species, adults migrate upstream to spawn during the wet season and the eggs and larvae are conveyed downstream to floodplains. Spent adults undergo a return migration downstream to suitable habitats for feeding (Pompeu et al., 2012). Biotelemetry data of Prochilodus costatus in the upper São Francisco River, Brazil revealed external drivers of the migration, specifically a preference for initiating migration at the beginning of the rainy season, when river discharge is low, on days with increased water level, and at times of new or waxing moon (de Magalhães Lopes et al., 2018). After spawning, most fish returned to the same location where they were captured/released, and for those tracked for two consecutive years, both upstream and downstream migration timings occurred only a few days apart (de Magalhães Lopes et al., 2019). Such homing behavior and temporal fidelity still needs to be confirmed for most other Neotropical migratory species. However, these findings pose additional challenges to the use of (predominantly upstream-directed) fishways as a management tool in the South American context, which is already controversial (Pompeu et al., 2012; Pelicice et al., 2015; photo: Pompeu).

Arapaima arapaima is a migratory osteoglossiform fish that moves between the flooded forest in the rainy season and floodplain ponds in the dry season in several neotropical rivers, including Guyana's Essequibo watershed. The extent of its movements, migratory tendencies, and site fidelity are unknown, challenging efforts to establish protected areas, for example. The fish is threatened by overexploitation and although listed as 'unassessed' by IUCN it is a protected species in Guyana (Watson et al., 2016). Illegal fishing has historically been challenging to manage and legal fishing tourism has the potential to offer some relief if arapaima are resilient to catch-and-release fishing pressure (Lennox et al., 2018b). Given uncertainty about internal states and external drivers of migration, and that climate change could affect the length and intensity of the dry season, the future of arapaima will depend upon an adequate understanding of its movements to enable effective management (photo: Lennox). Mahseer (Tor spp.) such as the Critically Endangered cauvery

(Continued)

humpbacked mahseer Tor remadevii (pictured), are iconic fishes exhibiting potamodromous migrations, most often to facilitate successful spawning (Nautiyal et al., 2001; Pinder et al., 2019). Mahseer are distributed in the monsoonal rivers of South and Southeast Asia, many of them heavily modified and fragmented due to hydropower dams. There is an urgent need to understand and resolve the impacts of river engineering projects on mahseer migrations. Particularly important are the Mekong and Ganges-Brahmaputra river systems which harbor many of the conservation-concern and data-deficient mahseer species. Preliminary understanding of Tor putitora revealed large-scale migrations (>50 km in a 48 h period) to warmer (non-snow fed) tributaries for spawning and homing behavior of individual fish to distinct tributaries on an annual basis (Fisheries Conservation Foundation and World Wildlife Fund-Bhutan Unpublished; photo: John Bailey).

The Murray-Darling Basin in Australia is home to 56 fish species, all of which migrate at some stage of their life (Koehn and Lintermans, 2012). Golden perch are known to traverse thousands of kilometers when they migrate upstream in high densities (Reynolds, 1983). Movement of fish within and between river systems remains significantly restricted by over 10,000 dams and weirs without adequate fish passage (Baumgartner et al., 2009). Further, many fish, as well as sensitive eggs and larvae, are either diverted into water distribution canals, or pumped onto irrigation crops and die (Gilligan and Schiller, 2003). Significant numbers of fish also die when they pass through sluice type weirs (Baumgartner et al., 2006). These observations demonstrate that physiological traits may be important to understand migratory species. Passage requirements for adults are significantly different to those for early life history stages but all should be considered in a holistic sense when considering fish migration behavior (photo: Baumgartner).

Understanding the factors influencing fish migration can also be essential for management of invasive species. The sea lamprey (Petromyzon marinus) is a prolific invader in the North American Great Lakes that imparts substantial economic damage on fisheries (Smith and Tibbles, 1980; Christie and Goddard, 2003). Parasitic lamprey hatch in Great Lakes tributaries and the larvae metamorphose and sub-adults move to the lakes where they parasitize many different fish species including native lake trout (Salvelinus fontinalis, pictured). Lamprey control has benefited from installing unpassable barriers that block lamprey return migrations to spawning habitat, but the same barriers have impacted migrations of the diverse fish fauna other than leaping salmonids (McLaughlin et al., 2006). Understanding that lamprey use conspecific pheromones to navigate has allowed development of semiochemicals to distract them during migration (Siefkes, 2017; photo: Wikimedia Commons).

Eight species of yellowfish in southern Africa have been referred to as potamodromous (O'Brien et al., 2014). Conflicting literature regarding the migration behavior and distances traveled of the largemouth yellowfish (Labeobarbus kimberleyensis) suggests that studies may need to focus on detailing their behavior and the internal and external drivers of their migrations. The yellowfish are just one example of the many fish species in southern Africa that are sensitive to the increasing impacts of development, most notably from instream barriers (O'Brien et al., 2013). Without the necessary information on the migratory behavior of these freshwater fishes, water managers are not able to implement the necessary measures required to mitigate issues arising from development. It is thus critical that studies relating to the migratory strategies of fish in southern Africa, and indeed the entire continent, need to be prioritized (image: Brink).

The Mekong River is especially significant because migratory species, many of which provide substantial food security and economic benefits, are expected to decline in the next 20 years (Dugan et al., 2010). It has been long suspected that several large upstream migrant species in the Mekong might originate from the ocean (Ferguson et al., 2011). If this is true, then mainstem dam development on the Mekong may effectively extirpate entire endemic species by blocking access to critical habitat (Hogan et al., 2004). The Krempfii catfish (Pangasias krempfii) is so far the only described anadromous species in the Mekong (Hogan et al., 2007). It commences its spawning migration in February each year and spends up to 4 months reaching its spawning grounds above the Khone Falls, Laos. Upon hatching, the juveniles then commence a seaward migration. The migrations are cyclic, annual, and important for the long term sustainability of this species, which can grow to 1.4 m long and fetch up to \$8 USD per kilo on the local markets. The main threats to these species are hydropower dams, especially on the mainstem in Cambodia and Vietnam that may block access to the upstream spawning grounds (photo: Baumgartner).

can shape an understanding of the hierarchical structure of fish populations, from within rivers and watersheds to regional and landscape scales, and designations of evolutionarily significant units (Dionne et al., 2008). Establishing the role of genetics in the migration of fish can then assist in informing how climatic and anthropogenic selective pressures may influence the genetics of fish populations, with implications for how activities structuring fish populations are managed (Quinn et al., 2007; Kovach et al., 2012).

Ecosystems are not isolated but interconnected by species that cross boundaries and transport matter and energy. Many terrestrial and aquatic food webs rely on migratory fishes directly or indirectly and we still have rudimentary understanding of many of these functional roles but are informed by a few well-studied species that are not necessarily representative of the diversity of migratory species. Salmonid migrations, specifically, have been extensively studied in the context of nutrient subsidies that adults bring from the marine environment into freshwater (Naiman et al., 2002). Recently, it was shown that salmon populations in freshwater correlate with geographically overlapping forest bird abundance and diversity, suggesting crucial linkages among species and habitats that merit further investigation (Wagner and Reynolds, 2019). Similar studies are needed on other migratory species that connect distant habitats, particularly in terms of forming metapopulations of their own species or their parasites and pathogens. How this operates vertically for fish that migrate between the shallows and depths or in many tropical and subtropical aquatic systems is particularly uncertain. The challenge, noted independently by many of the authors in this exercise, is developing international cooperation and an understanding of how to integrate this information into policy that can adequately and fairly protect species that cross jurisdictions and political boundaries (Dallimer and Strange, 2015; Midway et al., 2016).

Distinguishing migration from other movements is a challenge. Partial migration theory (Chapman et al., 2012a,b; Secor, 2015a), which draws on studies of birds and fishes to explain a latent capacity for all taxa to exhibit phenotypic variation in their migration behaviors, may help encapsulate the many different forms that migration can take in fishes. Physiological research, such as the genetics of seawater tolerance, will also contribute to an evolutionary perspective on the origins of fish migration (Ishikawa et al., 2019). Recent discoveries of diverse modes of seasonal and lifetime migrations within populations fit well with the theory of partial migration, and aligns well with recent research agendas on population connectivity (Cowen and Sponaugle, 2008), biocomplexity (Ruzzante et al., 2006), and resilience (Hilborn et al., 2003; Kerr et al., 2010). Partial migration may be fixed at the individual level (obligate partial migration) or conditiondependent (facultative partial migration; Boyle, 2008). The latter is especially pertinent to many fishes and to management of aquatic environments because long-lived individuals may migrate under some conditions (or in some years) but the same individuals may exhibit non-migratory behavior under other circumstances (or in other years), emphasizing their flexible and presumably adaptive responses to varying environmental conditions (Lucas and Baras, 2001). This also may inform the potential for irruptive behavior in some species. Developments in explanatory frameworks such as movement ecology and partial migration will be informed by increased attention to central mechanisms (e.g., locomotion, navigation, internal, and external drivers) and common emergent properties (schooling, population structure, range shifts, speciation) across taxa. Exploring how migration and dispersal interact along a continuum will assist in categorizing species and a better understanding fish migration in the future.

Much migration research focuses on active movement of animals but currents may transport eggs, larvae, and juveniles such that passive transport can form an integral component of many fish life cycles (e.g., Zeng et al., 2019). Although passive transport, particularly of fish larvae, has been a long-term focus of research and modeling in coastal and ocean systems (Harden Jones, 1968; Sinclair, 1988; Secor, 2015a), it has not received as much attention in freshwater ecosystems. Pelagic transport of eggs and larvae is common in many tropical freshwater fish taxa (Lucas and Baras, 2001) and those of goliath catfish (Brachyplatystoma) drift hundreds to thousands of kilometers toward estuarine reaches of South America's largest rivers (Barthem et al., 2017). In the ocean, adult plaice (Pleuronectes platessa) use selective tidal stream transport by moving vertically into the water column to select the direction of movement, potentially saving energy or assisting their conveyance to suitable habitat (Metcalfe et al., 1990). Hydraulics in both the marine and freshwater environments therefore have significant relevance to fish migration. There is strong potential for synergistic research between fish migration biology, river hydrogeomorphology and physical oceanography to study the role of currents on the fate and behavior of migratory species, with generation of predictions for climate change impacts.

Collaboration with Indigenous nations and local stakeholders/interests can help steer the research agenda to prioritize questions that we have set out here. Many projects have shown that strong collaborations yield active knowledge exchange. For example, in the Penobscot River, USA, access to thousands of kilometers of river were restored following dam removals primarily as a result of the active communication and involvement of all stakeholders from the beginning of the project (Opperman et al., 2011). However, a lot still needs to be learned about interdisciplinary research and community involvement in research and conservation efforts (Nguyen et al., 2016). This specifically includes improving communication among researchers, engineers, water managers, and authorities; reaching out to politicians; improving collaborations and commitment; and creating awareness and inspiring citizens (Young et al., 2016). The current focus on flyways for avian conservation has provided an instrument for international cooperation (Runge et al., 2015) and parallel efforts should be developed by identifying and protecting key spatiotemporal swimways for migratory fish (e.g., Pracheil et al., 2012). Global initiatives such as World Fish Migration Day (https://www. worldfishmigrationday.com/), International Year of the Salmon (https://yearofthesalmon.org/), and the emerging Swimway Global initiative (https://www.worldfishmigrationfoundation. Lennox et al. Fish Migration Questions

com/projects/4/swimway-project) aim to improve public knowledge and unify organizations at a global level. Policy instruments and international cooperation are critical to realizing protection of migrants and the services and values they support.

We have generalized many of our questions to apply to the broad spatial, temporal, and geographic scales, but many of the initial formulations of the questions noted very specific threats to particular fish migrations that required investigation. Indeed, humans continue to look to the aquatic systems as resources to be tapped for solutions to global problems such as energy production, commerce, sewage treatment, stormwater impoundment, food and water security, recreation, and more (Fang and Jawitz, 2019; Reid et al., 2019). The myriad stressors emerging from the associated infrastructure affects fish habitat with noise, light pollution, electromagnetic interference, temperature and flow alteration, chemical pollution, water abstraction, and other threats, all of which have the potential to interfere with internal state, navigation, and locomotion. These and other threats are being explored but require additional attention and replication with different species, because many uncertainties remain and different species frequently respond very differently to common threats (e.g., Gill et al., 2012; Hellström et al., 2016; Filous et al., 2017). Introduction and spread of non-native species facilitated by humans and climate change have unknown consequences on migratory fish and their ecosystems. Novel predators and parasites are being introduced that can negatively affect migration (Boulêtreau et al., 2018).

A final challenge emerging from this exercise is the importance of prioritization. We have identified many questions with myriad implications for understanding, managing, and conserving ecosystem integrity in a changing world, and it is a great ambition to answer them all. Much of the research now conducted on fish is motivated to address threats from climate change or human activities but the efforts could significantly benefit from abetter fundamental understanding of migration and migratory species. Research necessarily tends to focus on species of economic importance or species at risk, but we must not lose sight of the importance of all species (e.g., Cooke et al., 2006). Studying migratory fishes in ecosystem contexts (i.e., with predators, prey, competitors, and pathogens) will be essential to understand how ecosystem processes operate and how migration functions to modulate the biotic and abiotic interactions that migratory species have in their environment. This directs attention to integrative and flexible models, which can make use of best available empirical studies and evaluate likely responses to future scenarios of change in probabilistic frameworks (e.g., Heath et al., 2008; Kerr et al., 2010; McGilliard et al., 2011). Ultimately, it is our hope that this list of questions is used to shape future projects, highlight the importance of migratory fishes, and to inform conservation decisions.

### AUTHOR CONTRIBUTIONS

SC, CP, and RL conceived and outlined the study aims and objectives. All authors contributed questions, had input on organization into themes, and contributed to the direction and framing of the manuscript and writing. LB, KBr, YC, AF, RL, PP, and RR compiled **Box 1**. AH conceptualized and illustrated **Figure 1**.

### FUNDING

MA-M and SC were funded by the Natural Sciences and Engineering Research Council of Canada (NSERC), and the Canada Research Chairs program. EM was funded by NSERC. Several authors were part of HYCANOR, GLATOS, and the Ocean Tracking Network. JD is funded by the Research Council of Norway. AH was supported by Research Initiation Award #16000691 from the US National Science Foundation. YC was funded by the Chinese Academy of Sciences (grants Y45Z04, Y62302, ZDRW-ZS-2017-3-2, Y55Z061201, and QYZDB-SSW-SMC041) and WWF (grants 10002550 and 10003581). The Missouri Cooperative Fish and Wildlife Research Unit was sponsored jointly by the U.S. Geological Survey, Missouri Department of Conservation, University of Missouri, the Wildlife Management Institute, and the U.S. Fish and Wildlife Service. John Bailey provided the image of Tor remadevii.

#### ACKNOWLEDGMENTS

SJ-H acknowledges funding from the Welsh European Funding Office and European Regional Development Fund under project number 80761-SU-140 (West). Select imagery for **Figure 1** was obtained courtesy of the Integration and Application Network, University of Maryland Center for Environmental Science (ian.umces.edu/imagelibrary/).

# REFERENCES


Altizer, S., Bartel, R., and Han, B. A. (2011). Animal migration and infectious disease risk. Science 331, 296–302. doi: 10.1126/science.1194694


Harden Jones, F. R. (1968). Fish Migration. London: Edward Arnold.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Lennox, Paukert, Aarestrup, Auger-Méthé, Baumgartner, Birnie-Gauvin, Bøe, Brink, Brownscombe, Chen, Davidsen, Eliason, Filous, Gillanders, Helland, Horodysky, Januchowski-Hartley, Lowerre-Barbieri, Lucas, Martins, Murchie, Pompeu, Power, Raghavan, Rahel, Secor, Thiem, Thorstad, Ueda, Whoriskey and Cooke. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Prevalence and Mechanisms of Partial Migration in Ungulates

#### Jodi E. Berg<sup>1</sup> \*, Mark Hebblewhite<sup>2</sup> , Colleen C. St. Clair <sup>1</sup> and Evelyn H. Merrill <sup>1</sup>

<sup>1</sup> Department of Biological Sciences, University of Alberta, Edmonton, AB, Canada, <sup>2</sup> College of Forestry and Conservation, University of Montana, Missoula, MT, United States

Partial migration, a phenomenon wherein only some individuals within a population migrate, is taxonomically widespread. While well-studied in birds and fish, partial migration in large herbivores has come into the spotlight only recently due to the decline of migratory behavior in ungulate species around the world. We explored whether partial migration in ungulates is maintained at the population level through frequency-dependence, an environmental-genetic threshold, or a conditional strategy. Through a review of studies describing individual variation in migratory behavior, we then addressed how density-dependent and -independent factors such as social constraints, competition for forage, and escape from predators or pathogens, alone or together, could lead to occurrence of both migrants and residents within a population. We searched for evidence that intrinsic and extrinsic factors could combine with genetic predispositions and individual differences in temperament or life experience to promote migratory tendencies of individuals. Despite the long-held assumption for ungulates that migration is a fixed behavior of individuals, evidence suggested that flexibility in migratory behavior is more common than previously thought. Partial migration maintained by a conditional strategy results in changes in movement tactics as state-dependent responses of individuals. Data are needed to empirically demonstrate which factors determine the relative costs and benefits to using migratory vs. resident tactics. We outline what types of long-term data could address this need and urge those studying migration to meet these challenges in the interest of conserving partially migratory populations.

Keywords: ungulate, partial migration, density-dependence, frequency-dependence, condition, review

# INTRODUCTION

Dramatic declines in populations of migratory ungulates and the disappearance of migratory behavior in many ungulate species are now recognized as a global conservation challenge (Berger, 2004; Bolger et al., 2008; Tucker et al., 2018). Population reductions have been well-documented in migratory species ranging from antelope (Antidorcas marsupialis, Child and Le Riche, 1969; Saiga tatarica, Milner-Gulland et al., 2001) and buffalo (Syncerus caffer caffer, Bennitt et al., 2016) to wildebeest (Connochaetes taurinus, Gasaway et al., 1996) and zebra (Equus burchelli antiquorum, Bartlam-Brooks et al., 2013). Loss of migratory behavior in ungulates is attributed primarily to human-induced changes to landscapes, which may be exacerbated by climate change (Lendrum et al., 2013). Loss of migration can have significant ecological impacts, potentially resulting in collapse of whole ecosystems, extending from alteration of plant composition and ecosystem

#### Edited by:

Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway

#### Reviewed by:

Arne Hegemann, Lund University, Sweden Inger Maren Rivrud, Norwegian Institute for Nature Research (NINA), Norway Marco Festa-Bianchet, Université de Sherbrooke, Canada

> \*Correspondence: Jodi E. Berg jberg@ualberta.ca

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 29 March 2019 Accepted: 13 August 2019 Published: 29 August 2019

#### Citation:

Berg JE, Hebblewhite M, St. Clair CC and Merrill EH (2019) Prevalence and Mechanisms of Partial Migration in Ungulates. Front. Ecol. Evol. 7:325. doi: 10.3389/fevo.2019.00325 processes such as grassland production and nitrogen mineralization (McNaughton et al., 1988; Frank, 1998; Holdo et al., 2006), to declines in other species including apex predators (Packer et al., 2005; Lee et al., 2016), to loss of wildlife tourismbased dollars normally used for environmental protection of Africa's most iconic species (Harris et al., 2009; Holdo et al., 2011). Given the potential severity of these ecological impacts and their associated economic consequences, identifying the processes that lead to migratory behavior should be a primary focus of biodiversity research and conservation efforts to address the loss of migration in ungulate populations (Bolger et al., 2008).

Migratory movements of individuals are expected to arise in variable environments wherein ungulates migrate to enhance lifetime reproductive fitness by gaining access to critical resources such as nutrients or water, reduce their likelihood of predation, or escape parasites (Fryxell and Sinclair, 1988a; Mysterud et al., 2011, 2016; Qviller et al., 2013). However, anthropogenic disturbances and environmental changes have sometimes altered the relative benefits of migrating in large herbivores to make residency more profitable (Berger, 2004; Hebblewhite et al., 2006; Jones et al., 2014). Partial migration is a populationlevel phenomenon in which a population is comprised of both resident and migrant individuals (Chapman et al., 2011a). Partial migration has become a focus for studies on ungulates only recently, and is presumed to result from trade-offs between the costs and benefits of migration (Eggeman et al., 2016). Although several studies have described the pattern of partial migration, the underlying ecological processes, which we review below, for maintaining partial migration are theoretical or empirically correlative. Experimental manipulations needed to identify mechanisms driving migratory tendency in large mammals may be unethical and are difficult (but see below), which creates an urgent need to better synthesize existing information on partial migration in ungulates. A better understanding of the worldwide decline in migratory behavior of ungulates will offer directions for future studies and inform associated conservation actions (Bolger et al., 2008).

We explore this topic with a review that begins by defining migrant and resident behavior in the context of partial migration. We then review the evidence for population-level mechanisms described by others to explain why partial migration occurs and is maintained in diverse populations of ungulates that inhabit variable environments. We explain how changes in proportions of migrants and residents within populations might occur both across generations, through either a frequency- or densitydependent fitness equilibrium, and within generations, via behavioral switching between migrant and resident behavior by individuals. Then we review the factors operating on individuals that might promote migration vs. residency. We focus primarily on genetic variability, social interactions and cultural inheritance, intrinsic factors such as age and nutritional condition, and extrinsic or environmental factors such as forage and predation risk. We conducted the review by searching the published literature for all ungulate species listed in Ultimate Ungulate (Huffman, 2018) and by Groves and Grubb (2011) within the orders Perissodactyla (odd-toed ungulates) and Cetartiodactyla (even-toed ungulates). We used "Web of Science" and "Google Scholar" search engines to find articles by the common and Latin name and/or genus in combination with "migra<sup>∗</sup> ," "resid<sup>∗</sup> ," "partia<sup>∗</sup> migra<sup>∗</sup> ," "facultative migra<sup>∗</sup> ," or "conditional migra<sup>∗</sup> ." In particular, we retained any article that described partial migration (i.e., the article needed to state that a portion of the population remained resident/sedentary, and another portion of the population migrated, irrespective of the form of migration observed) and addressed or speculated on the reasons behind the observed differences in migratory behavior. We chose not to include gray literature due to variability in data types and rigor. The hypotheses we evaluated are not mutually exclusive and two or more proximate mechanisms for migration are likely to operate simultaneously (Ketterson and Nolan, 1983; Smith and Nilsson, 1987; Avgar et al., 2014). The review focused on migration in female ungulates because adult female survival is thought to have the greatest influence on large ungulate population dynamics (Gaillard et al., 1998; Raithel et al., 2007) and because few articles concentrated on males or compared factors affecting migratory behavior between the sexes; we included migratory tendency in males if new mechanisms arose and there were adequate data (see **Supplementary Tables 1**, **2**). We end by challenging researchers to collect the long-term data necessary to test the mechanisms underlying maintenance of partial migration to bring us closer to conserving ungulate populations in the face of ongoing environmental change.

# WHAT IS A MIGRANT?

Migration as a phenomenon is not easy to define because of variation in both terminology and types of animal movement among taxa (Sinclair, 1983; Fryxell et al., 2011). The term migration is also used differently when it is applied to individuals vs. populations (Dingle and Drake, 2007; Dingle, 2014b). In either context, associating migration with a trait or a behavioral syndrome (sensu Sih et al., 2004) requires that migration responds to natural selection (Dingle, 2014b), but it may do so as part of a correlated suite of behavioral, physiological, or life history traits (Réale et al., 2010). In this review, we define migration as a behavioral tactic (sensu Dominey, 1984; Gross, 1996; Dawkins, 1999) describing a movement type that is exhibited by individuals (**Table 1**). We call it a tactic, rather than assume it is a genetically fixed strategy (sensu Maynard Smith, 1982) because of the information we synthesized during our review (below). Consistent with this definition as a tactic, the migratory tendency of an individual could be rigid and result from conditions during a key developmental window (i.e., phenotypic plasticity or reaction norm) or change over time (i.e., ongoing behavioral flexibility; Piersma and Drent, 2003). We explore the evidence for these mechanisms below.

Additional confusion about the meaning of migration stems from spatial definitions. Ungulates are among the taxa for which migration is thought to be movement, most commonly, but not always, as a round-trip between discrete seasonal ranges (Sinclair, 1983; Fryxell and Sinclair, 1988a). The spatiotemporal separation between ranges and the emphasis on return movement makes migration different from: (1)

#### TABLE 1 | Definitions of words used in discussion of migration in ungulates.




Note that populations exhibiting these non-exclusive forms may also be described as partially migratory.

dispersal, a relatively short-term, one-time movement to a new population or a new range primarily for the purpose of reproduction; (2) nomadism or roaming, where animals follow resource pulses with little spatial predictability; and (3) residency, where there is continuous, overlapping use of the same range (McPeek and Holt, 1992; Hjeljord, 2001; Abrahms et al., 2017). Distinguishing between migratory tactics using seasonal ranges becomes challenging when individuals exhibit more idiosyncratic or mixed movements, such as returning to a seasonal range soon after leaving it (Dingle and Drake, 2007; Dingle, 2014b). Describing migration as a round-trip is problematic when individuals switch among multiple ranges and do not return to the same seasonal range they used the summer or winter before (e.g., Eggeman et al., 2016). Variation in the spatial extent of migratory movement reinforces that partial migration is not the simple dichotomy that is implied by terms like migrant vs. resident or short-distance (<10–50 km) vs. long-distance migrant (>50–150 km; **Table 2**). Indeed, some authors consider the choice to migrate as one point in a continuum of movement responses that occur over multiple scales of spatiotemporal variability (Cagnacci et al., 2011).

Greater latitude in the way migration is defined, behaviorally and spatially, may lessen the need for several quantitative methods used to distinguish migration from other types of movements and to classify variation in migratory movements (Cagnacci et al., 2016; Singh et al., 2016; Abrahms et al., 2017; Peters et al., 2019). Migrants are often distinguished from residents based on criteria such as the amount of seasonal home range overlap (Mysterud, 1999; Ball et al., 2001; Fieberg and Kochanny, 2005), trajectory segmentation (Buchin et al., 2013), or algorithms that cluster seasonal locations (Cagnacci et al., 2011, 2016; Damiani et al., 2016). A second approach is based on Correlated Random Walk (CRW) models (Bergman et al., 2000), including the increasingly popular Net Squared Displacement (NSD), measured as the cumulative squared displacement from a starting point (Turchin, 1998; Nouvellet et al., 2009; Bunnefeld et al., 2011). The drawback to NSD is that it can be computationally complex and often requires ad hoc reclassification of the migratory status of an individual (Spitz et al., 2017). On the other hand, this method is capable of quantifying different types of movement along a continuum, overcoming the problem of simplistic dichotomies (Singh et al., 2016). Despite the limitations in methodologies, quantifying animal movements as migratory behavior is a first step in exploring how partial migration is maintained.

#### MAINTENANCE OF PARTIAL MIGRATION IN UNGULATE POPULATIONS

Historically, partial migration was simply described as a kind of within-population variation in movement behavior in which just a part of the population migrates (Lack, 1943) with speculation about causation (e.g., Lack, 1943; Lundberg, 1988). Modern assessments have since evolved to developing theoretical frameworks for hypotheses that need to be tested with empirical data (Kokko, 2007, 2011; Lundberg, 2013). Both past and modern interpretations assume that migration results from natural selection such that the occurrence of partial migration requires the long-term balancing of Darwinian fitness between migrant and resident tactics under different ecological conditions. Such polymorphisms in life history tactics are maintained over evolutionary time only if fitness varies with population densities, environmental conditions, or similar phenomena (Swingland and Lessells, 1979). More specifically, natural selection could favor the maintenance of partial migration within a population via: (1) a frequency-dependent mixed evolutionarily stable strategy (ESS; Swingland, 1983; Dingle, 2014b), (2) an environmentalgeneticthreshold, a variant of a gene-environment interaction that accommodates changing environments (Pulido, 2011), or (3) a conditional strategy in which an individual's choice of migratory tactic varies with other aspects of phenotype, individual state, or the behavior of other individuals in the population (Lundberg, 1987; Chapman et al., 2011b, 2012; Pulido, 2011). Each of these mechanisms might prevail under different environmental conditions.

A frequency-dependent evolutionarily stable state (ESSt) assumes that migratory behavior is fixed, and residents are favored when migrants are at a high frequency and vice versa. At some specific equilibrium frequency, the migratory and non-migratory alternatives should have the same average pay-off; that is, if one alternative increases in frequency, its


<sup>a</sup>See also (Hebblewhite and Merrill, 2011).

<sup>b</sup>See also (Gurarie et al., 2017; Peters et al., 2017).

<sup>c</sup>Considered conditional migrants: migrating at least once, but failing to migrate during any 1 season, or migrating briefly within 1 season.

pay-off should decrease (i.e., fitness is negatively frequencydependent; Swingland, 1983; Dingle, 2014a). The evolution of partial migration has been examined using frequency-dependent ESS modeling especially in birds (Lundberg, 1987; Kaitala et al., 1993; Kokko, 2011). However, empirical support for frequencydependent ESSts in most species is lacking (Chapman et al., 2011b; Lundberg, 2013), perhaps because negative frequencydependence may be observable only when the population is at or above the carrying capacity.

In partially migratory ungulates, many authors assume that migration is a fixed trait (Hebblewhite and Merrill, 2011; Gaillard, 2013; Middleton et al., 2013b). Fixed migration would necessarily mean that the ratio of migrants and residents in a population would need to be balanced by density- or frequency-dependence in a mixed-ESS at the population level (Lundberg, 1988; Kaitala et al., 1993), as described above. That is, individuals are not able to change their behavior, but the relative demographic success of each separate tactic determines the relative fitness of each behavior, which then changes in some stabilizing way as densities or frequencies change. Without such a stabilizing mechanism, a population would be expected to reach fixation for a single behavior. The rarity of "pure" migrant or resident populations itself rejects this notion. Further, partial migration through an ESSt could not happen if there is switching between tactics, which has been reported in deer (Odocoileus virginianus, Nelson, 1995), elk (Cervus elaphus, Eggeman et al., 2016), impala (Aepyceros melampus, Gaidet and Lecomte, 2013), moose (Alces alces, White et al., 2014), pronghorn (Antilocapra americana, White et al., 2007), Sierra Nevada bighorn sheep (Ovis canadensis sierrae, Spitz, 2015), and Svalbard reindeer (Rangifer tarandus platyrhynchus, Hansen et al., 2010; Meland, 2014; **Table 3**). In these studies, the average annual rate of switching was ∼20%, although most studies had limited ability to detect switching due to inadequate sample size or infrequent monitoring over the course of entire lifetimes. If the results of these few switching studies are representative of the many long-lived ungulates with lifespans >10 years, the evidence suggests that individuals may switch tactics several times during their lifetime.

In contrast, the environmental-genetic threshold describes a mechanism in which a number of additive, environmental variables may interact with a number of genes to contribute toward expression of an underlying phenotypic, behavioral liability (i.e., migratory tendency) or trait that is normally distributed within a population (**Figure 1**, Pulido, 2011). According to the environmental-genetic threshold model, individuals have a genetically determined propensity for migration that is triggered, or not, by environmental conditions. A threshold exists below which individuals are sedentary, whereas those above the threshold are migratory (Berthold, 1991; Pulido et al., 1996). Migratory traits may not be fixed, even under strong, directional selection, because as the distribution of migratory propensity shifts below the threshold, migratory traits will not be phenotypically expressed (Pulido, 2011). Environmental variables such as food, social dominance, or body condition may affect individuals with liability values close to the threshold, causing them to change migratory

tactic. This conceptual model has not been used to address partial migration in ungulates, and testing its predictions would require long-term studies once the genetic basis or a correlate for migration propensity was identified. Even if further work identifies genetically controlled, regulatory pathways of complex traits linked to migration, monitoring the interaction of these traits with environmental conditions over a sufficiently long period in free-ranging ungulates remains a formidable challenge (Pulido, 2011).

environmental shift.

The alternative to genetically fixed traits or liabilities is the possibility that migration varies between individuals as a function of state, such as age, nutritional condition, or other circumstances. As we discuss below, state-dependent migration may be relatively fixed intrinsically (e.g., dependent on an individual's age or sex or personality), or highly plastic based on nutritional state (e.g., fat reserves or other physiological mechanisms of the metabolic, immune, or endocrine systems) or extrinsic conditions (e.g., predation risk, parasite loads, or climate). If fitness varies temporally with environmental conditions (Rolandsen et al., 2017), then fitness balancing is not necessary over the short term. In this case, a single conditiondependent strategy could produce 2 (or more) tactics. Each individual should adopt the migratory tactic that is best for it at the time (Swingland, 1983), in some cases, making the "best of a bad job" (Lundberg, 1987) and resulting in relative pay-offs that may not be equal across individuals. For example, dominant or more competitive individuals may optimize fitness by remaining resident, whereas less competitive or sub-dominant individuals may optimize fitness by trading the cost of migration in return for a habitat where there is less competition (Swingland, 1983; Lundberg, 1987; Chapman et al., 2011b).

Consequently, both migratory and non-migratory tactics may be maintained within a population due to differential density-dependent regulation of vital rates that must counteract each other over the long term, such that any differences in reproductive success between migrants and residents must be countered by differences in survival (**Figure 2**). Hebblewhite and Merrill (2011) found that despite higher pregnancy rates and winter calf weights, migratory elk were more at risk during migration. In contrast, residents reduced predation risk by remaining in areas of human activity, which resulted in lower pregnancy and calf weights, but slightly higher adult and calf survival. Similarly, White et al. (2014) also found that calf survival was higher in migratory moose, but that there was no difference in body fat accumulation between residents and migrants. Both studies were suggestive of demographic balancing between the two tactics (Hebblewhite and Merrill, 2011; White et al., 2014). Peters et al. (2019) suggested that the probability of migrating should increase under highdensity conditions; with increasing density, density-dependent or environmentally-driven switching between tactics would maintain partial migration within a population. Indeed, recent evidence from elk supports the notion of density-dependent migration being a potentially stabilizing mechanism regulating partial migration in populations (Eggeman et al., 2016). On the other hand, stochastic environmental events could cause mortality for the more successful tactic, independent of density, but if the increase in mortality is only to the level of survival of the alternative behavior, partial migration can be maintained (Grayson et al., 2011). The balance between these conflicting costs and benefits leads to individuals remaining in a range yearround, or moving to new areas. In the next section, we identify and assess the support for the most commonly hypothesized mechanisms shaping individual variation in migratory tendency in ungulates.

# WHY DO SOME INDIVIDUALS MIGRATE?

In this section, we summarize results from a range of field studies focused on ungulate migration to address what factors promote migration in an individual animal. We summarize evidence for a genetic basis to migration, evidence for the role of learning and cultural transmission, and factors related to individual state or environmental conditions and/or their interactions (**Tables 4**, **5**).

#### Genetics

Evidence for a direct genetic basis for migration would require that behavioral traits of individuals were linked to specific alleles that differentiated groups or showed heritability, as demonstrated for migratory restlessness in captive birds (Berthold and Querner, 1982; Terrill, 1987; Berthold, 1991; Berthold and Pulido, 1994). Such experiments showing restless behavior related to migration have not been attempted and may not be feasible in ungulates, which are harder to hold in captivity and often express less spatial and temporal synchrony in their migration. Nonetheless, some authors have attempted more correlative approaches for exploring indirect genetic effects by using genetic surveys to distinguish individuals with different migratory tendencies. For example, authors used microsatellites to identify genetic differentiation between GPScollared pronghorn antelope defined as migrants vs. residents in the Yellowstone Ecosystem (Barnowe-Meyer et al., 2013). Similar uses of microsatellites have revealed genetic structure in ungulates (e.g., Coltman et al., 2003; Colson et al., 2016), but inferences from microsatellite differentiation based on a few multi-loci (typically <20) were generally limited (**Table 5**). This scanning approach might better distinguish behavioral differences among individuals with new genomics approaches, such as amplified fragment length polymorphism (AFLP) markers in whole genome scans (Liedvogel et al., 2011).


TABLE 4 | State- or condition-dependent hypotheses to explain individual variation in migratory tendency within partially migratory ungulate populations.

Support for these hypotheses can be found in Table 5.

A second approach for identifying genes associated with migration could attempt to isolate aspects of mitochondrial genotypes. For example, the probability of being migratory in a hybrid swarm of caribou (R. tarandus) in the Canadian Rockies was higher in individuals carrying a Beringian– Eurasian haplotype, which was mainly associated with the migratory, barren-ground subspecies, compared to the typically non-migratory woodland caribou (McDevitt et al., 2009). Interestingly, these animals could not be distinguished with microsatellite data, perhaps owing to interbreeding between diverged lineages since the last glaciation (McDevitt et al., 2009). The promise of an mtDNA approach was amplified by the correlation reported by Northrup et al. (2014) between the timing of migration in mule deer (O. hemionus) from 4 distinct winter ranges in the Piceance Basin of Colorado. These authors attributed the correlation to differences in mitochondrial efficiency associated with metabolic demands of migration.

Two other classic approaches for identifying the genetic basis of any behavior would be to compare parent-offspring pairs in long-term studies with known individuals (e.g., Gaillard, 2013) or to conduct cross-fostering experiments. To our knowledge, no authors have applied either technique to address migration in ungulates. Perhaps measuring gene expression in a species with fixed migrants, fixed residents, and individuals that switch migratory tactics within their lifetime could shed some light. A further challenge would be to consider alternative explanations for genetic correlations. For example, in the case of timing of migration in mule deer, Northrup et al. (2014) were able to reject a causative effect of sociality by controlling for the source of the individual's winter range, which showed little spatial clustering of haplotypes. Clearly, it would be challenging to disentangle alternative explanations such as fat levels or physiological status and social or cultural factors, which we discuss next, in correlative studies to support a genetic component for migration. In many cases, particularly in species for which animals switch migratory tactics within their lifetime, it is likely that genetic tendencies are moderated by environmental circumstances.

#### Learning, Culture, and Personality

Being able to discriminate genetic mechanisms from learning and cultural transmission is difficult but could be possible via studying mother-offspring pairs for long periods. Within and beyond these pairs, it is likely that information about navigation and migratory routes are passed from more experienced, key individuals to those that are less experienced (Dodson, 1988; Couzin et al., 2005; Fagan et al., 2012). Nelson (1998) reported that white-tailed deer fawns mimicked the migratory behavior of their mothers. Particularly in the first year of life, residency or migration can be assumed to be dependent on the migratory status of the parent because of the offspringparent bond (Andersen, 1991b). However, we found few studies that addressed the potential effects of early learning or cultural inheritance on migration beyond the first year in ungulate populations (Sweanor and Sandegren, 1988; Nelson, 1998), although it has been documented in whales (Valenzuela et al., 2009). Translocated bighorn sheep and moose learned to increase knowledge and exploit green waves of forage growth in new environments where they had no previous knowledge of the landscape; as knowledge increased, so did the propensity to migrate (Jesmer et al., 2018).

Social learning that promotes migration does not need to be heritable to evolve (Boyce, 1991), although the ability to learn and mimic migratory behavior is likely partially hereditary (Nelson, 1998). Indeed, behavioral flexibility itself appears to be highly heritable (Laughlin et al., 2011) and might be especially important for partial migration. In the Canadian Rockies, TABLE 5 | Support ( + positive/likely, ? potentially but untested/suppositional, – negative/evidence against) for mechanisms explaining individual variation in migratory tendency within partially migratory ungulate populations, including genetics, learning, personality or cultural transmission, and state- or condition-dependence.


(Continued) Partial Migration in

Ungulates

**251**

TABLE 5 | Continued


Berg et al.

resident elk exhibited bolder personalities that included greater exploration of novel objects relative to migratory individuals (Found and St. Clair, 2016). Bolder elk also exhibited lesser lateralization of hoof preferences when pawing the snow to forage, which potentially signals greater cerebral flexibility (Found and St. Clair, 2017). The same authors suggest that less lateralized animals had genetically determined temperaments that made them more responsive to environmental stimuli, which resulted in greater likelihood of them realizing the increasing benefits of residency and abandoning previous histories of migration (Found and Clair, in review). Similar metrics for studying personality traits in wild animals have proliferated in recent years (reviewed by Dingemanse et al., 2010 and Sih et al., 2012), creating much potential to explore their correlations with both migratory tactics and genotypic variation.

#### State and Physiological Condition

In reviewing potential intrinsic factors promoting migration, we found studies primarily addressed one specific hypothesis related to age, and very few studies directly tied physiological mechanisms to partial migration. The Terminal Investment Hypothesis states that older (past their prime) individuals are more likely to devote more resources toward ensuring successful reproduction than younger (yearling or prime-aged) individuals because they anticipate fewer future reproductive events (Clutton-Brock et al., 1982; Clutton-Brock, 1984; Bercovitch et al., 2009). When applied to migration, this hypothesis predicts that ungulates might have a propensity to remain resident while young so as to prioritize their own survival by avoiding risks that can occur during migration (e.g., from predation or anthropogenic factors; Nicholson et al., 1997; Hebblewhite and Merrill, 2011; Schuyler et al., 2019); as they age, they might accept greater risks to migrate to increase access to resources for investment in their terminal offspring (Fryxell and Sinclair, 1988a; Albon and Langvatn, 1992). The age at which this hypothesis might occur in ungulates could be quite old; indeed, Eggeman et al. (2016) showed potential evidence that elk became more likely to migrate with age in Alberta, Canada, but migrants rarely switched to a resident tactic after aging (>15 years old). However, the opposite pattern appears to occur in both pronghorn antelope, which became non-migratory as they aged (White et al., 2007), and moose, which migrated when young but were less likely to migrate as they aged (Singh et al., 2012). Evidence to support predictions of the terminal investment hypothesis could be confounded with other factors. For example, increasing costs of movement are associated with age-related changes in physiological condition (Ericsson and Wallin, 2001), and home ranges may become smaller with age due to experience gained (Allen et al., 2016).

Migration may also be state-dependent (Visscher and Merrill, 2018). If individuals were able to meet their nutritional demands satisfactorily without migrating, there may be no need to migrate if an individual were to incur additional costs or stress related to movement, predation risk, or social conflict (but see below). Because ungulate survival and reproductive efforts are closely tied to body fat reserves (Cook et al., 2004, 2016; Monteith et al., 2014), we would expect to see the propensity to migrate closely linked to nutritional state if condition buffers consequences (Spitz, 2015). Other physiological mechanisms differing between migratory tactics might include metabolic, immune, and endocrine systems, or oxidative stress associated with intense physical activity or fatigue (Jachowski and Singh, 2015; Hegemann et al., 2019). Although Jachowski et al. (2018) found individual mule deer occupying areas closer to peak forage quality during migration had decreased levels of fecal glucocorticoid metabolites, to our knowledge, there have been no studies comparing these mechanisms as related to partial migration in ungulates. Experiments would most likely be needed to identify the specific physiological mechanism, but even then, these could differ among species and environments (Hegemann et al., 2019). Further, recent evidence shows that transfer of the nutritional benefits that are normally associated with migration to residents, as can occur when irrigated agriculture supplements elk feeding, can promote resident behavior (Jones et al., 2014; Barker, 2018). In fact, reproducing and migrating every other year (Morrison and Bolger, 2012) may be a better tactic for ensuring survival and lifetime reproduction, and decisions surrounding migration in ungulates might be driven almost primarily by nutrition and reproductive status (e.g., Festa-Bianchet, 1988, described below).

# Competition, Forage, Predation, and Pathogens

Competition may promote migration but how and where competition influences the tendency for an individual to migrate may vary. The Dominance or Competitive Release Hypothesis (Ketterson and Nolan, 1976; Fudickar et al., 2013) is based on intraspecific competition, with an individual's propensity to migrate expected to increase at higher density on sympatric range. Although competition for food on high-density, sympatric range is likely, it is difficult to demonstrate directly, but might be inferred. For example, white-tailed deer have shown flexible migratory behavior in which they do not remain after fall arrival on sympatric winter ranges, and instead move back to summer ranges during years of little snow and mild weather, suggesting avoidance of competition on the less nutritional winter ranges (Nelson, 1995). Similarly, the distance migrated by elk and red deer in summer has been shown to increase with density, suggesting avoidance of competition on seasonal ranges (Mysterud et al., 2011; Eggeman et al., 2016). Sawyer et al. (2016) also showed that long-distance migrants spent more time migrating and may have initiated spring migration 3 weeks earlier than moderate- or short-distance migrants to escape intraspecific competition by lessening time spent on winter range. In contrast to competition that occurs on sympatric winter ranges, if population densities also increase on allopatric summer ranges, leading to occupation of all the summer areas, migration tendency can be restricted due to competition and social aggression according to the Social Fence Hypothesis (Matthysen, 2005). For example, Mysterud et al. (2011) reported that a lower proportion of red deer migrated at high density during summer, consistent with this hypothesis. However, the authors only contrasted areas of differing densities and did not measure

variation in habitat quality, which is needed to determine the level of competition (Fretwell and Lucas, 1969). Because fall migration was delayed at high density, Mysterud et al. (2011) further suggested that a combination of the competitive release and social fence hypotheses were needed to explain migratory tendency in ungulates. Constraints on distribution and changes in sociality and aggressive behaviors of individuals would need to be documented on both sympatric and allopatric ranges as ungulate densities increased to support these hypotheses.

In seasonal environments, the Forage Maturation Hypothesis predicts that spatiotemporally varying resources promote migration to maximize nutrient intake where there are phenological gradients of plant development (Fryxell and Sinclair, 1988a; Fryxell et al., 1988; Albon and Langvatn, 1992; Hebblewhite et al., 2008). Whereas the classic example may be the Serengeti wildebeest following new green growth to the plains during the wet season (Holdo et al., 2009), many cervids in temperate systems show migrations tied to elevational gradients in plant green up (Sawyer and Kauffman, 2011; Bischof et al., 2012; Merkle et al., 2016; Aikens et al., 2017). If migrants "surf " or "jump" an altitudinal green wave, they are predicted to enter winter with heavier masses and in better body condition than residents as a consequence of higher-quality forage (Albon and Langvatn, 1992; Hebblewhite et al., 2008). Only a handful of studies focused on partially migratory ungulates have demonstrated that females or their young were fatter when they were migratory (e.g., Mysterud et al., 2001; Hebblewhite et al., 2008; Hebblewhite and Merrill, 2011). Yet, this conclusion for elk in the Greater Yellowstone Ecosystem was driven largely by non-lactating females with no data on calf survival and whether release from nutritional costs associated with calf loss contributed to their better condition; in addition, an influence of surfing on autumn fat levels was not detected for lactating elk so results remained somewhat inconclusive (Middleton et al., 2018). Even fewer efforts have linked the tactic of migration to life-time reproductive success. Such studies would require not only long-term studies but additionally evaluating other costs or benefits of migration.

The major hypothesis posed as an alternative to ungulate migration as a response to forage maturation is the Predation Risk Hypothesis, which states that ungulates migrate to escape or minimize predation or other risk factors, such as human hunting or parasites (Fryxell and Sinclair, 1988a; Bergerud et al., 1990; Hebblewhite and Merrill, 2007). Evidence we found to support this hypothesis focused on ungulates moving outside of predator ranges and denning territories (Bergerud, 1988) or by using terrain where predators travel less frequently (Bergerud and Page, 1987). For example, pregnant bighorn sheep in Alberta moved from relatively higher-quality forage to rugged highelevation summer range earlier than non-pregnant ewes and before plant growth started, which Festa-Bianchet (1988) argued was to avoid predation on vulnerable newborn lambs. On the coast of Alaska, migrant moose showed almost 3 times higher neonatal calf survival by migrating to avoid predation but did not obtain nutritional benefits through accumulation of body fat (White et al., 2014). Recent theoretical work suggests that parasites and pathogens could be drivers of partial migration, either as escape from infected areas or individuals, through loss of infected individuals during migration, or as recovery from infection when parasites cannot adjust to environmental changes that occur during migration (Altizer et al., 2011; Fritzsche McKay and Hoye, 2016; Shaw and Binning, 2016). In support of these mechanisms, Pruvot et al. (2016)showed that migratory elk herds in Canada were potentially less likely to be infected with giant liver flukes (Facioloides magna) when compared with resident elk, and lower intensities of warble fly larvae (Hypoderma tarandi) were found in reindeer the farther they migrated post-calving (Folstad et al., 1991).

While comparing the costs and benefits of migratory tactics represents an important first step to understanding what promotes the tendency to migrate, explaining migration by only 2 hypotheses (predation risk avoidance vs. forage maturation, which tend to be the focus of many studies) limits the possibility that other intrinsic or extrinsic factors could also be influential as we've described above. However, results do demonstrate that there may be no straightforward, easy answer because the top-down benefits of avoiding risk through migration may be complicated by life history trade-offs (the cost of rearing offspring to subsequent fecundity), or which may be at times compete with, or modulate, the bottom-up effects of increased access to forage.

# CONCLUSIONS AND FUTURE DIRECTIONS

We have shown that flexibility in migratory behavior by ungulates is more common than previously appreciated, amplifying the suggestion by others that migration should evolve under widely varying environmental conditions in response to the advantages and disadvantages of different life-history strategies (Holt and Fryxell, 2011; Fryxell and Holt, 2013; Avgar et al., 2014). Migration is a complex phenomenon (Alerstam et al., 2003) determined by a number of traits, in turn affected by several genes with pleiotropic effects (Sutherland, 1998). We conclude that migration is not determined by a direct mapping of genotype to phenotype, making it a flexible tactic adopted within a broader strategy. Establishing that partial migration is common in ungulates, and that it appears to respond to diverse genetic, environmental, and demographic correlates, increases the range of techniques that might be applied to study it. Achieving these advances will require use of clear, universal definitions (Avgar et al., 2014; Cagnacci et al., 2016) and classification methods (e.g., Bunnefeld et al., 2011; Naidoo et al., 2012). In fact, the longer individual white-tailed deer were monitored, the more likely they were to be classified by researchers as conditional migrants as opposed to fixed migrants or residents (Fieberg et al., 2008).

Limitations of past studies of migration might be overcome with an understanding that migration is often flexible. Very few of the studies we found were set up to examine how density could lead to a long-term demographic balancing of migrants and residents within a population, but viewing migration as a conditional tactic in a broader strategy to maximize forage intake increases the range of experimental studies that might be applied to this problem. For example, related hypotheses might be tested by manipulating forage or ungulate access to it in protected areas (e.g., Most et al., 2015) or in managed herds (Rivrud et al., 2016). Similarly, few studies have tested explicitly for a genetic basis for differences in migratory and resident individuals within partially migratory populations. The few studies that mentioned learning or cultural inheritance (Singer et al., 1981; Sweanor and Sandegren, 1988; Andersen, 1991a; Barnowe-Meyer et al., 2013) did not conduct them with detailed behavioral observations or controlled experiments to test related hypotheses. A broader view of the genetic and environmental correlates of migratory tactics increases the relevance of many associated metrics.

The decision to migrate or not is made by individuals, but rarely do studies examine individual decision-making in migratory populations (Ball et al., 2001). Nonetheless, some authors attempt such an approach, as with the characterization of multiannual movement patterns by more than 300 moose in 10 different populations (Allen et al., 2016). But many authors whose studies we reviewed characterized migration dichotomously at the level of single populations. Emerging is the view that migration may be a continuum (Ball et al., 2001), both as a behavior (e.g., individuals may exhibit intermediate tactics or variability in timing and distance) and as a population metric (i.e., 1 to 99% of the population may be migratory). Based on our review, migration as a continuum means the reasons for migration were often hard to detect and characterize (Cagnacci et al., 2016). In particular, instances in behavioral switching between migratory tactics should be explored for their potential intrinsic and extrinsic correlates.

Unfortunately, it is difficult to link multiple, interacting intrinsic and extrinsic variables to the occurrence of migration when there is strong environmental variation (Fieberg et al., 2008). In contrast to fixed migrants in other species that show predictable movements as a result of physiological processes (neuroendocrine and endocrine systems), linking environmental cues (day length, photoperiod) to the mechanisms controlling facultative migration in highly variable environments is challenging (Ramenofsky et al., 2012). We found speculative support for state- or condition-dependent migration in ungulates in our review, but relatively little experimental data, despite several indirect lines of evidence. We know that differences in habitat quality can lead to corresponding differences in physiology, nutritional condition, and reproductive success in ungulates, and that these can be modified by density (Weber et al., 1984; Becker et al., 2010). More studies are needed that relate habitat use to resulting nutritional acquisition, and measures of body condition and reproductive success, to identify the fitness consequences of migratory tactics. Given new advances in remote monitoring of physiological traits in free-ranging animals, studies on not only how body fat at time of capture, but also physiological mechanisms, differ between migrants and residents and contribute to switching between tactics are warranted (Hegemann et al., 2019). Further, studies that track migratory traits of mothers and their offspring could separate the genetic and learned components of migratory behavior from environmental effects.

Current knowledge of partial migration in ungulates is sometimes limited by their large size, long lives, and wideranging use of habitats, but these traits also confer advantages of observability, long-term study, and generalization across spatial scales. These advantages will be further amplified by using methodologies that are increasingly cost-effective and tractable over the long term in space and time, and in remote environments, to test the relative fitness-related consequences of partially migratory behavior (Bolger et al., 2008; Gaillard, 2013). Long-term, demographic studies and population models tracking the life-history traits of co-existing individuals along the resident-migrant gradient through the year will allow for calculating the costs and benefits of their migration patterns (Bolger et al., 2008). Given the potential ecological and evolutionary significance of partial migration, and that everincreasing anthropogenic disturbance and environmental change may alter or eliminate the benefits of migration altogether (Bischof et al., 2012), understanding the genetic, environmental, and density-driven trade-offs underlying partial migration is of the utmost importance.

# AUTHOR CONTRIBUTIONS

JB designed and wrote the first draft of the manuscript. All authors contributed to manuscript revision and read and approved the submitted version.

### FUNDING

This work would not have been possible without the financial and in-kind support from: Alberta Environment and Parks; Parks Canada; Natural Sciences and Engineering Research Council; University of Alberta; University of Montana; Alberta Conservation Association; Minister's Special License— Hunting for Tomorrow and Alberta Fish & Game Association; Rocky Mountain Elk Foundation; Safari Club International Foundation; Safari Club—Northern Alberta Chapter; NASA (USA, NNX11A047G to MH); National Science Foundation (USA) Long-term Research in Environmental Biology grant (LTREB) to MH and EM (1556248); and the Nestor and Sue Cebuliak, Bill Samuel, and William Wishart Graduate Awards.

# ACKNOWLEDGMENTS

We thank Tal Avgar, Andrew Derocher, and Kevin Monteith for comments on an earlier version of the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00325/full#supplementary-material


REFERENCES

doi: 10.1034/j.1600-0706.2003.12559.x

patterns. Ecosphere 7:e01524. doi: 10.1002/ecs2.1524

Sweden. Wildl. Biol. 7, 39–47. doi: 10.2981/wlb.2001.007

Manage. 83, 167–174. doi: 10.1002/jwmg.21564

712–721. doi: 10.1644/09-MAMM-A-057.1

108–115. doi: 10.1016/j.biocon.2013.09.022

doi: 10.2307/3803279

Abrahms, B., Seidel, D. P., Dougherty, E., Hazen, E. L., Bograd, S. J., Wilson, A. M., et al. (2017). Suite of simple metrics reveals common movement syndromes across vertebrate taxa. Mov. Ecol. 5:12. doi: 10.1186/s40462-017-0104-2 Aikens, E. O., Kauffman, M. J., Merkle, J. A., Dwinnell, S. P. H., Fralick, G. L., and Monteith, K. L. (2017). The greenscape shapes surfing of resource waves in a large migratory herbivore. Ecol. Lett. 20, 741–750. doi: 10.1111/ele.12772 Albon, S. D., and Langvatn, R. (1992). Plant phenology and the benefits of migration in a temperate ungulate. Oikos 65, 502–513. doi: 10.2307/3545568 Alerstam, T., Hedenström, A., and Åkesson, S. (2003). Long-distance migration: evolution and determinants. Oikos 103, 247–260.

Allen, A. M., Månsson, J., Sand, H., Malmsten, J., Ericsson, G., and Singh, N. J. (2016). Scaling up movements: from individual space use to population

Altizer, S., Bartel, R., and Han, B. A. (2011). Animal migration and infectious disease risk. Science 331, 296–302. doi: 10.1126/science.1194694 Andersen, R. (1991a). Habitat changes in moose ranges: effects on migratory behavior, site fidelity and size of summer home-range. Alces 27, 85–92. Andersen, R. (1991b). Habitat deterioration and the migratory behaviour of moose (Alces alces L.) in Norway. J. Appl. Ecol. 28, 102–108. doi: 10.2307/2404117 Avgar, T., Street, G., and Fryxell, J. M. (2014). On the adaptive benefits of mammal migration. Can. J. Zool. 92, 481–490. doi: 10.1139/cjz-2013-0076 Ball, J. P., Nordengren, C., and Wallin, K. (2001). Partial migration by large ungulates: characteristics of seasonal moose Alces alces ranges in northern

Barker, K. J. (2018). Home is where the food is: Causes and consequences of partial migration in elk (MSc thesis). University of Montana, Missoula, MT, USA. Barker, K. J., Mitchell, M. S., Proffitt, K. M., and Devoe, J. D. (2018). Land management alters traditional nutritional benefits of migration for elk. J. Wildl.

Barnowe-Meyer, K. K., White, P. J., Davis, T. L., Smith, D. W., Crabtree, R. L., and Byers, J. A. (2010). Influences of wolves and high-elevation dispersion on reproductive success of pronghorn (Antilocapra americana). J. Mammal. 91,

Barnowe-Meyer, K. K., White, P. J., Waits, L. P., and Byers, J. A. (2013). Social and genetic structure associated with migration in pronghorn. Biol. Conserv. 168,

Barten, N. L., Bowyer, R. T., and Jenkins, K. J. (2001). Habitat use by female caribou: tradeoffs associated with parturition. J. Wildl. Manage. 65, 77–92.

Bartlam-Brooks, H. L. A., Beck, P. S. A., Bohrer, G., and Harris, S. (2013). In search of greener pastures: using satellite images to predict the effects of environmental change on zebra migration. J. Geophys. Res. Biogeosci. 118,


Lack, D. (1943). The problem of partial migration. Br. Birds 37, 122–130.


in a forest-grassland mosaic in Tierra del Fuego, Chile. Oryx 49, 30–41. doi: 10.1017/S0030605312001238


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Berg, Hebblewhite, St. Clair and Merrill. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Energetic Status Modulates Facultative Migration in Brown Trout (Salmo trutta) Differentially by Age and Spatial Scale

Samuel J. Shry † , Erin S. McCallum\*, Anders Alanärä, Lo Persson and Gustav Hellström

Department of Wildlife, Fish, and Environmental Studies, Swedish University for Agricultural Sciences (SLU), Umeå, Sweden

#### Edited by:

Nathan R. Senner, University of South Carolina, United States

#### Reviewed by:

Martin Hage Larsen, The Danish Centre for Wild Salmon, Denmark Yolanda E. Morbey, University of Western Ontario, Canada

> \*Correspondence: Erin S. McCallum erin.mccallum@slu.se

> > †Present address:

Samuel J. Shry, Länsstyrelsen Gävleborg, Gävle, Sweden

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 29 March 2019 Accepted: 15 October 2019 Published: 13 November 2019

#### Citation:

Shry SJ, McCallum ES, Alanärä A, Persson L and Hellström G (2019) Energetic Status Modulates Facultative Migration in Brown Trout (Salmo trutta) Differentially by Age and Spatial Scale. Front. Ecol. Evol. 7:411. doi: 10.3389/fevo.2019.00411 Fish display a remarkable diversity in juvenile migration strategies and behavior. Intra-species variation in migration can be considerable, and understanding the driving force of such variation is important for effective management and conservation of migratory fish. In facultative migratory species, such as many salmonid fish, energetic status is known to affect migration strategy and behavior. However, we currently lack a full understanding of how energetic status affects juvenile development and migration over different environmental contexts. In this study, we examined the effect of energetic status on juvenile migration initiation and migratory behaviors in 1 and 2 year old brown trout (Salmo trutta). By manipulating feeding regimes, we created a large variation in trout energetic status (using condition factor as a proxy). We then studied behavioral changes in migration in both a controlled environment (large-scale migration pools) as well as a natural river system using both passive integrative transponder tags (PIT-tags) as well as acoustic telemetry tags. In the laboratory setting, 1 year old trout with higher energetic status were more likely to initiate migration and migrated faster. For 2 year old trout, energetic status did not affect the initiation of migration (the large majority migrated), but high energetic status fish migrated faster. In a small-scale natural creek system, few age one fish migrated (11%); however, these few migrators were within the upper range of energetic status. In 2 year old trout, a high percentage became migrants (79%), and those with higher energetic status migrated at a faster speed. In a large-scale river system, successful downstream seaward migration for 2 year olds was low (9%) and independent of energetic status. Our findings provide valuable data for fisheries management because we show that age at release and energetic status prior to release can impact migration initiation and behaviors. Our findings also indicate that migration measured in the laboratory may over estimate migration in the wild, especially for younger, age one fish. More broadly, this work advances our understanding of this complex life history stage and the mechanisms involved in the initiation, behavior, and survival of migrating brown trout.

Keywords: behavior, smolt, salmonid, acoustic telemetry, PIT-tag, feeding, condition factor, hatchery

# INTRODUCTION

Intra-population variation in migration behaviors are commonly observed in animal populations and can be an important component shaping population structure and viability (Chapman et al., 2011; Pulido, 2011; Avgar et al., 2013). Movement can be a response to adversity (Taylor and Taylor, 1977) and migration should be favored when the benefits outweighs the costs (Gross et al., 1988). Partial migration refers to the situation where both resident and migratory individuals breed in the same population. There is evidence to support that partial migration can be condition dependent and driven by environmentally responsive genetic thresholds that shape the migratory tendency across a range of animal species, including fish, mammals, and birds (reviewed in Chapman et al., 2011; Pulido, 2011; Avgar et al., 2013). In fish, partial migration has received particular attention, especially for anadromous salmonid populations where the consequences of migration may be profound in terms of fitness and ecosystem production (Chapman et al., 2012). In anadromous salmonids, juvenile migration from freshwater habitats to the sea is preceded by a physiological process preparing the fish for ocean life, i.e., smoltification. Smoltification generally takes place during the spring and affects the morphology, physiology, and the behavior of the fish; for instance, they become silvery in color, lose their positive rheotaxis and territoriality, increase their salinity tolerance and olfactory sensitivity among other things (McCormick et al., 1998; Jonsson and Jonsson, 2011). The decision to migrate, and hence undergo smoltification, is believed to occur during time periods (decision windows) several months before actual smoltification and downstream migration occurs (Thorpe et al., 1998). The decision may be controlled via genetically-based thresholds, where an individual's length, body condition, or energetic status is measured against a genetic threshold that varies between sexes and among individuals and populations (McCormick et al., 1998; Thorpe et al., 1998; Dodson et al., 2013).

Currently, we do not fully understand the effect of physiological state on juvenile migration in salmonids, especially for highly facultative migrating species such as the brown trout (Salmo trutta). Brown trout display a large variation in migration strategies, and there can be considerable variation in migration life-histories both within and between populations (Klemetsen et al., 2003; Cucherousset et al., 2005). Juvenile trout will either migrate to access more productive habitat (e.g., the sea or lake), sexually mature as a small resident, or potentially wait as a juvenile until a later decision window (Jonsson and Jonsson, 1993; Satterthwaite et al., 2009; Dodson et al., 2013). Ultimately, these alternative life history strategies have evolved to maximize reproductive success and has led to population resilience in variable environmental conditions (Jonsson and Jonsson, 2011). Proximately, the decision to migrate in brown trout can be affected by juvenile energetic status and energy (food resource) limitation in the natal habitat (Jonsson and Jonsson, 2011; Boel et al., 2014). Measures such as metabolic rate, growth rate, lipid levels, and condition factor are all indicators of an individual's energetic status, and changes in these measures are often associated with smoltification before downstream migration (reviewed in Wedemeyer et al., 1980; Jonsson and Jonsson, 2011; Ferguson et al., 2017). For wild fish, environmental food limitation (Olsson et al., 2006; O'Neal and Stanford, 2011), low body condition (Boel et al., 2014; Peiman et al., 2017), and high metabolic or growth rate (Forseth et al., 1999; Cucherousset et al., 2005) have all been associated with an increased likelihood of migration in brown trout; altogether, indicating that energetic status is an important proximate determinant of juvenile migration.

It is important to understand how energetic status affects juvenile brown trout migration propensity, distance, speed and success depending on fish size and age, especially in an aquaculture framework where effective production of hatchery reared smolt is highly warranted. Hatchery reared brown trout are released in large numbers globally to compensate for loss of natural production due to, for example, damming of rivers for hydropower production. The well-fed hatchery smolt are generally reported to be less successful at reaching the sea during downstream migration when compared to wild smolt (Serrano et al., 2009; Larsson et al., 2012). Excessive feeding in hatcheries also produces fast growing juveniles that quickly reach the size or condition that may activate smoltification (Thorpe et al., 1998). This has led to younger smolts being released from many hatcheries (e.g., at age one instead of age two; Hedman, 2011). Studies using reduced feeding regimes or periods of starvation (i.e., producing fish with lower energetic status) have found that more fish smoltified (Wysujack et al., 2009; Jones et al., 2015), migrated earlier (Vainikka et al., 2012), or migrated faster downstream (Lans et al., 2011; Larsson et al., 2012; Vainikka et al., 2012). But not all studies with hatchery-reared fish have found clear links between feed restriction and smoltifcation and/or migration speed and success (Davidsen et al., 2014; Näslund et al., 2017; Persson et al., 2018). The degree of feed deprivation, age of the fish, and precise timing of when treatments are applied may underlie the apparent variation in findings and be important factors contributing to migration "decisions" being made by juveniles (Thorpe et al., 1998).

To date, the majority of studies that have manipulated the quantity and/or quality of feed have done so to change long-term energy reserves (i.e., manipulations on the order of months), which would significantly reduce lipid stores, body mass, and condition (McCue, 2010; Bar, 2014). Few studies have manipulated short-term energy reserves (i.e., on the order of days) that would alter immediate hunger levels, which is—in part—triggered by an empty digestive track and rapidly declining glycogen reserves (Fletcher, 1984; McCue, 2010; Bar, 2014). Short-term feed deprivation may signal a less productive habitat to juveniles and may induce migratory behavior (Brodersen et al., 2008). Implementing short-term feed restrictions would be logistically practical for hatchery producers, and it would avoid any welfare concerns associated with long-term feed deprivation (e.g., increased fin damage, Persson and Alanärä, 2014). In this study, we tested how changes in long and/or short-term energetic status affected a suite of migration behaviors in hatchery reared brown trout. We induced four levels of energetic status by manipulating time when food was withheld: one with depleted short-term energy reserves (fed long-term, deprived short-term), one with depleted long-term energy reserves (deprived longterm, fed short-term) one with depleted short- and long-term energy reserves (deprived for entire study duration), and one undeprived control (fed for entire study duration). We measured the effect of these regimes on energetic status using fish condition factor, a good and non-invasive proxy for energetic status (Persson et al., 2018). We implemented this study for both 1 and 2 year old fish and measured migration across three spatial scales using controlled laboratory migration pools, in the field across a small spatial scale in a creek, and across a large spatial scale in a river system. Most previous studies have also tended to exclusively focused on a single age/size class and only addressed migration effects in the field, which may mask behavioral effects due to influence of biotic and abiotic factors beyond control, such as predation.

#### MATERIALS AND METHODS

The study was conducted in three parts (**Table 1**). Part one assessed migration in the laboratory using experimental migration pools and passive integrated transponder tags and antennas (i.e., a PIT tag tracking system). Part two assessed migration across a small spatial scale in the wild using PIT tag antennas and tagged individuals in a small creek. Part three assessed large-scale migration in the wild using acoustic telemetry to track migration of tagged individuals in a river. The three parts of this study used the same fish husbandry and feeding treatments. The migration portions of this study were timed to coincide with peak wild brown trout smolt migration in this system, which is typically late May to the end of June with some variation each year related to temperature. We first outline the general methods common to all experiments (husbandry, tagging, feeding treatments), and then follow this with specific details for each migration experiment.

#### Fish Source and Husbandry

The study took place at Norrfors research laboratory at the Norrfors fish hatchery (63◦ 52'N 20◦ 01'E), alongside Ume River outside Umeå, Sweden in 2017. Brown trout are annually released as smolts from the hatchery at both age one and age two. Fish were produced at the Norrfors fish hatchery following standard procedures: the hatchery population of juveniles is derived from a mix of sea-run returning spawners that are of wild and hatcheryorigin (released as smolts) and are caught at the fish ladder at the Norrfors hydropower dam every year. On February 23, a subsample of individuals from the age one and age two cohorts were collected from the hatchery stock and moved to the research laboratory. The hatchery and research laboratory use a flow through circulation system from the adjacent Ume River, causing water temperature to vary with river conditions. Numerous windows allowed for semi-natural circadian light rhythms (63◦N).

#### Fish Tagging and Feeding Treatments

Before placing fish in a feeding treatment, they were tagged using PIT-tags (HPT12, Biomark USA; 12.5 × 3 mm, 0.1 g) and/or acoustic transmitters (v5-180 khz, Vemco; 11 × 3.6 × 5.7 mm, 0.24 g) for later identification. Both tag types did not weight more than 5% of fish body mass (see **Supplementary Table S1** for further details on tag dimensions and weights). Fish were anesthetized using diluted tricaine methanesulfonate (MS-222) and tagged using scalpel incision (tag placed in the intraperitoneal cavity). Morphological measurements of total length (mm) and body mass (g) were recorded. Fish were then placed in separate flow through tanks based on age cohort. Tanks were made of opaque glass fiber with a diameter of 1 m and had a water depth of 40 cm. Age one individuals were placed in eight tanks with a density of N = 75 per tank. Age two individuals were placed in 16 tanks with a density of N = 38 per tank. Fish were drawn from these housing tanks for the laboratory and creek migration experiments (further described, sections Part One: Laboratory Migration and Part two: Small-Scale Field Migration, below).

After tagging, water temperature was too cold to induce feeding (∼0 ◦C), and fish were therefore not administered their feeding treatments until after April 1 (or later for acoustic tagged fish). Four feeding treatments were created to produce fish with varying energetic status by manipulating long- and short-term energy reserves (measured using the change in condition factor from pre- to post-feeding treatment). The start of the feeding treatment, duration, and the final number of fish per treatment varied with experiment, see **Tables 1**, **2** and **Supplementary Table S2** for details. The first treatment group comprised of individuals that were food deprived for the entire duration of the treatment period to create fish with low longand short-term energy reserves (DD, "deprived"-"deprived"). The second treatment group was food deprived until 72 h before the behavioral measures commenced, at which point the fish were fed standard daily portions (described below) to create fish with low long-term energy reserves and high short-term energy reserves (DF, "deprived"-"fed"). The third treatment group was fed standard daily portions until 72 h before the behavioral measures commenced, at which point they were food deprived (FD, "fed"-"deprived") to create fish with high long-term energy reserves but low short-term energy reserves. The fourth treatment group were fed standard daily portions for the entire duration of the treatment period (FF, "fed"- "fed") to create fully fed fish with high long- and short-term energy reserves. The 72 h duration of the short-term treatment ensured that the fish had an empty stomach (i.e., being hungry), because trout have a gastric evacuation rate of <24 h at the water temperature during our treatments (10–12◦C, Elliot, 1972). Treatments were administered equally over the holding tanks and age cohorts. Feed was 1.1 mm sinking pellets for the age 1 fish and 2 mm floating pellets for the age 2 fish (Inicio plus and Inicio 917, BioMar; www.biomar.se) administered in daily standard portions of 45 g until June 16, where the ration increased to 55 g. This amount and quality of the feed has been shown to be sufficient for hatchery reared salmonid juveniles (Alanärä et al., 2014; Persson et al., 2018). Feed was distributed using automatic feeders (TDrum 2000 feeders from Arvo-Tec; www.arvotec.fi) regulated with timer control (Sterner Fish Tech AS; www.fishtech.no).

TABLE 1 | A timeline detailing when each part of the study was conducted over the winter/spring of 2017.


Each part had specific dates when the tagging and feeding treatments began and when the data was collected.

TABLE 2 | Total number of fish used, mean ± (s.d.) pre- and post-treatment mass, mean ± (s.d.) pre- and post-treatment total length (TL), and number and percentage of fish that were classified as migrating during each part of this study.


Information is divided by cohort (age) and treatment group (FF, fed-fed; FD, fed-deprived; DF, deprived-fed; DD, deprived-deprived; see section Fish Tagging and Feeding Treatments for further treatment details)\* .

\*Within experiment and age cohort, all fish did not differ in mass or length at the start of the experiment, except for the 2-year old fish used in Part 2 (creek migration), where the DF group was shorter than the FF and FD groups and lighter than the FF group at the time of tagging.

#### Part One: Laboratory Migration

To measure migration in the laboratory, two identical circular concrete pools (diameter: 11 m) were modified into experimental streams following previous methods (Hellström et al., 2016; Persson et al., 2018; **Figure 1**). Briefly, boundaries were constructed to form a stream course along the outer edge of the pool, which measured 30.1 m in length, 1.5 m wide, and had a water depth of ∼33 cm. For each pool, a portion of the stream concaved (**Figure 1**) and had a shelter structure (40-cm polyvinyl chloride (PVC) pipe cut lengthwise). Two pass-through PIT-tag antennas (Biomark Inc.; www.biomark.com) with accompanying HPR plus tag readers (Biomark Inc.; www.biomark.com) were placed within each stream, ∼6 m apart. Stream flow was counterclockwise throughout the study and kept at a constant velocity for the duration of the study (∼ 0.17 m/s, electromagnetic flow meter, Valeport Model 801).

The laboratory migration trials were conducted between June 15—June 29, 2017. Three trials were conducted per migration pool (i.e., six replicates) with a staggered start (see **Supplementary Table S2**). Within a trial, 20 individuals from each of the four treatment groups were haphazardly selected from each cohort and released into each pool, resulting in 160 fish per pool, per trial. Each trial lasted for 72-h, during which fish migratory behavior was continuously measured using PITtag antenna detections. Fish were not fed in the migration pools. After the 72-h period, individuals were removed and euthanized (via cerebral concussion). Total length (mm) and body mass (g) were again collected (**Table 2**).

Individuals were assigned as either a "migrator" if they obtained at least 10 detections on both PIT-tag antennas in an alternating, consecutive order (i.e., signifying a direction of movement pass the first antenna to the second, back to the first), or "non-migrator" if they had <10. In most individuals a clear pattern of either "migrator" or "nonmigrator" was observed, with most migrators having over 200 detections (see **Supplementary Figures S1A,B** for examples). See

the **Supplementary Materials** for information on the detection efficiency of the antennas. Number of detections per individual was used as a proxy measure of migration distance under laboratory conditions. Individual lap times were calculated by taking the difference in time between detections on the same antenna. Because some individuals lingered in the antenna detection range causing multiple detections within a short timespan a minimal lap time of 10 s was enacted to reduce detection noise. Additionally, not all individuals migrated continuously. To ensure an accurate measure of active swimming that excluded rest time, laps that were longer than 90 s were excluded. Visually inspecting detection-time histograms and identifying the common time when the trough following the first peak of detections occurred determined this cut-off.

#### Part Two: Small-Scale Field Migration

On July 7 2017, a subsample of trout from all treatment groups and both cohorts were released into a natural creek habitat (**Table 1**). The creek was 1–3 m in width, of variable depth from 0.3 to 0.7 m in depth, contained natural complex structure (e.g., boulders, cobble), overhanging branches from the surrounding forest environment, and curvature (e.g., many small bends in the creek's path). Individuals were anesthetized (MS-222) 24 h before release and measured for total length (mm) and body mass (g). The creek was ∼240 m in length, with the release site located at the origin. Within this creek, two PIT-tag antennas and accompanying HPR plus tag readers were installed to track fish downstream migration. The first antenna was located ∼110 m downstream from the release site, whilst the second antenna was located ∼87 m downstream from the first antenna (**Figure 2A**). Both antennas were run continuously for 10 days after release. Fish released in the creek were later categorized as either "migrators" or "non-migrators" based on whether they were detected crossing the first and second downstream PITtag antenna. Migration due to density dependent reasons was thought to be negligible since most of the 2 year olds were expected to leave the creek soon after release and the complex habitat from the release site to the first antenna offered plenty of holding opportunities [additionally, the highest density of 1+ brown trout in the wild in Västerbotten, Sweden was estimated at 55 fish per 100 m<sup>2</sup> (unpublished results), and the density of fish in the creek in our study would have been ∼60 fish per 100 m<sup>2</sup> if all fish stayed and did not leave the creek, see Results]. However, we cannot rule out that some migration movements were by residents exploring the creek environment or that some fish may have migrated past the antennas locations after the 10 day postrelease monitoring period. See the **Supplementary Materials** for information on the detection efficiency of the antennas in the creek.

### Part Three: Large-Scale Field Migration

On May 3 2017, 42 individuals from the age two cohort were selected, anesthetized (MS-222), and surgically tagged in the intraperitoneal cavity with both a PIT-tag (12 mm) and an acoustic tag (Vemco v5-180 khz). The tag transmitted every 0.7 s on the HR coding scheme, and every 30 s on the PPM coding scheme (Guzzo et al., 2018). After tagging the fish were returned to their tank where one of the four feeding treatments began for five more weeks. On June 7, all individuals were again anesthetized (MS-222) and measured for length (mm) and weight (g). After a 24-h recuperation period, they were released into the same natural habitat used for the small-scale field migration. The creek flows directly into the Ume River, which in turn enters the Gulf of Bothnia 30 km downstream. The creek flows into the old river bed of the Ume River about 32 km from the coast of the Bothnian Bay. The annual discharge of the main Ume River is ∼460 m<sup>3</sup> /s and due to the hydropower plant Stornorrfors there is this 8 km long section of old river bed where the discharge varies between 10 and 50 m<sup>3</sup> /s between May 20 and October 1 with higher flows at spring flood and special occasions. For more detailed description of the river system please see Persson et al. (2019). To track the migration in the river, 14 acoustic receivers (Vemco 180 khz VR2W & HR2) were deployed at seven locations dispersed along the river to the sea (**Figure 2B**). One receiver was also deployed 1.27 km upstream the mouth of the creek to monitor if the fish migrated upstream in the river. Receivers operated continuously from the date of release until September 27, 2017.

# Statistical Analyses

#### General Approach

Feeding treatment was used as the explanatory variable and treated as a four-level fixed factor (DD, DF, FD, FF) in all analyses outlined in detail below, unless otherwise stated. The main effect of feeding treatment was inferred using an ANOVA or likelihood ratio test, followed by Tukey's post-hoc analyses to distinguish between treatment levels if needed. In all analyses, we conducted separate models for each age cohort. Model assumptions were visually checked using quantile–quantile and residuals-vs.-fitted diagnostic plots, as well as Shapiro-Wilk and Breusch-Pagan tests. Data were transformed when necessary to meet model assumptions (log or power transformations). Analyses were done in the statistical software R (R Core Team, 2017), using base R

and the packages lme4 (Bates et al., 2015), interval (Therneau, 2015), and survival (Fay and Shaw, 2010) when necessary.

#### Effect of Feeding Treatment on Body Condition

map created with Google Maps.

To determine whether feeding treatment affected fish body condition, the change in Fultons condition factor (Fulton, 1904; hereafter, condition factor) between the start and end of the study was calculated for each fish.

$$Condition\,factor = \,\,\,\,\frac{Weight}{Total\,length} \times 100\,\,\,$$

In all three parts of the study, linear mixed effect model (LMM) or linear model (LM) was used to assess if feeding treatment affected body condition, where the change in condition was the response variable. A random effect of trial ID (migration pool replicate) was included as a random effect for the laboratory study analysis to account for staggered start trial start times (see **Supplementary Table S2**).

#### Part One: Laboratory Migration

For the age two cohort in Part one, feeding treatment was modeled as a two-level factor ("fed" and "deprived") as the treatment groups FD and FF could not be unambiguously distinguished due to a logistical error during the final shortterm feeding treatment. The same applied to the DF and DD treatments. Fish were hence pooled into a "fed" (i.e., FD & FF) and a "deprived" (i.e., DF & DD) treatment group (see **Table 2**) using relative condition factor (see **Supplementary Materials**).

To determine the effect of treatment on the probability that a fish from either cohort would migrate or not, a generalized linear mixed effects model (GLMM) with binomial errors was applied, using migration as a binary response variable (migrating/not migrating) and feeding treatment as a fixed effect. A separate binomial GLMM was used to test if final condition factor predicted probability of migration, with migration as a binary response and final condition factor as a continuous predictor. The effect of feeding treatment on migration distance (number of detections) was tested using a negative binomial GLMM appropriate for over-dispersed count data. To test the effect of feeding treatment on lap time, the mean lap time of an individual was used as a response variable in a LMM. In all the above analyses, trial ID (migration pool replicate) was included as a random effect.

#### Part Two: Small-Scale Field Migration

The effect of feeding treatment on the probability to migrate was tested using a GLM with binomial errors, treating migration as a binary response variable (migrating/non-migrating). A separate binomial logistic GLM was used to test if final condition factor predicted probability of migration, with migration as a binary response and final condition factor as a continuous predictor. The effect of feeding treatment on migration speed in the creek was tested using a LM, treating the difference in time between detection at the first and the second antenna per individual as a response variable and feeding treatment as fixed effect.

#### Part Three: Large-Scale Field Migration

Migration success (detection from one receiver to the next) of the released individuals was performed using an interval-censored survival analysis. Each receiver acted as intervals and time of unique last detection for each individual at each receiver was used to determine an individual's last interval of survival. To test whether the survival (or distance moved) of the juveniles varied with feeding treatment, we used a non-parametric maximum likelihood (NPMLE) permutation test suitable for intervalcensored data with small sample sizes (Fay and Shaw, 2010; Therneau, 2015).

## RESULTS

#### Part One: Laboratory Migration

Feeding increased condition factor in both age cohorts [Age one: LMM N = 237, F(3, 229) = 229, P < 0.0001; Age two: LMM N = 255, F(1, 249) = 221, P < 0.0001; **Figures 3A,B**]. Within the age one cohort, fish from the FF and FD treatments had a greater change in condition factor than the DF and DD treatment groups, which were not statistically different from each other (Tukey, all contrasts P < 0.001 except DD-DF P = 0.91; **Figure 3A**).

There was a large difference between the age cohorts in the proportion of fish migrating (**Table 2**). For the age one cohort, only 39% of the fish migrated in the migration pools, while 94% of the age two fish migrated. In the age one cohort, FF and FD individuals were more likely to migrate than both the food deprived treatments (Binomial GLMM N = 237, LRT = 126, P < 0.0001, **Figure 3C**; **Table 2**; Tukey, FD-FF Z = 5.23 P < 0.001, DF-FF Z = 6.34 P < 0.001, DD-FF Z = 6.54 P <0.001, DF-FD Z = 2.62 P = 0.04, DD-FD Z = 3.04 P = 0.012). Regardless of treatment, age one fish with higher condition factor at the end of the experiment were more likely to migrate (Binomial GLMM, N = 237, LRT = 146.91, P < 0.0001; **Figure 4A**). In the age two cohort, almost all fish were migrating and feeding treatment did not affect whether or not a fish migrated (Binomial GLMM, N = 255, LRT = 0.95 P = 0.33; **Figure 3D**). Likewise, condition factor at the end of the experiment did not predict if age two fish would migrate (Binomial GLMM, N = 255, LRT = 0.51 P = 0.47; **Figure 4B**).

Of the fish that were actively migrating, age one fish that were fed (FF, FD) migrated a further distance than fish that were fooddeprived (DF, DD) (number of detections; Negative Binomial GLMM, N = 92, LRT = 43.4, P < 0.0001; **Figure 3E**; Tukey: DF-FF Z = −6.38 P < 0.001, DD-FF Z = −4.56 P < 0.001, DF-FD Z = −4.44 P < 0.001, DD-FD Z = −2.74 P = 0.03). Each treatment group migrated the equivalent of 30.8 (FF), 27.5 (FD), 7.3 (DF), or 9.2 (DD) km. In the age two cohort, fish from the fed treatment group tended to migrate a greater distance than the food deprived treatment group, but this did not reach significance (Negative Binomial GLMM N = 239, LRT = 3.65, P = 0.056; **Figure 3F**). Age two fish migrated the equivalent of 28.4 or 24.5 km in the fed and deprived treatments, respectively.

Feeding increased average lap time in age one fish [LMM N = 92, F(3, 88) = 14.4, P < 0.001; **Supplementary Figure S2A**]. The FF treatment group completed an average lap faster than both of the food deprived treatments (Tukey: DF-FF Z = 5.97 P < 0.001, DD-FF Z = 3.38 P = 0.004, DF-FD Z = 4.14 P < 0.001), with each treatment group taking an average ± s.d. of 1.02 ± 0.1 (FF), 1.07 ± 0.1 (FD), 1.20 ± 0.1 (DF), or 1.12 ± 0.2 (DD) min to complete a lap, which corresponds to a swimming speed of 0.49 (FF), 0.46 (FD), 0.42 (DF), and 0.44 (FF) m/s (relative to ground distance). Feeding also increased average lap time in age two fish [LMM N = 239, F(1, 232) = 18.2, P < 0.0001; **Supplementary Figure S2B**], with each treatment group taking an average ± s.d. of 0.93 ± 0.05 (fed) or 0.96 ± 0.06 (deprived) min to complete a lap, which corresponds to a swimming speed of 0.53 (fed) and 0.50 (deprived) m/s (relative to ground distance).

#### Part Two: Small-Scale Field Migration

Similar to the laboratory study, feeding treatment affected the change in condition factor for both age cohorts. In the age one fish, both fed treatment groups (FF, FD) had a more positive change in condition factor compared to the food deprived [LM, N = 194, F(3, 190) = 88.7, P < 0.0001; Tukey: all contrasts P < 0.001, except DD-DF P = 0.38; **Figure 5A**]. In the age two cohort, fed treatment groups also had larger positive change in condition factor compared to food deprived treatment groups [LM, N = 127, F(3, 123) = 94, P < 0.0001. Tukey: all contrasts P < 0.001 except FD-FF P = 0.0049; **Figure 5B**].

The proportion of age one fish migrating was low, with 11% of fish migrating past antenna 1 and only 4% of fish migrating past antenna 2. Feeding increased the probability that age one fish would migrate to antenna 1 in the field (Binomial GLM, N = 194, LRT = 22.3, P < 0.0001; **Figure 5C**). This effect was similar for antenna 2, but did not reach statistical significance (Binomial GLM, N = 194, LRT = 7.27, P = 0.064). Tukey posthoc contrasts among all treatments were not possible because no fish migrated in the DF group (i.e., data separation; DD-FF Z = −2.38 P = 0.06, FD-FF Z = −1.22 P = 0.56, DD-FD Z = −1.26 P = 0.54; **Figure 5C**). In contrast, most fish migrated in the age two cohort with 79% of the fish migrating past antenna 1 and 77% migrating past antenna 2. There was no significant effect of feeding treatment on probability to migrate past antenna 1 (Binomial GLM N = 127, LRT = 5.28, P = 0.15; **Figure 5D**) or antenna 2 (Binomial GLM N = 127, LRT = 3.24, P = 0.36) in the age two cohort. Regardless of treatment, age one fish with higher condition factor at the end of the experiment were more likely to migrate (Antenna 1: Binomial GLM, N = 194, LRT = 19.6, P < 0.0001, **Figure 4C**; Antenna 2: Binomial GLM, N = 194, LRT = 4.65, P = 0.031). Condition factor at the end of the experiment did not predict if age two fish would migrate (Antenna 1: Binomial GLM, N = 127, LRT = 2.27, P = 0.13; **Figure 4D**; Antenna 2: Binomial GLM, N = 127, LRT = 0.26, P = 0.61).

The effect of feeding treatment on migration speed could not be analyzed in the age one cohort due to small sample size (i.e., very few individuals detected at the first or second antenna, **Table 2**). In the age two cohort, feeding increased migration speed (LM, N = 92, χ <sup>2</sup> = 98.4, P = 0.0003; **Figure 5E**), with both fed treatment groups migrating faster than the food deprived treatment groups (Tukey: DF-FF t = 2.6 P = 0.05, DD-FF t = 3.7 P = 0.002, DD-FD t = 3.5 P = 0.005).

#### Part Three: Large-Scale Field Migration

As with the first two studies, feeding resulted in an increased change in condition factor [LM, N = 42, F(3, 38) = 7.07, P = 0.0007; **Figure 6A**], with change in condition factor declining with less feed (Tukey, DD-FF t = −4.2 P = 0.001, DD=FD t = −3.8 P = 0.003, remaining contrasts P > 0.1).

After release, 33 fish were detected leaving the creek and entered the river between 1 and 48 h later. Once in the river,

detection numbers dropped to 12 unique individuals (28%) by 0.8 km from release site. By 8 km downstream from release site, only six unique individuals were detected and these six continued 20 km downstream, where only four were then detected in the river delta 29 km downstream. It took between 5 and 17 days for these four fish to migrate out of the river system. Three of the four out migrating fish were from the treatment group FF, while one was from treatment group DD. Feeding treatment had no effect on migration success (NPMLE permutation test, N = 42, χ 2 = 6.84, P = 0.07; **Figure 6B**). In addition, five of the 42 released fish migrated upstream through a fish ladder to be detected on a receiver located in the reservoir above a large hydropower dam

(1.3 km upstream the release site). Upstream migrating fish were not included in the survival analysis. Of these five fish, two came from the treatment group DD, two came from the treatment group FD, and one came from the treatment group FF.

#### DISCUSSION

#### Efficacy of Feeding Treatments in Manipulating Energetic Status

As expected, the feeding treatments used in our study manipulated long-term energy reserves in brown trout (as measured by change in condition factor), even when the duration of treatment was as little as 5 weeks. The greatest difference was between the treatment groups with or without depleted long-term energy reserves (FF vs. DD groups). In contrast, the energetic status created from fish with altered short-term energy reserves (FD vs. DF) was often intermediate, less consistent across experimental contexts, and sometimes not statistically different from the long-term treatments. Previous studies have similarly used altered feeding regimes to manipulate long-term energy reserves (Wysujack et al., 2009; Lans et al., 2011; Persson et al., 2018), but the best of our knowledge, this is the first study to include very short-term feeding regime manipulations aimed at altering short-term energy reserves.

#### Laboratory and Small-Scale Field Studies: Migration Decisions and Migratory Behaviors

In both the laboratory and the small-scale field study, the majority of the age two fish migrated independent of their feeding treatment. This finding is in contrast to previous studies showing that starvation for a period of at least 5 months

increased the proportion of trout that smoltified (Wysujack et al., 2009) and the proportion of migrating juvenile trout (Davidsen et al., 2014, see Midwood et al., 2016 for short-term effects). Our results suggest that migration for age two trout was independent of the energetic status in the spring (i.e., ish may have surpassed a size or condition threshold) and/or the decision to smoltify and migrate had been made before the feeding treatments started in April. It is well-established for Atlantic salmon that the switch for smoltification is activated during the late summer or early autumn prior to smoltification

the following spring, and it is dependent on the physiological status of the fish and the rate of change of that status in relation to genetic thresholds (Thorpe et al., 1998). Regional environmental conditions may also influence these thresholds creating a geneby-environment interaction (Økland et al., 1993; Jonsson and Jonsson, 2011). Persson et al. (2018) did not reverse the pathway toward smoltification and migration in Atlantic salmon despite very harsh starvation treatments during the winter and spring prior to migration. The authors concluded that the decision to migrate or threshold for migration may have been made or surpassed before their treatments began. Midwood et al. (2014) suggested that brown trout have an autumn activation window similar to Atlantic salmon, because the proportion of brown trout migrating in their study were unaffected by cortisol treatments in the spring. Näslund et al. (2017) found that reducing the feed ration in the laboratory for 1 month in October prior to migration decreased the proportion of wild juvenile brown trout that smoltified in spring. However, Jones et al. (2015) found that starvation over the prior autumn and winter did not predict which juvenile fish would smoltify, while starvation in the spring increased the likelihood of smoltification.

In contrast to the age two fish, feeding increased the proportion of age one fish migrating in both the laboratory and the small-scale field study. Thus, the proportion of migrating age one trout was modified by different feeding regimes during the spring, but in an opposite direction to the majority of previous studies with hatchery-reared brown trout (Wysujack et al., 2009; Davidsen et al., 2014; Jones et al., 2015). At the time of migration, the fed age one fish weighed 2.5–3 times more and had substantially higher condition factor compared to the age one fish that had been food deprived. Smoltification and juvenile migration are energetically costly processes for trout (Folmar and Dickhoff, 1980), and the age one fish with low energetic status might not have had enough energy to migrate. The average size of wild brown trout smolts from the rivers in the region of this study is 150–225 mm (Larsson et al., 2012; data collected within the European Data Collection Framework, ICES International Cooperation for Exploration of the Sea, 2018), which is larger than the size of the food deprived age one trout in our study (average length 116–119 mm; **Table 2**). Deprived age one fish in our study may not have reached sufficient size and/or energetic status to become smolt and therefore postponed migration, unlike age two fish that migrated despite a period of starvation. Most studies investigating effects of energetic status on hatchery reared trout migration have used age two juveniles. Yet, age one smolt are very common in hatcheries today (ICES International Cooperation for Exploration of the Sea, 2018) and age one smolts also occur in nature at southern latitudes (L'Abée-Lund et al., 1989). Our results show the importance of also considering younger age classes when conducting smolt migration studies as the response may differ across ontogeny.

For both ages in our study, high energetic status increased the lap time and the distance migrated (together, per unit time: migration speed). High migration speeds may have implications for fitness, because smolt that spend more time in the river have been shown to experience decreased energetic status and survival (Thorpe and Morgan, 1978; Peake and McKinley, 1998; Aarestrup et al., 2005; Salminen et al., 2007). In our study, fish with low energetic status may have been too depleted of energy to keep a similar migration speed as fish of higher energetic status. Also, in a natural river environment, fish with low energetic status may divert time from migration to feed and sustain the energy needed for migration. Boel et al. (2014) found that smolt with the lowest energetic status were more likely to stop at the first feeding opportunity compared to smolt of intermediate energetic status that migrated the longest distance (fish with the highest energetic status stayed as resident in the river in their study). Of the few studies that have investigated impact of energetic status on migration speed in trout, both Lans et al. (2011) and Larsson et al. (2012) found that starvation increased migration speed, while Davidsen et al. (2014) could not determine any effect. Noteworthy is that all of these studies were performed in the field by tracking smolt migration in the river and may therefore not be fully comparable with our laboratory and small-scale field study.

We did not find any clear effects of the treatment aimed to manipulate short-term energy reserves or "hunger" on the probability, speed or distance of trout migration. As the levels of glycogen influence the capacity for physical activity and endurance in fish (Hammer, 1995), we expected these shortterm depleted fish to migrate slower or a shorter distance than fish that were fed. Contrarily, one could also expect "hunger" to increase migration speed as "hunger" is known to increase risk taking behavior in fish (Pettersson and Brönmark, 1993), and increased risk-taking has been linked to increased migration intensity in Atlantic salmon smolt (Hellström et al., 2016). The lack of effect seen in this study may indicate that the 72 h starvation period was too short to alter migration due to "hunger" in the well fed fish (FD), or due to replenished energy reserves in food deprived fish (DF). It is also possible that the behavioral effects of temporary "hunger" vary more among individuals than their responses to the long-term feeding treatment.

#### Large-Scale Field Migration

We could not detect any effect of feeding treatment on migration success in age two fish in the large-scale field study. However, the sample size was very low because only four fish managed to reach the coast. The majority of fish did begin migrating (i.e., they left the release site in the creek), but quickly disappeared once reaching the main river, most likely due to avian or piscivorous predators that are abundant in the area. Mortality can be high during in-river migration for brown trout smolt (Thorstad et al., 2012), especially for hatchery reared smolt who lack predator experience (Huntingford, 2004). High predation may mask any effect of energetic status on migration, and a larger sample size would be needed to discern any difference between treatments in the river system used in this study. It is noteworthy that five of the fish undertook a substantial upstream migration in the river, even navigating a long fish ladder. Similar extensive upstream migration could not be detected either in the laboratory migration pool or in the small creek, highlighting the importance of investigating aspects of salmonid migration over multiple scales. We also cannot exclude that this behavior was a consequence of being predated by a larger upstream migrating predator.

# CONCLUSIONS

Our findings show that high energetic status in 1 and 2 year old brown trout increased swimming speed and migration distance. We also showed that the proportion of migrating age one fish can be modified by manipulating the energetic status during early spring, but that the same manipulation does not affect the proportion of migrating age two fish. These findings suggest that there are different thresholds or windows of activation (or potentially inhibition) that vary with age or physiological aspects related to age (size, growth, energetic status etc.). Specific to our study, fed 1 year old fish that were large and in good condition migrated, whereas deprived 1 year old fish that were smaller and in poorer condition appeared to delay migration for another year, and almost all 2-year old fish migrated regardless of condition because a threshold or migration decision was surpassed prior to the treatments. This study is one of very few that has tested the effect of energetic status on juvenile migration over multiple spatial scales, and the fact that the results were largely repeatable both in the laboratory setup and in the smallscale field study suggests high confidence in the results. Our results also suggest reconsidering the previous recommendations for more restrictive feeding regimes in hatcheries to enhance migration (Serrano et al., 2009; Larsson et al., 2012). Future studies should aim to identify when in time juvenile trout makes the decision to smoltify by tracking the energetic status of juveniles over longer periods, if possible from birth until smoltification and migration.

# DATA AVAILABILITY STATEMENT

The datasets generated for this study are available in the **Supplementary Material** and on request from the corresponding author.

# ETHICS STATEMENT

This study and the methods used within it were approved under animal ethics permit a11-13 to AA and a20-18 to GH from Jordbruksverket.

# AUTHOR CONTRIBUTIONS

SS, GH, LP, and AA designed this study. SS and EM collected the data and conducted the statistical analyses. GH and EM wrote the manuscript together, with input from SS, LP, and AA. GH and AA secured the funding for this study.

# FUNDING

This research was supported by an Energiforsk grant VK12006 to AA and GH.

# ACKNOWLEDGMENTS

The authors would like to thank the Vattenfall Norrfors fish hatchery staff and Daniel Cerveny for laboratory assistance.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00411/full#supplementary-material

# REFERENCES


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Shry, McCallum, Alanärä, Persson and Hellström. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# The Interplay Between Extrinsic and Intrinsic Factors in Determining Migration Decisions in Brown Trout (Salmo trutta): An Experimental Study

Louise C. Archer 1,2 \*, Stephen A. Hutton1,2, Luke Harman1,2, Michael N. O'Grady <sup>3</sup> , Joseph P. Kerry <sup>3</sup> , W. Russell Poole<sup>4</sup> , Patrick Gargan<sup>5</sup> , Philip McGinnity 1,4 and Thomas E. Reed1,2

*<sup>1</sup> School of Biological, Earth and Environmental Sciences, University College Cork, Cork, Ireland, <sup>2</sup> Environmental Research Institute, University College Cork, Cork, Ireland, <sup>3</sup> Food Packaging Group, School of Food and Nutritional Sciences, University College Cork, Cork, Ireland, <sup>4</sup> Marine Institute, Galway, Ireland, <sup>5</sup> Inland Fisheries Ireland, Dublin, Ireland*

#### Edited by:

*Nathan R. Senner, University of South Carolina, United States*

#### Reviewed by:

*Jonathan Paul Velotta, University of Montana, United States Frederick Gilbert Whoriskey, Dalhousie University, Canada*

> \*Correspondence: *Louise C. Archer l.archer@umail.ucc.ie*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *28 March 2019* Accepted: *29 May 2019* Published: *14 June 2019*

#### Citation:

*Archer LC, Hutton SA, Harman L, O'Grady MN, Kerry JP, Poole WR, Gargan P, McGinnity P and Reed TE (2019) The Interplay Between Extrinsic and Intrinsic Factors in Determining Migration Decisions in Brown Trout (Salmo trutta): An Experimental Study. Front. Ecol. Evol. 7:222. doi: 10.3389/fevo.2019.00222* Many species are capable of facultative migration, but the relative roles of extrinsic vs. intrinsic factors in generating diverse migratory tactics remain unclear. Here we explore the proximate drivers of facultative migration in brown trout in an experimental laboratory setting. The effects of reduced food, as a putative environmental cue, were examined in two populations: one that exhibits high rates of anadromy (sea-migration) in nature, and one that does not exhibit anadromy in nature. Juveniles derived from wild-caught parents were reared for 2 years under four environmental treatments: low food in years 1 and 2 (Low-Low); high food in years 1 and 2 (High-High), low food in year 1 and high in year 2 (Low-High), and vice versa (High-Low). Food restriction had a significant effect on migratory tactics, with the frequency of smolts (juveniles choosing migration) highest in the Low-Low treatment in both populations. No individuals became smolts in the High-High treatment, and intermediate smolting rates were observed in the Low-High and High-Low treatments. Higher overall smolting rates in the naturally anadromous population suggested an inherited component to anadromy/migration decisions, but both populations showed variability in migratory tactics. Importantly, some fish from the naturally non-anadromous population became smolts in the experiment, implying the capacity for migration was lying "dormant," but they exhibited lower hypo-osmoregulatory function than smolts from the naturally anadromous population. Tactic frequencies in the naturally anadromous population were more affected by food in the 2nd year, while food in the 1st year appeared more important for the naturally non-anadromous population. Migratory tactics were also related to sex, but underpinned in both sexes by growth in key periods, size, and energetic state. Collectively these results reveal how migration decisions are shaped by a complex interplay between extrinsic and intrinsic factors, informing our ability to predict how facultatively migratory populations will respond to environmental change.

Keywords: climate change, partial migration, anadromy, aquatic, brown trout, genotype by environment, Salmo trutta, proximate drivers

# INTRODUCTION

Intraspecific phenotypic variation accounts for much of the diversity of form and function in nature (Roff, 1996). Understanding the mechanisms generating and maintaining divergent phenotypes and life histories within and among populations is thus a fundamental goal of evolutionary ecology, with applied relevance to conservation and wildlife management (Naish and Hard, 2008). A particularly striking example of alternative phenotypes is the phenomenon of facultative migration, whereby individuals within a population vary in their migratory tendencies. Facultatively migratory populations can comprise a mixture of migrant and resident individuals (sometimes called "partial migration"), with migration at specific life stages occurring typically to take advantage of alterative foraging opportunities or avoid adverse abiotic (e.g., climatic) conditions (Chapman et al., 2011a). Despite its widespread occurrence across taxa and regions, fundamental gaps still exist in our understanding of proximate and ultimate drivers of facultative migration. In particular, there is a dearth of studies addressing how facultatively migratory species respond to environmental change (Doswald et al., 2009; Chapman et al., 2011b), limiting our ability to generalize about the impacts of anthropogenic factors on migratory species and to effectively manage their populations.

Polymorphisms such as facultative migration are potentially underpinned by a complex mapping between genotype and phenotype, i.e., phenotypic similarity can arise from different genotypes, or the same genotypes can produce dramatically different phenotypes through plasticity mediated by environmental cues (Roff, 1996). As such, migration and residency have often been considered as environmentallytriggered alternative phenotypes/tactics produced by an evolvable conditional strategy, where optimal tactic choice in a given context is conditional on extrinsic or intrinsic cues (Chapman et al., 2011b). This interplay between proximate and ultimate drivers of conditional strategies has been formalized as the so-called "environmentally cued threshold model" (Tomkins and Hazel, 2007). Within this framework, alternative tactics are controlled by an environmentally-sensitive status trait (e.g., physiological condition, energy state) and an inherited threshold, or "switch point," which is assumed to be genetically variable. An individual assesses their status trait and, for example, adopts a resident tactic if it exceeds their inherited switch point, otherwise it switches to a migratory tactic. Individual physiological condition/energy state is strongly influenced by the environment, and so the assessed status trait can vary relative to the intrinsic threshold depending on external conditions; for this reason, the status trait can be thought of as an "environmental cue" and the step function relating tactic expression to cue as a "threshold reaction norm" (Tomkins and Hazel, 2007; Piche et al., 2008; Pulido, 2011; Buoro et al., 2012). There is some evidence for genetic variation in thresholds for alternative tactics, e.g., in blackcaps Sylvia atricapilla (Pulido et al., 1996) and Atlantic salmon Salmo salar (Piche et al., 2008), but detailed understanding of how external environmental variation is translated into internal physiological signals, on which migratory decisions are then based, is lacking.

Salmonine fishes (salmons, trouts, and charrs) are excellent models for disentangling causes of facultative migration as they display wide variation across a continuum of migratory strategies, coupled with obligate freshwater spawning (Klemetsen et al., 2003; Ferguson et al., 2019). Individuals can remain in freshwater post hatching for their entire life cycle, either staying in their natal stream or lake (residency tactic) or undertaking an adfluvial migration that takes them to a larger river or lake (potamodromous tactic) (Dodson et al., 2013; Ferguson et al., 2019). Facultative anadromy is an extreme form of this conditional migration strategy, where some individuals adopt the residency tactic whilst others from the same population undertake a marine migration (involving anywhere from tens to thousands of kilometers of directed movement between freshwater and saltwater). This is followed by a period of marine or estuarine feeding and growth (from months to years), before returning to spawn in natal streams (Jonsson and Jonsson, 1993). Populations can contain both resident and migratory (anadromous or potamodromous) forms, or be dominated by one life history type (Chapman et al., 2012). Both forms can breed freely in sympatry, and although offspring tend to track the tactics of their parents, either life history can be produced from a given migratory phenotype (Zimmerman and Reeves, 2000; Berejikian et al., 2014). Such flexibility indicates an interplay between genetic predisposition and environmental conditions experienced i.e., genotype by environment interactions, underpinning facultative migration (Hutchings, 2011).

The threshold reaction norm framework has been useful in understanding migratory decisions in salmonines (Hutchings and Myers, 1994; Thorpe et al., 1998; Thériault et al., 2007). If during a key decision window an individual's status trait exceeds their predetermined threshold, the fish adopts a residency tactic leading to maturation in freshwater; if not, maturation is deferred in favor of migration (Dodson et al., 2013; Kendall et al., 2014; Ferguson et al., 2017). However, the proximate factors on which individuals base the migration decision remain unclear. Previous studies have focused on a range of aspects of physiological state/energy status that may influence migratory tactics such as body size (Thériault and Dodson, 2003), lipid reserves (Jonsson and Jonsson, 2005), body condition (Hecht et al., 2015), growth (Jonsson, 1985), growth efficiency (Forseth et al., 1999; Morinville and Rasmussen, 2003), and metabolism (Sloat and Reeves, 2014). While body size is often used as a surrogate for, or argued to itself be, the status trait triggering alternative migratory tactics, the associations here have been varied and inconclusive. Larger sizes and faster growth rates have been associated with early age at migration (Jonsson, 1985), whereas others have found no size-based differences between migrants and non-migrants at a given age (Thériault and Dodson, 2003), or conversely found larger sizes (and higher lipid reserves) to be associated with freshwater maturation in lieu of anadromy (McMillan et al., 2012). These inconsistencies could reflect species' specific responses, and thus require further exploration to establish potential status traits for a given species. Studies might also be inconclusive because size is typically measured sometime after the migratory decision itself, perhaps at the parr-to-smolt transformation stage, and size at migration may not accurately reflect size when the decision was made. For example, residents may have meanwhile diverted energy into maturation and gonadal development at the expense of somatic growth (Tocher, 2003), while migrants may undergo accelerated growth as the migration itself approaches (Metcalfe, 1998).

Moreover, there may be at least two separate threshold decisions: an early one determining whether a fish will migrate per se or not, and a later one determining whether fish on a migratory trajectory actually migrate this year or defer migration to an older age (Ferguson et al., 2019). Size may be the cue used for the second decision, given that survival on entry to the sea or a lake is typically positively related to size (Klemetsen et al., 2003; Phillis et al., 2016). Yet, size at the migration point may be unrelated to, or inconsistently related to, the status trait triggering the initial migration decision, which could occur considerably earlier than the point at which migrants and resident become phenotypically distinguishable (Beakes et al., 2010). Identifying the key proximate drivers of migration is therefore complicated by the fact that the exact time windows for each of these putative decisions may not be known a priori, while correlations among physiological, energy status and growth traits may be variable across ontogeny or contexts. In the particular case of facultative anadromy, sea-migration requires a suite of adjustments in preparation for life in saltwater and therefore the physiological remodeling process, which includes changes in osmoregulation, coloration, and body shape (Tanguy et al., 1994), is likely to begin sometime in advance of the migratory period. The existence of early "decision windows" that initiate divergent life-history trajectories in salmonine fishes (Thorpe and Metcalfe, 1998; Thorpe et al., 1998) has some empirical support; for example, body condition of anadromous O. mykiss was found to be significantly lower than resident counterparts within a year of hatching and a full 12 months prior to emigration (Hecht et al., 2015).

Although the proximate drivers of migration in salmonines are unresolved, there is some consensus that potamodromous or anadromous migratory tactics are promoted by energetic limitation in natal rivers, which prevents fish reaching the inherited physiological threshold for maturation as residents (Kendall et al., 2014). Energetic limitation can arise through an interplay between environmental factors and intrinsic physiological state; for example, if freshwater food resources are insufficient to support growth rates or metabolic demands, then migration could be triggered that takes the fish to a better feeding environment such as the sea or a large lake (O'Neal and Stanford, 2011; Sloat and Reeves, 2014; Jones et al., 2015). Food limitation arising from competition at high population densities has also been shown to increase the proportion of adfluvial migratory brown trout, whereas low population densities have been associated with residency and maturation (Olsson et al., 2006; Wysujack et al., 2009). It remains largely unknown, however, during which ontogenetic stages food limitation is most important to migration decisions.

Brown trout (Salmo trutta) are an interesting model for understanding facultative migration as they exhibit highly variable strategies, with some individuals/populations remaining resident in their natal stream their entire lives, while others migrate to a larger river, a lake, an estuary, or the sea (Jonsson and Jonsson, 1993; Klemetsen et al., 2003; Cucherousset et al., 2005; Ferguson et al., 2019). Here we present the results of an experimental laboratory study of brown trout that involved F1 progeny of wild-caught parents from two populations that exhibit divergent migratory life-histories in nature. Our primary aim was to explore the interaction between intrinsic proximal factors (which may encompass both inherited and non-inherited variation) and the extrinsic environment in generating alternative migratory tactics in brown trout. Specifically, we aimed to: (i) assess the relative importance of food availability and inherited differences between populations in determining alternative migratory tactics; (ii) determine whether food restriction was more important in the first year or second year of freshwater rearing; (iii) test for differences between our two populations in their response to food restriction and its timing, which may be indicative of genotype-by-environment interactions influencing tactic frequencies, and (iv) explore associations between status traits (length, weight, condition factor) and migratory tactics. We expected that food restriction would increase the frequency of the migratory tactic overall. While we expected migratory tactic frequencies to vary overall between fish from our two population backgrounds, we also anticipated that the naturally non-anadromous stock might produce migratory phenotypes when subjected to reduced food, given that migration may only be expressed under certain environmental conditions (Roff, 1996; Pulido, 2011).

# MATERIALS AND METHODS

#### Study Populations

Wild-origin brown trout brood stock were obtained by seine netting from the Burrishoole (53◦ 57′ N: 09◦ 35′ W) and Erriff (53◦ 37′ 0.00′′ N: 09◦ 40′ 17.10′′ W) catchments in the west of Ireland in November 2015. Burrishoole brood stock were caught in Lough Bunaveela (46 ha, **Figure S1**) in the headwaters of the catchment. A local population of non-anadromous trout remain resident in Lough Bunaveela for most of their lifecycle, bar very short-distance directed movements (on the order of 10–100 s of meters) between the lake and two spawning rivers (one inflowing to the lake, the other outflowing). No obvious genetic structure at neutral microsatellite markers is evident between these spawning rivers, implying trout from Lough Bunaveela comprise a single panmictic population (R. Finlay, pers. comm.). A large run of sea trout (typically 2000+ anadromous recruits annually) occurred in the Burrishoole catchment up to 30 years ago. The Burrishoole anadromous trout run collapsed in the late 1980s, coinciding with sea-lice outbreaks following the establishment of salmon aquaculture farms in the downstream estuary. The exact spawning locations of the historic anadromous individuals within the Burrishoole catchment remain uncertain, and we cannot exclude the potential for some anadromous fish having contributed to the Bunaveela population before the anadromous population collapse. Nevertheless, despite Bunaveela spawning streams being accessible to anadromous migrants, there is little to no evidence that the Bunaveela population produced anadromous trout historically or recently (Poole et al., 2007; Magee, 2017) and we thus consider it a population that rarely, if ever, expresses anadromy.

Erriff brood stock were caught in Tawnyard Lough, a small upland lake (56 ha) on the western side of the Erriff catchment (the National Salmonid Index catchment) that is fed by a primary inflowing stream, the Glendavoch River and a number of smaller tributaries (**Figure S1**). The vast majority of trout spawned in the Glendavoch River are believed to disperse as fry or parr to Tawnyard Lough (a distance of a few 100 meters to a few kilometers, depending on how far up the Glendavoch River spawning occurred), although a small fraction remain permanently resident in the natal stream (P. Gargan, pers. comm.). A large run of out-migrating anadromous juveniles (in the range of 500–3,000 smolts per year over the last 30 years) is enumerated annually in a trap at the outflow of Tawnyard Lough (Gargan et al., 2016). The remaining fish never go to sea but instead spend several years growing in the lake, before returning to spawn in the Glendavoch River and smaller tributaries once mature. Brood stock from the Tawnyard population used in this experiment putatively comprised a mix of anadromous and non-anadromous fish, assumed to represent naturally occurring frequencies of anadromous and non-anadromous tactics (see **Table S1** for details of brood stock), with local expertise indicating that the Tawnyard population in general shows high rates of anadromy (P. Gargan, pers comm.). In summary, we consider the Tawnward population to have a strong migratory/anadromous background, and the Bunaveela population to have essentially no (recent) anadromous background and to exhibit only limited local movements. For ease of reading, juveniles derived from Tawnyard parents are hereafter referred to simply as the "anadromous-background" population and juveniles from Bunaveela parents as the "nonanadromous background" population.

#### Fish Rearing

Females were stripped of eggs, and the eggs of each female were divided into two batches, each fertilized by the milt of a single male from the same source population (i.e., Tawnyard or Bunaveela; see **Table S1** for full details on crossing). Fertilized eggs were then incubated in standard Heath trays in a hatchery facility located within the Burrishoole catchment. Surviving unfed fry (2–3 weeks prior to exogenous feeding) were transferred to a rearing facility at University College Cork (Aquaculture and Fisheries Development Center). While transitioning to exogenous feeding, fry were held in 100 L growth tanks on a recirculating aquaculture system (RAS) with bio filtration, and fed ad libitum to satiation using commercially available trout pellets (Skretting Ltd, Norway). The populations were kept separately in two 100 L tanks during this initial rearing phase and maintained under a natural temperature regime regulated by a single conditioning unit. Once the fry had transitioned to exogenous feeding (June 2016), they were fed ad libitum with commercial trout pellets for a period of 2 months. All fish experienced the same constant photoperiod regime (12 h of light and 12 of dark) during this initial rearing phase.

In September 2016, fish were randomly allocated into four 100 L tanks in the same RAS as described above (two tanks for Tawnyard and two tanks for Bunaveela), at which point the experimental phase began and food manipulations were initiated (see next section for experimental treatments). A random subset of fish (n = 200 per population) were given individual identifier tags using unique color combinations of visible implant elastomer tags (Northwest Marine Technology Ltd., USA). To facilitate growth, in December 2016 the fry were transferred (within their experimental groups) to 520 L growth tanks in a larger RAS in the same aquaculture facility. Continuous through flow of water prevented any waste accumulation in tanks, with returning water passed to a central holding sump and treated via mechanical filtration, protein skimming, bio filtration, and ozone and UV sterilization. Water quality in the system was monitored weekly, and levels of pH, nitrate, nitrite, and ammonia were within acceptable ranges for optimal fish health. During the experimental phase, the fish experienced a seasonally-changing photoperiod and temperature regime typical of the west of Ireland, simulated via an automated lighting system of LED lights (BioLumen, UK) above each tank and a single conditioning unit. Negligible natural mortality occurred during the experimental phase but to maintain total biomass in the RAS at acceptable levels from a water quality perspective, fish were randomly culled (n = 120 in total across all tanks) over the course of the 2 years of tank rearing, with equal fish densities maintained between food treatments. Fish that were prematurely culled were excluded from all analyses. Full details on the stripping, crossing and rearing procedures are given in **Supplementary Information**.

#### Experimental Design

The experimental phase ran for a 22 month period, from September 2016 to June 2018, with all fish humanely euthanized at the end of the experiment under license (the study and all associated procedures were carried out with ethical approval from Health Products Regulatory Authority (HPRA) Ireland, under HPRA project license AE19130/P034, and HPRA individual licenses AE19130/1087, AE19130/I200, AE19130/I201, and AE19130/I202).

To investigate the relative importance of the extrinsic environment (food supply) and intrinsic inherited factors (population-of-origin) in determining migratory tactics, juveniles from the anadromous and non-anadromous background populations were divided evenly and allocated randomly across four tanks receiving water from the same recirculating source, each experiencing a different feeding regime over the experimental phase. Populations were kept separately for the duration of the study (n = 90 per feeding treatment per population, at the beginning of the experimental phase). Great care was taken to ensure that all measured variables other than feeding regime (fish densities, temperature, photoperiod, lux, flow rates) were constant across the tanks. The four feeding regime treatments were designed to test the effects of food restriction in the early vs. late periods of this experimental phase, with each period corresponding to ∼11 months [chosen because similar periods of c. 9 months have been reported to alter adfluvial migration rates in trout (Olsson et al., 2006)]. These four food regimes were as follows: (i) High-High treatment: fish fed recommended daily pellet rations for optimal growth in both periods, calculated as a percentage of their body weight and adjusted for seasonally-changing temperatures (Skretting Ltd, Norway); (ii) Low-Low treatment: fish fed 25% of recommended optimal rations in both periods; (iii) High-Low treatment: fish fed 100% of optimal daily rations in the first period and 25% of optimal daily ration in the second period; and (iv) Low-High treatment: fish fed 25% of optimal daily rations in the first period and 100% of optimal daily ration in the second period. A value of 25% of optimum levels was chosen for the Low feeding regime because similar reductions have previously been shown to reduce the frequency of the resident tactic in adfluvial brown trout (Wysujack et al., 2009). Rations were reduced down to 25% of optimal gradually over a 4-week period, to minimize stress. Within each food treatment, absolute rations were adjusted according to manufacturer's instructions (see **Table S2**) on a monthly basis to account for changes in body mass and temperature (i.e., there was no variation in daily rations within months, within groups).

### Life History Determination and Data Collection

In the spring of 2017 and 2018 (March–June in year 1 and 2 of the experimental phase of the study), fish were routinely assessed for morphological indicators of "smoltification": the series of morphological, physiological and behavioral changes that is generally considered a precursor to downstream migration of juvenile salmonids (Tanguy et al., 1994). Here we use "smolt" to simply mean a fish showing external morphological features consistent with preparing for a migration, and we use saltwater tolerance tests (see below) to further assess physiological aspects of smoltification. We visually assessed morphological smoltification (silvered flanks/loss of parr marks, pronounced lateral line, colorless fins and fusiform shape) according to Tanguy et al. (1994). No fish matched the morphological criteria of smolts in the spring of 2017, the very earliest point at which we expected any smoltification (Poole et al., 2007; Gargan et al., 2016). Individuals that matched the morphological criteria for smolts in spring 2018 were transferred to saltwater at 30 ppt for 24 h to assess their hypo-osmoregulation as a further indicator of anadromy capacity. We used 30 ppt salinity (following Tanguy et al., 1994) because trout often spend large amounts of time in brackish water/estuaries when migrating, hence trout smolts are typically less saltwater tolerant than other salmonids e.g., Atlantic salmon (Urke et al., 2010). After the 24-h immersion in saltwater, a period proposed to induce hypo-osmoregulation in euryhaline species (Schultz and McCormick, 2012), fish were euthanized with an overdose of MS-222 and a blood sample was taken from the caudal vasculature using a 21 G needle and a 2.6 ml heparinized syringe. Blood samples were transferred to 2 ml epindorphs and centrifuged at 8,000 rpm for 3 min. The plasma aliquot was then siphoned off and stored at −80◦C before being measured for plasma chloride concentration as an indicator of hypo-osmoregulatory ability.

All fish, whether identified morphologically as smolts or nonsmolts, were dissected to visually determine sex and maturation status according to gonad development. Males were classed as sexually mature if they had enlarged white testes or had running milt. Males that had visible testes that were moderately enlarged but not running milt were classed as maturing. Females were classed as mature or maturing if the body cavity contained identifiable eggs. Fish with immature gonads, or that could not be identified as either male or female by visual inspection were classed as immature at the time of sampling, and their genotypic sex was later determined using a microsatellite sex marker (P. Prodöhl, unpublished). In the wild, the natural spawning period for these brown trout populations is in late autumn/early winter, and the migratory period is in the spring (Poole et al., 2007; Gargan et al., 2016). Fish showing signs of maturity in freshwater without having first gone to sea, were considered to be on a non-anadromous trajectory, while smolts migrating to sea in a given spring were all immature. Fish in our experiment were thus classed as smolts (migratory tactic) if they were morphologically assessed as smolts and were immature, and were classed as mature (freshwater maturation tactic) if they were mature or maturing at the time of sampling. Fish that were classed as immature, but did not have morphological indicators of smoltification, were considered to have an unknown life history tactic at the time of sampling. A small number of fish (n = 12) had significant skin/fin damage at the time of sampling, and were excluded from the analysis. Whole body lipid content (%) was measured for all smolts, and for a random sample of mature fish (n = 111), using a SMART Trac 5 system (CEM GmnH, Kamp-Lintfort, Germany) of integrated microwave heating and nuclear resonance on homogenized samples.

#### Statistical Analysis

To assess whether food treatment and population influenced life history tactics (Aims 1 and 2), we constructed generalized linear models (GLMs) with a logit link function and binary life-history response variables. One GLM was created to predict smolt status (binary response: 1 = smolt, 0 = non-smolt) using the brglm package in R (Kosmidis, 2019) to account for separation in the data (no smolts recorded in the High-High treatment) (Heinze and Schemper, 2002). A second GLM was created to predict maturation (binary response: 1 = mature or maturing, 0 = immature). Categorical explanatory variables in both of these GLMs included food treatment (High-High, Low-High, Low-Low, High-Low), population (anadromousbackground vs. non-anadromous-background), and sex (male or female) as predictors. We constructed a third GLM to test for treatment/population effects on likelihood of being classed as "unassigned" (i.e., not having expressed a migratory/resident phenotype by the end of the study (binary response: 1 = unassigned, 0 = smolt or mature). We included an interaction term between food treatment and population to determine if life history responses in each population were similar under the different food regimes (Aim 3). To test whether food restriction was more important in the early or late rearing periods (Aim 2), we conducted Tukey post-hoc tests of all possible pairwise comparisons among the levels of food treatment using the emmeans package in R (Lenth, 2019). Overall, one expects the strongest difference in life-history tactics to be found between the High-High and Low-Low treatments. If the effects of food restriction are additive and the timing of food restriction does not mater, then one expects life-history tactics in the Low-High and High-Low treatments to be intermediate between the High-High and Low-Low treatments, and not significantly different from each other. Conversely, if food restriction is more important in the first period, then one expects tactic frequencies in the Low-High treatment to be closer to those in the Low-Low treatment (and the High-Low treatment should be more similar to the High-High treatment), while if food restriction is more important in the second period, the High-Low treatment should be closer to the Low-Low treatment and the Low-High treatment to the High-High. To further explore factors influencing variation in saltwater tolerance (Aims 1–3) a key component of life-history tactics—we constructed a linear model (normal errors) with plasma chloride concentration as the continuous response, and population, food treatment, sex, and an interaction between population and food treatment included as predictors.

To address Aim 4, we explored factors influencing variation in the length, weight and condition factor of fish at different measurement time points across the study period within a mixedeffects modeling framework [nlme package (Pinheiro et al., 2019)]. Measurement time points were September and November in 2016, February, April, June, July, September, and December in 2017, and April 2018. Condition factor was calculated as Fulton's K where:

$$Condition\ \left(K\right) = \frac{mass\ \left(g\right)}{fork\ length\ \left(cm\right)^3} \times 100\ \frac{kg}{kg}$$

For the subsequent analyses of status traits, we created a new categorical variable called 'life-history tactic' with two levels: migratory (i.e., immature smolts) or mature/maturing (hereafter simply called mature). Fish which were neither classified as migratory nor mature (unassigned fish) were not included in the status trait analyses, as it could not be determined which life history trajectory they might adopt [i.e., these fish could have displayed either migratory or mature tactics the following spring (a full 3 years after hatching), but the experiment was terminated the previous spring (2 years after hatching)]. In addition to life-history tactics, month (continuous variable), population (categorical variable with two levels), food treatment (categorical variable with four levels), and sex (categorical variable with two levels) were included as fixed effects, and individual identity was included as a random effect to account for multiple measurements on some individuals. We included an interaction between life-history tactics and month (to test whether individuals on different life-history trajectories diverged through time in their length/weight/condition factor), an interaction between life-history tactics and population (to test whether average differences in length/weight/condition factor between the two tactics was similar across the two populations), and an interaction between population and food treatment (to test whether the effects of food regime were similar across populations). Temporal autocorrelation of the response variable was accounted for by modeling an autoregressive error structure as a first order lag function of month. Separate models were constructed each for length, weight, and condition factor and normal errors were assumed in each case.

We also explored factors influencing variation in final length, K and whole body lipids (i.e., the final measurements for these status traits at the end of the study) in a mixed effects modeling framework, where life-history tactics, food treatment, population, and sex were included as fixed effects, and date of terminal sample (categorical variable with 11 sampling dates) was modeled as a random effect. We included two interaction terms (life-history tactics × population, and food treatment × population), to explore whether the patterns for each population were similar across tactics and food treatments, respectively. Separate models were constructed each for length, K and whole body lipids and normal errors were assumed in each case. Marginal R 2 values for mixed effect models were calculated using the MuMIn package in R (Barton, 2018).

For all of the above models, statistical significance at a 5% alpha level of predictor variables was assessed using likelihood ratio tests (LRT), and non-significant interaction terms were omitted so the main effects could be interpreted.

Finally, to assess whether variation in growth was associated with life-history tactics (Aim 4), we compared growth trajectories of migratory and mature fish by fitting three typical models of fish growth: the von-Bertelanffy growth curve, the Gompertz growth curve and a logistic growth curve. The logistic growth curve best described the data according to AIC (1AIC = 0), and was used for all further growth trajectory analysis. The logistic growth equation models asymptotic growth as:

$$L = \frac{L\_{\infty}}{1 + e^{(-\lg(T - I))}}$$

Where L is fork length, L∞ is asymptotic fork length (cm), gi is the growth rate (cm/day), T is time (days) and I is the inflection point. The logistic model was fitted using non-linear least squares to length data collected on individually-identifiable fish during the experiment, with separate models fitted for smolts and mature fish. As non-linear least squares regression is sensitive to starting values of parameters, the model was fitted using the nls\_multstart function from the nls.multstart package in R (Padfield and Matheson, 2018). This allowed for starting values for each parameter to be randomly selected from a bounded distribution over 1,000 iterations of the model, with the best available model then selected by AIC. To determine the fit of the most parsimonious model to our data, we bootstrapped with replacement 10,000 times and constructed 95% confidence intervals from the bootstrapped fits.

All analysis was carried out in R version 3.5.3 (R Core Team, 2019), and all statistical models were checked against assumptions of the given model (independence, non-normality of residuals, heteroscedasticity and multicollinearity).



*Values correspond to percentages for each category, broken down by sex, of the total number of fish per tank (where each tank corresponds to a given population background by food treatment combination, i.e., a single row in the table). Sample size (n) given in brackets after the %.*

#### RESULTS

#### Life-History Tactics

By the end of the experimental phase, a total of 567 fish had been categorized as either smolts i.e., putatively migratory (n = 36 females and n = 18 males) or non-smolts (n = 277 females and n = 236 males). All of the smolts were by definition immature, and 15.52% of the non-smolt females and 28.39% of the non-smolt males were immature. See **Table 1** for a full breakdown of life-history tactics by population background, food treatment and sex. The proportion of smolts varied according to food treatment and population (**Figure 1**). Highest proportions of smolts were seen in the Low-Low food treatment, in which 26.56% of the anadromous-background population, and 15.71% of the non-anadromous background population, were classified as smolts. The lowest rates of smolting were found in the High-High food treatment, in which no fish from either population were categorized as smolts. Intermediate smolting rates were observed in the other two treatments, with 6.45% of fish from the anadromous-background population and 13.75% of fish from the non-anadromous-background population classified as smolts in the Low-High treatment, and 15.87% and 1.22% of fish from each population, respectively, classified as smolts in the High-Low treatment.

The probability of smolting was described by a GLM retaining food treatment (χ <sup>2</sup> = 44.57, df = 3, p < 0.001), population (χ <sup>2</sup> = 3.46, df = 1, p = 0.063), sex (χ <sup>2</sup> = 4.40, df = 1, p = 0.036), and an interaction between food treatment and population (LRT for the model with and without interaction term: χ <sup>2</sup> = 11.66, df = 3, p = 0.009). Overall across the two populations, there appeared to be an additive effect of food treatment on the probability of smolting — that is, the percentages of smolts in the Low-High and High-Low treatments were similar, and approximately intermediate to the percentages in the Low-Low and High-High treatments, when population was ignored (**Figure 1**). However, when population was taken into account, the life-history response to food treatment varied by population and appeared to be non-additive within each

under varying food restriction treatments. Food treatment is denoted in the format "food in year one—food in year two," where "high" refers to optimal food rations and "low" refers to 25% of optimal rations. *P*-values shown are Tukey *post-hoc* pairwise comparisons across all levels of food treatment for each population.

population (**Figure 1**; **Table 2**). Fish from the anadromousbackground population exhibited a relatively high percentage of smolts (15.87%) under the High-Low treatment that was closer to the Low-Low treatment (26.56% smolts) than to the High-High treatment (0% smolts) and post-hoc comparisons of High-Low against Low-Low were not significant (p = 0.377). The opposite was true for the anadromous-background population in the Low-High treatment (6.45% smolts) with significant post-hoc



*The reference level of each factor is in brackets, i.e., effects in both models were contrasted against female fish from the anadromous-population background in the Low-Low food treatment. Statistical significance was assessed at p* < *0.05.*

comparisons of Low-Low and Low-High (p = 0.016). In contrast, fish from the non-anadromous-background population exhibited a relatively high percentage of smolts (13.75%) under the Low-High treatment that was closer to the Low-Low treatment (15.71% smolts) than to the High-High treatment (0% smolts) (post-hoc contrasts between Low-High and Low-Low were non-significant, p = 0.994), while the opposite was true for this population in the High-Low treatment (1.22% smolts) (post-hoc contrasts between High-Low and Low-Low were significant, p = 0.042). This implies that food restriction was more important in the second period for fish from the anadromous-background population, while food restriction in the first period was more important for the non-anadromous-background fish.

Maturation tactics in freshwater were also significantly affected by food treatment (χ <sup>2</sup> = 33.03, df = 3, p < 0.001), population (χ <sup>2</sup> = 12.14, df = 1, p < 0.001), and sex (χ 2 = 4.54, df = 1, p = 0.033) but there was no significant interaction between food treatment and population (LRT for the model with and without interaction term: χ <sup>2</sup> = 5.31, df = 3, p = 0.150).

lines in each box.

Food restriction had a negative effect on maturation probability, in direct contrast to food restriction effects on smolting rates. Fish in the Low-Low food treatment had the lowest probability of maturing (p < 0.001, **Table 2**), and the highest rates of maturity were observed in the High-High food treatment (p < 0.001, **Table 2**). Fish from the anadromous-background population were significantly more likely to mature than fish from the non-anadromous-background population in all food treatments (p = 0.001, **Table 2**). See **Table 2** for all parameter estimates and associated standard errors. The probability of having been unassigned a life history showed similar patterns to maturation tactics, and was similarly significantly affected by food treatment (χ <sup>2</sup> = 16.95, df = 3, p = 0.001), population (χ <sup>2</sup> = 30.74, df = 1, p < 0.001), and sex (χ <sup>2</sup> = 16.21, df = 1, p < 0.001), see **Table 2**. The interaction between food treatment and population was marginally not significant (LRT for the model with and without interaction term: χ <sup>2</sup> = 7.75, df = 1, p = 0.052).

We found a significant effect of population on plasma chloride levels of fish classified as smolts (F = 9.47, df =1, 48, p = 0.003), but the interaction term between population and food treatment was not significant (LRT for model with and without interaction term: F = 1.39, df = 2, p = 0.259). Fish from the anadromous-background population had significantly lower plasma chloride concentrations than non-anadromousbackground fish (p = 0.003, **Figure 2**; **Table 3**). There was no significant effect of food treatment (F = 2.95, df = 2, 48, p = 0.062) or sex (F = 0.01, df = 1, 48, p = 0.991) on plasma chloride levels (**Table 3**).

# Factors Explaining Variation in Status Traits at Different Time Points

At the time at which the food treatments were first applied, fish from both populations were in similar condition (F = 0.41, df = 1, 137, p = 0.523), however, anadromous-background fish were heavier (F = 17.14, df = 1, 137, p < 0.001) and longer (F = 16.31, df = 1, 137, p < 0.001) than non-anadromous-background fish. A mixed model analysis indicated further divergence in these status traits over the study period that was related to lifehistory tactics, food treatment, and population effects (**Figure 3**; **Table 4**). The models for length (marginal R <sup>2</sup> = 0.77), weight (marginal R <sup>2</sup> = 0.62), and K (marginal R <sup>2</sup> = 0.35) retained a significant interaction between food treatment and population, and a significant interaction between life-history tactics and month (**Table 4**). Sex did not have a significant effect on length (χ <sup>2</sup> = 0.024, df = 1, p = 0.877), weight (χ <sup>2</sup> = 0.050, df = 1, p = 0.823), or condition factor (χ <sup>2</sup> = 0.082, df = 1, p = 0.774). After accounting for growth between measurement periods (i.e., the fixed effect of measurement period), smolts tended to be shorter, lighter and have lower condition than mature fish (**Table S3**). The differences in length, weight and K were similar for both populations (an interaction between population and lifehistory tactics was not retained in any of the final models, see **Table 4**). The significant interaction between food treatment and population indicated that fish from the anadromous-background were larger, and heavier (but in similar condition) than fish from the non-anadromous-background under both High-Low and High-High treatments (**Table S3**). However, in the Low-Low and Low-High treatments, there were negligible differences in length, weight and K between populations (**Table S3**). The significant interaction between month and life-history tactics indicated that changes in length, weight and K through time varied between smolts and mature fish. Mature fish tended to increase in length and weight quicker (**Figure 3B**; **Table S3**), while smolts tended to be in worse condition (lower K) earlier (**Figure 3C**; **Table S3**). See **Table S3** for all model outputs.

## Factors Explaining Variation in Final Values for Status Traits

At the end of the study, fish differed in length, condition and lipid content according to food treatment, life-history tactics and population (**Figure 4**). The model describing length (marginal R 2 = 0.50) retained a significant interaction between food treatment and population (**Table 5**) but did not indicate a significant effect of life-history tactics (χ <sup>2</sup> = 2.83, df = 1, p = 0.093), or sex (χ 2 = 0.005, df = 1, p = 0.947). The models describing condition (marginal R <sup>2</sup> = 0.56) and whole body lipids (marginal R 2 = 0.73, **Table 5**) each retained an interaction between population and food treatment (**Table 5**), and included a significant effect of life-history tactics on condition (χ <sup>2</sup> = 64.58, df = 1, p < 0.001), and whole body lipids (χ <sup>2</sup> = 7.71, df = 1, p = 0.005). Sex did not have a significant effect on condition (χ <sup>2</sup> = 3.43, df = 1, p = 0.064) or whole body lipids (χ <sup>2</sup> = 2.18, df = 1, p = 0.140). Overall, smolts were of similar length to mature fish at the end of study (**Figure 4**), but tended to be in poorer condition (p < 0.001, **Table S4**) and have slightly higher whole TABLE 3 | Parameter estimates with associated standard errors (SE) for the linear model testing effects of population, sex, and food treatment on plasma chloride concentration (mmol/L) of brown trout classified as smolts (*n* = 54).


*The reference level of each factor is in brackets, i.e., effects were contrasted against female fish from the anadromous-population background in the Low-Low food treatment. Note that no individuals were classed having adopted the anadromous tactic in the High-High food treatment, and this category was dropped for this analysis. Statistical significance was assessed at p* < *0.05.*

body lipids (p = 0.008, **Table S4**). We detected an interactive effect of food treatment and population, where fish from the anadromous-background population were larger than fish from the non-anadromous-background population, but similar under Low-Low food conditions (**Table S4**). However, nonanadromous-background fish were overall in better condition (p = 0.011, **Table S4**) and had higher whole body lipids (p < 0.001, **Table S4**), and these differences between populations were strongest under conditions of Low-Low food (**Figure 4**; **Table S4**). The lack of significant interactions between lifehistory tactics and population in the models for length, K, and whole body lipids indicated that differences between populations were similar for both mature fish and smolts (**Table 5**). See **Table S4** for all model outputs.

#### Growth Rate Differences

The somatic growth of fish during the experiment was welldescribed by a logistic growth model. Initial model fitting indicated the most parsimonious model included separate growth parameters for smolts and mature fish. Mature fish had higher intrinsic growth rates (g<sup>i</sup> = 0.0050, SE = 0.0006, p < 0.001), a smaller asymptotic size (L<sup>∞</sup> = 25.44, SE = 0.86, p < 0.001), and a lower point of inflection (I = 172.7, SE = 13.8, p < 0.001) than smolts, where g<sup>i</sup> = 0.0039 ± SE 0.0009 (p < 0.001), L<sup>∞</sup> = 27.31 ± SE 4.13 (p < 0.001), and I = 305.7 ± SE 89.9 (p = 0.001). Mature individuals were relatively larger earlier in life than smolts, and had faster overall growth (**Figure 5**).

Growth differences between the two populations were also identified, where fish from the anadromous-background population were relatively larger earlier in the study than fish from the non-anadromous-background population, and grew faster (**Figure 6**). Anadromous-background fish had higher intrinsic growth rates (g<sup>i</sup> = 0.0045, SE = 0.0009, p < 0.001), similar asymptotic size (L<sup>∞</sup> = 26.83, SE = 1.68, p < 0.001), and a lower point of inflection (I = 184.1, SE = 26.9, p < 0.001) than non-anadromous-background fish, where g<sup>i</sup> = 0.0043 ± SE 0.0007 (p < 0.001), L<sup>∞</sup> = 26.45 ± SE 1.65 (p < 0.001), and I = 236.3 ± SE 32.9 (p < 0.001).

#### DISCUSSION

Salmonine fishes exhibit some of the most striking examples of animal migration, but uncertainty still surrounds the mechanisms by which alternative migratory tactics can be expressed, or inhibited, across salmonine populations. A principle aim of our study was to assess the importance of food availability at different time points during early ontogeny in determining migratory/life-history tactics in two populations of brown trout. Food reduction across almost 2 years led to increased rates of smolting (migratory tactic) in fish from both population backgrounds, whilst no fish were classed as having adopted the migratory tactic in either population after 2 years of experiencing high food, i.e., optimal rations (**Figure 1**). Migratory/life-history tactics were also influenced by population background, consistent with an inherited component to migratory/life-history decisions—fish derived from a naturally anadromous population were more often classed as smolts in our experiment, while offspring derived from a naturally non-anadromous population were more often classed as nonsmolts, or having undergone freshwater maturation consistent with a residency tactic. Intriguingly, the populations responded differently to the timing of food restriction, with fish from an anadromous population background seemingly having been more affected by food restriction in their second year, whilst fish from a non-anadromous population background were more affected by food restriction in their first year. Females were more likely than males to become smolts under all food treatments. Collectively, these results indicate both extrinsic (food-driven) and intrinsic effects (related to population background, sex TABLE 4 | Results of the mixed effect model analysis for length, weight and condition factor (*K*) trajectories of brown trout in the experiment with life-history classed as either smolts (i.e., migratory) or freshwater mature across the study period.


*The results of the model selection procedure on interaction terms are given, and the selected model for each response is highlighted in bold. The models included a random effect of individual identify and a first-order autoregressive correlation structure with respect to month was also modeled.*

FIGURE 4 | Effects of food treatment on final (A) length, (B) condition factor (*K*), and (C) whole body lipids at the end of the experimental study (Spring 2018) of brown trout offspring classed as either smolts (migratory) or freshwater maturing (non-migratory/resident). Offspring were derived from wild-caught parents from an anadromous-background population (AB) and a non-anadromous-background population (non-AB). The median is represented by the white horizontal lines in each box.



*The results of the model selection procedure on interaction terms are given, and the selected model for each response is highlighted in bold. The models included a random effect of sample date.*

and other individual-level attributes) on migratory/life-history tactics in brown trout, that may interact in complex ways and influence how populations respond in the wild to changing environmental conditions.

Differences in growth and body condition were apparent from an early stage between fish adopting different lifehistory/migratory tactics, and were maintained across the full (almost 2-year) duration of the study. These differences were in turn also driven by both extrinsic and intrinsic effects. Extrinsic effects were evidenced by the fact that

backgrounds (anadromous or non-anadromous). Fitted lines are based on the best-fitting parameters from the logistic growth model, fitted using non-linear least squares regression. Shaded areas represent the 95% confidence intervals constructed by bootstrapping for 10,000 iterations.

large differences in fork length, mass, body condition and whole body lipids were apparent between fish reared under different food treatments, which in turn contributed to fish adopting different life-history tactics via phenotypic plasticity. Intrinsic differences among individuals in "status traits" clearly also contributed to migratory/life-history outcomes, given that differences in body size, condition and lipids were apparent between populations, and between fish from each population that adopted different tactics within each food treatment where the external environment was the same. Such intrinsic variation within and between populations could reflect inherited genetic effects, inherited non-genetic effects (e.g., parental effects, epigenetic inheritance), or non-inherited differences driven by early-life environmental influences that have a relatively long-lasting effect on phenotype (Burton and Metcalfe, 2014). Expanding our approach to incorporate even earlier life stages (e.g., post-hatching/fry) could further illuminate how factors in early life influence life history.

#### Extrinsic Factors

The observed increases in smolting in the face of food restriction, together with decreases in maturation, suggested that the reduction in food supply prevented individuals from meeting an intrinsic (e.g., genetically determined) threshold for residency and maturity in freshwater, which is in agreement with previous studies (Olsson et al., 2006; Wysujack et al., 2009; O'Neal and Stanford, 2011; Jones et al., 2015). Indeed, the absence of any smolts under conditions of high food supply was surprising, particularly within fish from the Tawnyard population (anadromous-background), which has a naturally high frequency of anadromy in the wild (Gargan et al., 2016). This suggests that, in nature, a large number of fish in the Tawnyard system must typically experience relatively low food availability as freshwater juveniles, as otherwise anadromy rates would be lower in the wild. Moreover, the balance of fitness cost and benefits of migration in the system must be such that natural selection has caused a relatively high threshold for residency to evolve (an ultimate mechanism; Hazel et al., 1990; Tomkins and Hazel, 2007; Pulido, 2011), meaning a minority of Tawnyard fish in the wild typically surpass their intrinsic freshwater maturation threshold and the anadromous tactic is more frequent.

Manipulation of the timing of food reduction revealed that life-history responses of a given population to environmental change might depend on the point during ontogeny at which the change is experienced. This could come about via two non-mutually exclusive mechanisms: populations could exhibit variation in sensitivity to cues experienced during given fixed "decision windows," and/or the timing of the decision windows themselves may vary across populations. In our study, food restriction in the first year (Low-High treatment) was a more important driver of smolting rates than food in the second year (High-Low) for fish from the non-anadromousbackground population, whereas food in the second year was more important for the anadromous-background population. This was an intriguing outcome, and hints at a complex interplay between extrinsic environment and intrinsic or populationspecific factors. The apparently greater importance of food restriction in the first year for the non-anadromous-background population could perhaps be related to lower intrinsic growth rates in this population in the wild. Given their low potential growth rates, individuals in the non-anadromous-background population might be constrained to make a life-history decision (i.e., choose future migration or residency) early in life in order to divert energy intake toward meeting the associated demands of the chosen tactic. Because residents must accumulate sufficient lipid reserves to be converted into reproductive tissue before spawning (McMillan et al., 2012), in the wild, Bunaveela fish may have experienced selection for adopting a maturation trajectory relatively early in order to allow sufficient time for growth and energy accumulation, with early decision windows evolving as a consequence. In contrast, fish from the anadromousbackground population with higher intrinsic growth potential may be less constrained in this regard, and may defer choice of migratory tactics to the second year of life, or indeed have flexibly reversible life-history trajectories where, for example, fish choosing residency based on high food in year one may switch to migratory tactics in response to low food in year two. There is some evidence for conditions in the second year of life being a key driver of migratory tactics in a naturally facultatively anadromous brown trout population to support this (Cucherousset et al., 2005).

Coupled with a later "decision window"/higher sensitivity to conditions in year two, a naturally high intrinsic growth propensity in the anadromous-background population could have facilitated high levels of compensatory growth when receiving optimal food resources in year two in the Low-High treatment. If growth, or some aspect of energy usage related to growth such as body condition, is used as a cue for migratory tactic choice, this may then have translated into more individuals from this population meeting their threshold for maturation in the Low-High treatment. Strong compensatory responses after periods of food restriction have been observed in salmonids in general, and interestingly, the compensatory response has often appeared to be directed toward restoring body condition, rather than size. Nicieza and Metcalfe (1997) found food restricted fish recovered similar condition to controls within a year of food supply restoration, and Alvarez and Nicieza (2005) further found a compensatory response that resulted in restoration of condition and energy status rather than skeletal growth in brown trout post food restriction.

Alternatively, we cannot rule out the presence of multiple migration vs. residency decision windows, that re-occur annually or more frequently, whereby an individual repeatedly re-assesses its status trait relative to its inherited freshwater maturation threshold and can remain "undecided" at the first or even second windows, though there is little empirical evidence for this. A simpler explanatory model is that there is a single, initial decision determining migration vs. residency, and then subsequent decision windows occur for fish on each trajectory (migrants and resident) related to the timing of expression of the adopted life-history tactic, where for example migrants must decide at what age to actually migrate (determined by pressures of size-dependent sea survival), or indeed where to migrate (Ferguson et al., 2019). Similarly, a resident individual must also decide when to mature (Thorpe and Metcalfe, 1998; Thorpe et al., 1998), a decision shown to be affected by lipid reserves in Atlantic salmon (Rowe et al., 1991; Jonsson and Jonsson, 1993, 2005) and possibly triggered by similar threshold type mechanisms in brown trout. These timing decisions could be further influenced by extrinsic environmental conditions, giving rise to a temporal continuum of migration and maturation tactics. This may explain why some fish in our study were classified as having an undetermined life-history (neither smolt nor mature) by spring of year two: these individuals may simply have been delaying expression of a migratory or freshwater maturing phenotype until the following year. These caveats must be born in mind when interpreting our experimental results, as the life-history tactic frequencies we measured in year 2 could be indicative of age-specific tactic frequencies, rather than overall rates of migration vs. residency across all ages. However, the basic conclusions were the same in the GLMs where the data were analyzed as either smolt vs. non-smolt, or immature vs. mature, giving us confidence that the patterns reflect the migration decision per se.

### Variation in Status Traits Underpinning Alternative Tactics

Size-based differences between migrating individuals (those classified at the end of the study as smolts) and resident fish (those classified at the end of the study as mature) were established relatively early, with differences in weight, length, and condition that were maintained during the course of the study. The early divergence in physiological condition between migrants and residents supports the energy limitation scenario, where fish adopt migration as a result of failing to meet the necessary condition in early life to mature as residents in freshwater (Jonsson and Jonsson, 1993). Maturing fish reached an apparent size asymptote earlier than migrating fish (i.e., had smaller inflection point in **Figure 5**, and were larger earlier in the study). Size appears to be a potential status trait that regulates, or correlates with factors regulating early sexual maturation, as has been documented in Atlantic salmon, where anadromous males are smaller than their counterparts that mature early in freshwater as so-called "precocious parr" (Whalen and Parrish, 1999; Garant et al., 2002). However, although body size has been suggested as a major component of the status (cueing) trait for anadromy in brook trout (Thériault et al., 2007), the divergence in mass and condition we find here in our study suggests that other factors beyond size also contribute to the maturation vs. migration/anadromy decision. It seems increasingly likely that a suite of interlinked physiological components is assessed (e.g., overall energetic status or rate of change in energy), and no single trait controls the migratory/anadromy decision. Genetic covariance between life history traits such as growth, size, metabolism, and other morphological traits further suggests that migration decisions are associated with a suite of inter-linked phenotypic traits (Doctor et al., 2014; Hecht et al., 2015).

Fish on a migratory trajectory here appeared to maintain growth rates during the experiment (and had a higher inflection point), such that they were similar in length to mature fish by the end of our study. Constant, or even accelerated growth in pre-migratory fish (Metcalfe, 1998) has been explained by size-dependent survival at sea (Klemetsen et al., 2003) due to better osmoregulation ability (Finstad and Ugedal, 1998) and reduced predation on larger anadromous individuals (Dill, 1983; Jonsson et al., 2017). Interestingly here, although skeletal growth (i.e., length) was maintained, migratory fish were considerably lighter and in worse condition than mature fish at the end of study, which suggests that once on a migratory trajectory, resources were primarily allocated to meeting a size-based threshold for surviving actual migration. The maintenance of growth rates in migrants as such does not contradict the energy limitation scenario, but rather suggests that migratory fish redirect what resources they obtain into becoming large enough to survive the migration, at a cost to their overall body condition.

The diminished body condition of migratory individuals was not, however, reflected in levels of whole-body lipids at the end of the study. Contrary to our expectations, migratory fish had marginally higher levels of whole body lipids than mature fish. Lipid storage has been identified previously as an important precursor of maturation in fish (Tocher, 2003) and an indicator of a residency life history in salmonids (Tocher, 2003; Sloat and Reeves, 2014; and references therein). The unexpected trend we observed in lipids may have been a consequence of measuring lipids during the smolt migration period, at which stage fish that have initiated maturation might have already converted some of their energy stores into gonadal tissue, and hence show depleted lipids levels relative to migrants (Tocher, 2003; Sloat and Reeves, 2014). Alternatively, higher lipid levels in migrants could reflect accumulation of reserves, as either a bet-hedging strategy if resources in the migration destination are uncertain, or to fuel the migration journey itself (Stefansson et al., 2003). Pre-migratory "fattening" strategies are relatively common in migratory birds (Piersma et al., 2005) but less so in salmonines (Jonsson and Jonsson, 2005)

#### Intrinsic Factors

We had predicted that the two populations in our study would show variability in adopting migratory tactics across all food restriction scenarios and indeed, overall, the probability of smolting was higher in the anadromous background population than in the non-anadromous population. Moreover, higher hypoosmoregulatory function (lower plasma chloride concentration) was documented in smolts from the former population relative to the latter, implying that smolts from the anadromousbackground population were physiologically better prepared for transition to marine conditions. In contrast, although some fish from the non-anadromous-background population were classified as smolts in the experiment, these putative smolts exhibited relatively lower saltwater tolerance. A potential explanation for the reduced hypo-osmoregulatory function of non-anadromous-background smolts might be that they are poorly adapted to saltwater given their lack of (recent) evolutionary exposure to marine conditions. Relaxed selection leading to degradation of hypo-osmoregulation has similarly been observed in non-anadromous populations of landlocked Atlantic salmon (Nilsen et al., 2008; McCormick et al., 2019) and alewife Alosa pseudoharengus (Velotta et al., 2014, 2015). Alternatively, reduced saltwater tolerance could be evidence of an emerging migration continuum whereby putative smolts may have chosen a potamodromous (freshwater migratory) tactic and hence were unprepared physiologically for transitioning to saltwater. Nevertheless, the causal mechanisms underpinning anadromy and potamodromy are proposed to be similar, e.g., reduced food availability has previously been reported to increase adfluvial migration in freshwater brown trout transplanted to streams of high population density (Olsson et al., 2006). All brown trout in Ireland presumably have anadromous ancestral origins, since they would have had to recolonize the island after the Last Glacial Maximum via the sea (Ferguson et al., 2019). It thus seems more likely that the capacity for anadromy (or at least migration), albeit somewhat deteriorated in terms of saltwater tolerance, lay dormant in the Bunaveela fish, with anadromy re-expressed under experimental conditions of energy limitation.

The putative re-emergence of an anadromous life history in our Bunaveela fish is of particular interest from a fisheries management perspective, as it suggests the capacity for anadromy (or at least migration) may lie dormant within apparently resident populations. Such populations may thus have the potential to contribute to the restoration of anadromous stocks that have experienced widespread reductions, as evidenced by Gargan et al. (2006) in two formerly anadromous populations that suffered collapses. Anadromous phenotypes arising from resident genotypes have similarly been documented in O. mykiss (Kelson et al., 2019), and from common garden experiments with lake resident O. mykiss which were formally anadromous but were prevented from migrating by impassable dams or waterfalls (Thrower et al., 2004). These findings make sense within the framework of the conditional threshold model (Tomkins and Hazel, 2007), where environmental factors can affect life history tactic frequency by changing the distribution of the realized physiological state relative to inherited switch points (a proximate mechanism). Environmental factors could also drive longer term changes in tactic frequency via natural selection acting to shift the genotypic distribution of underlying switch points (an ultimate mechanism) (Hazel et al., 1990; Tomkins and Hazel, 2007; Pulido, 2011); for example, if survival or growth at sea is poor then migration may become less prevalent in the population if residents attain higher overall relative fitness than migrants. Within the Burrishoole system, the establishment of an Atlantic salmon farm in the estuary was implicated in the collapse of the anadromous life history from this catchment over a period of 30 years due to high rates of sea lice transmission (Poole et al., 1996, 2007). Reduced marine survival rates may have imposed strong selection against anadromy, and hence caused the evolution of lower mean threshold values for freshwater maturation within the Burrishoole catchment as a whole. Our current results are consistent with this evolutionary explanation, in that we demonstrated heritable differences (or at least phenotypic differences among genetically divergent populations in a common garden experiment) a pre-requisite for evolutionary responses. However, they also show that phenotypic plasticity can drive changes in migratory tactics, which may contribute to observed lifehistory changes in natural populations (Gargan et al., 2006; Sandlund and Jonsson, 2016).

Early-life differences in length and mass between the two populations may proximately cause different anadromy propensities, as has been seen in brook trout, where size of juvenile fish was negatively related to probability of future residency (Thériault et al., 2007). Interestingly, though our populations differed in size early in the study (before food restriction), they were in similar condition at this time, suggesting that both populations had similar energy intake vs. output, at least initially. Higher intrinsic growth rates in the anadromous background population may have increased the likelihood of eventual energetic limitation in freshwater, thus reducing relative condition and increasing anadromy propensity (exemplified in our Low-Low food treatment). Conversely, when food resources are in ample supply, high intrinsic growth rates could hasten freshwater maturity instead of anadromy in this population (c.f. the scenario of optimal food resources in our study). Such variability in migratory tactics is a feature of salmonines in general [e.g., "retirement" from anadromy in Dolly Varden Salvelinus malma (Bond et al., 2015)] which may buffer species from increasing anthropogenic pressures in the marine environment (Russell et al., 2012).

# CONCLUSIONS

Collectively, the results of this study show that the adoption of migratory tactics in brown trout involves an interplay between inherited components and environmentally cued physiological condition, in line with previous salmonines studies (Chapman et al., 2012; Dodson et al., 2013; Kendall et al., 2014). The differences we observed in population responses to food restriction and its timing suggest a complex relationship between intrinsic and extrinsic factors that may allow for a continuum of migratory tactics to exist. These population differences, together with the fact that putative anadromy emerged within offspring of a naturally non-anadromous population, emphasize that a range of life history outcomes are possible even within a single species, which can contribute to so-called portfolio effects that cushion the species as a whole from rapidly changing environmental conditions (Schindler et al., 2015). Although our study offers some important insight into how extrinsic and intrinsic factors interactively shape life-history tactics, we have only considered one element of the freshwater environment here, and future studies should expand to consider how other proximate drivers such as temperature, which influences a range of physiological and life-history traits in salmonines (Satterthwaite et al., 2010; McMillan et al., 2012; Doctor et al., 2014; Kendall et al., 2014; Sloat and Reeves, 2014), govern migratory tactics in fish from different genetic backgrounds. Moreover, it is now important to expand this approach into natural systems using, for example, common garden or reciprocal transplant experiments, to assess whether these findings hold up under real world complexities.

Finally, our results have important implications for the conservation of facultatively migratory species, which are in global decline due to in-stream barriers, habitat degradation, climate change, overfishing, and the expansion of aquaculture (Costello, 2009; Limburg and Waldman, 2009). Knowledge of how extrinsic and intrinsic factors affect fish migratory tactics may aid in successful management and restoration of facultatively migratory populations, and in doing so maintain important intraspecific biocomplexity, which offers increased resilience to effects of global change (Schindler et al., 2015).

#### DATA AVAILABILITY

The datasets generated for this study are available on request to the corresponding author.

#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of Health Products Regulatory Authority (HPRA) Ireland, under HPRA project license AE19130/P034, and HPRA individual licenses AE19130/I087, AE19130/I200, AE19130/I201, and AE19130/I202.

#### AUTHOR CONTRIBUTIONS

TR, PM, LA, and WP conceived the study. LA, SH, TR, PG, and LH collected data and contributed to experimental design, as did MO'G and JK. LA conducted statistical analysis and led the manuscript writing. All authors contributed to interpretation of results and revisions of the manuscript.

#### REFERENCES


#### ACKNOWLEDGMENTS

The authors would like to thank Brian Clarke, Deirdre Cotter, members of the FishEyE team at UCC, and the staff of Inland Fisheries Ireland and the Marine Institute for obtaining brood stock and for assistance in fish rearing, along with Robert Wynne, Ronan O'Sullivan, Peter Moran and Adam Kane for assistance in fish husbandry, and Jamie Coughlan for genotyping work**.** This research was supported by an ERC Starting Grant (639192-ALH) and an SFI ERC Support Award awarded to TER. PMcG was supported in part by grants from Science Foundation Ireland (15/IA/3028 and 16/BBSRC/3316) and by grant-in-aid (RESPI/FS/16/01) from the Marine Institute (Ireland) as part of the Marine Research Programme by the Irish Government. We thank the Associate Editor and two reviewers for comments that improved the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00222/full#supplementary-material


western Ireland. Aquacult. Environ. Interact. 8, 675–689. doi: 10.3354/aei 00211


marine survival. ICES J Mar Sci. fsr208. 12, 1563–1573 doi: 10.1093/icesjms/ fsr208


derived freshwater populations of steelhead. J. Fish Biol. 65, 286–307. doi: 10.1111/j.0022-1112.2004.00551.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Archer, Hutton, Harman, O'Grady, Kerry, Poole, Gargan, McGinnity and Reed. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Individual Variability in Migration Timing Can Explain Long-Term, Population-Level Advances in a Songbird

Kevin C. Fraser <sup>1</sup> \*, Amanda Shave<sup>1</sup> , Evelien de Greef <sup>1</sup> , Joseph Siegrist <sup>2</sup> and Colin J. Garroway <sup>1</sup>

*<sup>1</sup> Department of Biological Sciences, University of Manitoba, Winnipeg, MB, Canada, <sup>2</sup> Purple Martin Conservation Association, Erie, PA, United States*

#### Edited by:

*Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway*

#### Reviewed by:

*Jason Courter, Malone University, United States Kristen Covino, Loyola Marymount University, United States*

> \*Correspondence: *Kevin C. Fraser kevin.fraser@umanitoba.ca*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *27 March 2019* Accepted: *13 August 2019* Published: *06 September 2019*

#### Citation:

*Fraser KC, Shave A, de Greef E, Siegrist J and Garroway CJ (2019) Individual Variability in Migration Timing Can Explain Long-Term, Population-Level Advances in a Songbird. Front. Ecol. Evol. 7:324. doi: 10.3389/fevo.2019.00324* Migratory animals may be particularly at-risk due to global climate change, as they must match their timing with asynchronous changes in suitable conditions across broad, spatiotemporal scales. It is unclear whether individual long-distance migratory songbirds can flexibly adjust their timing to varying inter-annual conditions. Longitudinal data for individuals sampled across migration are ideal for investigating phenotypic plasticity in migratory timing programs, but remain exceptionally rare. Using the largest, repeat-tracking data set available to date for a songbird (*n* = 33, purple martin *Progne subis*), we investigated individual variability in migration timing across 7,000–14,000 km migrations between North American breeding sites and South American overwintering sites. In contrast to previous studies of songbirds, we found broad, within-individual variability between years in the timing of spring departure (0–20 days), spring crossing of the Gulf of Mexico (0–20 days), and breeding site arrival (0–18 days). Spring departure and arrival dates were fairly repeatable across years (depart *r* = 0.39; arrive *r* = 0.32). Fall migration timing was more variable at the individual level (depart range = 0–19 days; gulf crossing range = 1–15 days; arrive range = 0–24 days) and less repeatable, with fall crossing of the Tropic of Cancer being the least repeatable (*r* = 0.0001). In this first, repeat-tracking study of a diurnal migratory songbird, the high within-individual variability in timing that we report may reflect the greater influence of environmental and social cues on migratory timing, as compared to the migration of more solitary, nocturnally migrating songbirds. Further, large, within-individual variability in migration dates (0–24 days) suggest that advances in spring arrival dates with climate change that have been reported for multiple songbird species (including purple martins) could potentially be explained by intra-individual flexibility in migration timing. However, whether phenotypic plasticity will be sufficient to keep up with the pace of climate change remains to be determined.

Keywords: phenotypic plasticity, spring phenology, repeatability, climate change, avian, long-distance migration, songbird

# INTRODUCTION

Phenotypic plasticity in animal migration timing could provide the means for rapid acclimation to environmental change, as compared to adaptive responses through genetic change (Charmantier and Gienapp, 2014). To what extent phenotypic plasticity and/or micro-evolution are the mechanisms responsible for population-level advances in the spring migration timing of some landbirds has been hotly debated (Knudsen et al., 2011; Charmantier and Gienapp, 2014). Steep population declines among migratory species (Both et al., 2010), lends urgency to determining whether constraints on adaptive timing are a contributing factor.

Longitudinal data are ideal for research on these themes because they provide the opportunity to investigate phenotypic variation within individuals in response to varying environmental conditions across years (Charmantier and Gienapp, 2014). For songbird migration, most previous studies have focused on the use of observational data to determine the individual repeatability (r) of spring migration departure and arrival dates. These studies report broad variation (r = 0.04–0.51), both within and among migratory species (Potti, 1998; Brown and Brown, 2000; Moller, 2001; Ninni et al., 2004; Cooper et al., 2009; Studds and Marra, 2011). Direct-tracking technologies (Stutchbury et al., 2009) provide the means to examine complete annual migration tracks, but studies where multiple migrations by the same individual are monitored are still rare (Both et al., 2016). In a Neotropical songbird (wood thrush, Hylocichla mustelina), within-individual spring migration timing was remarkably repeatable, with a mean difference of just 3 days in spring arrival date between years, suggesting limited plasticity (n = 10; Stanley et al., 2012). In a Palearctic example (red-backed shrikes, Lanius collurio), within-individual variability was similarly low, where mean within-individual differences were 3–12 days, and breeding arrival date (n = 2) varied by only 1–4 days (Pedersen et al., 2018). Thus, repeatability of migration timing of songbirds using direct tracking is generally reported to be high, particularly in spring. However, studies to date have relied on low sample size (<20 individuals), are thinly spread across species and migratory systems, and have focused on nocturnally migrating species (Both et al., 2016).

In many migratory species, population-level advances in spring migration timing have been observed over decadal scales and linked to temperature increase with climate change (Butler, 2003; Mayor et al., 2017; Lehikoinen et al., 2019). Across European and North American migratory landbird systems, the mean advance over several decades in spring migration timing was 1 week (Lehikoinen et al., 2019); with advances within some species reported to be >2 weeks (Butler, 2003). There is much debate as to whether these timing advances are the result of phenotypic plasticity, micro-evolutionary change, or both (Knudsen et al., 2011). For species with high, within-individual repeatability in migration timing, these rapid population-level advances in timing are difficult to reconcile. In Icelandic black-tailed godwits (Limosa limosa islandica), population-level arrival dates advanced by approximately 17 days over 20 years (Gunnarsson and Tomasson, 2011) but over this same time span individual arrival dates were highly consistent (r = 0.51; Gill et al., 2014). These results for black-tailed godwits suggest that advances were not driven by individual plasticity of adult migrants, but rather ontogenetic effects during development could underlie these rapid advances in timing (Gill et al., 2014). Such comparisons of individual variability vs. populationlevel advances in timing remain rare, and need to be further explored in other species and systems. There is currently a deficit of knowledge about the mechanisms of adaptive change (microevolutionary and/or phenotypic plasticity) in response to environmental conditions during migration (Pulido and Berthold, 2004) and the impact of longer-scale climatic effects on the flexibility of migration patterns (Knudsen et al., 2011), especially for long-distance migratory songbirds.

We investigated the phenotypic plasticity of migration timing by using a diurnal, long distance, Neotropical migratory songbird (purple martin, Progne subis) that travels 10, 000–20, 000 km annually between North American breeding sites and South American overwintering sites (Fraser et al., 2012, 2013a,b). Purple martins are aerial insectivores, and like other swallows, are thought to use a fly-and-forage strategy during their diurnal migration (Brown and Tarof, 2013). At the population-level, this species has advanced spring migration timing by 8–20 days over the last 100 years (Arab and Courter, 2015). However, arrival dates did not advance in response to a record-setting, early spring in 2012 (Fraser et al., 2013a). Data on the variation in individual timing across multiple migrations are therefore required to further investigate the potential for phenotypic plasticity in purple martins in response to environmental conditions and to determine whether this can explain long-term advances in timing. We used the largest repeat-track data set available for a songbird, comprised of 33 individuals tracked across 2 years by using light-level geolocators. Our objectives were to (1) determine within-individual variation and repeatability (r) in timing across both spring and fall migration, and (2) assess whether the degree of within-individual variability provides a potential mechanism to explain population-level advances in spring timing reported for martins and whether this species can serve as a model for similar investigations in other songbirds.

# METHODS

#### Geolocator Analysis Methods

Light-level geolocators were deployed on adult purple martins at 8 North American breeding sites (latitudinal range 38.36◦N to 53.02◦N; **Supplementary Table 1**) using a leg-loop backpackstyle harness made of Teflon ribbon (Rappole and Tipton, 1991; Stutchbury et al., 2009) and retrieved in the following year (or subsequent year, n = 2) at the same locations (2009–2016). This study was conducted in accordance with the recommendations of the Ornithological Council's Guidelines to the Use of Wild Birds in Research' and was approved by the University of Manitoba and York University Animal Care Committees (2009- 2 W, F14-009/1–3).

We defined sunrises and sunsets (twilights) from the raw geolocator light data using the preprocessLight function in the R-package BAStag version 0.1.3 (Wotherspoon et al., 2016). We used a light intensity threshold of 32 to define the separation of day and night. Events that influenced the geolocator's light sensor outside of sunrise and sunset times (e.g., shading during day, light during night) indicated false twilights. We used the initiation of heavy shading (false twilights during daylight periods) in spring, that clearly indicated entrance and exits of nest cavities, to identify breeding arrival date. After defining arrival date, all false twilights were removed.

The twilight dataset was used to define daily locations and movement periods by using the R-package GeoLight version 2.0 (Lisovski and Hahn, 2012). We used the coord function to determine spatial coordinates throughout entire migratory tracks. We calculated an appropriate sun elevation angle using twilight data at each bird's known breeding location, before fall migration. Latitudes impacted by spring and fall equinox periods were omitted by using a tolerance level of 0.13 (Lisovski and Hahn, 2012). The resulting data were used to determine daily coordinates and movement periods and to identify spring and fall arrival and departure dates, as well as the date individuals crossed the Tropic of Cancer (23.4◦N). We used the changeLight function to determine residency and movement periods, and shifts in latitudinal and longitudinal coordinates, to identify departure and fall arrival dates. We defined overwintering locations as a tenure of >7 days within the known non-breeding range. Most stopovers in this region were <7 days (Van Loon et al., 2017), thus this provided a conservative estimate of when birds had completed migration and arrived at their overwintering destination.

#### Repeatability of Spring and Fall Migration Phenology

To investigate if individuals are consistent in their migration timing between years, repeatability was examined for birds tracked for at least 2 years (individuals = 33, geolocator tracks = 67). We examined the repeatability of migration departure date (fall tracks = 66, spring tracks = 67), date passing the Tropic of Cancer (23.4◦N; fall tracks = 34, spring tracks = 61), and date of arrival at the breeding grounds or overwintering grounds (fall tracks = 67, spring tracks = 67). We also included 144 single-tracked (1 year only) individuals in the analysis to better account for population level variability in the analysis, resulting in a total of 5–57 tracks per breeding location (**Supplementary Table 1**). Repeatability was calculated as the fraction of variation in behavior between individuals, as compared to the sum of phenotypic plasticity and measurement error (Nakagawa and Schielzeth, 2010). Repeatability is a proportion between 0 and 1, where low values indicate most of the variation is due to plasticity and error. The adjusted repeatability (value of repeatability calculated after controlling for confounding effects) of aspects of migration timing were calculated in linear, mixed-effects models using the package MCMCglmm (Hadfield, 2010). In this case the confounding effects of sex and age were set as fixed effects (males and older birds may have earlier timing) with year, individual, and breeding colony as random effects to control for repeated measures. We did not include temperature or other weather factors in our analysis owing to limitations in the number of repeat-track birds per site. Confidence intervals for repeatability were estimated by parametric bootstrapping with 1,000 replications. Results were replicated with an uninformed prior which produced quantitatively similar results (**Table 1**) with overlapping 95% credibility intervals. All analyses were done in R version 3.5.3 (R Core Team, 2018).

## RESULTS

We found that spring migration timing (departure, crossing 23.4◦N, arrival) was more repeatable between years at the individual level than timing during fall migration (spring range, r = 0.32–0.39; fall range, r = 0.0001–0.001; **Table 1**, **Figure 1**). The timing of spring departure was the most consistent across years (r = 39, CI = 0.08–0.50), perhaps owing to strong endogenous control of migration initiation. Spring crossing of the Tropic of Cancer (23.4◦N) and breeding arrival date were also fairly consistent across years (cross r = 0.32; arrive r = 0.32). Fall arrival date was much less repeatable than breeding arrival date (0.0009 vs. 0.32) (**Figure 1**). Variance explained by the random factors of colony and year ranged from 32.47 to 90.76 and 3.09–30.94, respectively (**Supplementary Table 2**). Age had a significant effect on timing across both spring and fall migration, with ASY birds departing and arriving earlier by 5.52–7.18 days as compared to SY birds. Sex impacted the timing of spring departure and spring cross only, with males migrating 6.46 and 6.08 days earlier than females (**Supplementary Table 3**).

Within-individual variability between the first and second year of tracking was broad (0–24 days), with individual timing earlier (1–24 days), or later (1–23 days) in the second year of tracking (**Figure 2**). In spring, departure date varied by 0–20 days, spring crossing of 23.4◦N by 0–20 days, and arrival at the

TABLE 1 | Adjusted repeatability estimates and 95% credibility intervals for spring and fall migration (2008–2016) including departure dates (*n* = 67; *n* = 66), spring and fall crossing the Tropic of Cancer (*n* = 61, *n* = 34), and spring and fall arrival dates (*n* = 67; *n* = 67).


*Individuals were tracked for two spring migrations, except one individual tracked for three spring migrations. Estimates and credibility intervals were calculated using MCMCglmm (Hadfield, 2010)*\**. We included fixed effects of sex and age and controlled for nonindependence of year and individuals within breeding colonies by including them as random effects. Uninformed prior distributions (V* = *1, nu* = *0.002) were used for all variables.*

*Hadfield (2010)*\**.*

breeding site by 0–18 days. Fall migration timing was generally more variable at the individual level (depart range = 0–19 days; cross range = 1–15 days; arrive range = 0–24 days). Overall within-individual variability over the year was therefore up to 24 days, which spans the 8–20 day population-level advance in spring arrival date over 100 years, reported for purple martins (Arab and Courter, 2015) (**Figure 2**).

# DISCUSSION

Individual patterns of migration timing across years provide invaluable clues regarding the phenotypic plasticity of migration timing. We show that spring migration timing of a long-distance migratory songbird was more repeatable, from start-to-finish, than fall migration, with spring departure date being the most consistent between years. We found broad, within-individual variability in migration timing (year 1 as compared to year 2) at key points around the annual cycle (start and stop dates, and approximate midway points). We show that broad, intraindividual variability (up to 24 days earlier or 23 days later in the second year of tracking), was a feature of both spring and fall migration. Our results therefore suggest individual plasticity as a potential mechanism to account for population-level advances in spring arrival date (8–20 days) reported for purple martin (Arab and Courter, 2015). The degree of plasticity we show in individual martins also exceed mean population-level advancements (∼1 week) for spring migration reported for North American and European migratory landbird systems, including several species of long-distance aerial insectivores (Lehikoinen et al., 2019).

It has been difficult to reconcile the decadal-scale advances in spring migration timing at the population level, with intra-individual data that show high consistency of migration timing (e.g., Gill et al., 2014). It is debated whether strong selection for advanced timing and rapid micro-evolution could be responsible for population-level change (Knudsen et al., 2011) because advances via these mechanisms are generally predicted to take much longer. For example, using quantitative genetic models an observed 14-days advance in laying date in great tits (Parus major) was estimated to require more than two centuries to attain via micro-evolution (Charmantier and Gienapp, 2014). The predicted time period is considerably longer than the scale of the 10–100-years advances reported for multiple landbird species across North American and European migration systems (Lehikoinen et al., 2019). If microevolutionary responses cannot occur this quickly (Charmantier and Gienapp, 2014), and individuals do not exhibit a high level of phenotypic plasticity in spring timing, then how do we explain population-level advances in timing over short timescales? Our results show, that at least in purple martins, individual variation is a potential explanation for the kinds of advances reported in spring timing (Arab and Courter, 2015). We found within individual variation of up to 24 days whereas population level advancements for this species over 100 years are between 8 and 20 days (Arab and Courter, 2015). Our results contrast those for Icelandic godwit, where consistency in individual timing precluded individual plasticity of adult birds as a viable explanation for population-level advances in timing (Gill et al., 2014). Further studies are required across species and systems to determine whether individual plasticity is a

potential explanation for observed advances in timing. It would also be valuable for future studies to investigate the influence of additional factors on individual plasticity, such as nocturnal vs. diurnal migratory strategies, foraging guild, short vs. longdistance migration (Lehikoinen et al., 2019), nest strategy, and within-species patterns.

We found moderate repeatability of spring migration timing (range: r = 0.32–0.39) that is lower than reported in other studies of nocturnally migrating songbirds (e.g., range: r = 0.49–0.71, Stanley et al., 2012) and many long-distance migrants generally (Both et al., 2016). We found the highest repeatability of timing at spring departure from the non-breeding grounds as has been shown for red-backed shrikes, Icelandic whimbrels, and blacktailed godwits; perhaps owing to the strong role of endogenous cues in migration initiation (Gwinner, 1996; Pedersen et al., 2018; Carneiro et al., 2019; Senner et al., 2019). Sex and age were also important factors influencing spring departure timing, but differences between the sexes diminished by the time of arrival at breeding areas, whereas older birds were consistently earlier than younger ones. Relatively high repeatability of spring arrival date may reflect strong selection on timing at the breeding ground. In martins, high competition for nest cavities may further contribute to higher repeatability of spring arrival dates (Brown and Tarof, 2013). In contrast, fall timing was much less repeatable (r = 0.0001–0.001). Particularly low repeatability of fall migration arrival date (r = 0.0009), may reflect relaxed selection on this trait in purple martins; a species that is nonterritorial in winter and joins large communal roosts, in contrast to a songbird that is territorial in winter, where repeatability was relatively high (r = 0.62; Stanley et al., 2012). Intra-individual variation in nest success may also have contributed to low fall repeatability values in our study, if birds with failed nests depart earlier than birds attending young from successful broods. The overall, lower repeatability of migratory timing in our study of a diurnal migrant as compared to results for some nocturnally migrating songbirds (Both et al., 2016) may be influenced by migratory strategy (diurnal vs. nocturnal). Phenological and repeatability studies of songbirds have tended to focus more on nocturnally migrating species and comparisons of short and long-distance migrants (Both et al., 2016; Lehikoinen et al., 2019), however, diurnal and nocturnal migrants may exhibit different amounts of plasticity in timing to environmental change and should be further investigated.

We found higher repeatability in spring than fall for crossing of the Tropic of Cancer (23.4◦N), which is generally associated with crossing of the Gulf of Mexico in martins, as most individuals make a >800 km open-water crossing of this "barrier" during spring and fall migration (Fraser et al., 2013a,b). Lower repeatability in fall may indicate greater, population-level synchronization of the timing of crossing in this season. In wood thrushes, crossing of the gulf showed low repeatability during both spring and fall (r = 0.12, Stanley et al., 2012). In spring, the timing of gulf crossing in martins is not largely impacted by weather conditions (Abdulle and Fraser, 2018), thus we infer that higher consistency in individual timing at this barrier is a result of a carry-over effect of spring departure timing, rather than an influence of conditions at this stage of migration.

We speculate that the generally larger, intra-individual variation (up to 24 days) that we found for martins as compared to other songbirds (Stanley et al., 2012; Both et al., 2016; Pedersen et al., 2018), may reflect the nature of migration and social behavior in martins, and possibly other swallows. The intraindividual variability we found for purple martins is more similar to broad, within-individual variation recently shown for some shorebirds (Senner et al., 2019; Verhoeven et al., 2019), than to data reported for nocturnally migrating songbirds. Purple martins are diurnal migrants that roost in large flocks during stopovers across migration (Brown and Tarof, 2013), and may use island-like habitats for stopover (Fraser et al., 2017; Fournier et al., 2019). Large social aggregations and suitable stopover habitat are unevenly distributed across a migratory landscape, thus martin stopover decisions may be influenced by these social factors which could contribute to variation in their individual timing. In contrast, diurnal songbirds that migrate singly during the night may not require social stopover cues to the same degree, which may favor more independent and consistent migration schedules. While it has been demonstrated that shortdistance migrants may exhibit swifter phenological shifts in response to environmental change than long-distance migrants (Hurlbert and Liang, 2012; Kullberg et al., 2015; Takuji et al., 2017), whether diurnal, long-distance migrants exhibit greater phenotypic plasticity than nocturnal ones would be valuable to determine.

Our data did not provide the opportunity to examine variation in repeatability across populations breeding at different latitudes, but such within-species investigations are an important frontier. Such research would be particularly important for purple martins and other aerial insectivores, where strong north-south patterns of population decline are reported (Nebel et al., 2010), and where relative limitations in behavioral plasticity between populations could be playing a role. More northern breeding populations may exhibit larger advances in spring arrival dates than more southern ones (Arab and Courter, 2015), which should be investigated in concert with individual-level patterns.

#### CONCLUSION

In an era of rapid, global environmental change, it is critical that we address the degree to which migratory birds can mount phenotypic responses to change. In this first investigation using a diurnal migrant and the largest repeat-tracking data set for a songbird, we show that phenotypic plasticity in migration timing is a potential mechanism to explain decadal-scale, populationlevel advancement in spring migration timing. It remains to be determined whether the degree of individual plasticity we show is connected to, or cued by, temperature and whether any advances in migration timing are sufficient to match advances in seasonal phenology of lower trophic levels. Future research should also investigate the role of environmental cues and

#### REFERENCES


other mechanisms contributing to within-individual variation in migration timing.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

#### ETHICS STATEMENT

This study was conducted in accordance with the recommendations of the Ornithological Council's Guidelines to the Use of Wild Birds in Research' and was approved by the University of Manitoba and York University Animal Care Committees (2009- 2W, F14-009/1-3).

#### AUTHOR CONTRIBUTIONS

KF, AS, and JS conducted fieldwork. KF, AS, EG, and CG analyzed the data. KF, AS, EG, JS, and CG wrote the manuscript.

#### ACKNOWLEDGMENTS

For field assistance and support we thank Nanette Mickle, Paul Mammenga, Tim Shaheen, Kelly Applegate, Michael North, Larry Leonard, Edward Cheskey, Megan McIntosh, Pat Kramer, Cassandra Silverio, Lee Bakewell, Richard Doll, Myrna Pearman, Alisha Ritchie, Bridget Stutchbury, and John Tautin. Funding and support were provided by University of Manitoba, NSERC, Nature Canada, Central Lakes College, Minnesota Audubon, Minnesota Ornithologists' Union, Brainerd Lakes Audubon Society, Mille Lacs Band of Ojibwe, and Ellis Bird Farm.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00324/full#supplementary-material


passerine. Oikos 105, 55–64. doi: 10.1111/j.0030-1299.2004.12 516.x


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Fraser, Shave, de Greef, Siegrist and Garroway. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Sex-Specific Spatiotemporal Variation and Carry-Over Effects in a Migratory Alpine Songbird

#### Devin R. de Zwaan<sup>1</sup> \*, Scott Wilson<sup>2</sup> , Elizabeth A. Gow1,3 and Kathy Martin1,4

*<sup>1</sup> Department of Forest and Conservation Sciences, University of British Columbia, Vancouver, BC, Canada, <sup>2</sup> Wildlife Research Division, Environment and Climate Change Canada, National Wildlife Research Centre, Ottawa, ON, Canada, <sup>3</sup> Department of Integrative Biology, University of Guelph, Guelph, ON, Canada, <sup>4</sup> Environment and Climate Change Canada, Pacific Wildlife Research Centre, Delta, BC, Canada*

#### Edited by:

*Nathan R. Senner, University of South Carolina, United States*

#### Reviewed by:

*Rien Van Wijk, Van Wijk Eco Research, Denmark Kristina Paxton, University of Hawaii at Hilo, United States*

> \*Correspondence: *Devin R. de Zwaan drdezwaan@gmail.com*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *27 March 2019* Accepted: *15 July 2019* Published: *31 July 2019*

#### Citation:

*de Zwaan DR, Wilson S, Gow EA and Martin K (2019) Sex-Specific Spatiotemporal Variation and Carry-Over Effects in a Migratory Alpine Songbird. Front. Ecol. Evol. 7:285. doi: 10.3389/fevo.2019.00285* For migratory animals, events at one stage of the annual cycle can produce constraints or benefits that carry over to subsequent stages. Differing life-history strategies among individuals can influence the expression of these carry-over effects, leading to pronounced within-population variation in migration. For example, reproductive roles can drive spatiotemporal segregation during the non-breeding season and promote sex-specific carry-over effects, such as reproductive effort affecting autumn migration behavior. For an alpine breeding population of horned larks *Eremophila alpestris* in northern British Columbia, Canada, we addressed sex-specific variation in migration behavior and carry-over effects during both autumn and spring migration using light-level geolocators. Males spent more time farther north and arrived an average of 6 days earlier at the breeding site in spring. Females delayed autumn departure following greater reproductive effort, in turn demonstrating flexible migration behavior by increasing migration speed and decreasing stopover use. Males maintained autumn migration behavior regardless of reproductive effort or departure date. Finally, both sexes used staging areas in spring (average stopover = 41 days), with consequences for breeding success. Individuals that used staging areas during spring migration exhibited greater nest success and produced 1.8 more fledglings on average than those that migrated directly from their winter site. Consistent use of staging areas may allow individuals to monitor environmental conditions and optimize their breeding arrival date to acquire high-quality territories while avoiding the cost of arriving too early in a harsh alpine habitat. Overall, our results indicate: (1) sex-specific flexibility in migration strategies that carry-over to and from the reproductive period, and (2) spring staging areas may be a critical component of the annual life-cycle for alpine breeding larks. These behaviors may be particularly important for alpine and arctic birds because the stochastic nature of their breeding habitat likely selects for flexible responses to prevailing conditions.

Keywords: alpine, flexible life-history, Eremophila alpestris, light-level geolocation, phenology, plasticity, protandry, stopover ecology

# INTRODUCTION

Migratory birds can spend up to 75% of the annual cycle away from the breeding site (Webster et al., 2002), often using multiple habitats in different locations and for variable time periods (Marra et al., 1998; Briedis et al., 2018). How individuals disperse spatially across the landscape and the timing of migration events can have important consequences for both individual fitness and population dynamics (Møller et al., 2008; Legagneux et al., 2012; Runge et al., 2014; Wilson et al., 2018). Migratory strategies can vary among populations (Gilroy et al., 2016; Knight et al., 2018; Gow et al., 2019), in response to inclement weather (Morganti et al., 2011; Schmaljohann et al., 2017), and among individuals of different age or sex classes (Tøttrup et al., 2012; McKinnon et al., 2014; Woodworth et al., 2016; Briedis et al., 2019). Understanding drivers of within-population variation in migratory strategies and how differences are linked across seasons is critical to understanding the ecological and evolutionary processes shaping life-history dynamics throughout the full-annual cycle (Marra et al., 2015; Paxton and Moore, 2017).

Spatial and temporal segregation of the sexes during the nonbreeding season occur in avian populations, but the drivers of these patterns are not well-understood (McKinnon and Love, 2018). For sexually dimorphic species, the larger sex (often male) may winter farther north or closer to the breeding site (Cristol et al., 1999; Macdonald et al., 2015) and exhibit protandry (i.e., earlier spring arrival for males; Morbey and Ydenberg, 2001). Non-mutually exclusive hypotheses such as the "body size," "arrival time," and "social dominance" hypotheses, all predict that larger individuals can withstand harsher winter conditions that enable them to remain closer to the breeding site (Ketterson and Nolan, 1976; Gauthreaux, 1978) and better monitor environmental cues to match breeding site arrival with optimal weather conditions (Saino et al., 2010). The selective advantage of an earlier arrival includes improved territory and mate acquisition (Kokko, 1999; Reudink et al., 2009) and reproductive success (Norris et al., 2004; Smith and Moore, 2005; Gienapp and Bregnballe, 2012), and is most apparent for the more territorial sex (Møller, 2004; Kokko et al., 2006). Thus, the reproductive roles of each sex can produce spatial and temporal differences among individuals during the non-breeding season (Gow and Wiebe, 2014; Meissner, 2015), and consequently may lead to sex-specific fitness consequences of variation in migration behaviors (Saino et al., 2017).

The annual cycle is partitioned into specific life-history stages that are linked across seasons and conditions that influence energy reserves in one season can "carry-over" to influence subsequent stages (Marra et al., 1998). Conditions during the non-breeding season can influence breeding site arrival and success (Norris et al., 2004; Norris, 2005; Harrison et al., 2011). Similarly, breeding success may affect autumn departure date (Stutchbury et al., 2011; Meissner, 2015; van Wijk et al., 2017) which in-turn can influence stopover ecology and arrival at the winter site (Briedis et al., 2016; Gow et al., 2019). Thus, carry-over effects can cascade through temporallylinked phases of the annual cycle during both autumn and spring migration (Piersma, 1987). Behavioral adjustments such as altering stopover duration can allow individuals to buffer cascading effects, but the ability to adjust behavior may be condition-dependent (Gómez et al., 2017). Thus, during autumn migration (post-breeding), individuals that invested more in reproduction may have a reduced capacity for flexible migration behavior, consequently exhibiting greater delays in arrival at non-breeding sites. Similarly, energy-constrained individuals that delay spring migration and arrive late to the breeding site may delay clutch initiation or experience reduced breeding success. Given differences in reproductive investment, males and females may have differing abilities to compensate for delayed migration events.

Variation in migratory strategies among individuals in response to reproductive effort and timing is a type of phenotypic flexibility, or the capacity to reversibly alter phenotypic expression (i.e., behavior) in response to prevailing conditions (Piersma and Drent, 2003). Populations may differ in their capacity for flexible migration behavior based on the breeding and non-breeding habitat to which they are adapted. For example, short-distance migrants exhibit greater flexibility in phenology, potentially by monitoring environmental conditions near the breeding site (Usui et al., 2017; Lehikoinen et al., 2019). Additionally, phenotypic flexibility should be adaptive in stochastic conditions (Piersma and Drent, 2003), and as such birds that breed in highly variable environments, like alpine or arctic habitats, may be more likely to demonstrate and benefit from flexible migration behaviors. Recent studies from far northern latitudes indicate strong potential flexibility in migration timing and stopover behavior (e.g., Krause et al., 2016; Schmaljohann et al., 2017). Addressing within-population variation of migratory traits for alpine birds will advance our understanding of sexspecific flexible migration behaviors and thus the capacity for populations to respond to rapidly changing environments (Shaw, 2016; Beever et al., 2017).

In this study, we examined sex-specific variation in migration behavior and the carry-over effects between reproductive output and migration strategies for an alpine breeding population of horned larks Eremophila alpestris. Specifically, we tested four predictions involving spatial and temporal differences between sexes across the full-annual cycle. First, since larks are sexually dimorphic (Beason, 1995) we predicted spatial sex segregation during the non-breeding season such that the larger males would remain closer to the breeding site. Second, we expected that males would depart and arrive at the breeding site earlier than females, demonstrating protandry. Third, we predicted greater nesting effort or success would influence departure dates from the breeding site, and that a later departure would influence subsequent arrival at non-breeding sites. We expect breeding effort to more strongly influence female migration timing because female larks invest more in reproduction, being solely responsible for nest building, incubation, and half of offspring provisioning (Goullaud et al., 2018). Finally, we predicted that migration distance and departure date from the non-breeding site would influence spring arrival date and subsequent breeding success. We expected this interdependency to be most pronounced in males due to the territorial advantage of arriving early on the breeding site (Reudink et al., 2009).

### METHODS

#### Study System and Migratory Behavior

From 2015 to 2018, we studied a population of horned lark in ∼4 km<sup>2</sup> of alpine tundra (elevation: 1,650–2,000 m) on Hudson Bay Mountain (HBM) in northern British Columbia, Canada (54.8◦N, 127.3◦W). Horned larks are open-country, groundnesting songbirds (28–40 g) that breed in sparsely-vegetated habitats from 0 to over 4,000 m above sea level (Beason, 1995). Snow-melt at HBM often extends into mid-June, resulting in short breeding seasons from mid-May or early-June to late July (41–57 days; MacDonald et al., 2013; Martin et al., 2017). Females lay 2–5 eggs per nest (average = 3.6) and initiate 1–3 clutches per season, including re-nests (Camfield et al., 2010). Most females produce a single brood; on average 7.7% of females have two broods in one season (de Zwaan, unpublished data).

Horned lark populations throughout the United States are partially migratory or resident, while those in Canada are obligate migrants (Beason, 1995). Wintering locations are predominantly below 49◦ latitude and extend as far south as the Chihuahuan grasslands in Mexico (Sullivan et al., 2009). Larks have been observed arriving in southern Canada as early as February (Cannings and Threlfall, 1981). The migratory connectivity or phenology of alpine lark populations is unknown, but they are assumed to be at least altitudinal migrants (Beason, 1995).

### Field Methods

Nests were located during the nest building or incubation stage using behavioral observations. If discovered during incubation, clutch initiation was estimated by back-calculating from hatch date using an average incubation period of 12 days (de Zwaan et al., 2019). Nests were monitored every 2 days to record nest fate (fledged, depredated), clutch size, and number of fledglings. Both males and females were captured at the nest during the nestling period using bownets that were triggered when adults entered to provision the nestlings (de Zwaan et al., 2018). Each adult was measured for body size and condition traits (i.e., wing, tarsus, mass, fat) and ringed with one U.S. Geological Survey numbered aluminum ring and three color rings for subsequent identification.

From 2015 to 2017, we fit 37 males and 22 females with Intigeo-P65B1-11 light-level geolocators (Migrate Technology Inc.). The geolocators were attached using a leg-loop harness (Rappole and Tipton, 1990) tied using 45 lb test nylon string (Lee Valley Tools Ltd., Ottawa, Canada), allowing the fit of each harness to be adjusted for individuals in the field. The string was double knotted and glued with epoxy to prevent geolocator loss. The total weight of the geolocator (0.77 g) plus harness (0.20 g) was 0.97 g which, on an average lark for this population (35.1 g; Camfield et al., 2010), is 2.8% of body mass. All birds were captured on fair-weather days and released in under 12 min. We monitored individuals for ∼15 min after release to ensure there were no immediate effects of the geolocator. The following year, returning individuals with geolocators were captured at the nest during the nestling stage (males) or the incubation stage (females) using a noose-line trap surrounding the nest (de Zwaan et al., 2018). We removed the geolocator from recaptured birds, measured body size traits, and released them in under 5 min.

### Geolocator Analysis

All analyses were performed in R version 3.5.1 (R Core Team, 2018). Based on re-sights, the average return rate for geolocator birds was 59.3% (35 of 59 tagged birds) vs. 68.2% for ringed birds without geolocators. Due to significant trap avoidance in the first season, we retrieved 17 geolocators (8 males, 9 females). Pre-processing of drift-adjusted.lux files and calibration were conducted using the package "GeoLight" (Lisovski and Hahn, 2013). We used a light level threshold of 1 to estimate twilights and removed outliers if they differed by more than 30 min from adjacent twilight estimates. To calibrate the data, we used the "Hill-Ekstrom method" which estimates the zenith angle based on the lowest amount of variance in latitude estimates during stationary periods (Hill and Braun, 2001; Ekstrom, 2004). This method provides more accurate location estimates than both "roof-top" and "on-the-bird" calibration when a lengthy stationary period occurs near an equinox (Lisovski et al., 2012). We identified likely stationary periods (>10 days) during the non-breeding season using the "changeLight" function. Stationary periods were distinguished from movement based on high probability changes in twilight times, where a change point probability greater than the 90th percentile of all probabilities calculated within the entire migration period indicated movement. We then estimated the zenith angle based on the longest stationary period in the non-breeding season.

Location estimates were derived using the package "SGAT" (Sumner et al., 2009), because it allows prior distribution knowledge and bird behavior to be incorporated in a Bayesian framework and tends to be more accurate than location estimates derived from "GeoLight" (Roberto-Charron, 2018). "SGAT" uses the curve method to estimate locations based on the difference between twilight times, in combination with movement behavior priors which apply constraints to the estimates. We first identified likely stopover locations where the bird was stationary for ≥3 days within a distance threshold of 150 km. We created a prior movement model of flight speeds applied to each day of the migration period where most speeds were from 0 to 40 km/h but allowed for speeds up to 80 km/h with rapidly diminishing likelihood. This allows travel distances of 200 km per 5 h flight to be common, up to a maximum of 400 km, which is comparable to migration estimates for other songbirds (Hall-Karlsson and Fransson, 2008; Macdonald et al., 2015; Wright et al., 2018).

We also developed a land mask to restrict the locations of individuals to suitable habitats. We used a 1 km<sup>∗</sup> 1 km landcover database from Tuanmu and Jetz (2014) which classifies the globe into 12 categories based on a consensus from multiple remotesensing sources. Since horned larks are obligate open-country birds that avoid forested areas, we used "herbaceous vegetation or grassland" and "barren" (often alpine) landcover categories as identified by a remote-sensing consensus of >25%. Agricultural fields are also potential non-breeding habitat (Beason, 1995), and therefore we included the category "cultivated or managed

vegetation" which includes fallow fields and rangeland. Birds were restricted to stopping in these habitat types but could move freely between stopover sites over all other landcovers, including water.

sake of clarity. Each individual is represented by at least one winter site and up

to one or two stopover sites (*n* = 17 birds).

We fit a group Estelle model incorporating the behavior model and land mask to estimate the location of stationary and movement periods with a multivariate normal error distribution. The model was trained using 1,000 initial iterations with a 20 iteration burn-in, followed by MCMC chains including 10 rounds of 300 iterations each. The final model was run with 2,000 iterations.

The average geolocator error based on location estimates from a known location—the breeding site—was 59 ± 56 (s.d.) km (range: 3–224 km) in latitude and 57 ± 50 km (1–185 km) in longitude. Similarly, the average 95% credible interval for point estimates of non-breeding stationary periods (unknown locations) was 58 km (range: 30–101 km) for latitude and 61 km (range: 34–122 km) for longitude. As a result, we defined the non-breeding season as beginning when an individual moved >150 km from the breeding site in a consistent direction. Similarly, we considered estimated stopover locations to be distinct if they were separated by at least 150 km, otherwise they were grouped by calculating the centroid weighted by the number of days at each location. Location estimates indicating large latitudinal fluctuations 2 weeks prior-to and after an equinox were removed because at this time estimates of latitude can be inaccurate, although longitude is still relevant (Hill, 1994; Ekstrom, 2004). As a result, if a bird arrived or departed from a stationary point during an equinox period, the arrival and departure times were approximated based on longitudinal shifts and when the longitude stabilized to match the stationary site. Normally, estimates of departure and arrival dates are accurate to ±2 days. However, when estimating phenology during an equinox period based on longitudinal shifts, we assumed accuracy to be ±5 days. This inaccuracy predominately occurred when estimating arrival at prolonged stopovers during autumn migration.

#### Migration Terminology

Following location estimation, we calculated several migration traits that require definition:

Winter site—The non-breeding site farthest to the south and with the longest duration of stay.

Stopover—Any stationary period ≥ 3 days between the breeding and winter site. Due to location estimate error, stopovers represent general but distinct regions within which an individual may use several locations.

Spring staging area—Stopovers during northward migration where a bird remained stationary for >20 days. The length was chosen to be consistent with other examples in the literature (Renfrew et al., 2013; Gow et al., 2019) and to account for error in location estimates. We acknowledge that staging areas can be considered alternative wintering locations, similar to observations for Neotropical migrants like purple martin (Progne subis; Stutchbury et al., 2016) and veery (Catharus fuscescens; Heckscher et al., 2011). However, since they generally occurred on the migration route between the wintering and breeding site, it is reasonable to treat these locations as staging areas (Warnock, 2010; Bayly et al., 2018).

Absolute distance—Calculated as the linear distance between the breeding and winter site using the Vincenty ellipsoid method to account for curvature in the earth (Vincenty, 1975).

Route distance—Absolute distance does not account for route variation. Therefore, we also calculated the one-way sum of the linear distances between the breeding site, stopovers, and winter site for both southward and northward migration as a measure of realized distance.

Staging distance—The linear distance between the staging area (or winter site if an individual did not demonstrate staging behavior) and the breeding site. This allowed us to address proximity to the breeding site and spring phenology in a manner


TABLE 1 | Migration behavior for an alpine breeding population of horned larks in northern British Columbia for both autumn and spring migration.

*Top values are the mean* ± *SD, while those in brackets depict the range. The location that departure and arrival refer to depends on the migration season. For example, departure in autumn is from the breeding site, but in spring it is from the winter site. Duration refers to the staging area in spring, often in the Thompson-Okanagan or northern Columbia Plateau regions.*

that is comparable among larks who did or did not exhibit staging behavior.

Migration speed—Measured as the route distance divided by the duration, and thus incorporates flight speed and stopover duration. Due to error in location estimates, distance and speed variables should be considered relative for comparison among individuals and not exact differences.

#### Statistical Analysis

To test sex-specific differences in spatial distribution and phenology, we used four separate one-way Analysis of Variance (ANOVA) tests with sex as the explanatory variable and absolute distance, staging distance, breeding site departure, or breeding site arrival date as the response variables. Separate ANOVAs were necessary because each response variable reflects different migration events that are distinct in time and space.

We fit simple linear models to address: (1) the influence of breeding success on departure date, (2) subsequent effects of departure date on autumn migration speed, stopover use, and non-breeding arrival dates, and (3) effects of using spring staging areas and breeding site arrival date on breeding success. Breeding success was calculated as the proportion of nests that successfully fledged offspring in a season. Total nesting attempts ranged from 1 to 3 but most individuals attempted two nests (12/17 birds; 5 males, 7 females). For autumn stopover behavior, we fit a logistic regression where an individual either did (1) or did not (0) remain stationary at a prolonged stopover for >20 days. Since several individuals migrated during part of the autumn equinox, identifying extended stationary periods was more reliable than the total number of stopovers. All other response variables were fit to a Gaussian distribution.

For all associations, we fit separate models for males and females. If neither were significant, we combined the data from both sexes and report the association from a model with sex as an additive term. We used this process because we were specifically interested in sex-specific differences in flexible migration behaviors. For effects of spring migration behavior on breeding parameters, we addressed two response variables: (1) nest success, and (2) the total number of fledglings in a season. An R <sup>2</sup> was calculated for each model to indicate the strength of the association.

#### RESULTS

Horned larks from an alpine breeding population in northern British Columbia, Canada wintered predominantly east of the Cascade mountain range in Washington and Oregon, USA, with one female overwintering in southeast Idaho (**Figure 1**). Most males and females exhibited "loop" migrations, traveling southward at higher elevations along the Coast Mountains to the northern Columbia Plateau region before dispersing farther south or east (**Figure 1A**). Northward migration tended to be at lower elevation through the Thompson-Okanagan and up through the Central Interior Plateau of British Columbia (**Figure 1B**). Individuals departed the breeding grounds within a span of 31 days from early August to early September and arrived at the wintering grounds over a span of 81 days (**Table 1**), subsequently spending an average of 170 days or 5.5 months at the wintering site (range: 56–242 days). Both sexes demonstrated faster spring than autumn migration speeds (**Table 1**).

The variation in arrival and duration of stay at winter sites followed from variable stopover use. During autumn migration, at least 11 individuals (∼65%) appeared to remain stationary at a prolonged stopover site prior to continuing to the wintering site. However, because this period tended to occur at least partially within the fall equinox, we could not accurately estimate arrival date or exact duration of stay, although it was usually possible to estimate departure date. During spring migration,

both males (5/8; 63%) and females (5/9; 56%) stopped at staging areas on their northward route, predominantly within the Thompson-Okanagan or northern Columbia Plateau regions (**Figure 2**). Individuals remained at staging areas for periods of 21−66 days (**Table 1**).

#### Spatial and Phenological Segregation of the Sexes

The mean winter coordinates for males (46.8 ± 1.5◦N, −120.1 ± 0.9◦W; mean ± SD) tended to be farther north than for females (45.8 ± 1.3◦N, −119.6 ± 1.6◦W), although the average distance between the breeding and wintering sites did not differ between sexes (**Table 2**). However, when considering the spring staging area, the spatial segregation between males (48.4 ± 1.3◦ N, −120.0 ± 0.6◦W) and females (46.9 ± 1.3◦N, −119.8 ± 0.9◦W) was more distinct (**Figure 2**). Thus, males spent less time at their winter sites and significantly more time closer to the breeding site (**Table 2**). Autumn departure dates from the breeding grounds were similar for males and females (**Table 2**). During spring migration, males arrived at the breeding grounds an average of 6 days earlier than females (**Table 2**).

#### Effects of Reproductive Success on Autumn Migration Behavior

Females that had greater breeding success departed earlier from the breeding site (t = −2.4, P = 0.04, R <sup>2</sup> = 0.46), while males did not exhibit an association (t = 0.1, P = 0.90, R <sup>2</sup> = 0.00; **Figure 3A**). Departure date was positively associated with arrival at the first prolonged stationary site (i.e., stopover > 20 days or wintering site) for males (t = 3.2, P = 0.02, R <sup>2</sup> = 0.63), but not for females (t = 1.2, P = 0.26, R <sup>2</sup> = 0.17; **Figure 3B**). However, when just the winter site was considered, arrival was not associated with departure from the breeding site for either sex (t = −1.5, P = 0.16, R <sup>2</sup> = 0.15; **Figure 3B**), likely due to variable stopover use. Later departing females increased migration speed to the first stopover site (t = 2.4, P = 0.05, R <sup>2</sup> = 0.45; **Figure 3C**), and were less likely to exhibit prolonged stopover behavior (t = −5.3, P < 0.01, R <sup>2</sup> =0.82; **Figure 3D**). In contrast, regardless of departure date, males maintained migration speed (t = 0.2, P = 0.83, R <sup>2</sup> = 0.01; **Figure 3C**) and stopover use (t = 0.0, P = 1.00, R <sup>2</sup> = 0.00; **Figure 3D**).

### Effects of Spring Migration Behavior on Breeding Success

While breeding season arrival date was positively associated with departure date from the last prolonged stationary site (i.e., staging area or wintering site) for both sexes (t = 4.6, P < 0.01, R 2 = 0.69; **Figure 4A**), arrival date did not correlate with clutch initiation date (Pearson's product-moment correlation: r<sup>p</sup> = 0.32, t = 1.3, d.f. = 15, P = 0.21). However, larks that stopped at staging areas while traveling northward demonstrated greater breeding success (t = 2.7, P = 0.02, R <sup>2</sup> = 0.39; **Figure 4B**) and consequently produced more fledglings than those that stopped farther south or migrated directly from the winter site (t = 2.4, P = 0.03, R <sup>2</sup> = 0.30). On average, larks that stopped at staging areas farther north produced 1.8 more fledglings over the season (2.2 ± 0.6; mean + SE; n = 10) than individuals that remained farther south (0.4 ± 0.4; n = 7).

# DISCUSSION

#### Sex-Specific Differences in Spatial Distribution and Phenology

In an alpine breeding population of horned larks, males spent a significantly longer time farther north and closer to the breeding site than females. In contrast to expectations for a short-distance migrant, males arrived on the breeding grounds an average of 6 days earlier, demonstrating relatively minor protandry. Shortdistance migrants often exhibit greater protandry (∼2 weeks) compared to long-distance migrants (2–8 days; Tøttrup and Thorup, 2008; Briedis et al., 2019). Closer proximity to the breeding site should allow individuals to monitor environmental cues (Ouwehand and Both, 2017; Lehikoinen et al., 2019), to arrive as early as possible and gain territorial benefits (Reudink et al., 2009). However, the competitive advantage of arriving early is balanced with the cost of arriving too early when harsher weather may limit available resources and deplete energy reserves (Kokko et al., 2006; Coppack and Pulido, 2009). In a stochastic alpine habitat, the cost-to-benefit ratio of arriving early may be greater than in a less seasonal and more environmentally consistent low elevation habitat. At our alpine site, daytime temperatures often remain at or below 0◦C throughout the first half of May, with frequent storms and high wind speeds (Camfield and Martin, 2009; Martin et al., 2017). Therefore, both males and females may experience stabilizing selection to arrive at similar times.

TABLE 2 | Analysis of Variance (ANOVA) results for sex-specific differences in spatial distribution and phenology during the non-breeding season.


*Absolute distance is the linear distance between two points rather than route distance to facilitate comparison of spatial segregation. Staging distance is the distance between the breeding site and the staging area (or winter site if the individual did not exhibit staging behavior). Departure and arrival dates refer to the breeding season only. Values in brackets depict the range. P-values with a double asterisk are considered significant, while a single asterisk is marginal.*

points indicate males and females separately for comparison. Additionally, (C) late departing females increased migration speed to the first stopover, and (D) reduced use of prolonged stopovers following delayed departure. Residuals were calculated by subtracting from the mean such that positive is greater and negative is less than average.

We did not find a relationship between breeding site arrival and clutch initiation date, which contradicts patterns observed in migrating songbirds breeding at low elevation (Norris et al., 2004; Woodworth et al., 2016). A selective advantage for early spring arrival is predicated on expected high-quality territory acquisition and subsequent benefits for breeding success (Morbey and Ydenberg, 2001). In the alpine, regardless of arrival date, ground-nests cannot be initiated until late May due to extensive snow cover. Even after onset of snowmelt, the higher probability of severe and cold early-season storms that lead to nest failure can limit the fitness benefits of breeding early (Martin et al., 2017). Alpine and arctic birds likely have greater flexibility to respond to proximate environmental conditions to advance or delay both breeding site arrival and clutch initiation date (Bears

et al., 2009; Boelman et al., 2017). As a result, a relationship between arrival and clutch initiation date is likely moderated by variable environmental conditions, regardless of any potential competitive advantage to arriving early.

# Cascading Effects of Breeding Success on Autumn Migration

Breeding effort and success showed sex-specific associations with autumn migration behavior, indicating potentially different mechanisms driving the timing of autumn migration, and subsequently different migration strategies. During autumn migration, breeding site departure date was delayed with reduced breeding success for females but not males. Since females will renest multiple times following failed attempts, this suggests greater energy investment during the breeding season can influence timing of autumn migration. Although similar evidence is limited, individuals that invest more in reproduction may be energetically constrained to depart later (Wojczulanis-Jakubas et al., 2013). For example, dunlin (Calidris alpina) delay breeding site departure following greater breeding success (Meissner, 2015), and wood thrush (Hylocichla mustelina) in low body condition at the end of the breeding season tend to remain farther north for longer, potentially because these individuals have delayed molting behavior (Stutchbury et al., 2011). Horned larks molt at the breeding site before departing (Beason, 1995), and thus molt could exacerbate departure delays for females considering the substantial energy investment required when late reproduction and molt coincide (Flinks et al., 2008; Borowske et al., 2017). Regardless of the mechanism, our results indicate that greater breeding effort has the potential to delay departure from the breeding site.

Males and females commonly differ in their capacity for flexibility across a wide range of behaviors (Nakagawa et al., 2007), including migration strategies (Both et al., 2016). We demonstrated sex-specific abilities to compensate for delayed departure during autumn migration, or to buffer cascading constraints so they did not carry over indefinitely (Conklin and Battley, 2012; Senner et al., 2014). Termed "reversible state effects," conditions that influence energy reserves in one stage of the annual cycle can produce temporary constraints that influence subsequent stages but are corrected over time (Senner et al., 2015). While males maintained consistent stopover behavior, later departing females increased migration speed and decreased stopover use, providing some support for reversible state effects. Short-distance migrants likely have a greater capacity for flexible stopover behavior (Schmaljohann and Both, 2017), which can compensate for a late departure (Stutchbury et al., 2011). Speeding up migration implies an urgency to arrive at a destination and this selective pressure is thought to be greater in spring than autumn (Horton et al., 2016). It is unclear why female larks speed up autumn migration, but one possibility is that poor weather conditions at northern stopovers later in autumn may cause less territorial individuals, like females, to continue migrating south (Stutchbury et al., 2016; Schmaljohann et al., 2017).

# Spring Staging Behavior

We observed exceptionally prolonged staging behavior during spring migration, with some individuals remaining at northern staging areas for up to 2 months. The traditional expectation for stopovers is that their duration should be minimized, such that birds stay just long enough to refuel (Alerstam et al., 2003). While evidence for staging in songbirds originally stemmed from long-distance migrants preparing to cross major landscape features (e.g., the Sahara; Arlt et al., 2015), recent observations indicate this behavior may be more prevalent in songbirds than once thought (Bayly et al., 2018). For example, bobolink Dolichonyx oryzivorus may stop for more than 20 days during both southward and northward migration (Renfrew et al., 2013),

are delayed and negative values are advanced.

while rusty blackbirds Euphagus carolinus exhibited average stationary periods of 25.5 days (max = 62 days) prior to arriving at the breeding site (Wright et al., 2018). Interestingly, staging behavior is most commonly observed in capital breeders like snow goose Anser caerulescens when preparing to arrive at unpredictable, high latitude breeding sites (Bêty et al., 2004). While songbirds are income breeders, staging behavior may allow larks to improve fat reserves to respond to variable, early season conditions upon arriving in the alpine.

For larks, staging areas may be key components of the annual cycle (Pledger et al., 2009), particularly because individuals that exhibited staging behavior farther north demonstrated greater reproductive success. The use of a specific staging area may have intrinsic value that carries over to reproduction. For example, high-quality individuals may be able to move earlier to northern staging areas to maximize limited resources in a harsher environment and thus be in a better position to arrive at the breeding site at an optimal time to acquire high-quality territories and mates. This may explain why two males remained in the region for the entire non-breeding season. In support of this mechanism, black-and-white warblers Mniotilta varia that arrive early at stopovers close to the breeding grounds remain longer and accumulate more fat than later arriving birds (Paxton and Moore, 2017). Future research that addresses the habitat quality of staging areas, as well as, the body condition of individuals that remain at these sites for prolonged periods of time would help identify the drivers of this behavior in an alpine population of horned larks.

#### CONCLUSION

We report sex differences in spatial distribution during the non-breeding season and flexibility in migration behavior. Since our sample size is modest, with 17 individuals over 3 years, we acknowledge that our results, although intriguing, may not capture the full range of variability in certain parameters measured in this study (e.g., departure date, nest success). Nevertheless, the sex-specific patterns we observed indicate the potential importance of flexible migration behaviors in shaping individual life-history strategies and fitness, with implications for population dynamics. If individuals from one sex are more constrained to specific migration behaviors or non-breeding sites, they may be more susceptible to changing environmental and land-use conditions which may influence sex-biased mortality or dispersal rates. As a result, identifying critical non-breeding sites, as well as, within-population variation in how these locations are used is an important step in the conservation of declining open-country birds (Cohen et al., 2017). To this end, we observed striking spring staging behavior and its potential influence on subsequent breeding success, suggesting staging areas in the Thompson-Okanagan and Columbia Plateau may

# REFERENCES

Alerstam, T., Hedenström, A., and Åkesson, S. (2003). Long-distance migration: evolution and determinants. Oikos 103, 247–260. doi: 10.1034/j.1600-0706.2003.12559.x

be critical components of the annual cycle. Staging behavior is difficult to recognize and poorly understood among songbirds owing to their small size and often greater dispersion across the landscape, but future studies like ours will improve our ability to identify staging areas and the factors influencing this behavior.

# DATA AVAILABILITY

Raw geolocator data will be made available in the Movebank data repository. All other datasets are available upon request. The data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

# ETHICS STATEMENT

All procedures and protocols for this study were approved by the University of British Columbia's Animal Care Committee (A15-0027) and are in accordance with the Canadian Council on Animal Care's national guidelines. All data were also collected under a Scientific Permit for Capture and Banding of Migratory Birds from Environment and Climate Change Canada (10365 DS and 10761 J).

# AUTHOR CONTRIBUTIONS

DdZ, EG, SW, and KM conceived the ideas and designed the methodology. DdZ collected and analyzed the data, and led writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

# FUNDING

Funding for this research was provided to DdZ by the Northern Scientific Training Program, Werner and Hildegard Hesse Research Award (University of British Columbia), Northwest Scientific Association, American Ornithological Society, and Society of Canadian Ornithologists (Taverner Award), to EG by the Izaak Walton Killam Memorial Fund, to KM and DdZ by the Natural Sciences and Engineering Research Council of Canada and to KM and SW by Environment and Climate Change Canada.

# ACKNOWLEDGMENTS

We thank A. Sulemanji, D. Maucieri, S. Hudson, N. Morrel, T. Altamirano, N. Froese, N. Bennett, K. Hicks, and C. Rivas for their contributions to data collection. Special thanks to J. Fox and S. Lisovski for helpful feedback on geolocator application and analysis.

Arlt, D., Olsson, P., Fox, J. W., Low, M., and Pärt, T. (2015). Prolonged stopover duration characterises migration strategy and constraints of a long-distance migrant songbird. Anim. Migr. 2, 47–62. doi: 10.1515/ami-2015-0002

Bêty, J., Giroux, J. F., and Gauthier, G. (2004). Individual variation in timing of migration: causes and reproductive consequences in greater snow geese (Anser caerulescens atlanticus). Behav. Ecol. Sociobiol. 57, 1–8. doi: 10.1007/s00265-004-0840-3


modulated by environmental conditions. Proc. R. Soc. B 279, 876–883. doi: 10.1098/rspb.2011.1351


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 de Zwaan, Wilson, Gow and Martin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Effects of Natal Dispersal and Density-Dependence on Connectivity Patterns and Population Dynamics in a Migratory Network

#### Caz M. Taylor\*

*Department of Ecology and Evolutionary Biology, Tulane University, New Orleans, LA, United States*

Migratory species can be visualized as occupying spatial networks with nodes representing regions and the populations that seasonally occupy them and links between seasonal subpopulations representing migratory connectivity. Connectivity is often regarded as a static property of a migratory network and visualized to evaluate the vulnerability of migratory populations to changes in specific regions. However, if the network itself is a dynamical system, its connectivity can be an output of the system that may be changed by perturbations to the network. I constructed a regulated, tripartite network population model with breeding, winter, and migration route nodes that also includes natal dispersal and in which connectivity goes to a dynamical equilibrium. I investigated how natal dispersal as well as the strength of density-dependent population regulation during breeding and non-breeding seasons affects connectivity patterns and the responses of the network population to simulated habitat loss. I found that when the population is primarily regulated by availability of habitat in only one season and natal dispersal was geographically constrained, connectivity patterns were unsymmetrical with weak (diffuse) connectivity from the non-regulating to regulating season and stronger connectivity in the other direction. Less-constrained natal dispersal always resulted in weak connectivity throughout. The overall magnitude of declines caused by habitat loss was determined by relative regulation and generally was not affected by natal dispersal although it was possible, with high natal dispersal, for loss of low-quality nodes in a non-regulating season to cause increases in network population size since the low-quality nodes were acting as an ecological trap. Although we expect that localness (i.e., the extent to which declines resulting from local winter habitat loss was concentrated in a small breeding area vs. spread across a larger area) should be predicted by connectivity, localness was in fact hugely variable and affected by both density-dependence and natal dispersal and generally quite difficult to predict from the connectivity pattern. In summary, the complexity of the system meant that visualization of a network by itself, without knowledge of the underlying processes causing connectivity patterns, often does not provide a good indication of the vulnerability of the network or individual node populations to habitat loss.

Keywords: migratory connectivity, natal dispersal, serial residency, density-dependence, habitat loss and degradation, migratory birds, network

Edited by:

*Yolanda E. Morbey, University of Western Ontario, Canada*

#### Reviewed by:

*James Gilroy, University of East Anglia, United Kingdom Henry Streby, University of Toledo, United States*

> \*Correspondence: *Caz M. Taylor caz@tulane.edu*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *27 March 2019* Accepted: *09 September 2019* Published: *24 September 2019*

#### Citation:

*Taylor CM (2019) Effects of Natal Dispersal and Density-Dependence on Connectivity Patterns and Population Dynamics in a Migratory Network. Front. Ecol. Evol. 7:354. doi: 10.3389/fevo.2019.00354*

# INTRODUCTION

In recent years, a great deal of research has used various tracking technologies to determine the migratory connectivity, i.e., the connections between breeding and non-breeding locations (Webster et al., 2002) of species of migratory animals, particularly birds. Migratory connectivity is referred to as strong when individuals from a single breeding location are close together during the winter and weak if they spread out over a large geographic distance and use multiple wintering locations (Webster et al., 2002). Most of the publications resulting from this work state that knowledge of migratory connectivity and/or its strength is essential for understanding declines and setting conservation priorities (e.g., Rushing et al., 2014; Trierweiler et al., 2014; Hallworth et al., 2015; El-Arabany et al., 2016; Dhanjal Adams et al., 2017; Knight et al., 2018). A few studies have found correlations between connectivity and declines which leads to the suggestion of causal relationships. For example, Hewson et al. (2016) suggests that connectivity plays a role in the declines of Common cuckoos (Cuculus canorus) in the U.K. since the proportion of birds that used one migration route correlated with the degree of population decline across nine breeding populations. In another example, Kramer et al. (2018) found that Golden-winged warblers (Vermivora chrysoptera) from declining breeding populations were spending the winter in South America, while birds from stable populations were wintering in Central America and suggest that this strong connectivity explains declines. In a review of 45 species of longdistance, terrestrial migrant landbird, Finch et al. (2017) found that connectivity is often weak and say that the conservation implications of weak connectivity are that the loss (or protection) of any non-breeding site will have a diffuse and widespread effect on many breeding populations.

The overall assumption here is that by visualizing or measuring connectivity in a population, we can make at least some general spatial recommendations for conservation. But, using connectivity information in this direct way makes an underlying (and usually unstated) assumption about the fixedness, at least over some period of time, of migratory connectivity. But we know that it is possible for connectivity to be the outcome of a migration system such that perturbations, such as habitat loss, that affect population size also change connectivity, and this raises the possibility that conservation actions will change connectivity and perhaps lead to unpredicted and unintended consequences. One approach to model dynamics of connectivity applied a maximum flow algorithm to several species of shorebirds moving through networks of non-breeding (winter and stopover sites). This model assumes that "population flow" can be modeled analogously to water flow in a pipe under the assumption is that the population will fill a network to capacity. Habitat loss (in this case due to sea level rise) caused a drop in capacity at some nodes in the network and therefore changed the overall capacity of the network and the connectivity in the network (Iwamura et al., 2013). Another approach, Migratory Flow Networks also use the analogy of water in pipes but model population flows as following general physical flux laws such that the movement between nodes depends on the attractiveness of source and destination nodes as well as the resistance to movement between them (Taylor et al., 2016). Changing the attractiveness or resistance in a migratory flow network results in altered connectivity.

Connectivity can also be determined by modeling the network as a dynamical system and solving for a steady-state equilibrium of the system. Taylor and Norris (2010) constructed a bipartite network model which had two set of nodes representing breeding and over-wintering locations where the population was regulated by density-dependent breeding success and winter survival while survival during migration declined with distance (or cost-distance) between nodes. With these assumptions, the connectivity in the migratory network goes to a stable steady state or equilibrium that is potentially altered by any change to any node or parameter. Simulated habitat loss at a single winter node showed that local winter habitat loss can cause declines even in unconnected breeding habitat regions (Taylor and Norris, 2010). This framework was used to predict the unmeasured migratory connectivity in Mexican free-tailed bats (Tadarida brasiliensis mexicana) (Wiederholt et al., 2013). The Taylor and Norris (2010) framework allows a continuum of relative strengths of density-dependence (affected by the amount of habitat/resources) in winter vs. breeding, such that the network population may be strongly winter-regulated, equally regulated by winter and breeding habitat, or strongly breeding-regulated. A version of the Taylor and Norris (2010) model was fitted to trend and tracking data for a migratory songbird, Wood thrush (Hylocichla mustelina), and found that the population was strongly winter-regulated and the pattern of declines across the breeding range could be explained primarily by habitat loss that has occurred in one winter region (Taylor and Stutchbury, 2016). Relative strength of regulation across seasons will clearly have consequences for the overall effects of habitat loss; in a strongly winter-regulated population, winter habitat loss will cause larger population declines than equivalent breeding habitat loss but how, or whether, regulation affects connectivity patterns in a network has not been explored.

One process that is certain to affect connectivity, but was not included in the modeling approaches described above, is dispersal. In migratory birds, two forms of dispersal have been described, natal dispersal is the displacement of first-time breeding birds from their natal location while breeding dispersal is the displacement of adults from their previous breeding location (Greenwood and Harvey, 1982). Cresswell (2014) put forward the hypothesis of serial residency, where offspring on their first migration move randomly (perhaps with some geographical constraints) to a non-breeding, then subsequently to a breeding location and, following those first two migrations, remain highly faithful to their selected winter and breeding locations for the rest of their lives. A great deal of evidence supports this hypothesis in migratory birds, including low juvenile natal site fidelity but high adult breeding-site fidelity to breeding as well as high adult fidelity to wintering and even staging or stopover locations. The consequence of serial residency will be to generally make connectivity patterns more diffuse (weaker), especially at small scales such that strong migratory connectivity will only be apparent at a large scale (Cresswell, 2014). How dispersal and serial residency affects the consequences of habitat loss has not been explored but, as discussed above, it is often assumed that when connectivity is weak, the loss (or protection) of any nonbreeding site will have a diffuse and widespread effect on many breeding populations (Finch et al., 2017).

Here, I explore how natal dispersal (with serial residency) and the strength of density-dependent population regulation during breeding and non-breeding seasons affect connectivity and the consequences of local winter habitat loss. I explore both the trend, which is the magnitude of the species-level breeding decline resulting from winter habitat loss, as well as the localness of those declines, which is whether the declines are spatially localized vs. widespread across the breeding range. To do this, I use a version of the Taylor and Norris (2010) equilibrium network population model with several modifications. First, in order to explicitly include the migration seasons, I extended the model from a bipartite to a tripartite network with the inclusion of migration route nodes that represent, as single nodes, generalized routes that animals take to move from their breeding to winter regions and back (Cooke, 1905). Stopover sites are incorporated within a route rather than modeled as separate nodes. Migration survival is assumed to be negatively related to distance and positively related to the quality of the migration route used. Limited numbers of available routes of differing qualities is a move toward a more realistic representation of a migration system and routes can greatly affect connectivity patterns. Second, I added the process of natal dispersal under the assumption of serial residency. Natal dispersal can be constrained by distance and is controlled by a continuous parameter that varies from 0 (no dispersal) to 1 (offspring disperse everywhere in the network). Values between 0 and 1 are distance-constrained natal dispersal. Third, I vary the relative regulation in the network by changing the average carrying-capacity (which is inversely related to strength of density dependence) in nodes in each season. In a set of networks with randomly generated node locations and parameters, I vary the level of natal dispersal and relative regulation and measure how regulation and natal dispersal affect connectivity, the overall magnitude, and the localness of declines. I discuss to what degree visualization of connectivity in a migratory network allows us to predict the consequences of habitat loss or other changes in the network.

#### QUANTIFYING CONNECTIVITY IN A MIGRATORY NETWORK

To be able to summarize and compare connectivity across networks, we need to quantify connectivity. Metrics that have been developed that summarize the strength of connectivity are based on individual-level tracking data (Ambrosini et al., 2009; Cohen et al., 2017). As described above in the terminology of migratory connectivity, strong connectivity means individuals from a given location in one season are remaining close together in another season while weak (sometimes called diffuse) connectivity means they are spreading out and the metrics used to quantify strength of connectivity reflect this definition in that high values indicate strong connectivity (Webster et al., 2002; Ambrosini et al., 2009; Cohen et al., 2017). It has been pointed out, however, that migratory connectivity actually has multiple components that are often conflated. Finch et al. (2017) define two: population spread—the geographic distance over which individuals from a single breeding population spread out in the non-breeding season and inter-population mixing—the degree to which individuals from different breeding populations mix on the non-breeding grounds. It is possible for breeding populations to have high population spread but still have low mixing or overlap among wintering regions (Finch et al., 2017).

Describing a migratory system as a network leads to a somewhat different way of thinking about connectivity. The metrics mentioned above are based on measured geographic distances from tracking data on individual animals that have generally been tracked from the breeding to the non-breeding season and not often the other way around. In a network to quantify connectivity, we need graph-theoretical metrics (Rayfield et al., 2015) and we can separately quantify the strength of connectivity from a node in any season to nodes in any other season. In a bipartite network (e.g., Taylor and Norris, 2010), with breeding and winter seasons, there are two metrics of connectivity breeding-to-winter connectivity that could be calculated for any breeding node and winter-to-breeding connectivity that could be calculated for any winter node.

Here I develop a network-based season X-to-season Y connectivity metric based on the average diversity of connections from X to Y nodes. If animals from a single node in season X migrate to a large number of Y nodes then X-to-Y connectivity is weak but if animals from nodes in X are predominantly using one or a small number of Y nodes, X-to-Y connectivity will be strong. Network connectivities do not have to be symmetrical; it is possible for X-to-Y connectivity to be weak but Y-to-X connectivity to be strong. Breeding-to-winter and winter-tobreeding connectivity are related to the concepts of population spread and inter-population mixing, respectively (Finch et al., 2017) but the former are network or graph-theoretic metrics that assume a network or graph structure in which space has been discretized into nodes whereas the latter are geographic metrics that assume contiguous ranges. Weak breeding-towinter connectivity corresponds to high population spread and weak winter-to-breeding connectivity corresponds to high interpopulation mixing. With more than two seasons, there are other possible metrics. Cohen et al. (2018) measured breedingto-winter, breeding-to-spring migration, and breeding-to-fall migration connectivity in Neotropical migratory birds. In general with k seasons, there are k! possible measures of the strength of connectivity.

Formally, following the terminology of other connectivity metrics, I define Network Migratory Connectivity NMCXY, as the strength of connectivity from season X to season Y in a migratory network. NMCXY is 1—the normalized Shannon diversity index, H2 ′ (Dormann et al., 2009) for the connections from nodes in X to Y averaged over nodes in season X,

$$\text{NMC}\_{\text{XY}} = 1 - H2'\_{\text{XY}} = 1 - \frac{1}{N\_{\text{X}}} \sum\_{\mathbf{x}=1}^{N\_{\text{X}}} \frac{H2\_{\mathbf{x}\mathbf{y}}}{\log(N\_{\text{Y}})}.\tag{1}$$

N<sup>X</sup> and N<sup>Y</sup> are the numbers of nodes in seasons X and Y, H2xY is the Shannon Diversity index for node x with respect to connections to nodes in season Y (i.e., H2xY = −P y pxylog(pxy) where pxy is the proportion of the population at node x that links to node y in season Y. NMCXY takes into account the number and evenness of connections between X and Y and is normalized by the number of nodes so that it ranges between 0 (completely diffuse or weak connectivity; all possible connections are populated evenly) to 1 (strong connectivity; one connection between each node in X and Y).

#### POPULATION MODEL DESCRIPTION

I developed a spatially structured, full annual-cycle population model of a migratory species. The annual cycle is comprised of four seasonal time-steps, Breeding, Fall migration, Wintering and Spring migration but space is modeled as a tripartite graph in which nodes, which represent spatial regions as well as the populations of animals that seasonally inhabit those regions, are classified into one of three types: breeding nodes occupied during the breeding season, winter nodes occupied during a stationary non-breeding period (referred to as "winter"), and migration route nodes which are used during both migration seasons. All nodes have a point location and an associated quality. Winter and breeding nodes also have a carrying capacity and density is defined as the population size at the node divided by carrying capacity. Fecundity is modeled as densitydependent and declines with density at the breeding node. Winter survival is also density-dependent and declines with density at the winter node. Migration route nodes do not have a carrying capacity since the model assumes no densitydependence in migration survival. Links represent movements between or through nodes. The model is an extension of the bipartite model presented in Taylor and Norris (2010). During a year, an individual moves through a cycle of 3 nodes: one breeding, one winter, and one route—for simplicity, one route node represents the geographic routes used by an individual during both migrations—and repeats the same cycle in the following year.

The "trick" to modeling population dynamics on such a network is to model the dynamics of each component of the global population that inhabits each cycle in the network. I use the notation P = (b,w,r) to denote a "cycle subpopulation" P that seasonally inhabits breeding node b, winter node w, and route node r. Nodes are not unique to cycles but are part of several cycles and so the cycle subpopulations overlap and compete with each other when there are density-dependent vital rates. Natal dispersal is modeled as displacement from one cycle subpopulation to another. In this model formulation, population dynamics, including natal dispersal, are completely deterministic. A cycle P has a length, L<sup>P</sup> , which is the migration distance and is defined as the sum of the Euclidean distance from b to r plus r to w. The size of cycle subpopulation P at the start of the breeding season in year t is NP,<sup>t</sup> and the population dynamics during the year from t to t + 1 are given by the following:

**Breeding**: During the breeding season, the number of offspring produced is given by,

$$R\_{\mathcal{P},t} = R\_{\text{max}}(Q\_b)e^{-\left(\frac{N\_{b,t}}{K\_b}\right)}\tag{2}$$

where Rmax, maximum reproductive success, is a function of the quality, Q<sup>b</sup> of the breeding node. Reproductive success declines with density ( <sup>N</sup>b,<sup>t</sup> Kb ) at the breeding node, where Nb,<sup>t</sup> is the population size at node b obtained by summing over all cycles that include node b and K<sup>b</sup> is the carrying capacity of breeding node b.

**Natal Dispersal:** Following the breeding season, offspring are redistributed uniformly among cycles according to the level of natal dispersal ND, which is a value between 0 and 1 that represents how far offspring will disperse. The distance between cycles (b1,r1,w1) and (b2,r2,w2) is defined as the sum of the Euclidean distances (w1 to w2) + (b1 to b2) + (r1 to r2). Offspring from any cycle P will be distributed evenly to all cycles that are no farther than ND ∗ Dmax from P, where Dmax is the maximum distance between any two cycles in the network. So, the proportion of offspring that moves from cycle Q to cycle P, fND(Q, P), will be 1/(number of cycles that are no farther than ND.Dmax from Q) if P is within (ND.Dmax), and 0 otherwise. If ND = 0, all offspring will stay on their natal cycle and if ND = 1, all offspring will be uniformly distributed among all cycles in the network. Natal dispersal is thus modeled here as a deterministic rather than stochastic process. ND is a dimensionless (unitless) parameter as it is expressed as relative to the maximum distance between cycles in the network.

All adults are assumed to survive through the breeding season and the distributed offspring are added into the cycle subpopulations to give the population sizes at the end of the breeding season,

$$N'\_{\mathcal{P},t} = N\_{\mathcal{P},t} + \sum\_{Q} R\_{\mathcal{Q},t} f\_{\text{ND}}(Q, \mathcal{P}) \tag{3}$$

**Fall Migration**: Migration survival is not density dependent but declines with migration distance (i.e., cycle length, L<sup>P</sup> ) at rate α<sup>r</sup> which is a decreasing function of the quality of the route node Q<sup>r</sup> . Following Fall migration at the start of winter the population size is given by

$$N\_{\mathcal{P},t}^{\prime\prime} = N\_{\mathcal{P},t}^{\prime}e^{(-\alpha\_{\mathcal{P}}(Q\_{\mathcal{I}})L\_{\mathcal{P}})} \tag{4}$$

**Winter**: Winter survival is density dependent and at the end of winter, the population size is given by

$$N\_{\mathcal{P},t}^{\prime\prime\prime} = N^{\prime\prime} \mathcal{S}\_{\max}(Q\_{\mathcal{W}}) \mathcal{e}^{-\frac{N\_{\mathcal{W},t}^{\prime\prime}}{\mathcal{K}\_{\mathcal{W}}}} \tag{5}$$

where the maximum survival, Smax,<sup>w</sup> is a is a function of the quality, Q<sup>w</sup> of the winter node. N ′′ w,t is the population size at node w obtained by summing over all cycles that include node w and K<sup>w</sup> is the carrying capacity.

**Spring Migration**: Survival during spring is assumed to be identical to fall migration survival and, following Spring migration, we have arrived back at the start of the next breeding season; the population size is given by,

$$N\_{\mathcal{P},t+1} = N\_{\mathcal{P},t}^{\prime\prime\prime} e^{(-\alpha\_{\ell}(Q\_{\ell})L\_{\mathcal{P}})} \tag{6}$$

#### Solving for Equilibrium

The model described above is a collection of deterministic processes and, for any given network configuration (including node parameter values), will go to the same fixed, equilibrium state from any starting point that has all cycles populated. This equilibrium solution of the model was obtained by starting with all cycles populated, N<sup>P</sup> > 0 for all P and simulating Equations (2)–(6) for multiple generations until the population stops changing NP,t+<sup>1</sup> − NP,<sup>t</sup> < 10−<sup>8</sup> for all P (Taylor and Norris, 2010). From the solution, the total population size, the populations size at each node, and all 6 connectivity metrics were calculated, NMCBW, NMCWB, NMCBR, NMCRB, NMCWR, and NMCRW.

#### Network Parameterization

Networks were generated inside a region defined by a unit square using the following steps:


#### Regulation

The relative regulation between breeding and winter in the network was quantified as a function of the ratio of the means of breeding K<sup>B</sup> and winter K<sup>W</sup> carrying capacity,

$$\text{regulation} = \frac{\sqrt{4\overline{K\_B}/\overline{K\_W}}}{\sqrt{4\overline{K\_B}/\overline{K\_W}} + 1} \tag{7}$$

Because carrying capacities in the two seasons affect differentlyscaled vital rates (reproductive success (0 to ∞) and survival (0 to 1), we found that when K<sup>W</sup> was 4 times the size of K<sup>B</sup> (regulation = 0.5), the population was equally regulated by winter and breeding. When K<sup>W</sup> = ∞, regulation = 0 and the population is completely regulated by breeding. When K<sup>B</sup> = ∞, regulation = 1 and the population is completely regulated by winter.

#### Model Runs

To explore the effects of natal dispersal and relative regulation on connectivity and on the consequences of winter habitat loss, K<sup>B</sup> and K<sup>W</sup> were set to 21 different values of regulation from 0 and 1 inclusive. Natal dispersal, ND, was set to one of 5 different values, which were: None: ND = 0, Low: ND = 0.05, Moderate: ND = 0.1, High: ND = 0.5, and Complete: ND = 1. For each combination of regulation and ND, 100 network configurations were generated using the steps above (resulting in 100 × 21 × 5 = 10,500 total model runs). Each network was solved for equilibrium and population sizes and connectivity metrics were calculated. One winter node was then randomly selected from all those that were occupied, i.e., population size at the node was at least 0.1% of the global population, and removed from the network. The network with the node removed was then re-solved for the new equilibrium value. The results of this perturbation were recorded as trend, the magnitude of the percentage decline (or increase) in the global breeding population size after node was removed. Also, the localness of the effect of habitat loss was measured as the diversity of individual breeding node trends measured as a normalized Shannon's index, the diversity among the proportions of the total population change that occurred at individual nodes. When localness is high, this means there is high variation in the effects of habitat loss, i.e., there are big changes in population size at a small number of nodes and no change or small changes at others. When localness is low, changes are more evenly distributed across the network.

#### RESULTS

#### Connectivity Patterns

Density-dependence and natal dispersal interacted to affect connectivity patterns. When density-dependent regulation was skewed toward one season, this caused an asymmetry in connectivity but only when there was no or low natal dispersal. With low natal dispersal in a strongly winter-regulated network, several breeding nodes were unoccupied but all winter nodes were occupied which led to strong winter-tobreeding (and migration-to-breeding) connectivity and relatively weak breeding-to-winter (and migration-to-winter) connectivity (**Figures 1A, 2**). The asymmetry in connectivity lessened as the regulation because less winter-skewed and with equal regulation all breeding and winter habitat was occupied and winter-tobreeding and breeding-to-winter connectivity were the same strength (**Figures 1D**, **2**). In a network that was breedingregulated (still with no or low natal dispersal), the situation was reversed and there was unoccupied winter habitat, strong breeding-to-winter (and migration-to-winter) connectivity and

proportional to the size of the node population to show the distribution of the population within each season.

relatively weak winter-to-breeding (and migration-to-breeding) connectivity (**Figures 1G**, **2**).

The spatial configuration of nodes, which was randomly generated, caused connectivity strengths and occupancy of nodes to be highly variable among networks with the same levels of ND and regulation but again only when there was low or no natal dispersal (**Figure 2**). Unoccupied nodes, which only occurred in networks with no or low natal dispersal and in seasons where regulation was relatively weak, tended to be lower quality than occupied nodes but occupancy also depended on the location of the node within the network. Since there is no regulation during migration season, occupancy and population size at route nodes were related to route node quality although location and distance from other routes also caused variation in route usage.

(w.to.b; *NMCWB*; Equation 1) connectivity in networks where *regulation* (Equation 7) varies from 0 to 1 at four levels of natal dispersal, *ND*. Points show the results of model runs and trendlines are fitted locally weighted polynomial (loess) curves.

High quality routes that were distant from others had high usage but nodes close to slightly higher quality routes had low usage (**Figure 1**).

Any amount of natal dispersal weakened connectivity overall in the network (**Figures 1**, **2**) and narrowed the gap between winter-to-breeding and breeding-to-winter connectivity (**Figure 2**). Once natal dispersal became moderate or high, all connectivities were zero or close to zero for all values of regulation (**Figures 1C,F,I**, **2**). High or moderate natal dispersal reduced the variation in the strengths of connectivities and also caused all nodes to be occupied irrespective of spatial location, quality, or season.

# Effects of Winter Habitat Loss: Size of Declines

Winter habitat loss at a local scale, modeled as the removal of an occupied winter node, unsurprisingly led to a small population decline in a breeding-regulated network and large decline in winter-regulated network. The size of the decline was controlled almost entirely by the relative regulation of the network population and was largely unaffected by natal dispersal. There was one exception: in a strongly breedingregulated network with no natal dispersal, winter habitat loss would cause small global declines but with high natal dispersal loss of winter habitat, in some cases, actually caused a small increase in overall population (**Figure 3**).

(loess) curves.

#### Effects of Winter Habitat Loss: Localness of Declines

There was a large amount of variation in the distribution of declines, so much so that it is difficult to generalize results (**Figure 3**). When natal dispersal was high, declines caused by local winter habitat loss were always distributed throughout the breeding range (localness was low) as would be expected by the weak connectivity. With low natal dispersal, declines tended on average to be more localized when regulation was skewed toward one season and less localized when the network was equally regulated. The most localized declines were seen when there was no natal dispersal and strong winter-regulation. The most diffuse declines were either when natal dispersal was relatively unconstrained or when regulation was equal between breeding and winter. **Figure 4** shows that although there are some cases where the network showed strong breeding-to-winter connectivity and localized declines due to habitat loss and other cases with weak connectivity and diffuse effects of habitat loss, there are also networks with strong connectivity where declines resulting from habitat loss were diffuse and localness was high.

# DISCUSSION

Both population regulation and natal dispersal affected connectivity in a migratory network and so the question arises as to what degree we can infer regulation or natal dispersal levels from connectivity patterns. When a network population was primarily regulated by one season, this caused an asymmetry in winter-to-breeding compared to breeding-to-winter connectivity (measured as the diversity of connection strengths) due to unoccupied habitat in seasons where regulation was relatively weak but natal dispersal wiped out this difference and generally made all connectivities weak. Rappole and McDonald (1994) suggested that observations of unoccupied habitat in one season could be used as indication of skewed regulation toward the opposite season so perhaps an asymmetry in connectivity could be used in the same way? However, studies that measure connectivity in both directions are rare since they involve tracking individuals from both breeding and winter locations. Stanley et al. (2015) did this for Wood thrushes and showed weak breeding-to-winter connectivity (i.e., non-breeding ranges from any breeding site were large and overlapping) and slightly stronger winter-to-breeding connectivity (breeding ranges from a given winter site were also large but there was less overlap among them), suggesting equal to winter regulation in this species. Two studies showed that Ovenbirds (Seiurus aurocapilla) exhibit weak breeding-to-wintering connectivity from individual breeding location (individuals tracked from 4 breeding locations had non-overlapping but large spread in non-breeding ranges) and similarly weak winter-to-breeding connectivity from three winter locations, suggesting equal, perhaps slightly skewed toward breeding, regulation in this species (Hallworth and Marra, 2015; Hallworth et al., 2015).

of localness.

Such inferences must be made with great caution, however, since even quite constrained natal dispersal tends to equalize seasonto-season connectivities (**Figure 2**) and since the connectivity metrics depend on the geographic scales at which nodes are defined.

The overall pattern of connectivity could be an indicator of the level of natal dispersal as suggested by Cresswell (2014). One pattern that emerges from multiple avian tracking studies that we often see is strong connectivity in the network when measured at a large scale or coarse resolution (i.e., if we cluster nodes together within geographic regions) but weak connectivity at a small scale. This pattern, which is observed, for example, in Common nightingales (Luscinia megarhynchos) (Hahn et al., 2013), Ovenbirds (Hallworth and Marra, 2015; Hallworth et al., 2015), and Tree swallows (Tachycineta bicolor) (Knight et al., 2018), suggests that natal dispersal in these species is constrained since their connectivity resembles the networks in panels A, D, G, B, E, and H rather than panels C, F, or I of **Figure 1**.

The magnitude of declines in the species caused by local habitat loss was, as expected, strongly determined by the relative strength of density-dependence (Sherry and Holmes, 1996; Sutherland, 1996). Local winter habitat loss in a winter-regulated network population caused a large decline while the same loss caused only a small to zero decline in a breeding-regulated network population. Natal dispersal generally had no effect on the magnitude of declines with one surprising exception: in a strongly breeding-regulated network with some degree of natal dispersal, the removal of a winter node could cause the overall population size to increase. This occurred when the habitat lost was low quality and its removal causes a shift to higher quality habitat which supported a higher equilibrium cycle subpopulation size. In this case, natal dispersal could be said to be buffering the population from effects of habitat loss (Cresswell, 2014).

It is thought that if migratory animals show weak migratory connectivity then loss of habitat or other unfavorable localized events in the non-breeding season will have a diffuse effect on the size of the global breeding population and thus a small effect on any single breeding population, whereas breeding populations with strong connectivity will be more vulnerable to localized unfavorable events in the non-breeding area (Finch et al., 2017). I tested this by measuring "localness," the diversity and evenness of declines across breeding nodes, following local winter habitat loss. While there were some general patterns, I found that localness was highly variable and difficult to predict. Natal dispersal did reduce localness (making breeding decline more diffuse) as well as making connectivity weak and thus if comparing two networks in which the connectivity difference was due to differences in natal dispersal, the relative localness of declines could be correctly inferred from the connectivity patterns. However, regulation had a non-linear effect on localness as well as affecting connectivity. With low or no natal dispersal, localness was lowest when the regulation was equal between winter and breeding and higher when skewed toward either season while breeding-to-winter connectivity decreased as winter-regulation increased. This means that if comparing two networks where the connectivity differences were due to different

regulation, localness of declines would not be predictable from connectivity (**Figures 3**, **4**). Even knowing the connectivity in a migratory network, we cannot be sure whether local habitat loss in a population will result in small or large declines or whether the declines will be localized vs. widespread unless we are also sure of the underlying processes that generate specific connectivity patterns.

The model developed here advances that presented in Taylor and Norris (2010) by demonstrating how to include natal dispersal, which has generally been overlooked in migration (Cresswell, 2014). As expected, natal dispersal had a large, easilypredictable effect on connectivity, a less-predictable but still large effect on the localness of declines resulting from habitat loss, and little-to-no effect on the magnitude of declines. Natal dispersal is modeled here as a deterministic (rather than stochastic) process in which offspring are uniformly distributed to all cycles that are within a distance threshold. The assumed dispersal kernel (the proportion or probability that an animal disperses a particular distance) is a step function with equal dispersal to all cycles closer than the threshold and no dispersal to cycles past the threshold. It would be straightforward to use a differently-shaped dispersal kernel, for example a function that declines exponentially with distance, and we might expect that this would affect the resulting connectivity patterns. However, since we know so little about natal dispersal in migratory animals, there does not appear to be a basis for assuming a different kernel shape. Natal dispersal and connectivity strength in this model are expressed as dimensionless metrics and are relative to the breeding and wintering range areas of the species. Across multiple species, high natal dispersal (ND ≈ 1.0) might be expected for a species with small breeding and winter ranges and low natal dispersal for wide-ranging species, although this pattern has not, to my knowledge, been demonstrated. The network model predicts weaker connectivity within its known range in a species with higher natal dispersal relative to its range. If it can be shown that species with smaller ranges do typically have higher relative natal dispersal, then the model predicts that, within their ranges, connectivity will typically be weaker in those species.

A second advance on Taylor and Norris (2010) is the extension to explicitly consider migration seasons, often believed to be the most critical seasons for migratory species although direct evidence for this is still rare (Sillett and Holmes, 2002; Newton, 2006; Klaassen et al., 2013). Other migratory network models have taken the approach of including stopover locations (places where animals stop to re-fuel during migration) as separate nodes in a migratory network (e.g., Iwamura et al., 2013; Knight et al., 2018). While this has the advantage of being able to examine the importance of individual stopover locations, and therefore is the appropriate approach for many questions, it complicates population modeling of these networks because individuals within the same migration route may stop at a variable number of locations. Instead, here I adopt the concept of migration routes, included in the model as single nodes. This idea follows from a long history of bird migration studies that describe routes of individual species and flyways used by multiple species (Cooke, 1905; Lincoln, 1935). Although the concept of flyways is a simplification, it has proved highly useful in regulation and conservation planning for migratory birds, including the management of migratory waterfowl and shorebirds by the U.S. Fish & Wildlife Service and its partners (https://www.fws.gov/birds/management/bird-management-

plans.php) and conservation work by non-profit organizations (e.g., https://www.audubon.org/birds/flyways). Furthermore, several recent studies that track migratory birds from multiple sites have identified disjunct migration routes (e.g., Stanley et al., 2015; Brown et al., 2017; Hahn et al., 2019). The idea of migration routes, or corridors, is also important in describing the migration and planning conservation of non-avian species including elephants (Roever et al., 2013), sea turtles (Morreale et al., 1996), and ungulates (Berger and Cain, 2014; Coe et al., 2015). The simplification of route nodes allows a more general, compact, and tractable formulation of the population model as a tripartite migratory network. This model could be used to explore breeding-to-migration or winter-to-migration connectivity as well as the effects of loss or degradation of migration routes on population size. Preliminary investigations along these lines indicate that geography matters a great deal for predicting the effects of loss of migration routes. Loss of a route that is near to another of reasonably good quality (whether occupied or not) will have a small effect whereas loss of isolated routes can have large impacts. A very important assumption in this model is that the population depends on distance but is not regulated by the migration season, i.e., survival during migration depends on the quality of the route but is independent of the number or density of individuals occupying a route node. Inclusion of density-dependence in migration survival would likely have large effects on all the results presented here.

The network could easily be extended beyond tripartite to multipartite to include more seasons to represent the spring and fall migration separately or perhaps to divide the winter season to accommodate winter movements, etc. While resembling metapopulations in many ways, multipartite migratory networks function somewhat differently in that a subpopulation rather than occupying a single node is more properly thought of as occupying a set of spatial nodes (one for each season) and the links that connect them. Using graph-theory terminology, this would be termed a path or, in this case, a cycle, since the path connects back to the first node (for a general habitat network, this has also been termed a "pathway"; Wiederholt et al., 2017). I use the term cycle and the model describes population dynamics on each cycle. I modeled natal dispersal as flux between cycles. The results described above can be more deeply understood by realizing that, when not at equilibrium, some cycles are sources and some sinks. With no natal dispersal, at equilibrium, sink cycles become extinct and some nodes in a season where regulation is relatively weak are unoccupied. But with natal

#### REFERENCES


dispersal, there is continual dispersal into sink cycles so that, at equilibrium, most habitat in both seasons (all if natal dispersal is very high) remains occupied and connectivity is weakened.

Other possible future extensions to the model could introduce other processes that might affect connectivity or population dynamics including seasonal interactions (e.g., when quality of winter habitat affects breeding success (Norris and Taylor, 2006; Harrison et al., 2010), orientation processes, i.e., how animals orient themselves or choose routes (Thorup and Rabøl, 2001; Willemoes et al., 2014), and differential migration by sex (Briedis and Bauer, 2018), which might require coupled but separate networks for males and females. The ecological model presented here could be adapted to explore evolutionary dynamics, e.g., evolution of natal dispersal. One limitation of the multipartite network approach is that it does not consider within-year timing of processes and therefore is not perhaps the best approach to use to explore consequences of changes in phenology and mismatch. There are other network approaches, specifically migratory flow networks that can be used to explore this (Taylor et al., 2016).

In summary, the tripartite migratory network model presented here shows that the connectivity pattern in a migratory network is altered by both regulation and natal dispersal, especially the latter. Since regulation is the primary determinant of the effects of habitat loss and natal dispersal has the largest effect on connectivity, visualization of connectivity patterns is not a reliable predictor of whether declines caused by local habitat loss will be localized vs. diffuse and connectivity alone provides almost no information about the expected magnitude of declines following habitat loss. This does not mean that visualization of connectivity is not important—of course it is—but it is only the first step and understanding the processes that lead to different connectivity patterns is vital to understand population trends in migratory species.

#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the manuscript/supplementary files.

# AUTHOR CONTRIBUTIONS

The author confirms being the sole contributor of this work and has approved it for publication.

#### FUNDING

This research was funded by a Scholar award from the James S. McDonnell Foundation.


connectivity. Methods Ecol. Evol. 9, 513–524. doi: 10.1111/2041-210X. 12916


**Conflict of Interest:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Taylor. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Migration Distance and Body Condition Influence Shorebird Migration Strategies and Stopover Decisions During Southbound Migration

Alexandra M. Anderson<sup>1</sup> \*, Sjoerd Duijns <sup>2</sup> , Paul A. Smith<sup>2</sup> , Christian Friis <sup>3</sup> and Erica Nol <sup>4</sup>

*<sup>1</sup> Environmental and Life Sciences Graduate Program, Trent University, Peterborough, ON, Canada, <sup>2</sup> National Wildlife Research Centre, Environment and Climate Change Canada, Ottawa, ON, Canada, <sup>3</sup> Canadian Wildlife Service, Environment and Climate Change Canada, Toronto, ON, Canada, <sup>4</sup> Department of Biology, Trent University, Peterborough, ON, Canada*

Technological constraints have limited our ability to compare and determine the proximate and ultimate drivers of migratory behavior in small-bodied birds. Small VHF transmitters (<1.0 g) paired with automated radio telemetry allowed us to track the movements of six small shorebird species and test hypotheses about migratory behavior in species with different migration distances. We predicted that during southbound migration, species with longer migration distances (>9,000 km; pectoral sandpiper, *Calidris melanotos*, and white-rumped sandpiper, *Calidris fuscicollis*) would be more likely to migrate with characteristics of a time-minimizing migration strategy compared to species migrating intermediate distances (5,000–7,500 km; semipalmated sandpiper, *Calidris pusilla*; and lesser yellowlegs, *Tringa flavipes*) or shorter distances (∼5,000 km; least sandpiper, *Calidris minutilla*; semipalmated plover, *Charadrius semipalmatus*), which would migrate with more characteristics of an energy-minimizing strategy. Our results indicate that migration and stopover behaviors for adults matched this prediction; longer distance migrants had longer stopover lengths, departed with higher relative fuel loads, flew with faster ground and airspeeds, and had a lower probability of stopover in North America after departing the subarctic. The predicted relationship between migration distance and migratory strategy was not as clear for juveniles. Despite our prediction that longer distance migrants would be less wind selective at departure and fly into headwinds *en route*, all species and age classes departed and migrated with supportive winds. Birds with higher estimated fuel loads at departure were less likely to stop in North America after departing the subarctic, indicating that some birds attempted non-stop flights from the subarctic to the Caribbean or South America. Additionally, within species, adults with higher relative fuel loads at departure had a higher detection probability after departing the subarctic, which we interpret as evidence of higher survival compared to juveniles. This study shows that migratory behavior of shorebirds has predictable patterns based on migration distance that are moderated by body condition of individuals, with potential implications for fitness.

Keywords: automated telemetry, body condition, carryover effects, flight speed, migration distance, optimal migration, stopover

#### *Edited by:*

*Yolanda E. Morbey, University of Western Ontario, Canada*

#### *Reviewed by:*

*Phil Taylor, Acadia University, Canada Simeon Lisovski, Swiss Ornithological Institute, Switzerland*

*\*Correspondence: Alexandra M. Anderson aande763@gmail.com*

#### *Specialty section:*

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> *Received: 24 March 2019 Accepted: 17 June 2019 Published: 09 July 2019*

#### *Citation:*

*Anderson AM, Duijns S, Smith PA, Friis C and Nol E (2019) Migration Distance and Body Condition Influence Shorebird Migration Strategies and Stopover Decisions During Southbound Migration. Front. Ecol. Evol. 7:251. doi: 10.3389/fevo.2019.00251*

# INTRODUCTION

Many animals migrate to exploit spatial and temporal increases in prey abundance (Alerstam et al., 2003; Teitelbaum et al., 2015) while also reducing predation risk (Hebblewhite and Merrill, 2007; McKinnon et al., 2010). Despite the benefits of migration, mortality can be high during this life stage (Sillett and Holmes, 2002; Calvert et al., 2009; Piersma et al., 2016) because animals encounter variable environments, habitat limitation, inclement weather, and other risks (Klaassen et al., 2012) throughout their migratory range.

Migration distances vary across species, and this variation may influence migratory behaviors and strategies. Long-distance migrants must rely on local conditions, circannual clocks, and photoperiod to make migratory decisions (Gwinner, 1996) about far-away destinations. This may, in part, explain more consistent and less plastic timing of migration for long-distance migrants compared to short-distance migrants (Rubolini et al., 2007; Miller-Rushing et al., 2008). Long-distance migrants must balance high energetic demands of migration with predation risk and time constraints to complete farther migrations (Alerstam et al., 2003). Optimal migration theory provides clear predictions about migratory strategies for individuals under different energy, time, and predation constraints (Alerstam and Lindström, 1990; Hedenström and Alerstam, 1997; Alerstam, 2011), but the theory is less clear about how total migration distance influences migratory behavior and the currency individuals use to maximize fitness (i.e., time, energy, and predation). Few empirical studies have investigated the effects of migration distance on migratory behavior. A recent study found support for the hypothesis that long-distance migrants are more time-constrained than short-distance migrants because of farther travel distance (Nilsson et al., 2014).

Optimal bird migration theory predicts the consequences for stopover ecology of different migration strategies. The energyminimization hypothesis predicts that migrants minimize the total energy cost of migration, whereas the time-minimization hypothesis predicts that animals migrate to reduce total migration time (which is more costly energetically). Timeminimizers are predicted to depart stopover sites with higher fuel loads despite the high energetic costs of carrying more weight (Pennycuick, 1969, 1975; Hedenström and Alerstam, 1997). To avoid delays, they may depart stopover sites with less wind assistance than energy minimizing migrants (Nilsson et al., 2014; McCabe et al., 2018). They also are predicted to be more goal oriented during migration and should have a higher propensity to fly into headwinds toward their destination (Alerstam, 1979; Liechti, 1995). Lastly, they should migrate at higher airspeeds (Hedenström and Alerstam, 1997; Nilsson et al., 2014) and make fewer stops en route compared to energy-minimizing migrants (Hedenström and Alerstam, 1997; Alerstam, 2011).

Larger relative fuel loads in the form of fat (Ramenofsky, 1990; McWilliams et al., 2004) can increase flight range and reduce the number of stops necessary en route (Hedenström and Alerstam, 1997). This is favorable for longer distance migrants, because it reduces energy and time costs associated with search and settling at each stopover site (Alerstam, 2011), such as rebuilding and subsequently catabolizing digestive tracts and other organs (Piersma and Lindström, 1997).

In this study, we use automated radio telemetry to compare southbound migration strategies of six shorebird species with variable migration distances (**Table 1**) from a key subarctic stopover site in North America. We examine the relationship between migration distance and stopover length, departure fuel loads, wind selectivity, ground speeds and airspeeds, and subsequent stopover probability and determine if these patterns match previously observed patterns of time-minimizing migration in longer-distance migrants (e.g., Nilsson et al., 2014). More specifically, we predict that species with longer migrations (white-rumped sandpiper, Calidris fuscicollis, and pectoral sandpiper, Calidris melanotos; ∼9,000–11,000 km from the subarctic) will exhibit migratory behaviors more consistent with a time-minimizing migration strategy (i.e., higher fuel loads at departure, less wind selectivity and tailwind support en route, faster ground speeds and airspeeds, and lower probability of subsequent stopover after departing the subarctic). By comparison, we predict that species with intermediate migration distances (semipalmated sandpiper, Calidris pusilla and lesser yellowlegs, Tringa flavipes; ∼5,000–7,500 km from the subarctic) or shorter migration distances (least sandpiper, Calidris minutilla and semipalmated plover, Charadrius semipalmatus; ∼5,000 km from the subarctic) will show more characteristics of an energyminimizing strategy.

We examine these patterns as a function of age class (adult or juvenile) because juvenile shorebirds tend to have shorter, rounder (Fernández and Lank, 2007), and more convex (Anderson et al., 2019) wings than adults, a shape that is less efficient for long migratory flights (Rayner, 1988; Lockwood et al., 1998). Because of these differences in wing shape, juvenile shorebirds may need to take more stops en route than adults. Alternatively, the migration behavior of juveniles may show less clear patterns than adults because they have no previous migration experience. Lastly, body condition is known to influence migratory behavior and outcomes (e.g., Duijns et al., 2017), so we explore how it influences migration strategies and determine if it affects detection probabilities, and hence potentially survival, of individuals outside of James Bay.

# MATERIALS AND METHODS

# Banding and Relative Fuel Loads

Shorebirds were captured with mist nets at four remote field camps in 2014–2018, from mid-July through mid-September each year along the southwestern coast of James Bay, Ontario, Canada (**Figure 1**). The sampling period corresponded with the bulk of southbound migration for shorebirds at James Bay, except least sandpiper adults and white-rumped sandpiper juveniles, which we excluded from the study. We banded birds and recorded mass (± 0.1 g), maximum flattened wing length (± 1 mm) (Gratto-Trevor, 2004), and subcutaneous fat score [0–7 scale, (Meissner, 2009)]. Birds were aged as juvenile (hatched that year) and adult (> 1 year of age) by examining the color and shape of the median wing coverts (Gratto-Trevor, 2004). In 2014–2017, we attached



*Wintering areas are summarized from the literature as well as from* \**historical band recoveries (banded 1974–1982) and* \*\**present-day band recoveries (Patuxent Wildlife Research Center Bird Banding Laboratory), flag resightings (www.bandedbirds.org), and* <sup>+</sup>*present-day nanotag telemetry detections. Recoveries, flag resightings, and telemetry locations may include detections from late fall migration routes and southern stopover sites. Lean mass was calculated in Supplementary Methods 1.1.*

digitally coded VHF nanotags (Lotek Wireless, Newmarket, Ontario, Canada; **Supplementary Table 1**) to skin on the lower back of each bird above the uropygial gland (Warnock and Warnock, 1993) using cyanoacrylate glue (Loctite <sup>R</sup> Super Glue ControlTM UltraGelTM).

We collected blood from the brachial vein of most birds for molecular sexing because many shorebirds cannot be sexed by morphometrics or plumage (Baker et al., 1999; Dos Remedios et al., 2010). We used 27-gauge needles and capillary tubes to collect samples, and samples did not exceed 1% of body mass. Samples were stored on ice and in 95% ethanol prior to DNA extraction. DNA was extracted and amplified using primers and molecular methods designed for shorebirds (van der Velde et al., 2017). Capture, banding, and blood sampling were approved by Trent University and Environment and Climate Change Canada's Animal Care Committees and carried out under permit from Environment and Climate Change Canada.

We calculated relative fuel loads (f; ratio of fat mass to lean mass) at capture and departure by subtracting lean mass (m0) from capture mass (mcap) or estimated departure mass (mdep) and dividing by lean mass f = (mcap or mdep − m0)/m<sup>0</sup> (Delingat et al., 2008). We calculated m<sup>0</sup> (fat score zero) for each bird from regression equations (**Supplementary Methods 1.1**) of mass predicted by fat score, wing length, species, and interactions between species and fat score and species and wing length. Mass at departure was calculated as mdep = mcap + mchange ∗ L where mchange is the species and age specific rate of daily mass change (g/day) at the population level, and L is the individual's minimum

FIGURE 1 | Locations of Motus Wildlife Tracking System automated VHF radio telemetry receiver stations used to track shorebirds during stopover along the southwestern coast of James Bay, Ontario, Canada, and during southbound migration. Towers mapped were active in at least 1 year between 2014 and 2017. Darker blue dots indicate multiple towers in close proximity. One tower at Asunción Bay, Paraguay, is not shown in the figure.

length of stay (days) in James Bay determined from nanotags (see Length of Stay below). We determined mchange using linear mixed effects models for each age group with mass as the response variable and species, capture day of year, and an interaction between capture day of year and species as predictor variables (**Supplementary Methods 1.2**). Wing length was included as a covariate and year as a random factor. We used population rates of mass change to estimate departure masses because it was not possible to recapture individuals. Population rates of mass change were low (**Supplementary Methods 1.2**) and may underestimate individual rates of mass gain as a result of the arrival of thin birds or departure of fat birds. Although they obscure individual differences in refueling rates, population level rates allow for conservative estimates of mass change in individuals with long length of stay (several weeks) and little mass change in individuals with short length of stay (days). We compared relative departure fuel loads using linear mixed effects models with species as a predictor variable. We only included birds that we could confirm departed from James Bay (**Supplementary Methods 1.3**).

#### Automated Radio Telemetry

We used automated radio telemetry paired with VHF nanotags to obtain high temporal resolution estimates of length of stay, departure decisions, and flight speeds. Nanotags were the best option for this study on small shorebirds because they are lightweight (<1.0 g) and provide data without requiring recapture, which is difficult at this study site. Nanotags operate on a single frequency (166.380 MHz) and transmit unique, identifiable bursts every 4.7 to 10.1 s for ∼80–160 d depending on battery size and burst rate (**Supplementary Table 1**). Nanotags were monitored through the Motus Wildlife Tracking System, a network of > 325 automated radio tower receivers (Taylor et al., 2017). Nanotags were automatically recorded by SRX receivers (Lotek Wireless, Newmarket, Ontario, Canada) or Sensorgnome receivers (www.sensorgnome.org) when a tagged bird was within range of tower antennas (∼50 km; **Supplementary Table 2**). Birds were detected in James Bay by 5 to 8 towers (henceforth the "local array") and at towers south of James Bay (the "southern array"; **Figure 1**). In 2016 and 2017, tags also were detected in James Bay with an SRX800 receiver and a 3-element Yagi antenna mounted to the base of a helicopter.

We removed detections with < 3 consecutive bursts at intervals of a tag's burst rate (Brown and Taylor, 2017; Duijns et al., 2017), which removed most false detections; however, some towers were prone to noise, which resulted in systematic false detections of tags (e.g., detections of multiple birds at the same tower and time hundreds of kilometers away from their last known location). These false detection patterns were identified by examining plots of detections for each bird by latitude and time and longitude and time and were subsequently removed.

#### Length of Stay and Migratory Departure

We estimated minimum length of stay from the capture time and the last detection of the individual in the local array. For birds captured in 2014 and 8 birds captured in other years of the study, capture times were not recorded, so we set the capture time to 12:00 p.m. on the day of capture (resulting in a maximum error of 12 h). We compared length of stay using linear mixed effects models with species, relative fuel load at capture, capture day of year, and all interactions as predictors. Only birds for which we could confidently identify departure detection patterns from nanotags (**Supplementary Methods 1.3**) were included in this analysis because length of stay could be biased shorter by undetected mortality or by birds traveling outside of the detection zone of the local array during the stopover period. We considered the last detection of an individual in the local array to be the time of migratory departure, and we evaluated weather conditions at departure for birds with confirmed departure (**Supplementary Methods 1.3**).

Wind and precipitation data at departure were obtained from a weather station attached to the Piskwamish tower (**Figure 1**) in 2015–2017. The Piskwamish weather station was not erected in 2014, so we used weather data from a nearby weather station in Moosonee. The data were comparable between stations, except for wind speeds, which we calibrated to ensure similar estimates (**Supplementary Methods 1.4**). The weather stations had different temporal resolutions (Moosonee: hourly point observations; Piskwamish: 2 h averages), so we selected wind and precipitation data from the hour closest to departure (2014) or from the 2 h time-period in which the bird departed. For each bird, we compared wind profit (see below) and precipitation at the time of departure with the same weather variables 48 h prior. We chose 48 h as a comparison because wind speeds are correlated for up to 32 h in this region (**Supplementary Methods 1.4**), and we aimed to sample wind at the same time of day to avoid confounding results with daily temporal patterns in wind speed.

We estimated wind profit, wind support toward a migratory goal, at departure and 48 h prior following Erni et al. (2002) where wind profit = D − p D<sup>2</sup> + W<sup>2</sup> − 2DWcos(α) and D = airspeed (m/s), W = wind speed (m/s), and α = wind direction (degrees)–orientation direction (degrees) converted to radians. Positive wind profit values indicate wind assistance whereas negative values indicate wind hindrance. We assumed birds would fly at an airspeed of 16 m/s (Alerstam et al., 2007; Grönroos et al., 2012; the mean airspeed of shorebirds detected by radar). For each species and age class, we used the median bearing of the first migratory flight from James Bay to the southern array (**Supplementary Figure 1**) as the migratory goal. We compared departure probability using generalized linear mixed effects models with a binomial response variable (departed yes/no, where "yes" represented predictors at the time of departure and "no" represented predictors 48 h prior). We included fixed predictors of wind profit, precipitation (yes/no), departure day of year, relative fuel load at departure, and species. We also included pairwise interactions between wind profit and all other predictors.

# Migration Tracks and Flight Speeds

We partitioned migration data into "tracks": the great circle trajectories between sequential tower detections for each bird. Partitioning flights into tracks allowed for multiple estimates of flight speed and wind assistance during a single migratory flight for some individuals. We calculated ground speeds (speed of the bird relative to the ground) for each track as the time elapsed between tower detections divided by the track distance. We considered ground speeds between 9 and 42 m/s to be typical of shorebird migratory flight (Grönroos et al., 2012; **Supplementary Figure 2**) and excluded tracks with ground speeds outside of these ranges. Ground speeds <9 m/s may indicate undetected stops en route or a longer flight path than the great circle trajectory. Speeds > 42 m/s were typical of detections on nearby towers (typically <140 km; **Supplementary Table 2**) and represent a small proportion (<10%) of the southbound migratory distance traveled by these birds from the subarctic to the southern array.

We compared ground speeds using generalized linear mixed effects models. Species, relative departure fuel load at departure, departure day of year, and interactions between fuel load and species and species and departure day were included as predictors. We considered a quadratic effect of tailwind support for the track (see below) as a covariate in the models because the relationship between tailwind and ground speed was non-linear.

We used the NCEP.flight function and tailwind equation in the RNCEP package (Kemp et al., 2012b) in R (R Core Team, 2018) to estimate wind assistance along the great circle trajectory for each track assuming 16 m/s airspeed (Alerstam et al., 2007; Grönroos et al., 2012). We extracted wind data from the NCEP/NCAR Reanalysis I dataset (https://www.esrl.noaa. gov/psd/data/gridded/data.ncep.reanalysis.html), which has a 2.5◦ latitude by 2.5◦ longitude spatial resolution and 6 h temporal resolution (Kemp et al., 2012a). Wind conditions were interpolated in space and time every 3 h along the route using inverse weighting distance. We considered wind conditions at pressure levels of 1,000, 925, 850, and 700 hPa (corresponding to ∼100, 700, 1,500, and 3,000 m above sea level respectively) as candidate flight altitudes because radar studies have recorded shorebirds migrating within this range (Grönroos et al., 2012; La Sorte et al., 2015b). We assumed birds would migrate at the altitude with the highest wind assistance at each interpolated point. We compared tailwind support with linear mixed effects models containing species, relative fuel load at departure, departure day of year, and interactions between species and fuel load and species and departure day as predictors.

We estimated airspeeds for each track by subtracting the average tailwind support (m/s) for the track from its ground speed (m/s). We compared airspeeds with linear mixed effects models containing species, relative fuel load at departure, departure day of year, and interactions between species and fuel load and species and departure day as predictors.

#### Stopover and Detection Probability

We classified individuals to two categories: made at least one stop or did not stop in North America after departing James Bay. Category assignments were made using a combination of back-and-forth transmitter detection patterns on nearby towers and/or slow speed thresholds between sequential tower detections (**Supplementary Methods 1.5**). We examined stopover probability with generalized linear mixed effects models. The response variable for each model was binomial (the bird stopped or did not stop) and contained predictors of species, relative fuel load at departure (or relative fuel load at time of last detection at James Bay for individuals that did not have clear departure detection signals), and an interaction between species and fuel load. We did not include departure day as a predictor in this model because we detected stopovers for some birds with unknown departure.

Individuals that were not detected in the southern array either died, took a route not monitored by receivers, or nanotags malfunctioned or were lost. In this study, we cautiously interpret detection probability as a metric of apparent survival. Detection probability will underestimate true survival, because of tag loss or migration through areas without receivers (such as central North America and Newfoundland, Canada). It should, however, show reliable patterns for the effect of relative fuel load at departure on probability of detection unless skinny or fat birds migrate using

TABLE 2 | Sample sizes of shorebirds tagged with VHF radio transmitters in James Bay, Ontario, Canada in 2014–2017 and detected on departure and in the southern automated radio telemetry with the Motus Wildlife Tracking System.


*Some birds were never detected, primarily a result of tag activation errors in 2014. Lower sample sizes for confirmed departure mostly were the result of radio tower malfunction at Northbluff Point at periods in 2015 and 2016.*

different routes. For birds with confirmed departure from the subarctic, we compared detection probability using generalized linear mixed effects models with a binomial response variable (detected in the southern array or not) and predictors of species, departure day of year, and relative fuel load at departure. Each model included an interaction between species and departure day of year and species and fuel load. For adults, all semipalmated plovers with confirmed departure from James Bay were detected in the southern array; therefore, we did not include this species in this analysis.

#### Statistical Analyses

Age classes were analyzed separately for all models because not all age classes were sampled for each species. All mixed models included sex and year as random factors. The models predicting tailwind, ground speed, and air speed also included bird ID as a random factor to account for multiple migration tracks per individual. We used the "drop1" function with a Wald chisquare test (in a backwards stepwise approach) to remove model parameters from each global model that were not significant (α = 0.05). Model predicted means and slopes were calculated and compared with Tukey HSD post-hoc tests using the emmeans package (Lenth, 2019). Analyses were conducted with program R version 3.5.1 (R Core Team, 2018), and we made figures with ggplot2 (Wickham, 2016) and sjPlot (Lüdecke, 2018).

#### RESULTS

We tagged 335 adult and 275 juvenile shorebirds with nanotags at James Bay (**Table 2**). A subset of those tags (19%) was never detected (**Table 2**) because of tag activation errors in 2014 and malfunctions of the Northbluff Point tower in 2015 and 2016. Of birds that were detected at least once, 62% were detected in the southern array where they were detected for, on average, 5 ± 9 days. Most detections in the southern array occurred in North America, and none of these birds were detected south of 35.7◦ N in North America (northern North Carolina, USA) despite the presence of towers south of this latitude (**Figure 1**). Migration routes were variable across species and age classes (**Supplementary Figure 3**).

#### Length of Stay, Relative Fuel Loads, and Departure

Minimum length of stay in James Bay was explained by species, relative fuel load at capture, and capture day for both adults and juveniles. For all adults, length of stay differed by species (χ <sup>2</sup> = 38.8, df = 2, p < 0.001), and the pattern indicated longer stopover lengths with increasing migration distance. Semipalmated plovers had the shortest length of stay (5.9 days ± 3.0 SE), followed by semipalmated sandpipers (14.4 ± 1.7 days), and white-rumped sandpipers (21.4 ± 1.2 days; **Figure 2**). There also was an interaction between capture day and relative fuel load at capture for adults (χ <sup>2</sup> = 11.3, df = 1, p < 0.001) but not juveniles (**Figure 2**). Adult birds with high fuel loads captured early in the season stayed longer in James Bay than birds with lower fuel loads. This pattern changed later in the season;

birds with high fuel loads stayed fewer days than birds with low fuel loads.

Juveniles with higher fuel loads at capture had shorter stopover lengths in James Bay (χ <sup>2</sup> = 11.0, df = 1, p < 0.001), and this pattern did not change throughout the season (**Figure 2**). A species by capture day of year interaction (χ <sup>2</sup> = 15.1, df = 4, p < 0.01) indicated that, while controlling for relative fuel

load at capture, pectoral sandpipers captured later in the season had shorter stopover lengths. This contrasted with the other species in which later captures had slightly longer stopovers (**Figure 2**; **Supplementary Figure 4**). This indicated longer stopover durations for the longest distance migrant juveniles (pectoral sandpipers) captured during the peak migration period (∼August 7th, the time of first juvenile arrivals) and late August but not during early September.

For both age classes, relative fuel loads at departure differed by species (juveniles: χ <sup>2</sup> = 38.5, df = 4, p < 0.001; adults: χ <sup>2</sup> = 50.0, df = 2, p < 0.001; **Figure 3**). For adults, individuals with longer migration distances had larger relative fuel loads at departure (semipalmated plover: 0.20 ± 0.08 (mean ± SE); semipalmated sandpiper: 0.53 ± 0.05; white-rumped sandpiper: 0.68 ± 0.04). For juveniles, there was no clear association between departure fuel loads and migration distance. Pectoral sandpipers (the longest distance migrant juvenile) did not have higher departure fuel loads than other species (**Figure 3**), and differences only were detected between least and semipalmated sandpipers (Tukey HSD: t = −4.0, df = 81.3, p < 0.01) and semipalmated plovers and semipalmated sandpipers (Tukey HSD: t = −5.1, df = 73.4, p < 0.001).

Wind profit, precipitation, relative fuel load at departure, and an interaction between wind profit and fuel load explained departure decisions in adult shorebirds. There was no difference in departure decisions by species; therefore, wind selectivity at departure did not differ by migration distance. Individuals were more likely to depart on nights with higher wind support, but those with lower fuel loads at departure were more likely to depart in unfavorable winds (juveniles: χ <sup>2</sup> = 10.3, df = 1, p < 0.01; adults: χ <sup>2</sup> = 14.4, df = 1, p < 0.001; **Figure 4**). On nights with precipitation, adults were less likely to depart (χ 2 = 112.8, df = 1, p < 0.001), but juveniles would depart regardless of precipitation (N.S. term removed from model). For juveniles, species had different patterns of wind selectivity at departure (χ <sup>2</sup> = 9.5, df = 4, p = 0.0498). This was driven by uncertainty in the relationship for lesser yellowlegs (though there were no statistically significant differences in wind selectivity by species in post-hoc tests), and all other species had higher departure probability with increasing wind profit.

#### Tailwinds *En Route* and Flight Speeds

Tailwind support for adults during southbound migration was best explained by a model with species, departure day of year, and an interaction between the two variables. Overall, there was no difference in mean tailwind support en route between species (χ <sup>2</sup> = 1.6, df = 2, p = 0.44); however, there was a species by departure day interaction (χ <sup>2</sup> = 6.2, df = 2, p = 0.04) that was driven by an increase in tailwind support with later departure dates for semipalmated plovers (1.1 m/s ± 0.5 increase in tailwind support per day over the 8 d departure window for the species). For juveniles, no predictors explained tailwind support en route. Like adults, all species of juvenile shorebirds migrated with tailwind support (adults: 7.6 ± 6.3 m/s; juveniles: 7.2 ± 7.4 m/s).

For adult shorebirds, a model with species and a quadratic variable of tailwind explained ground speeds. Ground speeds differed by species (χ <sup>2</sup> = 11.5, df = 2, p < 0.01; **Figure 5**), and there was a pattern of faster ground speeds with increasing migration distance. As predicted, white-rumped sandpipers achieved faster ground speeds than semipalmated sandpipers (20.1 ± 0.7 SE and 17.5 ± 0.8 m/s respectively; Tukey HSD: t = −2.6, df = 36.8, p = 0.04), but there was no difference between semipalmated plovers and the other species. Ground speeds were fastest for individuals flying with high tailwind support (**Supplementary Figure 5**). For juveniles, species and relative fuel load at departure remained in the final model. Ground speeds differed by species (χ <sup>2</sup> = 16.4, df = 4, p < 0.01; **Figure 5**). Pectoral sandpipers, the longest distance migrant juvenile, had faster ground speeds (24.0 ± 1.6 m/s) than least sandpipers (17.3 ± 1.3 m/s) and semipalmated plovers (18.4 ± 1.2 m/s), but no other pairwise comparisons were significantly different. Across species, juvenile birds with higher fuel loads at departure migrated with higher ground speeds (χ <sup>2</sup> = 6.6, df = 1, p = 0.01).

Airspeeds differed by species for adult shorebirds (χ <sup>2</sup> = 6.3, df = 2, p = 0.04). Adult white-rumped sandpipers migrated with faster airspeeds (13.5 m/s ± 0.7 SE) than semipalmated sandpipers (11.5 ± 0.9 m/s) indicating faster airspeeds in species with longer migration distances, though the relationship was no longer significant in post-hoc means comparisons (Tukey HSD: t = −0.1, df = 47.8, p = 0.10). Semipalmated plover airspeed (13.1 ± 1.5 m/s) did not differ from the other species (p > 0.05). For juveniles, no model predictors explained airspeeds; therefore, all species had similar airspeeds (mean 19.3 m/s ± 5.3 SD) and there was no clear association with migration distance.

#### Stopover and Detection Probability

Approximately 38% (n = 118) of all birds detected in the southern array stopped in North America at least once (**Table 2**). For adults, stopover probability differed by species (χ <sup>2</sup> = 14.6, df = 2, p < 0.001). White-rumped sandpiper, the longest distance migrant adult, had the lowest stopover probability in North America, and this was lower than that of semipalmated sandpipers (Tukey HSD: z = 3.3, p < 0.01) and semipalmated plovers (Tukey HSD: z = 2.6, p = 0.02). Individuals with higher relative fuel loads at departure were less likely to make a stop in North America for both adults (χ <sup>2</sup> = 10.8, df = 1, p < 0.01) and juveniles (χ <sup>2</sup> = 8.4, df = 4, p < 0.01; **Figure 6**; **Supplementary Figure 6**). For juveniles, species was not a significant predictor in the final model, but there was a pattern of a lower stopover probability for the longest distance migrant juvenile, pectoral sandpiper, compared to other species (**Figure 6**; **Supplementary Figure 6**).

Detection probability of adult semipalmated and whiterumped sandpipers in North America only was explained by relative fuel load at departure (χ <sup>2</sup> = 15.8, df = 1, p < 0.001). For both species, individuals with higher departure fuel loads were more likely to be detected in the southern array (**Figure 7**; **Supplementary Figure 7**). Adults with high fuel loads at departure (relative fuel load = 1) had approximately three times higher odds of detection in the southern array than birds with no fat mass at departure (relative fuel load = 0). For juveniles, only species remained in the final model (χ <sup>2</sup> = 11.1, df = 4, p = 0.03). Pectoral sandpipers, the longest distance migrant juvenile, had a lower detection probability than semipalmated plovers (Tukey HSD: z = −0.9, p = 0.03; **Figure 7**), but no other pairwise comparisons were significant. For juveniles, relative fuel load at departure did not remain in the final model (χ <sup>2</sup> = 2.1, df = 1, p = 0.14), though there was a trend of higher probability of detection for birds with higher fuel loads at departure (**Figure 7**; **Supplementary Figure 7**).

### DISCUSSION

Stopover and migration behaviors of adult shorebirds were associated with migration distance and matched behaviors consistent with a time-minimizing migration strategy. Adult white-rumped sandpipers, the longest distance migrant, had longer stopovers in James Bay, departed James Bay with higher relative fuel loads, migrated with faster airspeeds and ground speeds, and had a lower probability of stopover in North America after departing James Bay than semipalmated sandpipers and semipalmated plovers, species that migrate shorter distances. The relationship between migration strategies and migration distance was not as clear in juvenile shorebirds. The longest distance migrant juvenile, pectoral sandpipers, did not depart James Bay with higher fuel loads than shorter distance migrant juveniles, nor migrate with faster airspeeds than other species. They did, however, have longer stopovers in James Bay earlier in the migratory period, migrate with faster ground speeds, and tended to have a lower stopover probability outside of James Bay than shorter distance migrants.

The less clear relationship between migration distance and migratory behavior for juvenile birds than adults simply may be a result of inexperience. We found that adults were less likely to depart the subarctic on nights with precipitation if the

winds were supportive, but juveniles would depart regardless of precipitation. Possibly juveniles are more time-constrained because they tend to arrive at the stopover site later in the season than adults. Late arrival may coincide with the peak of southbound raptor migration (Lank et al., 2003; Ydenberg et al., 2004) or declines in dipteran larvae and oligochaete prey abundance at intertidal marsh habitats at James Bay (Morrison et al., 1982), perhaps because of seasonal weather patterns and/or prey depletion by shorebirds (Székely and Bamberger, 1992; Salem et al., 2014). Additionally, individuals departing from the subarctic later in the migratory period may encounter unfavorable wind patterns along the Atlantic coast (La Sorte et al., 2015a), which could increase departure probability under poor conditions. Ultimately, the decision to depart under poor weather conditions could result in juvenile mortality (Newton, 2007) and selection favoring departure under favorable weather conditions.

In contrast to predictions of time-minimization (e.g., McLaren et al., 2012; Nilsson et al., 2014; McCabe et al., 2018), we found that all shorebird species were wind selective at departure regardless of migration distance. Wind selectivity may indicate that all groups were attempting to minimize energy expenditure during migration. This could be a result of a broader tendency toward energy-minimizing strategies on southbound compared to northbound migration (Karlsson et al., 2012; Nilsson et al., 2013; Horton et al., 2016; Duijns et al., in press). Alternatively, departure with favorable wind conditions could reduce total migration time by reducing energy costs and increasing flight range. Supportive winds during migration can cut flight energy expenditures in half and double a bird's flight range (Liechti and Bruderer, 1998), and this could reduce the number of stopovers and subsequent associated search and settling time.

We identified a pattern of lower wind selectivity at departure for birds with lower relative fuel loads. This matches theoretical predictions of wind selectivity in optimal migration theory if refueling opportunities are poor (Liechti and Bruderer, 1998). Lean birds with low refueling rates should be more likely to depart in headwinds because there is a higher energetic cost of migrating into headwinds with heavier fuel loads (Liechti and Bruderer, 1998). If foraging opportunities are poor, lean individuals may leave the subarctic in anticipation of better refueling opportunities elsewhere (i.e., the "expectation rule"; Alerstam and Lindström, 1990; Alerstam, 2011), whereas individuals in good condition may be able to afford to

wait for favorable winds to curtail costs of carrying high fuel loads.

Body condition (relative fuel load at capture) also moderated length of stay and stopover and detection probability of shorebirds outside of James Bay. Juveniles with high fat mass (relative fuel load = 1) remained in James Bay approximately 17 fewer days than birds with no fat at capture (relative fuel load = 0). This is consistent with other studies (e.g., Matthews and Rodewald, 2010; Seewagen and Guglielmo, 2010; Cohen et al., 2014), though some studies do not detect such a relationship (Skagen and Knopf, 1994; Lyons and Haig, 1995; Lehnen and Krementz, 2007). For adults, individuals with high fuel loads at capture early in the migratory period stayed longer at the stopover site than individuals with low fuel loads. This pattern changed later in the migratory period; individuals with high fuel loads captured late in the migratory period had shorter stopover durations. This pattern simply could be the result of more time available to forage prior to the arrival of migratory birds of prey (Lank et al., 2003; Ydenberg et al., 2004) or the onset of freezing temperatures and declining prey availability (Morrison et al., 1982).

Across species, individuals in better condition were less likely to make a subsequent stop in North America. Longer distance migrants were less likely to make a stop than shorter distance migrants with the same relative fuel loads at departure, which could indicate less flexibility in migratory strategies for longer distance migrants. Surprisingly, many individuals were not detected making a stop in North America. Stopover probability may be higher than the levels observed in our study because of tag loss or stopover outside of the southern array after a tag's last detection; however, we found evidence that some individuals attempted non-stop transoceanic flights to the Caribbean or South America. In this study, one white-rumped sandpiper made a non-stop flight from James Bay to Barbados via the Bay of Fundy (Nova Scotia, Canada), a trip of ∼4,800 km in ∼5 days, after which it stayed in Barbados for at least 2.5 d. This result mirrors that of a semipalmated sandpiper with a geolocator (Brown et al., 2017) which made a non-stop transoceanic flight from James Bay to Venezuela (5,270 km). This phenomenon of non-stop flights was not identified in historical studies for James Bay, which identified shorebirds stopping along the Atlantic seaboard prior to transoceanic flights (Morrison, 1978, 1984; Morrison and Harrington, 1979). Future work should examine if body condition has changed for shorebirds during stopover at James Bay resulting in non-stop flights or if these flights went undetected in historical studies.

Body condition also was related to detection probability in the southern array, which we cautiously interpret as a metric reflecting apparent survival of individuals after their departure from James Bay. This pattern of higher detection of birds in better body condition was clear for white-rumped and semipalmated sandpiper adults, and though non-significant in juveniles, there was a positive relationship between relative fuel load at departure and detection probability. The pattern may be less clear for juveniles than adults because juvenile shorebirds may be more prone to mortality from predation (Whitfield, 2003; Van Den Hout et al., 2008) or inclement weather during migration (Newton, 2007). Alternatively, juveniles may migrate through more variable routes because of inexperience (Able and Bingman, 1987; Chernetsov, 2016), resulting in lower detection probability.

Given the high energetic demands of migration and physiological limitations of powered flight, it is not surprising that migration distance is associated with different migratory strategies, or suites of migratory behaviors in birds. These migratory strategies have not evolved independently of other traits, such as body size (La Sorte et al., 2013; Zhao et al., 2017, 2018; Horton et al., 2018) and wing shape (Minias et al., 2015; Vágási et al., 2016). In this study, we cannot disentangle migration distance from other species differences such as body size and shape. For closely related sandpipers in this study, migration distance tends to scale positively with lean mass (**Table 1**; least sandpipers traveling the shortest distance, semipalmated sandpiper intermediate distances, and whiterumped and pectoral sandpipers traveling the farthest). Similarly, longer-distance migrants (white-rumped and pectoral sandpiper) have more pointed wings compared to shorter distance migrants (e.g., least sandpiper), which may allow for more efficient long-distance migratory flights and offset costs of larger body size (Rayner, 1988; Hedenström, 2007). Future studies should continue to investigate the complex relationships between these traits and migratory strategies.

The use of a widespread automated radio telemetry network to study migratory strategies is not without limitations. The data obtained from this system are constrained to the spatial extent of receiver stations, and the status of individuals prior to capture is unknown. At our study site, species with farther migratory destinations also tend to breed at higher latitudes (**Table 1**); therefore, the migratory strategies we observed may be a result of distance traveled prior to arrival to James Bay as well as the distance yet to travel. Similarly, migratory strategies may be influenced by events prior to arrival at the stopover site, such as reproductive success (Inger et al., 2010). In this study, we made inferences about flights and stopovers from biologically relevant flight speeds, but we cannot exclude the possibility of misclassification of stopover decisions of birds with circuitous routes (and therefore low flight speeds). Future studies should compare this classification approach with true bird location data, such as from small GPS tags, to validate these inferences. Despite these constraints, this system provides high temporal resolution estimates of length of stay, departure times, and flight speeds of individuals at a site where recapture and monitoring of individuals otherwise is difficult.

This work is one of the first to track small shorebirds with fine temporal resolution during southbound migration over a broad spatial scale. It is the first migration tracking study of white-rumped sandpipers, least sandpipers, and semipalmated plovers outside of stopover or breeding sites, and it is the first to link body condition of individuals at a stopover site directly to future stopover probability for small shorebirds. Overall, this study shows that migration strategies of small shorebirds are linked to migration distance, but stopover, departure, and flight behaviors are moderated by body condition.

#### DATA AVAILABILITY

The raw data supporting the conclusions of this paper will be made available by the authors, without undue reservation, upon request.

#### ETHICS STATEMENT

Capture, banding, and blood sampling were approved by Trent University and Environment and Climate Change Canada's Animal Care Committees and carried out under permit from Environment and Climate Change Canada.

#### AUTHOR CONTRIBUTIONS

AA, EN, and PS: conceptualized and designed the study. AA and CF: collected the data. AA: analyzed the data and wrote the paper. EN, PS, SD, and CF: contributed significantly to data interpretation and writing. AA, EN, PS, CF, and SD: approved the final version of the manuscript.

#### FUNDING

This study was funded by Environment and Climate Change Canada, the United States Fish and Wildlife Service Neotropical Migratory Bird Conservation Act (Award

#### REFERENCES


#F14AP00405 and #F17AP00668), Ontario Ministry of Natural Resources and Forestry Species at Risk Stewardship Fund, W. Garfield Weston Fellowship for Northern Research, and the Ontario Trillium Scholarship.

#### ACKNOWLEDGMENTS

We thank the many volunteers, staff, and collaborators from the James Bay Shorebird Project for their assistance with field work and data management. We thank Moose Cree First Nation for their collaboration and continued work to protect shorebirds in their Traditional Territory. We appreciate the support of Bird Studies Canada with the Motus Wildlife Tracking System, field work, and nanotag deployment. We also thank the Royal Ontario Museum Ornithology Collection for field work support and assistance with molecular sexing analyses. We thank the Ontario Ministry of Natural Resources and Forestry for logistical support with field work and helicopter surveys. We are grateful for shorebird photographs from S. Bonnett, A. Lenske, and M. Peck.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00251/full#supplementary-material


class, and annual cycle. Auk 124, 1037–1046. doi: 10.1642/0004- 8038(2007)124[1037:VITWMO]2.0.CO;2


Pennycuick, C. J. (1975). Mechanics of flight. Avian Biol. 5, 1–75.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Anderson, Duijns, Smith, Friis and Nol. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Individual Spatial Consistency and Dietary Flexibility in the Migratory Behavior of Northern Gannets Wintering in the Northeast Atlantic

W. James Grecian<sup>1</sup> \* † , Hannah J. Williams 2,3 \* † , Stephen C. Votier <sup>4</sup> , Stuart Bearhop<sup>2</sup> , Ian R. Cleasby 2,5, David Grémillet 6,7, Keith C. Hamer <sup>8</sup> , Mélanie Le Nuz <sup>9</sup> , Amélie Lescroël 6,10, Jason Newton<sup>11</sup>, Samantha C. Patrick <sup>12</sup>, Richard A. Phillips <sup>13</sup> , Ewan D. Wakefield<sup>14</sup> and Thomas W. Bodey 2,15 \*

#### Edited by:

Yolanda E. Morbey, University of Western Ontario, Canada

#### Reviewed by:

Kyle Elliott, McGill University, Canada Jaime Albino Ramos, University of Coimbra, Portugal

#### \*Correspondence:

W. James Grecian wjg5@st-andrews.ac.uk Hannah J. Williams h.williams@swansea.ac.uk Thomas W. Bodey t.w.bodey@exeter.ac.uk

†These authors have contributed equally to this work

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 14 March 2019 Accepted: 22 May 2019 Published: 12 June 2019

#### Citation:

Grecian WJ, Williams HJ, Votier SC, Bearhop S, Cleasby IR, Grémillet D, Hamer KC, Le Nuz M, Lescroël A, Newton J, Patrick SC, Phillips RA, Wakefield ED and Bodey TW (2019) Individual Spatial Consistency and Dietary Flexibility in the Migratory Behavior of Northern Gannets Wintering in the Northeast Atlantic. Front. Ecol. Evol. 7:214. doi: 10.3389/fevo.2019.00214 <sup>1</sup> Sea Mammal Research Unit, Scottish Oceans Institute, University of St Andrews, St Andrews, United Kingdom, <sup>2</sup> Centre for Ecology and Conservation, University of Exeter, Cornwall, Penryn, United Kingdom, <sup>3</sup> Swansea Lab for Animal Movement, Department of Biosciences, College of Science, Swansea University, Swansea, United Kingdom, <sup>4</sup> Environment and Sustainability Institute, University of Exeter, Cornwall, Penryn, United Kingdom, <sup>5</sup> RSPB Centre for Conservation Science, Inverness, United Kingdom, <sup>6</sup> Centre d'Ecologie Fonctionnelle et Evolutive, UMR 5175, CNRS - Université de Montpellier - Université Paul-Valéry Montpellier - EPHE, Montpellier, France, <sup>7</sup> Percy FitzPatrick Institute of African Ornithology, DST/NRF Centre of Excellence at the University of Cape Town, Cape Town, South Africa, <sup>8</sup> School of Biology, University of Leeds, Leeds, United Kingdom, <sup>9</sup> Réserve Naturelle des Sept-Iles, Ligue pour la Protection des Oiseaux, Pleumeur Bodou, France, <sup>10</sup> Point Blue Conservation Science, Petaluma, CA, United States, <sup>11</sup> Natural Environment Research Council Life Sciences Mass Spectrometry Facility, Scottish Universities Environmental Research Centre, East Kilbride, United Kingdom, <sup>12</sup> School of Environmental Sciences, University of Liverpool, Liverpool, United Kingdom, <sup>13</sup> British Antarctic Survey, Natural Environment Research Council, Cambridge, United Kingdom, <sup>14</sup> Institute of Biodiversity, Animal Health & Comparative Medicine, College of Medical, Veterinary & Life Sciences, University of Glasgow, Glasgow, United Kingdom, <sup>15</sup> School of Biological Sciences, University of Auckland, Auckland, New Zealand

Migration is a fundamental behavioral process prevalent among a wide variety of animal taxa. As individuals are increasingly shown to present consistent responses to environmental cues for breeding or foraging, it may be expected that approaches to migration would present similar among-individual consistencies. Seabirds frequently show consistent individual differences in a range of traits related to foraging and space-use during both the breeding and non-breeding seasons, but the causes and consequences of this consistency are poorly understood. In this study, we combined analysis of geolocation and stable isotope data across multiple years to investigate individual variation in the non-breeding movements and diets of northern gannets Morus bassanus, and the consequences for changes in body condition. We found that individuals were highly repeatable in their non-breeding destination over consecutive years even though the population-level non-breeding distribution spanned >35◦ of latitude. Isotopic signatures were also strongly repeatable, with individuals assigned to one of two dietary clusters defined by their distinct trophic (δ <sup>15</sup>N) and spatial (δ <sup>13</sup>C) position. The only non-breeding destination in which the two dietary clusters co-occurred was off the coast of northwest Africa. The majority of individuals adopted a consistent foraging strategy, as they remained within the same dietary cluster across years, with little variation in body mass corrected for size among these consistent individuals. In contrast, the few individuals that switched clusters between years were in better condition relative to the rest of the population, suggesting there may be benefits to flexibility during the non-breeding period. Our results indicate that a consistent migratory strategy can be effective regardless of wintering region or diet, but that there may be benefits to those individuals able to display flexibility. This appears to be an important behavioral strategy that may enhance individual condition.

Keywords: individual variation, carry-over effects, Geolocator (GLS), stable isotope analysis (SIA), animal migration

#### INTRODUCTION

Animal migration is a fundamental behavioral process that involves seasonal movements between habitats in response to resource heterogeneity. Although prevalent among a wide variety of animal taxa, there is enormous variation in migration strategy in terms of the distance traveled and the degree of seasonal site fidelity, which ranges from strong philopatry to the loose tracking of seasonal resources (Webster et al., 2002; Newton, 2008). Variation among individuals is often attributed to age, sex or morphology (Marra, 2000; Alerstam et al., 2003; Bailleul et al., 2010) but may also be the result of differences in foraging behavior, breeding success or endogenous control (Bradshaw et al., 2004; Phillips et al., 2005; Broderick et al., 2007; Dias et al., 2011). Despite increasing evidence for individual differences in migratory behavior, the degree of consistency or plasticity and their causes and consequences remain incompletely understood (Chapman et al., 2011; Phillips et al., 2017).

A high degree of consistency in migration strategy with strong individual non-breeding site fidelity can be advantageous if this allows access to predictable foraging resources (Bradshaw et al., 2004; Weimerskirch, 2007) and reduces risks associated with exploring novel habitat (McNamara and Dall, 2010). Relatively inflexible strategies are seen in a number of groups, including passerines (Cuadrado et al., 1995), waterfowl (Hestbeck et al., 1991), cetaceans (Calambokidis et al., 2001), pinnipeds (Bradshaw et al., 2004), seabirds (Phillips et al., 2005, 2006), turtles (Broderick et al., 2007), and sharks (Jorgensen et al., 2010). Conversely, if food availability is unpredictable, or environmental conditions are prone to deteriorate in particular regions during the non-breeding period, selection should favor migratory flexibility or nomadism (Andersson, 1980) and facilitate plastic responses within individuals (Switzer, 1993; Sutherland, 1998). Such strategies are seen in groups including seabirds (Dias et al., 2011), waterbirds (Pedler et al., 2014), ungulates (Morrison and Bolger, 2012), and fish (Tibblin et al., 2016). Thus, the extent to which individuals respond to biotic and abiotic variation across time and space can select for clear individual differences in both movement and foraging strategies.

Recent studies of migrant birds show that individual differences in habitat selection and foraging behavior can influence diet quality during the non-breeding season and impact subsequent breeding traits such as body condition, timing of breeding, egg volume, and breeding success (Bearhop et al., 2004; Inger et al., 2008; Sorensen et al., 2009; Hoye et al., 2012), with important fitness consequences (Marra et al., 1998; Crossin et al., 2010; Inger et al., 2010; Harrison et al., 2011). Thus, individuals that pursue a non-breeding strategy that produces strong negative carry-over effects might be expected to preferentially switch strategies in subsequent years, reducing within-individual consistency (Switzer, 1993; Dias et al., 2011; Morrison and Bolger, 2012). Understanding the incidence and implications of individual consistency or flexibility in nonbreeding behavior is therefore a key issue in animal ecology, yet there are few long-term studies that quantify these individual differences over multiple seasons or migration periods (Araújo et al., 2011; Phillips et al., 2017).

Marine predators such as seabirds provide an ideal model for examining such questions as they exhibit a broad spectrum of individual differences in behavior (Votier et al., 2010; Patrick et al., 2014). Recent work suggests these differences are likely to develop through ontogeny (Votier et al., 2017) as individuals learn to target profitable habitat (Grecian et al., 2018). In addition, many species display site fidelity to broadly productive regions during the non-breeding period (Grecian et al., 2016; Phillips et al., 2017). Disentangling individual differences in non-breeding foraging behavior and site fidelity may provide insights into how carry-over effects shape the annual cycle of an individual (Furness et al., 2006). For example, when local conditions are poor, individuals may switch non-breeding region while targeting the same preferred prey or, alternatively, may remain within the same preferred non-breeding region and instead switch prey types (Orben et al., 2015).

In this study, we combine multi-year deployments of geolocation loggers with stable carbon and nitrogen isotope analysis of winter-grown feathers to investigate the degree of individual consistency in the non-breeding destination and foraging behavior of a generalist marine predator, the northern gannet (Morus bassanus), tracked from four breeding colonies in the NE Atlantic. Gannets exhibit a southward-oriented chain migration following a flyway running along the coast of western Europe and Africa (Fort et al., 2012). Variation in migratory behavior, the migration path, final non-breeding destination and foraging behavior during these periods, occurs both among and within populations (Kubetzki et al., 2009; Fort et al., 2012; Deakin et al., 2019), and one recent study has shown that individuals in the NW Atlantic exhibit consistent behavioral strategies in successive years (Fifield et al., 2014). Additionally, the nonbreeding distributions of gannets may have changed in recent decades (Kubetzki et al., 2009), suggesting a degree of plasticity in migratory behavior. Such shifts could be linked to changes in human fishing activity as many seabirds are attracted to the foraging opportunities afforded by fisheries (Pichegru et al., 2007; Votier et al., 2010; Bodey et al., 2014a; Patrick et al., 2015).

This behavior may come at a cost; as well as increasing the risk of bycatch (Bicknell et al., 2013), diets high in discards can have reduced lipid content compared to pelagic fishes, with the potential for adverse effects on body condition and breeding success (Grémillet et al., 2008; Votier et al., 2010). Should dependency on this resource also be evident in the nonbreeding season, there may be further fitness consequences via carry-over effects. We therefore examine whether differences in non-breeding destination and diet affect individual body condition (as a short-term fitness proxy) during the subsequent breeding season.

# MATERIALS AND METHODS

#### Study System and Data Collection

We collected data between 2008 and 2012 from gannets at four colonies in the northeast Atlantic: Bass Rock, Scotland; Grassholm, Wales; Great Saltee, Ireland; and Rouzic, France (**Figure 1**). In total, 187 breeding adults with chicks aged between 2 and 7 weeks (egg laying is poorly synchronized across the breeding colony) were caught at the nest during changeover of brood-guard duties using a brass noose or crook attached to the end of a carbon fiber pole (**Table 1**). On capture, the mass (to the nearest 50 g) and bill length (to the nearest 0.1 mm) of each individual was measured, and sex was subsequently assigned from DNA using 2550F, 2718R, or 2757R primers (Griffiths et al., 1998; Fridolfsson and Ellegren, 1999) following Stauss et al. (2012).

#### Non-breeding Destination

Combined geolocation-immersion loggers (Mk 19, 15, and 5, British Antarctic Survey, Cambridge UK) were deployed on 77 of these individuals across the four colonies. Loggers were attached TABLE 1 | Summary of samples collected from 187 northern gannets across four breeding colonies between 2008 and 2012 including geolocation loggers and feather stable isotope analysis (SIA).


Numbers in parenthesis indicate individuals sampled in the previous year.

with two cable ties to a plastic ring, which was then fitted to the tarsus and remained in place for up to 2 years before the bird was recaptured at the breeding colony. The total mass of the attachment did not exceed 10 g, representing <0.35% of average adult body mass, and so unlikely to have any adverse effects (Bodey et al., 2018a). The loggers sampled ambient light every minute and recorded the maximum value every 2, 5, or 10 min (Mk 19, 15, and 5 loggers, respectively).

Positions were calculated from logger data following established methods (Wilson et al., 1992; Phillips et al., 2004). Briefly, the timings of sunset and sunrise were estimated using TransEdit2 (British Antarctic Survey, Cambridge UK) using a light-intensity threshold of 16. A minimum dark period of 4 h was set to remove any light-dark transitions created by shading or cloud cover. Latitude was derived from day length, and longitude from the timing of local midday and midnight, with respect to Greenwich Mean Time and Julian day, providing two positions per day with an accuracy of ∼200 km (Phillips et al., 2004). Examination of individual migration tracks revealed latitude to be the major axis of movement, with birds tending to migrate southward from the breeding colonies toward northwest Africa (Fort et al., 2012; **Figure 2**). Plots of displacement from the colony indicated that all individuals reached their final non-breeding destinations by December and remained in this region for a minimum of 1 month before commencing their return migration. The mean latitude and longitude for December was therefore used as the non-breeding destination of each bird.

# Non-breeding Stable Isotopes

Small samples from the 8th primary feather were taken from 148 individuals for stable isotope analysis, with 43 of these individuals sampled a second time when loggers were removed the following year (**Table 1**). Gannets perform a complete annual molt after the breeding season (from September; Ginn and Melville, 1983), suspending molt by the time the return to the breeding colony (January to March) to invest in nest attendance and foraging trips (see Nelson, 2006). Thus, as feathers are metabolically inert after formation and larger feathers grow over a protracted period, the stable isotope ratios of primary feathers were assumed to largely

represent prey consumed at the non-breeding grounds (between October and December).

Feather samples were thoroughly washed with distilled water and placed in a drying oven at ∼40◦C until dry. The barbules were cut into fine pieces and subsamples of 0.7 ± 0.1 mg were weighed into tin cups. Stable isotope analysis of these subsamples was then conducted at the East Kilbride Node of the Natural Environment Research Council Life Sciences Mass Spectrometry Facility via continuous flow isotope ratio mass spectrometry, using a Thermo Fisher Scientific Delta V Plus with a Costech ECS 4010 elemental analyser configured for simultaneous <sup>13</sup>C/12C and <sup>15</sup>N/14N isotope analysis. Stable isotope ratios are reported in δ notation, expressed as parts per thousand (‰) deviation according to the equation δX = [(Rsample/Rstandard)-1], where X is <sup>13</sup>C or <sup>15</sup>N, R is the corresponding ratio <sup>13</sup>C/12C or <sup>15</sup>N/14N, and Rstandard is the ratio of the international references VPDB for carbon and AIR for nitrogen. At set intervals, standards of GEL, <sup>14</sup>N ALA, glycine and tryptophan were analyzed between feather samples in the IRMS. The measurement precision, calculated as the standard deviation of multiple analyses of these standards, was ± 0.1 ‰ for δ <sup>13</sup>C and ± 0.2 ‰ for δ <sup>15</sup>N.

### Consistency in Non-breeding Strategies and Isotopic Clustering

To examine the consistency of non-breeding destination and stable isotope ratios we calculated the repeatability of these traits based on the intra-class correlation coefficient from linear mixed-effect models fitted with bird ID as a random intercept, using the package "rptR" v. 0.9.21 (Stoffel et al., 2017). We used repeatability as a proxy for behavioral consistency, testing the hypothesis that between-individual variance in a particular trait was greater than within-individual variance (Patrick et al., 2014). To estimate the consistency of non-breeding destinations we calculated the repeatability of mean December latitude and longitude for those individuals from Bass Rock (n = 22) and Rouzic (n = 21), that were tracked over two consecutive years. To estimate the consistency in stable isotope ratios during the non-breeding season we calculated the repeatability of δ <sup>13</sup>C and δ <sup>15</sup>N in primary feathers of individuals from Bass Rock (n = 27) and Grassholm (n = 13) sampled in two consecutive years. The three individuals from Great Saltee were excluded from the estimate of isotopic repeatability due to the small, multi-year sample size (**Table 1**).

To test for the occurrence of distinct dietary clusters in stable isotope ratios we fitted a multivariate normal mixture model to δ <sup>13</sup>C and δ <sup>15</sup>N values using the package "mixtools" v. 1.1.0 (Benaglia et al., 2009). The best-fitting model was selected by comparing the log-likelihood of candidate models with differing numbers of clusters. Feather samples were assigned to a dietary cluster based on a probability of assignment >0.5.

#### Consequences of Non-breeding Strategy

We estimated body condition using a scaled mass index (SMI, Peig and Green, 2009) with bill length as a linear measurement of body size in relation to body mass. However, given that 53 of the individuals were measured in multiple years, we extended this approach to a mixed-effects model with an individual level random intercept, fitted using the package "lme4" v. 1.1-18- 1 (Bates et al., 2015). The correlation between body mass and bill length accounting for repeated measures was estimated using the package "rmcorr" v. 0.3.0 (Bakdash and Marusich, 2017). SMI allows for the comparison of the relative size of energy reserves of individuals within a population, avoiding the assumption that larger animals have better body condition due to a higher absolute mass (Peig and Green, 2009). While reproductive success would be a more robust measure of fitness, chick survival from hatching to fledging is over 90% and the majority of offspring mortality occurs during the post-fledging and juvenile period at sea (Nelson, 1966).

The implications of alternative non-breeding strategies (destination and dietary cluster) at the individual level were explored by examining the effects of sex, breeding colony, dietary cluster and non-breeding destination on scaled mass using linear regressions. In cases where there were two observations of an individual's scaled mass in consecutive years, these were fitted as mixed-effects models with individual as a random intercept term. Model selection of linear regressions was based on the F statistic, model selection of mixed-effects linear regression was based on the Chi-squared statistic using likelihood ratio tests. Post-hoc comparisons were made using the package "lsmeans" v. 2.30-0 (Lenth, 2016). All analyses were carried out in R v. 3.4.3 (R Core Team, 2018).

#### RESULTS

#### Consistency in Non-breeding Destination and Stable Isotope Ratios

Gannets spent the month of December in one of three regions: a northern region (>36◦N), around the British Isles and the Bay of Biscay (n = 26); a southern region (<36◦N) from Gibraltar to Mauritania (n = 47); and the Mediterranean Sea (n = 4, all from Rouzic; **Figure 2**). Individuals from all four colonies TABLE 2 | Comparison of the log-likelihoods of multivariate normal mixture models fitted with k distributions.


were present in both the northern and southern regions during the non-breeding period. The 43 birds from Bass Rock and Rouzic that were tracked over two consecutive years exhibited a high degree of consistency in non-breeding destination and were highly repeatable in both mean December latitude (R = 0.91; CI = 0.83, 0.95; P = 0.001) and longitude (R = 0.92; CI = 0.87, 0.96; P = 0.001; **Figure 2**).

Stable isotope values in primary feathers from individuals sampled in two consecutive years at Grassholm (n = 13) were repeatable with respect to both δ <sup>13</sup>C (R = 0.73; CI = 0.35, 0.90; P = 0.004) and δ <sup>15</sup>N (R = 0.57; CI = 0.06, 0.82; P = 0.028). Individuals from Bass Rock (n = 27) also showed significant repeatability in both δ <sup>13</sup>C (R = 0.59; CI = 0.27, 0.79; P = 0.002), and δ <sup>15</sup>N (R = 0.52; CI = 0.20, 0.74; P = 0.002).

#### Isotopic Clustering

Stable isotope ratios in primary feathers sampled from 148 individuals were best described by a mixture of k = 2 multivariate normal distributions (**Table 2**). One cluster centered on −13.9 δ <sup>13</sup>C and 13.2 δ <sup>15</sup>N, and a second cluster centered on −16.1 δ <sup>13</sup>C and 15.8 δ <sup>15</sup>N. The 95% ellipses of the two multivariate normal distributions did not overlap (**Figure 3**). Seventy-three individuals were assigned to cluster 1 and 75 individuals to cluster 2. Of the 43 individuals (Bass Rock n = 27; Grassholm n = 13; Great Saltee n = 3) that were sampled in consecutive years, most were consistent in their cluster assignment, with 16 assigned to cluster 1 and 20 assigned to cluster 2 in both years. Nevertheless, seven individuals switched between clusters from 1 year to the next (Bass Rock n = 3; Grassholm n = 2; Great Saltee n = 2). Six of these were female and switched from cluster 1 to cluster 2 (lower δ <sup>15</sup>N to higher δ <sup>15</sup>N) and one male from Grassholm switched from cluster 2 to cluster 1 (higher δ <sup>15</sup>N to lower δ <sup>15</sup>N).

#### Isotopic Clustering Controlling for Winter Destination

Both non-breeding destination and stable isotope data were available for 56 individuals (Bass Rock n = 35, Grassholm n = 13, Great Saltee n = 8). Colony of origin was unrelated to cluster assignment (χ <sup>2</sup> = 0.51, P = 0.78) or non-breeding region (χ <sup>2</sup> = 0.13, P = 0.94). However, individuals that wintered in the northern region were all assigned to cluster 2 (higher δ <sup>15</sup>N) whereas individuals that wintered in the southern region were assigned to either isotopic cluster (**Figure 4**). No isotope data were available for the four individuals that wintered in the Mediterranean.

FIGURE 3 | Primary feather δ <sup>13</sup>C and δ <sup>15</sup>N values of 148 individual gannets sampled at four colonies in the northeast Atlantic. Crosses represent the mean of the two isotopic clusters identified from a multivariate mixture model and dotted lines represent the 95% ellipses for each distribution. Points are colored based on a probability of group assignment >0.5. Gray lines connect individuals sampled over two consecutive non-breeding periods (n = 43).

# Consequences of Non-breeding Strategy

Female gannets were estimated to be on average 89.4 g (95% CI; 43.9, 134.9) heavier than males (χ 2 <sup>1</sup> = 12.9, P < 0.001). Scaled mass differed slightly between the four colonies (χ 2 <sup>3</sup> = 9.7, P = 0.02) and post-hoc comparisons indicated that, when averaging over sex differences, individuals sampled at Grassholm were 179.6 g ± 58.0 g lighter than individuals sampled at Bass Rock (z = 3.1, P = 0.01) with no other significant differences between colonies.

There was no difference in scaled mass between individuals using either the northern or southern non-breeding region (χ 2 1 = 0.58, P = 0.44), nor were there differences in scaled mass between birds in the two isotopic clusters (χ 2 <sup>1</sup> = 0.0, P = 0.96). Data on scaled mass and feather stable isotope ratios in two consecutive years were available for 34 individuals sampled at Bass Rock (n = 21) and Grassholm (n = 13). In this sample, five individuals switched between the two isotopic clusters and had higher scaled mass (∼200 g heavier) compared to those that did not switch isotopic cluster, after accounting for both colony and sex differences (**Figure 5**). Post-hoc comparisons of marginal means indicated individuals that switched were significantly heavier than individuals in the high δ <sup>15</sup>N cluster (z = 2.53, P = 0.03). Of these five switching individuals, four were female and switched from the low to high δ <sup>15</sup>N cluster and one was male and switched from the high to

low δ <sup>15</sup>N cluster. Scaled mass was unavailable for two other switching individuals.

### DISCUSSION

In this study, we reveal that the non-breeding behavior of individual northern gannets is highly repeatable over consecutive years, with a high degree of site fidelity and consistency in stable isotope ratios during successive non-breeding seasons. Despite substantial differences in destination and variation in foraging strategy among individuals, consistent behaviors during the nonbreeding period had no apparent carry-over effect on scaled mass in the subsequent breeding season.

#### Consistency in Non-breeding Destination and Stable Isotope Ratios

Gannets tracked in this study tended to migrate uniformly southward on a known flyway (Kubetzki et al., 2009; Fort et al., 2012), and spent the non-breeding period in a wide variety of marine habitats including the North Sea, Bay of Biscay, Mediterranean Sea, and Canary Current Upwelling region (Grecian et al., 2016; **Figure 2**). These three regions differ in their environmental conditions, yet individuals tracked over two consecutive years displayed a high degree of nonbreeding site fidelity. The range of ∼4 ‰ in δ <sup>13</sup>C and ∼6 ‰ in δ <sup>15</sup>N in stable isotope data from the broader sample of the population are larger than the estimates of baseline isotopic variation across the differing non-breeding

destinations (McMahon et al., 2013; Magozzi et al., 2017). This suggests that while the population winters across a range of locations, prey are targeted across trophic levels within locations (Inger and Bearhop, 2008).

Adult gannets display consistency in foraging movements and diet within breeding seasons (Patrick et al., 2014; Wakefield et al., 2015; Votier et al., 2017; Bodey et al., 2018b), and the high isotopic consistency observed in individuals in our study that were sampled in consecutive years suggests a similar degree of consistency in both wintering region and the trophic level of prey consumed. Non-breeding site fidelity has been documented in other migratory marine vertebrates (Bradshaw et al., 2004; Broderick et al., 2007; Jorgensen et al., 2010; Phillips et al., 2017) and could allow individuals to increase knowledge of a specific area and thus improve foraging efficiency (Dall et al., 2012).

#### Isotopic Clustering

Pooling the stable isotope data from all colonies indicated two clusters, indicative of alternative foraging strategies that differed in both spatial (δ <sup>13</sup>C, δ <sup>15</sup>N) and trophic (δ <sup>15</sup>N) characteristics. One cluster was described by higher δ <sup>15</sup>N and depleted δ <sup>13</sup>C, consistent with a higher trophic level diet and offshore prey, respectively (Hobson et al., 1994; Post, 2002; Inger and Bearhop, 2008). In contrast, the second cluster was more representative of a diet of inshore (higher δ <sup>13</sup>C) prey at a lower trophic level (depleted δ <sup>15</sup>N). Although δ <sup>15</sup>N can also vary with geographic location (Seminoff et al., 2012; McMahon et al., 2013), the observed difference between these two clusters is greater than the baseline variation between non-breeding destinations (McMahon et al., 2013; Magozzi et al., 2017). In addition, the co-occurrence of individuals from both dietary clusters in the southern wintering area indicates that cluster assignment is not purely driven by the local environment. However, there may be other drivers of the observed isotopic differences, for example individual variation in molt location or feather growth rate could result in a shift in feather isotope signature.

While we lack conventional samples of gannet diet during the winter, the higher trophic level cluster may represent prey obtained primarily as fisheries discards, as δ <sup>15</sup>N values are elevated in demersal relative to pelagic fish (Votier et al., 2010; Bicknell et al., 2013). In contrast, the second cluster is suggestive of a more inshore diet in pursuit of small forage fish (Garthe et al., 2000; Nelson, 2002). The majority of individuals that were sampled in two consecutive years remained in the same isotopic cluster from one year to the next. Therefore, these clusters may represent foraging strategies that reduce competition among individuals though niche differentiation (Phillips et al., 2009; Young et al., 2010; Bodey et al., 2014b). Foraging specializations have been documented during the breeding season for many seabird species (Annett and Pierotti, 1999; Bearhop et al., 2006; Woo et al., 2008; Phillips et al., 2017) including northern gannets where individuals can vary in the extent of their reliance on high trophic level prey such as fisheries discards (Votier et al., 2010; Patrick et al., 2014; Wakefield et al., 2015; Bodey et al., 2018b).

#### Consequences of Non-breeding Strategy

Based on our metric of scaled mass, we did not detect any consequences for individuals consistently pursuing different non-breeding strategies. Neither non-breeding destination nor isotopic cluster was significantly related to scaled mass; instead, sex and colony effects drove the observed differences. This is in contrast to patterns seen in thick-billed murres Uria lomvia, where over-wintering foraging strategies are strongly dependant on body size (Orben et al., 2015). Differences in energetic demands over the breeding season may also lead to variation in body condition (Moe et al., 2002) but all individuals in this study were sampled during the chick provisioning period. Sex-linked differences in scaled mass have been observed previously in Northern gannets and may reflect the differing physiological demands of reproduction, and breeding role specialization, as well as more subtle differences between the sexes in prey-capture techniques, nutritional requirements and fine-scale habitat and prey selection (Stauss et al., 2012; Cleasby et al., 2015; Machovsky-Capuska et al., 2016; Bodey et al., 2018b). The difference in scaled mass between individuals at Grassholm and Bass Rock may reflect variation in the prey resources and environmental conditions accessible to individuals from their respective colonies. For example, the North Sea differs in oceanography to the Celtic Sea and supports fewer competing gannet colonies, though the colony at Bass Rock is much larger than at Grassholm and so within-colony competition will be more severe (Nelson, 2002; Wakefield et al., 2013).

Some individuals remained close to the breeding colonies during the non-breeding period and were consistent in this behavior over the two years (**Figure 2**). Remaining in these areas may decrease energy expenditure by reducing migration costs (Flack et al., 2016), however, this may be offset by the increased energetic requirement for thermoregulation in these colder more northerly waters (Garthe et al., 2012). These individuals were all assigned to the higher δ <sup>15</sup>N cluster which suggests a greater consumption of fisheries discards or a lack lower δ <sup>15</sup>N prey available during the winter period (e.g., shoaling forage fish).

The small number of individuals that switched between the two dietary clusters were in better body condition, after accounting for colony and sex effects, than those that were consistent in their cluster assignment. This difference was largest compared to individuals in the higher δ <sup>15</sup>N cluster, which had relatively low scaled mass. The switching strategy was observed in seven of the 43 individuals (ca. 16%) that were sampled in two consecutive years. Six of these were female and all switched from the lower to higher δ <sup>15</sup>N cluster. The only individual to switch from the higher to lower δ <sup>15</sup>N cluster was male. Although foraging on discards brings an additional risk of mortality via incidental bycatch (Bicknell et al., 2013), previous work suggests such a diet may not be detrimental to adult body condition in Cape gannets (Grémillet et al., 2008). Our findings suggest that individuals capable of switching between higher and lower trophic level diets between non-breeding seasons may benefit when compared to individuals specializing in a diet likely to consist of a high proportion of fisheries discards. Alternatively, individual in better condition may be the only ones capable of investing in more risky behaviors (Geary et al., 2019). The majority of individuals switched to the higher δ <sup>15</sup>N cluster, so this may indicate a short-term benefit of switching to a diet based on fisheries discards or other alternative higher δ <sup>15</sup>N resources within non-breeding region. The higher δ <sup>15</sup>N cluster represents one third of those individuals wintering off the coast of northwest Africa, a region known to have experienced a recent intensification of fishing activity (Worm et al., 2009; Grecian et al., 2016).

# CONCLUSIONS

Our results reveal strong individual consistencies in movement and diet during the non-breeding season, and it is this consistency rather than the strategy itself, that may be important for long-lived species (Ceia et al., 2012; Gilmour et al., 2015). Indeed, such consistency has been demonstrated to result in similar life-time reproductive success among Brünnich's guillemots (Uria lomvia) pursuing different foraging strategies (Woo et al., 2008). Individual repeatability is frequently seen in marine vertebrates despite strong between-year variation in environmental variables and prey fields (Cherel et al., 2007). Importantly however, such consistency could come at a price for highly specialized individuals; for example, changes to anthropogenic subsidies disproportionately affect sub-sections of populations that specialize on such resources (Whitehead and Reeves, 2005; Bicknell et al., 2013). The findings here further highlight the importance of research that links different aspects of behavior between seasons or across annual cycles to understand ecological differentiation among individuals, populations and species (Friesen et al., 2007; Bodey et al., 2014b; Wakefield et al., 2015); and the need to consider the degree of flexibility of individuals and populations to changes in resource availability (Grémillet and Boulinier, 2009).

# DATA AVAILABILITY

Telemetry data are available through the BirdLife International Seabird Tracking Database: http://www.seabirdtracking.org. Biometric data are available through the University of St Andrews Research Portal: https://doi.org/10.17630/b3c6dc92- 13eb-447d-82d3-acf01d029bc9 (Grecian et al., 2019).

# ETHICS STATEMENT

Birds were ringed and loggers deployed with permits and approval from the British Trust for Ornithology and Scottish Natural Heritage. Tissue samples were collected under license from the UK Home Office (PPL 30/3065 and 40/3408).

# AUTHOR CONTRIBUTIONS

SV, SB, and KH conceived the study. WG, HW, and TB wrote the first draft of the manuscript. WG, HW, SV, SB, IC, DG, KH, ML, AL, SP, EW, and TB collected data. WG, HW, IC, JN, RP, EW, and TB conducted analyses. All authors contributed to manuscript revision, read and approved the submitted version.

# FUNDING

This study was funded by the UK Natural Environment Research Council (NERC Standard Grant NE/H007466/1). SB is currently funded by an ERC consolidator's grant (310820: STATEMIG). EW is funded by the NERC (Independent Research Fellowship NE/M017990/1). Tissue samples were collected under license from the UK Home Office (PPL 40/3408).

# ACKNOWLEDGMENTS

We thank Sir Hew Hamilton-Dalrymple (Bass Rock), the RSPB (Grassholm), the Neale family (Great Saltee) and the Ligue pour la Protection des Oiseaux and Réserve Naturelle Nationale des Sept- Îles (Rouzic) for permitting and facilitating fieldwork. We are grateful to Fraser Bell, Nadja Christen, Armel Deniau, Richard Inger, Mark Jessopp, Greg & Lisa Morgan, Claudia Stauss, Alyn Walsh, Emma Wood, the Scottish Seabird Center, Maggie Sheddan, and Venture Jet for field assistance.

#### REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Grecian, Williams, Votier, Bearhop, Cleasby, Grémillet, Hamer, Le Nuz, Lescroël, Newton, Patrick, Phillips, Wakefield and Bodey. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Ecological and Evolutionary Consequences of Environmental Change and Management Actions for Migrating Fish

Carl Tamario<sup>1</sup> , Johanna Sunde<sup>1</sup> , Erik Petersson<sup>2</sup> , Petter Tibblin<sup>1</sup> and Anders Forsman<sup>1</sup> \*

<sup>1</sup> Department of Biology and Environmental Science, Linnaeus University, Kalmar, Sweden, <sup>2</sup> Department of Aquatic Resources, Institute of Freshwater Research, SLU, Drottningholm, Sweden

Migration strategies in fishes comprise a rich, ecologically important, and socioeconomically valuable example of biological diversity. The variation and flexibility in migration is evident between and within individuals, populations, and species, and thereby provides a useful model system that continues to inform how ecological and evolutionary processes mold biodiversity and how biological systems respond to environmental heterogeneity and change. Migrating fishes are targeted by commercial and recreational fishing and impact the functioning of aquatic ecosystems. Sadly, many species of migrating fish are under increasing threat by exploitation, pollution, habitat destruction, dispersal barriers, overfishing, and ongoing climate change that brings modified, novel, more variable and extreme conditions and selection regimes. All this calls for protection, sustainable utilization and adaptive management. However, the situation for migrating fishes is complicated further by actions aimed at mitigating the devastating effects of such threats. Changes in river connectivity associated with removal of dispersal barriers such as dams and construction of fishways, together with compensatory breeding, and supplemental stocking can impact on gene flow and selection. How this in turn affects the dynamics, genetic structure, genetic diversity, evolutionary potential, and viability of spawning migrating fish populations remains largely unknown. In this narrative review we describe and discuss patterns, causes, and consequences of variation and flexibility in fish migration that are scientifically interesting and concern key issues within the framework of evolution and maintenance of biological diversity. We showcase how the evolutionary solutions to key questions that define migrating fish—whether or not to migrate, why to migrate, where to migrate, and when to migrate—may depend on individual characteristics and ecological conditions. We explore links between environmental change and migration strategies, and discuss whether and how threats associated with overexploitation, environmental makeovers, and management actions may differently influence vulnerability of individuals, populations, and species depending on the variation and flexibility of their migration strategies. Our goal is to provide a broad overview of knowledge in this emerging area, spur future research, and development of informed management, and ultimately promote sustainable utilization and protection of migrating fish and their ecosystems.

Keywords: biodiversity, climate change, developmental plasticity, evolution, fish migration, fishway, phenotypic flexibility, spawning migration

#### Edited by:

Yolanda E. Morbey, University of Western Ontario, Canada

#### Reviewed by:

Christer Brönmark, Lund University, Sweden Sue Katherine Lowerre-Barbieri, University of Florida, United States

> \*Correspondence: Anders Forsman anders.forsman@lnu.se

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 08 March 2019 Accepted: 28 June 2019 Published: 10 July 2019

#### Citation:

Tamario C, Sunde J, Petersson E, Tibblin P and Forsman A (2019) Ecological and Evolutionary Consequences of Environmental Change and Management Actions for Migrating Fish. Front. Ecol. Evol. 7:271. doi: 10.3389/fevo.2019.00271

# INTRODUCTION

Migratory fish showcase a scientifically interesting example of biological diversity that is of considerable ecological and socioeconomic importance (Leggett, 1977; Lynch et al., 2016; Oke and Hendry, 2019). Given the alarming situation for migratory fish worldwide, there is a need for a better knowledge and understanding of the patterns, causes and consequences of variation of their migratory behavior. Important challenges addressed in this contribution include to identify how different ecological drivers influence the evolution and variation in migratory behavior, and to illuminate how genetic polymorphism, developmental plasticity, and intra-individual flexibility of migratory behavior influence the response, and ability of individuals, populations and species to cope with environmental change (**Figure 1**).

#### What's at Stake?

The variation and flexibility in migration strategies in fishes that move between habitats to fulfill competing needs provides a rich and fascinating example of how biological diversity manifests between and within communities, species, populations and individuals (Leggett, 1977; Roff, 1988; Lucas and Baras, 2001; Nathan et al., 2008; Mehner, 2012; Brönmark et al., 2014). As such, fish offer good model systems for investigating how biological systems respond to and cope with environmental heterogeneity and change. Being important predators, competitors, and prey to other species, migrating fish affect the functioning of lakes, rivers, coastal ecosystems and open oceans (Post et al., 2008; Brodersen et al., 2015; Donadi et al., 2017). In some areas, migrating fish represent important "vectors" by transferring nutrients or pathogens between habitats, as in the case of mass-migration and postspawning death of Pacific salmon that brings energy from resource rich marine habitats to less productive rivers (Naiman et al., 2002). Migrating fish also comprise an important resource of considerable socioeconomic value targeted by commercial and recreational fisheries throughout the world (Oke and Hendry, 2019).

#### What Are the Key Hazards to Migrating Fish?

Migrating fish are under threat by habitat modification, fragmentation and destruction of spawning and nursery habitats, pollution, and overexploitation (Waldman et al., 2016; Forseth et al., 2017). Apart from immediate negative effects associated with declining populations, changes in distribution ranges, and local extinctions (Dudgeon et al., 2006), exploitation can induce long-term evolutionary shifts in behaviors, individual growth trajectories and life-history strategies. These in turn may affect the recruitment, size-structure and dynamics of populations (Beacham, 1983; Kuparinen and Merilä, 2007; Uusi-Heikkila et al., 2008; Lowerre-Barbieri et al., 2017).

Perhaps counterintuitively, migrating fish are potentially also under threat by various management actions. Even efforts designed to compensate for overexploitation and mitigate the devastating effects of dispersal barriers via removal of dams, construction of fishways, compensatory breeding and supplemental stocking may have unintentional and unforeseen negative consequences. For example, alterations in river connectivity caused by the building and removal of dams or the construction of fishways may bring about changes in community composition and species interactions (Ngor et al., 2018), and in rapid loss of local adaptations (Thompson et al., 2019). Connectivity changes can also affect the directions and rates of gene flow with consequences for genetic diversity and interpopulation hybridization (Lynch, 1991; McClelland and Naish, 2007; Whitlock et al., 2013; Rius and Darling, 2014). Similar to fisheries induced evolution (Kuparinen and Merilä, 2007), the altered severity of migration caused by constructed fishways may influence the characteristics of successful migrants and impose selection and evolutionary shifts in traits that directly define migration or dispersal capacity, as well as in other traits that may impair population growth (as discussed and exemplified below).

Selection that gives rise to local adaptations generally reduces phenotypic and genetic variance. This can be detrimental because diversity brings many benefits. Theory and empirical evidence concur that flexibility and variance reducing bethedging strategies within individuals and genotypes can increase geometric mean fitness in changing and heterogeneous environments (Slatkin, 1974; Seger and Brockmann, 1987; Forsman et al., 2007). Earlier work unanimously show that among-individual variation contributes to improved establishment, more stable populations, and reduced extinction risk of populations and species, via complementarity and/or variance reducing effects (Hughes et al., 2008; Simberloff, 2009; Forsman, 2014; Forsman and Wennersten, 2016; Des Roches et al., 2018). Lastly, portfolio effects associated with variation among populations across environments or with high species diversity may increase stability, productivity and resilience of species and ecosystems (Schindler et al., 2010, 2015; Waldman et al., 2016; Hui et al., 2017; Lowerre-Barbieri et al., 2017).

Designing adaptive management for protection and sustainable utilization of migrating fish is complicated by ongoing climate change that brings changes in salinity, temperature, precipitation, sea surface levels, and species distribution ranges (Roessig et al., 2004; IPCC, 2013, 2018; Reusch et al., 2018; Cheng et al., 2019), thereby resulting in modified, novel, more variable and extreme selection regimes (Parmesan and Yohe, 2003; Root et al., 2003; Forsman et al., 2016b). The situation for migratory fish is worsened by the challenges (e.g., increased harvesting and habitat destruction) that accompany the increasing demands of a growing human population.

#### Questions Addressed in This Review

An important task for research is to investigate how the key hazards outlined above disrupt eco-evolutionary processes and the diversity of migrating fish. Scientific output on fish migration has grown tremendously from < 100 papers per year prior to 1970 to nearly 3300 papers in 2018 (**Figure 2A**). The portion of studies addressing aspects of variation and flexibility among and within populations or individuals is relatively low (< 10%), but this emerging field has increased 7-fold from < 50

papers per year prior to 1990 to > 350 papers per year in 2018 (**Figure 2B**). This growing appreciation of the potential importance of flexible migration strategies in fish is comparable to that in other organism groups (**Figure 2C**), and evident also relative to total research output (**Figure 2D**).Given the rich literature on variation and flexibility in fish migration (**Figure 2**) it is impossible to provide an all-inclusive summary of current knowledge, and there are already more than 300 reviews touching on various facets of this emerging area. Previous reviews typically focus on specific hypothesis, biomes, taxa, or migratory behaviors to summarize knowledge within a restricted area.

In this narrative review we provide a broad overview, in which we describe and discuss aspects of variation and flexibility in fish migration of basic scientific interest that concern key issues within the framework of evolution and maintenance of biological diversity (**Figure 1**). We consider key questions (whether or not to migrate, why to migrate, where to migrate, and when to migrate?) that define migrating fish and other organisms (Nathan et al., 2008), and exemplify how the evolutionary solutions to these questions may vary and change depending on ecological conditions, environmental settings, and individual characteristics. In particular, we explore links between environmental change, and migration strategies, and discuss whether and how threats associated with overexploitation, environmental makeovers, and management actions are likely to differently influence individuals, populations and species depending on the variation and flexibility of their migration strategies. In so doing, we aim to advance knowledge, spur future research and critical evaluation of management strategies, to ultimately promote sustainable utilization and protection

of migrating fish and their ecosystems. The disproportionate attention given to different subsections below reflects our subjective interests and concerns, not necessarily the relative importance or state of knowledge.

### VARIATION IN FISH MIGRATION–WHAT'S AT STAKE?

Fish migration encompasses a broad range of behaviors and life-history strategies by which individuals, populations and species cope with challenges associated with different scales of temporal and spatial environmental heterogeneity (**Figure 1**). The growing literature (**Figure 2**) has resulted in a rich flora of terms and phrases pertaining to various aspects of fish migration (Myers, 1949; Lucas and Baras, 2001; Secor and Kerr, 2009). The increasing interest in developmental plasticity and phenotypic flexibility has also been accompanied by numerous classifications and definitions (Piersma and Drent, 2003; West-Eberhard, 2003; O'Connor et al., 2014; Forsman, 2015; Senner et al., 2015). Below we provide a brief overview and reintroduce some definitions and key concepts related to variation and flexibility of fish migration.

### Definitions and Key Concepts in Fish Migration

Migration involves bi-directional large- or small-scale movements by individuals between habitats that fulfill competing needs that may occur within and between different life-stages. The habitats and resources that maximize growth, survival and reproductive success during different life history phases are typically separated in time and space (Gross et al., 1988). Migration is often interpreted as an adaptive response, although discriminating adaptive optimal migration solutions from "non-adaptive" movements induced by external or internal stressors can be difficult. Benefits from migratory movements may come in the forms of refuge from predators, access to resources, or strategic positioning of gametes in locations that offer advantageous conditions for the developing embryos and offspring. Potential costs include the energy expenditure associated with moving, predation risk, osmoregulation, erroneous navigation, and impaired reproductive success owing to genetic incompatibility associated with inter-population hybridization.

Migration tactics vary between species, among populations, and among individuals within populations. In "Migration of Freshwater Fishes," Lucas and Baras (2001) define migration as: "a strategy of adaptive value, involving movement of part or all of a population in time, between discrete sites existing in an ndimensional hypervolume of biotic and abiotic factors, usually but not necessarily involving predictability or synchronicity in time, since inter individual variation is a fundamental component of populations." However, the classification and understanding of fish migration is complicated further by an intra-individual component of variation, meaning that migration strategies can change also over an individual's life. Despite the extensive variability, some general migration patterns can be discerned.

#### Main Migration Modes

Fish migration modes can be described on the basis of the freshand salt water biomes used (**Figure 3**). These include holobiotic lifestyles, meaning that the fish spend their entire lifespan in either salt or fresh water, and amphibiotic lifestyles, meaning that the fish move between water bodies with different salinities (Lucas and Baras, 2001).

Oceanodromous fishes live and migrate wholly in the sea (Myers, 1949; Lucas and Baras, 2001). Well-known examples include small prey fish such as sardine (Sardina pilchardus), anchoveta (Engraulis encrasicolus), herring (Clupea harengus) but also larger fishes at higher trophic levels, pelagic species with wide distributions such as tuna, sailfish, marlin, swordfish, sharks, and rays that undertake variable but often long-distance migrations for feeding or reproduction.

Potamodromous fishes migrate between natal areas and feeding grounds entirely within fresh water. Although these fish typically migrate relatively shorter distances, these movements across habitats within freshwater may be just as important for survival, growth and reproduction as the typically larger scale migrations partaken by oceanodromous or diadromous species. There are also potamodromous species with extensive migrations; spawning migration distances of 300 km have been recorded for the endangered Colorado pike minnow of the Colorado River system (Lucas and Baras, 2001).

Diadromous fishes migrate between fresh and salt water environments to complete different parts of their life cycle (Lucas and Baras, 2001; Griffiths, 2006, 2010) (**Figure 3**). Catadromous fish spend the majority of the time feeding and growing in freshwaters and migrate into the saline sea water as adults to reproduce. Famous examples are the freshwater eels of the genus Anguilla, including the iconic European eel (Anguilla anguilla L.), which spawns in the Sargasso Sea and whose offspring

spend most time in freshwater but migrate to marine environments to reproduce, and as catadromous if they instead spend most time in the sea and migrate into freshwater to reproduce. The figure was created in Adobe Photoshop CC 2015 v. 16.0.1.

drift across the Atlantic Ocean to the coasts and freshwaters of Europe and North Africa where they will grow and mature, before returning to the Sargasso Sea to reproduce (Moyle, 2004; Aoyama, 2009). Another catadromous species is the Indo-Pacific barramundi (Lates calcarifer) that inhabits rivers before returning to the river mouths or estuaries to spawn, and where the larvae and juveniles live in the associated brackish temporary swamps (Russell and Garrett, 1983). Anadromous fishes spend the majority of the time feeding and growing in the sea and move into freshwater to reproduce. Well-known examples can be found among salmonids, such as Atlantic salmon (Salmo salar) that exploit the rich resources of the ocean as adults, only to return to the natal river or stream to reproduce. Additional examples include various species of Pacific salmon, striped bass (Morone saxatilis), and sea lampreys (Petromyzon marinus) (Moyle, 2004; Silva et al., 2014).

Not all fish species fall easily into the above categories. Species showing pronounced intraspecific variation include some salmonids (S. trutta), the three-spine stickleback (Gasterosteus aculeatus), and the northern pike (Esox lucius) in which different populations of the same species can be classified as belonging to at least two of the oceanodromous, potamodromous, and the anadromous lifestyles (Jonsson and Jonsson, 1993; Fleming, 1996; Lucas and Baras, 2001; Dodson et al., 2013; Forsman et al., 2015; Leitwein et al., 2016). There is also extensive variation in timing and distance of migration among species and populations (McDowall, 1997; Hendry and Day, 2005; Kuparinen and Merilä, 2009; Griffiths, 2010; Seamons and Quinn, 2010; Kovach et al., 2015; Forsman and Berggren, 2017; Bloom et al., 2018).

#### Spawning Migration

While fish migration takes countless shapes and can be described based on utilization of different biomes (**Figure 3**), it is sometimes fruitful to analyze and classify them from a functional viewpoint. In principle, the main drivers of large scale fish migrations are to reproduce, find food, and avoid enemies. Although any habitat shifts must be interpreted as representing the outcome of these competing needs, fish migrations are typically classified based on reproduction.

Spawning-, reproductive- or breeding migrations involve the movements of reproductively mature fish from foraging areas to a location where they will place their gametes. For a spawning environment to be productive, it should provide abiotic and biotic conditions that are favorable for the development and survival of fertilized eggs, embryos, hatched larvae, and young juveniles (Lowerre-Barbieri et al., 2017) (**Figure 1**). Because of differential needs and demands depending on size and age, the nursery habitat progressively becomes suboptimal. As the fish grow larger and older, they eventually leave the nursery grounds in favor of more productive foraging grounds where they likely join the adult population. Spawning migration may involve the crossing of the borders between fresh, brackish, and saline water bodies, but can occur within such biomes, for example between or within lakes and rivers (**Figures 3**, **4**).

TABLE 1 | Overview of potential correlates and putative internal (left column) and external (right column) drivers of variation and flexibility in migration behavior in fishes.


#### Homing Behavior and Navigation

Some fish display homing behavior. After having reached maturity, the adults may return to spawn where they were born. In iteroparous species, the adults may reuse the same spawning grounds for multiple reproductive cycles (e.g., Tibblin et al., 2016b). Homing is not an obligatory part of fish migratory behavior (Lucas and Baras, 2001). However, it can allow for evolution of genetic structure, local adaptations, and divergence of early life-history traits among subpopulations that use different spawning areas, and thereby reinforce the benefits of homing (Jensen et al., 2008; Kavanagh et al., 2010; Petersson, 2015; Tibblin et al., 2015, 2016a; Berggren et al., 2016; Mäkinen et al., 2016; Sunde et al., 2018a). This showcases how varying environmental conditions and behaviors can shape biodiversity even on small spatial scales.

The mechanism(s) involved in navigation, identification and habitat recognition that allow for homing behavior in fish have been reviewed elsewhere (Lucas and Baras, 2001; Odling-Smee and Braithwaite, 2003; Keefer and Caudill, 2014; Petersson, 2015). Receptors sensitive to electric and geomagnetic fields, light, temperature, olfactory and visual cues together with information based on landmarks, water flow, and sound seem to be involved to various degrees by different species (Lucas and Baras, 2001; Keefer and Caudill, 2014).

Below, we illustrate how migratory behavior may vary among and within species of fish (**Table 1**). We also exemplify how variation and flexibility in migratory behavior may be associated with, and possibly depend on, spatiotemporal environmental heterogeneity and vary according to individual characteristics (**Table 1**; **Figure 1**).

#### Variation Among Species

Patterns and strategies of migration vary extensively among species with regards to function (e.g., spawning, feeding, and refuge from predators and other environmental stressors), migration mode (diadromous, potamodromous and oceanodromous), mode of parity (semelparous-iteroparous), timing of migratory events (phenology), and migratory distance (McDowall, 1997; Griffiths, 2010; Seamons and Quinn, 2010; Mehner, 2012; Forsman and Berggren, 2017; Nilsson et al., 2019). As for diadromy, inter-specific comparisons have uncovered that anadromous species predominate in temperate latitudes where productivity in freshwater is generally lower than in marine environments whereas catadromy dominates in tropical latitudes where the highest productivity is found in freshwater habitats (Gross et al., 1988; McDowall, 1997). Similarly among potamodromous fish, many species utilize rivers as spawning and nursery grounds whereas foraging occurs in more productive areas such as lakes. Drivers other than productivity are also important in shaping the mode or direction of migration in fish (Bloom and Lovejoy, 2014). Recent evidence from a comparative analysis indicate that across Clupeiformes (anchovies, herring, shad and allies) diadromous species are larger than nondiadromous species, whereas no association was found with trophic position (Bloom et al., 2018).

With regards to migration timing, Kovach et al. (2015) report that temporal trends in the direction of the shift in the median migration date, as well as in duration and inter-annual variation in migration timing are highly variable across species and populations of Pacific salmon. The drivers resulting in the diversity of migration strategies seen across fish species are poorly understood but presumed to be the result of improved access to resources in a patchy system or decreased predation. With regards to distance, pike, and goliath catfish (Brachyplatystoma rousseauxii) offer an example of a striking difference in freshwater migration distance between anadromous species. Both species inhabit coastal estuarine areas and migrate to spawning locations upstream. For pike, maximum migration distance in freshwater is probably < 50 km (Larsson et al., 2015), whereas the goliath catfish that spawns in the western Amazon travels 11,500 km, the longest fish freshwater migration in the world (Barthem et al., 2017). In the catadromous Anguilla eels, migration distances from freshwater to the marine spawning areas range from 750 to > 8,000 km (see Table S1 in Forsman and Berggren, 2017). Results from a comparative analysis indicate that the evolutionary increments of migration distances in Anguilla have been accompanied by shifts in dispersal enhancing phenotypic traits, such as larger body size (Forsman and Berggren, 2017). Phylogenetic comparative analysis also point to a role of thermal biology for migration distance. Watanabe et al. (2015) showed that species that are able to maintain red muscles warmer than ambient temperatures swim faster and have longer annual migration distances compared with similar sized species of fish without red muscle endothermy (i.e., the vast majority of fishes).

Even though different species may share the same modes of migration there can be differences in where and when alternative strategies (residents and anadromous) are sympatric or allopatric. For example, salmonid and esocid species are phylogenetically relatively close (Rondeau et al., 2014), and their resident and anadromous populations are partially sympatric (Craig, 1996; Fleming, 1996; Quinn, 2005; Jonsson and Jonsson, 2011; Skov and Nilsson, 2017). However, salmonids are sympatric during spawning and early life-stages in the recruitment habitat, whereas in esocids the resident and anadromous phenotypes are sympatric during the adult life-stage in the foraging habitat (Engstedt et al., 2010; Forsman et al., 2015; Tibblin et al., 2015).

Insights about the causes and consequences of migration behavior in fishes might be gained by studying species and populations that do not migrate, or do so to a lesser extent. This opens for phylogeny based comparative approaches (Felsenstein, 1985) that may inform about large scale evolutionary dynamics of migration behavior in fishes. Although tedious to perform, the compilation and analysis of data within a phylogenetic framework may pay dividends in the long run. For example, such approaches may uncover how migratory behavior data deficiency is distributed across and within different clades of fishes, and thereby help identify taxa and geographic regions in particular need of further investigation. Given a sufficient number of independent evolutionary modifications, phylogenetic comparative approaches can help identify why certain species migrate whereas others do not (McDowall, 1997; Bloom et al., 2018). Apart from uncovering associations of migration behaviors with external environmental factors, there is potential for phylogeny based comparisons to reveal whether evolutionary shifts in migration have been accompanied by correlated modifications of morphological, physiological, or behavioral phenotypic dimensions (Watanabe et al., 2015; Forsman and Berggren, 2017; Bloom et al., 2018) (**Figure 1**).

The diversity of migration behaviors among species outlined above is impressive. However, identifying generality is complicated by the extensive variation seen also within species.

#### Variation Among Populations

There is considerable variation in spawning migratory patterns among populations within species (Jonsson and Jonsson, 1993; Griffiths, 2006; Dodson et al., 2013). Different populations have different evolutionary histories, and are exposed to different selection pressures depending on the environment they inhabit (Berggren et al., 2016; Sunde et al., 2018a). Accordingly, different populations can adopt different migration tactics and, in some cases, display a level of variation that is comparable to that observed between species. In several salmonid species (e.g., S. salar, S trutta, O. mykiss, O. tshawytscha, and Salvelinus alpinus) populations differ in migration mode (ranging from anadromous, potamodromous to residents in either streams or lakes) (Jonsson and Jonsson, 1993; Fleming, 1996; Lucas and Baras, 2001; Dodson et al., 2013; Leitwein et al., 2016), and the number of migratory events vary according to mode of parity (Unwin et al., 1999; Narum et al., 2008; Seamons and Quinn, 2010; Dodson et al., 2013). This large-scale variability at the population level has been attributed to life-history evolution being shaped by stage-specific mortality and resource availability (McDowall, 1997; Kindsvater et al., 2016). Variation in migration modes among populations has also been documented in cyprinids, esocids, gasterosteids, gadids, and percids (Nordahl et al., (in press); Lucas and Baras, 2001; Tibblin et al., 2012).

Populations commonly vary also in the timing and distance of migratory events (Kinnison et al., 2001; Hodgson and Quinn, 2002; Quinn, 2005; Kuparinen and Merilä, 2009; Kennedy and Crozier, 2010; Jonsson and Jonsson, 2011). This has been suggested to reflect in part phenotypic flexibility (Forsman, 2015) in response to environmental conditions (e.g., temperature, precipitation, light regime and water flow) along latitudinal and altitudinal gradients and local climate (Hodgson and Quinn, 2002; Dodson et al., 2013), but a growing body of evidence suggests that genetic components are also involved (Skov et al., 2010; Plantalech manel-la et al., 2011; Kovach et al., 2012; Thompson et al., 2019). Crossin et al. (2004) showed that migratory distance of populations of sockeye salmon (O. nerka) within the Fraser river ranged from <100 km to >1,100 km, and that the severity of migration (distance and elevation) was associated with higher densities of somatic energy and a more fusiform, streamlined body shape. A similar pattern has been documented in roach (Rutilus rutilus) with migratory populations having a more slender body shape than resident ones (Chapman et al., 2015).

An important task for future research is to determine whether the occurrence of populations with different migration strategies within a species buffers against environmental challenges. Predictions from theory, evidence from manipulation experiments, and results from comparative analyses concur that populations and species with higher phenotypic and genetic diversity are better able to cope with environmental changes and less extinction prone (Hughes et al., 2008; Bolnick et al., 2011; Wennersten and Forsman, 2012; Forsman, 2014, 2015; Forsman and Wennersten, 2016). However, it has not yet been systematically investigated whether these benefits apply also to variation and flexibility of migratory behavior in fishes. To achieve this, information on migration behaviors must first be compiled for multiple populations and species. The large number of studies of variation and flexibility in fish migration identified by our literature search (**Figure 2**) opens for such future systematic reviews and for meta-analytical approaches that can be used to summarize information, identify patterns, and evaluate potential drivers of variation in migration mode, migration timing, and migration distance among populations (Gurevitch et al., 2018). Results from such endeavors may also help identify the need for and inform population specific management and conservation efforts.

#### Variation Among and Within Individuals

Variation in migratory behaviors among individuals within populations can also provide insights into the underlying mechanisms and functional significance of migration (Wilson, 1998). Spawning migrating and resident phenotypes sometimes coexist within the same population, a population level phenomenon called partial migration (Brodersen et al., 2007; Chapman et al., 2011a; Dodson et al., 2013; Brönmark et al., 2014). Such partial migration may offer good opportunities to study both the causes and consequences of migration, and suggests that sometimes not migrating is adaptive for an individual in an otherwise migratory population, and further that partial migration is an evolutionary stable strategy. Whether individuals chose to migrate or not is influenced by numerous interacting environmental variables (e.g., resource availability, predation risk, water flow and temperature) and individual characteristics (e.g., growth rate, size, age, lipid content, life history stage, personality, and previous reproduction efforts), as well as by genetic variation in the sensitivity to the external and internal cues (Chapman et al., 2011a,b; Skov et al., 2011; Dodson et al., 2013; Brönmark et al., 2014) (**Table 1**; **Figure 1**). For example, Olsson et al. (2006) showed that migration could be environmentally induced by translocating individuals between two habitat patches that differed in density and opportunities for individual growth.

In iteroparous species that engage in multiple migratory spawning events there is potential for phenotypic flexibility (Forsman, 2015), such that individuals change and modify their migratory behavior (Brodersen et al., 2014). Intra-individual flexibility in migratory behavior has recently received increased scientific attention (**Figure 2**), especially in birds. Evidence is accumulating that flexibility is key to cope with the challenges associated with anthropogenic impacts such as climate change and exploitation (Arnaud et al., 2013; Winkler et al., 2014). Yet, individual flexibility in migratory behavior and timing of fish remains largely overlooked (Tibblin et al., 2016b). Studies of roach, an iteroparous species that displays partial migration, suggest that individuals are consistent rather than flexible across years in whether to migrate or not, thus implying that residency and migration can be stable strategies (Brodersen et al., 2014). This consistency can either be attributed to genetic differences or to initial plasticity, possibly caused by variation in somatic condition, followed by canalization with the latter gaining some support in the roach system (Brodersen et al., 2014).

With regards to phenotypic correlates of timing of spawning migration (**Table 1**; **Figure 1**), a common pattern is that males migrate, and subsequently arrive in the spawning habitat, earlier than females (Morbey and Ydenberg, 2001; Tibblin et al., 2016b), possibly because males strive to maximize the number of mating opportunities. Migratory timing may also be associated with body size. Larger size is associated with early migration in both juvenile and adult life-stages of salmonids (Heim et al., 2016; Jonsson et al., 2017), but Tibblin et al. (2016b) report the opposite pattern in pike. Reversible phenotypic flexibility can be selected for and evolve in environments that change throughout an individual's lifetime. Models predict that organisms that are longlived relative to the rate and frequency of environmental changes should be more flexible, compared with short-lived organisms. It has been suggested that causality may be bidirectional because flexibility itself may select for longevity. Simulation models suggest that under highly auto-correlated environmental fluctuations, reversible flexibility should coevolve with lifespan (Ratikainen and Kokko, 2019). To our knowledge, it has not yet been investigated whether reversible flexibility in migration strategies is more common in long-lived species of fish.

Besides the long-term and often larger scale seasonal migratory movements between areas used for breeding and nonbreeding purposes, many fish engage in migrations at smaller spatial, and temporal scales (Lucas and Baras, 2001; Mehner, 2012). Daily migratory movements for utilizing reoccurring and predictable windows of available resources and favorable conditions are particularly common. Many marine-, brackish-, and freshwater fish show such diel vertical migrations, rising to the surface to feed during night and diving to deeper layers during the day (reviewed in Lucas and Baras, 2001; Mehner, 2012). Some species instead utilize the near surface waters during the daytime and return to bottom layers in the evenings to feed. Other proximate triggers of vertical migrations include light intensity and water temperature, and ultimate drivers encompass bioenergetics efficiency, foraging opportunities and predator avoidance (Mehner, 2012; Nordahl et al., 2019). Horizontal fish migrations include movements between shallow, inshore littoral areas and offshore pelagic areas performed by fishes in larger lakes. Such horizontal movements are often cyclical on a daily basis, with shifts from offshore to inshore areas at night, or in the reverse direction. It is generally believed that such rhythmical diel shifts are driven by a trade-off between foraging and avoiding being fed upon (Lucas and Baras, 2001; Mehner, 2012).

Migrating between water bodies also offers a means to buffer against changing external physicochemical conditions, maintain internal homeostasis and regulate body temperature to conserve energy expenditure or to maximize aspects of performance (Reynolds and Casterlin, 1980; Nakamura et al., 2015; Pépino et al., 2015; Nordahl et al., 2018, 2019). Observations of diel horizontal migrations in juvenile coho salmon (Oncorhynchus kisutch) indicate that individuals that moved to warmer habitats after feeding processed their food more quickly and grew faster compared with individuals that adopted other behaviors (Armstrong et al., 2013). A behavioral study of pike has shown that individuals surface during daytime and seek out deeper waters during night in the summer, whereas the direction is reversed during winter, thus pointing to a possible role of sun basking (Nordahl, 2018; Nordahl et al., 2019).

A recent study of carp (Cyprinus carpio) demonstrates that sun basking close to the surface during sunny conditions enables fish to increase their body temperature above that of the ambient water, and further indicates that the temperature excess gained by basking enabled the fish to grow faster (Nordahl et al., 2018), thereby putting the individual in a favorable situation compared to those not expressing this behavior. The discovery that sun basking can offer thermoregulatory benefits even in aquatic environments (Nordahl et al., 2018, 2019) is likely to spur future research and may ultimately change the way we think about fish ecology and evolution, in particular with regards to behaviors and migrations.

Longitudinal studies have contributed with knowledge regarding how migratory behavior may be modified in response to environmental cues (**Table 1**; **Figure 1**). Forsythe et al. (2012a) and Forsythe et al. (2012b) studied associations between external factors and individual timing of spawning migration in lake sturgeon (Acipenser fulvescens) across 8 years and showed that individuals adjust their timing according to lunar cycle, water flow and temperature. These last results might be interpreted as an indication that flexibility is adaptive, but firm evidence to that effect is scarce, mainly for logistical reasons (Forsman, 2015). However, a recent study of pike migratory behavior has shed some light on this matter. Data on recapture rates of pike suggests that the timing of arrival to the spawning area is under stabilizing viability selection, and that individuals that are more flexible in their timing during the 1st years survive longer compared with less flexible individuals (Tibblin et al., 2016b). Besides extensive research on how abiotic cues influence migratory timing it has been proposed that timing may be modulated by social interactions. Work by Berdahl et al. (2017) suggests that migratory timing in sockeye salmon was better explained by social interactions (group migration) than by abiotic cues such as temperature and river flow. Environmental influences aside, there is evidence emerging that timing can be under genetic control and undergo rapid evolutionary change (Thompson et al., 2019). There is also potential for variation among individuals in the timing of spawning migration to contribute to population genetic structure; isolation by time rather than isolation by distance (Hendry and Day, 2005). Whether isolation by time is a common driver of genetic divergence and adaptation in fish, and whether differences in the timing of spawning migration contributes more or less to population structure in different species depending on their lifehistory (e.g., discrete or overlapping generations) remains to be investigated.

# Phenotypic Correlates of Migratory Performance

A plethora of studies have aimed to identify phenotypic correlates of swimming performance and the evolution of adaptations facilitating migratory behavior. Collectively, results point to important roles of morphological (e.g., body size, body shape, number of vertebrae, spool width, and size and shape of fins) and physiological traits that influence aspects of performance (e.g., swimming capacity, acceleration, endurance, and ability to sustain high water velocities), and of behavioral (boldness, and latency to pass obstacles) traits (Webb, 1975; Swain, 1992; McDowall et al., 1994; Fleming, 1996; McDowall, 2003; Crossin et al., 2004; Haugen et al., 2008; Long et al., 2011; Chapman et al., 2015; Podgorniak et al., 2016, 2017; Tibblin et al., 2016a; Forsman and Berggren, 2017; Hall, 2018; Aguirre et al., 2019). There is also potential for indirect evolutionary responses of phenotypic dimensions that are genetically or developmentally correlated with dispersal enhancing traits (see "Construction of Fishways").

### On Genetic Polymorphism, Developmental Plasticity and Phenotypic Flexibility

The differences in migration behaviors, or any other traits, between species, populations, and among individuals within populations discussed above may be seen as representing the combined outcomes of underlying genetic polymorphisms, developmental plasticity and phenotypic flexibility (Piersma and Drent, 2003; West-Eberhard, 2003; O'Connor et al., 2014; Forsman, 2015; Senner et al., 2015). The concept of phenotypic plasticity is deceptively simple, and has been previously defined in numerous ways by different authors [see for instance Box 1 in Whitman and Agrawal (2009)]. The consequences of plasticity and flexibility for the performance and success of individuals, populations and species continue to attract a growing interest (see Figure 1 in Forsman, 2015). Here, we distinguish between irreversible developmental plasticity and reversible intra-individual phenotypic flexibility (Piersma and Drent, 2003; Forsman, 2015).

Developmental plasticity is used primarily for irreversible phenotypic variation in traits of individuals (or genotypes) that result from environmentally induced modifications of development and growth (Stearns, 1989). Developmental plasticity can also involve mechanisms that operate across generations. When the phenotype is induced by the female parent, the plasticity is usually referred to as maternal effects (Roff, 1997; Mousseau and Fox, 1998). Cross generational plasticity can also be mediated by the male parent (e.g., Kekalainen et al., 2018).

Phenotypic flexibility is used for reversible changes within individuals of labile, context-dependent physiological, morphological, or life-history traits (Piersma and Drent, 2003; Forsman, 2015). It is applicable also to behavioral traits, for instance as a result of previous history, learning, and experience, or adjustments to external conditions that influence current responses and behaviors in given situations (Dingemanse et al., 2010; Tuomainen and Candolin, 2011; Snell-Rood, 2013).

Plasticity and flexibility are not fundamentally distinct from genetic polymorphisms (Leimar et al., 2006; Forsman, 2015). Crossing norms of reaction, when different genotypes display different phenotypic responses to environmental change, are manifestations of underlying genetic polymorphisms (Pigliucci, 2001; West-Eberhard, 2003). It is often difficult to disentangle the contribution of genetic and non-genetic sources of variation. Demonstrations of trait heritability alone do not provide conclusive evidence that differences among individuals or populations have a genetic basis. Conversely, failure to demonstrate a role of developmental plasticity for a given trait in response to a given environmental factor does not necessarily imply that the trait is insensitive also to other factors.

As we have seen, variation in fish migratory behaviors manifests at different hierarchical levels and at different spatiotemporal scales, and can be of genetic and/or environmental origin. In the following sections, we discuss how this may contribute to the viability of species and resilience of ecosystems. Safeguarding against key hazards requires management actions that do not raze, but promote variance-coping mechanisms. Unfortunately, management and conservation actions aimed to mitigate the devastating effects of key hazards for migrating fish can themselves disrupt natural processes and threaten biodiversity, as discussed below.

#### KEY HAZARDS AND HOW THEY DISRUPT THE NATURAL PROCESSES THAT UNDERLIE DIVERSITY

The environmental heterogeneity that has shaped evolution of fish migration behaviors is modified by anthropogenic activities and climate change. Threats associated with such makeovers, overexploitation and management actions may differently influence individuals, populations and species depending on their migration strategies (**Figure 1**). The variance reducing effects that diversity at different hierarchical levels of biological organization have on productivity (see Introduction for references) must inform decision making regarding utilization and protection of migratory fish and the ecosystem services they provide.

# On the Roles of Exploitation, Environmental Makeovers, and Management Actions

#### Dams and Hydroelectric Power Plants

Damming is a major threat to migratory fish, biodiversity, and ecosystem functioning. Damming is one of the most widespread environmental alterations of river ecosystems, affecting about half of all large river systems globally (Nilsson et al., 2005; Grill et al., 2015). Consequences include habitat fragmentation, loss and degradation, and changed hydrological regimes. Fragmentation resulting from damming in rivers is particularly troublesome because aquatic organisms are limited to linear pathways and cannot find another route unless one is provided. River systems comprise diverse communities of fish with many migration modes, partaken on different spatial and temporal scales and between different habitats (**Figure 4A**). Dams and other obstacles reduce river connectivity and hinder both small and large migratory movements for most species (**Figure 4B**). Although likely to be more common than recorded in the scientific literature, there are examples indicating that dams and inability to migrate cause local extirpations of populations (Winston et al., 1991; Holmquist et al., 1998; Morita and Yamamoto, 2002; Locke et al., 2003). Obstacles can potentially also constrict larger scale migrations such as poleward or altitudinal range shifts that many species are undertaking to evade effects of climate change (Comte and Grenouillet, 2013).

When connection between freshwater and marine habitats is removed, the persistence of anadromous species depends on whether they can switch to a more resident strategy. Species that would have utilized the ocean as foraging grounds but gets landlocked may change to a freshwater resident behavior or disappear from the freshwater system altogether. Such switches may lead to evolutionary divergence. For example, comparisons of juvenile alewives (Alosa pseudoharengus) have shown that anadromous life history forms are more robust compared with fish in landlocked freshwater resident populations that have a more fusiform body shape, pointing to a parallel divergence mediated by shifts in zooplankton prey (Jones et al., 2013). Catadromous species that utilize freshwater habitats as foraging and nursery grounds may get locked out in the ocean and extirpated from inaccessible freshwater systems (Harris et al., 2016). Potamodromous species are also affected by migration barriers (Branco et al., 2017) (**Figure 4B**) as most species migrate between habitats used for growth, survival or reproduction (Lucas and Baras, 2001). Fish with flexible migration strategies are likely more persistent during such severe environmental makeovers as they may adjust migratory behaviors to novel regimes.

#### **Habitat Fragmentation, Conversion and Loss**

Obstacles can be definitive or partial dispersal barriers, depending on the severity of the obstacle and the swimming capabilities of the fish. Naturally, obstacles are often harder to traverse in the upstream direction, while weirs and spillways may allow for some downstream dispersal. This unidirectional dispersal constricts gene flow in the upstream direction and reduces genetic diversity in the upstream direction (Gouskov et al., 2016; Van Leeuwen et al., 2018). Small populations upstream of dams, with no possibility for immigration or recolonization from downstream populations, may also be extirpated (Morita and Yamamoto, 2002). Depending on the severity of the upstream and downstream barriers, movement becomes restricted and gene flow between fragments reduced. This can lead to population differentiation among fragments and manifest as local population structures between barriers (Van Leeuwen et al., 2018). Reductions in the number, size, and type of available habitats (**Figure 4C** vs. **Figure 4D**) will reduce the size of the local populations that can be sustained between barriers, with consequences for genetic diversity, divergence, and viability of populations (Carim et al., 2016).

Inundation, the creation of reservoirs upstream dams (**Figures 4B,D**), can impact river communities (Geist, 2011) and cause a shift from lotic to lentic fish assemblages.

inhabiting the system. Fish migrate (a) between lakes and rivers, (b) between larger and smaller parts of the river, (c) between sea and lakes, and (d) between the sea and river. (B) When dams (e) are added to a river system, previous migration routes become disconnected and the total amount of freely available habitat fragments becomes smaller. Damming structures create impoundments (f) that store water, converting lotic habitats to lentic habitats. Connectivity can be partially restored by adding fish passages (g) that enable fish to pass obstacles, but the impoundments created upstream of damming structures persist even though fish passages are built. The height profiles (C,D) drawn from point X to point Z illustrate how damming changes the large-scale structure of a river system to a series of steps. The potential migration length for freshwater fish in the river system is severely shortened, limiting access to areas that may provide better opportunities for growth, survival, or reproduction for fish.

Impoundments upstream of damming structures persist despite attempts to restore connectivity through fish passage solutions (**Figures 4B,D**), and the lentic habitats created upstream can in themselves continue to pose large migratory challenges (Jepsen et al., 1998; Olsson and Greenberg, 2004). As a consequence of complete or partial conversion of lotic to lentic habitats by inundation (**Figure 4**), lotic habitats also become less frequent and spaced further apart (Aarts et al., 2004), reducing available suitable habitats for species that depend on running waters. Local populations whose structure and temporal dynamics is governed by meta-population processes may be particularly sensitive to river fragmentation (Rieman and Dunham, 2000) because damming increases isolation of "islands."

#### Construction of Fishways

Management actions to alleviate the negative impacts on migrating fish of impaired connectivity, river fragmentation and habitat destruction discussed above include the fitting of fauna passage solutions to damming structures and compensatory breeding, both of which may also have undesirable consequences. A fishway is a type of passage that, usually, consists of engineering solutions that reroute part of the water around obstacles to offer an alternative migration route and "free passage" for the fish (Birnie-Gauvin et al., 2018), a goal that is practically unreachable because fishways themselves entail a barrier of sorts. A more grounded goal would be that fishways should enable a wide range of genotypes and phenotypes to pass, such that populations can maintain their evolutionary potential. The innate tradition of fishway retrofitting to avoid negatively affecting the damming structures or the hydroelectric power production generally results in compromised designs and a performance that is suboptimal.

Although fishways improve possibilities for spawning migration (Gouskov et al., 2016; Tamario et al., 2018), they seldom result in the desired level of connectivity restoration (Brown et al., 2013; Foulds and Lucas, 2013; Birnie-Gauvin et al., 2018; Silva et al., 2018; Tamario et al., 2019) and are not fully and equally permeable for all species, ages and phenotypes (Haugen et al., 2008; Volpato et al., 2009; Birnie-Gauvin et al., 2018). The altered severity of migration caused by fishways, and other types of partial dispersal barriers (Newton et al., 2018), may thus impose selection by favoring certain phenotypes and disfavoring others, and thereby impact on the phenotypic and genetic composition (**Figure 5**). Fishways that are harsh to traverse may cause size selection with evolutionary consequences similar to that of size-selective fishing, with average size and variation in sizes decreasing over time (Haugen et al., 2008; Maynard et al., 2017). The loss of phenotypic diversity can be surprisingly rapid and observable over just a few decades (Haugen et al., 2008). Similarly, fish passage solutions for eels usually consist of ramps lined with a homogeneous climbing substrate that may favor the sinusoid movements and climbing performance of eels of a certain size (Podgorniak et al., 2017). Podgorniak et al. (2017) report that eels upstream of fish passage solutions showed less variation in size than eels below, and that different climbing substrates may select for different sizes. Such climbing substrates vary widely in their efficiency (Watz et al., 2019), and ignorance of best technical solutions in management likely leads to reduced fishway performance, stronger selection, and higher culling of variation.

Selection on migratory performance may have evolutionary consequences that extend beyond the phenotypic dimensions that directly influence migratory capacity. This is because phenotype sorting on dispersal enhancing traits may result in indirect correlated responses and induce evolutionary transitions in morphological, physiological, behavioral, and reproductive life-history traits that are developmentally, functionally or genetically associated with the traits directly involved in migration and dispersal (Lande and Arnold, 1983; Roff, 1997; Walsh and Lynch, 2012), and that may impair population growth. For example, a study on the effect of body length and arrival timing on reproductive success in wild pink salmon (O. gorbuscha) indicated that these traits are under stabilizing selection (Dickerson et al., 2005). Therefore, if small size enhances the ability to overcome migration obstacles [as in Maynard et al. (2017); Newton et al. (2018)], this might not only impact the evolutionary trajectory for body length but also population productivity. Changes in the severity of migration may thus have consequences similar to fisheries induced evolution (Kuparinen and Merilä, 2007).

Because fishways and other partial barriers can be difficult to find and pass through, migrating fish may be delayed (McLaughlin et al., 2013; Newton et al., 2018). Longer delays may lead to aggregations that promote disease transmission, create predatory hotspots, and leave individuals with less energy available for reproduction (McLaughlin et al., 2013). Tagging studies suggest that low attraction is often a limiting factor (Dodd et al., 2017), partly because fish rely on water flow dynamics as a cue to initiate upstream migration and to find the fishways (Hall, 2018). Mismatches between fishway operation (van Leeuwen et al., 2016) and the evolved migratory timing may have consequences for both individuals (e.g., late arrival, suboptimal conditions for breeding, not finding a partner) and populations (loss of adaptation of migratory timing) (Dickerson et al., 2005). For example, populations may become reproductively isolated by utilizing the same spawning grounds at different times (Quinn et al., 2000). Delays associated with passing of fishways can potentially cause admixture between temporally isolated subpopulations.

As for recommendations, there should be less focus on the number of fish passing, and more focus on maintaining diverse and viable fish populations (Birnie-Gauvin et al., 2018; Silva et al., 2018). Designing optimal fish passage solutions is complicated by the differential demands of different species, life-history stages and phenotypes (Birnie-Gauvin et al., 2018). There is a growing concern that fishways may relax selection or select for phenotypic dimensions, such as certain life-stages or sizes (Haugen et al., 2008; Maynard et al., 2017) or boldness (Cote et al., 2010), and trait value combinations that are different from those that are beneficial in un-manipulated water courses (Newton et al., 2018) (**Figure 5**). When deciding on the design and operation of fishway passages, it is important to consider that selectivity may apply to each of the approach, entry and passage components, as well as to post-passage behaviors and performances (Silva et al., 2018). The phenotypic and genetic structure of fish populations may be further influenced by selection operating on individuals as they embark on the downstream journey to complete their life-cycle in the lake or sea. We have in mind the risky and often deadly passage through the created impoundments as well as turbines of hydroelectric power-plants (Jepsen et al., 1998; Calles et al., 2010). If the phenotypic trait values that are favored by selection on juveniles during this downstream passage are different from those that are favored in spawning migrating adults during the upstream journey then this will magnify the variance reducing effect (comparable to stabilizing selection), which can detrimentally impact long-term population persistence. Perhaps the key question regarding connectivity restoration is whether the persistence of the dam or migration barrier is at all defendable, and whether it can be removed instead of installing inherently imperfect fishways? With barrier removal comes also the complex issue of how the capacity for re-colonization and range expansion may vary among species depending on migratory behavior and life-history characteristics (Pess et al., 2014), and the possible establishment of invasive migratory species, such as the sea lamprey, that may disrupt local communities (Smith and Tibbles, 1980; McLaughlin et al., 2013).

#### Captive Breeding, Supplemental Stocking and Aquaculture

The release of captive reared fishes might be seen as a quick and feasible fix for declining wild fish stocks to compensate for overfishing and losses due to dam construction (Hórreo, 2015), but it does not come without problems. Releasing large numbers of captive-bred fishes might expose wild fish populations to elevated competition and predation, and it can do so even if the stocked fish do not reproduce in the wild, as exemplified by escapes of farmed S. salar in Norway (Anonymous, 1999). The escaped farmed fish have low reproductive success (Fleming et al., 1996), and probably do not replace what they destroy neither in

numbers nor quality of offspring. An example from the North American west coast further indicates that the consequences of stocked fish may vary according to environmental conditions. Levin et al. (2001) report that the productivity of wild Chinook salmon (O. tshawytscha) was affected by the interaction between ocean conditions and the number of stocked hatchery spring chinooks. Nickelson (2003) reports on a similar negative relationship between hatchery spawners and wild productivity in coho salmon (O. kisutch).

Captive breeding has the advantage over wild reproduction that fewer parental fishes are needed for producing a certain number of juveniles of a certain age. However, captive breeding programs rarely use a sufficient number of breeding individuals, and studies indicate that the genetic variation declines in populations exposed to repeated captive breeding (Hansen et al., 2001; Säisä et al., 2003) thereby reducing their performance and adaptability in the wild (Araki et al., 2007). In addition, released captive reared fish, and escapers from aquaculture cages, may interbreed with wild stocks and result in genetic admixture.

#### Genetic Admixture

Migration behavior may result in reproductive interactions between fishes from different populations. Mixing of previously separated gene pools, admixture (Lynch, 1991), can occur both between species and between populations within species. Intraspecific admixture may be a natural outcome of dispersal and non-natal adult straying (Keefer and Caudill, 2014). It can also result from anthropogenic activities, including management actions aimed at protecting biodiversity, such as removal of migration barriers, installation of fishways, compensatory breeding, supplementary stocking, and translocations (Gjedrem et al., 1991; Berg et al., 1997; Søndergaard et al., 2000; McClelland and Naish, 2007; Seddon et al., 2007; Frankham, 2008; Service USFW, 2012).

Admixture will increase the genetic diversity in the receiving population, but fitness consequences can vary from positive to negative. By contributing new alleles and enabling creation of novel genotypes and haplotypes, admixture can counteract inbreeding depression, conceal deleterious recessive alleles, and result in heterosis (Lynch, 1991; Fenster and Galloway, 2000; Keller and Waller, 2002; Facon et al., 2005; Drake, 2006; Lavergne and Molofsky, 2007; Weeks et al., 2011). Conversely, the introduction of new genetic material can dilute favorable alleles, break up co-adapted gene complexes (Lynch, 1991; Rhymer and Simberloff, 1996; Fenster and Galloway, 2000; Edmands, 2007; Verhoeven et al., 2011; Whitlock et al., 2013) and reduce fertility and offspring viability (Gilk et al., 2004; Turner et al., 2012; Sunde and Forsman, 2016), thereby impairing population performance (Fleming et al., 2000; McGinnity et al., 2003).

The outcome of admixture affects both the genetic diversity within populations and genetic differentiation between populations, which might have consequences for the viability, and adaptability of the populations and species (McGinnity et al., 2009). From a management perspective it is therefore problematic that the direction and magnitude of responses to admixture can differ between species (Hardiman and Culley, 2010; Molofsky et al., 2014; Rollinson et al., 2014), among populations within species (Escobar et al., 2008; Tortajada et al., 2010; Hufford et al., 2012; Sunde and Forsman, 2016; Tinnert et al., 2016; Shi et al., 2018), and even vary depending on the sex of the immigrants (Sunde et al., 2018b). That the effects of admixture can be sex-specific (Sunde et al., 2018b) might impact on dispersal behavior; if the responses to admixture depend on the sex of the immigrant, it is likely that the impact on spawning migratory behavior may also differ between the sexes. Predicting the outcome of admixture is further complicated by that responses can differ also between generations (Huff et al., 2011; Tinnert et al., 2016) and environments (Lynch, 1991; McClelland and Naish, 2007).

Evolutionary divergence following reproductive isolation can occur in just a few generations (Christie et al., 2012; Thompson et al., 2019), and is thus potentially relevant for recent population sub-divisions. Anadromous fish populations that have been split into reproductively isolated subpopulations due to migration barriers have resulted in the evolution of freshwater forms in landlocked subpopulations (e.g., McDowall, 1997; Littrell et al., 2018). Evolutionary divergence and population genetic structure may also be modified by admixture associated with captive breeding programs, stocking, and escapes of farmed individuals (Christie et al., 2014), and this can affect migratory behavior and evolution of populations that have not previously been much influenced by gene flow.

Admixture is likely more important for species and populations that display spawning migration, compared with resident forms (Keefer and Caudill, 2014) for example owing to non-natal adult straying (Keefer and Caudill, 2014). Spawning migrating species and populations may also be affected more strongly by admixture resulting from management actions. Comparisons of dispersal probability between wild and captivebred individuals have generated mixed results; some studies report that captive bred individuals are more likely to disperse, while others have found that wild individuals are more dispersive [reviewed in Quinn (1993)]. Jonsson et al. (2003) showed that wild populations have a higher probability of homing and that captive-bred individuals tend to stray more. Studies investigating whether and how migratory behavior is affected by hybridization between different migratory forms are scarce. However, Saint-Pé et al. (2018) investigated genetic structure and spatial patterns of admixture in brown trout (S. trutta) within a small watershed in France, and report that dispersal was admixture-biased. In conclusion, populations can differentiate rapidly, selection can modify migratory behavior, and admixture between different migratory forms can impact on dispersal probability, population differentiation and genetic structure of migratory fish. Besides the immediate relevance for the understanding and management of biodiversity, this has implications for the productivity, functioning and resilience of ecosystems.

#### Overfishing and Fishing Regulations

Overfishing is a major threat to aquatic biodiversity globally (Pauly et al., 1998; Jackson et al., 2001). In addition to aggregations in spawning habitats, migratory species may be particularly vulnerable to overfishing through aggregations during migration (Allan et al., 2005). This is especially relevant to anadromous and catadromous species that pass through confined waterbodies represented by streams on their way toward the spawning habitats. It is plausible that species, populations and individuals may be differently affected by fisheries harvest depending on migration patterns and fishing regulations regarding timing (Diaz Pauli and Sih, 2017). For instance, it may result in skewed harvesting of migratory phenotypes in comparison to resident phenotypes, and ultimately change population dynamics and evolutionary trajectories. Fisheries may also impose differential mortality due to variation in timing and size of migratory fish. For example, it is a common practice to regulate fisheries with closed seasons (Wilen, 1985), and such actions might render early or late migrants disproportionately vulnerable to fisheries. If timing co-varies with body size (Tibblin et al., 2016b; Jonsson et al., 2017; Morita, 2019), regulations involving closed seasons may also translate into size-selective mortality, with potentially dramatic ecological and evolutionary side effects (Kuparinen and Merilä, 2007). Together, this calls for adaptive fisheries management where variation in migratory behavior is incorporated in management strategies and actions to prevent loss of biodiversity and unique migratory patterns. For example, given the protective variance reducing portfolio effect that population and life history diversity may have in exploited species, such as sockeye salmon, it will be important to minimize the homogenizing effects that hatcheries may have on genetic structure and to protect weak and declining populations from exploitation. This is essential both because it can stabilize productivity of individual species (Schindler et al., 2010), and because there can be a critical threshold for the number of populations below which regional extinction is likely (Hui et al., 2017). Maintaining options and portfolios for organism and their ecosystems is a means of spreading the risk and maintaining productivity, biodiversity and ecosystem functioning in the face of future uncertainties (Schindler et al., 2015; Waldman et al., 2016; Lowerre-Barbieri et al., 2017).

#### Responses to Changing Water Temperatures, Sea Surface Fluctuations, and Salinity Gradients Associated With Climate Change

Climate change constitutes a major threat to biodiversity in both terrestrial and aquatic environments. Environmental conditions (hydro geography, temperature, precipitation, ice coverage, sea surface levels, acidity, flow regimes, currents, and salinity gradients) are changing rapidly worldwide due to ongoing global warming (Mackenzie et al., 2007; IPCC, 2013, 2018; Cheng et al., 2019). In the wake of climate change, organisms will be exposed not only to increasing averages but also to more variable and extreme conditions (IPCC, 2018), with changes in both the strength and direction of selection over time. While there is little doubt that climate change is happening, it remains unclear how biodiversity and ecosystem services will be affected—particularly in aquatic systems that are less well studied compared with terrestrial systems (see Figure 1 in Forsman et al., 2016a).

Altered water temperatures, sea surface levels, flow regimes, and salinity gradients may modify the opportunities for dispersal and affect connectivity among populations (**Figure 1**). This too may induce changes in the timing of events, local adaptations, and distribution shifts, potentially with far reaching implications and indirect effects mediated via species interactions, modified community species compositions and altered ecosystem functioning.

#### Distribution Shifts

Because of limited potential for temperature regulation, body temperatures of fish generally conform closely to surrounding temperatures. Some species of fish [such as tunas (Scombroidei) and sharks (Lamniformes) (Dickson and Graham, 2004), and the opah (Lampris guttatus) (Wegner et al., 2015)] can evade the temperature boundaries of ambient water by generating and conserving metabolic heat internally, but this capacity is restricted to about 0.1% of the known fish species (Dickson and Graham, 2004). Most fish instead rely on external heat from the environment and on behavioral thermoregulation, including both larger scale migrations between colder and warmer environments and smaller scale vertical movements involved in sun basking and when fish take advantage of temperature differences among strata in stratified lakes and oceans, to regulate their internal temperature (May, 1979; Reynolds and Casterlin, 1980; Hertz et al., 1993; Gillooly et al., 2002; Mehner, 2012; Ma et al., 2018; Nordahl et al., 2018, 2019).

Mobile organisms (including fishes) may respond to temporal environmental changes (or altered demands) by dispersing to habitats with more suitable conditions, which might ultimately result in range expansions, distribution shifts (Parmesan and Yohe, 2003; Root et al., 2003; Cooke et al., 2004; Perry et al., 2005; Forsman et al., 2016b) and spatiotemporal modifications of migration routes (Crozier and Hutchings, 2014). That climate change is driving poleward distribution shifts in marine fish species that attempt to escape warm waters and enables fishes that cannot tolerate too cold water to colonize new regions complicates management, governance, and international fishing regulations. For example, recent modeling results suggest that the system for allocating fish stocks is being outpaced by the movement of fish species in response to climate change (Pinsky et al., 2018).

#### Phenology Shifts

In fish, changes in the timing of adult migration and reproduction, age at maturity and in age at juvenile migration seem to be common responses to temperature shifts (Crozier and Hutchings, 2014). Cooke et al. (2004) report that the timing of peak upriver spawning migration of sockeye salmon in the Fraser River shifted forward more than 6 weeks from 1995 through 2002, and that the earlier migration was associated with higher pre-spawning mortality. Such temporal shifts in the onset of spawning migration in salmonids are typically interpreted as responses to climate change. However, it has also been suggested that it might instead reflect a fisheries-induced evolutionary response because late-spawning brood lines are being fished for longer time periods (Morita, 2019). Predictions regarding future changes of migration timing in the face of global warming are further complicated by the heterogeneity in long-term shifts in migration timing seen across species and populations of Pacific salmon, with some postponing and others migrating earlier (Kovach et al., 2015). Environmental challenges in the form of warmer waters and altered flow velocities associated with climate change may also directly influence locomotor performance and the costs of migration, moderate energetic trade-offs, and limit the amount of resources available for other facets of the reproductive cycle (Fenkes et al., 2016).

An investigation of an anadromous pike population in the Baltic Sea shows that the timing of arrival to the spawning area may vary among years by as much as 3 weeks. Despite this year-to-year flexibility, the relative timing of spawning migration differed considerably and in a consistent manner among individuals (Tibblin et al., 2016b). Whether this variation has a genetic component remains unknown, but estimates of repeatability point to an upper bound of heritability of about 0.25 (Tibblin et al., 2016b), indicating that evolutionary responses to selection on timing of spawning in pike are possible. That the timing of spawning migration in pike is highly flexible, with individuals fine-tuning migratory timing between years (Tibblin et al., 2016b), indicates that temporal behavioral adjustments are used to ensure that embryos and larvae develop when temperature conditions are favorable. Such phenotypic flexibility may buffer populations against rapid unpredictable environmental changes and potentially prevent the loss of genetic diversity (Wennersten and Forsman, 2012; Forsman, 2015).

#### Adaptations of Migratory Fish to Changing Conditions

Migrating fishes cross habitat borders and move along environmental gradients (**Figure 3**). Individuals are exposed to environmental changes also if they remain for prolonged periods within a given habitat or limited area. Depending on the spatial and temporal scales of the environmental changes relative to the dispersal capacity, generation time and reproductive mode of the organisms, this may maintain a diversity of specialists within populations, or promote the evolution of generalist strategies that perform reasonably well across a range of environments (Levins, 1968; Kassen, 2002; Forsman et al., 2011). Generalist strategies include plastic or flexible phenotypes that adjust to conditions via developmental modifications or reversible intra-individual physiological or behavioral modifications (Pigliucci, 2001; West-Eberhard, 2003; Forsman, 2015; Tibblin et al., 2016b).

With regards to early life-history traits, it has been demonstrated in some species of spawning migrating fish that populations with a history of exposure to different thermal conditions during incubation of eggs and embryos display local adaptations and respond differently to temperature manipulations in the laboratory. Examples include cold water specialist salmonids such as brown trout (Salmo trutta) (Jensen et al., 2008) and grayling (Thymallus thymallus) (Kavanagh et al., 2010; Thomassen et al., 2011; Mäkinen et al., 2016), and temperate latitude species such as pike (Sunde et al., in press). Given that early and late spawning phenotypes coexist within populations (Tibblin et al., 2016b), and that the offspring produced by early arrivers likely develop in lower temperatures, it can be hypothesized that correlational selection (Arnold and Wade, 1984; Forsman and Appelqvist, 1998) on the combination of spawning timing and temperature tolerance has favored the evolution of genetic covariance and phenotypic integration between these behavioral and physiological traits. This hypothesis could be evaluated by comparing temperature related performance of eggs and embryos produced by gametes collected from adults arriving at the beginning and toward the end of the spawning period. Temperature related adaptations may potentially reinforce reproductive isolation by time (Hendry and Day, 2005), and contribute to further population structuring.

Spawning migrating fish not only have to cope with changing water temperatures. Global warming also brings modifications in sea surface levels and salinity regimes that may be particularly challenging for some anadromous species. Depending on the altitude of spawning areas proximate to the sea, sea surface fluctuations may result in that saline or brackish water temporarily enters the freshwater areas used for spawning and early life-history stages (Sunde et al., 2018a). A recent study suggests that different anadromous populations of pike in the Baltic Sea vary in their ability to cope with fluctuating salinity levels, and that the effects of salinity differed among families within populations, consistent with the notion that intrapopulation genetic variation for developmental plasticity offers buffering capacity and adaptive potential (Sunde et al., 2018a). Changing conditions in the areas used for reproduction can also affect population size and density, with consequences for intensity of competition, cannibalism, sexual selection, and for population genetic diversity and structure that together may influence the relative success of alternative migration strategies.

## FUTURE DIRECTIONS

Scientific output on variation in fish migration has increased tremendously over the past 50 years (**Figure 2**). This review highlights patterns, causes, and consequences of variation and flexibility of migration behaviors that are of relevance for the understanding, protection, and sustainable utilization of migratory fishes and of their ecosystems (**Figure 1**). Despite extensive previous research (**Figure 2**), important knowledge gaps and unanswered questions remain that require future investigations. We briefly summarize these points and future directions below:

Research on fish migration has traditionally focused on a few species of high socio-economic importance, primarily salmonids. In most species and populations, the relative contributions of genetic and environmentally induced variation in migratory behavior remain unresolved. Knowledge of how internal attributes, social interactions, and environmental factors (**Table 1**; **Figure 1**) influence variation in migratory behavior, timing and distance among species, populations and individuals of fish (including many salmonids) is incomplete. However, recent technological developments in bio-logging have advanced the ability to obtain high-resolution data on fish movements, inform about internal and external drivers of movements, help illuminate the consequences of movements for individual performance and population fitness, and provide answers to the questions how, where, when, and why fish migrate (Nathan et al., 2008; Cooke et al., 2013; Wilmers et al., 2015; Nordahl et al., 2018, 2019; Lowerre-Barbieri et al., 2019). This may improve our understanding of diversity, and allow for more reliable predictions of the consequences that exploitation, management actions and climate change may have for migratory fish.

New insights into the causes, consequences, and evolutionary dynamics of migratory behavior in fishes might be gained by phylogeny based comparative approaches (Felsenstein, 1985; Bloom et al., 2018). Besides uncovering the distribution of data deficiency and identifying taxa and geographic regions in particular need of further investigation, phylogenetic comparisons can inform why certain species migrate whereas others do so to a lesser degree, and uncover associations of evolutionary shifts in migration behaviors with environmental factors and with morphological, physiological, or behavioral phenotypic dimensions (e.g., McDowall, 1997; Watanabe et al., 2015; Forsman and Berggren, 2017; Bloom et al., 2018).

Another potentially rewarding line of future research is to focus on variation among populations. Information on migration behaviors for multiple populations and species compiled using systematic reviews and meta-analytical approaches (Gurevitch et al., 2018) can illuminate patterns and evaluate potential drivers of variation in migration mode, migration timing, and migration distance among populations. Besides summarizing information and advancing knowledge, results can inform when the evidence is sufficient for adaptive population specific management and conservation efforts.

The timing of spawning migration varies considerably both among and within fish populations. Whether isolation by time (Hendry and Day, 2005) is a common driver of reproductive isolation, genetic divergence and adaptation in fish, and whether differences in the timing of spawning migration contribute more or less to population structure in different species depending on their life-history remains to be investigated.

Several studies have aimed at identifying phenotypic correlates of variation in migratory behavior and performance. However, we need to know more about how alterations in migratory challenges brought about by exploitation, management and climate change modifies the intensity and direction of selection and evolutionary shifts in dispersal related physiological, morphological, behavioral traits, and correlated life-history attributes.

Improved sequencing technologies have enabled genomic resources to be generated with increasing efficiency and speed, such that non-mainstream fish species can now be utilized as models. The NCBI (2019) database (http://www.ncbi.nlm.nih. gov/genome/browse/ - last accessed 18 February 2019) currently includes genome sequence assemblies for 258 fish species. Recent developments in genomic tools [e.g., RAD-sequencing, WGS, GWAS and SNP-genotyping (Andrews et al., 2016)] together with information for closely related species and populations in different environments may help identify genes and underlying genomic regions under selection involved in shaping the diversity of migratory behavior. There is also potential for such approaches to clarify the contributions of stochastic processes, gene flow, selection, and plasticity and to pinpoint the role of specific genes in shaping genetic structure and phenotypic evolution in migratory fishes (e.g., Barson et al., 2015; Momigliano et al., 2017; Thompson et al., 2019).

Changes in river connectivity alters selection pressures and may have implications for species interactions, community composition and ecosystem functioning. To evaluate the consequences that selection associated with fishway passages may have for spawning migrating fish requires a combination of methodological approaches, such as analysis of otolith elemental chemistry, collection of high-resolution spatiotemporal movement data using mark-recapture studies, telemetry, and passive integrative transponders, behavioral monitoring using fishways equipped with submerged cameras, as well as longitudinal and cross sectional comparisons of phenotype distributions (Nathan et al., 2008). When evaluating the design and operation of fishway passages, it is important to consider that selection may operate during each of the approach, entry, and passage components, as well as on post-passage behaviors and performances; it is the cumulative effect that may modify evolution of behavior, reproduction, population genetics and population dynamics. As the number of studies grows, the opportunities for systematic reviews and meta-analytical approaches to provide new insights will increase.

The directions and rates of genetic exchange between populations may change over time owing to natural processes, anthropogenic environmental makeovers and management actions. The consequences of genetic admixture can vary from positive to negative, affect genetic differentiation and diversity between and within populations, and ultimately influence viability and adaptability of populations and species (McClelland and Naish, 2007; McGinnity et al., 2009). An important question to consider is whether and how migratory strategies employed by the populations subjected to management might affect the outcome of admixture. Conversely, few (if any) studies have examined whether and how genetic admixture affects migratory behavior. It is therefore necessary to investigate the reciprocal feedback loop between migration and admixture, and how it may influence dynamics, genetic structuring, and viability of populations. Further, admixture effects can be sex-specific (Sunde et al., 2018b), and it might be hypothesized that this should impact on spawning migratory behavior. To our knowledge, however, it has not yet been examined whether sex-biased dispersal or migration is associated with sex-specific responses to admixture.

Water temperature influences migratory behavior and performance of fishes. Recent studies show that fish can reap thermoregulatory rewards and elevate their body temperature above that of ambient water by sun basking near the water surface (Nordahl et al., 2018, 2019). There is little doubt that the capacity for aquatic thermoregulation by sun basking is important for fish. However, more work is needed in this emerging area to clarify the consequences of sun basking for fish performance, migratory behavior, spatiotemporal distribution shifts, and whether and how it will modify predictions regarding responses to climate change.

To understand the effects that climate change may have on migratory fish, future research needs to expand beyond considering effects of the projected increase in average water temperatures. There is a need for better knowledge of how more extreme and fluctuating water temperatures and sea levels may affect the development of early lifehistory stages or the growth and survival of adult fish, and whether this influences the relative success of spawning migrating forms.

Lastly, we offer a cautionary note regarding management. It remains uncertain how human activities and climate change will influence environmental conditions and selective regimes for fish. A particular challenge when addressing consequences of climate change is to find the balance between realism and methodological tractability (Forsman et al., 2016a), and this applies also to the evaluation of management actions. It is difficult to foresee which genetic makeups, phenotypic trait value combinations, and behaviors that will be most successful in the future. To safeguard against this uncertainty, management actions should be designed to maintain genetic and phenotypic diversity with regards to migratory behavior, seasonal timing of reproduction, place of spawning, growth trajectories, size and age at maturity, and reproductive allocation strategies both among and within populations and species. Theory and empirical evidence concur that this may promote Tamario et al. Fish Migrations

establishment success, reduce extinction risk, enable populations and species to cope with environmental change and adapt to novel conditions, and increase productivity, functioning and resilience of ecosystems (Hughes et al., 2008; Schindler et al., 2010, 2015; Bolnick et al., 2011; Wennersten and Forsman, 2012; Forsman, 2014, 2015; Forsman and Wennersten, 2016; Waldman et al., 2016; Lowerre-Barbieri et al., 2017). Perhaps future research should aim to develop a 'best practice' regarding adaptive management and how to safeguard against uncertainty; there might be good solutions waiting to be discovered.

#### AUTHOR CONTRIBUTIONS

The first outline of the review was conceived of by AF. AF coordinated the writing, and revisions. All authors contributed to developing the idea and to the writing of separate sections of the first draft and provided

#### REFERENCES


feedback on earlier drafts and approved the final version for submission.

#### FUNDING

Funding was provided by the Swedish Research Council Formas (Dnr. 2017-00346 to AF, EP, and PT; Dnr. 2018- 00605 to PT), Stiftelsen Oscar och Lili Lamms Minne (DO 2017-0050) to AF, EP, and PT, and by Linnaeus University (AF, PT).

#### ACKNOWLEDGMENTS

We thank Brett Sandercock and Nathan Senner for the invitation to write this review. We would like to thank Oscar Nordahl and two reviewers for comments on an earlier version of the manuscript.


Independent Project Course, Master in Science, Second Level, 15 ECTS, Linnaeus University, Faculty of Health and Life Sciences, Department of Biology and Environmental Science (Kalmar).


populations: depressed recruitment and increased risk of climate-mediated extinction. Proc. R. Soc. B 276, 3601–3610. doi: 10.1098/rspb.2009.0799


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Tamario, Sunde, Petersson, Tibblin and Forsman. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A Migratory Divide Among Red-Necked Phalaropes in the Western Palearctic Reveals Contrasting Migration and Wintering Movement Strategies

#### Edited by:

Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway

#### Reviewed by:

Stephen Brown, Manomet Center for Conservation Sciences, United States Monique De Jager, Netherlands Institute of Ecology (NIOO-KNAW), Netherlands

#### \*Correspondence:

Rob S. A. van Bemmelen rvanbemmelen@gmail.com

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 23 November 2018 Accepted: 06 March 2019 Published: 04 April 2019

#### Citation:

van Bemmelen RSA, Kolbeinsson Y, Ramos R, Gilg O, Alves JA, Smith M, Schekkerman H, Lehikoinen A, Petersen IK, Þórisson B, Sokolov AA, Välimäki K, van der Meer T, Okill JD, Bolton M, Moe B, Hanssen SA, Bollache L, Petersen A, Thorstensen S, González-Solís J, Klaassen RHG and Tulp I (2019) A Migratory Divide Among Red-Necked Phalaropes in the Western Palearctic Reveals Contrasting Migration and Wintering Movement Strategies. Front. Ecol. Evol. 7:86. doi: 10.3389/fevo.2019.00086 Rob S. A. van Bemmelen1,2 \*, Yann Kolbeinsson<sup>3</sup> , Raül Ramos <sup>4</sup> , Olivier Gilg5,6 , José A. Alves 7,8, Malcolm Smith<sup>9</sup> , Hans Schekkerman<sup>10</sup>, Aleksi Lehikoinen<sup>11</sup> , Ib Krag Petersen<sup>12</sup>, Böðvar Þórisson<sup>7</sup> , Aleksandr A. Sokolov <sup>13</sup>, Kaisa Välimäki 11,14 , Tim van der Meer <sup>1</sup> , J. David Okill <sup>15</sup>, Mark Bolton<sup>16</sup>, Børge Moe<sup>17</sup>, Sveinn Are Hanssen<sup>18</sup> , Loïc Bollache5,6, Aevar Petersen<sup>19</sup>, Sverrir Thorstensen<sup>20</sup>, Jacob González-Solís <sup>4</sup> , Raymond H. G. Klaassen<sup>21</sup> and Ingrid Tulp<sup>1</sup>

<sup>1</sup> Wageningen Marine Research, IJmuiden, Netherlands, <sup>2</sup> Resource Ecology Group, Wageningen University, Wageningen, Netherlands, <sup>3</sup> Northeast Iceland Nature Research Centre, Husavik, Iceland, <sup>4</sup> Biodiversity Research Institute (IRBio) and Department of Evolutionary Biology, Ecology and Environmental Sciences (BEECA), Faculty of Biology, Universitat de Barcelona, Barcelona, Spain, <sup>5</sup> Chrono-environnement, Université de Bourgogne Franche-Comté, Besançon, France, <sup>6</sup> Groupe de Recherche en Ecologie Arctique, Francheville, France, <sup>7</sup> Department of Biology & Centre for Environmental and Marine Studies CESAM, University of Aveiro, Aveiro, Portugal, <sup>8</sup> South Iceland Research Centre, University of Iceland, Laugarvatn, Iceland, <sup>9</sup> RSPB Scotland, Shetland, United Kingdom, <sup>10</sup> SOVON, Nijmegen, Netherlands, <sup>11</sup> The Helsinki Lab of Ornithology, Finnish Museum of Natural History, University of Helsinki, Helsinki, Finland, <sup>12</sup> Department for Bioscience, Aarhus University, Aarhus, Denmark, <sup>13</sup> Arctic Research Station of Institute of Plant and Animal Ecology, Russian Academy of Sciences, Labytnangi, Russia, <sup>14</sup> Department of Psychology, Centre for Interdisciplinary Brain Research, University of Jyväskylä, Jyväskylä, Finland, <sup>15</sup> Trondra, Shetland, United Kingdom, <sup>16</sup> RSPB Centre for Conservation Science, Royal Society for the Protection of Birds, Sandy, United Kingdom, <sup>17</sup> Norwegian Institute for Nature Research (NINA), Trondheim, Norway, <sup>18</sup> Norwegian Institute for Nature Research (NINA), Framsenteret, Tromsø, Norway, <sup>19</sup> Independent Researcher, Reykjavik, Iceland, <sup>20</sup> Independent Researcher, Akureyri, Iceland, <sup>21</sup> Conservation Ecology Group, Groningen Institute for Evolutionary Life Sciences (GELIFES), Groningen University, Dutch Montagu's Harrier Foundation, Groningen, Netherlands

Non-breeding movement strategies of migratory birds may be expected to be flexibly adjusted to the distribution and quality of habitat, but few studies compare movement strategies among populations using distinct migration routes and wintering areas. In our study, individual movement strategies of red-necked phalaropes (Phalaropus lobatus), a long-distance migratory wader which uses saline waters in the non-breeding period, were studied using light-level geolocators. Results revealed a migratory divide between two populations with distinct migration routes and wintering areas: one breeding in the north-eastern North Atlantic and migrating ca. 10,000 km oversea to the tropical eastern Pacific Ocean, and the other breeding in Fennoscandia and Russia migrating ca. 6,000 km—largely over land—to the Arabian Sea (Indian Ocean). In line with our expectations, the transoceanic migration between the North Atlantic and the Pacific was associated with proportionately longer wings, a more even spread of stopovers in autumn and a higher migration speed in spring compared to the migration between Fennoscandian-Russian breeding grounds and the Arabian Sea. In the wintering period, birds wintering in the Pacific were stationary in roughly a single area, whereas individuals wintering in the Arabian Sea moved extensively between different areas, reflecting differences in spatio-temporal variation in primary productivity between the two wintering areas. Our study is unique in showing how habitat distribution shapes movement strategies over the entire non-breeding period within a species.

Keywords: flexibility, itinerancy, migration strategy, Phalaropus lobatus, plasticity, red-necked phalarope

#### INTRODUCTION

Animal movements are strongly linked to habitat and resource availability. For example, during the non-breeding season, migratory birds cover long distances to track spatio-temporal peaks in resource abundance and avoid unfavorable conditions over large spatial scales (Newton, 2010; Thorup et al., 2017). Nonbreeding movements may consist of a long-distance migration to wintering areas and movements within the wintering areas, which can cover long distances and—in some species—are carried out by only a portion of the individuals (Trierweiler et al., 2013; Kolecek et al., 2018 ˇ ). Components of movement strategy, such as the number of staging sites, the distances between them, and the time spent at each staging site, are thought to be dictated by the availability and relative quality of habitat (Alerstam and Lindström, 1990; Gudmundsson et al., 1991). Movement strategies are expected to be flexibly adjusted to habitat availability and quality within the boundaries set by, for example, maximum fattening rates and fuel loads, resulting in different strategies among individuals or populations using different migration routes and wintering areas. Indeed, migration strategies do differ between populations in many species (Buehler and Piersma, 2008; Delmore et al., 2012; Alves et al., 2013). At the same time, surprisingly similar migration strategies have been observed in other species, despite geographically distinct migration routes (Fraser et al., 2013; Trierweiler et al., 2014). With so few studies, the extent to which species flexibly adjust large-scale movement strategies to habitat remains poorly understood.

The red-necked phalarope Phalaropus lobatus is a small wader which breeds in fresh water ponds in arctic and subarctic tundra. The species is probably best known for its unusual mating system, with reversed sex-roles where polyandry takes place and the male cares for eggs and chicks (Reynolds et al., 1986). In the nonbreeding period, it adopts a pelagic lifestyle in three disjunct tropical ocean areas: the tropical eastern Pacific, the Arabian Sea, and off the East Indies (Cramp and Simmons, 1983). Due to the challenges involved in studying small seabirds at sea, its nonbreeding movements remained elusive until recently, when lightlevel geolocators revealed individual non-breeding movements. Four Swedish males were shown to winter in the Arabian Sea in the north-west Indian Ocean (van Bemmelen et al., 2016), confirming what was already expected based on a small number of ring recoveries (Schiemann, 1977). No ring recovery exists for populations breeding on islands in the north-eastern North Atlantic (Scotland, Faroe Islands, Iceland, and Greenland), despite substantial ringing efforts (Schiemann, 1977; Wernham et al., 2002). Recently, three male red-necked phalaropes were tracked from the Scottish breeding area to the northern Humboldt Current in the Pacific (Smith et al., 2014, 2018), suggesting that north-eastern North Atlantic populations migrate westwards after breeding. The migration routes to the Pacific and Arabian Sea differ in distance (respectively, ca. 11,000 km vs. ca. 6,000 km one-way; Smith et al., 2014; van Bemmelen et al., 2016), and habitat (mostly oversea vs. largely overland), given that rednecked phalaropes depend on saline waters for staging (Cramp and Simmons, 1983; van Bemmelen et al., 2016). In addition, the wintering grounds contrast in spatio-temporal variation in primary productivity, with stable conditions in the Pacific sites but strong fluctuations in the Arabian Sea (Longhurst, 2006). The occurrence of two distinct breeding populations of red-necked phalarope with divergent migration routes and wintering areas provides a rare opportunity to study how, within a single species, habitat distribution shapes movement strategies over the entire non-breeding period.

A first objective of the current study is to investigate the generality of previous tracking studies (Smith et al., 2014, 2018; van Bemmelen et al., 2016) using non-breeding movements of red-necked phalaropes from nine breeding locations between East-Greenland (22◦W) and Western Siberia (69◦E) recorded by light-level geolocators. Subsequently, both migration strategies and wintering movement strategies are compared between the two routes and wintering areas. Based on optimal migration theory and assuming that suitable feeding conditions are more widespread along the marine western migration route, rednecked phalaropes migrating to the Pacific are expected to make shorter flights with more frequent but briefer stopovers on migration to and from the breeding grounds, thereby avoiding the costs of carrying large fuel loads (Alerstam and Lindström, 1990). In contrast, red-necked phalaropes migrating to the Arabian Sea are expected to show a more direct flight, with few but longer stopovers, as more widely separated saline waters along their route forces them to build up larger fuel loads to cover longer flights (Alerstam and Lindström, 1990). Different movement strategies may favor selection toward certain morphological adaptations, such as longer, more pointed, wings for longer migration distances (Alerstam, 1990; Leisler and Winkler, 2003). Thus, we also tested for morphological variation potentially associated with the two migration strategies. After arrival at the wintering grounds, birds wintering in the Pacific are expected to move only short distances within the wintering grounds ("residency"), given the high and constant primary productivity of the northern Humboldt Current (Chavez and Messié, 2009). In contrast, birds wintering in the Arabian Sea are expected to move between several distant areas during the wintering period ("itinerancy"), reflecting large seasonal and spatial variation in primary productivity (Longhurst, 2006). In addition to the above, individual consistency between years and the potential effect of sex on migration timing and strategies are explored. We expect females to arrive earlier than males to maximize their potential of obtaining males (Oring and Lank, 1982; Reynolds et al., 1986).

# MATERIALS AND METHODS

#### Catching and Geolocator Deployments

Red-necked phalaropes were captured and fitted with geolocators from 2012 to 2017 in Greenland (Constable Pynt: 70◦ 45′N−22◦ 38′W), two sites in Iceland (Flói: 63◦ 56′N−21◦ 15′W and Aðaldalur: 65◦ 51′N−17◦ 04′W), Scotland (Fetlar: 60◦ 36′N−0 ◦ 52′W), Sweden (Ammarnäs: 65◦ 59′N−16◦ 01′E), Finland (Enontekiö: 68◦ 58′N−21◦ 16′E), Norway (Slettnes: 71◦ 05′N−28◦ 13′E), and two sites in Russia (Tobseda: 68◦ 36′N−52◦ 19′E; Erkuta: 68◦ 14′N−69◦ 9 ′E), see **Figure 1**. Adults were captured while foraging in tundra ponds using mist nets or at their nests using walk-in traps or spring traps. Different geolocator types and deployment methods were applied because studies in Scotland, Finland and Iceland each began as independent projects before joining the project covering Greenland, Sweden, Norway and Russia. Geolocators were either leg-mounted using a modified darvic ring with flag (Greenland and the first year at Flói, Iceland) or back-mounted using leg-loop harnesses (all other sites and years, Rappole and Tipton, 1991). Leg-loops were constructed from 1 mm silastic tubing (WMQ 60) with an elasticated core (only first study year in Scotland and in Finland) or 1 mm wide flat braided shelf-string (British Trust for Ornithology, UK). Geolocator models used were the Mk10 model (British Antarctic Survey, Cambridge, UK), (a modified version of) the Intigeo P65A (Migrate Technology Ltd, Cambridge, UK) and the SOI-GDL2 v1.3 in Finland and v2.3 in Scotland (Swiss Ornithological Institute, Sempach, Switzerland), weighing, respectively 1.0 g, 1.0 g and 0.7 g. At a mean body mass of 38.3 g for females (n = 170) and 33.4 g for males (n = 468), a geolocator mass of 1 g represents about 2.6% of mean body mass in females and 3.0% in males, or 1.7 and 2.0% in case of a geolocator weight of 0.7 g. We are not aware of biases in position estimations inherent to specific geolocator types. The following biometrics were collected: wing length (to the nearest 1.0 mm), bill length (to the nearest 0.1 mm), total head length (head + bill; to the nearest 1.0 mm), tarsus length (to the nearest 1.0 mm) and body mass (to the nearest 0.1 g). In addition to collecting biometrics for the birds equipped with geolocators, biometrics were obtained from other individuals at the study sites, and also from individuals captured in Iceland (Mývatn: 65◦ 36′N−16◦ 60′W and Flatey Island: 65◦ 22′N−22◦ 55′W) and Russia (Medusa Bay: 73◦ 04′N−80◦ 30′E). Using annual encounter data from 212 individuals with geolocators, we estimated apparent survival ϕ and resighting probability p using Cormack-Jolly-Seber (CJS) models for live encounter data in the "Rmark" package in R, an interface to the program Mark (Laake, 2013). AICc was used to select the best model from a range of CJS models fitted with parameters for ϕ, p and study site either fixed or varying per year. Four CJS models had 1AICc values <2, including the simplest model (1AICc = 0.64), which estimated p at 0.52 (CI = 0.31–0.71) and ϕ at 0.32 (CI = 0.23–0.43). An apparent survival rate of 0.32 is notably low for a wader species (Méndez et al., 2018) but is within the range of 0.17 to 0.56 reported from sites in Alaska (Colwell et al., 1988; Schamel and Tracy, 1991; Sandercock, 1997) and likely reflects a low site fidelity rather than survival. Nevertheless, a weak effect of geolocator deployment cannot be ruled out, considering we could not compare against a control group without a geolocator, and knowing that a geolocator attached to a leg flag and weighing >2.5% of the body mass led to lower return rates in other small wader species (Weiser et al., 2016). Trapping and tagging red-necked phalaropes in Greenland has been approved by the Ministry of Fisheries, Hunting and Agriculture (Government of Greenland); in Scotland by the Special Methods Technical Panel of the British Trust for Ornithology and Scottish Natural Heritage; in Finland by the Lapland Center for Economic Development, Transport and the Environment; in Norway by the Norwegian Food Safety Authority (FOTS ID 6328, 7421, 8538) and in Sweden by the Malmö-Lunds djurförsöksetiska nämnd (M160-11, M470-12).

#### Geolocator Data Analysis

Data were downloaded from retrieved loggers and processed using BASTrack software (BAS, UK) and time was adjusted for clock drift. Twilights events were determined using the "twilightCalc" function in GeoLight package version 2.0 (Lisovski and Hahn, 2012), in R version 3.4 (R Core Team, 2017) using a light threshold of two and subsequently checked for errors by plotting the date against time of sunrise or sunset. For a range of potential sun angles (−7 to 2◦ ), we plotted (for each track) the position estimates and selected the sun angle that (1) minimized the amplification of latitudinal error close to the equinoxes while (2) resulting in similar latitudes at both sides of the equinox, and (3) where positions fitted the shape and position of the oceans and inland seas. Final sun angles were between −5.5 to −5.0◦ for BAS loggers, −5.0 to −3.5◦ for Migrate Technology loggers without stalks, −5.5 to −4.5◦ for Migrate Technology loggers with stalks, and 0◦ for the Swiss Ornithological Institute loggers.

Stationary periods were delineated based on patterns in time of sunset and sunrise, using the function "changeLight" from the "GeoLight" package in R, with a minimum staging duration of 2 days. For each stationary period, the geographical centroid was calculated. For stationary periods with latitudes greatly affected by proximity to the equinox (e.g., resulting in positions in the central Indian Ocean), a latitude was assumed at the coastline north of the original position estimate, which was always in close proximity to other staging locations. Subsequently, stationary periods with geographical centroids closer than 200 km were joined. Stationary areas with centroids in the Arabian Sea, Persian Gulf or Pacific were defined as the wintering period. Staging areas preceding the wintering period were assigned to the autumn migration and those following the wintering period to spring migration. Departure from and arrival at the breeding area were calculated as the time needed to cover the great-circle distance

to the first or last position estimate, respectively, at a speed of 13.3 ms−<sup>1</sup> (Alerstam and Gudmundsson, 1999). Migration distance was measured using the sum of great-circle distances, using the "rdist.earth" function in the "fields" package in R, covered from the breeding site via the centroids of each staging site to the centroid of the first (autumn migration) or last (spring migration) wintering area. Similarly, distance traveled during winter was calculated as the great circle distance between staging areas assigned to the winter period. The fat load needed for the initial migratory movement in still air was estimated using the software package Flight v. 1.25 (Pennycuick, 2008). For each potential body mass (at increments of 1 g), the fat fraction in which the migration simulation finished with a lean body mass was estimated by an iterative process. Pennycuick (2008) defined lean body mass as the "body mass of a bird with zero fat reserves, but not actually starving." Here, lean body mass was taken to be the minimum weight for each sex in our database, excluding one male of 24 g and one female of 28.0 g: 26.0 g for males (n = 440) and 29.3 g for females (n = 164). Note that taking a lower lean body mass results in higher fat fractions and thus larger potential flight ranges. After estimating the body mass required to cover the distance to the first stopover, the time required to fuel for this flight was calculated assuming a fat deposition rate (FDR) of 3.2% of lean body mass (Sikora

and Zielinksi, 2000; Lindström, 2003). As no FDR values are known for red-necked phalaropes at the wintering areas, the sensitivity of migration speed was explored by plotting migration speed for a range of possible FDR values at each wintering ground (**Supplementary Figure 1**). In two individuals migrating in spring from the Pacific to the north-eastern North Atlantic, no stopovers of 2 or more days were identified. In these cases, the first leg was taken as the distance to the centroid of three clustered position estimates off Florida, which is near the first staging area of other individuals following the western route (**Figure 1**). Migration duration was taken as the number of days in transit between the breeding and the wintering area, plus the estimated time needed to fuel for the initial leg. Migration speed was calculated as the total migration distance divided by the migration duration.

The following parameters were derived for each track: migration distance, total migration duration (including stopovers and fuelling time for the first leg, see above), distance to the first staging area, length of the longest stopover, number of stopovers, migration speed, and timing of departure from and arrival at both the breeding and wintering area. First, t-tests were used to test whether any of the aforementioned parameters differed between sexes among Swedish birds, the only area with sufficient females to compare sexes (see Results). ANOVAs were then constructed comparing each of the parameters between four "regions:" "Atlantic" (Greenland, Iceland and Scotland), "Sweden" (Ammarnäs), "Norway" (Slettnes), and "Russia" (Tobseda and Erkuta). A contrast between "Atlantic" (hereafter referred to as "western" birds, populations or migration route) and the three other regions (hereafter referred to as "eastern" birds, populations or migration route) was used to test for differences in migration strategy parameters between the two flyways. The number of stopovers was compared between the flyways using the same contrast structure in a Poisson Generalized Linear Model (GLM). The number of stopovers per unit distance was tested using the same Poisson GLM with the log of the migration distance included as an offset. In addition to the analysis of the geolocator data, body mass measurements from breeding sites were compared in a Linear Mixed-effect Model (LMM) with flyway as the only main effect and site as a random effect, using the nlme package in R (Pinheiro et al., 2018). From these body mass measurements, maximum flight ranges were estimated using the relationship as derived from Pennycuick's model (see above). For comparison of fuel loads at staging sites, we obtained body mass measurements at the Bay of Fundy, Canada, along the western migration route, from Mercier (1985) and from saline lakes in Kazakhstan, along the eastern migration route, from Gavrilov et al. (1983).

To compare winter movement strategies between the two wintering areas, the number of staging areas was compared using a Poisson GLM and the duration of the longest stationary period and the total distance covered within the winter period were compared using ANOVAs with the same contrast structure as used in the analysis of migration characteristics. In addition, we linked phalarope movements to spatio-temporal patterns in food availability. First, net ocean primary productivity for 8 day periods, based on the Vertically Generalized Production Model (VGPM) algorithm (Behrenfeld and Falkowski, 1997), were obtained from O'Malley (2015). Second, a 300 km wide polygon parallel to the coast was drawn and split into four main areas: (1) northern Humboldt Current to the Pacific coast off Central America, (2) Red Sea via the Gulf of Aden and Oman to Pakistan, (3) Persian Gulf to the Gulf of Oman, and (4) East coast of Somalia. Areas were subdivided bins perpendicular to the coast, thus in longitude bins (Arabian Sea) or latitude bins (Pacific), each 100 km in width. Finally, to test whether tagged individual red-necked phalaropes were associated with areas of high primary productivity, for each 8-day period and longitude/latitude bin, the log<sup>e</sup> of the 95% quantile of primary productivity values was correlated with the number of staging red-necked phalaropes, using Poisson LMMs, with "area" and "longitude/latitude bin" as random effects to account for spatial correlation. To test whether wintering movements are directed toward areas with higher primary productivity, or a steeper decline in primary productivity (which could be due to higher grazing intensity by zooplankton), we calculated the difference in primary productivity and its slope between departure from one site and arrival at the next site and tested whether this differed from zero using an intercept-only LMM, with individual as a random effect.

In the Greenlandic bird, the logger was mounted to a leg flag and also recorded submersion in saline water every 3 s and summed the number of "wets" every 10 min. This "wet/dry" data was used to delineate migration flights (Battley and Conklin, 2017). Other loggers were back-mounted and only occasionally recorded submersion, and were unsuitable for delineating long flights.

#### Biometrics

We tested for differences in morphology between the two populations using biometric data from eleven breeding locations from Greenland to Taymyr, Russia. As females are larger than males (Cramp and Simmons, 1983), we built separate LMMs for each sex, comparing wing length, bill length and tarsus length between western and eastern birds, and with "breeding site" as a random effect. Wings can be longer either due to more pointedness of wings or due to larger body size. In absence of data on pointedness, we took tarsus length as a measure of body size and fitted a model of wing length by flyway and tarsus length.

# RESULTS

#### Sample Size

In total, 34 geolocator tracks (each track referring to 1 year of data between two breeding periods) of 26 individuals were obtained from eastern populations. From the western population, 10 tracks of 8 individuals were obtained. Nine tracks (1 from a western bird and 8 from eastern birds) were incomplete due to premature failure or a flat battery of the logger at the wintering areas; 5 of these failures happened in the second year of tracking. Repeated tracks were obtained for 2 western and 7 eastern individuals (see below). To compare movement strategy variables between western and eastern birds, only the first track of each individual was included. Also, the single Finnish bird was excluded as it had much higher variation in estimates of twilights and positions than the other loggers. The final dataset for comparisons of movement strategies thus included 8 full tracks from western birds (1 female, 7 males) and 22 full tracks plus 3 autumn migrations from eastern birds (6 females, 19 males).

#### Migration Routes

The data show two distinct migration routes from Western Palearctic red-necked phalaropes (**Figure 1**). All Fennoscandian-Russian birds migrated to the Arabian Sea (n = 26, including the Finnish bird), whereas those breeding at North Atlantic islands wintered in the Pacific; either in the northern Humboldt Current (n = 6) or off the westcoast of Central America (n = 2).

#### Sex Differences in Migration Strategy

Only the Swedish sample contained sufficient females (n = 5) and males (n = 7) to compare sexes. Between Swedish males and females, no migration strategy variables were significantly different, either during autumn or spring migration. Although the trends in our data comply with the expectation of earlier migration in females, overlap between sexes in timing rendered them non-significant (p > 0.1), but females arrived nearsignificantly earlier than males at the breeding area [t(1, 9) = −2.0, p = 0.078]. Like the Swedish birds, the single female from Norway fell within the range of values for all migratory strategy variables. She was earlier than most males in the timing of departure from the breeding area, arrival at and departure from the wintering area. However, she arrived later than most Norwegian males at the breeding area. Among the western birds, the single female departed from the Icelandic breeding area and arrived at and departed from the wintering area at the same time as the males, but arrived earlier at the breeding area. She performed the migration faster than most but not all males, using four stopovers in autumn and none in spring, resulting in a relatively short autumn migration and the shortest spring migration. Considering that no significant sex-differences in migration strategies were detected and the small overall sample size (especially for western birds), no sexeffect was considered in the results we present for subsequent tests comparing western and eastern populations. Nevertheless, analyses were repeated with females excluded to see if the results would change qualitatively.

#### Autumn Migration Strategy

Mean departure time from the breeding grounds was slightly earlier for western than eastern birds [xwest = 9 July, xeast = 19 July, t(3, 29) = −2.7, p = 0.012, **Table 1**]. However, mean arrival at the wintering grounds did not differ between the two flyways [xwest = 10 September, xeast = 31 August, t(3, 29) = 0.9, p = 0.400]. Migration duration was slightly longer among western birds [xwest = 73.5 d, xeast = 52.9 d, t(3, 29) = 3.3, p = 0.002, **Figure 2**], while autumn migration distance was 1.5 times longer in western than in eastern birds [xwest = 9,238 km, xeast = 6,069 km, t(3, 29) = 12.1, p < 0.001]. The initial migration leg from the breeding area to the first stopover site was similar in length in western and eastern birds [xwest = 2,916 km, xeast = 2,638 km, t(3, 29) = 1.3, p = 0.218]. During autumn migration of both western and eastern birds, the longest stopover occurred at about 45◦N; western birds staged in (the vicinity of) the Bay of Fundy, Canada, whereas eastern birds used areas near the Black Sea, NW and NE of the Caspian Sea, the Aral Sea, and areas in the vicinity of these seas. The longest stopover was shorter in western than in eastern birds [xwest = 12.4 d, xeast = 22.2 d, t(3, 29) = −2.8, p = 0.008]. It should be noted, however, that some birds had several stopovers around 45◦N. If the duration of the stopovers between 40 and 50◦N is summed for each bird, western birds stage significantly shorter between these latitudes than eastern birds [xwest = 16.2 d, xeast = 24.2 d, t(3, 29) = −2.2, p = 0.037]. In addition to these long stopovers at around 45◦N, shorter stops were made both before and after the main stopover. Significantly more stopovers were made by western birds than eastern [xwest = 3.9, xeast = 2.0, t(3, 29) = 2.6, p = 0.010], but this difference disappeared when correcting for the migration distance [t(3, 29) = 0.7, p = 0.468], indicating that mean migration distance between staging sites was similar between the flyways. The number of days on migration, not classified as staging, was higher among western than eastern birds (xwest = 37.3 d, xeast = 15.5 d, z3, 29 = 5.0, p < 0.001). Overall migration speed was similar between western and eastern birds [xwest = 132 km d−<sup>1</sup> , xeast = 120 km d−<sup>1</sup> , t(3, 29) = 1.4, p = 0.165]. FDR before departure from the breeding grounds effected slower migration speeds, but this effect leveled off at higher migration speeds. In addition, FDR had little effect on differences in migration speed between the two populations: only when FDR would be much lower in eastern breeding areas than in western breeding areas, would the populations differ in migration speed (**Supplementary Figure 1**). Qualitatively identical results were found when excluding females from the autumn migration analysis, except that both breeding departure timing [t(3, 20) = −2.0, p = 0.062] and number of stopovers were near-significantly different between eastern and western birds [z(3, 20) = 1.7, p = 0.096].

#### Maximum Range of First Autumn Leg

The distance from the breeding area to the first stopover was similar in western and eastern birds (see above) and consistent with body mass measurements from the breeding grounds, which were not significantly different between the areas for both males [t(10, 455) = 1.4, p = 0.195] and females [t(5, 162) = −0.5, p = 0.644]. Predicted flight range for model-based averages of body mass are 2,532 km for western and 2,642 km for eastern males and 4,260 km for western and 3,928 km for eastern females. Repeating this for the 75% quantile body mass per flyway translates into 3,214 km for western and 2,675 km for eastern males, and 3,033 km for western and 3,300 km for eastern females. Similarly, flight range for individuals with maximum body mass was predicted to be 4,573 km for western and 5,036 km for eastern males, and 4,646 km for western and 5,036 km for eastern females (**Figure 3**). Flight ranges predicted for mean and 75% quantile body mass were similar to the ranges inferred from our geolocator data, but flight range estimates based on maximum body mass were higher than our longest recorded first legs (**Figure 2**).

#### Spring Migration Strategy

Departure time from the wintering grounds did not differ between the two wintering areas [xwest = 8 May, xeast = 8 May1, t(3, 26) = 0.7, p = 0.511], and arrival at the breeding grounds was also similar between western and eastern birds [xwest = 1 June, xeast = 3 June, t(3, 26) = −1.1, p = 0.293, **Table 1**]. Like in autumn, spring migration distance was ca. 1.7 times longer in western birds as compared to eastern birds [xwest = 9,103 km, xeast = 5,366 km, t(3, 26) = 15.3, p < 0.001, **Figure 2**], but migration duration did not differ [xwest = 37 d, xeast = 32 d, t(3, 26) = 1.5, p = 0.147]. The migration strategy in spring was similar to the autumn migration strategy, with a main stopover at about 45◦N (using the same or nearby areas as in autumn). At the main stopovers, western birds staged about as long as eastern birds [xwest = 4.8 d, xeast = 8.0 d, t(3, 26) = −1.3, p = 0.191]. The number of stopovers was not different between western and eastern birds [xwest = 2.0, xeast = 2.0, z(3, 26) = −0.1, p = 0.900], but when corrected for migration distance, a near-significant difference emerged [z(3, 26) = −1.9, p = 0.063]. Interestingly, two western birds (including the single female) and one eastern male from Slettnes did not stage for as many as 2 or more days anywhere along the spring migration route. Visual


TABLE 1 | Characteristics of autumn and spring migration of red-necked phalaropes migrating between north-eastern North Atlantic breeding areas and the Pacific ("West") and between Fennoscandian-Russian breeding areas and the Arabian Sea ("East").

For the number of stopover/staging sites and the longest stopover/staging period, median and range are given instead of mean ± SD and range. Statistical significance of difference between Western and Eastern birds is indicated as N.S, not significant; \*p < 0.05, \*\*p < 0.01, and \*\*\*p < 0.001.

inspection of their tracks suggest they staged once (western male), twice (western female) or thrice (eastern male) for 1.5 days, i.e., under the threshold of 2 days used in delineation of staging periods. Nevertheless, no more days between staging areas (i.e., days that were not classified as staging periods) were spent by western birds than eastern birds [xwest = 19.6 d, xeast = 19.4 d, z(3, 26) = 0.1, p = 0.883], but this became significant when correcting for migration distance [z(3, 26) = −2.8, p = 0.005]. Thus, western birds covered more distance without stopovers than eastern birds. In spring, their first leg from the wintering grounds was significantly longer than that of eastern birds [xwest = 3,304 km, xeast = 1,052 km, t(3, 26) = 5.9, p < 0.001]. Their overall migration speed was also significantly higher [xwest = 255 km d −1 , xeast = 172 km d −1 , t(3, 26) = 4.3, p < 0.001]. Particularly the migration speed of western birds was affected by FDR before departure from the wintering grounds (**Supplementary Figure 1**). FDR had little effect on the difference in migration speed: the difference in migration speed would become non-significant only if FDR would be low in the Pacific wintering area (**Supplementary Figure 1**). Excluding females from the spring migration analysis led to qualitatively identical results.

Wet/dry data from the Greenlandic bird with the geolocator attached to a leg flag showed that long (>3 h) dry periods, assumed to reflect periods in flight, usually started within an hour of sunset and lasted into the morning of the next day (**Supplementary Figure 2**). Hence, this individual migrated mainly at night, but also performed a non-stop flight of 48 h when crossing the Caribbean in spring. In this individual, the number of flights was much greater than the number of stopovers of two or more days recorded for its migration using the methods outlined above (21 flights vs. 6 staging periods in autumn 2013, 10 flights vs. 2 staging periods in spring 2014, and 21 flights vs. 3 staging periods in autumn 2014). Our results suggest that distances between main staging areas were covered by migrating only by night.

#### Winter Movement Strategies

Winter movement strategies differed markedly between birds in the two wintering areas (**Figures 1**, **4**; **Table 1**). Birds wintering in the Pacific used fewer sites than those wintering in the Arabian Sea [xwest = 2.9, xeast = 8.8, z(3, 26) = −4.926, p < 0.001]. Among western birds, the longest staging period was longer [xwest = 240 d, xeast = 251 d, t(3, 26) = 4.5, p < 0.001]. Their summed great-circle distances between staging areas were shorter [xwest = 1,549 km, xeast = 4,934 km, t(3, 26) = −3.8, p < 0.001]. Excluding females led to qualitatively identical results. Winter movements of western birds mainly occurred at the start and end of the wintering period, when birds moved between the northern Humboldt Current and areas off central America. In the Arabian Sea, itinerant individuals often arrived at the winter quarters off Oman before moving to either off Pakistan or off Somalia (in particular the Gulf of Aden). Such itinerant individuals shifted location up to several times (4–13), before moving to stage in the Persian Gulf or the Gulf of Oman in early spring. More or

Pacific ("West", white boxplots) or migrating from Fennoscandian-Russian breeding sites to the Arabian Sea ("East", gray boxplots): migration distance (A), duration of the migration periods (B), migration speed (C), distance of the first migration leg (D), number of stopovers of at least 2 days (E), and the duration of the longest stopover (F).

less resident individuals in the Arabian Sea staged in the Gulf of Aden, Gulf of Oman or off the coast of Pakistan during the entire winter. In summary, individuals wintering in the Pacific mostly stayed in the same area (residency), whereas individuals wintering in the Arabian Sea showed a larger range of behaviors, from using only a small area to multiple, widely separated areas (itinerancy). Wintering movements appear to correspond to spatio-temporal patterns in ocean primary productivity (**Figure 5**). Primary productivity correlated significantly with number of phalaropes both in the Arabian Sea (z = 6.0, p < 0.001) and in the Pacific (z = 3.0, p = 0.003). However, we did not find evidence for higher primary productivity at arrival at a new site relative to departure from the previous site (z = 0.9, p = 0.369), or for higher primary productivity during staging at a new site compared to the same period at the previous site (z = 0.5, p = 0.607).

#### Repeated Tracks

Data from 2 years were obtained for one individual from Greenland (male), one from Iceland (male), three from Sweden (females), and three from Norway (one female, two males). In addition, data for 3 years were obtained for one male from Norway. Halfway along the final track, four loggers failed prematurely: one in late January and three in February. Locations of and time spent at staging and wintering sites appear fairly consistent between years (**Figure 6**). Individuals wintering in the Arabian Sea mostly had very similar routes between the same main sites within the wintering period, although some exceptions occurred (for example, **Figure 6F**). The number of individuals with repeated tracks was too small for statistical tests of repeatability.

#### Biometrics

In both sexes, wings were longer in western populations [females: xwest = 116.6 mm, xeast = 112.9, t(4, 152) = 4.4, p = 0.013; males: xwest = 111.8, xeast = 109.0, t(9, 415) = 5.1, p < 0.001, **Figure 7**]. In males, tarsi were longer in western populations; in females, this was only near-significant [females: xwest = 21.5, xeast = 20.4, t(3,111) = 3.0, p = 0.056; males: xwest = 21.4, xeast = 20.2, t(7, 297) = 9.7, p < 0.001]. Wings remained significantly longer in western populations in each sex when including tarsus length in the model [females: flyway t(3, 109) = 3.6, p = 0.037; tarsus t(3, 109) = 2.8, p = 0.006; males: flyway t(7, 294) = 4.0, p = 0.005; tarsus: t(7, 294) = 3.2, p = 0.002]. Bill lengths were

not different [females: xwest = 21.9, xeast = 21.4, t(3, 137) = 0.7, p = 0.511; males: xwest = 21.1, xeast = 21.0, t(7, 316) = −0.1, p = 0.914].

# DISCUSSION

Our study demonstrates the existence of two distinct populations of red-necked phalaropes within the Western Palearctic: one that breeds in Fennoscandia and Russia and winters in the Arabian Sea (Indian Ocean) and another that breeds on islands in the north-eastern North Atlantic (Greenland, Iceland and Scotland) and winters in the Pacific. The migration route across the Atlantic and into the Pacific was already suggested in part or wholly by earlier authors (Alerstam, 1990; Smith et al., 2014), but is now confirmed for birds breeding in Greenland and Iceland. The eastern route to the Arabian Sea was already shown for Swedish males (van Bemmelen et al., 2016), but is now also shown for other Fennoscandian and Russian populations. The contrast in the availability of suitable saline stopover habitat along the two routes and the contrast in spatio-temporal variability in ocean productivity between the two wintering areas provided the rare opportunity to study the effect of non-breeding habitat on largescale movement strategies within a species. Both autumn and spring migration strategies differed between the oversea route to the Pacific and the overland route to the Arabian Sea. Rednecked phalaropes wintering in the strongly seasonal Arabian Sea moved between several areas whereas those wintering in the non-seasonal eastern Pacific remained roughly in a single area.

The migration of red-necked phalaropes migrating oversea to the Pacific was more evenly spread over staging periods along the migration route than those heading overland to the Arabian Sea, which showed a prolonged staging period at about 45◦N. According to optimal migration theory, whenever suitable stopover habitat is abundant, short flights and refueling periods should be alternated as this minimizes the costs of carrying fuel loads (Alerstam and Lindström, 1990). Thus, when habitat is uniformly distributed and of equal quality along the migration route, regular spacing of many stopovers, each visited for a similar time period, is expected. However, variation in quality between potential staging sites can induce skipping behavior and unequal staging durations (Gudmundsson et al., 1991; Klaassen et al., 2011). In red-necked phalaropes following the western migration route, variation in quality of habitat is indicated by the non-uniform distribution of staging periods. Whereas no staging areas were identified in the early and late parts, a prolonged stopover was identified at or near the Bay of Fundy, a well-known staging site where our tagged birds remained for about 12 days, somewhat less than the 15 or 20 days estimated

using other methods (Mercier, 1985; Hunnewell et al., 2016). Here, staging red-necked phalaropes fatten up considerably more than what would be expected for a strategy of making short flights with several short stops along the western Atlantic coast (**Figure 3**, Mercier, 1985), indicating a strategy of overloading (Gudmundsson et al., 1991). Despite the high cost of transporting fuel loads, overloading seems a common strategy among waders (Piersma, 1987; Gudmundsson et al., 1991; Alves et al., 2012) and can be expected in time-minimized migrations when fuelling rates at successive stopovers are lower. Overloading at the Bay of Fundy would enable red-necked phalaropes to stop and refuel shortly at sites south to Florida and then to rapidly traverse the relatively unproductive waters of the Caribbean Sea, an area where we identified only two short autumn stopovers among all individuals (**Figure 1**). In contrast to the western route in autumn, overloading does not seem to occur at the main staging areas along the eastern route, as indicated by body masses obtained in Kazakhstan and the Caspian Sea (**Figure 3**, Gavrilov et al., 1983) that seem enough to cover the last stretch to the Arabian Sea. Field observations on fuelling rates at all potential staging sites along the migration route are needed to ultimately understand the migration behavior in the red-necked phalarope.

Migration strategies also differed between the two flyways in spring. A two-day non-stop flight allowed western birds to quickly traverse the unproductive waters of the Caribbean Sea (**Supplementary Figure 2**), after which migration was continued in shorter, mainly nocturnal, flights (Rubega et al., 2000). In contrast, a strategy with several stopovers at scattered wetlands was adopted by eastern individuals to reach the main stopover areas around 45◦N. Spring migrations were also notably fast among red-necked phalaropes migrating from the Pacific to the north-eastern North Atlantic, with few or no stops longer than 2 days. A rapid spring migration is usually explained by selection acting at the breeding grounds, e.g., timely arrival at the breeding grounds to compete for mates or territories, or to achieve optimal timing of reproduction relative to seasonal peaks in food abundance (Nilsson et al., 2013). However, as both populations have similar advantages of timely arrival at the breeding grounds, the difference in spring migration speeds is more likely to be explained by differences along the migration route or at the wintering grounds. Explanations for faster speeds may include the occurrence of favorable winds, such as the westerlies prevailing north of the Tropic of Cancer, and potentially higher fuelling rates along the western route between Florida and the Bay of Fundy, than at similar latitudes along the eastern route.

The differences in migration strategies between red-necked phalarope populations suggest flexible adjustment of migration strategies within species. In contrast, the migration strategies of thrushes, swallows and raptors showed consistent autumn migration strategies among populations following different routes (Delmore et al., 2012; Fraser et al., 2013; Trierweiler et al., 2014), demonstrating rigid migration strategies, as has been suggested in several other studies on passerines (Irwin and Irwin, 2004; Bensch, 2009). The effect size of habitat distribution on movement strategies likely depends on the flexibility of habitat requirements for a given species, the absolute degree to which the distribution of suitable habitat differs between compared routes or sites, and the relative quality (fuel rate and predation risk) between sites along each route. Thus, differences in movement strategies are more likely to be detected in diet or habitat specialists, such as the red knot Calidris canutus (Buehler and Piersma, 2008) or the red-necked phalarope, than for generalist species for which habitat is more widespread, and more evenly distributed along the migration route. As specialists will have less alternative migration routes and strategies when habitat quality changes, they are required to make larger behavioral adjustments than generalist species, rendering specialists particularly vulnerable. Large-scale tracking studies like ours—in particular of specialist species—is thus critical to understand the potential effect of threats to migrants (Sutherland et al., 2012).

The westward migration route to the Pacific is surprising considering that the migration distance to the Pacific is longer than to the Arabian Sea. The longer migration distance and different migration strategy employed by the western birds may explain why the wings of red-necked phalaropes breeding at Greenland, Iceland and Scotland are longer than those of birds breeding in Fennoscandia and Russia. Assuming longer wings are also more pointed, an association between migration distance and wing length is consistent with other studies (Alerstam, 1990; Leisler and Winkler, 2003; Fiedler, 2005; Altizer and Davis, 2010),

and would be explained by selection for aerodynamically more efficient pointed wings for longer migration. As shown by Minias et al. (2015), wing pointedness is a better predictor for migration strategy among wader species than wing length, but only wing length was measured in our study. That variation in wing length of red-necked phalaropes is associated with migration distance is additionally supported by the wing lengths of Canadian populations, which are shorter than our samples from Greenland, Iceland and Scotland and similar to Fennoscandian/Russian red-necked phalaropes. The Canadian populations presumably winter in the eastern Pacific and do not have to cross the Atlantic (Reynolds, 1987).

Red-necked phalaropes wintering in the Pacific showed only minor movements, whereas most individuals wintering in the Arabian Sea moved around considerably throughout winter (van Bemmelen et al., 2016). Wintering movements have received growing attention in recent years, especially in species wintering in the Neotropics and in Africa (Fraser et al., 2012; Heckscher et al., 2015; Norevik et al., 2019). Also in seabirds, wintering movements occur (Phillips et al., 2005; Hedd et al., 2012; Orben et al., 2015). Although we are not aware of studies contrasting wintering movement strategies between distinct wintering areas in a single species, differences in wintering movement strategies between populations of the same species with overlapping wintering areas have been reported in terrestrial species (Stutchbury et al., 2016; Kolecek et al., ˇ 2018). Itinerant strategies occurred among great reed warblers Acrocephalus arundinaceus wintering in sub-Saharan non-breeding sites; with translocations over larger distances by those breeding and wintering further eastwards (Kolecek et al., 2018 ˇ ). Wintering movement strategy of common swifts Apus apus wintering in Africa appears to correlate with breeding origin, with both Swedish and Dutch birds sharing a major wintering area in the Congo basin, but only Dutch birds vacating this area midwinter to make a round trip to south-east Africa (Åkesson et al., 2012; Klaassen et al., 2014). In red-necked phalaropes, the occurrence of different wintering movement strategies within the same species suggests flexibility of movement behavior within the species.

The difference in movement strategies between red-necked phalaropes wintering in the Pacific and in the Arabian Sea is consistent with our expectation of how birds should respond to differences in spatio-temporal variation in primary productivity between the two wintering areas. Primary productivity shows only minor seasonal variation in the Pacific (Chavez and Messié, 2009), but large spatio-temporal variability driven by monsoon winds in the Arabian Sea (Longhurst, 2006). Although phalaropes mainly occupied areas with high primary productivity, movements did not appear to result in higher experienced primary productivity. In great reed warblers

Iceland (B), Norway (C–E,I), and Sweden (F–H), with breeding sites indicated by red stars. Shaded circles show staging areas in the first (red), second (blue), or third (orange) year of tracking, with circle size proportional to staging duration and staging areas connected by great-circle lines (which do not necessarily represent routes taken) for autumn and early winter (solid lines) and late winter and spring (dotted lines).

and pallid swifts Apus pallidus, intra-tropical movement of individuals is explained by improved conditions (inferred from remotely sensed indices of food availability) at destinations in comparison to where they initially staged (Kolecek et al., ˇ 2018; Norevik et al., 2019). For red-necked phalaropes, however, remotely sensed primary productivity may not be a reliable proxy of food availability, as red-necked phalaropes feed on zooplankton which may not be directly related to primary productivity (Cramp and Simmons, 1983; Brown and Gaskin, 1988). Data on distribution and abundance of zooplankton is however scant and scattered in both time and space. A further complication when linking movement to food availability in our study species is the spatial resolution of both the tracking data (which in the case of geolocators deployed on seabirds is ca. 185 km; Phillips et al., 2004) and the remotely sensed productivity (cell size

of 1/6 a degree, approximately 8 km in the Arabian Sea). Both spatial resolutions are coarser than the small-scale ephemeral phenomena, such as thermal oceanic fronts, that may be targeted by red-necked phalaropes at sea (Haney, 1985; Brown and Gaskin, 1988).

The large variation between individuals in the number of movements within the Arabian Sea suggests that itinerancy is facultative, as has been shown for several other species wintering in the Neotropics or Africa (Stutchbury et al., 2016; Kolecek et al., ˇ 2018). At the same time, wintering movement patterns seem consistent within individual red-necked phalaropes between years (**Figure 6**). High individual consistency in non-breeding movement patterns appears to be widespread among seabirds (Dias et al., 2011; McFarlane Tranquilla et al., 2014; van Bemmelen et al., 2017) and is hypothesized to develop during an explorative phase in the pre-breeding years (Pulido, 2007; Guilford et al., 2011; Senner et al., 2015). If this is the case, then given that the pre-breeding period lasts several years in most seabirds (Weimerskirch, 2002), but only 1 year in red-necked phalaropes(Schamel and Tracy, 1988), individuality in itineraries may arise in a single non-breeding season.

Migratory divides can co-occur in areas where multiple species have secondary contact zones after recolonization of northern breeding areas from southern ice age refugia, or in areas that are at similar migration distances from suitable wintering areas or at locations that present a barrier to migration (Alerstam and Gudmundsson, 1999; Newton, 2010; Møller et al., 2011). For example, a migratory divide at a longitude of about 100◦E is shared among many Palearctic birds breeding in the Arctic (Alerstam and Gudmundsson, 1999) and central Europe or Scandinavia among European passerines (Møller et al., 2011). Among Nearctic birds, no area with co-occurring migratory divides has been identified, but several species with a circumpolar breeding distribution have a migratory divide in Arctic Canada and/or have evolved into different (sub)species that migrate either within the New World or to the Old World. However, the position of the migratory divide among Western Palearctic red-necked phalaropes is not shared with other species. The south-eastward migration of Scandinavian and Russian populations of phalaropes is shared with a small number of species, such as broad-billed sandpiper Calidris falcinellus, little ringed plover Charadrius dubius and redspotted bluethroat Luscinia svecica svecica (Verkuil et al., 2006; Hedenström et al., 2013; Lislevand et al., 2015), but the westward migration of populations from the north-eastern North Atlantic to the Pacific is unique. The migration route to the Pacific is highly suggestive of a biogeographic legacy, wherein the Nearctic breeding population expanded eastwards while retaining the migration route to the Pacific wintering area. Considering no geographic plumage or biometry variation has been described for the red-necked phalarope (Cramp and Simmons, 1983), this hypothesis can best be tested in a future population genetic study. Migratory divides delineating the breeding populations migrating to the East Indies are still unknown (Mu et al., 2018) but could be revealed by tracking Red-necked Phalaropes from breeding areas in Siberia and the Nearctic. The divide between birds wintering in the East Indies and eastern Pacific will indicate what part of the Canadian population, besides the populations from the north-eastern North Atlantic, may have been affected by the huge but unexplained population crash observed in the Bay of Fundy in the 1980s (Nisbet and Veit, 2015). Siberian and Nearctic breeding populations may migrate oversea as well as overland to the wintering areas in the East Indies and eastern Pacific (Jehl, 1986; Rubega et al., 2000; Mu et al., 2018), providing further opportunities to test the effect of habitat on migration strategies. The observed migration pattern of Western Palearctic birds also raises the question of why red-necked phalaropes apparently do not winter in substantial numbers in productive waters in the tropical Atlantic, such as the Canary Current (Camphuysen and van der Meer, 2005). Potential explanations for their absence in the Atlantic include a lack of suitable stopover sites along the route to reach these areas, or competition with the larger but ecologically similar gray phalarope Phalaropus fulicarius, with which no overlap in wintering area occurs in the Pacific (Cramp and Simmons, 1983). Despite being absent from the tropical Atlantic, the differences in both migration strategies and wintering movement strategies between populations indicate adjustment of movement strategies to habitat distribution within the species.

By tracking individual migratory birds from two distinct populations and capitalizing on international collaboration, our study provides a rare demonstration of how habitat distribution shapes large-scale movement strategies over the entire nonbreeding period. The added value to previous studies is that we (1) demonstrate population differences in movement strategies within a species, (2) base our results on tracking data at the individual level, and (3) show that movement strategies are largely consistent within individuals. The individual consistency of movement strategies in adults suggests variation between individuals may best be regarded as the result of developmental plasticity (Piersma and Drent, 2003; Gill et al., 2014). Developmental plasticity may be an important mechanism for population-level adjustments of movement strategies to habitat distribution in the red-necked phalarope. Thus, mapping the ontogeny of movement strategies and how individual strategies affect fitness will be key to understanding the origin of flexibility in movement strategies. Understanding flexibility will be important when considering a species' ability to respond to (climate-induced) environmental change (Parmesan and Yohe, 2003; Chen et al., 2011). Given the decline of many populations of long-distance migratory species (Møller et al., 2008; Both et al., 2010), more studies like ours are needed to understand how species characteristics affect the flexible adjustment of movement strategies to habitat availability at individual and population-levels.

# DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

# AUTHOR CONTRIBUTIONS

YK and JG-S initiated our collaboration. RvB, IT, RR, and RK conceived the focus of the manuscript. RvB and RR analyzed the data. RvB wrote the first draft of the manuscript. All authors collected geolocator and/or biometric data and provided critical feedback on the manuscript.

# FUNDING

RvB was funded by the Netherlands Organization for Scientific Research (project number 866.13.005) and supported by the LUVRE project. We also thank the European Union (FP7- PEOPLE-2013-CIG, 618841) for funding our research. BM and SH were supported by Fram Center and RR was supported by a postdoctoral contract of the Juan de la Cierva program, from the Spanish Ministerio de Ciencia e Innovación (JCI-2012-11848, CGL2013-42585-P, CGL2016-78530-R). Work at Constable Pynt by OG and LB (part of program 1,036 Interactions) were supported by the French Polar Institute, IPEV. Work at Fetlar was funded by the RSPB and the Shetland Ringing Group. AL received financial support from the Academy of Finland (grant 275606). AS was supported through the grant of RFBR #18-05-60261 Arctic, JAA from FCT (SFRH/BPD/91527/2012) and BÞ from RANNIS (152470-052) and University of Iceland equipment fund. KV received support from Kone foundation.

# ACKNOWLEDGMENTS

Our study would not have been possible without the help and enthusiasm of the many fieldworkers. Ammarnäs: Piet Admiraal, Christian Brinkman, Michiel Elderenbosch, Christian Hoefs, Vincent Hin, Johannes Hungar, Guido Keijl, Fons de Meijer, Morrison Pot, and Bram Ubels. Also, thanks to Martin Green and Åke Lindström for help in various ways. Erkuta: Jasper Koster, Brigitte Sabard, Vladimir Gilg, and Vadim Heuacker. Enontekiö: The Kilpisjärvi waterbird monitoring expedition: Daniel Burgas, Heikki Eriksson, Sara Fraixedas, Sanna Mäkeläinen, Hanna Laakkonen, Petteri Lehikoinen, Mari Pihlajaniemi, Jarkko Santaharju, and Jenni Santaharju. Slettnes: Daniel van Denderen, Jan van Dijk, Daan Liefhebber, Maria van Leeuwe, Morrison Pot, Marc van Roomen, Janne Schekkerman, Cees Tesselaar, and Rinse van der Vliet. Tobseda: Jasper Koster, Thomas Lameris, Stefan Sand, and Kees Schreven. Constable Pynt (part of program 1,036 Interactions, supported by the French Polar Institute, IPEV): Brigitte Sabard, Adrian Aebischer, Antoine Dervaux, Eric Buchel, Vadim Heuacker, Mickael Sage, and Vladimir Gilg. Fridland (Flói): Verónica Méndez, Camilo Carneiro, Harry Ewing, and Einar Gunnlaugsson. Fetlar: George Petrie, Pete Ellis, Phil Harris, and Roger Riddington. Francine Mercier kindly provided the raw body mass measurements of phalaropes from her 1980s study in the Bay of Fundy. Thanks

#### REFERENCES


to Christoph Mayer and Felix Liechti (Swiss Ornithological Institute) for providing loggers and to Foto Fennica for funding the Finnish loggers. Thanks to Linda McPhee for improving the readability of the manuscript, and to the editor and two referees for helpful comments that improved the quality of the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00086/full#supplementary-material


Newton, I. (2010). The Migration Ecology of Birds. London: Academic Press.


the Scandinavia-Arabian Sea connection. J. Avian Biol. 47, 295–303. doi: 10.1111/jav.00807


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The handling editor declared a shared affiliation, though no collaboration, with two of the authors, BM and SAH, at the time of review.

Copyright © 2019 van Bemmelen, Kolbeinsson, Ramos, Gilg, Alves, Smith, Schekkerman, Lehikoinen, Petersen, Þórisson, Sokolov, Välimäki, van der Meer, Okill, Bolton, Moe, Hanssen, Bollache, Petersen, Thorstensen, González-Solís, Klaassen and Tulp. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Variation From an Unknown Source: Large Inter-individual Differences in Migrating Black-Tailed Godwits

Mo A. Verhoeven<sup>1</sup> \*, A. H. Jelle Loonstra1†, Nathan R. Senner 1,2†, Alice D. McBride<sup>1</sup> , Christiaan Both<sup>1</sup> and Theunis Piersma1,3

*<sup>1</sup> Conservation Ecology Group, Groningen Institute for Evolutionary Life Sciences, University of Groningen, Groningen, Netherlands, <sup>2</sup> Department of Biological Sciences, University of South Carolina, Columbia, SC, United States, <sup>3</sup> NIOZ Royal Netherlands Institute for Sea Research, Department of Coastal Systems and Utrecht University, Texel, Netherlands*

#### Edited by:

*Carlos Alonso Alvarez, Spanish National Research Council (CSIC), Spain*

#### Reviewed by:

*James Dale, Massey University, New Zealand Wendy Hood, Auburn University, United States*

#### \*Correspondence:

*Mo A. Verhoeven m.a.verhoeven@rug.nl orcid.org/0000-0002-2541-9786*

*†A.H. Jelle Loonstra orcid.org/0000-0002-5694-7581 Nathan R. Senner orcid.org/0000-0003-2236-2697*

#### Specialty section:

*This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution*

> Received: *16 October 2018* Accepted: *28 January 2019* Published: *26 February 2019*

#### Citation:

*Verhoeven MA, Loonstra AHJ, Senner NR, McBride AD, Both C and Piersma T (2019) Variation From an Unknown Source: Large Inter-individual Differences in Migrating Black-Tailed Godwits. Front. Ecol. Evol. 7:31. doi: 10.3389/fevo.2019.00031* Variation in migratory behavior is the result of different individual strategies and fluctuations in individual performances. A first step toward understanding these differences in migratory behavior among individuals is, therefore, to assess the relative contributions of inter- and intra-individual differences to this variation. We did this using light-level geolocators deployed on the breeding grounds to follow continental black-tailed godwits (*Limosa limosa limosa)* throughout their south- and northward migrations over multiple years. Based on repeated tracks from 36 individuals, we found two general patterns in godwit migratory behavior: First, migratory timing in black-tailed godwits varies mostly because individual godwits migrate at different times of the year. Second, individuals also exhibit considerable variation in timing within their respective migratory windows. Although the absolute amount of inter-individual variation in timing decreased over the course of northward migration, individual godwits still arrived at their breeding grounds across a span of more than 5 weeks. These differences in migratory timing among individuals are larger than those currently observed in other migratory bird species and suggest that the selective forces that limit the variation in migratory timing in other species are relaxed or absent in godwits. Furthermore, we could not attribute these individual differences to the sex or wintering location of an individual. We suggest that different developmental trajectories enabled by developmental plasticity likely result in these generally consistent, life-long annual routines. To investigate this possibility and to gain an understanding of the different selection pressures that could be acting during migration and throughout a godwit's life, future studies should track juvenile godwits and other migratory birds from birth to adulthood while also manipulating their spatiotemporal environment during development.

Keywords: migratory behavior, repeatability, shorebird, developmental plasticity, light-level geolocators

# INTRODUCTION

Long-term mark-recapture studies and the rapid development of tracking technologies have revealed the migratory patterns of many avian migrants (Berthold, 2001; Newton, 2008; Bridge et al., 2011). These migratory patterns are always characterized by some degree of variation, such as individuals migrating at different times and toward different destinations (Berthold, 2001; Newton, 2008). Such populationlevel variation in migratory patterns is the result of both inter- and intra-individual differences (Vardanis et al., 2011; Conklin et al., 2013). The amount of consistent variation among individuals (i.e., inter-individual variation) is subject to selection: only those strategies that ensure survival will remain in the population and over the long-term those strategies that maximize fitness will be selected (Alerstam et al., 2003).

For instance, the timing of arrival on the breeding grounds, in particular, is thought to be under strong selection in migratory birds in order for individuals to procure high-quality breeding territories and breed in synchrony with consistently timed local resource peaks (Alerstam et al., 2003). Inter-individual variation in this component of migration is therefore usually expected to be small (Kokko, 1999; Bety et al., 2004; Both et al., 2006). However, selection can also favor multiple canalized strategies and thus lead to large inter-individual variation within a population. This can happen as a result of fluctuating environmental conditions (e.g., serial residency; Cresswell, 2014) or frequency-dependent processes (e.g., partial and differential migration; Lundberg, 1988; Chapman et al., 2011).

Most environments, however, are neither entirely consistent nor entirely predictable, which can affect the consistency with which individuals are able to perform their migrations (e.g., Studds and Marra, 2011). In addition, an individual can exhibit different migration strategies with increasing experience (e.g., individual improvement; Sergio et al., 2014), because the environment requires flexibility (e.g., nomadism; Pedler et al., 2018), or because the environment allows flexibility (e.g., the absence of carry-over effects; Senner et al., 2014). Differences in an individual's migratory behavior across years (i.e., intra-individual variation) therefore also contribute to migration variation at the population level (sensu Conklin et al., 2013). Thus, the amount of observed variation in migratory behavior within a population can result from (1) differences among individuals, which are consistent, and (2) differences within individuals, which are expected to vary according to the predictability and consistency of the environment and the individual's ability to respond to environmental changes.

A first step toward understanding why migratory patterns vary within populations is to consider the relative contributions of both inter- and intra-individual variation to the amount of variation at the population level (Senner et al., 2015b). To do this, the performances of multiple individuals must be measured across multiple years. These repeated measures allow for the calculation of repeatability (r), which reflects the proportion of population-level variation that can be attributed to interindividual differences (Nakagawa and Schielzeth, 2010). The non-repeatable fraction (1-r) therefore reflects the contribution of intra-individual variation. However, a high r value—where inter-individual variation is proportionally larger than intraindividual variation—can result from either large variation among individuals, high consistency within individuals, or both (Conklin et al., 2013).

Black-tailed Godwits (Limosa limosa limosa; hereafter "godwits") are long-distance migratory birds that breed in Europe and have a large non-breeding range—a quarter of the population winters north of the Sahara on the Iberian Peninsula (Márquez-Ferrando et al., 2014), while the majority winters south of the Sahara in the Sahel zone of West Africa (Hooijmeijer et al., 2013; Kentie et al., 2017). There is also large variation in the migratory timing of godwits (Lourenço et al., 2011; Senner et al., under review). This is especially true during northward migration: at the population level, variation in departure dates from the African wintering grounds and Iberian stopover sites span more than 10 weeks, and even arrival at the breeding grounds can vary by up to 5 weeks (Lourenço et al., 2011; Senner et al., under review). Inter-individual differences play an unexpectedly important role in this considerable variation accounting for the majority of observed variation in departure to the north (r = 0.76) and nearly half the observed variation in arrival on the breeding grounds (r = 0.49; (Senner et al., under review).

Because selection determines the amount of variation in migratory timing among individuals, this raises two major questions about godwit migration: (1) Why isn't the role of inter-individual variation small, as it is in most other longdistance migrants (Newton, 2008; Stanley et al., 2012; Conklin et al., 2013)? and (2) What is the source of this surprisingly large amount of inter-individual variation? (Senner et al., under review) address the first question, suggesting that the large variation among individuals in godwits exists because of relaxed selection on migratory timing. This study addresses the second question, investigating the source of inter-individual variation in migratory timing in godwits. We describe the timing of migration and the wintering location of 70 individuals, of which 36 individuals were followed for multiple years. We calculate the repeatability of migratory timing and wintering location to assess whether individuals consistently winter either north or south of the Sahara and to identify the relative contributions of interand intra-individual variation to total population-level variation. Then, we use sex and wintering location to explain some, but not all, of the large amount of inter-individual variation. Ultimately, we are unable to account for the remaining variation that occurs between individuals and so we discuss in detail other sources such as differences in developmental trajectories—that may be contributing to this phenomenon.

# MATERIALS AND METHODS

#### Fieldwork

Fieldwork occurred from March through July 2012–2017, in our long-term study area in southwest Fryslân, The Netherlands (Senner et al., 2015a). This area, which encompasses 12,000 ha, stretches from 53.0672◦N, 5.4021◦E in the north to 52.8527◦N, 5.4127◦E in the south, and from 52.9715◦N, 5.6053◦E in the east to 52.8829◦N, 5.3607◦E in the west. In this area we located godwit nests and used the flotation method (Liebezeit et al., 2007) to determine lay dates. To reduce the chance of nest abandonment and increase the chance of capturing an adult, we caught breeding adults toward the end of their incubation period (24 ± 3.84 days after laying). In each of the six field seasons, we outfitted 42– 69 individuals with geolocators; this corresponded to 26–61% of all captured adults each year. We used geolocators from Migrate


TABLE 1 | Summary of the linear mixed effect models evaluating whether the timing of each crossing on both south- and northward migrations was a result of the sex (male/female) or wintering location (north/south of Sahara) of an individual.

*We also included individual and year as random effects. Significant p-values for fixed effects are in bold. The marginal R*<sup>2</sup> *, conditional R*<sup>2</sup> *, and sample size are also given for all models. # r denotes the number of individuals with repeated measurements.*

*<sup>a</sup>Reference level for Sex is female.*

*<sup>b</sup>Reference level for Sahara is North.*

Technology, Ltd: the 0.65 g Intigeo W65A9 model in 2012–2013, and the 1g Intigeo C65 model thereafter.

These geolocators were attached to colored flags and placed on the tibia. The total weight of this attachment was ∼3.3 g in 2012–2013 and ∼3.7 g in 2014–2017, resulting in a loading factor of 1–1.5% of an individual's body mass at capture. For molecular sexing, we took ∼30 µl of blood from the brachial vein of each individual. We were able to use these blood samples to sex 67 of the 70 individuals; we sexed the remaining 3 individuals based on morphological measurements as in Schroeder et al. (2008). In the years following geolocator deployment, we put considerable effort into recapturing godwits carrying geolocators. Recapture probability was nonetheless low; over the course of six field seasons, we managed to recapture 92 out of 305 deployed geolocators. Some geolocators did not record full annual cycles. For this reason, our data contains more tracks of southward migration (n = 117) than northward migration (n = 79; see **Table 1** for more details).

#### Analyzing Geolocator Data

Using package "BAStag" (Wotherspoon et al., 2016) in Program R (R Core Team, 2017), we started with the function "preprocesslight," which automatically detects sunrises and sunsets. We set the threshold light value to 2. Next, we visually inspected the slope of each sunrise and sunset and excluded those slopes that were strongly biased over time, i.e., showed abrupt changes in light level (Rakhimberdiev et al., 2016). We then used package "FLightR" (Rakhimberdiev et al., 2017) to reconstruct the annual schedules of godwits from this light-level data. Detailed examples of this analytical routine using our own godwit data can be found in Rakhimberdiev et al. (2016, 2017). These examples use data from a godwit that wintered north of the Sahara (≥28◦N). Our sample also includes birds that wintered south of the Sahara (<28◦N), with the only difference between the published examples and our own analyses being that we constrained the spatial extent of the particle filter to 18◦W−13◦E and 11–57◦N instead of the 14◦W−13◦E and 30– 57◦N boundaries used by Rakhimberdiev et al. (2016, 2017).

Next, using the FLightR function "find.times.distribution" we estimated when individual godwits crossed certain arbitrary spatial boundaries. For this, we designated eight spatial boundaries which were spaced 4◦ of latitude apart across the entire godwit migration corridor, from 52 (the breeding grounds) to 20◦N (just north of the southernmost African wintering grounds; **Figure 1**). We used the same eight spatial boundaries for both south- and northward migrations. In our analyses, we excluded the crossing of the spatial boundary at 36◦N (the Strait of Gibraltar) because we could not distinguish between birds stopping in northern Morocco and birds stopping in southern Spain. In 26 out of 79 cases, we were also unable to estimate arrival at the breeding grounds (≥52◦N) using this method. In these cases, migration and arrival on the breeding grounds coincided with the spring equinox, a period during which it is difficult to reliably estimate latitude from light-level data (Fudickar et al., 2012; Rakhimberdiev et al., 2015). Longitude, however, is much less affected (Rakhimberdiev et al., 2015, 2016), and godwits fortunately migrate from west to east as well as from south to north during their northward migration. For these

26 cases we could therefore use a spatial boundary of 5◦E to estimate arrival on the breeding grounds (sensu Rakhimberdiev et al., 2015, 2016). Similarly, in 16 of 79 cases, we were unable to estimate the crossing of 48◦N due to the spring equinox, and used a boundary of 0.75◦E instead.

Lastly, we used the FLightR-function "stationary.migration. summary" to provide an overview of the stationary periods occurring throughout an individual's annual schedule. This allowed us to infer whether an individual wintered north (≥28◦N) or south of the Sahara (<28◦N).

#### Analyzing Annual Schedules

We first grouped individual godwits according to where they spent the non-breeding period. We considered individuals that crossed the Sahara (<28◦N) during migration to have wintered "South" of the Sahara, and individuals that never crossed the Sahara (≥28◦N) to have wintered "North" of the Sahara. To determine whether individuals were flexible in their overwintering behavior, we looked at whether they consistently wintered on the same side of the Sahara from year to year. We also used a binomial Generalized Linear Model (GLM) with wintering area as the dependent variable and sex as the independent variable to test whether the proportion of males and females that crossed the Sahara differed. Because some individuals winter north and others winter south of the Sahara, our sample sizes differed among spatial boundaries (**Table 1**).

For both south- and northward migrations, we calculated population-level variation in the timing of each crossing of our arbitrary spatial boundaries. We did this by calculating the difference between the earliest crossing and all subsequent crossings, and then calculating the 5 year mean and standard deviation of this difference (**Figures 2A,B**). We then used a GLM to test whether the amount of variation differed between the spatial boundaries. Our data includes repeated measures of 26 individuals followed for 2 years, nine individuals followed for 3 years, and one individual followed for 4 years. We calculated individual variation in the timing of crossings by identifying the largest difference between the crossings of each individual over the course of the time that they were tracked. Next, we calculated the mean and standard deviation across all individuals (**Figures 2C,D**) and used a GLM to test for differences in the amount of intra-individual variation among spatial boundaries. When differences among spatial boundaries were found, we used a Tukey post-hoc test with a 95% confidence level to establish how the timing differed between pairs of boundaries. Additionally, we calculated the repeatability of each barrier crossing during south- and northward migration (**Figures 2E,F**). We did this by including individual as a random effect in the linear mixed model method of the function "rpt," which is part of the R package "rptR" (Stoffel et al., 2017). To evaluate whether individuals consistently shifted their timing earlier or later over the course of our study, we plotted for every spatial boundary the first observed timing of crossing vs. the last observed timing of crossing for each individual godwit (**Figure S1**).

Finally, using the R package "lme4" (Bates et al., 2015), we fitted linear mixed effect models for each crossing during both south- and northward migrations. In these models we used the timing of crossing of a spatial boundary as the response variable, and the wintering location (north/south of Sahara) and sex of an individual (male/female) as fixed effects. We also included individual and year as random effects. We assessed whether the fixed effects improved the model significantly by means of a likelihood ratio test. We also calculated the marginal R 2 to describe the amount of variance that is explained by the fixed effects using package MuMIn (Barton, 2016), following the method established by Nakagawa and Schielzeth (2013).

#### RESULTS

Among all individuals (n = 70), 30 females and 26 males crossed the Sahara (80%), whereas 9 females and 5 males did not (20%). The proportion of males and females that crossed the Sahara did not differ (χ <sup>2</sup> = 0.53, df = 1, p = 0.47). Of the 36 individuals for which we obtained repeated measures—23 females and 13

during southward migration 2012–2016 and (B) during northward migration 2012–2017. Boxplots show 25, 50, and 75th percentiles; whiskers indicate 5 and 95th percentiles (day 1 = earliest observation for each crossing). (C) Intra-individual variation in the timing of southward migration and (D) of northward migration. Boxplots show 25, 50, and 75th percentiles; whiskers indicate the entire range of values. (E) Individual repeatability of timing on southward migration and (F) on northward migration. Plots show the repeatability estimate and the 95% confidence interval. The different colors are used for visual purposes only.

males—all 36 were consistent in wintering either north (n = 7) or south (n = 29) of the Sahara over the course of the time they were tracked.

The smallest difference among individuals between the earliest and latest crossings during southward migration was 62 days for the barriers at both 48 and 44◦N, whereas the largest difference was 106 days for crossing 20◦N (**Figures 1B**, **2A**; **Figure S2**). During northward migration, the smallest difference was 38 days for crossing 52◦N, and the largest difference was 153 days for crossing 28◦N (**Figures 1C**, **2B**; **Figure S2**). The average amount of variation among individuals did not vary among spatial boundaries during southward migration [F(7,832) = 0.96; p = 0.46; **Figures 1B**, **2A**; **Figure S2**], but did vary during northward migration [F(7,596) = 108.4; p < 0.001; **Figures 1C**, **2B**; **Figure S2**]. A Tukey post-hoc test with a 95% confidence level found that population-level variation was greatest for crossing 20–32◦N (the Sahara), decreased for crossing 40◦N (departing the Iberian Peninsula), and was smallest for crossing 44–52◦N (France, Belgium, and The Netherlands; **Figures 1C**, **2B**; **Figure S2**).

Intra-individual differences between years for timing at the same latitude varied from 0–73 days during southward migration (**Figure 2C**). The biggest differences, 62 and 73 days, occurred when crossing 40◦N (**Figure 2C**). This was due to two individuals stopping over north of this boundary 1 year and south of it the other. The intra-individual differences in timing between those years includes the durations of these stopovers and is deceptively large as a result. During northward migration, intraindividual differences varied from 0 to 42 days (**Figure 2D**). The biggest difference, 42 days when crossing 40◦N, was again the result of an individual stopping over on opposite sides of the boundary in different years (**Figure 2D**). Thus, the average amount of intra-individual differences did not vary between spatial boundaries during either southward [F(7,252) = 0.70; p = 0.68; **Figure 2C**] or northward migration [F(7,186) = 0.57; p = 0.78; **Figure 2D**]. Furthermore, individuals did not consistently shift their timing earlier or later over the course of our study during either southward or northward migration (**Figure S1**).

Individual repeatability during southward migration varied from 0.1–0.6 and was highest when crossing the Sahara (20– 32◦N; **Figure 2E**). During northward migration, repeatability varied between 0.3–0.9 and was again highest when crossing the Sahara (**Figure 2F**). Repeatability therefore increased over the course of southward migration and decreased over the course of northward migration (**Figures 2E,F**). This could be the result of individuals wintering south of the Sahara being more consistent in their timing than individuals wintering north of the Sahara. However, the amount of intra-individual variation is non-significantly larger during Sahara crossings (**Figures 2C,D**); this indicates that the repeatability is higher because interindividual differences are larger for Sahara crossings, not because these individuals are more consistent.

During southward migration, males departed the Netherlands (52◦N) on average 5 d earlier than females (χ <sup>2</sup> = 4.47, df = 1, p = 0.03, n = 117; **Table 1**). This difference held true for the crossing of 48 and 44◦N, but not for more southerly boundaries (40–20◦N; p > 0.1; **Table 1**). Whether or not an individual crossed the Sahara did not explain a significant amount of the variation in the timing of southward migration (**Table 1**). During northward migration, neither the sex of the individual (p > 0.1 for all spatial boundaries; **Table 1**) nor whether it crossed the Sahara (p > 0.05; **Table 1**) explained a significant amount of the variation in their timing. Thus, the amount of variance explained by our fixed effects—as indicated by the marginal R <sup>2</sup>—was never higher than 0.06 (**Table 1**). The marginal R <sup>2</sup> was highest when crossing 44–52◦N during both southward and northward migration (**Table 1**). Not surprisingly, these were the southward migration boundary crossings for which a significant amount of the variation was explained by sex, and the northward migration crossings for which at least some of the variation (p = 0.06; **Table 1**) was explained by whether an individual crossed the Sahara or not.

# DISCUSSION

We found that the large amount of population-level variation in the migratory timing of continental black-tailed godwits is mostly the result of individual godwits exhibiting consistent differences from one another in the timing of their movements during both north- and southward migration. In addition, we found that a given individual can exhibit considerable flexibility while still adhering to its own particular schedule. These inter-individual and intra-individual differences in timing are large compared to other species of migratory birds (e.g., Alerstam et al., 2006; Vardanis et al., 2011; Stanley et al., 2012; Conklin et al., 2013). This suggests that the selective forces that limit the variation in migratory timing in other species are likely relaxed or absent in godwits (see also Senner et al., under review) and that the unexplained but consistent differences among godwits may be the result of different developmental trajectories.

#### Population Variation

We found that approximately 80% of black-tailed godwits breeding in Fryslân cross the entirety of the Sahara Desert during migration, whereas 20% do not cross any portion of it, and that this was a consistent behavior across years. Furthermore, although the repeatability in the timing of flights across the Sahara was higher than that of other migratory flights, this was driven by the relative influence of inter-individual variation, which was also highest at this point in the migration. In other words, individual godwits consistently time their Sahara crossings differently from one another. This suggests both that this major ecological barrier is traversable for a long period of time and that other temporal constraints—for example, the availability of resources at sites to the north of the Sahara do not influence the time at which individuals make this flight (Moore and Yong, 1991; Baker et al., 2004). This is surprising, as the crossing of the Sahara during both south- and northward migrations in most other migratory bird species takes place over a shorter period of time (e.g., Vardanis et al., 2011, max = 64 days; Lindström et al., 2015, max = 33 days; Briedis et al., 2016, max = 25 days; Jacobsen et al., 2017, max = 35 days; Ouwehand and Both, 2017, max = 37 days), although Sergio et al. (2014) found that black kites (Milvus migrans) also cross the Sahara over a 5 month period as a result of the sequential departure from the wintering grounds by different age classes. It is not clear why the Sahara crossing of the other migratory birds appears to be under generally stronger temporal selection, but these species must either face stronger temporal constraints in relation to the crossing itself or during subsequent events in their annual cycle.

Once past the Sahara Desert during northward migration, inter-individual variation in timing decreased toward the breeding grounds and was smallest when crossing the region between 44–52◦N (France, Belgium and The Netherlands). Levels of intra-individual variation did not decrease simultaneously, but were smallest when crossing 44◦N. As a result, the repeatability of timing for these stages (40–52◦N) differed from zero only when crossing 44◦N. Individuals are thus relatively consistent in their timing of departure from the Iberian Peninsula (44◦N), but not their timing of arrival at the breeding grounds. Given that the intra-individual variation increased for the two most northerly crossings, 48 and 52◦N, this is probably due to the flexible adjustment of their migratory schedule in response to environmental conditions encountered en route. For instance, in 2013, a rare spring snowstorm delayed the arrival of godwits to the breeding grounds by an average of 19 days (Senner et al., 2015a).

Both the tightening of migratory schedules toward the breeding grounds (e.g., Hasselquist et al., 2017; Wellbrock et al., 2017) and the flexible adjustment of migratory schedules (e.g., Nuijten et al., 2014; Briedis et al., 2017) have been shown in other migratory bird species. Nonetheless, the arrival of godwits at the breeding grounds spans more than 5 weeks—which is a larger range than that currently observed in other migratory bird species (e.g., Senner et al., 2014; Lindström et al., 2015; Briedis et al., 2016; Jacobsen et al., 2017; Ouwehand and Both, 2017). Potentially, the absence of a strong temporal constraint on arrival at the breeding grounds is what allows godwits to cross the Sahara over such a long period of time. If this is true, other species making similar flights, but over a shorter period of time, may not face stronger temporal constraints for the crossing of the Sahara itself, but rather for their arrival at the breeding grounds. Accordingly, in other species, individuals from different breeding populations wintering in the same region of sub-Saharan Africa depart their wintering areas at different times, and these departure windows are correlated with their breeding-site specific reproductive timing (Briedis et al., 2016; Ouwehand et al., 2016).

# The Relative Importance of Intra-Individual Variation

Although each godwit appeared to migrate within its own migratory window, individual godwits also displayed considerable flexibility in their timing of migration within their own specific windows. This degree of flexibility was not the result of directional changes in migratory timing, and is greater than the amount of intra-individual variation reported in other studies (e.g., Conklin et al., 2013, <5 days; Senner et al., 2014, <5 days; Hasselquist et al., 2017, <20 days; Wellbrock et al., 2017, <15 days). The relatively large intra-individual variation during migration can therefore be interpreted as an individual decision that balances migrating at a specific time and leaving when endogenous and exogenous conditions are best (e.g., Senner et al., 2015a). For example, crossing the Sahara is possible over a long period of time, but the right conditions might not present themselves consistently each year at the same time; waiting for the right conditions could thus result in considerable intra-individual variation in the timing of the initial portion of northward migration. If godwits lack a strong temporal constraint during northward migration, this might enable them to exhibit such flexibility without fitness consequences (Senner et al., under review). In this scenario, it is important that godwits be able to reliably predict the conditions characterizing the flight ahead of them (Winkler et al., 2014). However, Senner et al. (under review) found that in three of the 5 years studied, the survival of godwits was reduced while crossing the Sahara during northward migration; this could indicate that godwits cannot always reliably predict the conditions for this crossing or that the Sahara crossing invariably has a survival cost (see also Klaassen et al., 2010).

# The Control of Migratory Timing

How can individual godwits consistently depart West Africa at different times? Individual godwits could depart at different times as a result of variation in their speed of migratory preparation or as a result of variation in their condition when they begin preparing for migration. Both options are likely to occur in godwits, through consistent differences in individual and environmental quality (Studds and Marra, 2005; Paxton and Moore, 2015). However, it is highly unlikely that these options could result in a difference of up to 5 months in migratory timing among individuals. Alternatively, unpredictable cues or a less rigid endogenous programme could also lead to variation among individuals within a given year (Aloni et al., 2017). However, if the cue or programme were so variable as to lead to a difference of up to 5 months in a given year, it is improbable that differences in migratory timing among individuals would be consistent across years, as is observed in godwits. For these reasons, we believe that godwits must make use of a predictable cue or have a relatively rigid circannual programme, or that both factors apply (Gwinner, 1989, 1996).

If we assume that godwits, like other migratory birds, use photoperiod to reliably keep track of time, then individual godwits must be responding differently from each other to the same photoperiod cues in order to maintain their differences in migratory timing (Gwinner, 1996). For instance, some godwits begin their northward migrations while day length is still decreasing, whereas others migrate once day length has begun increasing again. Thus, the inter-individual variation in the migratory timing of godwits from the same wintering location must be the result of individually-specific reaction norms to the same environmental stimuli. What might be the source of these large inter-individual differences in reaction norms? They are unlikely to be the result of inheritance or adaptation, as they appear to have no fitness consequences (Kentie et al., 2017; Senner et al., under review). They are also not likely to be the result of inter-individual differences in experience, since godwits did not shift their migration earlier or later over the course of our study. Instead, different developmental trajectories are likely the source. For instance, godwits have shifted their spring staging site through developmental plasticity (Verhoeven et al., 2018), which makes it plausible that the observed individual differences in migration are also the result of different developmental trajectories (Senner et al., 2015b).

# Future Directions

Future research should therefore investigate whether differences in developmental trajectories are the source of the large interindividual differences observed, and whether the variation in migratory timing in other migratory bird species is limited by stronger temporal constraints. To accomplish this, researchers could track godwits and other migratory bird species from birth to adulthood while also performing translocation and delay experiments (Perdeck, 1958; Chernetsov et al., 2004; Thorup et al., 2007). Additionally, researchers could simultaneously perform a captive study during development in which selective disappearance is absent and photoperiod is manipulated (Helm and Gwinner, 2006; Maggini and Bairlein, 2012). All of these experiments should manipulate the spatiotemporal environment during development, thus enabling an evaluation of whether the environment does or does not affect the migratory behavior of juveniles. If it does not, this would be evidence for innate migratory behavior (Perdeck, 1958; Thorup et al., 2007). If it does, this would suggest that environmental variation brings about differences in migratory behavior (Chernetsov et al., 2004; Piersma, 2011; Meyburg et al., 2017). Tracking these individuals into adulthood would then show whether these environmentallyinduced differences are plastic or flexible and whether there is selective disappearance as a result of temporal constraints. The combination of these results would allow researchers to discern whether the narrower window of migratory timing in other bird species is the result of stronger innate control, stronger temporal constraints, or both.

# DATA AVAILABILITY

The raw data supporting the conclusions of this manuscript will be stored on the University of Groningen data repository and made available by the authors to any qualified researcher upon reasonable request.

## AUTHOR CONTRIBUTIONS

All authors designed and carried out the study. MAV, AHJL, and ADM performed the analyses. MAV, AHJL, NRS, and ADM wrote the paper with contributions from all authors.

#### FUNDING

Funding for geolocators and their analysis was provided by NWO-ALW TOP grant Shorebirds in space (854.11.004) and the Spinoza Premium 2014 of the Netherlands Organization for Scientific Research (NWO), both awarded to TP. The longterm godwit research project was funded by the Kenniskring weidevogels of the former Ministry of Agriculture, Nature Management and Food Safety (2012, 2016) and the Province of Fryslân (2013–2017). Additional financial support came from the Prins Bernhard Cultuurfonds (through It Fryske Gea), the Van der Hucht de Beukelaar Stichting, the Paul and Louise Cook Endowment Ltd., the University of Groningen, BirdLife-Netherlands, and WWF-Netherlands.

### REFERENCES


#### ACKNOWLEDGMENTS

We thank the members of our field crews for recapturing godwits, Eldar Rakhimberdiev for teaching us how to use FLightR, Jos Hooijmeijer for curating our research database and Julie Thumloup, Marco van der Velde and Yvonne Verkuil for their help with the molecular sexing. We are grateful to many farmers, most of whom are organized in the Collectief Súdwestkust, and the conservation management organizations It Fryske Gea and Staatsbosbeheer for cooperation and granting us access to their properties. This work was done under license number 6350A following the Dutch Animal Welfare Act Articles 9 and 11.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00031/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Verhoeven, Loonstra, Senner, McBride, Both and Piersma. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# High Migratory Survival and Highly Variable Migratory Behavior in Black-Tailed Godwits

Nathan R. Senner 1,2 \*, Mo A. Verhoeven<sup>1</sup> , José M. Abad-Gómez <sup>3</sup> , José A. Alves 4,5 , Jos C. E. W. Hooijmeijer <sup>1</sup> , Ruth A. Howison<sup>1</sup> , Rosemarie Kentie1,6,7, A. H. Jelle Loonstra<sup>1</sup> , José A. Masero<sup>3</sup> , Afonso Rocha<sup>8</sup> , Maria Stager <sup>9</sup> and Theunis Piersma1,7

<sup>1</sup> Conservation Ecology Group, Groningen Institute for Evolutionary Life Sciences, University of Groningen, Groningen, Netherlands, <sup>2</sup> Department of Biological Sciences, University of South Carolina, Columbia, SC, United States, <sup>3</sup> Conservation Biology Research Group, Department of Anatomy, Cell Biology and Zoology, Faculty of Sciences, University of Extremadura, Badajoz, Spain, <sup>4</sup> DBIO/CESAM Centre for Environmental and Marine Studies, University of Aveiro, Campus Universitário de Santiago, Aveiro, Portugal, <sup>5</sup> South Iceland Research Centre, University of Iceland, Fjölheimar, Selfoss, Iceland, <sup>6</sup> Department of Zoology, University of Oxford, Oxford, United Kingdom, <sup>7</sup> NIOZ Royal Netherlands Institute for Sea Research, Department of Coastal Systems and Utrecht University, Den Burg, Netherlands, <sup>8</sup> Department of Life Sciences, Marine and Environmental Research Centre, University of Coimbra, Coimbra, Portugal, <sup>9</sup> Division of Biological Sciences, University of Montana, Missoula, MT, United States

#### Edited by:

David Costantini, Muséum National d'Histoire Naturelle, France

#### Reviewed by:

Diego Rubolini, University of Milan, Italy Cas Eikenaar, Institute of Avian Research, Germany Michael P. Ward, University of Illinois at Urbana-Champaign, United States

> \*Correspondence: Nathan R. Senner nathan.senner@gmail.com

#### Specialty section:

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> Received: 23 September 2018 Accepted: 11 March 2019 Published: 09 April 2019

#### Citation:

Senner NR, Verhoeven MA, Abad-Gómez JM, Alves JA, Hooijmeijer JCEW, Howison RA, Kentie R, Loonstra AHJ, Masero JA, Rocha A, Stager M and Piersma T (2019) High Migratory Survival and Highly Variable Migratory Behavior in Black-Tailed Godwits. Front. Ecol. Evol. 7:96. doi: 10.3389/fevo.2019.00096 Few studies have been able to directly measure the seasonal survival rates of migratory species or determine how variable the timing of migration is within individuals and across populations over multiple years. As such, it remains unclear how likely migration is to affect the population dynamics of migratory species and how capable migrants may be of responding to changing environmental conditions within their lifetimes. To address these questions, we used three types of tracking devices to track individual black-tailed godwits from the nominate subspecies (Limosa limosa limosa) throughout their annual cycles for up to 5 consecutive years. We found that godwits exhibit considerable inter- and intra-individual variation in their migratory behavior across years. We also found that godwits had generally high survival rates during migration, although survival was reduced during northward flights across the Sahara Desert. These patterns differ from those observed in most other migratory species, suggesting that migration may only be truly dangerous when crossing geographic barriers that lack emergency stopover sites and that the levels of phenotypic flexibility exhibited by some populations may enable them to rapidly respond to changing environmental conditions.

Keywords: repeatability, phenotypic flexibility, seasonal survival, migration, annual cycle

# INTRODUCTION

Many migratory species are experiencing rapid and dramatic population declines (Wilcove and Wikelski, 2008). These declines have been linked with a variety of anthropogenic changes, including habitat loss (Rushing et al., 2016), and land-use (Gill et al., 2007) and climatic change (Both et al., 2006). Nonetheless, linking a species' or population's decline with specific environmental drivers remains difficult given the vast distances that separate areas occupied during different phases of the annual cycle (Piersma et al., 2016). The advent of new tracking technologies over the past two decades has begun to bridge this gap, but these new technologies remain expensive—meaning that many studies of migration are carried out over short timescales or with small sample sizes—and thus a plethora of questions about migration and the causes of declines in migratory species still remain (Hebblewhite and Haydon, 2010). This is especially true in regards to our understanding of how migratory behaviors may change over the course of an individual's lifetime and how migration itself may influence the population dynamics of migratory species (Piersma, 2011).

Recent work, however, has begun to deepen our understanding of the degree of variation in migratory behavior that individuals can be expected to display over the course of their lives. For instance, a number of studies have shown that individual migratory birds can exhibit highly repeatable migratory timing (Stanley et al., 2012; Gill et al., 2014). In some cases, individuals may vary their departure dates from non-breeding sites by as little as 3 days over 4 consecutive years (Conklin et al., 2013). In contrast, other studies have found that individuals of some species can exhibit marked improvement and flexibility in their migratory timing and performance, enabling them to adjust their behaviors to prevailing environmental conditions (Sergio et al., 2014; Pedler et al., 2018). What remains unclear, though, is how flexible migratory behaviors are generally and how both historical and contemporary selection pressures may mold the levels of flexibility exhibited by individuals within a population.

Similarly, our understanding of the location and timing of mortality events during the annual cycles of migratory species has developed rapidly over the past two decades. While most efforts to determine the seasonal survival rates of migratory species have relied on color-marking schemes and mark-recapture analyses to infer when and where individuals die (Sillett and Holmes, 2002; Lok et al., 2015), some recent studies have used satellite tracking devices to monitor the survival of individuals continuously throughout their annual cycles (Hebblewhite and Merrill, 2011; Klaassen et al., 2014; Hewson et al., 2016; Watts et al., 2019). In general, these studies have identified migration as the period during the annual cycle with the highest mortality rates. Nonetheless, work with two different sub-species of red knots (Calidris canutus islandica and C. c. canutus) has found the opposite—higher survival during migration than in stationary periods—indicating that migration may not be universally dangerous (Leyrer et al., 2013; Rakhimberdiev et al., 2015). As a result, the seasonal survival patterns of a larger number of migratory species still need to be documented in order to more fully understand how migration may influence a species' population dynamics.

The nominate subspecies of the black-tailed godwit (Limosa limosa limosa) breeds predominantly in The Netherlands and spends the non-breeding season disjunctively in sub-Saharan West Africa and the southern Iberian Peninsula (Hooijmeijer et al., 2013; Kentie et al., 2016). Previous work has indicated that individuals exhibit significant repeatability in the timing of their departure from staging sites during northward migration (r = 0.30–0.42; Lourenço et al., 2011) and arrival at the breeding grounds (r = 0.24; Kentie et al., 2017), irrespective of their nonbreeding location. Individuals can also exhibit marked flexibility in their migratory behaviors, however. For instance, in response to a recent early spring snowstorm, many godwits were able to alter their arrival timing and use of stopover sites in order to avoid the most inclement conditions (Senner et al., 2015a). Similarly, recent work has found little evidence that godwits migrating longer distances incur costs that carry-over to affect reproduction, indicating the possibility that migration may not be the limiting event during the godwit annual cycle (Kentie et al., 2017). Godwits thus represent an intriguing opportunity to assess levels of flexibility in migratory timing across an individual's life, as well as to broaden our understanding of how mortality events are spread across migratory annual cycles.

To address these knowledge gaps, we used three types of tracking devices to monitor godwit migration timing and seasonal survival over the course of 6 years, from 2012 to 2017. Given their ability to flexibly respond to conditions encountered mid-migration and the lack of reversible state effects linking migration to reproductive success (Senner et al., 2015a; Kentie et al., 2017), we predicted that godwits would exhibit high levels of flexibility and high survival during migration. By broadening the spectrum of species for which seasonal survival estimates and measures of migratory repeatability have been determined, we aim to improve our understanding of the ways in which migratory species may be able to respond to environmental change.

## METHODS

#### Study Species

Godwits of the nominate subspecies breed across much of Western Europe, but nearly 80% of their population now breeds in The Netherlands (Kentie et al., 2016). Historically, the entire population was thought to spend the non-breeding season in West Africa and then migrate northward via stopover sites in Italy, Morocco, Portugal, and France (Beintema and Drost, 1986). However, coinciding with shrinking populations and the creation of fish ponds and seasonally flooded rice fields in Spain and Portugal in the 1980s, godwits have altered both their non-breeding distribution and migration routes (Lourenço and Piersma, 2008). Currently, nearly a quarter of the population spends the non-breeding season in and around Doñana Natural and National Parks in southern Spain (Márquez-Ferrando et al., 2014). Together with staging sites in Extremadura, Spanish sites now host approximately half of the population during northward migration as well (Masero et al., 2011), with the remainder using Portuguese staging sites (Lourenço et al., 2010; Verhoeven et al., 2018). Moreover, Italy is now rarely visited and the use of French and Moroccan stopovers has declined dramatically (Lourenço and Piersma, 2008; Alves and Lourenço, 2014).

In general, godwits spending the non-breeding season in West Africa depart from January to February and then join the remainder of the population in Iberia, where they can stage for as long as 4–5 weeks (Masero et al., 2011). Godwits begin leaving Iberia in early March and can fly directly to their breeding sites or stopover as many as four times en route (Senner et al., 2015a). Breeding ground arrival then spans a period from early March to mid-May. Once on the breeding grounds, a period of 5 weeks can elapse between arrival and

clutch initiation (Senner et al., 2015b). Finally, adult godwits depart breeding areas on southward migration from mid-June onwards (Hooijmeijer et al., 2013).

#### General Methods and Tracking Devices

We employed three different tracking technologies: solar geolocation devices ("geolocators"), satellite transmitters, and GPS trackers. Geolocators and GPS trackers were both deployed during the breeding season (Apr–Jun) at our long-term demographic study area in southwest Friesland, The Netherlands (Senner et al., 2015a). This area encompasses 10,280 ha spanning from 53.0672◦N, 5.4021◦E in the north, to 52.8527◦N, 5.4127◦E in the south, 52.9715◦N, 5.6053◦E in the east, and 52.8829◦N, 5.3607◦E in the west. Satellite transmitters were deployed during northward migration (Jan–Feb) at staging sites in Extremadura, Spain (39.0364◦N, 5.9112◦W; Masero et al., 2011) and Santarém, Portugal (38.8525◦N, 8.9695◦W; Lourenço et al., 2011).

We deployed geolocators from Migrate Technology, Ltd. on adults captured on nests using walk-in traps or mist-nets placed over the nest (n = 126; 2012–2013: 0.65 g W65A9, 2014: 1 g Intigeo C65). Geolocators were attached to colored flags placed on the upper tibia; the combination of the geolocator and flag weighed ≤3.3 g, which was ≤1.5% of an individual's mass at the time of capture. In subsequent years, we then attempted to recapture geolocator-carrying individuals using similar methods as during the initial capture. We also attached 9.5 g solar-powered PTT-100s satellite transmitters from Microwave Technology Inc. (n = 60; 2013–2015) and 7.5 g solar-powered UvA-Bits GPS trackers (n = 20; 2013) developed by the University of Amsterdam (Bouten et al., 2013) using a leg-loop harness made of 2 mm nylon rope (see Senner et al., 2015a for more details). Because of the weight, we deployed these devices only on individuals weighing >300 g, meaning the majority were placed on females (n = 1 male). In total, the tracking device and harness weighed ∼12 and 10 g, respectively, representing 3.43 ± 0.22 and 2.86 ± 0.19% of an individual's mass at the time of capture.

# Tracking Data

Geolocators measure ambient light levels in order to identify the timing of sunrise and sunset, and, ultimately, estimate an individual's position on the globe. To transform our raw light data into twice-daily position estimates, we first passed it through the program IntiProc (v. 1.03; Migrate Technology, Ltd.) and then processed our transformed light data using the "BAStag" package (Wootherspoon et al., 2013) in the R software environment (R Development Core Team, 2016). We used a light threshold of 1.5 to demarcate all sunrises and sunsets and discarded sunrises/sunsets that had non-random shading events, such as when a geolocator was shaded during either the beginning or end of a twilight period. Finally, we processed our light data using the R package "FLightR" (Rakhimberdiev et al., 2017) following Rakhimberdiev et al. (2016). In FLightR, we used the period during which an individual was known to be on the breeding grounds (from resighting data) as a calibration period. We then analyzed the data without land or behavioral masks or automated outlier exclusion, but with movements constrained to the region between 18◦W−13◦E and 11–57◦N (Hooijmeijer et al., 2013). All models were optimized with 1 million particles.

We programmed satellite transmitters to collect locations for 10 h and recharge for 48 h, which allowed us to identify locations used for ≥2 d. We retrieved all location fixes via the CLS tracking system (www.argos-system.org) and passed them through the Douglas Argos-filter (DAF) algorithm (Douglas et al., 2012). We retained all standard class locations (i.e., LC 3, 2, 1) and excluded all auxiliary class locations that did not meet our predefined threshold for maximum movement rate (120 km h −1 ). On average, this resulted in 8 ± 1 locations per 10-h duty cycle for each individual.

We programmed the GPS trackers to record an individual's location once every 5 min when the tracker's battery was fully charged and once every 15 min in all other instances (see Senner et al., 2018 for more details). Although GPS trackers generate many fewer erroneous locations than satellite transmitters, spurious locations are recorded. We thus also filtered our GPS tracker data with the DAF and used the same movement thresholds as those imposed on our satellite transmitter data.

#### Statistical Analysis

To detail the temporal flexibility of godwit annual cycles, we first characterized the timing of all movements made by tracked individuals. Because both the spatial and temporal resolution of the three types of tracking devices differed (Bouten et al., 2013; Boyd and Brightsmith, 2013; Rakhimberdiev et al., 2016), we took a conservative approach to identifying movements in order to make the data comparable across devices. Therefore, given that the average location estimates generated by FLightR have a deviation of ± ∼40 km from an individual's true location and that our satellite transmitters were only able to detect stopovers lasting ≥2 d, we defined a migratory movement as a direct flight of ≥80 km and a stopover as a stationary period lasting ≥2 d. Additionally, geolocators have difficulty estimating latitude—but not longitude—within a week of the equinoxes (Rakhimberdiev et al., 2016). Unfortunately, the spring equinox coincides with godwit northward migration and the period of their arrival at their breeding sites. However, godwits migrate in both a northerly and easterly direction during this period, enabling us to document each individual's arrival date by identifying the date on which they crossed a longitude of 5◦E, which corresponds to The Netherlands when arriving from the west (see Verhoeven et al., 2019 for more details).

Given these "decision rules" for identifying migratory movements, following our previously published studies (e.g., Hooijmeijer et al., 2013), we delineated the godwit annual cycle into 10 separate annual cycle events that are performed by most godwits—the breeding season, post-breeding season, southbound migration over Europe, Iberian stopover period, southbound migration over the Sahara Desert, the non-breeding season, northbound migration over the Sahara Desert, Iberian staging period, northbound flight over Europe, and European stopover period. Briefly, we defined non-breeding sites as the site where an individual was located on 20 September, while breeding sites were those sites where an individual was located on 15 May. We defined the post-breeding period as beginning for geolocator-carrying individuals the day they were last resighted in our study area (see Loonstra et al., 2019 for detailed information about our hemisphere-wide resighting efforts); for all other individuals, the post-breeding period began when they moved >25 km away from their breeding site. Finally, not all individuals migrate to sub-Saharan Africa. As a result, we only made comparisons among individuals performing the same event (e.g., northbound flight over Europe).

We then calculated: the repeatability (r) of the timing and number of movements with the R package "rptR" (Stoffel et al., 2017) using the linear mixed model method and including individual as a random-effect; the amount of variation exhibited by an individual in their migratory behavior across years (difference in the timing of an event between consecutive years; "intra-individual" variation); and the amount of populationlevel variation across all tracked individuals (difference in timing between earliest and latest individual within a single year ± SD; "inter-individual" variation). To ensure that the observed levels of intra-individual variation were not a byproduct of stochastic environmental conditions or the type of transmitter an individual carried, we used linear mixed models with individual included as a random effect, and year and type of tracking device as predictor variables. Additionally, as an anecdotal comparison, we obtained data on the amount of intra-individual variation in northward migratory timing exhibited by two other species of godwits—Hudsonian (L. haemastica) and bar-tailed godwits (L. lapponica baueri)—from the published literature (Conklin et al., 2013; Senner et al., 2014).

We also used satellite transmitter data to calculate eventspecific survival rates. Retrieving data from both the GPS trackers and geolocators was dependent upon an individual returning to the breeding grounds, meaning we could not pinpoint when mortality occurred in individuals carrying those types of devices that did not return to The Netherlands. To calculate the eventspecific survival rates, we first determined which type of annualcycle event an individual was engaged in on each day that its transmitter provided location estimates using an individual's location and movement patterns (see above). Then, we identified on which day each individual died via its transmitter's activity sensor. Although the death of a transmitter does not necessarily imply that the individual carrying that transmitter also died, in all but five cases in which we documented the death of a transmitter, we also failed to subsequently observe the associated individual. In those cases in which we did observe the individual after the transmitter had stopped functioning, two birds had lost their transmitters, while the transmitters of the other three individuals had failed. These individuals were removed from our analyses in the season during which their transmitter failed or was lost; our estimates of godwit survival rates may therefore be biased low. Finally, to test for differences in daily survival rates among years and annual cycle events, we used an Andersen-Gill model—a type of hazards model—in the R package "survival" (Therneau and Lumley, 2015). In these models, we included the annual cycle event as the (categorical) predictor variable, survival between consecutive time steps as the dependent variable, year as the strata, and individual as a random effect. Because the hazards calculated by the model depend on which event of the annual cycle was used as the reference/baseline level, we reran the model with each separate event as the reference and then used model averaging in the R package "AICcmodavg" (Mazerolle, 2013) to determine their model-averaged coefficients.

For our linear regression model testing the effects of year and type of tracking device on intra-individual variation, we compared models including predictor variables to an interceptonly "null" model within an AIC<sup>c</sup> framework, where the model with the lowest AIC<sup>c</sup> score was considered the most wellsupported model (Burnham and Anderson, 2002). Predictor variables whose 95% confidence intervals did not include zero were considered biologically relevant (Grueber et al., 2011). The regression models were run using the R package "lme4" (Bates et al., 2014); the statistical significance of random effects was tested using the R package "lmerTest" (Kuznetsova et al., 2015). All results are reported as mean ± SD unless otherwise noted.

# RESULTS

#### Return Rates and Migration Routes

We deployed 60 satellite transmitters, 20 GPS trackers, and 126 geolocators from 2012 to 2015 (**Figure 1**). Three satellite transmitters were inadvertently placed on Icelandic-breeding black-tailed godwits (L. l. islandica) and not included in our analyses; the remaining 57 individuals were tracked for ≤4 southward and ≤5 northward migrations per individual. Fourteen individuals with GPS trackers returned the following year, of which 4 provided data and 1 eventually provided two complete annual cycles. One-hundred and eighteen individuals with geolocators returned to the breeding grounds at least once, 43 of those individuals were recaptured, 28 of those geolocators provided data for at least one full migration period, and 2 individuals were tracked for two complete annual cycles.

Among the individuals with satellite transmitters, all but 10 bred in The Netherlands (n = 47); the remaining individuals bred in Germany (n = 3), Belgium (n = 2), and Poland (n = 1), or were not tracked long enough to determine their breeding location (n = 4). Across all individuals tracked to their nonbreeding grounds (n = 64), irrespective of tracking device type, all but eight spent the non-breeding season in sub-Saharan West Africa, with the remainder spending that period either on the Iberian Peninsula (n = 7) or in Morocco (n = 1); 13 individuals died before reaching their non-breeding grounds. All individuals exhibited broad fidelity to both their breeding and non-breeding sites, as no individual changed either the province in which they bred nor whether they spent the non-breeding season north or south of the Sahara Desert (**Figure 1**; see also Kentie et al., 2017; Verhoeven et al., 2019).

#### Repeatability

The repeatability of migratory timing and behavior ranged from r = 0.78 (95% CI = 0.55, 0.91; **Table 1**) for the date of departure from the non-breeding grounds to r = 0.00 (95% CI = 0.00, 0.40) for the number of stops made during southward migration. In only three cases were aspects of migration not repeatable—the

number of stops made during southward migration (see above), number of stops made during northward migration (r = 0.20, 95% CI = 0.00, 0.52), and duration of northward migration (r = 0.21, 95% CI = 0.00, 0.52).

#### Inter-individual Variation

Levels of inter-individual variation for the timing of migratory events ranged from an average of 96 ± 29 d (n = 6 years; **Table 1**) for the timing of departure from the non-breeding grounds to an average of 47 ± 9 d (n = 6 years) for the timing of departure from the breeding grounds. Inter-individual variation in the number of stops made was 4 ± 2 stops (n = 6 years) during northward migration and 4 ± 2 stops during southward migration (n = 5 years), while for the duration of migration, it was 25 ± 11 d (n = 6 years) for northward migration and 63 ± 23 d (n = 5 years) for southward migration.

#### Intra-individual Variation

Levels of intra-individual variation in the timing of migratory events ranged from an average of 17 ± 12 d (n = 17 individuals; **Table 1**) for the timing of arrival at non-breeding sites following southward migration to an average of 10 ± 11 d (n = 22 individuals) for the timing of arrival at the breeding grounds. Intra-individual variation in the number of stops made was 1 ± 1 stops (n = 33 individuals) during northward migration and 2 ± 1 stops during southward migration (n = 19 individuals), while, for the duration of migration, it was 6 ± 8 d (n = 33 individuals) for northward migration and 15 ± 9 d (n = 19 individuals) for southward migration. These levels of intra-individual variation did not vary across years or with the type of tracking device an individual carried (**Supplementary Information Tables 1, 2**). Additionally, blacktailed godwits exhibited qualitatively more flexibility than did



Intra-individual variation is the difference in migratory timing for an individual between consecutive years; inter-individual variation is the time span between the first and last individual performing an event. The standard deviation is presented for the inter- and intra-individual variation in migratory timing and the 95% confidence interval for r.

either Hudsonian or bar-tailed godwits (**Figure 2**) over the course of northward migration.

#### Seasonal Survival

Across all years and periods of the annual cycle, daily survival rates for godwits carrying satellite transmitters averaged 0.998 ± 0.001 (n = 57 individuals and 24,366 days), leading to an annual survival rate of 0.52 ± 0.12. Hazard rates were highest during the flight from non-breeding areas in sub-Saharan West Africa to staging areas in Spain and Portugal (β = 18.97, 95% CI = 11.75, 26.19; **Table 2**; **Supplementary Information Tables 3, 4**). In contrast, hazard rates were lowest during the post-breeding staging period (β = −15.6, 95% CI = −27.29, −3.91). No other portion of the annual cycle had hazard rates that differed significantly from the baseline (**Table 2**). Finally, the breeding season accounted for 29.8 ± 20.1% (n = 5 years; **Table 2**) of all mortality events, while northward flights over Africa accounted for 12.8 ± 16.0% (n = 5 years), and the post-breeding period 0.0 ± 0.0% (n = 4 years).

### DISCUSSION

We found that black-tailed godwits breeding in northwestern Europe exhibited high-levels of inter- and intra-individual variation in migratory timing—but also high repeatability of departure and arrival dates—and generally high survival rates during migratory flights. Nonetheless, survival rates were lower during migratory flights over the Sahara Desert. Our results thus suggest that the relative danger of migration may be context dependent and that, under the right circumstances, some migratory species may be readily capable of responding to contemporary environmental changes.

#### Seasonal Survival in Migratory Species

Given the difficulty of tracking migrants throughout their annual cycles, few studies have explored their seasonal survival. The majority of studies that have succeeded in developing such estimates have identified migration as the period with the lowest daily (and seasonal) survival rates of the annual cycle (Sillett and Holmes, 2002; Hebblewhite and Merrill, 2011; Klaassen et al., 2014; Romer et al., 2015; Watts et al., 2019). Our results are not entirely consistent with these findings: migratory flights over Europe during both north- and southward migration were consistently characterized by hazard rates that were equivalent to those exhibited during other, stationary, portions of the godwit annual cycle. Nevertheless, hazard rates during the northward flight traversing the Sahara Desert were significantly higher—and thus survival rates lower—overall than during other periods of the annual cycle. Our results thus suggest that the relative danger of migration is likely context dependent and migration may only be truly dangerous during flights over geographic barriers that lack potential emergency stopover sites (see also Lok et al., 2015).

Two potential caveats, however, should also be noted: First, even after accounting for the potential biases in our survival estimates resulting from transmitter failures, the overall survival


TABLE 2 | The duration, model-averaged β coefficients from an Andersen-Gill model, and total proportion of mortality events observed during each annual cycle event.

Sample sizes for the mean duration refer to the number of episodes of each event documented across all individuals; for the proportion of morality events it refers to the number of years included.

rates documented here are considerably lower than those found in other studies with color-marked godwits (Roodbergen et al., 2008; Kentie et al., 2016, 2018; Loonstra et al., 2019). This suggests our use of satellite transmitters weighing 9.5 g may have served as a handicap that reduced survival. Survival was lower than expected across nearly all annual-cycle events, however, indicating that these effects were not just experienced during migration. Second, our sample of godwits carrying satellite transmitters was heavily biased toward females and it is possible that males and females differ in their seasonal survival patterns. Previous godwit studies, though, have suggested that migratory patterns are roughly similar between the two sexes (Lourenço et al., 2011; Kentie et al., 2016), although adult females do have slightly lower annual survival rates (Loonstra et al., 2019). We therefore believe that our results are robust and broadly representative of the seasonal survival rates and migratory patterns exhibited by godwits breeding in northwest Europe.

Given these results and those from recent studies with red knots (Leyrer et al., 2013; Rakhimberdiev et al., 2015), how strongly should migration be expected to determine the population dynamics of migratory species? Previous studies of migratory birds making flights across large geographic barriers, such as the Sahara, have found that mortality events experienced during these flights may result from a suite of potentially interacting and sometimes unpredictable processes, including extremely high temperatures (Schmaljohann et al., 2007), violent sandstorms (Klaassen et al., 2010), poor body condition (Ward et al., 2018), and predation (Gangoso et al., 2013). In some cases, the mortality events experienced during these flights play an important role in determining a species' overall population dynamics (Lok et al., 2013). Many migratory species never cross such barriers, however. For these species, our results suggest that migration should not necessarily be more dangerous than any other activity, so long as high-quality stopover sites exist (Alves et al., 2013).

Even for those species whose migrations do include geographic barriers, it is unclear how strongly migration should generally be expected to regulate population dynamics. For instance, although hazard rates were highest during trans-Saharan flights in our study, these flights accounted for only a relatively small proportion of the total number of mortality events experienced by godwits across their annual cycle. Instead the breeding season, which is substantially longer in duration than a trans-Saharan flight (µ = 75 and 2 d, respectively), resulted in more than twice the number of mortality events as did trans-Saharan flights. Given that trans-Saharan flights occupy such a short period of time, survival rates during these flights would have to be severely reduced before they become the limiting component of the godwit annual cycle (but see Sillett and Holmes, 2002; Leyrer et al., 2013). Nonetheless, understanding what is influencing mortality rates during trans-Saharan flights, and during flights over geographic barriers more generally (e.g., Ward et al., 2018), is necessary to predict how changing conditions may impact the population dynamics of migratory species.

#### Flexibility in Migratory Timing

Godwits also exhibited significant inter- and intra-individual variation in their migratory timing within and among years. For instance, individual godwits varied their departure timing from non-breeding sites by an average of 17 days and their arrival on the breeding grounds by an average of 11 days between consecutive years. Furthermore, within a single year, departure dates across the population from non-breeding sites in Africa could span nearly 5 months and arrival dates at breeding sites in The Netherlands could cover almost 2 months (see also Verhoeven et al., 2019). In turn, our previous work has indicated that this variation is neither influenced by an individual's nonbreeding site—as godwits spending the non-breeding season in sub-Saharan Africa exhibit similar levels of intra-individual variation and repeatability to those spending it in Iberia (see Kentie et al., 2017; Verhoeven et al., 2019)—nor the length of its life—as godwits do not directionally change the timing of their migrations over the course of their lives (see Verhoeven et al., 2019). To the best of our knowledge, godwits thus display the largest degree of variation in migratory timing yet documented among obligate migratory birds (Both et al., 2016). For example, in other godwit species, departure and arrival dates can span a period as small as 10 days across entire breeding populations (Senner et al., 2014), while individuals can vary their migratory timing by as little as 4 days over the course of their lifetimes (**Figure 2**; Conklin et al., 2013; Gill et al., 2014; Senner et al., 2014). More broadly, even among those bird species exhibiting low levels of repeatability in migratory timing, their absolute levels of intra- and inter-individual variation are lower than those observed in godwits (Both et al., 2016).

Intriguingly, in addition to their high levels of intra- and interindividual variation, godwits also exhibited significant levels of repeatability in nearly all of their migratory behaviors. As discussed extensively by Conklin et al. (2013), high levels of repeatability can arise for a number of different reasons, as the measure represents the ratio between intra- and inter-individual variation within a population. In godwits, their measures of high repeatability appear to result from the large degree of interindividual variation exhibited by the population (see Verhoeven et al., 2019 for more details). Thus, individual godwits tend to time their annual cycles differently from one another, yet show significant flexibility in the timing of their movements within their separate migratory windows.

This high degree of individual-level flexibility and populationlevel variability could mean that godwits are better able to respond to environmental change than other long-distance migrants. For instance, unlike Hudsonian and bar-tailed godwits (Conklin et al., 2010; Senner et al., 2017), black-tailed godwits do not appear to be time constrained during their northward migration. This means that the fitness benefits for an individual of flexibly altering its migration timing, as well as the number of stops it makes during migration, may outweigh those of arriving at a specific site at a specific time. In turn, this may enable them to minimize the initiation of reversible state effects that carry-over the conditions experienced during previous time periods, potentially affecting their survival and fitness (sensu Senner et al., 2015c). Accordingly, we have previously shown that many godwits delayed their migrations by arriving more than 20 days later than normal to their breeding sites in response to an early spring snowstorm in northwest Europe, enabling individuals that delayed their migrations to avoid the most inclement storm-related conditions, but also subsequently achieve high reproductive success (Senner et al., 2015a).

Nonetheless, their flexibility does not enable godwits to adequately respond to all types of environmental change. During the past nine decades, for example, godwits have failed to shift the onset of their breeding season earlier (Meltofte et al., 2018) and have become increasingly mismatched with the local insect phenology on their breeding grounds, leading to reductions in their reproductive success that have compounded the reductions simultaneously incurred by broadscale agricultural intensification (Kleijn et al., 2010; Schroeder et al., 2012; Kentie et al., 2018). As such, heightened flexibility may enable godwits to respond to some, but not necessarily all, environmental changes.

### The Drivers of Migratory Flexibility

More broadly, there is mounting evidence that plasticity can drive advances in migratory timing in response to climate change (Gill et al., 2014), as well as the colonization of novel migratory routes within only a few generations (Eichhorn et al., 2009; Verhoeven et al., 2018). Flexible migratory behaviors are also not unique to godwits, as some ungulates are able to flexibly alter whether or not they migrate in a given year in response to environmental conditions and herd size (Eggeman et al., 2016). Furthermore, the farthest flying migratory birds can fly more than 12,000 km non-stop and maintain annual survival rates among the highest recorded across all bird species, suggesting the ability to flexibly respond to a wide range of conditions in flight (Conklin et al., 2013). These studies thus suggest that under the right circumstances, many migratory populations can exhibit significant levels of plasticity and flexibility. But, what are those circumstances?

We propose that high levels of plasticity and flexibility in migratory behaviors could be related to: (1) high variability in the cues used to time migration (Senner, 2012; Winkler et al., 2014); (2) a relaxation of selection on migratory timing in response to a lack of density dependent selection pressures (Day and Kokko, 2015); or, (3) strong selection on flexibility in migratory behaviors (Nussey et al., 2005). In the case of godwits, we hypothesize that a relaxation of selection on migratory timing resulting from the on-going godwit population decline is the most likely scenario. For instance, if the cues godwits use to migrate were highly variable, we would expect to observe that godwits exhibit low repeatability in addition to their considerable population-level variation in migratory timing (Conklin et al., 2013). Instead, individual godwits, while displaying high flexibility in their migratory behaviors, still appear to time their migrations very differently from each other (Verhoeven et al., 2019). Similarly, if density dependence were acting strongly on godwits, we would expect survival rates to vary with the size of the godwit population (Rakhimberdiev et al., 2015). However, we found little inter-annual variation in survival rates despite the fact that the godwit population size has fluctuated in recent years (Kentie et al., 2016). Furthermore, an individual's migratory timing is not influenced by the number of other godwit pairs breeding nearby and is uncorrelated with their subsequent reproductive success (Senner et al., unpublished data), suggesting that selection on migratory timing has been relaxed. Finally, the creation of new, artificial, wetlands throughout their range has led to an expansion of the amount of habitat available for adult godwits during the non-breeding season and enabled a series of rapid changes to godwit migration patterns (Márquez-Ferrando et al., 2014; Verhoeven et al., 2018).

In combination, density dependent pressures do not currently appear to be influencing godwit survival or migratory timing and this may now mean that adult godwits exist below their carrying capacity throughout the year. Unfortunately, however, we currently lack the data to robustly assess these three possibilities and future studies should therefore endeavor to identify those circumstances that may enable migratory populations to exhibit high levels of flexibility and plasticity in their migratory behaviors. Developing a deeper understanding of why some populations are more flexible than others will both aid conservation efforts, but also help revise our view of what is normal and possible for migrants (Conklin et al., 2017).

#### AUTHOR CONTRIBUTIONS

NS, MV, and TP designed the project. All authors carried out the fieldwork. NS and MS carried out the analyses. NS wrote the manuscript. All authors contributed edits.

#### ACKNOWLEDGMENTS

We thank the many members of our field crews from 2004 to 2015 for their assistance in the field. We also thank S. Pardal, M. Parejo-Nieto, A. Villegas-Sánchez, and the rest of the teams from Badajoz and Lisboa for help with satellite transmitter instrumentation. A. Stokman, W. Nauta, S. Venema, Staatsbosbeheer, It Fryske Gea, ANV Súdwesthoeke, and Kuststripe, and many other land managers and farmers were gracious in allowing us access to their land. Local bird conservation communities—including Fûgelwachten Makkum, Warkum, Koudum-Himmelum, Oudega, Gaastmeer, and Stavoren-Warns—provided locations of many nests.

#### REFERENCES


E. Rakhimberdiev greatly assisted with the geolocator analyses, J. R. Conklin helpfully made available his bar-tailed godwit repeatability data, and T. L. Tibbitts curated the satellite tracking data from 2013 to 2015. Funding for NS, MV, and their fieldwork was provided by NWO-ALW TOP grant Shorebirds in space (854.11.004) awarded to TP. RK is funded by the Royal Society. JA benefited from a Fundação para a Ciência e Tecnologia grant (SFRH/BPD/91527/2012). Long-term godwit research was funded by the Kenniskring weidevogels of the former Ministry of Agriculture, Nature Management and Food Safety (2007–2010, 2012, 2016); the Province of Fryslân (2013–2016); and the Spinoza Premium 2014 of the Netherlands Organization for Scientific Research (NWO) awarded to TP. Additional financial support came from the Prins Bernhard Cultuurfonds (through It Fryske Gea), the Van der Hucht de Beukelaar Stichting, the Paul and Louise Cook Endowment Ltd., the University of Groningen, BirdLife-Netherlands, and WWF-Netherlands. This work was done under license number 6350A, C, and G following the Dutch Animal Welfare Act Articles 9 and 11.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00096/full#supplementary-material


staging site use by a long-distance migratory bird. Biol. Lett. 14:20170663. doi: 10.1098/rsbl.2017.0663


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Senner, Verhoeven, Abad-Gómez, Alves, Hooijmeijer, Howison, Kentie, Loonstra, Masero, Rocha, Stager and Piersma. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Interspecific Variation in Seasonal Migration and Brumation Behavior in Two Closely Related Species of Treefrogs

#### Amaël Borzée1,2 \*, Yoojin Choi <sup>2</sup> , Ye Eun Kim<sup>2</sup> , Piotr G. Jablonski 1,3 \* and Yikweon Jang<sup>2</sup> \*

<sup>1</sup> Laboratory of Behavioral Ecology and Evolution, School of Biological Sciences, Seoul National University, Seoul, South Korea, <sup>2</sup> Division of EcoScience, Department of Life Sciences, Ewha Womans University, Seoul, South Korea, <sup>3</sup> Museum and Institute of Zoology, Polish Academy of Sciences, Warsaw, Poland

Most amphibians migrate between flooded habitats for breeding and dry habitats for non-breeding activities, however, differences in closely related species may highlight divergent evolutionary histories. Through field surveys, Harmonic Direction Finder tracking and laboratory behavioral experiments during the wintering season, we demonstrated differences in seasonal migration and hibernation habitats between Dryophytes suweonensis and D. japonicus. We found that D. japonicus migrated toward forests for overwintering and then back to rice paddies for breeding in spring. By contrast, D. suweonensis was found to hibernate buried in the vicinity of rice paddies, its breeding habitat. We also found that the difference in migrating behavior matched with variation in microhabitat use during brumation and hibernation between the two species. Our findings highlight different ecological requirements between the two species, which may result from their segregated evolutionary histories, with speciation potentially linked to species use of a new breeding habitat. Additionally, the use of rice paddies for both breeding and hibernation may contribute to the endangered status of D. suweonensis because of the degradation of hibernation sites in winter.

#### *Edited by:*

Brett K. Sandercock, Norwegian Institute for Nature Research (NINA), Norway

#### *Reviewed by:*

Kristine Elizabeth Hoffmann, St. Lawrence University, United States Ivan Gomez-Mestre, Estación Biológica de Doñana (EBD), Spain

#### *\*Correspondence:*

Amaël Borzée amaelborzee@gmail.com Piotr G. Jablonski piotrjab@behecolpiotrsangim.org Yikweon Jang jangy@ewha.ac.kr

#### *Specialty section:*

This article was submitted to Behavioral and Evolutionary Ecology, a section of the journal Frontiers in Ecology and Evolution

> *Received:* 13 May 2018 *Accepted:* 15 February 2019 *Published:* 11 March 2019

#### *Citation:*

Borzée A, Choi Y, Kim YE, Jablonski PG and Jang Y (2019) Interspecific Variation in Seasonal Migration and Brumation Behavior in Two Closely Related Species of Treefrogs. Front. Ecol. Evol. 7:55. doi: 10.3389/fevo.2019.00055 Keywords: brumation, *Dryophytes japonicus*, *Dryophytes suweonensis,* hibernation, hylids, migration

# INTRODUCTION

Most amphibians migrate between flooded habitats for breeding and dry habitats for non-breeding activities. Differences among species in the details of these seasonal variations may highlight divergent evolutionary histories (Wake, 1982), although intraspecific variations in life-history strategies are widespread (Collins, 1981; Miaud et al., 1999), and traits covary with environmental gradients such as elevation and latitude (Morrison and Hero, 2003). In amphibians, aquatic breeding is the shared ancestral character for all species (Wake, 1982; Reiss, 2002; Schoch, 2009). Thus, migration between breeding and overwintering habitats is an evolutionary requirement tied to the biphasic life cycle of a large number of amphibians (Duellman and Trueb, 1986). The medium used for breeding typically reflects the ancestral character (Duellman, 1989), whereas migration toward different environments is a more recent evolutionary trait (Semlitsch, 2008). Exceptions do exist, however, such as for plethodontid salamanders, which retain the larval stage (Chippindale et al., 2004), and for one of the focal hylid species of this study for yet unknown reasons.

Differences in life cycles are one of the indicators of divergence in evolutionary origin between related species (West-Eberhard, 2003, 2005). Novel phenotypes stem from the reorganization of ancestral phenotypes, followed by the genetic accommodation of changes (Mayr, 1963; West-Eberhard, 2005). Selection acts on phenotypes (Mayr, 1963), and thus populations subjected to differential environmental pressure may see the apparition of specific phenotypes that are subsequently integrated into genotypes (West-Eberhard, 2005). This includes for instance the tropical vine Monstera sp., which displays varying leaf forms, resulting in species-specific ontogenies (Madison, 1977). It is also the case with role-reversed sandpipers species where males incubating eggs show the same increase in prolactin as incubating females (Beach, 1961; Oring et al., 1986). Another example is the interspecific variation in parental care due to ecological requirements in Microtus spp. (West-Eberhard, 2003). However, selected phenotypes may arise from the pressure exerted by environmental drivers. For instance, fisheries result in the selection of individuals through both direct and indirect pressures (Heino and Godø, 2002). Direct selection pressure includes elevated mortality of target species, while indirect selection is exemplified by ecosystem-level impacts as intakes result in depleted resources (Kaiser and De Groot, 2000). Similarly, phenotypic plasticity is one of the reasons for the expression of different phenotypes between populations. For instance, local adaptation to climate can result in different range dynamics (Atkins and Travis, 2010).

Seasonal migration is only one type of population displacement (Semlitsch, 2008; Cayuela et al., 2018), but it is the most common non-circadian migration, observed from whales (Kenney et al., 2001) to butterflies (Brower, 1995). Seasonal migration is also common in amphibians (Sinsch, 1990; Ryan and Semlitsch, 1998); for instance, Bufo bufo hibernates in terrestrial hibernacula (Van Gelder et al., 1986; Denton and Beebee, 1993) and migrates to water bodies to breed in spring (Gittins, 1983). Some amphibian species will migrate considerable distances from their breeding site to find shelter against climatic variation (Griffiths, 1984), although 15 km is considered the limit for direct migration and dispersion due to physiological requirements (Sinsch, 1990; Cayuela et al., 2018). A representative assessment of eight Central European amphibian species demonstrated migration distances between roughly 100 and 2,200 m from the breeding site (Kovar et al., 2009). Other species, such as Lithobates catesbeianus, hibernate in the vicinity of their breeding sites and thus do not require migration (Stinner et al., 1994). Migration is generally observed between breeding and over-wintering sites, where species will shelter from inclement condition through torpor. While over-wintering strategies are diverse (Storey and Storey, 2017), and include under-water (Penney, 1987) and below-ground sheltering (Borzée et al., 2018c), information about brumation or hibernation for a large number of species is still missing. Here, brumation refers to the pre-hibernation period, and although yet poorly described in metabolic terms in amphibians (Feder and Burggren, 1992; Balogová et al., 2017; Wilkinson et al., 2017; Kundey et al., 2018) with the exception of Salamandra salamandra (Catenazzi, 2016), it refers to the pre-winter reduced activity and reduced metabolic rates of poikilotherms (Hutchinson, 1979; Pratihar and Kundu, 2011; McEachern et al., 2015). When focusing on over-wintering in treefrogs, field observations provide the most reliable data for European species (reviewed by Stumpel, 1990), complemented by behavioral ecology studies on North American species (Mahan and Johnson, 2007; Johnson et al., 2008). In addition, some Indian species are known to shelter from the cold in banana stems (Iangrai, 2011). Dryophytes japonicus in North East Asia is comparatively well studied and known to start hibernating because of the rise of cirp RNA due to cold and photoperiod (Sugimoto and Jiang, 2008). In addition, the species is able to withstand massive temperature drops, down to −53◦C in laboratory settings (Berman et al., 2016), and to principally use forested hills for hibernation (Borzée et al., 2018c).

In amphibian species in the Republic of Korea, both aquatic and dry types of non-freeze-resistant hibernation types are known. For instance, Rana spp. can hibernate under water (Lee and Moon, 2011; Macias et al., 2018) and are the first species present at the breeding sites after ice thaw (Yoo and Jang, 2012; Macias et al., 2018). In contrast, Dryophytes japonicus hibernates under decaying vegetation on hills principally forested by oak trees (Borzée et al., 2018c). It is unclear if the other treefrog species from the peninsula, the endangered D. suweonensis, can follow the same pattern. Unlike D. japonicus, the species is not found in forests during the non-breeding season. Furthermore, there are anecdotal observations of D. suweonensis hibernating in the banks of rice paddies (Pers. Comm. Kim Hyun-Tae). Dryophytes suweonensis is known to have originally bred in low-altitude alluvial wetlands, but it is now restricted to rice paddies (Borzée and Jang, 2015), whereas D. japonicus breeds in a much wider range of environments as long as solid substrate is available for call production (Borzée et al., 2016b). In addition, the two species display microhabitat segregation during the production of advertisement calls, likely due to competition (Borzée et al., 2016b). Given the long list of traits linked to the species' breeding behavior (Roh et al., 2014; Borzée and Jang, 2015; Borzée et al., 2015a, 2016a, 2017a; Kim, 2015b, 2016), we hypothesize that D. suweonensis hibernates in rice paddies, where it also breeds, whereas D. japonicus is expected to migrate seasonally between breeding sites and the forests hills where it hibernates. Here, we tested the hypothesis through winter field surveys of the two Dryophytes species. Interestingly, the pattern hypothesized is known to be similar to the one displayed in other amphibians where one of the two closely related species is restricted to rice paddies (Pelophylax nigromaculatus and P. porosus brevipodus in Japan Maeda and Matsui, 1993). The absence of seasonal migration may indicate that the two species share the same breeding and overwintering habitats, and thus exploit the same environment, whereas differences in seasonal migration would suggest a different evolutionary history and the use of different environments.

#### MATERIALS AND METHODS

This project is composed of 5 distinct sections consisting of experiments, field tracking and observations for both Dryophytes suweonensis and D. japonicus: (1) field observations for brumation, (2) field orientation tracking for brumation, (3) laboratory brumation and hibernation observations, (4) winter field observations and finally, (5) spring orientation tracking. Here, we define brumation as the pre-hibernation period during which amphibians are partially active (Mayhew, 1968; Pratihar and Kundu, 2011; McEachern et al., 2015), and such as defined for one of the two focal species, D. japonicus, by Borzée et al. (2018c). The five sections used to distinguish between experiments are based on the seasonal succession of activities, from prehibernation to emergence from hibernation. We conducted all experiments with the agreement of the Ministry of Environment from the Republic of Korea under permit numbers 2013-16, 2014-04, 2014-08, 2014-20, 2015-3, 2015-4, 2015-6, 2015-28, and 2016-5.

Dryophytes suweonensis is slender and smaller than D. japonicus (Borzée et al., 2013), and the earlier species is active earlier in the afternoon than the latter, although both species are principally active at night (Borzée et al., 2016b). Dryophytes japonicus is widespread on the Asian mainland until central Mongolia and the Baikal lake region in Russia (Dufresnes et al., 2016; Kuzmin et al., 2017) but the two species only co-occur on the western lowlands of the Korean Peninsula, where the distribution of D. suweonensis is restricted to agricultural wetlands due to widespread habitat modification (Roh et al., 2014; Borzée and Jang, 2015; Borzée et al., 2015a, 2018a; Borzée and Seliger, 2018). The use of rice paddies impacts the breeding behavior of both species (Borzée et al., 2018b; Groffen et al., 2018), which display both temporal and spatial segregation during the breeding season (Borzée et al., 2016a,b). The population decline of D. suweonensis is principally linked to habitat loss (Borzée, 2018; Borzée et al., 2018a), but other factors such as hybridization (Borzée and Jang, 2018), sensitivity to water quality (Borzée et al., 2018f), behavior (Borzée et al., 2018g), and invasive species and pathogens (Borzée et al., 2017b) are involved (Borzée et al., 2017a; Borzée, 2018).

#### Field Observations for Brumation

We collected field observations on the brumation ecology of the two species at two localities in 2014 (n = 29) and four in 2015 (n = 32; **Figure 1**). The two localities from 2014 (#1 and #2; **Figure 1**) are included within the four localities from 2015 (together with remaining sites #3 and #4; **Figure 1**). Locality 1 is composed of one paddy site and two forested sites, and localities 2, 3, and 4 are each composed of one paddy site and a single forested site. We selected the localities following observations of calling males of both Dryophytes species during the breeding season (see Borzée and Jang, 2015).

We initiated the surveys in September of both years and continued until the first freeze. In 2014, we surveyed the sites in the first and third weeks of September, every week in October and the first week of November. In 2015, we surveyed all sites in the third week of September, the first and third weeks of October and the first week in November. We conducted the surveys through spotlight line transects (Smith and Nydegger, 1985), by which the researcher followed a predetermined transect and visually inspected the vegetation for individuals. The transects at the rice paddy sites were conducted for varying distances along the straight cement road at the center of rice-paddy complexes

FIGURE 1 | Spatial location of the sites surveyed in this study. The map was generated with ArcMap 9.3 (Environmental Systems Resource Institute, Redlands, California, USA; http://www.esri.com/) and the range of the species is extracted from Borzée et al. (2017a).

(**Figure 2**; Borzée and Jang, 2017). We conducted the transects at the forest sites along a 250-m approximately straight line due to the topology of the field. Each hylid frog found was hand caught, and the species identified based on morphology (Borzée et al., 2013) when calls were not available (Park et al., 2013). We detected D. suweonensis during 20 surveys, and D. japonicus during 29 surveys.

#### Field Orientation Tracking for Brumation

For all microhabitat use and directionality experiments in this study, we used a Harmonic Direction Finder (HDF; R2 RECCO AB; Lidingö, Sweden), relying on a passive dipole attached to the individual to be tracked with a gauze waist band. The HDF emits microwaves toward the dipole, which bounces them back through an antenna tailored for each individual and encodes for directionality and distance (Pellet et al., 2006; Pašukonis et al., 2014; Borzée et al., 2016b). We soldered a Schottky diode (model R2 RECCO AB; Lidingö, Sweden) on a tin-plated copper wire folded on itself at 180◦ in a fashion that created a loop 1 cm away from the bent, prolonged by two isolated segments

FIGURE 2 | Example of the brumation and winter field observations at site #1, Sihung, located 37.406◦N and 126.805◦E. The red lines represent the transects, in the rice paddies in the west and in the forests in the east. The red star indicates the site where the hibernating female was found during the winter field observations. This map was generated with Google Earth Pro (v7.1.2.2041, 2013) on maps from 2016 SKEnergy, Image Landsat Copernicus.

of the wire, resulting in an antenna. This design maintains the electric properties of the diode and provides mechanical elasticity in the antenna. The Schottky diode reflects the wave received at twice its frequency (see de Moura Presa et al., 2005; and Borzée et al., 2016b for details), which the HDF then translates into an acoustic signal of varying intensity in function of the direction and distance to the dipole, thus allowing for locating the organism that is wearing the antenna. To isolate the electrical dipole, we uniformly insulated each antenna with a silicone spray (NABAKEM, S-830 UL94 V-0, Seoul, Korea).

We selected the initial antennae with two 8-cm legs, resulting in an approximately 25-m effective range. We prepared the waistbands to which the antennae were attached with different lengths for each individual to ensure that each antenna weight was below the recommended 5% of each individual's body mass.

The protocols used here to investigate migration patterns (i.e., directionality) are following standard procedures, similar to those used to determine bird and insect directionality during migration. The protocols for birds are described in Emlen and Emlen (1966), Emlen and Emlen (1966) and have been widely used in birds and insects (Wiltschko et al., 1993, 2008; Able and Able, 1995; Benvenuti et al., 1996; Eng et al., 2017). Studies on amphibian migrations and orientation usually rely on setting such as drift-fence and pit-fall traps (Johnson, 2003; Todd et al., 2009; Santos et al., 2010; Lenhardt et al., 2015), emigration pattern field-data collection (Lenhardt et al., 2013), or indirectly through road-kills (Elzanowski et al., 2009). In addition, amphibian movements can also be studied through telemetry (Baldwin et al., 2006; Pellet et al., 2006; Borzée et al., 2018d) and orientation has also been studied through HDF tracking (Pašukonis et al., 2013; Pašukonis et al., 2014).

Tracking during brumation was conducted separately for the two treefrog species. HDF tracking was conducted in 2013 for D. japonicus because the species was already known to be hibernating on forested hills (Sugimoto and Jiang, 2008; Borzée et al., 2018c). However, we collected the first anecdotal observation on the overwintering of D. suweonensis in 2015, following which we conducted brumation tracking.

We released and tracked each frog for 24 h, thus including both diurnal and nocturnal activities. Releases were conducted after sunset, at least 15 m from each other to prevent detection overlap. For each tracking point, we placed a colored flag approximately 10 cm from the frog to measure the displacement and directionality of the movement between successive points. Every hour, we recorded temperature (◦C), luminosity (lux), relative humidity (%), height from the ground (cm), distance moved from the previous point (cm) and type of microhabitat. The five types of microhabitats were "grass," "rice," "buried," "ground," and "bush." We also noted the directionality of the movement, toward either the center of the forest or the center of the adjacent rice paddies. The movement was decomposed in the form of a vector, for instance, 50 cm toward the forest and 20 cm toward the rice paddies. We took all measurements at 5-cm resolution to avoid overly disturbing the frogs.

#### Dryophytes japonicus

For this experiment, the waistband to attach the antenna to the frogs was made of gauze sewed onto itself ventrally and it was thus adjusted for the size of each individual. We tracked nine males and one female D. japonicus in the city of Paju (red marker; **Figure 1**), on September 27 to 30, 2013. As it is difficult to catch adults once they have reached their hibernation habitat, individuals had been previously caught in the forest adjacent to the rice paddy complex (see Borzée et al., 2018c) and raised for 1 month under controlled conditions in the lab. We released each individual at the edge of a rice paddy between 174 and 209 m away from the edge of the forest where it had been caught. We tracked individuals for 2328.33 ± 719.63 min on average (ca. 39 h), resulting in an average of 125.98 ± 17.70 min (mean ± S. D.; ca. 2 h) between tracking points, for a total of 172 observations.

#### Dryophytes suweonensis

For this experiment, each waistband was made of silicone tubing (diameter = 1.8 mm) within which we inserted the loop of the antenna. The loop was connected ventrally with electrical paint (BareConductive 10 ML, Bare Conductive Ltd; London, UK) and thus worked as an electric dipole that was tailored for the size of each individual. We tracked four individuals of each sex at four independent sites over a 110-km range (green markers, **Figure 1**) between October 8 and 16, 2015. Individuals had been caught ∼1 month earlier at the same site (Kim, 2016; Borzée et al., 2018e) and had been raised in controlled laboratory conditions before their release. We released each individual at the edge of a rice paddy between 139 and 1,018 m away from the edge of the closest forest, and we tracked individuals for 1,332 ± 120 min on average (ca. 20 h), for a total of 305 observations.

# Laboratory Brumation and Hibernation Observations

We conducted two experiments under laboratory conditions for both species, the first one on microhabitat selection during brumation, referred to as "lab brumation," and the second one on microhabitat selection during hibernation, referred to as "lab hibernation." For both species, the individuals originated from egg masses that we collected from the wild (Kim, 2016; Borzée et al., 2018e) and were kept in the lab from hatching to release (permit 2015-4 issued by the Ministry of Environment of the Republic of Korea). For this experiment, we collected five individuals from five egg masses at five different locations (blue markers, **Figure 1**; n = 25 for each species) and raised the tadpoles in independent PVC aquariums (20 cm W × 30 cm L × 20 cm H). After metamorphosis, individuals were isolated, providing a sample size of five individuals raised independently for each of the five families for each species. Each metamorph was raised in a glass terrarium 45 × 45 × 45 cm with a lateral opening, transparent sides, and a screen top (PT2605, Exo-terra, Hagen, Montreal, Canada).

We set all terraria with wet towels at the bottom that were changed weekly or more often if needed. We also set a conic non-glazed terra cotta pot (25 cm height × 16 cm diameter) horizontally at the back left of each terrarium and set a glazed water-dish (3 cm deep, 12 cm diameter) at the front right, with a 5 × 3 cm non-glazed terra-cotta cylinder pot set upside down within the water dish. The arrangement allowed for the terra cotta pots to absorb water and release it through evaporation within the terrarium to keep the humidity relatively constant (48.48 ± 11.18% rH). Finally, we set a wooden cylinder (2.5 cm diameter) from the bottom front left corner to the top back right corner. We used oak cylinders on D. japonicus known preferences (Borzée et al., 2018c). We set the terraria onto two four-layered shelves, with their position randomized every second week. Each terrarium was illuminated by its own lighting (UV-B bulbs), and we estimated that the positions of the terraria did not result in any bias in the experiments.

Each of the 50 terraria was sprayed daily and the water dish refilled ad libitum with carbon filtered and 72-h evaporated water. The frogs were set on a circadian cycle that matched the natural one, readjusted weekly, under natural spectrum illumination. Crickets were the main diet items used to feed the frogs, but fruit flies and maggot supplements were also given when available. All prey items were powdered with calcium and multivitamins prior to use.

For both brumation and hibernation experiments, we recorded the position of the individual, i.e., substrate use, and its height from the bottom of the terrarium three times a day (variable: "time period"). We also collected date, time, temperature (◦C) and relative humidity (%) for each survey point. We noted temperature and humidity readings once for all 50 terrarium replicates because they were set in a common rearing room that was exposed to controlled climatic variations that followed natural variations (see Borzée et al., 2018e for details). There were five categories of microhabitat in each aquarium: big pot (on or within the large terra cotta pot, representative of sheltering behavior), ground (anywhere on the paper towel, representative of ground microhabitats), wood (sitting on the wood cylinder, representative of perching behavior on wooden microhabitats, as known to be important for D. japonicus brumation; Borzée et al., 2018c), pot in water (representative of flooded or damp habitat, as known to be important for D. suweonensis brumation, earlier in this study), and glass (when an individual was resting on a glass panel, not representative of any wild habitat but included to prevent any bias in further analysis). Whenever we made an observation, we noted the microhabitat the individual used.

We conducted the brumation experiment between September 23 and October 16, 2015, leading to a total of 1,906 observations, and we conducted the hibernation experiment between December 1 and 23, 2015, for a total of 1,612 observations.

#### Winter Field Observations

The purpose of this set of observations was to observe frogs of both species hibernating in their natural environments. We revisited the four sites where we had conducted brumation observations in the third week of January 2015 to look for buried individuals. We spent a total of 4 h at each of the four sites digging the soil down to 50 cm in the areas where the frogs of the two species had last been seen. The areas searched were about 2 m<sup>2</sup> for each observation, and matched with the areas where the frogs were last seen (Borzée and Jang, 2017), based on the assumption that home range and breeding ranges may be similar (Kim et al., in press).

#### Spring Orientation Tracking

For this section, the tracking methodology was the same as used for orientation experiments (Section Field Orientation Tracking for Brumation). The frogs used for this experiment were animals we had raised from hatching and had used for the two lab experiments. We had maintained them in the setting described above until the beginning of tracking experiments, and we released them after finishing the spring orientation tracking experiments. The sites for the experiment, which aimed at determining the post-hibernation behavior of the two treefrog species, were distributed along the whole range of D. suweonensis (blue markers, **Figure 1**). The sites were the same as the sites where we had collected the egg masses. The tracking procedure in the spring tracking experiment was the same as that of the fall tracking for D. suweonensis.

We conducted the fieldwork between May 25 and June 17, 2016, for a total of 20 individuals from each species. We released ten individuals of each species in rice paddies and the remaining ten in forested areas at the edge of rice paddies. The forest was composed of Chinese chestnuts (Castanea spp.) and pine trees (Pinus spp.) where D. japonicus is typically found during the nonbreeding season (Borzée et al., 2018c). We checked the position of each individual every hour and we took note of date, time of day, GPS coordinates, vegetation type, temperature, humidity, height, total movement and movement toward rice paddies and forest. We assessed the directionality based on movements toward the selected landscape features. We spent an average of 22.07 ± 2.32 h (n = 25 for each species) tracking per individual (duration ± SD). As all individuals were tracked for over 16 h, all were included in the subsequent analyses. The microhabitats recorded were grass, rice, buried, ground, and bush. We tracked a maximum of 5 individuals at the same time, and avoided tracking individuals within 50 m of each other to prevent overlap in HDF detection signal. We released each individual between 20:00 and 00:00 to minimize possible predation due to the tracking apparatus.

#### Data Analysis

After the tracking experiments, we mapped the GPS coordinates for all tracked points for each individual on Google Earth (Google, Mountain View, USA). For each data point, we then measured the directionality of the movement, such as the angle between the forest, the tracking position of the individual and its subsequent tracking position. Here, the forest was defined as a point situated at the center of gravity of 10 points randomly chosen on the edge of the forest. Because we had released all individuals at different points and their orientation toward landscape elements was our prime focus, and not directionality toward cardinal points, we defined 0 degrees as toward the forest. We set this point arbitrarily for ease of graphical interpretation and because individuals rarely expressed a total change (i.e., 180 degrees) in directionality. Thus, an angle of orientation between 270 and 90 degrees represented a displacement that factored positively toward the forest, an angle of 0 degrees was a straight line toward the forest and an angle of 180 degrees was a straight line away from the forest. We measured all angles on screen with the software imageJ (National Institutes of Health, Bethesda, USA). For ease of analysis for models assessing directionality of migration patterns but not involving interspecies comparisons, the directionality was binary encoded as toward or away from the forest. For other analyses where a binary encoding would not be adequate to describe the behavior of the species, variation in the angle of directionality was analyzed separately with circular statistics.

(1) To assess seasonal variations in occurrence for both species, and compare the occurrence patterns between the two species, we first analyzed the dataset through a repeated-measure ANOVA. To do so, the repeated surveys, continuous variable hereafter defined as "season," were set as the dependent variable with seven levels, corresponding to the seven surveys replicates, while occurrence for D. suweonensis and D. japonicus were set as factors. To run this analysis, we tested for homogeneity of variance with Levene's test, and because the error variance was not equal across groups for the sixth replicate only, we ignored the partial violation of assumption for the statistical analysis. We also tested the sphericity assumption with Mauchy's test, and ran the repeated-measures ANOVA with the Greenhouse-Geisser correction (Scheiner and Gurevitch, 2001) because the sphericity assumption was violated. Furthermore, we assumed compound symmetry (homogeneity of the variance-covariance matrix) for this analysis. We then graphically matched the variation in occurrence over weeks for the two species and the type of site with the results of the statistical analysis.

(2) Because there was no correspondence between the data for D. japonicus and D. suweonensis, the data for each species was first analyzed in relation with directionality to the forest (details below), and then compared. The fall tracking data for D. japonicus was first tested for the significance of directionality toward forests for individuals. Because data were either temporally or spatially independent, the directionality variable was set as dependent variable in a binary logistic regression. The last assumption was also met in that the individual variable, set as the independent variables, was on a nominal scale.

Once directionality was established, we analyzed the dataset using a univariate General Linear Model (GLM) to find the factors that were important for directionality. Thus, we set directionality as a dependent variable, encoded as toward paddies, toward forests, or no movement; distance traveled as a fixed factor; frog ID, day, time of day and microhabitat as random factors and temperature and luminosity as covariates under a main effect model. After visually testing for the absence of outliers by analyzing box-plots, we determined the normal distribution of the data with the Kolmogorov-Smirnov test for normality with the Lilliefors significance correction (0.14 ≤ D(151) ≤ 0.51, p < 0.001), and determined the homogeneity of variance with Levene's test for homogeneity of variances [F(134,116) = 2.19, p = 0.001]. We did not include relative humidity in the analysis because it was correlated with temperature (Pearson's Correlation, R = −0.16, p = 0.044). We subsequently explained all significant variables through descriptive statistics.

We analyzed the fall tracking for D. suweonensis in a similar way as that for D. japonicus, first testing for the significance of directionality toward forests for individuals. Because the data were either temporally or spatially independent, we set directionality as the dependent variable in a binary logistic regression. The last assumption was also met in that "individual," set as the independent variable, was on a nominal scale.

We then analyzed our dataset with a GLM to find the factors that were important for directionality. For this model, we set directionality as the dependent variable; distance traveled as a fixed factor; site, sex, and vegetation as random factors and date, time of day, temperature, height, luminosity, and frog ID as covariates, under a main factor effect model. After visually testing for the absence of outliers by analyzing boxplots, we determined the normal distribution of the data with the Kolmogorov-Smirnov test for normality with the Lilliefors significance correction (0.07 ≤ D(301) ≤ 0.47, p ≤ 0.001), and determined the homogeneity of variance with Levene's test for homogeneity of variances [F(136,164) = 4.28, p < 0.001]. In this analysis as well, we did not include relative humidity because it was correlated with temperature (Pearson's Correlation; R = −0.81, n = 304, p < 0.001). The significant variations were then explained through descriptive genetics.

Last, to understand the differences in movement patterns between the two species, we also analyzed the angles described by the movements in relation to forests using circular two-sample geometrical directional analysis. The data were temporally and spatially independent but did not meet the prerequisite assumption of the von Mises distribution (Watson's U2 test; U2 = 1.72, p < 0.005; Lockhart and Stephens, 1985), and we used the non-parametric Mardia-Watson-Wheeler (Mardia, 1972) test with angle as the dependent variable and species as independent variables. We ran the analysis under an axial (orientation) model. We conducted this additional test despite the experiments being conducted at different years and at different sites for the two species. We used the results to link the two analyses and highlight the differences between the two species.

(3) We first analyzed the lab brumation experiment dataset using a multinomial logistic regression to detect variation in microhabitat use between the two species. Thus, we set microhabitat as the dependent variable, and the independent variables were species, family, and individual ID nested within family as factors, and temperature, height, time and date as covariates. We did not use humidity in the model because it was correlated with temperature (Pearson's Correlation; r = 0.40, p < 0.001) and date (r = 0.11, p = 0.001). We ran the regression under a main-effects model and selected a multinomial logistic regression because assumptions were fulfilled: we did not detect outliers in our analyses of the box-plots. There was a linear relationship between the continuous independent variables and the logit transformation of the dependent variable, tested through the Box-Tidwell (Box and Tidwell, 1962) procedure with Bonferroni corrections (Tabachnick and Fidell, 2014), with p > 0.379 for all variables, which thus rejected the null-hypothesis. We then described the variation between the different microhabitat and other significant results.

We then analyzed the lab hibernation experiment dataset for microhabitat use in the same way as for the brumation period, because we conducted this experiment in the same setting and collected the type same data. We also ran the multinomial logistic regression in agreement with assumptions: we observed no outliers through the analysis of box-plots, and we did not use humidity in the model because it was correlated with temperature (Pearson's Correlation; r = 0.09, p < 0.001) and date (r = 0.08, p = 0.001). There was a linear relationship between the continuous independent variables and the logit transformation of the dependent variable with p > 0.116 for all variables.

As the same variables were significantly different for the two species for both phases of the experiment, as tested above, we proceeded to run an additional multinomial logistic regression to assess whether the two species differed in microhabitat use between the two phases of the experiment (i.e., brumation and hibernation). To do this, we set microhabitat as the dependent variable, and for the independent variables, we set phase and species as factors and height as a covariate. The model assumptions were met, we did not detect any outlier, no variables were significantly correlated and there was a linear relationship between the continuous independent variables and the logit transformation of the dependent variable with p > 0.358 for all variables.

(4) Due to the low number of hibernating individuals found during the field observations, we could not conduct any statistical analysis and the results are descriptive only.

(5) To determine microhabitat preferences and directionality post-hibernation for the two species, we first assessed the correlations between variables to avoid collinearity in subsequent analysis. We detected significant correlations between total displacement and displacement toward paddies (Pearson's correlation; r = 0.20, n = 873, p < 0.001) and between total displacement and displacement toward forest (Pearson's Correlation; r = −0.26, n = 873, p < 0.001). The variables ID, date (Pearson's Correlation; r = 0.95, n = 873, p < 0.001) and sites (Pearson's Correlation; r = 0.96, n = 873, p < 0.001) were also correlated. Finally, temperature and humidity followed the same trend (Pearson's Correlation; r = −0.79, n = 873, p < 0.001). Among the correlated variables, we included only one from each group in the subsequent analyses.

We then ran a GLM to determine the differences in directionality between the two species post-hibernation. Thus, the binary encoded directionality was set as the dependent variable, species and ID as fixed factors, habitat as a random factor, and date, time, temperature and height as covariates, under a main-effects model. A few more variables were not included in the model because of collinearity: humidity was correlated with temperature (Pearson's Correlation; r = −0.78; n = 873; p < 0.001), total displacement was correlated with directionality (Pearson's Correlation; r = 0.33; n = 873; p < 0.001), and site type with microhabitat (Pearson's Correlation; r = −0.41; n = 873; p < 0.001). We chose a GLM after visually testing for the absence of outliers through the analysis of the box plots, and we determined the normal distribution of the data with the Kolmogorov-Smirnov test for normality with the Lilliefors significance correction (0.14 ≤ D(151) ≤ 0.51, p < 0.001). Finally, we determined the homogeneity of variance with Levene's test for homogeneity of variances [F(68,804) = 6.03, p = 0.001]. The significant differences were highlighted by descriptive statistics, and in addition T-tests, circular statistic tests and ad hoc analyses were conducted if required. Due to significant differences in directionality between the two species, we analyzed the differences in displacement angles through a suite of circular two-sample geometrical directional analyses, two-by-two for the variables: species, release habitat and angles. Because the data were temporally and spatially independent, but did not meet the prerequisite assumption of von Mises distribution (Watson's U2 test; U2 = 2.65, p < 0.005), we used the non-parametric Mardia-Watson-Wheeler test (Mardia, 1972) with angle as a dependent variables and either species or release habitat as the independent variable. We ran the analyses under an axial (orientation) model. Specifically, we computed the biostatistical analyses using SPSS v 21.0 (IBM Corp., Armonk, USA) while the circular statistics were conducted under PAST v 3.17 (Hammer et al., 2001).

#### RESULTS

#### Brumation Field Observations

The field surveys during the brumation period highlighted a clear difference in habitat preference between the two treefrog species: during brumation, D. suweonensis was present at rice paddies only while D. japonicus occurred at both rice paddies and forests (**Figure 3**). However, D. japonicus was present in rice paddies until the last week of September only, after which the species was only seen in forests. Neither species was detectable by spotlight transects from the last week of October, temporally matching with the first freeze. The results of the repeated-measure ANOVA between the two species.

(n = 61) supported these results, showing that the occurrence of D. suweonensis significantly varied with season (χ² = 0.64, d = 6, p < 0.001), and D. japonicus followed the same pattern (χ² =

0.58, d = 6, p < 0.001). Thus, our results highlight a significant variation in habitat use: both species were present in rice paddies at the beginning of the surveys, whereas D. suweonensis was in the rice paddies and D. japonicus was in the forests only during the week preceding hibernation.

#### Field Orientation Tracking for Brumation

During the fall tracking (2) experiment for D. japonicus in 2013, there was a clear directionality pattern. The species moved away from the rice paddies in 63.6% of cases (n = 96), and toward the forest in 66.2% of cases (n = 100), while moving toward paddies in 36.4% of cases (n = 55), and away from forests in 33.8% of cases (n = 51). Here, away from paddy and toward forest conveys the same directionality vector, but the percentages do not equal 100% if combined two by two because the directionality is divided into vectors that can have multiple constituents. The logistic regression model was statistically significant, χ 2 (1) <sup>=</sup> 8.38, <sup>p</sup> <sup>=</sup> 0.004 and the model explained 78.0% (Nagelkerke pseudo-R 2 ) of the variance in directionality and correctly classified 58.9% of cases.

Dryophytes japonicus used the grass and bush microhabitats to move toward paddies in 1.8% of cases (n = 3) and while they used the rice microhabitat in 96.4% of cases (n = 146). Most movements toward forests were also made in rice (94%; n = 142), while remaining movements toward forest were in grass (4%; n = 6) and bush (2%; n = 3). The distances traveled in relation to directionality were also significantly different between the two species (**Table 1**; **Figure 4**), with an average distance moved toward the rice paddies of 48.10 ± 7.8 cm and an average distance moved toward the forest of 99.4 ± 11.2 cm. ID, day, TABLE 1 | GLM to test the relationships between directionality and other factors collected during the tracking experiment to investigate the brumation behavior of Dryophytes japonicus.


time of day, temperature, and luminosity were not significant to directionality (**Table 1**).

During the fall tracking experiment (2) for D. suweonensis in 2015, there was no clear difference in directionality, in contrast with D. japonicus (**Figure 4**). The species moved away from the rice paddies in 3.0% of cases (n = 9) but toward paddies in 41.14% of cases (n = 125), and moved away from forests in 20.4% of cases (n = 62) and toward forests in 16.4% of cases (n = 50). The percentages do not equal 100% if combined two by two because the directionality is divided into vectors that can have multiple constituents. Once tested statistically, the logistic regression model was not significant, χ 2 (1) <sup>=</sup> 0.56,

FIGURE 4 | Orientation of the paths taken by Dryophytes japonicus individuals (green) and D. suweonensis individuals (red) during the fall tracking experiment (A) corrected for release point for each individual. Dryophytes japonicus individuals were directed toward forests in 66.2% of cases, and D. suweonensis individuals moved toward rice paddies in 41.14% of cases and toward forests in 16.4% of cases. Rose diagrams show variation in the angles of displacement between Dryophytes japonicus and D. suweonensis in relation to the release habitats (rice paddies or forest; B). The differences in the angles of displacement are significantly different between the two types of environments for both species. The dark lines are kernel density estimates representative of the weighted relative directionality to forest, and the light shaded areas are abundances proportional to radius. For the analyses, direction to forest was used as 0 degrees, and is thus indicated at the top of each rose diagram.

p = 0.453, explaining 0.1% (Nagelkerke pseudo-R<sup>2</sup> ) of the variance in directionality and correctly classified 55.4% of cases. There was thus a difference between the two species, with D. japonicus displaying a significant preferential direction whereas D. suweonensis did not.

The results of the GLM explaining the factors related to directionality show that only the distance traveled was significant (**Table 2**). Despite the different distances to forest at the four sites (139, 1018, 470, and 404 m), no significant variation among sites were reported, and it was the same for the variation between individuals, highlighting a general behavior (**Table 2**). The average distance moved toward the rice paddies between two locations was 78.75 ± 235.98 cm while the distance traveled toward the forest was on average −38.18 ± 219.34.

We found a difference in directionality through angles of displacement for the two species in that the angle described by the movements of D. japonicus deviated from the forest by 34.79 ± 5.09 (mean ± kappa) degrees only on average, whereas the angle described by D. suweonensis was 87.55 ± 1.79 degrees away from the forest on average (**Figure 4**). We found the difference in directionality between the two species to be significant (Mardia-Watson-Wheeler test; W = 461.80, n = 288, p < 0.001).

# Laboratory Brumation and Hibernation Observations

#### Brumation

During the brumation period, we observed a significant difference in microhabitat use between the two species (**Table 3**) under a significant model (χ² = 3266, df = 212, p < 0.001), explaining 85.9% of variation (Nagelkerke



pseudo-R²). Dryophytes japonicus preferentially selected the wood microhabitat (20.0% use in D. japonicus and 10.0% in D. suweonensis), whereas D. suweonensis preferentially selected the ground microhabitat (24.0% use in D. suweonensis and 12.6% use in D. japonicus; **Figure 5**). These microhabitats were the most commonly used ones after removing the "glass" microhabitat used for displacements (37% in D. japonicus and 34.0% in D. suweonensis).

One of the other siignificant differences between the two species was the height at which the frogs were found in the terraria (**Table 3**). Dryophytes japonicus was on average 23.14 ± 4.33 cm high, whereas D. suweonensis was on average 20.37 ± 7.37 cm high. In addition, there was a significant difference between individuals and families, and the model for this analysis was significant (χ² = 373.48, F = 264.41, p < 0.001).

#### Hibernation

During the hibernation period, the variables microhabitat and height were also significantly different between the two species (**Table 4**), under a significant model (χ² = 2138.21, df = 232, p < 0.001) that explained 79.1% of the variation (Nagelkerke

TABLE 3 | Results of the multinomial logistic regression to investigate variation in microhabitat use between the two species during the brumation experiment (n = 2,055).


pseudo-R²). In this second phase of the experiment, the use of the flooded microhabitat was higher for D. japonicus (5.7%) than for D. suweonensis (1.7%) while D. japonicus moved about twice more than D. suweonensis, as seen by the greater use of glass walls (12.9 vs. 6.9%; **Figure 5**).

According to the results of the brumation experiment, the heights at which individuals were found was also significantly different (**Table 4**), with D. japonicus again higher on average (7.19 ± 9.47 cm) than D. suweonensis (5.21 ± 8.10 cm). Besides, there was a significant difference between families. The model for this analysis was also significant (χ² = 463.33, F = 663.08, p-value < 0.001).

#### Difference Between Brumation and Hibernation

The multinomial logistic regression to assess whether the two species differed in microhabitat use between brumation and hibernation were significant for species (χ² = 44.83, df = 4, p < 0.001), phase (χ² = 118.27, df = 4, p < 0.001) and height (χ² = 4362.12, df = 4, p < 0.001). The model was significant (χ² = 5069.81, df = 12, p < 0.001) and explained 78.9% of the variance (Nagelkerke preudo-R²). As seen earlier, the frequency of use for the wood microhabitat decreased between the two species between brumation and hibernation, while the frequency increased for the use of the ground microhabitat (**Figure 5**). The average height also decreased for the two species during the same period.

#### Winter Field Observations

Out of the 4 h spent digging at each of the sites, we found a single individual, a female D. suweonensis at the paddy site in Sihung (site 1; 37.410046◦N; 126.808053◦E; **Figure 2**). We found

FIGURE 5 | Microhabitat variation displayed by Dryophytes suweonensis and D. japonicus during brumation and hibernation with corresponding frequency. The variation in the total number of counts was due to the death of three individual D. suweonensis during the period before hibernations. The non-annotated stack "pot in water" for D. suweonensis is 1.7%.

TABLE 4 | Results of the multinomial logistic regression to investigate variation in microhabitat use between the two species during the hibernation experiment (n = 1,750).


TABLE 5 | Results of the GLM to test the factors of importance between the two species during the spring tracking experiment.


the individual in a burrow excavated by an unknown animal. The individual was buried between 27 and 30 cm deep. The absence of other findings does not reflect the absence of individuals, but only our inability to find them.

#### Spring Orientation Tracking

The directionality exhibited by D. japonicus and D. suweonensis during the spring tracking experiment was significantly different between the two species (**Table 5**). When released in the forest habitat, 60.1% of D. japonicus displacements were away from the forest release point, and 42.6% toward rice paddies (**Table 6**), a significant difference between the two directions for the species (T-test; t = −8.15, df = 127, p < 0.001). In contrast, there was no difference in directionality between the two species when released in the rice paddies (**Table 6**; T-test; t = −1.31, df = 45, p = 0.198). Oppositely, the movements of D. suweonensis toward and away from the forest when released in that habitat were not significantly different (**Table 6**; T-test; t = −3.49, df = 50, p = 0.186), neither than it was significant when released in rice paddies (**Table 6**; T-test; t = −0.32, df = 46, p = 0.749; **Figure 4**). This pattern was the same for the cumulated distances traveled by the two species toward either rice paddies (D. japonicus = 20.60 ± 128.62 and D. suweonensis = 15.82 ± 119.01 m; ANOVA; χ² = 402.09, F(1,872) = 0.07, p = 0.786) or forests [D. japonicus = −30.46 ± 119.36 and D. suweonensis 0.98 ± 115.80; ANOVA; χ² = 31439.84, F(1,872) = 6.61, p = 0.010; here cumulated by type of site for ease of understanding].

TABLE 6 | Descriptive statistics for the directionality of the movements exhibited by Dryophytes suweonensis and D. japonicus in relation to the type of site selected for the release (i.e., rice paddy or forest) during the spring tracking experiments.


Data pooled two-by-two do not equal 100% because frogs were sometimes immobile and thus did not provide any directionality data for a few hours.

The results of the first GLM also showed a difference in microhabitat use between the two species (**Table 5**), with grass preferentially used by D. japonicus, and grass and bush principally used by D. suweonensis. However, we never found D. japonicus buried, and we found D. suweonensis buried in only 3.3% of cases and on the bare ground in only 6.3% of cases (**Figure 6**).

When assessing the difference in directionality for the two species (**Figure 7**), the angles were significantly different (Mardia-Watson-Wheeler tests) when released in rice paddies [D. japonicus: 79.22 ± 1.69 (mean ± kappa); D. suweonensis: 68.02 ± 2.00], and forests (D. japonicus: 113.77 ± 1.87; D. suweonensis: 86.39 ± 1.57). In addition, the variables were significantly different (Mardia-Watson-Wheeler tests) for: habitat of release for both D. japonicus (W = 364.76, n = 225, p < 0.001) and D. suweonensis (W = 326.79, n = 221, p < 0.001) but also between both species for a given release habitat: rice paddies (W = 306.23, n = 192, p < 0.001) and forests (W = 387.23, n = 245, p < 0.001).

#### DISCUSSION

Extensively different traits can occur with little genetic change (West-Eberhard, 2005), and here we further developed the suspected divergence in evolutionary history between the two Korean species of Dryophytes treefrogs (Ham, 2014; Kim, 2016). Whereas, D. japonicus migrates twice yearly between rice paddies and forested areas, to breed and overwinter, D. suweonensis is present at rice paddies all year round and hibernates buried underground. The ancestors of the two species diverged during the Late Miocene (8.7∼6.5 Mya Duellman et al., 2016; Dufresnes et al., 2016), apparently when one of the two species adapted to a different environment. It is likely that this happened when D. suweonensis preferentially selected marshes for breeding, whereas the preference of the ancestral species is expected to have been broader, a characteristic shared by most Hyla and Dryophytes species. Breeding in a different habitat led to the acquisition of new behavioral traits, such as holding on vegetation while calling (Borzée et al., 2016a), but also to the loss of traits, here migration. It is unlikely that the seasonal migration is a newly acquired trait in D. japonicus given that the species breeds in most types of habitats, also at higher elevations (Roh et al., 2014), and thus migration is expected to have been the ancestral character. The migration distance may, however, have been modified by the development of agriculture, with the two species brought back into contact (Borzée et al., 2015b, 2017a). Dryophytes suweonensis breeding in a separate habitat is the preferred hypothesis as this scenario enables the development of pre-zygotic isolating traits, such as seen in Dryophytes cinereus (Höbel and Gerhardt, 2003), and the two species are able to hybridize (Kuramoto, 1984; Borzée et al., 2015b), and have thus not evolved post-zygotic isolation. Sympatric speciation is hypothesized as peri- and para-patric speciations cannot have occurred with the two species displaying sympatric ranges (Jang et al., 2011).

During the breeding season, both species call from rice paddies, and although calling is discontinued in early July (Roh et al., 2014), both species are present in the vicinity of rice paddies until mid-September, where individuals sometimes produce calls. Continued vocalizations are unlikely to be for mating purposes because juveniles attempt the same behavior in synchrony (Pers. Obs.). At this point in time, adult D. japonicus migrate toward forests, up to several hundred meters away in this study, although closely related species can migrate up to 8 km (Angelone and Holderegger, 2009). Males D. japonicus will be present on the tree canopy, favoring oak trees (Borzée et al., 2018c) and producing calls as observed during transects here, until they are not seen anymore, in late October/early November, temporally coinciding with the first frost. Adult and juvenile D. suweonensis stay in

the vicinity of rice paddies, favoring the upper leaves of planted beans (Borzée and Jang, 2017), hypothetically for feeding based on the high insect density seen in proximity. This, however, also makes the species susceptible to the bean harvest, which potentially affects recruiting young individuals into the breeding pool (Borzée and Jang, 2017). Individuals will then find shelter underground on the banks of rice paddies, where rice straw is stacked after harvest but also burnt before the thaw of ice, with unknown consequences for the species.

Both before and after overwintering, the two species displayed variations in the orientation of their displacements in relation to forests and rice paddies. Dryophytes japonicus was aiming at forested hills before winter, whereas D. suweonensis displayed non-directional displacements. After winter, the difference between the species was still present, with D. japonicus moving toward the breeding sites when released in the forest, whereas D. suweonensis did not display any directionality. It would seem that D. suweonensis was unable to find its way toward the breeding sites. When released in rice paddies, both of the species displayed non-directional displacements, an indication that they had reached their target sites, for breeding and feeding (Kim, 2015a), and that the directionality D. japonicus displayed toward forests before hibernation and toward rice paddies afterwards was not an artifact. Our results raise an interesting question regarding the methods used to display directionality. Amphibian species are known to rely on a set of methods to orientate their movement (see review by (Sinsch, 1990)). However, because all individuals had been kept in laboratory for a month prior to release, or were laboratory born, landscape and field recognition could not be learned. Also, rainy conditions during the fall orientation experiment reduced the possibility of celestial navigation or the use of polarized light.

The behavior expressed by the two species while they were kept in terraria during fall and winter differed clearly between species and between seasons following what would have been expected from wild individuals. During brumation, D. suweonensis was found on the ground of the terraria, whereas we found D. japonicus more often on wooden structures, at higher heights from the ground. The difference in height matches the brumating behavior we recorded in the wild, with D. suweonensis brumating in the vicinity of rice paddies, that is, in areas without high vertical structures, whereas D. japonicus is found on trees most of the time. The preference for bean leaves by D. suweonensis in the wild, around 50 cm high, may thus be more closely related to prey availability than microhabitat preference. The hibernation period saw the two species exploiting the ground microhabitat, and thus hints at the use of buried hibernacula for hibernating by the two species. Dryophytes japonicus was more active and was found higher up during the period, in agreement with the high freeze tolerance shown by the species (Storey and Storey, 2017).

Amphibian populations have greatly declined in recent decades (Blaustein et al., 1994), with approximately one third of all species currently under threat of extinction and more than 200 species already extinct (Hayes et al., 2002). The potential for extinctions in pristine environments such as Madagascar (Kolby, 2014) is still high, and basic ecological knowledge is still required for a high number of amphibians, as highlighted by the Data Deficient designation of species by the IUCN (International Union for Conservation of Nature) on its Red List of endangered species. In response to these losses, biodiversity conservation efforts have been deployed to tackle population decreases and extinctions (Marsh and Trenham, 2001; Gascon, 2007). However, and especially with amphibians, most conservation work is addressed to the breeding habitat, and thus, not all conservation efforts have been successful (Blaustein and Kiesecker, 2002). Failure is not necessarily attributable to the work carried out but because of limiting factors such as basic knowledge. We therefore urge the implementation of conservation measures for D. suweonensis, listed as endangered by the IUCN (IUCN,

#### REFERENCES


2018) and under Korean law (Ministry of Environment, 2012), highlighting that protecting the totality of the space used by the species is easier than for most species, as hibernation and breeding habitats overlap. Despite the areas to protect being only in rice paddies, international conservation plans such as the ones developed under the Ramsar convention will meet the dual objective.

#### AUTHOR CONTRIBUTIONS

AB designed the experiment, collected and analyzed the data and wrote the manuscript. YC analyzed data and participated in writing. YK collected data and participated in writing. PJ supervised. YJ designed the experiment, revised the manuscript, and supervised the project.

#### FUNDING

Our work was financially supported by a grant from the National Geographic Society Asia (Young Explorer #17-15), two Small Grants for Science and Conservation from The Biodiversity Foundation (2015 and 2016) to AB; a grant from Korea Environmental Industry & Technology Institute (KEITI RE201709001), a grant from the Korea Rural Community Corporation for the translocation of the Suweon treefrog individuals, a grant from the National Geographic Asia (#2-2016-1632-001-1), a grant (#PJ01228503) by the Rural Development Administration of Korea, a research grant (#2017R1A2B2003579) by the National Research Foundation of Korea to YJ; and by the BK21 program at the School of Biological Sciences.

#### ACKNOWLEDGMENTS

We apologize for difficulties encountered by color-blind readers and recommend printing the figures in shades of gray for easier understanding. We are grateful to Minwha Hong, Yeongseon Park, Miyeon Kim, Junyung Kim, and Taeho Kim for their help rearing and tracking the frogs.

Beach, F. A. (1961). Hormones and Behavior. New York, NY: Cooper Square.


Animal Syst. Evol. Divers. 31, 176–181. doi: 10.5635/ASED.2015. 31.3.176


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Borzée, Choi, Kim, Jablonski and Jang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.