# NORTH AMERICAN MONARCH BUTTERFLY ECOLOGY AND CONSERVATION

EDITED BY : Jay E. Diffendorfer, Wayne E. Thogmartin and Ryan G. Drum PUBLISHED IN : Frontiers in Ecology and Evolution and Frontiers in Environmental Science

#### 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 978-2-88966-118-3 DOI 10.3389/978-2-88966-118-3

#### 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

# NORTH AMERICAN MONARCH BUTTERFLY ECOLOGY AND CONSERVATION

Topic Editors:

Jay E. Diffendorfer, United States Geological Survey (USGS), United States Wayne E. Thogmartin, United States Geological Survey (USGS), United States Ryan G. Drum, United States Fish and Wildlife Service (USFWS), United States

Image: Jeremy Havens

Citation: Diffendorfer, J. E., Thogmartin, W. E., Drum, R. G., eds. (2020). North American Monarch Butterfly Ecology and Conservation. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88966-118-3

# Table of Contents


Victoria Marie Pocius, John M. Pleasants, Diane M. Debinski, Keith G. Bidne, Richard L. Hellmich, Steven P. Bradbury and Sue L. Blodgett

*19 An Evaluation of Studies on the Potential Threats Contributing to the Decline of Eastern Migratory North American Monarch Butterflies (*Danaus plexippus*)*

Alana A. E. Wilcox, D. T. Tyler Flockhart, Amy E. M. Newman and D. Ryan Norris


Leona K. Svancara, John T. Abatzoglou and Beth Waterbury


Beth Waterbury, Ann Potter and Leona K. Svancara

*88 Host Plants and Climate Structure Habitat Associations of the Western Monarch Butterfly*

Thomas E. Dilts, Madeline O. Steele, Joseph D. Engler, Emma M. Pelton, Sarina J. Jepsen, Stephanie J. McKnight, Ashley R. Taylor, Candace E. Fallon, Scott H. Black, Elizabeth E. Cruz, Daniel R. Craver and Matthew L. Forister

*105 The Importance of Shifting Disturbance Regimes in Monarch Butterfly Decline and Recovery*

Nathan L. Haan and Douglas A. Landis

*113 The Integrated Monarch Monitoring Program: From Design to Implementation*

Alison B. Cariveau, Holly L. Holt, James P. Ward, Laura Lukens, Kyle Kasten, Jennifer Thieme, Wendy Caldwell, Karen Tuerk, Kristen A. Baum, Pauline Drobney, Ryan G. Drum, Ralph Grundel, Keith Hamilton, Cindy Hoang, Karen Kinkead, Julie McIntyre, Wayne E. Thogmartin, Tenlea Turner, Emily L. Weiser and Karen Oberhauser

*121 Design Implications for Surveys to Monitor Monarch Butterfly Population Trends*

Karen E. Kinkead, Tyler M. Harms, Stephen J. Dinsmore, Paul W. Frese and Kevin T. Murphy

*132 The Role of Modeling in Monarch Butterfly Research and Conservation* Tyler J. Grant and Steven P. Bradbury

*146 Quantifying Pesticide Exposure Risk for Monarch Caterpillars on Milkweeds Bordering Agricultural Land*

Paola Olaya-Arenas and Ian Kaplan

*162 Expanding the Isotopic Toolbox to Track Monarch Butterfly (*Danaus plexippus*) Origins and Migration: On the Utility of Stable Oxygen Isotope (*d*18O) Measurements*

Keith A. Hobson, Kevin J. Kardynal and Geoff Koehler

*170 Does Nature Need Cities? Pollinators Reveal a Role for Cities in Wildlife Conservation*

Abigail Derby Lewis, Mark J. Bouman, Alexis M. Winter, Erika A. Hasle, Douglas F. Stotz, Mark K. Johnston, Karen R. Klinger, Amy Rosenthal and Craig A. Czarnecki

*178 Estimating Milkweed Abundance in Metropolitan Areas Under Existing and User-Defined Scenarios*

Mark K. Johnston, Erika M. Hasle, Karen R. Klinger, Marc P. Lambruschi, Abigail Derby Lewis, Douglas F. Stotz, Alexis M. Winter, Mark J. Bouman and Izabella Redlinski

*200 Western Monarch Population Plummets: Status, Probable Causes, and Recommended Conservation Actions*

Emma M. Pelton, Cheryl B. Schultz, Sarina J. Jepsen, Scott Hoffman Black and Elizabeth E. Crone


Nickolay I. Hristov, Dionysios Nikolaidis, Tatjana Y. Hubel and Louise C. Allen


Alison B. Cariveau, Erik Anderson, Kristen A. Baum, Jennifer Hopwood, Eric Lonsdorf, Chris Nootenboom, Karen Tuerk, Karen Oberhauser and Emilie Snell-Rood

*295 A Method to Project Future Impacts From Threats and Conservation on the Probability of Extinction for North American Migratory Monarch (*Danaus plexippus*) Populations*

Kristen J. Voorhies, Jennifer Szymanski, Kelly R. Nail and Mason Fidino

#### *308 Spatio-Temporal Distribution of Monarch Butterflies Along Their Migratory Route*

Saul Castañeda, Francisco Botello, Víctor Sánchez-Cordero and Sahotra Sarkar

#### *318 Recent Forest Cover Loss in the Core Zones of the Monarch Butterfly Biosphere Reserve in Mexico*

José Juan Flores-Martínez, Anuar Martínez-Pacheco, Eduardo Rendón-Salinas, Jorge Rickards, Sahotra Sarkar and Víctor Sánchez-Cordero

#### *326 Ecological Restoration of* Abies religiosa *Forests Using Nurse Plants and Assisted Migration in the Monarch Butterfly Biosphere Reserve, Mexico*

Aglaen Carbajal-Navarro, Esmeralda Navarro-Miranda, Arnulfo Blanco-García, Ana Laura Cruzado-Vargas, Erika Gómez-Pineda, Cecilia Zamora-Sánchez, Fernando Pineda-García, Greg O'Neill, Mariela Gómez-Romero, Roberto Lindig-Cisneros, Kurt H. Johnsen, Philippe Lobit, Leonel Lopez-Toledo, Yvonne Herrerías-Diego and Cuauhtémoc Sáenz-Romero

#### *342 Adult Monarch (*Danaus plexippus*) Abundance is Higher in Burned Sites Than in Grazed Sites*

Julia B. Leone, Diane L. Larson, Jennifer L. Larson, Nora Pennarola and Karen Oberhauser

#### *355 Configuration and Location of Small Urban Gardens Affect Colonization by Monarch Butterflies*

Adam M. Baker and Daniel A. Potter

*365 Is the Timing, Pace, and Success of the Monarch Migration Associated With Sun Angle?*

Orley R. Taylor Jr., James P. Lovett, David L. Gibo, Emily L. Weiser, Wayne E. Thogmartin, Darius J. Semmens, James E. Diffendorfer, John M. Pleasants, Samuel D. Pecoraro and Ralph Grundel

*380 Monarch Habitat in Conservation Grasslands* Laura Lukens, Kyle Kasten, Carl Stenoien, Alison Cariveau, Wendy Caldwell and Karen Oberhauser

# *393 Evidence for a Growing Population of Eastern Migratory Monarch Butterflies is Currently Insufficient*

Wayne E. Thogmartin, Jennifer A. Szymanski and Emily L. Weiser

# *398* Abies religiosa *Seedling Limitations for Passive Restoration Practices at the Monarch Butterfly Biosphere Reserve in Mexico*

Gerardo Guzmán-Aguilar, Aglaen Carbajal-Navarro, Cuauhtémoc Sáenz-Romero, Yvonne Herrerías-Diego, Leonel López-Toledo and Arnulfo Blanco-García

#### *408 Evaluating the Migration Mortality Hypothesis Using Monarch Tagging Data*

Orley R. Taylor Jr., John M. Pleasants, Ralph Grundel, Samuel D. Pecoraro, James P. Lovett and Ann Ryan

# Editorial: North American Monarch Butterfly Ecology and Conservation

#### Jay E. Diffendorfer <sup>1</sup> \*, Wayne E. Thogmartin<sup>2</sup> , Ryan Drum<sup>3</sup> and Cheryl B. Schultz <sup>4</sup>

*<sup>1</sup> Geosciences and Environmental Change Science Center, United States Geological Survey, Denver, CO, United States, <sup>2</sup> Upper Midwest Environmental Sciences Center, United States Geological Survey, LaCrosse, WI, United States, <sup>3</sup> U.S. Fish and Wildlife Service, Bloomington, MN, United States, <sup>4</sup> School of Biological Sciences, Washington State University, Vancouver, WA, United States*

Keywords: Danaus plexippus, habitat restoration, population ecology, migration, monarch butterfly, monitoring, strategic habitat conservation

**Editorial on the Research Topic**

#### **North American Monarch Butterfly Ecology and Conservation**

Spanning Canada, the United States, and Mexico, North America contains two populations of the migratory monarch butterfly (Danaus plexippus). The smaller "western" population overwinters in groves along the California coast and breeds west of the Rocky Mountains, while the much larger "eastern" population breeds east of the Rocky Mountains and overwinters in Oyamel fir forests in central Mexico. Both populations have declined in the last 20 to 30 years, leading to a formal petition in 2014 to list the species as threatened or endangered under the US Endangered Species Act (ESA) and a recommendation in 2016 for listing as endangered under the Canadian Species at Risk act.

#### Edited by:

*Maria L. Pappas, Democritus University of Thrace, Greece*

#### Reviewed by:

*Douglas Landis, Michigan State University, United States*

\*Correspondence: *Jay E. Diffendorfer jediffendorfer@usgs.gov*

#### Specialty section:

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

Received: *25 June 2020* Accepted: *25 August 2020* Published: *25 September 2020*

#### Citation:

*Diffendorfer JE, Thogmartin WE, Drum R and Schultz CB (2020) Editorial: North American Monarch Butterfly Ecology and Conservation. Front. Ecol. Evol. 8:576281. doi: 10.3389/fevo.2020.576281*

The response to monarch declines in North America includes trinational (CAN, MEX, and USA) conservation agreements, federal and state management actions, non-governmental organizations programs, and concerted effort by individual citizens. The concomitant rise in science devoted to monarch ecology and conservation was the motivation for this Research Topic. The editors participated in the Monarch Conservation Science Partnership (MCSP) from 2014 to present and attended trinational monarch meetings sponsored by the Commission for Environmental Cooperation. The MCSP convened meetings including leading academic and government scientists from all three countries, managers from federal and state agencies, and representatives of conservation organizations to identify, prioritize, design, and apply science needed to conserve and recover the eastern and western migratory monarch populations. These scientific endeavors were a holistic approach, encompassing all elements of strategic habitat conservation (biological planning, conservation design, habitat delivery, monitoring, and assumption-driven research, National Ecological Assessment Team, 2006), and collectively serve to strengthen the scientific foundation for the impending threatened or endangered listing decisions.

This Research Topic expands on some of the work stemming from the MCSP as well as contributions from a much larger international scientific community. We purposely cast a broad net, inviting scientists and practitioners working on all aspects of monarch ecology and conservation, with the intent of illuminating the front lines of monarch conservation science prior to the listing decision. We sent out an open invitation and personally invited 67 scientists who had published research on monarchs or monarch habitat. Ultimately 34 articles were published by >150 authors. This is a subset of monarch-related research, reflecting widespread efforts to equip decision makers with the best available science to inform the listing process. Based on a Web of Science search for either "monarch butterfly" OR "Danaus plexippus" from 2017 to 2019, scientists published over 200 papers on topics related to monarch biology and conservation (55 in 2017, 64 in

**6**

2018, and 96 in 2019). This collection in Frontiers includes basic monarch ecology and population dynamics, social science on attitudes toward monarch conservation, and the conservation and restoration of breeding, migratory and overwintering habitats.

Collectively, these papers demonstrate a vibrant, international community of scientists working diligently to fill key knowledge gaps associated with monarchs and their habitat. Much of this science is uniquely co-produced with managers and decision makers working alongside scientists. Like many species of concern, this work covers four main themes: habitat identification, management and restoration; population ecology; monitoring populations and habitats; and human dimensions and policy.

#### HABITAT

Monarch habitat, including breeding, overwintering, and migratory, is perhaps the largest focus of conservation efforts. In Mexico and coastal California, maintaining overwintering habitat and managing for climate change motivated some of our contributions. Forest cover loss in Mexican overwintering areas has been greatly reduced, due to shared conservation efforts with local communities (Flores-Martínez et al.), though some deforestation continues. In addition, contributions suggest reforestation practices using nurse plants enhance seedling survival of Oyamel fir (Carbajal-Navarro et al.) and new strategies for reforestation are constantly being improved (Guzmán-Aguilar et al.).

The loss of breeding habitat is considered a primary cause of decline for both eastern and western populations (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Oberhauser et al., 2017; Pleasants, 2017; Pelton et al.; Wilcox et al.) and so identification, conservation, and restoration of breeding habitat is a key management concern. Identifying habitat remains a key issue for the western population and Dilts et al. developed a seven-state map of non-overwintering habitat, demonstrating that habitat suitability is structured by both host plant habitat associations and climate variables. Waterbury et al. identified key breeding habitat in Idaho and Montana and showed, based on phenology, that it could support two or three generations of breeding per year. Finally, Svancara et al. suggested breeding habitat in Idaho may remain relatively stable under climate change given contrasting patterns of expansion and contraction of different milkweed species distributions.

For the entirety of North America, Castañeda et al. developed monthly distribution maps for monarchs, identifying key migratory habitat and showing large contractions of suitable areas from April to December during the monarch life cycle. Semmens and Ancona showed riparian buffer strips could make substantial contributions to monarch breeding habitat, while Johnston et al., Derby Lewis et al., and Baker and Potter emphasized various contributions of urban environs in conserving monarchs given the potential for high-density milkweed and nectar source plantings.

Restoring and managing breeding habitat is an area of active research. Contributions suggested monarch females lay more eggs on A. incarnata and A. syrica than other milkweed species in Iowa, emphasizing we need more studies on how monarchs select milkweed (Pocius et al.). In addition, adult monarchs have higher abundances in burned than in heavily grazed prairies, and high cattle stocking levels may negatively impact monarchs (Leone et al.). Haan and Landis argued more generally that implementing disturbance regimes into management may improve monarch habitat and called for more research on this topic. Lukens et al. demonstrated conserved grasslands provide much higher densities of milkweeds than previously thought and began to tease apart important relationships between management practices and habitat responses.

# POPULATION ECOLOGY AND MIGRATION

The monarch life cycle is a multi-generational, spatially complex process taking place across many habitats that involves both longdistance migration and a long period of diapause. Fundamental processes driving migration and population dynamics remain important areas of study, as does predicting population dynamics through time and understanding how populations are impacted by threats and will respond to management actions. Wilcox et al. reviewed the existing literature on demography and threats to the eastern population and concluded breeding and overwinter habitat loss, in addition to a decline in suitable environmental conditions, are the most likely threats to longterm viability. Nail et al. described the global distribution of the species, recording over 90 countries, islands, and island groups where monarchs occur; they also discussed important differences in morphology, migration, overwintering behavior, natural enemies, larval diet, and genetics among these populations. Pelton et al. describe the 97% decline in historical abundance of the western monarch population and argue it may be nearing an extinction vortex. Crewe et al. teased apart the role of migration vs. summer breeding factors on monarchs in southern Canada; they found that the Canadian population is likely affected by variation in reproductive condition caused by weather conditions experienced during the spring migration. They also found a strong correspondence between breeding population sizes in Canada and the following overwintering population size in Mexico. Larval exposure to pesticides has been identified as a potential threat to monarchs. Olaya-Arenas and Kaplan found 14 pesticide residues in milkweed plants near neighboring croplands; they also found high levels of spatiotemporal variation in pesticide occurrence and called for more detailed studies that can better evaluate risk to monarchs. Within the eastern population's breeding habitat, Dinsmore et al. studied factors affecting site occupancy, colonization, and extinction dynamics; occupancy declined at sites with high woodland cover while extinction and colonization of sites were driven by landscape attributes and site-level habitat.

Models of population and movement dynamics have added to our understanding of monarch populations and risk of decline (Grant and Bradbury). Voorhies et al. continued this work by developing a modeling methodology and tool linking monarch population responses to specific threats or conservation actions for both the eastern and western populations. Scenarios of future monarch populations suggested a continued declining trend, even in best-case scenarios. Thogmartin et al. pointed out how high variability in the eastern monarch overwintering population size makes trend detection difficult; they showed that the recently reported increase in overwintering population size, while a positive result for monarch conservation, does not have sufficient statistical support to suggest an increasing trend in the population.

Two contributions focused on migration. Taylor et al., analyzed citizen science tagging data to suggest that the timing and pace of fall migration is consistent with monarchs seeking a constant sun angle at solar noon and that successful migrants fall within a sun angle window of 57 degrees at the beginning to 46 degrees at the end of the fall migration. Mora Alvarez et al. estimate ∼200,000 monarchs are killed per season where they migrate across roads in two locations in Mexico. It is the first quantitative assessment of monarch road kills in Mexico.

#### MONITORING

Different aspects of monarch biology and its population status have been monitored for some time. For example, overwintering areas in Mexico and California have been routinely monitored since 1994 and 1997 respectively, while community science efforts monitor larvae (Monarch Larva Monitoring Program), the movement of tagged (Monarch Watch) and untagged (Journey North) adults, as well as sightings of immatures and adults throughout the year (iNaturalist and Western Monarch Milkweed Mapper). Efforts to develop a statistically rigorous monitoring program and rapid assessments that facilitate ease of data collection in breeding areas in the eastern breeding population are described by Cariveau, Anderson et al. and Cariveau, Holt et al. and included input by field biologists, monarch specialists, statisticians, land managers, and the US Fish and Wildlife Service. Similarly, Kinkead et al. analyzed monarch densities from a statewide inventory and monitoring program in Iowa and produced the first statewide density estimate of monarchs on breeding grounds. Monitoring of habitat in conservation sites (Flores-Martínez et al.; Lukens et al.) and along roadsides (Cariveau, Anderson et al.) will also be needed to understand how monarchs interact with these areas and how they support overwintering success, reproduction, and migration.

Two contributions describe relatively novel technological advances that may assist in both monitoring population status and other areas of monarch biology. Hristov et al. demonstrated ground-based lidar holds promise as a technology for surveying clusters of overwintering monarchs in both the US and Mexico. Hobson et al. investigated the utility of stable oxygen isotopes as a potential addition to the more commonly used stable hydrogen and carbon isotopes for inferring natal origins of adult monarchs found at overwintering sites.

## SOCIAL ISSUES AND MONARCHS

The monarch is an iconic species in North America, having deep cultural significance in Mexico and high levels of public support and interest in the US and Canada. An imbalance exists between rural and urban areas across the monarch range, with rural areas supplying the bulk of habitat supporting monarchs but people living in urban areas reaping the benefits (Semmens et al., 2018). Using a survey of citizens from the US and Canada along the eastern flyway, Solis-Sosa et al. showed that garnering support of urbanites for monarch conservation can be maximized if such support is led by a not-for-profit organization, strives for transboundary cooperation, and includes communication about anticipated ecological outcomes.

# CONCLUSIONS

Despite being one of the most-studied insects in the world, many important scientific mysteries of monarch butterflies remain, particularly regarding processes driving monarch population dynamics, the effectiveness of broad-scale conservation efforts, and related human dimensions in conservation policy. This system is incredibly complex. Data are limited and driving processes are very difficult to isolate. Thus, a great deal of important and challenging work remains—both in terms of scientific research and conservation work needed on the ground.

Moving forward, tackling the most important issues tracking the status and trends of the population throughout the annual cycle, better understanding western population dynamics, evaluating the effectiveness of conservation efforts, predicting and responding to climate change, and understanding other potential threats—will increasingly rely on collaborative partnerships. Successful collaborations will depend on ongoing coordination, leveraged funding/capacity, data sharing and interdisciplinary synthesis, community science contributions, technological innovations, and expanded applications of social science.

Regardless of forthcoming federal listing decisions in the US and Canada, it appears that this iconic species—and the fascinatingly complex system it inhabits—will continue to offer a prolific arena for applied conservation research. Recent efforts, including those resulting from the MCSP alongside many of the collaborations highlighted in this collection, offer a holistic and integrated framework, linking extinction risk to habitat goals at various scales. This framework was developed alongside a long-term monitoring strategy and directly tied to plans and tools that can guide strategic conservation planning and adaptive management throughout the annual cycle. Viewed as a whole, this framework and guidance is the cutting edge of monarch science and arguably represents the most robust, strategic and timely scientific foundation for applied decision-making possible at this time.

The advances captured in this Research Topic build on decades of successful research, standing on the shoulders of giants such as Lincoln Brower (Oberhauser et al.), among other pioneers of the field. Yet this progress also provides a fresh glimpse into the future frontiers of monarch conservation science. Scanning the horizon, we see a vibrant and increasingly international collaborative research community, an integrated and holistic approach guiding strategic conservation, exciting innovations drawing from interdisciplinary technological advancements, long-term biological monitoring data empowering new opportunities, and a growing awareness of the importance of social science woven into any foreseeable conservation solutions.

The state-led Mid-America Monarch Conservation Strategy (Midwest Association of Fish Wildlife Agencies, 2018) and the Western Monarch Butterfly Conservation Plan (Western Association of Fish Wildlife Agencies, 2019) offer reasons for cautious optimism. Furthermore, the recently signed candidate conservation agreement for monarchs on energy and transportation lands in the US provides a mechanism for habitat conservation and restoration intended to provide a net benefit for monarchs (Monarch CCAA/CCA Development Advisory Team, 2020). These plans and agreements portray the magnitude of both the opportunities and the challenges ahead. The ongoing implementation of these plans serves as a massive real-time experiment, offering a unique and timely opportunity to learn more about the drivers of the system and the effectiveness of our conservation interventions. Successfully capitalizing on this opportunity will require broad-scale collaboration to design research and implement long-term conservation, management, and monitoring efforts. This effort is no easy task to realize. For those mobilizing monarch conservation on the ground and

#### REFERENCES


those conducting applied conservation science, the only viable approach remains: "all hands on deck" (Thogmartin et al., 2017).

# AUTHOR CONTRIBUTIONS

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

### FUNDING

For JD, RD, and WT, this research was conducted in accordance with official duties as employees of the U.S. Government.

#### ACKNOWLEDGMENTS

We thank the U.S. Geological Survey (USGS) Powell Center for Analysis and Synthesis for hosting the Monarch Science Conservation Partnership meetings. This work was supported by the Ecosystems Mission Area and the Land Change Science program at the USGS. We thank M. Steinkamp and M. Wimer, USGS Ecosystems Mission Area, for providing publication support. The findings and conclusions in this article are those of the authors and do not necessarily represent the views of the U.S. Fish and Wildlife Service. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.


**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.

At least a portion of this work is authored by Jay Diffendorfer, Ryan Drum, and Wayne Thogmartin on behalf of the U.S. Government and, as regards Jay Diffendorfer, Ryan Drum, and Wayne Thogmartin and the U.S. Government, is not subject to copyright protection in the United States. Foreign and other copyrights may apply. 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.

# Monarch Butterflies Show Differential Utilization of Nine Midwestern Milkweed Species

Victoria Marie Pocius <sup>1</sup> \*, John M. Pleasants <sup>2</sup> , Diane M. Debinski 2,3, Keith G. Bidne<sup>4</sup> , Richard L. Hellmich<sup>4</sup> , Steven P. Bradbury <sup>5</sup> and Sue L. Blodgett <sup>5</sup>

*<sup>1</sup> Department of Entomology, The Pennsylvania State University, University Park, PA, United States, <sup>2</sup> Department of Ecology, Evolution and Organismal Biology, Iowa State University, Ames, IA, United States, <sup>3</sup> Department of Ecology, Montana State University, Bozeman, MT, United States, <sup>4</sup> Corn Insects and Crop Genetics Research Unit, United States Department of Agriculture, Agricultural Research Station, Department of Entomology, Iowa State University, Ames, IA, United States, <sup>5</sup> Departments of Natural Resource Ecology and Management and Entomology, Iowa State University, Ames, IA, United States*

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*Jennifer Lesley Silcock, The University of Queensland, Australia Panagiotis Milonas, Benaki Phytopathological Institute, Greece*

> \*Correspondence: *Victoria Marie Pocius vmpocius@gmail.com*

#### Specialty section:

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

> Received: *22 June 2018* Accepted: *04 October 2018* Published: *25 October 2018*

#### Citation:

*Pocius VM, Pleasants JM, Debinski DM, Bidne KG, Hellmich RL, Bradbury SP and Blodgett SL (2018) Monarch Butterflies Show Differential Utilization of Nine Midwestern Milkweed Species. Front. Ecol. Evol. 6:169. doi: 10.3389/fevo.2018.00169* Monarch butterfly overwintering numbers have declined over the past 20 years. Restoring habitat that includes milkweeds, the only host plants for larval monarch butterflies, is necessary to increase monarch numbers within the breeding range. The value of different milkweed species for restoration will depend, in part, on the extent to which they are utilized by ovipositing females. The number of eggs laid on different species over a season will be a function of plant size and phenology as well as female preference. We examined seasonal egg deposition and females' oviposition choices by comparing the number of eggs laid by free-flying wild monarchs on each of nine native milkweed species occurring in Iowa (*Asclepias syriaca, Asclepias tuberosa, Asclepias incarnata, Asclepias verticillata*, *Asclepias exaltata*, *Asclepias hirtella*, *Asclepias speciosa, Asclepias sullivantii,* and *Cynanchum laeve*). One plot, consisting of clusters of each of the nine species, was established at each of 14 sites across the state of Iowa. Eggs were counted weekly in June, July and August 2015–2017. The highest egg totals were recorded on *A. incarnata* and *A. syriaca* in all years. Fewer eggs were counted on *A. exaltata, A. hirtella, A. tuberosa, A. verticillata,* and *C. laeve*. Our results show that monarchs prefer some milkweed species over others, but that they can use all nine native milkweed species for oviposition.

Keywords: Danaus plexippus, milkweed species (Asclepias spp), oviposition preference, habitat restoration, conservation

# INTRODUCTION

Habitat loss is one of the leading causes of species decline for many taxa, (Means and Simberloff, 1987; Wilcove et al., 1998; Pimm and Raven, 2000; Ceballos and Ehrlich, 2002,?; Kerr and Cihlar, 2004; Venter et al., 2006; Xu et al., 2018). Over the past 20 years, monarch populations have experienced a significant decline in overwintering numbers (Brower et al., 2012; Espeset et al., 2016; Inamine et al., 2016; Schultz et al., 2017). Loss of milkweed within the breeding range is considered by many scientists to be the leading cause of the decline of the monarch population east of the Rocky Mountains (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Pleasants, 2017; Zaya et al., 2017). Restoration of Midwestern monarch habitat is essential to increase (population numbers Oberhauser et al., 2016 as many of the monarchs that overwinter in Mexico originate from this area Wassenaar and Hobson, 1998; Flockhart et al., 2017. Organizations federal, state, and non-profit) have started efforts to establish monarch habitat, especially adding milkweeds to the landscape in critical land cover/land use categories to enhance monarch reproduction (Thogmartin et al., 2017).

Knowledge of what species to include in habitat restoration is necessary to develop and implement an effective conservation program. Although monarch butterflies (Danaus plexippus) are dependent upon milkweeds (Asclepias spp.) as larvae, there are over 100 species of milkweeds in the U.S. (Woodson, 1954) and we need to know how available each species is throughout the season, which ones are better for larval growth and on which ones monarch females choose to lay eggs. Currently, the majority of monarchs in population east of the Rocky Mountains feed on Asclepias syriaca in the summer (Wassenaar and Hobson, 1998). Rather than reflecting a preference, this may be because disturbance from modern agricultural has made A. syriaca the dominant species on the landscape (Martin and Burnside, 1980). This species may not have been as prevalent in historic landscapes (Hayden, 1919; Pleasants, 2015). More information is needed about monarch butterflies' use of other native milkweed species both as larvae and adults beyond A. syriaca, the milkweed on which all current conservation recommendations are based (Landis, 2013; Pleasants and Oberhauser, 2013; Pleasants, 2017).

Prior work has contributed to our understanding of monarchs' oviposition choices and use of different milkweed species (Cohen and Brower, 1982; Malcolm et al., 1989; Zalucki et al., 1990; Haribal and Renwick, 1996, 1998a,b; Calvert, 1999; Bartholomew and Yeargan, 2002; DiTommaso and Losey, 2003; Ladner and Altizer, 2005; Casagrande and Dacey, 2007) as well as larval survival on different species (Cohen and Brower, 1982; Zalucki et al., 1990; Zalucki and Brower, 1992; Ladner and Altizer, 2005; Yeargan and Allard, 2005; Robertson et al., 2015; Baker and Potter, 2018). These studies have not compared larval survival and oviposition preference patterns across the same set of cooccurring milkweed species in both laboratory and field settings.

In our prior work (Pocius et al., 2017b, 2018), we compared larval survival on nine milkweed species and oviposition preference on four of these species in a laboratory setting. These nine species are native to Iowa, which is a high priority area for Midwestern conservation efforts (Flockhart et al., 2015; Thogmartin et al., 2017). Most milkweed species native to the Midwest have not been evaluated in field experiments. The species we tested included: A. syriaca (common milkweed), Asclepias incarnata (swamp milkweed), Asclepias tuberosa (butterfly milkweed), Asclepias verticillata (whorled milkweed), Asclepias speciosa (showy milkweed), Asclepias exaltata (poke milkweed), Asclepias sullivantii (prairie milkweed), Asclepias hirtella (tall green milkweed), and Cynanchum laeve (honeyvine milkweed). These milkweeds have overlapping ranges (Woodson, 1954; Kaul et al., 1991; Eilers and Roosa, 1994), but varying habitat needs as well as differing concentrations of phytochemicals including cardenolides (Roeske et al., 1976; Malcolm, 1991; Rasmann and Agrawal, 2011); and quercetin glycosides (Haribal and Renwick, 1996). The species also have different plant architecture (stem height, leaf width, leaf shape, stem branching, etc.; Woodson, 1954).

Our laboratory results suggest that monarch larvae will consume, survive, and eventually pupate on all nine Midwestern milkweed species (Pocius et al., 2017b); however, fewer larvae reached adulthood when they fed on A. hirtella and A. sullivantii (Pocius et al., 2017b, see **Table 1**). Larval survival was not significantly different among the other seven milkweed species; these species may provide equal benefits for larvae when included in habitat restorations within the native range of each milkweed species (Pocius et al., 2017b). Our laboratory oviposition results, using just A. incarnata, A. syriaca, A. tuberosa, and A. verticillata, suggest that monarch butterflies prefer to oviposit on A. incarnata and A. syriaca although they will utilize all four species (Pocius et al., 2018-see also Baker and Potter, 2018). Here, we build on prior laboratory work with a report of field oviposition using all nine milkweed species. We compare the total number of eggs laid on each species in June and July, in 2015 and 2016, to compare females' choices when all nine species were present, before senescence of three of the species. We also compared the total number of eggs laid on each of six species present in July and August, in 2015 and 2016, to capture females' choices during peak oviposition. Finally, we compare the total number of eggs laid on each species during the entire summer season, in 2015–2017, to provide estimates of monarch utilization of each milkweed species for habitat restoration purposes.

# MATERIALS AND METHODS

#### Field Oviposition

Experimental Milkweed Plots and Site Establishment Midwestern ecotype milkweed seeds of A. exaltata, A. hirtella, A. incarnata, A. speciosa, A. sullivantii, A. syriaca, A. tuberosa, A. verticillata, and C. laeve (Prairie Moon Nursery, MN, USA) were stratified in wet sand for 6 weeks. After stratification, seeds were sown into 128-cell plug trays (Landmark Plastics, Akron OH, USA) and transplanted into 8.9 cm<sup>2</sup> , deep perennial pots (Kord, Ontario Canada) at approximately 6 weeks following germination. When milkweed plants were 12 weeks old, five young plants of each species were transported to each location. Sites were established at ten Iowa State Research and Demonstration Farms (Newell, IA; Lewis, IA; Boone, IA; Ames, IA; Chariton, IA; Nashua, IA; Kanawha, IA; Sutherland, IA; and Castana, IA), Luther College (Decorah, IA), Pella High School (Pella, IA), Central College (Pella, IA), and on the Sorenson-Powell property (Adel, IA). At least one site was located in each quadrant of the state. Plants were distributed to each site and planted by the second week of June 2015.

Each of nine milkweed species was randomly assigned to a 1 m<sup>2</sup> plot within a single row at each site. Each plot consisted of 5 plants for a total of 45 plants at each site. Plots were separated from each other by a 1 m wide grass or stone path. Any plants that did not survive were replaced with young plants (6–8 weeks old) twice during the summer of 2015 and at the beginning of the


TABLE 1 | Summary of the utility of nine milkweed species examined in the current study.

*Habitat information is summarized from Kaul et al. (1991) and (Eilers and Roosa, 1994). Larval survivorship designated as high if over 60% of larvae reached adulthood (Pocius et al., 2017b); under 60% survival is designated as low. Oviposition use is designated as high if species were in the top third for both laboratory (Pocius et al., 2018) and field oviposition experiments, medium if species were in the second third for both experiments, and low if the species were in the bottom third of egg totals for both experiments. Species are designated as easy to establish if over 60% survived within the demonstration plots from 2015 to 2017, and are recommended for restoration if plants were easy to establish, had high larval survivorship, and medium to high oviposition use. Larval survivorship and oviposition use were determined as low, medium or high from laboratory data Pocius et al., 2017a; Pocius et al., 2018.*

season in 2016. A. hirtella plants were not replaced due to a lack of seed in 2016 and 2017.

#### Site Monitoring

Each site was monitored weekly from the first week of June 2015 through the end of August 2017 for a total of 42 visits to each site. Each week, the number of live milkweed plants, bloom presence, the number of blooms, the height of the tallest plant, the presence of seed pods, and the presence of mature seed pods was recorded for each milkweed species. Each plant was examined for the presence of monarch eggs, larvae, or other insects using a modified protocol from the Monarch Larva Monitoring Project (Oberhauser, 2013).

#### Statistical Analysis

The total number of eggs on each plot of five plants was summed across June, July, and August for each site and then averaged; the results were analyzed separately for each year. Only sites where observers recorded egg numbers for at least 8 weeks were included in the analysis of each year. Sites without any eggs during the summer within each year were removed from the analysis (N = 12 sites in 2015, N = 13 sites in 2016, and N = 10 sites in 2017). Egg counts were only reported for milkweed species with live plants at each site over the observation period. Differences in total egg counts in single years were determined using a Poisson regression with milkweed species (Pocius et al., 2018) as a fixed effect and site a random effect. Pairwise differences in egg counts were determined by comparing least square means for each milkweed species (Pocius et al., 2018); p-values were adjusted using Tukey's range test for multiple comparisons (Pocius et al., 2018). Concordance was determined using a Kendall coefficient of concordance. Correlation between average egg counts and average plant traits were determined using a Pearson correlation. R version 3.3.3 (R Core Team, 2014) was used for all statistical analyses.

To address preference directly, the total number of eggs in each plot of five plants were summed across June and July in 2015 and 2016 when all nine species were available and prior to senescence of A. exaltata, A. hirtella, and A. speciosa. The total number of eggs in each plot were also summed across the six milkweed species present in across July and August in 2015 and 2016 to include the timing of peak oviposition in the analysis of these years. The year 2017 was excluded from preference analyses because some species had disappeared from the plots by then. Only sites where eggs were laid were included in the analysis (N = 12 in 2015 and N = 11 in 2016). Differences in total egg counts in each year were determined using a Poisson regression with milkweed species (Pocius et al., 2018) as a fixed effect and site as a random effect. Plant height and bloom count were not significant predictors of the number of eggs laid per species and were excluded from the final model. Pairwise differences in egg counts were determined by comparing least square means for each milkweed species (Pocius et al., 2018); p-values were adjusted using Tukey's range test for multiple comparisons (Pocius et al., 2018).

#### Pocius et al. Monarch Butterfly Differential Oviposition

# RESULTS

#### Field Oviposition 2015

Milkweed species had a significant effect on the total number of eggs laid per milkweed species. A. incarnata had the highest egg totals when counts from all sites were combined across the entire breeding season (**Figure 1A**). Females laid 1.3 times more eggs on A. incarnata than A. syriaca, although this difference was not significant (z = 2.12, p > 0.4). One of the largest differences in total egg counts was between A. incarnata or A. syriaca and A. exaltata. Females laid 6.8 times more eggs on A. incarnata (z = −4.04, p < 0.001) and 5.4 times more eggs on A. syriaca (z = −6.59, p < 0.001) than on A. exaltata (**Figure 1A**). All other significant pairwise comparisons are shown in **Supplementary Table 1**.

For June-July, when all species were present, A. incarnata and A. hirtella had the highest average egg totals per site (**Figure 2A**). A. incarnata had significantly higher egg counts compared to A. exaltata, C. laeve, A. tuberosa, and A. verticillata (**Figure 2A**, **Supplementary Table 2**). A. exaltata had significantly lower egg counts than A. hirtella and A. sullivantii (**Figure 2A**, **Supplementary Table 2**). The number of eggs laid on A. syriaca was not significantly different from A. incarnata. All other significant pairwise comparisons are shown in **Supplementary Table 2**.

When the period from July through August was examined, with A. exaltata, A. hirtella, and A. speciosa removed due to senescence, A. incarnata and A. syriaca had the highest average egg totals per site (**Figure 3A**). The largest differences in egg counts were between A. incarnata and C. laeve (z = 7.02, p < 0.0001), A. tuberosa (z = 5.86, p < 0.0001), and A. verticillata (z = 6.49, p < 0.0001). A. verticillata (z = 5.02, p < 0.0001), A. tuberosa (z = 4.28, p = 0.0003), and C. laeve (z = −5.65, p < 0.0001) also had significantly fewer eggs than A. syriaca although A. sullivantii and A. syriaca were not significantly different from each other. All other significant pairwise comparisons are shown in **Supplementary Table 3**.

#### 2016

Milkweed species had a significant effect on the total number of eggs laid per milkweed species. A. syriaca had the highest average egg totals followed by A. incarnata (**Figure 1B**). Females laid 1.4 times more eggs on A. syriaca than A. incarnata although this difference was not significant (z = −1.55, p > 0.8). The largest difference in egg counts was observed between A. syriaca or A. incarnata and A. exaltata. Females laid over twenty times more eggs on A. syriaca (z = −4.21, p < 0.01) and A. incarnata (z =

FIGURE 1 | Average eggs counted on each milkweed species over the course of the summer breeding season in 2015 (A), 2016 (B), and 2017 (C). Each bar represents one milkweed species. EXA, *A. exaltata*; HIR, *A. hirtella*; INC, *A. incarnata*; LAE, *C. laeve*; SPE, *A. speciosa*; SUL, *A. sullivantii*; SYR, *A. syriaca*; TUB, *A. tuberosa*; VER, *A. verticillata;* error bars represent 95% confidence intervals. *N* = 12 sites in 2015, 12 sites in 2016, and 10 sites in 2017. Bars that do not share a letter within each panel are significantly different from each other. Females laid more eggs on *A. incarnata* and *A. syriaca* than on *A. exaltata, A. hirtella, C. laeve, A. tuberosa,* and *A. verticillata* in all years (*p* < 0.05). *P*-values were adjusted using the Tukey method for multiple comparisons.

comparisons.

−3.87, p < 0.01) than on A. exaltata in 2016. All other significant

pairwise comparisons are shown in **Supplementary Table 4**. For June-July, when all species were present, species had a significant effect on the average total number of eggs laid per milkweed species (**Figure 2B**). No eggs were laid on A. exaltata during this time period (**Figure 2B**). A. incarnata and A. syriaca had the two highest average egg totals (**Figure 2B**). A. syriaca had significantly higher egg counts than A. hirtella, C. laeve, A. tuberosa, and A. verticillata (**Figure 2B**, **Supplementary Table 5**). A. speciosa and A. sullivantii had comparable egg totals to A. incarnata and A. syriaca (**Figure 2B**). All other significant pairwise comparisons are shown in **Supplementary Table 5**.

When the period from July through August was examined, with A. exaltata, A. hirtella, and A. speciosa removed due to senescence, A. incarnata and A. syriaca had the highest average egg totals (**Figure 3B**), but the largest differences in egg counts were between A. syriaca and C. laeve (z = −5.23, p < 0.0001), A. tuberosa (z = 5.63, p < 0.0001), and A. verticillata (z = 5.68, p < 0.0001). A. verticillata (z = 4.67, p < 0.001), A. tuberosa (z = 4.88, p < 0.001), and C. laeve (z = 4.35, p = 0.0002) also had significantly fewer eggs than A. incarnata. A. incarnata, A. syriaca, and A. sullivantii are not significantly different from each other. All other significant pairwise comparisons are shown in **Supplementary Table 6**.

#### 2017

Milkweed species had a significant effect on the number of total eggs laid per milkweed species. A. incarnata had the highest egg totals while A. syriaca had the second highest egg counts when eggs from all sites were combined (**Figure 1C**). Females laid about 1.3 times more eggs on A. incarnata than A. syriaca, although this difference was not significant (z = 1.29, p > 0.9). Females laid eight times more eggs on A. incarnata than on A. exaltata (z = −4.44, p = 0.0003) and six times more eggs on A. incarnata than on A. hirtella (z = −4.44, p = 0.0003) in 2017. All other significant pairwise comparisons are shown in **Supplementary Table 7**.

#### Comparison Among Years

During each of the 3 years, over the entire summer season, female monarchs laid eggs on all nine milkweed species but a greater number of eggs were laid on some milkweed species than others (**Figure 1**). The species order of the number of eggs laid was highly concordant across years (W = 0.94). Across years the overall utilization of each species is summarized in **Table 1**. There was no significant correlation between the average number of blooms per plant and the average number of eggs per plant (r = 0.18, p = 0.25) or the average number of eggs per plant and species plant height (r = −0.07 to 0.09, p > 0.05). The total

number of eggs laid was 542 (41.7 eggs per site) in 2015, 221 (13 eggs per site) in 2016 and 136 (10.5 eggs per site) in 2017. When species were compared during a subset of the summer, species order of preference was moderately concordant between June-July 2015 and 2016 (W = 0.50) and highly concordant between July-August 2015 and 2016 (W = 0.70).

not significantly different from *A. incarnata*. *P*-values were adjusted using the Tukey method for multiple comparisons.

# DISCUSSION

The findings of our field-based oviposition preference experiment (June through July counts) were consistent across 2015–2016 and suggest that while monarch butterflies will oviposit on all milkweed species tested, some species consistently received fewer total eggs in the research plots; A. exaltata received few eggs across all years. The species on which females chose to oviposit in the June-July period, A incarnata and A. syriaca, were also preferred in July-August. A. incarnata and A. syriaca also had higher egg totals in the field study by Baker and Potter (2018). These two species were also preferred in the laboratory experiment which also included A. verticillata and A. tuberosa (Pocius et al., 2018). Contrary to Zalucki and Kitching (1982) and Baker and Potter (2018), we did not see an increase in egg counts with plant height within species in any year or an increase in the number of eggs laid with increasing bloom count.

Monarchs from the populations both east and west of the Rocky Mountains also have exhibited the same oviposition choices when exposed to the same array of milkweed species (Ladner and Altizer, 2005). Although monarchs exhibited egglaying patterns in this study, they did lay eggs on all nine species each year. This indicates that although monarchs make oviposition choices, they do not specialize on a single milkweed species. This is important for a species that encounters different sets of milkweed species on the landscape during its annual cycle (Zhan et al., 2014; Agrawal, 2017).

Interestingly, the species on which larvae performed well and those with high egg totals were not always correlated (Mayhew, 1997, 2001; Berdegué et al., 1998; Gratton and Welter, 1998 see **Table 1**). For example, both A. tuberosa, and A. verticillata were good larval food sources (Pocius et al., 2017b), but fewer eggs were laid on these species in the lab (Pocius et al., 2018) and in the field. This suggests that the factors that female monarchs use to make egg-laying decisions can be different from those that determine larval success.

We saw more eggs on all species in 2015 than 2016 and 2017. These higher egg totals could be due to the young plant age and smaller stature of first-year plants which made them more attractive (Zalucki and Kitching, 1982). Alternatively, the 2015 observations were reflective of the higher level of egg laying in the Midwest in 2015 as compared to 2016 and 2017 (J. Pleasants pers comm.). Eggs were not present at all sites each year, but no site had zero eggs in 2 consecutive years, demonstrating the variability of egg distribution across Iowa during these 3 years. These differences could be related to varying adult recruitment rates in the spring and subsequent habitat utilization across the state later in the summer.

Across years, fewer monarch eggs were deposited on A. exaltata, A. tuberosa, and C. laeve when compared to A. incarnata and A. syriaca. Both A. exaltata and A. hirtella were difficult to establish in these Iowa sites (**Table 1**). Only four sites had five live plants of both species by August 2017, but the differences in egg counts are apparent in 2015 and 2016 when each site still had 5 live plants of each species. A. exaltata senesced by late July in all years, before peak oviposition occurred. This is likely the primary explanation for its lower overall egg count. The few eggs that we did observe on A. tuberosa were located on flower buds; however, we saw 4th and 5th instars feeding on this species in August. Older larvae may have moved to these plants from the other milkweed species within the site. Because A. tuberosa was in better condition (greener leaves, no visible senescence) compared to A. incarnata, A. speciosa, and A. syriaca late in the growing season, A. tuberosa may be more valuable as a late-season larval food source than for oviposition in August. The utility of C. laeve may be underestimated in our analysis; we observed more eggs on this species anecdotally in September in central Iowa after plot monitoring across the state ended; data from September were not included here. However, it is unlikely that eggs laid that late will successfully produce adults that migrate to Mexico (Orley Taylor pers comm). There is also inherent variation because of the various locations of the research plots. An examination of these site differences is outside the scope of this study.

Annual and inter-annual variation of temperature and precipitation can affect milkweed quality. High-quality milkweed is essential for both larvae and ovipositing females throughout the breeding season. Because some milkweeds thrive in wet conditions (A. incarnata), and others grow well in drier conditions (A. hirtella and A. tuberosa), specialization on one milkweed species is not a viable strategy for ovipositing female monarchs because plant quality is highly variable across the landscape and the duration of the breeding season. Future work should investigate milkweed phenology, milkweed survival after planting, and monarch use across critical areas of the breeding range because the timing for peak oviposition and larval feeding likely differs by location. More information is needed about how monarchs find and use mature, naturally occurring milkweed plants. Understanding how females utilize these mature patches will allow researchers and managers to assess the worth of different milkweed species and the configuration of milkweed patches within habitat restoration sites.

As a whole, the results show that there are a few species that are most preferred for oviposition and would be best to use for restoration purposes (**Table 1**). Other considerations in choosing a species for restoration include matching the habitat preferences of species with the environmental conditions of the restoration site. Planting several milkweeds species with different habitat preferences may allow the persistence of milkweeds at a site despite variable weather conditions within and between years. Because larval survivorship is high on most species, with the exception of a couple (**Table 1**), planting a few species that are less preferred for oviposition will not compromise larval survival. See **Table 1** for a summary of milkweed species' habitat requirements, ease of plug establishment, and utility for larvae and ovipositing females. We designate a species as recommended for restoration if plugs were easy to establish, had high larval survivorship, and medium to high oviposition use.

# AUTHOR CONTRIBUTIONS

These studies were part of the Ph.D. project of VP; the plant and monarch monitoring were done under the supervision of DD and JP. VP, DD, and JP are responsible for the experimental designs. KB, RH, StB, SuB, DD, and JP contributed to site selection, experimental designs, and growing all milkweed species. The manuscript was prepared by VP and critically revised by JP, DD, RH, StB, and SuB.

# FUNDING

This work was partially funded by the USDA National Institute of Food and Agriculture, Hatch project number 1009926 (IOW05478), Prairie Biotics, Inc., The Center for Global and Regional Climate Research (CGRER), and by the USDA, Natural Resources Conservation Service's Conservation Innovation Grant program under Agreement Number 69-3A75- 16-006. The Iowa Monarch Conservation Consortium provided additional support.

#### ACKNOWLEDGMENTS

Mention of a proprietary product does not constitute an endorsement or a recommendation by Iowa State University or USDA for its use. The authors would like to thank Ali Ford, Nancy Shryock, Cory Haggard, Jacqueline Appelhans, Royce Bitzer, Kristen Siewert, Kirk Larsen, Linda Powell, Steve Sorensen, Lyle Rossiter, Dallas Maxwell, Randy Breach, Steve Jonas, Nick Howell, Logan Wallace, Myron Rees, Brandyn Chapman, Ken Pecinovsky, Matt Schnabel, Terry Tuttle, and Chris Beedle for their help with demonstration site establishment, plant replacement, and data collection. The authors would like to thank Nick Oppedal, Kelsey Fisher, Teresa Blader, and Niranjana Krishnan for their help planting, transplanting, and watering milkweed plants grown at Iowa State for this project, and Lincoln Brower for his comments which improved the clarity of this manuscript. We also thank Jay Diffendorfer and two reviewers for their constructive criticism which greatly improved this manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2018.00169/full#supplementary-material

cardenolide concentration. J. Kans. Entomol. Soc. 55, 343–348. DiTommaso, A., and Losey, J. E. (2003). Oviposition preference and larval performance of monarch butterflies (Danaus plexippus) on two invasive swallow-wort species. Entomol. Exp. Appl. 108, 205–209. doi: 10.1046/j.1570-7458.2003.00089.x

Agrawal, A. A. (2017). Monarchs and Milkweeds. Princeton, NJ: Princeton

Baker, A. M., and Potter, D. A. (2018). Colonization and usage of eight milkweed (Asclepias) species by monarch butterflies and bees in urban garden settings. J.

Bartholomew, C., and Yeargan, M. (2002). Phenology of milkweed (Asclepiadaceae) growth and monarch (Lepidoptera: Nymphalidae) reproduction in Kentucky and ovipositional preference between common

Berdegué, M., Reitz, S. R., and Trumble, J. T. (1998). Host plant selection and development in Spodoptera exigua: do mother and offspring know best? Entomol. Exp. Appl. 89, 57–64. doi: 10.1046/j.1570-7458.1998.00381.x Brower, L. P., Taylor, O. R., Williams, E. H., Slayback, D. A., Zubieta, R. R., and Ramirez, M. I. (2012). Decline of monarch butterflies overwintering in Mexico: is the migratory phenomenon at risk? Insect Conserv. Div. 5, 95–100.

Calvert, W. H. (1999). Patterns in the spatial and temporal use of Texas milkweeds (Asclepiadaceae) by the monarch butterfly (Danaus plexippus L.) during fall,

Casagrande, R., and Dacey, J. (2007). Monarch butterfly oviposition on swallowworts (Vincetoxicum spp.). Environ. Entomol. 36, 631–636. doi: 10.1603/0046-

Ceballos, G., and Ehrlich, P. R. (2002). Mammal population losses and the extinction crisis. Science 296, 904–907. doi: 10.1126/science.1069349 Cohen, J. A., and Brower, L. P. (1982). Oviposition and larval success of wild monarch butterflies (Lepidoptera: Danaidae) in relation to host plant size and

Insect Conserv. 1–14. doi: 10.1007/s10841-018-0069-5

doi: 10.1111/j.1752-4598.2011.00142.x

1996. J. Lepidopterists Soc. 53, 37–44.

225X(2007)36[631:MBOOSV]2.0.CO;2

and honeyvine milkweed. J. Kans. Entomol. Soc. 74, 211–220.


Oberhauser, K. S. (2013). Monarch Larva Monitoring Project. Minneapolis, MN.


REFERENCES

University Press.


**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 © 2018 Pocius, Pleasants, Debinski, Bidne, Hellmich, Bradbury and Blodgett. 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.

# An Evaluation of Studies on the Potential Threats Contributing to the Decline of Eastern Migratory North American Monarch Butterflies (Danaus plexippus)

Alana A. E. Wilcox <sup>1</sup> \*, D. T. Tyler Flockhart <sup>2</sup> , Amy E. M. Newman<sup>1</sup> and D. Ryan Norris <sup>1</sup>

*<sup>1</sup> Department of Integrative Biology, University of Guelph, Guelph, ON, Canada, <sup>2</sup> Appalachian Laboratory, University of Maryland Center for Environmental Science, Frostburg, MD, United States*

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*John Pleasants, Iowa State University, United States Douglas Landis, Michigan State University, United States*

> \*Correspondence: *Alana A. E. Wilcox awilco01@uoguelph.ca*

#### Specialty section:

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

Received: *20 December 2018* Accepted: *12 March 2019* Published: *05 April 2019*

#### Citation:

*Wilcox AAE, Flockhart DTT, Newman AEM and Norris DR (2019) An Evaluation of Studies on the Potential Threats Contributing to the Decline of Eastern Migratory North American Monarch Butterflies (Danaus plexippus). Front. Ecol. Evol. 7:99. doi: 10.3389/fevo.2019.00099* The migratory monarch butterflies (*Danaus plexippus*) of eastern North America have undergone large-scale declines, which may be attributable to a variety of underlying causes. The uncertainty about the primary cause of declines and whether individual threats are likely to increase in the future presents challenges for developing effective conservation management and policy initiatives that aim to improve population viability. This paper identifies five potential threats and classifies these threats according to the types of studies (observational, experimental, simulation/models) and their current impact and anticipated risk. Broadly, the threats can be classified into five categories: (1) change in suitable abiotic environmental conditions; (2) deforestation in the overwintering range; (3) exposure to contaminants including the bacteria *Bacillus thuringiensis*, herbicides, and insecticides; (4) loss of breeding habitat; and (5) predation, parasitism, and species-specific pathogens. The vast distribution of the monarch butterfly makes it likely that population declines are attributed to a suite of interacting factors that vary spatially and temporally in their contribution. Nonetheless, the published papers we reviewed suggest the decline in suitable environmental conditions in addition to overwintering (i.e., deforestation) and breeding habitat loss are the most likely threats to continue to affect the population viability of monarch butterflies.

Keywords: conservation threats, habitat loss, population declines, migration, Ophryocystis elektroscirrha, contaminants, climate change, deforestation

#### INTRODUCTION

Insect populations are experiencing rapid declines globally (Dirzo et al., 2014; Stork et al., 2015) that may have implications for ecosystem function and contributions to economic services (Allen-Wardell et al., 1998; Potts et al., 2010). Changes in the suitability of environmental conditions driven by extreme weather and climate change (Batalden et al., 2007; Barve et al., 2012; Brower et al., 2017), habitat loss (Didham et al., 1996; Fattorini, 2011; Thogmartin et al., 2017b), exposure to contaminants (Stanley-Horn et al., 2001; Thogmartin et al., 2017b), and changes in species interactions (e.g., invasive species, Burghardt and Tallamy, 2015) can have profound effects on biodiversity. The substantial loss of biodiversity within the insect taxon emphasizes their sensitivity to environmental perturbations, but also makes them ideal bioindicators (Lenhard and Witter, 1977; Nummelin et al., 2007) for testing the impact of threats and their downstream effects. By integrating an array of research methods and disciplines, the ramifications of declines on biological systems can be betteranticipated and incorporated into conservation management plans and policy initiatives (Vanbergen and Garratt, 2013).

A multi-disciplinary approach is needed to address the loss of insect diversity and to understand the potential mechanisms driving declines. Extending research initiatives beyond traditional economically significant species may allow for identification of shared mechanisms behind observed declines. In doing so, it is important to consider the type and strength of evidence for threats to a population, emphasizing varying susceptibility at different life stages and across the species range. Therefore, the dynamic and synergistic nature of potential threats can be evaluated and considered when developing conservation strategies, especially for animals that cross national and international boundaries.

The eastern migratory North American population of monarch butterflies (Danaus plexippus) undergoes an annual migration between the Sierra Madre Mountains of Mexico and the northern United States and southern Canada (Urquhart, 1960; Urquhart and Urquhart, 1978; Brower, 1995). In Mexico, monarchs overwinter in large colonies for 4–5 months and remain in a reproductive diapause. In spring, individuals begin mating and migrate north to lay eggs on emerging milkweed (Asclepias spp.) in the southern United States (Brower, 1995). Over successive generations the population colonizes much of the eastern and central United States and parts of southeastern and south central Canada (Brower, 1995; Flockhart et al., 2013). The complex nature of the annual cycle and habitat specialization provides a rare opportunity to investigate ecological pressures across a variety of temporal and geographical scales. Moreover, such an understanding can improve our knowledge of the international cooperation required to preserve this flagship species-at-risk.

Conservation management plans rely on the accurate estimation of species abundance. Over the last 2 decades the eastern population of monarch butterflies has declined more than 80% at overwintering sites (Thogmartin et al., 2017b). Arguably overwintering population size determined by the occupied surface area represents the most reliable estimates and denotes the effective population size (i.e., number of individuals contributing to the next generation, Davis, 2012; Ries et al., 2015b). However, counts taken in the northerly portion of the range during pre-migration do not always correspond with those at overwintering sites, which may suggest either methodological issues in estimating population size or high mortality during migration (Davis, 2012; Badgett and Davis, 2015; Ries et al., 2015b; Inamine et al., 2016; Pleasants et al., 2017). Regardless, despite substantial interannual variation in monarch population size (>10-fold, Swengel, 1995; Rendón-Salinas et al., 2014), summer and winter counts show consistent year-to-year fluctuations (Ries et al., 2015a,b). The discrepancy between population estimates in winter and summer highlights the need to distinguish independent threats contributing to potential declines observed throughout the annual cycle and at different developmental stages.

Changes in suitable environmental conditions (Barve et al., 2012; Thogmartin et al., 2017b) and habitat at both breeding (Pleasants, 2017; Thogmartin et al., 2017a) and overwintering (Oberhauser et al., 2017) sites, as well as contaminant exposure (Oberhauser et al., 2006, 2009; Pecenka and Lundgren, 2015), are thought to be foremost threats to monarch butterfly populations. Suboptimal environmental conditions during the overwintering period, such as unseasonably warm temperatures (Hunt and Tongen, 2017) or cold and wet microclimates that pose a risk of freezing (Anderson and Brower, 1996), can accelerate lipolysis that quickly depletes lipid stores needed for overwinter survival (Alonso-Mejía et al., 1997). Overwintering lipid stores may be further reduced by limited availability of nectar sources due to habitat loss and a northward shift in monarch movements expected with climate change (Batalden et al., 2007; Brower et al., 2015; Lemoine, 2015). The introduction and widespread adoption of glyphosate resistant corn and soybean has greatly increased the use of herbicide, causing up to 68% loss of milkweed in some areas of the central United States (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Zaya et al., 2017), and logging and rural development have limited suitable habitat for overwintering sites (Brower et al., 2011b). Finally, environmental contaminants, including genetically-modified (GM) pollen (i.e., Bacillus thuringiensis (Bt), Anderson et al., 2004, 2005) and insecticides (e.g., neonicotinoids, Pecenka and Lundgren, 2015; pyrethroids, Oberhauser et al., 2006, 2009), as well as pathogens [notably, Ophryocystis elektroscirrha (OE)], could contribute to elevated rates of mortality. As a species at risk (SARA 2017, US Fish & Wildlife Service 2017) it is imperative that research identifies the foremost threat(s) resulting in decreased fitness and survival and the relative contribution of each threat to the cumulative population decline.

To better understand the main drivers of monarch decline it is important to evaluate the type of study investigating the threat and strength of evidence (i.e., support) for each potential threat in terms of their effect on monarch butterflies and possible future risk to population viability. Though it is probable that other threats exist beyond those identified in this review, we focus on five broad potential threats commonly reported for the eastern migratory North American population of the monarch butterfly: (1) change in suitable abiotic environmental conditions, (2) deforestation in the overwintering range, (3) exposure to contaminants including the bacteria Bt, herbicides, and insecticides, (4) loss of breeding habitat, and (5) predation, parasitism, and pathogens. We modify a previously established scoring system (Godfray et al., 2013, 2014), sorting and summarizing peer-reviewed research papers into three groups based on the type of study. For each research paper we then assign the level of support for the potential threat affecting monarch butterflies and assess the likelihood of the threat persisting and/or increasing. Though a recent review highlighted anthropogenic impacts on monarch populations (Malcolm, 2017), our aim is to evaluate the strength of evidence available for each threats and highlight key gaps in research needed to guide conservation management plans and policy development.

TABLE 1 | Summary of search terms associated with the declines in the eastern migratory North American population of monarch butterflies (*Danaus plexippus*) used in the comprehensive literature review using Web of ScienceTM.


#### METHODS

A systematic review of available peer-reviewed literature was performed following the procedure outlined in Bechshoft et al. (2017), but using monarch-specific terms associated with hypotheses regarding population declines in the Web of ScienceTM (Reuters, 2016). An initial list of papers was established through a literature search of all databases, including the Web of Science Core Collection, Current Contents Connect, FSTA (The Food Science Resource), KCI (Korean Journal Database), MEDLINE, Russian Science Citation Index, SciELO Citation Index, and Zoological Record, with no limitations placed on the publication date. This list was later refined to retain only peer-reviewed research papers on the eastern migratory North American population of monarch butterflies in English. Search terms ("topic") were combined in pairs using the Boolean operator "and," with select terms having wildcard truncation (<sup>∗</sup> ) to allow for various word endings (**Table 1**).

We sorted and summarized individual peer-reviewed research papers (n = 115) published up to December 2018 by potential threat and assessed the type of study (e.g., [Control\_data], [Field\_data], [Mod]) using a scoring system modified from Godfray et al. (2013, 2014) with the following unranked categories:

[Control\_data] evidence involving controlled experimental studies.

[Field\_data] evidence involving data collected in the field but without experimental control.

[Mod] indirect evidence based on previously collected experimental and/or field collected data to assess the impact of threats on the population or make projections of future environmental and conservation scenarios. Models reflect a degree of uncertainty that cannot be fully accounted for as conditions encountered in the future may deviate from those defined in the model.

After classifying the type of study, we then identified whether research papers independently provided support (i.e., whether a threat was supported [S] or not [N]) for the threat as having a current impact on monarch butterflies and whether there was potential for the threat to pose continued risk to population viability. Support for future risk to the eastern migratory North American population of monarch butterflies was assigned based on when impacts were identified from multiyear and/or historical data or from model projections. Studies where no conclusion could be drawn about the future risk to the population were identified by [-]. We then calculated the percentage of research papers (i.e., number of research papers of the total available studies within each type of threat) for: (1) the type of study, (2) whether the study provided support for a threat to monarch butterflies, and (3) the future risk posed to the population. Papers classified in multiple categories (e.g., [Control\_data] and [Field\_data]) contributed an equivalent number of times to the total number of available studies. Where described in the Results, we also calculated the percentage of represented papers for subcategories within each threat category for the type of study or level of support. For example, within the category of changes in suitable environmental conditions, we calculated the percentage of studies addressing periodic, adverse weather conditions (relative to long-term climate change).

# RESULTS

The summary of peer-reviewed literature, sorted by threat and type of study, assessing the impact, and future risk imposed to the eastern migratory North American monarch butterfly population is presented in **Supplementary Table 1**. The percentage of research papers assigned to each type of study and their support for the specific threat on monarch butterflies and potential risk to the population is presented in **Table 2**.

# Change in Suitable Abiotic Environmental Conditions

Field studies constituted one of the principal methods documenting the effects of sudden changes in environmental conditions and adverse weather patterns in the decline of monarch butterflies (50% of total studies on adverse weather events, Brower et al., 2015, 2017), but only a single study considered the effect of extreme weather patterns before fall migration (25% of total studies on adverse weather events, Hunt and Tongen, 2017). Field studies examined the physiological response of monarchs to changes in environmental conditions in the southern portion of the migratory range (Brower et al., 2015), but controlled studies that assessed field-realistic, shortterm changes in environmental variables such as temperature, humidity, precipitation, or solar radiation on the physiological condition and survival during the breeding season were absent. Similarly, few studies applied modeling techniques that evaluated TABLE 2 | The proportion of peer-reviewed research papers classified by the type of study, effect on monarchs, and potential risk imposed by the threat to the eastern migratory North American population of monarch butterflies (*Danaus plexippus*).


*(Continued)*

TABLE 2 | Continued


the impact of weather extremes on monarch population viability (Flockhart et al., 2015; Hunt and Tongen, 2017).

The peer-reviewed literature suggested a negative impact of adverse weather patterns on monarch butterflies (50% of total studies on adverse weather events, Brower et al., 2017; Hunt and Tongen, 2017) and these conditions could impact monarchs at each stage of their life cycle (Hunt and Tongen, 2017). Though sporadic events may result in considerable losses, the timing of the events is also suggested to alter the severity of the impact. Brower et al. (2015) noted that nectar sources available in the southern portion of the migratory range might offset the energetic cost of adverse conditions experienced earlier in migration, therefore having less impact than extreme weather on overwintering populations. Further, though Hunt and Tongen (2017) showed a negative effect of increasing extreme weather events on monarch butterflies, no studies evaluated the effect of adverse weather patterns in long-term datasets or the extent to which populations are capable of recovering afterwards.

Few controlled experiments investigated the effects of predicted long-term climatic conditions on the condition, growth, and reproduction of monarch butterflies (9% of total studies on climate change) and only a single study explored how rising temperatures impacted host plants at different latitudes (9% of total studies on climate change, Couture et al., 2015). Though no multi-year field studies exist, a substantial number of predictive models (82% of total studies on climate change) attempted to disentangle the effects of long-term climate change on breeding habitat (Zipkin et al., 2012; Lemoine, 2015; Zalucki et al., 2015; Thogmartin et al., 2017b), overwintering conditions (Oberhauser and Peterson, 2003; Barve et al., 2012; Sáenz-Romero et al., 2012; Zalucki et al., 2015; Thogmartin et al., 2017b), and overall distribution (Batalden et al., 2007).

Climatic conditions are anticipated to change drastically overtime. In line with the temperature-dependent growth of monarchs (Zalucki, 1982), elevated temperatures are likely to positively affect larval growth and survival during the breeding season (Couture et al., 2015). Couture et al. (2015) predicted that larvae growth will increase under temperature- and water-stressed conditions, though it is unclear whether the shorter generation time will result in a greater number of generations overall during the breeding season. Beyond the direct effects on larval growth, models suggest climate change is anticipated to result in a northward expansion of the breeding range (Batalden et al., 2007; Lemoine, 2015) and that elevated temperatures (Zipkin et al., 2012) are likely to facilitate population growth. As such, other threats likely have a greater potential to drive monarch declines (Flockhart et al., 2015; Zalucki et al., 2015). Nonetheless, the effect of climate change on monarchs at overwintering sites in Mexico may contribute to lower population viability as rising temperatures may generate unsuitable conditions for diapause (Oberhauser and Peterson, 2003; Barve et al., 2012; Sáenz-Romero et al., 2012). Taken together, the majority of studies implied that the threat of climate change is likely to continue (73% of total studies on climate change, Oberhauser and Peterson, 2003; Batalden et al., 2007; Barve et al., 2012; Sáenz-Romero et al., 2012; Flockhart et al., 2015; Lemoine, 2015; Thogmartin et al., 2017b), but the analysis of multi-year datasets suggest that it may affect population viability (Zipkin et al., 2012; Flockhart et al., 2015); (Zalucki et al., 2015).

#### Deforestation in the Overwintering Range

The level and effect of deforestation is quantified in the peerreviewed literature primarily by means of field studies (82%, **Table 2A**). Field observations and aerial surveys assessed the extent of forest canopy loss (Brower et al., 2002; Ramírez et al., 2003; García, 2001; Honey-Rosés et al., 2011; Navarrete et al., 2011; Champo-Jiménez et al., 2012; Vidal and Rendón-Salinas, 2014; Vidal et al., 2014), microclimate suitability (Anderson and Brower, 1996; Alonso-Mejía et al., 1997; Brower et al., 2009, 2011b) and predation levels (Alonso-Mejía et al., 1998) under changing forest conditions. Models (18%, **Table 2A**), relative to controlled studies (no studies, **Table 2A**), were used to quantify the likelihood that forest loss contributed to monarch butterfly declines.

All field studies and models suggested that forest loss is a likely contributor to declines in individual condition and population viability of monarch butterflies through its effect on available overwintering habitat (Brower et al., 2002; Ramírez et al., 2003; García, 2001; Honey-Rosés et al., 2011; Navarrete et al., 2011; Champo-Jiménez et al., 2012; Vidal and Rendón-Salinas, 2014; Vidal et al., 2014; Flockhart et al., 2015; Hunt and Tongen, 2017; Oberhauser et al., 2017) and suitable environment conditions (Anderson and Brower, 1996; Alonso-Mejía et al., 1997, 1998; Brower et al., 2009, 2011b). However, studies varied on their assessment of the potential future risk to monarch populations. Multi-year data sets and modeling experiments showed continued forest loss within the Monarch Butterfly Biosphere Reserve (Brower et al., 2002; Ramírez et al., 2003; García, 2001; Navarrete et al., 2011; Champo-Jiménez et al., 2012; Vidal and Rendón-Salinas, 2014; Vidal et al., 2014; Flockhart et al., 2015) that may increase the exposure of monarchs and therefore the probability of a mass mortality event (53%, **Table 2C**). Improving habitat protection and availability at overwintering sites in Mexico is also considered a potential means to reduce losses (Oberhauser et al., 2017). Yet, other studies suggested that, though illegal logging and deforestation likely contributes to monarch declines, it is not the primary driver (Flockhart et al., 2015; Hunt and Tongen, 2017) and further forest canopy losses would be required to significantly impact monarch populations (Hunt and Tongen, 2017). The results from the remaining studies (Brower et al., 2009, 2011b; Honey-Rosés et al., 2011; 35%, **Table 2C**, Anderson and Brower, 1996; Alonso-Mejía et al., 1997, 1998) did not suggest a continued threat from deforestation.

#### Exposure to Contaminants

The type of study assigned to peer-reviewed research papers investigating the effects of environmental contaminants varied depending on the nature of the contaminant. Control (56% of total studies on Bt) and field-based (44% of total studies on Bt) studies were principally used to assess the effects of Bt exposure from GM crops. Control studies (75% of total studies on insecticides) pre-dominated for work on insecticides and modeling experiments examined the effect of insecticides (13% of total studies on insecticides) and herbicides (100% of total studies on herbicides) on population abundance. Though, certain types of studies are notably absent for each contaminant, perhaps most importantly the lack of field-based studies on the effect of contaminants limits extrapolation of results to field-realistic scenarios. Further, the majority of studies did not investigate multi-year datasets (57%, **Table 2C**) and control and field studies were limited to individual chemicals without considering the wide-range of potential agrochemicals or their interactions.

Controlled laboratory experiments showed a negative effect of Bt on larval development and survival (Losey et al., 1999; Hansen Jesse and Obrycki, 2000; Stanley-Horn et al., 2001; Anderson et al., 2004, 2005; Dively et al., 2004) and reproduction (Tschenn et al., 2001), but effects were dependent on age (Hansen Jesse and Obrycki, 2000; Hellmich et al., 2001) and Bt-transformation event (i.e., specific occurrence of the uptake of genetic material via transformation of cells of Bt, Stanley-Horn et al., 2001). Studies also confirmed that range overlap with Bt-exposed fields (Oberhauser et al., 2001; Pleasants et al., 2001) could contribute to lower reproductive output (Stenoien et al., 2015), though larval mortality is not always associated with proximity to Bt-exposed fields (Zangerl et al., 2001). Similarly, insecticide use showed effects on individual survival (pyrethroids, Oberhauser et al., 2006, 2009; clothianidin, Pecenka and Lundgren, 2015, λ-cyhalothrin, Stanley-Horn et al., 2001) and herbicide application (i.e., glyphosate) is known to influence population size (Thogmartin et al., 2017b). However, the strength of the effects varied depending on the agrochemical (i.e., significant effects not shown for dicamba and 2,4-dichlorophenoxyacetic acid use, Thogmartin et al., 2017b), geographic location (Thogmartin et al., 2017b), and life stage (Pan et al., 2017). Overall, few studies suggested a potential future risk to monarch population viability from Bt-exposure (31% of total studies on Bt, Hansen Jesse and Obrycki, 2000; Pleasants et al., 2001; Dively et al., 2004; Stenoien et al., 2015), dicamba (33% of total studies on herbicides, Thogmartin et al., 2017b), 2,4-dichlorophenoxyacetic acid (33% of total studies on herbicides, Thogmartin et al., 2017b), and glyphosate (33% of total studies on herbicides, Thogmartin et al., 2017b) applications. Neonicotinoids also did not contribute significantly to monarch declines in a population model (25% of total studies on neonicotinoids, Thogmartin et al., 2017b).

### Loss of Breeding Habitat

The effects of habitat availability and the influence of urbanization are rarely examined for their impacts on monarch butterflies. In fact, habitat fragmentation was only evaluated in the context of field studies of vehicular collisions or roadsides serving as ecological traps (McKenna et al., 2001; Mueller and Baum, 2014) and through 2 modeling experiments that examined the impact of fragmented landscapes on movement rate (Collingham and Huntley, 2000) and path (Grant et al., 2018). Likewise, only a single study investigated how the availability of nectar resources affected monarch population growth (Oberhauser et al., 2017). Projections on the threat of habitat fragmentation and urbanization were absent, but the reduction in nectar availability (Oberhauser et al., 2017) was suggested to pose a sustained risk.

The literature on optimal breeding habitat for monarch butterflies is dominated by studies on the potential declines in survival on different host plants (35% of total studies on breeding habitat loss) and the declines in common milkweed (Asclepias syriaca) in agricultural fields (47% of total studies on breeding habitat loss). Controlled laboratory experiments investigated the oviposition tendencies on different host plants and the effect on larval growth (DiTommaso and Losey, 2003; Mattila and Otis, 2003; Yeargan and Allard, 2005; Casagrande and Dacey, 2007; Pocius et al., 2017a,b), except for two studies that found higher oviposition on common and swamp milkweed (A. incarnata, Pocius et al., 2018) and greater numbers of immature larvae on tropical milkweed (A. curassavica; Malcolm and Brower, 1986) relative to other milkweed species. The effect of the loss of milkweed, principally on agricultural plots, was limited principally to field studies (40% of total studies on milkweed loss, Hartzler, 2010; Pleasants and Oberhauser, 2013; Inamine et al., 2016; Kasten et al., 2016; Zaya et al., 2017) and modeling experiments (60% of total studies on milkweed loss) relating overwintering population abundance to milkweed availability (Zalucki and Lammers, 2010; Flockhart et al., 2015; Zalucki et al., 2016; Hunt and Tongen, 2017; Oberhauser et al., 2017; Pleasants, 2017; Thogmartin et al., 2017a,c).

Studies provide competing evidence that select species of plants (e.g., dog-strangler vine (Cynanchum rossicum), swallowworts (Vincetoxicum spp.) resulted in changes in oviposition tendencies (DiTommaso and Losey, 2003; Mattila and Otis, 2003; Casagrande and Dacey, 2007), and larval survival (Mattila and Otis, 2003). Larval survival also varied across milkweed species and was generally higher on common milkweed (Yeargan and Allard, 2005; Pocius et al., 2017a,b). Tropical milkweed posed a more substantial threat as a greater number of larvae are found on this species relative to common milkweed (Malcolm and Brower, 1986) and year-round availability may alter migration patterns (Satterfield et al., 2015, 2018). At the same time, declines in common milkweed was almost unanimously agreed upon as a threat to monarchs, with the exception of a 22-year study that showed monarch population size is predictable along the migratory route and monarchs are capable of recovering during the breeding season (Inamine et al., 2016; but see Pleasants et al., 2017). Studies did not evaluate the sustained risk of the use of different host plant species during oviposition. The threat imposed by milkweed loss (Hartzler, 2010; Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Zalucki et al., 2016; Hunt and Tongen, 2017; Oberhauser et al., 2017; Pleasants, 2017; Pleasants et al., 2017; Thogmartin et al., 2017a,c; Zaya et al., 2017) to the eastern migratory North American population of monarch butterflies is anticipated to continue (80% in support from the total studies on milkweed loss), though present-day loss may occur at a lower rate than historical levels (Pleasants, 2017).

# Predation, Parasitism, and Species-Specific Pathogens

Predation and parasitism events experienced by monarch butterflies were recorded in controlled experiments (34%, **Table 2A**) and in field observations (60%, **Table 2A**). A variety of species were noted to prey upon monarch butterflies (e.g., flies, Arnaud, 1978; spiders, Borkin, 1982; orioles and grosbeaks, Fink and Brower, 1981). Moreover, a single quantitative risk assessment model determined the impact of the Asian lady beetle (Harmonia axyridis) on monarch butterflies (Koch et al., 2006). The effect of OE on body condition (Altizer and Oberhauser, 1999), flight capacity (Bradley and Altizer, 2005), reproduction (Altizer and Oberhauser, 1999), survival (Altizer and Oberhauser, 1999), and virulence, spore load and transmission (Leong et al., 1997; de Roode et al., 2008a,b, 2009; de Roode and Altizer, 2010; Satterfield et al., 2015) was principally quantified using controlled experiments (63% of total studies on OE), but OE detection was also available through in-field observations (32% of total studies on OE, Urquhart, 1966; Leong et al., 1992). Thogmartin et al. (2017b) provided the sole instance modeling the effect of OE on population size. Infield observations based on counts during migration determined the potential for migratory culling (80% of total studies on migratory culling, Altizer et al., 2000; Bartel et al., 2011; Badgett and Davis, 2015) and a single model determined that there was not a disconnect between monarch population estimates on summering and overwintering grounds, as would be predicted for migratory culling (20% of total studies on migratory culling; Pleasants et al., 2017).

A negative effect of predation and parasitism was found in all peer-review literature studying such events (94%, **Table 2B**), but few papers examined multi-year datasets or modeled the potential for the threat to pose future risk to the eastern migratory North American population of monarch butterflies (85%, **Table 2C**). Of those available, parasitism by OE was likely to have a continued negative effect, particularly with the increased availability of year-round tropical milkweed in the South (Satterfield et al., 2015, 2018; Thogmartin et al., 2017b). Though migratory culling due to OE infection may reduce population abundance at overwintering sites (Altizer et al., 2000; Bartel et al., 2011), other studies suggested this is unlikely the case (Pleasants et al., 2017). Badgett and Davis (2015) also highlight that monarch population abundance at survey sites in Michigan remained constant from 1996 to 2014, potentially due to the high concentration of monarchs in this region that were born in the Upper Peninsula and Canada, but also suggesting that larval survival during the breeding season could offset losses observed at overwintering sites.

## DISCUSSION

Our review focused on five broad threats to the eastern migratory North American population of monarch butterflies and highlights the dynamic factors that influence monarch butterfly reproduction and survival at different stages of their life cycle and throughout their range. Though evidence exists in support of each threat contributing to the declines in the eastern population of monarch butterflies, based on the potential future risk, we suggest that the change in suitable environmental conditions, specifically that related to climate change, and habitat loss on overwintering (i.e., via deforestation) and breeding grounds are likely the greatest threats.

For each threat, the most common methodology applied was somewhat different. Projections on the decline in the availability of suitable of environmental conditions were evaluated using models (73%, **Table 2A**) estimating range expansion and optimal abiotic conditions under climate change scenarios for both monarchs and their host plant. Studies quantifying deforestation, as expected, principally used observational field methods (82%, **Table 2A**). Models (18%, **Table 2A**) were then used to associate rates of loss and degradation to declines in monarch abundance. The toxicological effects of contaminants on monarch butterflies were principally evaluated using controlled designs (55%, **Table 2A**) and models (14%, **Table 2A**), though few studies were conducted and only 9 agrochemicals (i.e., herbicides and insecticides) were assessed. Study type was equally weighted in evaluating the impact and requirements needed for restoration of breeding habitat, though variability existed when assessing the influence of different host species ([Control\_data]: 82%, [Field\_data]: 18% of total studies on host plant species). More, field studies and models contributed the most in research on the effects of milkweed loss ([Field\_data]: 40%, [Mod]: 60% of total studies on milkweed loss). Finally, the effect of predation, parasites, and pathogens on monarch butterflies was determined primarily through field observations (60%, **Table 2A**), though the effect OE was quantified through controlled experiments (63% of total studies on OE).

Based on the current literature on potential threats in the declines of the eastern migratory North American population of monarch butterflies, the availability of suitable environmental conditions (i.e., climate change) and overwintering and breeding habitat loss arguably have the greatest impact on population viability and potential for continuing risk to monarch populations (Brower et al., 2011a). However, some threats are understudied and should not be discounted in their potential impact to the population. Contaminant exposure may also potentially drive declines based on evidence of the toxicological effects and potential for cumulative sublethal effects, but it is unknown whether the threat will rise given current high level of agrochemical use (Thogmartin et al., 2017b). Risk imposed from predation is also likely to continue given the interaction with climate warming and potential year-round residency by monarchs in the southern US (Sáenz-Romero et al., 2012; Satterfield et al., 2015, 2018).

The five threats highlighted in our review vary temporally (e.g., early vs. late migrants) and spatially (e.g., migrants vs. breeding populations) in their imposed risk. For instance, while exposure to Bt pollen generally reduces survival (Hansen Jesse and Obrycki, 2000; Dively et al., 2004), threat level may be minimized if larval populations do not occur at the same time as pollen shedding (Bartholomew and Yeargan, 2001) and/or contact with toxins is reduced during early development (Hansen Jesse and Obrycki, 2000; Sears et al., 2001). Each factor could interact synergistically, with the strength of effects varying over time. As the availability of milkweed declines around crops, the risk imposed by exposure to agricultural chemicals (e.g., Bt, insecticides, herbicides) is likely to decline in tandem, though no studies have assessed this long-term change. Interactions between threats may also vary in accordance with the pre-dominant threats in a particular region. For instance, climate change may result in phenological mismatch (e.g., milkweed availability during oviposition and nectar sources during breeding and migration) if environmental conditions drive changes in plant growth or the pattern of monarch butterfly migration. Simultaneously, if, as in other species (i.e., honeybees, Mason et al., 2013; Sánchez-Bayo et al., 2015), agrochemical exposure reduces immune system function, the potential elevated risk of exposure to pathogens with climate warming (Altizer et al., 2011) may reduce survival. Thus, a complex array of factors and their interactions must be examined with different methodological protocols to resolve how each potential threat contributes to declines.

Our results are based on the available published peer-reviewed literature, but bias may exist in the publication and dissemination of research that may unintentionally affect meta-analyses and systematic reviews. Though we conducted an extensive literature review, recently completed, unpublished literature may not yet be accessible and thus is unable to be accounted for in the results (Møller and Jennions, 2001). Further, publication bias during the submission, review, and editorial processes may also influence the likelihood of research being available and accessed (Møller and Jennions, 2001). Not only are significant results more likely to be submitted, but novel research with large sample sizes and greater statistical power are more likely to be published (Tregenza, 2002; Joober et al., 2012; Mlinaric et al., ´ 2017). Interestingly, publication record and identity of the author (e.g., gender, nationality, non-English surnames, alphabetical position of the surname in reference list) can also affect the likelihood of publication and subsequent citation rate (Tregenza and Wedell, 1997; Kothiaho, 1999a,b; Møller and Jennions, 2001; Einac and Yariv, 2006). The purpose of the research (e.g., natural history or multi-year modeling experiments) may also influence results. For instance, natural history studies on the effects of predators, parasites, and pathogens highlight the threat to monarchs, but were not intended as long-term studies and we therefore cannot extrapolate from these initial results. As research continues to expand reviews will need to incorporate new knowledge to properly evaluate the strength of evidence and potential threats to the eastern migratory North American population of monarch butterflies.

The threats examined in this review also pose considerable risk to other insect species. We suggest that the monarch butterfly is an ideal candidate to evaluate the contribution and the spatiotemporal interactions of each threat at different stages along the migratory route. Research should contribute to effective conservation management plans aimed at protecting habitat and raising population abundance, while also emphasizing the importance of international cooperation in the protection of species at risk. To accomplish this, studies should tackle questions in an interdisciplinary manner, taking a whole-systems approach, and integrate multiple biological disciplines that address major gaps in methodological procedures (i.e., type of study) and knowledge. An integrated approach to understanding the mechanisms underlying declines will be important in mitigating further losses under escalating and interacting threats and will be vital to developing management responses.

#### CONCLUSION

In this review, we sort and summarize 115 peer-reviewed research papers based on the type of study existing within five broad potential threats, evaluating the effect and potential risk imposed by each threat to the eastern migratory North American population of monarch butterflies. We recommend that research initiatives address hypotheses examining the spatiotemporal nature of each risk and how each factor interacts by integrating fields spanning a range of biological disciplines including,

#### REFERENCES


though not limited to, ecology, physiology, endocrinology, and conservation management. Without thorough knowledge and management action plans, declines in monarch butterflies may have unintended downstream effects. For example, butterflies are valued for their cultural economic services and socioeconomic benefits in terms of ecotourism in the southern portion of their range (Semmens et al., 2018). We encourage more detailed studies on the mechanisms driving declines, particularly those evaluating the relative contribution of each threat throughout the monarch life cycle and its geographic distribution. We also suggest that studies investigate potential interacting factors that may limit capacity to implement conservation management plans.

### AUTHOR CONTRIBUTIONS

The study was designed and conceived by all authors. AW wrote the initial draft of the manuscript and all other authors contributed to writing. All authors gave final approval to the revisions and approval for publication.

## FUNDING

This study is part of AW's Ph.D. thesis and was supported by scholarships from the University of Guelph and the Natural Sciences and Engineering Research Council (NSERC, Canada).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00099/full#supplementary-material


populations: a quantitative risk assessment. Biol. Invas. 8, 1179–1193. doi: 10.1007/s10530-005-5445-x


on monarch butterfly larvae in field studies. Proc. Natl. Acad. Sci. USA. 98, 11931–11936. doi: 10.1073ypnas.211277798


**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 Wilcox, Flockhart, Newman 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.

# Lincoln Brower, Champion for Monarchs

Karen S. Oberhauser <sup>1</sup> \*, Alfonso Alonso<sup>2</sup> , Stephen B. Malcolm<sup>3</sup> , Ernest H. Williams <sup>4</sup> and Myron P. Zalucki <sup>5</sup>

*<sup>1</sup> Arboretum, University of Wisconsin-Madison, Madison, WI, United States, <sup>2</sup> Smithsonian Conservation Biology Institute, National Zoological Park, Washington, DC, United States, <sup>3</sup> Department Biological Sciences, Western Michigan University, Kalamazoo, MI, United States, <sup>4</sup> Department Biology, Hamilton College, Clinton, NY, United States, <sup>5</sup> School of Biological Sciences, The University of Queensland, Brisbane, QLD, Australia*

#### Keywords: Lincoln Brower, automimicry, monarch butterfly conservation, cardenolides, monarch butterfly biosphere reserve

Lincoln Pierson Brower died in Virginia, USA, on July 17, 2018 at the age of 86. Many of the authors of papers in this special volume worked directly with Lincoln, and all were influenced by his work. In particular, for the past three decades, Lincoln worked extensively with his wife Linda Fink, who helped him in field research, in discussing monarch biology, and in critiquing all his written work. Multiple eulogies to Lincoln have been published; here, we describe ways in which Lincoln influenced us as scientists and the worlds of monarch science and conservation, from early in his career until its end.

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*Andrew K. Davis, University of Georgia, United States Robert Pyle, Independent Researcher, Grays River, WA, United States*

\*Correspondence:

*Karen S. Oberhauser koberhauser@wisc.edu*

#### Specialty section:

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

> Received: *15 January 2019* Accepted: *16 April 2019* Published: *08 May 2019*

#### Citation:

*Oberhauser KS, Alonso A, Malcolm SB, Williams EH and Zalucki MP (2019) Lincoln Brower, Champion for Monarchs. Front. Ecol. Evol. 7:149. doi: 10.3389/fevo.2019.00149*

### MONARCHS AND THE NEW FIELD OF CHEMICAL ECOLOGY

Lincoln was a founder of the International Society of Chemical Ecology (ISCE), which promotes the understanding of interactions between organisms and their environment that are mediated by naturally-occurring chemicals (International Society of Chemical Ecology, 2019). He was also the first ISCE president, primarily due to his ground-breaking research on the chemical ecology of tritrophic interactions among milkweeds, monarch butterflies, and bird predators, illustrated eloquently by his 1969 Scientific American article on ecological chemistry (Brower, 1969).

Like many creators of iconic scientific paradigm shifts, Lincoln was a product of his moment in time. In 1914, Oxford-based evolutionary biologist Edward Poulton presciently called for North American chemists to test the hypothesis of Haase (1896) that larvae of aposematic butterfly models in Batesian mimicry systems are toxic because they sequester chemical defenses from toxic hostplants. Poulton (1914) targeted two North American butterfly-plant systems as being particularly likely to yield results, Danaus plexippus feeding on Asclepiads and Battus philenor feeding on Aristolochias.

It took another 44 years before Lincoln's first wife, Jane Van Zandt Brower, published her research on bird predation of monarchs and mimetic viceroy butterflies (Brower, 1958). This coincided with a Fulbright-funded year at Oxford University for Lincoln and Jane after their doctorates at Yale. At Oxford, the Browers worked in the ecological genetics laboratory of E.B. Ford who was developing Poulton's legacy, and also collaborated with Miriam Rothschild, the Nobel prize-winner Tadeusz Reichstein, and graduate student John Parsons to study the chemical ecology of monarch butterflies and cardenolides, a group of toxic steroids found in their milkweed hostplants.

Building on the metaphor of the "ecological theater and the evolutionary play" coined by his Yale mentor G. Evelyn Hutchinson, Lincoln published a paper on the response of bird predators to monarchs that were reared on cardenolide-rich Asclepias curassavica, or controls that were reared either on a cardenolide-free milkweed vine, Gonolobus rostratus, or cabbage leaves (with much difficulty!). With the use of the famous "barfing blue jay assay," Lincoln and his colleagues found that monarchs reared on A. curassavica prompted blue jays to vomit, but monarchs reared on the control host plants did not induce an emetic response (Brower et al., 1967). These differences prompted them to develop the concept of "automimicry"; monarchs that fed as larvae on milkweeds with little cardenolide are palatable automimics of model monarchs that fed as larvae on milkweeds rich in cardenolides.

Although Lincoln was mostly interested in studying how different milkweeds influenced monarch defenses against bird predators, he did follow Poulton's advice of half a century earlier and developed collaborations with North American chemists to develop ecologically meaningful cardenolide measures in milkweeds and monarchs. Thus began seminal work showing how different milkweed host plants influenced the chemical defenses of monarchs (Brower et al., 1968, 1972; Brower, 1969; Brower and Glazier, 1975). Given the important contributions of both Lincoln and Jane Van Zandt Brower to our understanding of mimicry, Pasteur (1982) suggested that automimicry be named "Browerian" mimicry, a fitting tribute to insightful research that spans ecology, evolution, behavior, physiology, and chemistry!

Lincoln's work on cardenolides also spawned research on the role that milkweeds and cardenolides play in the annual cycle of monarchs, and also helped us understand how the annual cycle operates. His collaborations with chemical ecologists generated a series of papers in which the cardenolide "fingerprints" of monarchs reared on seven Asclepias species were described from thin layer chromatography separations and spectrophotometric quantifications (Nelson et al., 1981; Brower et al., 1982, 1984a,b; Seiber et al., 1986; Lynch and Martin, 1987; Martin and Lynch, 1988; Malcolm et al., 1989; Martin et al., 1992). These cardenolide fingerprints allowed Lincoln and his colleagues to describe variation in cardenolide sequestration through the annual cycle and to show that monarchs migrate each spring through successive broods to colonize their summer breeding habitat (Malcolm and Brower, 1989; Malcolm et al., 1993; Malcolm, 1995). The work on breeding also included field experiments that showed adverse effects of host plant characters on early monarch survival and oviposition behavior (Zalucki et al., 1990, 2001a,b; Zalucki and Brower, 1992; Zalucki and Malcolm, 1999). In sum, Lincoln's work with his students and colleagues provided a robust foundation for a wide range of research on sequestration, chemical defense, host plant use, migration, and life history variation, building a foundation for the interpretation of human impacts on monarch butterflies (Malcolm, 2018).

#### FOCUS ON MEXICO

The publication of Urquhart's National Geographic article (Urquhart, 1976) on the discovery of the overwintering sites of monarchs in Mexico was transformative for Lincoln. Because Urquhart did not share the location of the sites, Lincoln and Bill Calvert took up the challenge to find them (described in Brower, 1995). They focused their efforts on the known locations of the endemic oyamel fir trees (Abies religiosa) shown in the article, and Calvert quickly found dense aggregations of the butterflies on a mountain in Michoacán. Soon thereafter, in January 1977, Lincoln first viewed the awe-inspiring spectacle of millions of monarchs clustered on high elevation oyamels. He spent much of the next winter studying birds preying upon monarchs roosting on these Mexican mountains and the importance of food plant choice for chemical defenses against predators (Calvert et al., 1979; Fink and Brower, 1981). These experiences changed his career, generating an emphasis that shifted toward monarch conservation (Brower, 1995).

For the next 40 years, Lincoln worked in the Mexican overwintering sites, studying monarchs and their interactions with the forests that are their winter home. Those of us lucky enough to spend time with him there learned lessons about monarch biology and witnessed his incredible passion for these insects and their habitat. Collectively, the research conducted by Lincoln and dozens of colleagues resulted in major contributions to our understanding of the biology and conservation of monarchs in Mexico, including predation, microclimatic influences on survival, impacts of winter storms, monarch clustering behavior, and forest dynamics. He led research to demonstrate the dependence of monarchs on the forest microclimate during the overwintering season (summarized in Williams and Brower, 2015). The canopy protects clustered monarchs from night-time freezes and protects them from wetting, which lowers their resistance to low temperatures. But logging creates holes in the protective canopy and exposes overwintering monarchs to an increased likelihood of death from freezing.

Lincoln worked hard to convince national and local authorities of the need to stop logging at the overwintering sites. He described monarch migration as an endangered biological phenomenon (Brower and Pyle, 1980; Brower and Malcolm, 1991) and was instrumental in the creation of protected areas for overwintering monarchs. A 1980 presidential decree recognized the importance of the monarch overwintering phenomenon, but no area was delineated for protection. In 1986, a second presidential decree established the Special Monarch Butterfly Biosphere Reserve, protecting 16,000 ha of land in five separate locations (Brower, 1995). In 2000, the current Monarch Butterfly Biosphere Reserve was created to protect 56,000 ha, and Lincoln played a key role in helping to delineate the protected area based on monarch biology (summarized in Missrie, 2004).

#### CONSUMMATE COLLABORATOR AND MENTOR

As an engaged scholar, Lincoln joined and generated many discussions about science and conservation, welcomed different views from colleagues, served leadership roles in a number of professional organizations, and mentored dozens of students and younger scientists. He was generous in spirit, and throughout his career, openly and enthusiastically encouraged others to join him in research. It is no surprise, consequently, that the 167 peer-reviewed articles listed on his CV show a wide web of collaboration; they were written, entirely by coincidence, with 167 different coauthors (**Figure 1**). Lincoln's legacy is wide and enduring, not just through his published research but also

through the many people that he brought to the study and appreciation of monarchs and, more generally, all of nature.

Lincoln's natural interest and curiosity inspired students who were lucky enough to work with him, particularly during field expeditions; his influence on one of us (AA), illustrates the pivotal role that he played in many scientific careers. In the fall of 1985, Lincoln asked Dr. Jorge Soberon, a Mexican professor at the Universidad Nacional Autonoma de Mexico, to recommend two undergraduate students of biology to accompany his graduate students at the University of Florida for a research expedition he was organizing. Alfonso Alonso and Alfredo Arellano took on the opportunity, but not before debating the wisdom of skipping class for a semester to camp in the mountains of Mexico for 3 months. Their decision changed their lives forever. Lincoln came to the camp in mid-February 1986, and spent a week walking with the students all over one of the most important overwintering sites in the Sierra Chincua, in the State of Michoacán. They visited Lincoln's favorite spots and talked extensively about monarch biology and his ideas on how to conserve the forest in Mexico.

The following year, Lincoln invited Alfonso to visit his lab at the University of Florida for 3 months. Alfonso practiced English and applied to the graduate program in Zoology. Lincoln found funds to support his graduate work, and Alfonso began his PhD work in fall 1988. Upon his arrival in Gainesville, Lincoln literally took Alfonso under his wing. He taught him academic skills like scientific writing and public speaking, as well as personal traits, like generosity.

All of us feel lucky to have been mentored by Lincoln, as graduate students or post-docs, and to have watched his interest in the next generations of scientists continue as he interacted with our own students. One of us (KO) recalls a pair of her favorite memories of Lincoln, watching him interact with Middle School students she had invited to meetings in San Luis Obispo, California and Minneapolis, Minnesota. In both cases, he treated these students exactly as he would have treated a senior and esteemed colleague, and they came away from these conversations feeling that they had discovered some of the most interesting things in the world.

#### A CHAMPION FOR NATURE

As Lincoln built his lasting contribution to chemical ecology and to science in general, monarchs began to show signs of an uneasy relationship with humans. Once researchers started to monitor these highly mobile insects with an almost continentwide distribution, it was clear to Lincoln and many in the monarch scientific community that the numbers of monarchs reaching Mexico were declining with time and the development of profoundly different agricultural technologies across the North American landscape (Malcolm, 2018). Dramatically reduced numbers of overwintering monarchs prompted a shift in Lincoln's energies from the chemical ecology of monarchmilkweed interactions to the conservation of monarchs that navigate across landscapes dominated by human agriculture. During much of the last three decades of his career, Lincoln focused on conservation science and action (e.g., Pyle, 2019). His scientific work included both field and laboratory studies of monarch habitat requirements, especially during the winter

#### REFERENCES


(Williams and Brower, 2015); he worked with Mexican government and NGOs to design the current extent of the Monarch Butterfly Biosphere Reserve (Missrie, 2004); was a founding officer and board member of the Monarch Butterfly Fund until a few months before his death (Monarch Butterfly Fund, 2018); a board member of the citizen science project Journey North (Journey North, 2018); and a signatory of the petition to the USFWS as a threatened species (Center for Biological Diversity, 2014). He received accolades from the Mexican government, including the prestigious Reconocimiento a la Conservacion do la Naturaleza from the Mexican Federal Government.

Traveling with Lincoln in Mexico or through the central plains of the US was always filled with conservation lessons; he pointed out the importance of microhabitats caused by clouds over the mountains, endemic or rare flora and fauna, and amazing interactions between species and their living and non-living environment. And while he lamented the impacts of humans on these things, he was never too busy to take time to talk with kindness and respect to people from all backgrounds and knowledge levels; this compassion went a long way toward selling his conservation message. This ability to communicate so effectively and eloquently through has publications, films, and presentations is the basis of Lincoln Brower's lasting scientific legacy; we all win from his 86 years of passion dedicated to monarchs and nature.

#### AUTHOR CONTRIBUTIONS

KSO led the manuscript development process. KSO, AA, SBM, EHW, and MPZ contributed equally to the content. All authors benefited from their associations with Lincoln P. Brower throughout most of their careers.


**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 Oberhauser, Alonso, Malcolm, Williams and Zalucki. 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.

# Modeling Current and Future Potential Distributions of Milkweeds and the Monarch Butterfly in Idaho

Leona K. Svancara<sup>1</sup> \*, John T. Abatzoglou<sup>2</sup> and Beth Waterbury <sup>3</sup>

*1 Idaho Department of Fish and Game, Moscow, ID, United States, <sup>2</sup> Department of Geography, University of Idaho, Moscow, ID, United States, <sup>3</sup> Idaho Department of Fish and Game, Salmon, ID, United States*

Monarch butterflies (*Danaus plexippus*) are widespread in North America but have experienced large rangewide declines. Causes of recent declines likely involve multiple biotic and abiotic stressors including climate change and loss and degradation of native milkweed (*Asclepias* spp.), monarchs' obligate larval host plant. Recent broad-scale modeling efforts suggest milkweed and monarch distributions in the eastern United States will expand northward during summer months while fine-scale modeling of western population overwintering sites in California indicate shifts inland and upward in elevation. However, species' response to climate measures varies at sub-regional scales across its range and both the impacts of climate change and potential adaptation measures may be sensitive to the spatial scale of climate data used, particularly in areas of complex topography. Here, we develop fine-scale models of monarch breeding habitat and milkweed distributions in Idaho, an area at the northern extent of the monarch breeding range in North America and important in western overwintering population recruitment. Our models accurately predict current distributions for showy milkweed (*A. speciosa*), swamp milkweed (*A. incarnata*), and monarch with AUC (area under the receiver operating characteristic curve) = 0.899, 0.981, and 0.929, respectively. Topographic, geographic, edaphic, and climatic factors all play important roles in determining milkweed and, thus, monarch distributions. In particular, our results suggest that at sub-regional and fine-scales, non-climatic factors such as soil depth, distance to water, and elevation contribute significantly. We further assess changes in potential habitat across Idaho under mid-21st century climate change scenarios and potential management implications of these changing distributions. Models project slight decreases (−1,318 km<sup>2</sup> ) in potential suitable habitat for showy milkweed and significant increases (+5,830 km<sup>2</sup> ) for swamp milkweed. Projected amounts of suitable habitat for monarch are likely to remain roughly stable with expansion nearly equal to contraction under a moderate scenario and slightly greater when under the more severe scenario. Protected areas encompass 8% of current suitable habitat for showy milkweed, 11% for swamp milkweed, and 9% for monarch. Our study shows that suitable habitat for monarchs and/or milkweeds will likely continue to be found in managed areas traditionally seen as priority habitats in Idaho through mid-century.

Keywords: monarch, showy milkweed, swamp milkweed, climate change, species distribution model, Idaho, Asclepias, Danaus plexippus

#### Edited by:

*Wayne E. Thogmartin, United States Geological Survey, United States*

#### Reviewed by:

*Nathan Lemoine, Colorado State University, United States Tyler Flockhart, University of Maryland Center for Environmental Science (UMCES), United States*

\*Correspondence: *Leona K. Svancara leona.svancara@idfg.idaho.gov*

#### Specialty section:

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

Received: *29 December 2018* Accepted: *25 April 2019* Published: *15 May 2019*

#### Citation:

*Svancara LK, Abatzoglou JT and Waterbury B (2019) Modeling Current and Future Potential Distributions of Milkweeds and the Monarch Butterfly in Idaho. Front. Ecol. Evol. 7:168. doi: 10.3389/fevo.2019.00168*

# INTRODUCTION

Monarch butterflies (Danaus plexippus plexippus) are widespread in North America but appear to be experiencing large rangewide declines in both the eastern (Semmens et al., 2016; Agrawal and Inamine, 2018) and western populations (Jepsen et al., 2015; Schultz et al., 2017). Factors contributing to declines are increasingly understood and likely involve multiple biotic and abiotic stressors including climate change and loss and degradation of native milkweed (Asclepias spp.), monarchs' obligate larval host plant (Flockhart et al., 2015; Jepsen et al., 2015; Inamine et al., 2016; Thogmartin et al., 2017; Belsky and Joshi, 2018). Changes in climate patterns, in particular, may both favor and hinder species through complex, seasonal relationships. Monarchs, for example, appear to benefit from warm winters and warm, wet springs (Zipkin et al., 2012; Espeset et al., 2016; Thogmartin et al., 2017), while excessively hot, dry, cold, or wet conditions may be a detriment (Zalucki, 1982; York and Oberhauser, 2002; Zalucki and Rochester, 2004; Nail et al., 2015; James, 2016), but these relationships are region-specific (Zipkin et al., 2012; Flockhart et al., 2017). Similarly, warmer temperatures result in increased growth for some milkweed species while drought reduces their growth, survivorship, seed production and germination, and nutritional quality (Bowles et al., 1998; Woods et al., 2012; Couture et al., 2015). These patterns have led to suggestions that northern populations of both milkweed and monarchs may benefit from projected changes in climate (Couture et al., 2015) depending on if and how milkweed distributions change (Lemoine, 2015).

Species distribution models (SDMs) are increasingly used to assess the current "climate envelope" for species and project the potential effects of changes in those climate variables under different future scenarios (Porfirio et al., 2014). To date, broad-scale modeling efforts suggest milkweed and monarch distributions in the eastern United States (US) will expand northward during summer months (Batalden et al., 2007; Lemoine, 2015) while fine-scale modeling of western population overwintering sites in California indicate shifts upward in elevation and inland (Fisher et al., 2018). Such modeling efforts, however, are often limited due to a mismatch between the spatial resolution of climate data available (e.g., 1–30 km) and the scale relevant to the species of interest (e.g., 10–100 m), especially in areas of complex terrain (Randin et al., 2009; Austin and Van Niel, 2011; Suggitt et al., 2011; Franklin et al., 2013). In addition, several modeling efforts have shown that species' response to climate often varies across its range (e.g., Pearman et al., 2010; Hällfors et al., 2016; Ikeda et al., 2017; Nice et al., 2019). Given that monarchs are a widespread, highly migratory species, one may not expect significant variability in climate response across the species range, while milkweed may exhibit more local adaptability.

Currently at the northern extent of monarch breeding range, Idaho significantly contributes to recruitment of monarchs to the western population (Yang et al., 2016). Chosen as the Idaho state insect, yet classified as unprotected wildlife in the state, the monarch butterfly was recently designated as an Idaho Species of Greatest Conservation Need predominantly due to significant rangewide declines and a lack of information on Idaho-specific status and trends [Idaho Department of Fish and Game (IDFG), 2017]. In particular, knowledge of monarch breeding locations, distribution, and movement patterns as well as potential impacts of climate change and other threats are limited in the region. Initial mapping of western milkweed and monarch observations (Xerces Society, 2014) and modeling work (Stevens and Frey, 2010; USFWS/Xerces Society, 2016) identified probable breeding habitat across seven western states, including Idaho. These and other studies suggested that monarchs in the intermountain states (Idaho, Utah, Nevada) are comparatively sparsely populated and may be constrained by low milkweed species diversity, semi-arid climates, and unsuitable temperature regimes associated with elevation or latitude (Pyle, 1999; Stevens and Frey, 2010). However, surveys by the Idaho Department of Fish and Game (IDFG) (Waterbury and Ruth, 2015) identified monarch breeding activity in a climatic region previously considered to be thermally unsuitable for monarch reproduction (Stevens and Frey, 2010). This finding suggests that other regions of Idaho, also previously deemed thermally constrained for monarch breeding, may currently support suitable natal habitat. Beginning in 2015, IDFG engaged in several efforts aimed at improving the knowledge base for monarchs including collaborating with Washington Department of Fish and Wildlife on a project to investigate monarch distributions and status in Idaho and Washington (Waterbury and Potter, 2018). Here, we employ data collected through these efforts to (1) develop fine-scale models of current milkweed and monarch butterfly distributions in Idaho using the most scale-relevant data available, (2) assess potential changes under mid-century climate change scenarios, and (3) assess potential management implications of these changing distributions.

# MATERIALS AND METHODS

#### Milkweed and Monarch Observations

All known observations of monarch and milkweed in Idaho as of April 26, 2018, were compiled for this modeling effort. This included data from the Western Monarch and Milkweed Mapper Occurrence Database (museum and herbarium specimens, older survey efforts, and incidental observations) (Xerces Society, 2018), recently collected IDFG and College of Western Idaho survey data (summers 2016, 2017), and incidental observations recorded in iNaturalist and in the Idaho Fish and Wildlife Information System Species Diversity Database [Idaho Department of Fish and Game (IDFG), 2018]. We carefully evaluated all data for use in the distribution models to ensure spatial and temporal accuracy. As part of this evaluation, we determined sufficient data exist in Idaho to model distributions for monarch (n = 1603) and two native milkweed species, showy milkweed (A. speciosa) (n = 5258) and swamp milkweed (A. incarnata) (n = 313). Other milkweed species documented but not modeled included narrow-leaf milkweed (A. fascicularis) (n = 94), pallid milkweed (A. cryptoceras) (n = 84), and spider milkweed (A. asperula) (n = 7). All compiled data are available online [Idaho Department of Fish and Game (IDFG), 2018 data used for modeling can be requested from the lead author].

Compiled observation data such as these are prone to errors of observational certainty, spatial accuracy, and sampling bias both geographically (e.g., more samples in easily accessible areas) and environmentally (e.g., more samples at lower elevations). To maximize observational certainty, we used only observations classified as verified (e.g., specimen, DNA, or photograph) or trusted (e.g., documented by a biologist, researcher, or taxonomic expert). To maximize spatial accuracy, we limited our data set to those locations with ≤100 m accuracy. Even though the vast majority of our observations resulted from targeted inventory or survey efforts (93% of showy milkweed, 91% of swamp milkweed, and 88% of monarch observations), sampling bias was still an issue as data were highly clustered at fine scales in portions of the state. Species distribution models can be sensitive to such bias and several authors have suggested spatial filtering or subsampling of the presence data to account for it (Phillips et al., 2009; Veloz, 2009; Anderson and Raza, 2010; Kramer-Schadt et al., 2013; Radosavljevic and Anderson, 2013). The key to spatial filtering isto randomly subsample presence data with a minimum distance separating the sample points, thereby limiting spatial autocorrelation and reducing the environmental bias caused by uneven sampling. That minimum distance is somewhat arbitrary, however, and depends on the environmental conditions of the study area as well as the resolution of the data used for modeling. We reduced the locally dense sampling of monarch and milkweed by randomly subsampling the observations with a minimum distance of 270 m, which accounted for the coarsest spatial data resolution of the environmental variables and the sampling design of the majority of field surveys. These filtering procedures (trusted or verified, ≤100 m accuracy, and >270 m separation) resulted in a total of 1079 showy milkweed observations, 100 swamp milkweed observations, and 344 monarch observations available for use in our modeling efforts (**Figure 1**).

#### Environmental Variables

Previous modeling efforts have focused on a number of climatic, topographic, and edaphic environmental covariates at broader spatial extents (western US, eastern US) and resolutions (90 m−10 km cell sizes) (Batalden et al., 2007; Lemoine, 2015; Dilts et al., 2018). Learning from and building on these efforts, we compiled and/or developed finer-scale versions of these covariates (**Table 1**), striving to use data at scales most applicable to milkweed and monarch butterflies and to ensure variables represented ecophysiological processes known to influence plant growth either directly or as surrogates (Austin and Van Niel, 2011). Conducting all spatial analyses in ArcGIS 10.5.1 (ESRI, 2017), we ensured spatial data were in a common geographic coordinate system, spatial resolution (30 m), and extent (Idaho), and exported as ASCII files for input into R (3.5.0; R Core Team, 2018) and Maxent (Maxent 3.4.1; Phillips et al., 2006, 2017; Phillips and Dudík, 2008).

Topographic variables generally act as surrogates for factors influencing plant growth, but can also directly account for differences in local climate and be important in SDMs (Luoto and Heikkinen, 2008; Austin and Van Niel, 2011). We developed several topographic variables including elevation, slope, aspect, compound topographic index (CTI), roughness, and vector ruggedness measure (VRM) from the National Elevation Data (30 m) (US Geological Survey, 2016b). The CTI, a steadystate wetness index, measures the catenary topographic position represented by both slope and catchment size and aims to model soil water content (Moore et al., 1993). Roughness, similar to the terrain ruggedness index (Riley et al., 1999), calculates the amount of elevation difference between a grid cell and its neighbors, essentially the variance of elevation within the neighborhood (8 × 8 cells in this analysis). The VRM, which measures terrain heterogeneity within a neighborhood (9 × 9 cells in this analysis), captures variability in both slope and aspect into a single measure (Sappington et al., 2007). We calculated CTI and roughness using Evans et al. (2014) and VRM using Sappington (2012), both freely available ArcGIS tools. All of these topographic variables, to varying degrees, were selected to reflect temperature, water and light resources that help determine plant distributions and may contribute to monarch habitats. For example, CTI and roughness may serve as proxies for local temperature patterns (e.g., cold air drainage, Dobrowski et al., 2009) while VRM, slope, and aspect act as surrogates for light or solar radiation.

Edaphic measures developed were characteristics known to either affect the availability of nutrients or exert direct physiological limitations, or both, on plants and included percent sand, percent silt, percent clay, pH, available water supply, calcium carbonate, cation-exchange capacity, organic matter, and depth to a restrictive layer. To focus on the most critical soil for plant establishment, we used a weighted average based on percent composition for aggregating across all soil map units in the top 0–25 cm for all variables except soil depth. These data were developed primarily from the Soil Survey Geographic database (SSURGO, USDA Natural Resource Conservation Service, 2016a), with missing areas filled in with the U.S. General Soils database (STATSGO2, USDA Natural Resource Conservation Service, 2016b), following the national standard methodology and tools used for similar products (e.g., gSSURGO) (USDA Natural Resource Conservation Service, 2016c).

Climatic variables used in previous efforts relied on temperature and precipitation at moderate (∼1 km) spatial resolution (Hijmans et al., 2005; Daly et al., 2008; Wang et al., 2012). To better represent Idaho climate we used more recent temperature data developed at finer spatial resolution (250 m) for the Northern Rockies (Holden et al., 2015) in combination with precipitation data (originally 800 m, resampled to 250 m resolution using cubic convolution to match the temperature data) from the Parameterized Regression on Independent Slopes Model (PRISM, Version 14.1-20140502-1000) (PRISM Climate Group, 2012; Daly et al., 2015). Both of these datasets represent monthly 30 year normals covering the period 1981-2010, from which we calculated 19 bioclimatic variables following Nix (1986) and Hijmans et al. (2005). These bioclimatic variables have been used extensively in SDMs for decades (e.g., Elith et al., 2010, 2011; Anderson and Gonzalez, 2011; Stanton et al., 2011; Booth et al., 2014), as well as in previous monarch and milkweed modeling studies (e.g., Lemoine, 2015; Dilts et al., 2018), and

characterize climatic conditions best related to species physiology (O'Donnell and Ignizio, 2012; Booth et al., 2014).

To portray mid-century climate conditions we used projections from 20 Global Climate Models (GCM) participating in the Fifth Coupled Model Intercomparison Project that were statistically downscaled using the Multivariate Adaptive Constructed Analogs (MACA, Abatzoglou and Brown, 2012). Output from GCMs were downscaled using the historical training dataset of Abatzoglou (2013) to a 4-km spatial resolution. We calculated differences in monthly mean minimum temperature, maximum temperature, and accumulated precipitation between 1981-2010 and 2040-2069 for each of the 20 models given two emission scenarios, Representative Concentration Pathway [RCP] 4.5 and RCP 8.5. The latter, which we refer to herein as severe, assumes a business as usual emissions pathway whereas the former, referred to as moderate, assumes policies that lead to a plateau and eventual decline in emissions. While there is potential value in examining difference in the uncertainty in climate projections across models, we constrain our efforts on the 20-model mean change. Differences between ensemble mean future and baseline monthly climate data at a 4-km spatial resolution were interpolated to the 250-m resolution of the observed data and superposed to these 250-m gridded observed data to provide an estimate of the projected climate fields. We then recalculated the 19 bioclimatic variables using these projected values.

Other potentially informative landscape-related variables developed included distance to intermittent streams and distance to perennial streams and waterbodies based on the National Hydrography Data (US Geological Survey, 2017) (FCodes 46006 and 46003, respectively). We considered including land cover and percent natural land cover following Dilts et al. (2018), but instead chose to omit these variables because the spatial and thematic scale of the most current land cover data (US Geological Survey, 2016a) did not accurately reflect known milkweed occurrences. For example, >50% of both showy and swamp milkweed locations were mapped as developed, cultivated cropland, or open water, yet field surveys determined milkweed was rarely found in these types and instead preferred grasslandsherbaceous, emergent herbaceous wetlands, and deciduous TABLE 1 | Environmental variables used in monarch and milkweed distribution modeling in Idaho.


forests (Waterbury and Potter, 2018). In addition, even if found important, future projections of land cover were lacking. Finally, as monarch breeding habitat is generally found to be constrained by the occurrence of milkweed (Lemoine, 2015; Dilts et al., 2018), the outputs from the showy and swamp milkweed models were included with the previous covariates when modeling monarchs.

#### Current/Future Habitat Suitability

We used maximum entropy methods (Maxent 3.4.1; Phillips et al., 2006, 2017; Phillips and Dudík, 2008) to model current potential habitat suitability for monarch, showy milkweed, and swamp milkweed, as well as the potential habitat suitability for the mid-21st century. Given a set of environmental variables and presence-only species occurrences, Maxent identifies the correlations between each variable and the presence data, compares that with the range of environmental conditions available in the modeled region, and develops a continuous model of relative likelihood of suitable habitat across the study area based on environmental similarity to known occupied sites. We choose to use Maxent because it does not require absence data, it efficiently handles complex interactions, it has been shown to perform better than other approaches with similar data, and it effectively transfers model projections to future conditions (Phillips et al., 2006; Elith et al., 2010, 2011). We supplied Maxent with occurrence data as described above, as well as background points consisting of 10,000 randomly generated pseudo-absences across Idaho that were >270 m apart, >270 m from presence locations, and outside of waterbodies.

Following recommended approaches (Elith et al., 2010, 2011; Anderson and Gonzalez, 2011; Merow et al., 2013; Radosavljevic and Anderson, 2013; Yackulic et al., 2013; Porfirio et al., 2014; Wright et al., 2015; Searcy and Shaffer, 2016; Morales et al., 2017), we developed current distribution models for monarch and milkweed using species-specific model parameters, particularly with regard to collinearity, regularization multiplier, and feature types. In Maxent, the feature types represent mathematical transformations of the covariates to allow complex relationships while the regularization multiplier imposes penalties to the model to help prevent overfitting (Elith et al., 2010, 2011; Merow et al., 2013). In an iterative approach, we optimized each model for feature types (linear, quadratic, product, threshold, hinge, and interactions) and regularization multiplier (values tested included 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 6, 7, 8, 9, 10, 12.5, 15, 17.5, and 20) using the enmSdm package (Smith, 2017) in R 3.5.0 (R Core Team, 2018) and selected the best performing combination based on sample size-corrected Akaike Information Criteria (AICc) (Warren and Seifert, 2011; Wright et al., 2015). Beginning with a full model inclusive of all covariates (n = 37), we implemented 10-fold cross-validation and jackknifing to measure importance of each variable to the resulting model. Variables were then ranked based on their permutation importance (Searcy and Shaffer, 2016) and removed if <1% contribution. Correlated variables (Pearson correlation >|0.70|) were also removed keeping the variable with the higher permutation importance. This process of model optimization, development, and variable ranking and removal was repeated until all variables had a minimum contribution of ≥2%. The final model for each species represented the average of 10 crossvalidation replicates using the optimized parameters and most important variables in cloglog output format and was evaluated using AUC (area under the receiver operating characteristic curve). We then projected the final models for each species onto spatial data projected for the two mid-century climate scenarios.

We imported all mean model outputs into ArcGIS 10.5.1 (ESRI, 2017) and identified areas of suitable and unsuitable habitat based on the "balance training omission, predicted area and threshold value" threshold calculated by Maxent. This threshold uses weighting constants to provide a balance between over-fitting and over-estimating. For comparative purposes, we further binned the suitable habitat using an expert opinionbased threshold to identify marginal and optimal suitability for each species. The current and two future models for each species were then overlaid to calculate the proportion of suitable habitat projected to remain suitable in the future (persistence), the proportion of current suitable habitat projected to become unsuitable (contraction), and the proportion of future habitat projected in currently unsuitable areas (expansion). Lastly, we tabulated the areas of gain and loss for pertinent IDFG Wildlife Management Areas (WMA), US Fish and Wildlife Service (USFWS) National Wildlife Refuges (NWR), and the US Forest Service (USFS) Curlew National Grassland (NG) to identify potential areas of concern.

Uncertainty is inherent in species distribution modeling simply due to limited knowledge of species-habitat relationships and correlative nature of the modeling process. Modeling potential effects of climate change introduces additional sources of potential variability given the range of estimates provided by each GCM. To assess how variability in projected changes might influence both the calculated bioclimatic variables as well as the modeling of suitable species habitat, we addressed potential uncertainty in two ways. First, we calculated the monthly standard deviation of projected changes in each climate measure (minimum temperature, maximum temperature, precipitation) to map variability among the 20 GCMs. This provided a simple way of evaluating potential influences on the bioclimatic variables. Second, we evaluated results of a Multivariate Environmental Similarity Surface analysis (Elith et al., 2010, 2011) to identify areas where novel conditions (i.e., conditions outside of the training range of the covariates) exist in model predictions and which variables were most limiting. In all cases, we assumed the species relationships with mapped covariates and correlations among covariates would remain consistent through time.

# RESULTS

Maxent accurately predicted current distributions for showy milkweed, swamp milkweed, and monarch with AUC = 0.899, 0.981, and 0.929, respectively (**Figure 2**). The best-fit models based on AIC<sup>c</sup> for all three species employed linear, quadratic, and product features, with a regularization multiplier of 0.5 for showy milkweed and monarch and a regularization multiplier of 2.0 for swamp milkweed.

Suitable habitats for all three species were predominantly characterized as low in elevation (≤1,300 m) and near perennial water (≤1,000 m), particularly swamp milkweed. In addition, showy milkweed habitat occurred in areas of deeper soils (≥30 cm), lower precipitation (i.e., wettest month was relatively dry at ≤50 mm), and higher diurnal temperature range (≥14◦C) (**Figure 3**; **Table 2**). Jackknife tests for showy milkweed indicated that mean diurnal temperature range had the most useful information by itself and elevation appeared to have the most information that was not present in other variables. Similarly, suitable swamp milkweed habitat was also characterized by areas with moderate monthly and annual temperature variability (isothermality ≥33%, temperature seasonality ≥7.75◦C), moderate minimum winter temperatures (−1 to −9 ◦C), low winter precipitation (≤75 mm), and moderate annual precipitation variability (seasonality 26–52%). Jackknife tests for swamp milkweed indicated that precipitation of the coldest quarter had the most useful information by itself and distance to perennial streams appeared to have the most information that was not present in other variables. Monarch habitat was predominantly a function of modeled showy milkweed habitat (probability ≥0.53) but also depended on

shown in red with lower suitability in blue.

warmer average temperatures (≥10◦C) during the wettest part of the year. Jackknife tests for monarch indicated that the modeled prediction of showy milkweed had the most useful information by itself as well as appeared to have the most information that was not present in other variables.

Because selection of specific model thresholds is somewhat arbitrary and biologically meaningful thresholds can be difficult to determine, reporting a range of threshold values, or none at all, is often recommended (Liu et al., 2005; Merow et al., 2013). Although we considered several thresholds calculated by Maxent in the interpretation of habitat suitability (**Table 3**), the "Balance training omission, predicted area and threshold value area" threshold best reflected expert knowledge of known habitats across the state and captured 75% of monarch and showy milkweed occurrences and 88% of swamp milkweed occurrences. Thus, we defined suitable habitat for showy milkweed, swamp milkweed, and monarch using the threshold values of 0.4704, 0.1415, and 0.399, respectively. For all three species, we used an expert-opinion-based threshold of 0.7 to separate marginal and optimal habitat. Using these threshold values, models for present day conditions predicted 15,836 km<sup>2</sup> of Idaho (7.4%) as suitable showy milkweed habitat with 36.0% (5,701 km<sup>2</sup> ) of that as optimal, 3,125 km<sup>2</sup> (1.5%) of Idaho as suitable swamp milkweed habitat with 8.8% (275 km<sup>2</sup> ) as optimal, and 11,521 km<sup>2</sup> (5.4%) of Idaho as suitable monarch habitat with 44.6% (5,143 km<sup>2</sup> ) as optimal. Suitable habitats for all three species

Curves represent a model created using only that variable, and reflect the dependence of predicted suitability on the selected variable and on correlations between the selected variable and other variables. The mean response of 10 replicate runs is in red and the mean ± 1 standard deviation is in blue. Y-axis values are predicted probability of suitable conditions, as given by cloglog output format. See Table 1 for variable abbreviations.

occur primarily along the Snake River and tributaries in south Idaho, with smaller amounts of habitat in the Pend Oreille, Coeur d'Alene, and Clearwater River valleys of north Idaho and the Salmon River of east-central Idaho (**Figure 4**). Nearly all (>98%) predicted suitable habitat for monarchs and swamp milkweed is encompassed by that predicted as suitable habitat for showy milkweed. Similarly, predicted suitable monarch habitat includes 98% of swamp milkweed distribution, but only 72% of showy milkweed distribution.

Mid-century projections suggested relatively small (<8.5% of current habitat) changes in the cumulative area of suitable habitat statewide for showy milkweed and monarch under either the moderate or more severe emission scenario (**Figure 4**; **Table S1**). Models projected contraction of showy milkweed

TABLE 2 | Most important variables based on percent permutation importance for the final monarch and milkweed models.


\**Variable with highest gain when used in isolation (i.e., appears to have the most useful information by itself).*

\*\**Variable that decreases the gain the most when omitted (i.e., appear to have the most useful information not present in other variables).*

suitable habitat (1,259 and 1,349 km<sup>2</sup> for moderate and severe scenarios, respectively), with minor expansion (29 and 31 km<sup>2</sup> ) into previously unsuitable areas, predominantly due to increased precipitation in the wettest month. For monarch, projected expansion (543 and 859 km<sup>2</sup> ) was roughly similar to contraction (558 and 573 km<sup>2</sup> ) and largely due to changes in the showy milkweed model as well as increased temperature of the wettest quarter. Conversely, projections for swamp milkweed showed considerable expansion (3,444 and 5,830 km<sup>2</sup> ) with only minor contraction (1 km<sup>2</sup> under moderate and 0 km<sup>2</sup> under severe), a result of increased temperature seasonality, minimum temperature of the coldest month, and precipitation in the coldest quarter.

At a local scale, however, areas of expansion and contraction appear more substantial. This expansion and contraction is particularly apparent with projected increases of swamp milkweed into southeast Idaho, and contraction of both showy milkweed and monarch habitat in portions of north and southcentral Idaho (**Figure 5**). Projected changes for selected managed areas are also important (**Figure 6**; **Table S2**). For example, 14 WMAs, four NWRs, and the Curlew NG are all projected to gain suitable habitat for swamp milkweed, including >10 km<sup>2</sup> gains in Curlew NG, Minidoka NWR, and Mud Lake WMA. Conversely, while persistence of showy milkweed remains high, expansion of suitable habitat for the species is minor in extent (<0.2 km<sup>2</sup> ) and limited to only two WMAs (CJ Strike and Craig Mountain). In fact, 20 of the 30 managed areas assessed are projected to lose suitable showy milkweed habitat, although all projected losses are <4.5 km<sup>2</sup> . Monarch projections are more variable with 12 areas projected to experience habitat expansion (0.1– 15.2 km<sup>2</sup> ), 7 to experience habitat contraction (0.1–1.4 km<sup>2</sup> ), and two to experience both expansion and contraction. Overall, five managed areas in northern Idaho are projected to experience a cumulative loss of monarch and/or showy milkweed habitat by mid-century under a severe emission scenario (**Figure 6**).

Analyses of variability among GCMs and uncertainty in the SDMs suggest areas where these projections should be interpreted with caution. For showy milkweed, analyses indicated few areas with novel conditions under a severe emission scenario with only a few limiting variables in localized areas (**Figure 7**). For swamp milkweed, novel conditions are much more apparent,



*Threshold values used to differentiate suitable and unsuitable habitat are highlighted in bold.*

FIGURE 4 | Current and future predicted distributions of showy milkweed (A,D), swamp milkweed (B,E), and monarch (C,F) across Idaho. Future distributions represent 20 global climate model average ensemble projections under severe (RCP 8.5) climate change scenarios. Separation of non-suitable habitat with marginal habitat is based on the "Balance training omission, predicted area and threshold value area" threshold, and separation with optimal habitat is based on an expert-derived threshold.

particularly in the valley bottoms of northcentral Idaho and a few localized areas in south Idaho due, predominantly, to the influence of minimum temperature of the coldest month (particularly in north Idaho), temperature seasonality and precipitation seasonality. For monarchs, novel conditions are evident in the southwest and a few areas in northcentral, primarily limited by mean temperature of the wettest quarter. In addition, monthly standard deviation of the projected changes in each of the core climate variables (minimum temperature, maximum temperature, precipitation) indicated both seasonal and spatial variability among the 20 GCMs (**Figures S1**–**S3**). For both minimum and maximum temperature, variability among

the GCMs is greatest in early spring (Feb-April) in southcentral and southeast Idaho, regardless of emission scenario. For precipitation, variability among the models is greatest during the summer months (July-Sept) in southwest Idaho, which also has the lowest precipitation during this time. Given the seasonality of the variability among GCMs, the bioclimatic variables most likely affected include mean diurnal range, isothermality, temperature seasonality, and precipitation seasonality. Thus, SDMs using these variables may over- or under-predict possible changes in suitable habitats.

# DISCUSSION

Use of SDMs to depict current and potential future species distributions under changing climates is becoming commonplace, however local adaptation can result in significant variability in climate responses across a species range (Pearman et al., 2010; Hällfors et al., 2016; Ikeda et al., 2017; Nice et al., 2019). Our results indicated variability in milkweed response to projected climates in Idaho with the amount of showy milkweed suitable habitat decreased slightly statewide while swamp

milkweed potential habitat doubled under a moderate emission scenario and nearly tripled under a more severe emission scenario. Projected amounts of suitable habitat for monarch in Idaho are likely to remain roughly stable with expansion nearly equal to contraction under a moderate scenario and slightly greater when under the more severe scenario. As such, our assessment of potential changes in monarch and milkweed distributions at a finer scale and in the northern portion of the range represents important contributions to the long-term conservation and management of the species.

Overall, our models of current distributions of milkweed and monarch habitats are similar to other broad-scale modeling efforts (Lemoine, 2015; Dilts et al., 2018) in that monarch distribution was largely a function of milkweed occurrence with relatively minimal influence of climate variables, although spring temperatures were positively related in our case as in Espeset et al. (2016). While monarchs are not in Idaho during the wettest time of year (spring), the combination of warm/wet conditions likely facilitates milkweed habitat, and thus suitability of monarch habitat. Milkweed distributions in Idaho were influenced by variables emphasizing annual and seasonal variability in temperature and precipitation as opposed to the more broad-scale efforts (Lemoine, 2015; Espeset et al., 2016; Dilts et al., 2018) that highlighted mean and maximum variables (e.g., mean annual temperature). Perhaps more importantly, our results suggest that at a more local/state scale, non-climatic factors such as soil depth, distance to water, and elevation play important roles in determining milkweed and monarch distributions. Although these factors (soil, water, elevation) are likely not independent of climate, but rather interact with climate to create suitable conditions.

The primary resources of light, heat, water, and nutrients constrain species distributions at topo- and meso-scales (Mackey and Lindenmayer, 2001). As such, several studies have indicated inclusion of edaphic variables is important when modeling plant distributions (Bertrand et al., 2012; Beauregard and de Blois, 2014; Diekman et al., 2015; Velazco et al., 2017) and we found a positive relationship between soil depth and showy milkweed habitat. Surprisingly, soil variables were not significant in the model for swamp milkweed, a species adapted for moist soils (Woodson, 1954) and typically found in marsh and saturated meadow habitats in the state (Waterbury and Potter, 2018). Distance to perennial water, however, was significant for all three species distributions but particularly swamp milkweed. Dilts et al. (2018) also found swamp milkweed dependent on proximity to perennial water and it may be that this factor overrides other explanatory soil variables important for the species (e.g., hydric soils). This general relationship of milkweeds with water may also help explain the use of riparian corridors as typical migratory paths for western monarchs (Pyle, 1999; Dingle et al., 2005; Morris et al., 2015). Future modeling efforts that go beyond the typical bioclimatic variables might effectively couple soils with climate. For example, a simple water balance model that considers soil water holding capacity, precipitation, and potential evapotranspiration that captures the joint seasonality of energy and moisture (e.g., Abatzoglou et al., 2018) might more concisely synthesize climatic and edaphic factors important for habitat.

Our results support the inclusion of elevation as a necessary predictor for species distributions (Luoto and Heikkinen, 2008; Randin et al., 2009; Oke and Thompson, 2015), potentially due to reflecting environmental conditions not properly portrayed by climate data (Körner, 2007; Oke and Thompson, 2015), the bioclimatic factors used in this study, or other topographic variables (e.g., slope, aspect, CTI). Correlations of elevation with climate and biophysical variables may vary over space and time (Phillips et al., 2006) and we chose to follow a process of variable selection that allowed for expert input while maximizing variable importance in the SDMs and minimizing correlations. As such, we identified only five climate variables that were significantly correlated with, but of less importance in the SDMs, and thus excluded by elevation. Interestingly, these excluded variables reflected mean and maximum temperature characteristics identified as important in broad-scale models (Lemoine, 2015; Espeset et al., 2016; Dilts et al., 2018) but suggested as less relevant to butterfly distributions (Filz et al., 2013). The importance of elevation in our models suggests the included bioclimatic variables may still be inadequately capturing physical processes related to moisture and/or energetics of milkweed habitat either due to missing variables (e.g., downward shortwave radiation) or mismatches of scale. Ultimately, local fine-scale variability in topographic, climatic, and edaphic characteristics can result in extreme differences in growing conditions for plants (Austin and Van Niel, 2011; Lembrechts et al., 2018). In addition, the effects of climate change may be elevationally dependent (Nice et al., 2019), although that effect is not well-represented through our current approach using a statistically downscaled model ensemble. It remains unknown on how synoptic and land-surface factors will modify the occurrence and strength of cold air drainages and other elevation-dependent changes in climate in future climate (Daly et al., 2010; Pepin et al., 2015). Advances in understanding how climate change will manifest itself at these finer spatial scales may help improve estimates of changes in impacts.

To date, the rather substantial projected broad-scale northward expansions of showy milkweed and swamp milkweed as well as monarch breeding distributions (Batalden et al., 2007; Lemoine, 2015) are driven chiefly by increasing temperatures. Our results suggest that northward expansion might not be the case in Idaho. While we did project expansions of suitable swamp milkweed habitat, expansions were predominantly in the east and southeastern portions of the state. Conversely, showy milkweed expansions and contractions were minimal, but concentrated such that many of the managed areas in the northern portion of Idaho were projected to experience greater habitat losses than gains. Similarly, projected expansion of monarch distribution occurred mainly in the southeast with contractions in the north. These estimates reflect the spatial patterns of observed and projected increases in winter and spring temperature and precipitation across Idaho (Abatzoglou et al., 2014; Klos et al., 2015; Rupp and Abatzoglou, 2017). Our models indicated species expansion in areas with moderately increasing precipitation coupled with warming temperatures (i.e., southeast) but contraction in some areas, particularly in north Idaho, that may become too wet during spring to provide suitable habitat. Given future climate projections of progressively hotter, drier summers and warmer, wetter winters as well as the highly variable topography and amount of natural landscapes available in Idaho, one might hypothesize substantial expansion of showy milkweed suitable habitat due to upslope range shifts. While the iterative variable inclusion process works well modeling current distributions, it can be an important source of uncertainty for future predictions (Braunisch et al., 2013). Our models suggest current habitat suitability for all three species declines with increasing elevation. Given that elevation remains constant under future climate conditions, the future model projections are thus limited to lower elevations reducing the ability of higher elevation areas to potentially be modeled as expanded habitat under future scenarios. Development of a climate-only model, or inclusion of the elevation-correlated temperature variables mentioned above, may have resulted in just such changes in suitable habitat. However, projecting the effects of climate change on species in topographically diverse areas is complex. Climate-only models may not reflect realistic patterns as local topography can result in steep, fine-scale temperature gradients (e.g., insolation or cool-air pooling) (Daly et al., 2010; Holden et al., 2015) that can moderate the effects of rising temperatures by creating microclimates and refugia that enable species persistence (Luoto and Heikkinen, 2008; Austin and Van Niel, 2011; Suggitt et al., 2011; Lembrechts et al., 2018). In addition, variation in edaphic measures and habitat can also alter microclimates and refugia (Suggitt et al., 2011; Bertrand et al., 2012; Beauregard and de Blois, 2014). In fact, the presence of suitable microclimates due to substantial variability in these characteristics may be a key factor in the persistence of both milkweeds and monarch butterflies in Idaho.

In the absence of climate refugia, known physiological constraints of monarchs may further alter the extent of breeding distribution in Idaho beyond our model results. Monarchs typically arrive in Idaho in June, breed through August, then depart mid-August/mid-September (Waterbury and Potter 2018). Currently, maximum monthly temperatures during these months are generally within the optimal (27–29◦C) or sub-lethal (30–36◦C) range for monarch survival identified in laboratory studies (Zalucki, 1982; York and Oberhauser, 2002; Zalucki and Rochester, 2004; Nail et al., 2015) (**Figure 8**). Under a severe emission scenario, however, maximum temperatures are projected to increase substantially between June and September with areas in the optimal and sub-lethal range expanding greatly and a portion of current habitat in the southwest exceeding 36◦C (maximum 38.7◦C). While limited exposure to high temperatures (36–38◦C) is not detrimental to monarchs, extended exposure is lethal to immature stages (York and Oberhauser, 2002; Nail et al., 2015; James, 2016). Following a similar spatial pattern, minimum temperatures are also projected to increase such that the amount of area above the lower temperature threshold for larval development (≥11◦C, Zalucki, 1982) also expands substantially from June-September. Spatially extensive warming on the shoulder months of June and September may result in earlier and/or later breeding over broader areas of Idaho, potentially resulting in additional breeding generations (Batalden et al., 2007), but lethal temperature limits in the southwest may result in increased mortality in important managed areas such as Payette River WMA, Montour WMA, Fort Boise WMA, CJ Strike WMA, and Deer Flat NWR. In addition, increases in the frequency and duration of extreme temperatures may result in earlier senesce and/or reduced nutritional quality of important milkweed and other nectaring plants. Given climate variability, there will likely be years in which summer temperatures are above and below averages reported here. These model projections underscore the need to investigate western monarch population vital rates across all life stages (i.e., survival, individual growth, reproduction, recruitment) to better predict monarch response to changing climates and weather (Schultz et al., 2017; Belsky and Joshi, 2018).

#### Management Implications and Caveats

Mapped classes of suitable, marginal, and optimal habitats showcase how relatively small and fragmented suitable areas are for monarch and milkweed in Idaho, which appears to be the case in on-the-ground surveys (Waterbury and Potter, 2018). Even so, our study shows that suitable habitat for monarch and/or milkweeds will likely continue to be found in managed areas traditionally seen as priority habitats through mid-century. For example, Waterbury and Potter (2018) documented major monarch eclosure events at Bear Lake NWR, CJ Strike WMA, Sterling WMA, and Fort Boise WMA in recent years. Extensive areas of suitable habitats are projected to persist in all four of these managed areas, with sizeable habitat expansions (>4 km<sup>2</sup> ) in Bear Lake NWR. Additional managed areas with substantial expansions include Camas NWR, Minidoka NWR, Curlew NG, and Mud Lake WMA. Even though total amount of suitable habitat statewide is not projected to drastically diminish, geographic patterns in habitat contractions suggest that five areas in northern Idaho (Kootenai NWR, Boundary Creek WMA, CDA River WMA, Farragut WMA, and Pend Oreille WMA) may lose nearly all suitable habitat. Based on our results, protected areas currently encompass 8, 11, and 9% of suitable habitat for showy milkweed, swamp milkweed, and monarch, respectively, with 2.4% of showy milkweed habitat and 2.9% of swamp milkweed and monarch habitat protected under the IDFG, USFWS, and USFS managed areas we assessed. Under the most severe climate change scenario, this percentage increases for all three species to 2.5, 3.0, and 3.2%, respectively. These managed areas, however, represent a relatively small proportion of current and future suitable milkweed and monarch habitats in Idaho, indicating sizeable potential to engage private landowners and industry in expanding habitat protection, enhancement, and restoration to benefit monarchs and other pollinators in Idaho.

Currently, distribution of showy milkweed encompasses virtually all suitable swamp milkweed habitat. This relationship is seen in the field as well with swamp milkweed generally co-occurring with showy milkweed in Idaho. These areas of co-occurrence tend to have the highest monarch productivity due to extended phenology with showy milkweed early and swamp milkweed later (Waterbury and Potter, 2018). If swamp milkweed expands, but showy milkweed does not, these areas may experience reduced early-season monarch productivity but, conversely, increased later-season monarch productivity. Greater

availability of later-blooming swamp milkweed may also aid the migratory generation, both locally-produced and those stopping by from points further north. Although our results suggest broadscale overlap of milkweed and monarch occurrence, at finer scales, presence of milkweed is not synonymous with presence of monarchs. Breeding monarchs likely select for a range of variables beyond what we could model to maximize reproductive fitness. Such resource selection studies are lacking but urgently needed for the western monarch population. Further inventory is also needed to assess potential climate adaptability and habitat relationships of monarchs with other native milkweeds (narrowleaved, pallid, and spider) in Idaho. In addition, monitoring that incorporates areas of potential habitat expansion would ensure changes in distribution are documented and not just assumed a population loss due to declining numbers in specific areas.

Even though we employed best modeling practices recommended in the literature (Elith et al., 2010, 2011; Anderson and Gonzalez, 2011; Merow et al., 2013; Radosavljevic and Anderson, 2013; Yackulic et al., 2013; Porfirio et al., 2014; Wright et al., 2015; Searcy and Shaffer, 2016; Morales et al., 2017), our models and analyses are still subject to several caveats based on ecological and mathematical assumptions inherent in SDMs developed with Maxent and violation of these assumptions can affect model inferences to varying degrees (Phillips et al., 2006, 2017; Wiens et al., 2009; Merow et al., 2013; Yackulic et al., 2013). Key to Maxent, both sampling probability and detection probability are assumed constant across space. Although we attempted to account for multiple sources of uncertainty in our compiled occurrence data, including sampling bias, this key assumption may not be met. Similarly, detection probabilities for milkweed and monarch butterflies are unknown but, given both species are generally conspicuous on the landscape, we assumed detection probability was constant. Our choice of background extent and threshold values to determine suitable/non-suitable habitat likely also influenced model results, thus there may be more or less habitat available than we are suggesting. The true distribution of monarchs and milkweeds in Idaho is likely beyond the resolution our models provide, both spatially and thematically. We used the finest spatial data available (250 m climate, 30 m all others), yet even this resolution likely averaged over micro-environments important in the establishment of milkweed and/or the breeding of monarchs, as well as potential refugia.

Fundamental to SDMs, the species is assumed to be in equilibrium with its environment and its occurrence data representative of suitable habitat. In addition, SDMs are correlative in nature and, while selected model covariates are assumed to adequately reflect determinants of species distributions at relevant temporal and spatial scales, derived relationships may be due to other correlated variables not assessed. We modeled milkweed and monarch distributions based on the assumption that abiotic factors (light, heat, water, nutrients; Mackey and Lindenmayer, 2001) primarily controlled occurrence. However, other variables important in determining milkweed and monarch distributions may be lacking such as land cover or disturbance (Suggitt et al., 2011; Dilts et al., 2018) or light/solar radiation (Austin and Van Niel, 2011). Biotic interactions, genetic responses and geographic barriers limiting dispersal and/or colonization likely also dictate species distributions to some degree (Wiens et al., 2009). SDMs are also assumed to be spatially and temporally transferable. Projecting through time assumes the relationships among monarch, milkweeds, and their environs will remain consistent and that the correlation among this suite of variables will also remain consistent. In other words, we are assuming the species and their environment, as well as multiple components of that environment, will not become decoupled from each other (Walther, 2010). Furthermore, we still know little about monarch and milkweed direct relationships with climate and potential abiotic and biotic interactions (e.g., soils, parasites, monarch vital rates) at local scales, as well as species adaptive capacity (Dawson et al., 2011; Beever et al., 2016), both of which are likely to influence current and future distributions.

Lastly, uncertainty in the climate projections can be due to emission scenario, model structure, or natural climate variability (Hawkins and Sutton, 2011; Woldemeskel et al., 2016). We used a multi-model mean of 20 MACA downscaled GCMs because it performs well in areas of complex terrain (Abatzoglou and Brown, 2012). However, this does limit the SDMs to bulk-30 year averages and does not account for climate variability and extremes at finer temporal scales that may affect milkweeds and/or monarchs themselves, although assessments of SDMs and semi-process based models show similar performance (Parker and Abatzoglou, 2017). Both emission scenarios we assessed represent realistic scenarios of future conditions. Focusing on mid-century, as opposed to endof-century, projections limited the extent of uncertainty as well in that there is an average of only about 1◦C difference in temperature change between scenarios by the mid-century (Rupp and Abatzoglou, 2017). However, neither emission scenario takes into consideration potential cascading and/or interacting effects such as non-native plant invasions, changes in land use (e.g., extent of agricultural or urbanization), and changes in water usage. For instance, areas predicted to persist or expand under our two emission scenarios (i.e., Snake River Plain) are expected to favor cool-season species of exotic invasives such as cheatgrass (Bromus tectorum) (Smith et al., 2006; Bradley, 2009). Interactions between climate change and cheatgrass proliferation may combine to increase invasion risk to native rangeland and grassland ecosystems, thereby increasing fine fuels and risk of frequent and/or large-scale wildfires (Whisenant, 1990; D'Antonio and Vitousek, 1992). Likewise, changes in waterbody extent, distribution and water usage may play an important role in agriculture-intensive areas within an irrigation landscape (high water demand and diminishing aquifer), such as in areas of the Snake River Basin predicted as expansion for swamp milkweed. Because we limited our study extent to Idaho, future model projections may also not capture the full range of environments milkweed and monarch inhabit in other areas. In other words, areas and covariate values identified as "novel" in our analyses are only novel for Idaho and may not be in other areas.

Even with these caveats, our models accurately predict current and project future distributions of milkweed and monarchs and provide a means for assessing potential changes in habitat and distributions, identifying priority areas for conservation, and directing future research and monitoring efforts. While broad-scale modeling efforts provide a baseline understanding of species-habitat relationships and distribution, and can help guide finer-scale sampling efforts (such as our use of earlier versions of Dilts et al. (2018), or adoption of sampling schemes that collect presence/absence data), they fail to consider the potential for intraspecific variability in climate relationships and potential for local refugia. Understanding this variability is key in knowing just how vulnerable a species may be in a given area, as well as identifying the most appropriate adaptive management. Even state-level analyses, as ours, are likely to overlook more regional or local level variability in species responses (Pearman et al., 2010; Hällfors et al., 2016; Ikeda et al., 2017; Nice et al., 2019) and we recommend additional efforts to identify regions of smaller extents where climate responses of monarchs and milkweed are likely to be more homogeneous.

#### AUTHOR CONTRIBUTIONS

LS and BW contributed conception and design of the study and funding acquisition. JA developed climate scenarios. LS

#### REFERENCES


conducted analysis and wrote the first draft of the manuscript. BW and JA wrote sections of the manuscript. All authors contributed to manuscript revision and read and approved the submitted version.

## FUNDING

This research was funded by the USFWS Federal Aid in Wildlife Restoration Program (Grant Number F16AF01105, Amendment 1) with match from the IDFG Nongame Trust Fund.

#### ACKNOWLEDGMENTS

We thank the numerous field crews, citizen scientists, College of Western Idaho, and Xerces Society staff for data collection and compilation over the years. We also thank Adam Smith for modeling assistance and Tyler Flockhart, Nathan Lemoine and Wayne Thogmartin for comments on earlier drafts.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00168/full#supplementary-material


Iconic Butterfly, eds K. Oberhauser, S. Altizer, and K. Nail (Ithaca, NY: Cornell university Press), 99–108.


when can we trust the inferences? Methods Ecol. Evol. 4, 236–243. doi: 10.1111/2041-210x.12004


**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 Svancara, Abatzoglou and Waterbury. 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.

# Patterns of Monarch Site Occupancy and Dynamics in Iowa

Stephen J. Dinsmore<sup>1</sup> \*, Rachel A. Vanausdall <sup>1</sup> , Kevin T. Murphy <sup>2</sup> , Karen E. Kinkead<sup>3</sup> and Paul W. Frese<sup>3</sup>

*<sup>1</sup> Department of Natural Resource Ecology and Management, Iowa State University, Ames, IA, United States, <sup>2</sup> Center for Survey Statistics and Methodology, Iowa State University, Ames, IA, United States, <sup>3</sup> Wildlife Diversity Program, Iowa Department of Natural Resources, Boone, IA, United States*

#### Edited by:

*Wayne E. Thogmartin, United States Geological Survey, United States*

#### Reviewed by:

*Andrew William Byrne, Department of Agriculture, Food and the Marine, Ireland Richard Erickson, United States Geological Survey, United States*

> \*Correspondence: *Stephen J. Dinsmore cootjr@iastate.edu*

#### Specialty section:

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

> Received: *01 January 2019* Accepted: *26 April 2019* Published: *16 May 2019*

#### Citation:

*Dinsmore SJ, Vanausdall RA, Murphy KT, Kinkead KE and Frese PW (2019) Patterns of Monarch Site Occupancy and Dynamics in Iowa. Front. Ecol. Evol. 7:169. doi: 10.3389/fevo.2019.00169* The monarch butterfly (*Danaus plexippus*) is the focus of large-scale habitat restoration efforts because of recent population declines. From 2006-2017 we monitored monarchs at >400 sites throughout Iowa to link site occupancy and colonization/extinction dynamics to the presence of milkweed, site-specific habitat metrics, and landscape context at differing spatial scales. We used a robust design occupancy model in Program MARK and a hierarchical model selection approach to estimate site occupancy, extinction and colonization probabilities, and detection probability. Occupancy models revealed that monarchs responded differently to landscape features, environmental conditions, and local habitat conditions for site occupancy, extinction, and colonization probabilities. For site occupancy, the mean patch size of grassland at the 1-km spatial scale had a positive effect (β*GrassPS*1*<sup>K</sup>* = 0.94, SE = 0.54) while the percent of the landscape in woodland at the 200-m spatial scale had a negative effect (β*WoodPL*<sup>200</sup> = −1.68, SE = 0.34). For extinction, there were additive effects of the percent of the landscape in woodland at the 100-m spatial scale (β*WoodPLAND*<sup>100</sup> = 2.70, SE = 0.63), the interspersion of grassland at the 1-km spatial scale (β*GrassIJI*1*<sup>K</sup>* = −2.30, SE = 0.63), and litter depth (β*Litter* = 0.46, SE = 0.13). Finally, there were negative effects of the percent of the landscape in woodland at the 200-m spatial scale (β*WoodPLAND*<sup>200</sup> = −4.67, SE = 1.37) and the interspersion of grassland at the 100-m spatial scale (β*GrassPS*1*<sup>K</sup>* = −2.02, SE = 0.70) on colonization probability. Detection probability was affected by the additive effects of canopy cover and monarch density; no other detection model was competitive. In the top model there was a positive effect of monarch density (β*Density* = 0.28, SE = 0.05) and a negative effect of canopy cover (β*Canopy* = −0.18, SE = 0.03) on detection probability. In Iowa, monarchs are widespread on conservation lands where they avoid sites with lots of canopy cover. Colonization and extinction processes are influenced by an interplay of landscape attributes across multiple spatial scales and site habitat attributes. Our study provides the first comprehensive insight into monarch use of conservation lands in Iowa, and predicted responses to important covariates may be useful for future conservation efforts.

Keywords: butterfly, colonization, detection probability, extinction, habitat selection, Iowa, monarch, occupancy

# INTRODUCTION

The theory of Island Biogeography (MacArthur and Wilson, 1967) suggests that species diversity is a function of the processes of immigration and extinction, in addition to attributes of the habitat patch such as size (the species-area relationship) and distance from the nearest source population (the speciesdistance relationship). The concept of metapopulations (Levins, 1969) arose from this work to suggest how populations of the same species interact at some spatial scale. Through time, a metapopulation is generally thought to be stable while its distinct populations occupy suitable habitat patches and are subject to fluctuations resulting from the processes of extinction and colonization (Levins, 1969). Occupancy modeling estimates the probability a species is present in a habitat patch and is increasingly used as a long-term monitoring tool (MacKenzie et al., 2003; Manley et al., 2004; Bailey et al., 2007). Much work has focused on how to manage these populations for a species' long-term persistence. Some important considerations include the quality of habitat and degree of loss of habitat (Thomas, 1994), the size of the patch (Tscharntke et al., 2002), and the number and spatial arrangement of patches (Burkey, 1988; Hanski and Thomas, 1994; Hanski and Ovaskainen, 2000; Ouin et al., 2004). The spatial scale at which these effects operate is also an important consideration for butterflies (Loos et al., 2014; Olivier et al., 2016) and other pollinators (Murray et al., 2012), although few studies have addressed this topic. Understanding these processes is vital to the conservation of a species.

Efforts to manage the size of a habitat patch or the spatial arrangement of habitat patches are often not possible for economic reasons or because of land-use restrictions (Thomas, 1999; Zingg et al., 2018). Improving habitat quality in a patch is often the most feasible conservation measure available, and this applies to butterflies and also other taxa. Therefore, understanding species-habitat relationships is critical for management and conservation efforts of butterflies (Arnold, 1983; Dover and Settele, 2009). Grassland butterfly species respond to a host of local and landscape-level factors when selecting habitats for feeding and reproduction. In fragmented, agricultural landscapes such as the Midwestern U.S., butterflies must seek out patches of suitable grassland habitat from large areas of unsuitable habitat. This landscape can include native remnant grasslands (Davis et al., 2007), road right-of-ways (Ries et al., 2001), and other types of habitat patches. Attributes of these grasslands that are important include patch size (Tscharntke et al., 2002; Davis et al., 2007), permeability of the edge habitat (Ries and Debinski, 2001; Luppi et al., 2018), and measures of nectar sources and host plants (Clausen et al., 2001; Schneider and Fry, 2001; Pywell et al., 2004; Curtis et al., 2015; Luppi et al., 2018). It is also important to consider a multi-scale approach to habitat selection (McGarigal et al., 2016), although there are still relatively few examples for butterfly communities (Bergman et al., 2004; Olivier et al., 2016). Despite a widespread belief that selection factors operate at multiple spatial scales, many studies continue to consider just a single spatial scale (Bergerot et al., 2011; Öckinger et al., 2012) or use categorical local and landscape scales (Loos et al., 2014; Luppi et al., 2018). In a metapopulation context, management or conservation actions on even a subset of habitat patches may be important for a species' persistence.

The monarch butterfly (Danaus plexippus) has been the focus of intensive conservation efforts because of an estimated 80% decline in the eastern population in the last 20 years (Semmens et al., 2016; Oberhauser et al., 2017). Threats and causes for this population decline are many and include the loss and degradation of breeding habitat and resulting impacts on fecundity (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Pleasants, 2015), climate change (Zipkin et al., 2012; Zalucki et al., 2015), and threats on their wintering grounds (Oberhauser and Peterson, 2003; Brower et al., 2012). Much of the recent recovery focus has been directed at habitat restoration efforts on the breeding grounds (Thogmartin et al., 2017), especially by planting milkweed (primarily Asclepias sp.), which is the obligate host plant along with other nectar sources. Recent work suggests that most milkweed currently occurs on publicly owned grasslands, road right-of-ways, and in areas enrolled in conservation programs within the primary U.S. breeding range (Pleasants and Oberhauser, 2013; Pleasants, 2017; Thogmartin et al., 2017); the remaining milkweed occurs in marginal habitat such as agricultural fields. Adding 1.3–1.7 billion milkweed stems in the Midwestern region of the U.S. has been identified as a conservation priority to reduce quasi-extinction probability of the monarch (Pleasants et al., 2017; Thogmartin et al., 2017). This recommendation is informed by an extensive knowledge of monarch natural history (Zalucki and Kitching, 1982; Bull et al., 1985; Oberhauser, 1997; Flockhart et al., 2012; Zhan et al., 2014; Jepsen et al., 2015; Pitman et al., 2018) and emerging information about their movement patterns during the breeding season (Zalucki et al., 2016; Grant et al., 2018). Several authors have developed demographic models to predict future changes in the monarch population, ranging from fulllife-cycle models (Flockhart et al., 2015; Oberhauser et al., 2017) to those restricted to just the breeding season (Yakubu et al., 2004; Flockhart et al., 2013). An additional complexity is the annual cycle of the monarch, which includes multiple generations each year that are temporally and spatially explicit but with some overlap (Nail et al., 2015). But what is needed is a greater understanding of metapopulation dynamics—what proportion of habitat patches in a region are occupied during the breeding season, and at what rates do monarchs use ("colonize") and not use ("disappear from") those habitat patches between years?

In this paper we describe the metapopulation dynamics of monarch butterflies at >400 public and private properties in Iowa, USA. We investigate how local- and landscape-scale factors affect site occupancy by monarchs, and how those same factors influence the probability of colonization and extinction across our sample of sites. Collectively, this information provides greater insight into how monarchs exploit habitat patches in a region of intensive agriculture, and what attributes of those patches might benefit from future conservation efforts.

# MATERIALS AND METHODS

# Study Area

We studied monarchs at public and private properties located throughout Iowa (**Figure 1**) as part of the Iowa Multiple Species Inventory and Monitoring (MSIM) program (Iowa Department of Natural Resources, 2016). More than 60% of Iowa is comprised of agricultural cropland, with about 13% in grassland, 7% in woodland, and 0.7% in wetland (Iowa Department of Natural Resources, 2015). The state is comprised of seven landform regions (Northwest Iowa Plains, Missouri Alluvial Plain, Western Loess Hills, Des Moines Lobe, Southern Iowa Drift Plain, Iowan Surface, Paleozoic Plateau, and Mississippi Alluvial Plain; Prior, 1991). Most of the MSIM study sites are located on public land, which encompasses only about 2% of the state. Study sites were selected using a stratified random design (Kinkead, 2006), and this process is described in further detail in Harms et al. (2014). Briefly, sites that were >97 ha were classified based upon 19 different habitat classes according to the original Iowa Wildlife Action Plan (Kinkead, 2006; Zohrer, 2006). To ensure equal representation across the entire state, we split the state into four relatively equal management districts (northeast, southeast, northwest, and southwest). Each year, sites were randomly chosen without replacement within each habitat type within each management district. We also randomly selected a subset of "permanent" sites to be surveyed every year. In addition to sites on public land, a few of the sites were located on private land and were surveyed as part of the Landowner Incentive Program (https://wsfrprograms.fws.gov/ subpages/grantprograms/lip/lip.htm). Finally, a small number of sites (n = 88) were non-randomly selected for a variety of management reasons and are included in our analyses. These non-random sites were not chosen with any expectation that they were more or less likely to be occupied by monarchs, and they included a similar range of habitat types to the randomlychosen sites. At each site, primary habitat was delineated by a 10.4-ha area in the shape of a hexagon oriented in a north-south direction. Hereafter, we refer to this area of primary habitat as the "core area," and butterfly and habitat surveys always took place within this core area.

#### Butterfly Surveys

We utilized line transects to survey monarchs between 2006 and 2017 following a modified approach of Pollard and Yates (1993). These transects were used to survey all butterfly species, including monarchs. Transects were 400 m in length and mostly situated in a north-south direction so as to dissect the core habitat of the site. However, in some cases transects were placed to avoid barriers (e.g., rivers, major roads), so they were split or oriented in an east-west direction. Transects were divided into 10-m sections, resulting in 40 sections per transect. Observers slowly walked each section, taking ∼1 min per section, and counted all butterflies seen within 2.5 m of each side of the transect (so a 5 m transect width). There were typically 3–5 observers every year, and observers often differed between years. Observers usually remained consistent throughout a season and received training to identify butterflies to species prior to start of surveys. We assumed that training efforts combined with the ease of identifying a monarch in the field were sufficient to minimize observer effects, which we did not consider in our models. At the beginning of each survey we recorded temperature (◦C), wind speed (km/h), and percent cloud cover. We conducted surveys between 1000 h and 1830 h when the temperature was between 21 and 35◦C and wind speed was <16 km/h. For most properties we conducted four surveys between June 1 and August 31 with at least 2 weeks in between surveys, but in some years a cold spring or early summer resulted in a delay in the start of surveys and some surveys occasionally continued into September.

#### Local-Scale Variables

A primary focus of our work was to understand how sitespecific habitat attributes of a site influenced monarch use. To estimate these local-scale (or habitat) variables, we conducted annual vegetation surveys during the growing season (month of August) within the core area of each site. We were interested in three local-scale variables: average litter depth (cm), frequency of milkweed, and percent canopy cover. Vegetation surveys were conducted at eight plots within the core area: six were situated at each corner of the hexagonal core area, one was placed in the center of the core area, and another was randomly chosen each year at each site. During the early years of the MSIM program, three additional plots were surveyed instead of one, but we reduced the number of plots due to time constraints. Each plot had a radius of 17.3 m and vegetation transects were established at three aspects (30◦ , 150◦ , and 270◦ ) from the center of the plot. Different measurements were taken along each transect. For the first transect we measured litter depth (cm) at three distances (2.5, 5.0, and 7.5 m) from the center of the plot. We averaged litter depth across each of these distances for each site and year. Along the first and second transect we placed 1-m<sup>2</sup> quadrats at 4.6 m from the plot center. We estimated the percent aerial coverage of all plants. Most plants were identified to species but some were identified only to family or genus. We identified additional plant taxa by completing a 5-min search of the entire plot after each survey. Milkweed frequency was estimated by averaging the number of the larger plots that had milkweed of any species present. Finally, we measured the presence of canopy at the four cardinal directions 7.3 m from the center of each plot. The observer did this by looking up and recording the presence or absence of any visual vegetative obstruction. Canopy cover at each site was then estimated as the frequency at which canopy was recorded as present at those four sampling points within each larger plot (Iowa Department of Natural Resources, 2016).

#### Landscape-Scale Variables

The local-scale habitat attributes were paired with landscapelevel covariates to provide a clearer picture of how monarchs used each site. Unlike the local-scale variables, we calculated landscape-level variables using remotely-sensed data. Landscape data were calculated from the 2009 high resolution Iowa Landcover file (Ensminger et al., 2016) and analyzed using ArcMap 10.5.1 and FRAGSTATS (ver. 4.2; McGarigal et al., 2012). The 2009 Iowa Landcover file has a 3-m resolution and includes 16 landcover classifications, some of which we merged into broader categories of woodland (merged 4 classifications), grassland (merged 3 classifications), and agriculture (merged 2 classifications). We assessed all landscape covariates at four different spatial scales: 100, 200, 500 m, and 1 km. Our choice of this range of spatial scales was based on previous work illustrating how butterflies select habitat at differing scales (Bergman et al., 2004; Davis et al., 2007; Luppi et al., 2018). We generated buffers (Analysis Tools, Proximity, Buffer) at these four radii around each transect and clipped the landcover layer using the clip tool (Dana Management Tools, Raster, Raster Processing, Clip). The resulting raster layers were used in FRAGSTATS to estimate class specific covariates at each spatial scale. Specifically, we were interested in landscape metrics for "grassland," "woodland," and "agriculture" in the 2009 Iowa Landcover file. We were interested in six measurements for each landcover type: largest patch index (LPI), percentage of landscape (PLAND), edge density (ED), patch density (PD), mean patch size (PS), and interspersionjuxtaposition index (IJI). LPI measures the percentage of the landscape made up of the largest patch of the respective landcover type. PLAND is a measure of the percentage of the landscape made up the landcover type. ED is equal to the total length (m) of the edge of the landcover type in the landscape per ha, and PD is the number of patches in the landscape per ha. PS is the mean patch size (ha) within the landscape. IJI measures the degree of interspersion of the landcover type within all other landcover types in the landscape. High values of IJI result from landscapes in which the patch types are interspersed (i.e., equally adjacent to each other) while low values result from landscapes in which the patch types are poorly interspersed (McGarigal et al., 2012). Collectively, these metrics allowed us to characterize the landscape within which each site was situated, albeit for just a single year (2009) of classified habitat data. Remotely-sensed data were not available for all years of the study, nor were 3-m resolution data available for other years of the study than 2009. This is an important assumption of the landscape-level analyses, and we think it is reasonable because (a) the landscape classes are coarse and merge multiple landcover classes, (b) the likelihood of switching between these classes during the 12-year study seems rather small (e.g., it would be unlikely for a pixel to switch from agriculture to forest in 12 years because of limitations in plant succession), and (c) any changes that did occur were probably rare and thus unlikely to alter overall patterns. Thus, we assume this single year, situated in year four of our 13-year study, is representative of all years of the study.

#### Data Analysis

We used the robust design occupancy model (MacKenzie et al., 2003) in Program MARK (White and Burnham, 1999) to examine the effects of local- and landscape-scale variables on site occupancy and dynamics of monarchs. The model data are summarized in presence-absence encounter histories and do not make use of actual counts. This model produces estimates of site occupancy (ψ) while accounting for imperfect detection (p) over multiple primary sampling occasions (i.e., years). As with single-season occupancy estimation, the assumption of population closure applies within the sampling timeframe (i.e., secondary sampling occasions) but is relaxed between primary occasions in the robust design model (MacKenzie et al., 2002, 2003). From this model we can estimate site dynamics between years. This estimation is done by calculating site extinction probability (ε, the probability a site is unoccupied at time t + 1 given that it was occupied at t), and site colonization probability (γ, the probability a site is occupied at t + 1 given that it was unoccupied at t; MacKenzie et al., 2003). We acknowledge that all monarchs disappear from a site between years because they are migratory, so the interpretation of these two parameters differs slightly from the usual explanation. Both essentially represent changes in state (occupied to unoccupied, or unoccupied to occupied) as a rate change between years. In our study, the primary sampling occasions were the years 2006 to 2017 (12 primary occasions). The secondary sampling occasions were the surveys that occurred within each year.

A major challenge for our analysis was how to best approach the closure assumption that is required for the secondary sampling occasions. The monarch has a unique life cycle that involves multiple generations produced in quick succession as the species migrates north each spring (Batalden et al., 2007). The first generation is produced in spring in the southern U.S. (generally prior to 1 June; Nail et al., 2015) and their progeny move north to Iowa and the Midwest beginning in May (Schlict et al., 2007). An additional 2–3 generations are produced in Iowa during the summer (generally mid-May through September; Nail et al., 2015), the last of which eventually migrates to Mexico to overwinter and then migrate north in the spring (Solensky, 2004). There is also annual variation in the phenology of these generations, which further complicates the problem (Nail et al., 2015). This overlap of generations technically violates closure because new individuals can emerge during the sampling period. We argue that by restricting our sampling season we can minimize this assumption violation sufficiently to result in an analysis that is of value. To meet the assumption of population closure within years, we truncated the secondary sampling occasions to just the primary breeding season of monarchs (primarily second and third generations) in Iowa, which is between 1 June and 31 August on average (see Schlict et al., 2007). This timeframe excludes most migrant monarchs from the fourth and possible fifth generations (Nail et al., 2015).

We used two parameterizations of the robust design occupancy model: one that estimates occupancy, extinction probability, and detection probability, and another that estimates occupancy, colonization probability, and detection probability. These parameterizations are necessary because the likelihood function cannot include all estimates of occupancy, extinction, and colonization in a single model that allows covariates on each parameter (MacKenzie et al., 2006). Therefore, multiple formulations must be used to have covariates on each parameter. Luckily, different formulations of the model are directly comparable using model-selection procedures (e.g., Akaike's Information Criterion; AIC) because they have the same model likelihood. For each parameter and to determine important covariates, we used a step-wise modeling approach similar to Harms et al. (2014) and evaluated models using AIC adjusted for small sample sizes (AICc; Burnham and Anderson, 2002). We examined each parameter separately and in the following order: detection probability, occupancy, then extinction, and colonization probabilities. For each parameter, a suite of models was run (see below for details about what was included and in what combinations), keeping all other parameters constant. Detection probability was first, and the best model structure for detection probability was then used in the two model formulations that allowed us to model the remaining three parameters. All analyses were done using the "RMark" package (Laake, 2013) in Program R (R Core Team, 2018).

We first modeled covariates on detection probability while keeping all other parameters constant. We were interested in the effects of wind speed, temperature, and cloud cover on detection, along with canopy cover and yearly estimates of monarch densities. Because of their small size, the activity and behavior of butterflies can be greatly impacted by environmental covariates, including temperature and wind speed (Wikström et al., 2009; Cormont et al., 2011; Bried and Pellet, 2012). For example, butterflies are not as active in cooler temperatures (Wikström et al., 2009) and there is evidence that monarchs will avoid shaded regions during the breeding season (Zalucki and Kitching, 1982). Canopy cover may also affect detection for this reason, but it may also affect the ability of observers to see monarchs. Thus, we expected temperature, cloud cover, and canopy cover to each negatively influence detection probability. Alternatively, monarchs are strong flyers and we attempted to survey on days with calm winds (<16 km/h), so we did not expect wind speed to affect detection of monarchs. Additionally, detection probability tends to increase with an increasing number of individuals, and this has been demonstrated for several taxa, including insects (Mercader et al., 2012). We calculated a simple count, unadjusted for imperfect detection, for each year of the study as a simple index of monarch density. Counts were first averaged for all visits to a site within a year (typically four visits), and then averaged again across all sites sampled in a year. We used this as a covariate on detection probability, similar to the justification outlined by Royle and Nichols (2003) for a densitydependent effect on detections. We ran all possible combinations of single and additive covariate effect models, except that we did not include temperature and cloud cover in the same model because they were highly correlated. We used the highest-ranked model from this procedure as the detection probability structure in all further analyses.

Next, we examined the effects of three local-scale habitat covariates and five landscape-level covariates (each at four different spatial scales) on site occupancy, extinction, and colonization. We first assessed the main effects of each covariate on occupancy probability with extinction probability fixed with no effects. Because we were interested in the potential interaction between patch size and patch density, we also examined the interaction between these two metrics for each habitat variable within the same spatial scale. We identified covariates with strong (95% confidence intervals did not contain zero) coefficients. These covariates were combined into two- and three-factor additive models (four-factor for multiplicative interactions) for occupancy, but we did not include highly correlated variables (r > 0.60) in the same model. We used the top models (all those with 1AICc < 2) for occupancy in subsequent analyses for extinction and colonization, where we used the same methods described above for the main effects and additive models. At

TABLE 1 | Hypothesized effects of habitat and landscape-level covariates on the occupancy, colonization, and extinction probabilities of monarchs in Iowa, 2006–2017.


*For each habitat class, the six covariates were ED, edge density; IJI, interspersion and juxtaposition index; LPI, largest patch index; PD, patch density; PLAND, percentage of the landscape in that habitat; PS, mean patch size. "0" indicates an effect for which we predicted no effect. See Methods for full details.*

each step for all parameters, we scrutinized the model results and removed any models with uninformative parameters or numerical convergence problems. We also excluded models from the competitive model set if they were more complicated versions of nested models (Burnham and Anderson, 2002; Devries et al., 2008; Arnold, 2010). Based on preliminary analyses and concerns about the inability of models to numerically converge, we did not consider models with more than three covariates for occupancy and four covariates for extinction and colonization. This resulted in a moderate number of models for occupancy (n = 1,217), extinction (n = 1,478), colonization (n = 2,333), and detection probability (n = 21). We used the top model from each step to evaluate predictive patterns, which we illustrated only for strong effects (e.g., those with a 95% CI that did not include zero).

For each parameter we developed specific hypotheses concerning the local and landscape effects (**Table 1**). We expected grassland patch metrics to be particularly important to monarchs (Jepsen et al., 2015). Not only do grasslands harbor host plants, but they provide a greater diversity of nectar sources than agricultural land or woodlands, which has been shown to have a positive effect on abundance for other butterfly TABLE 2 | Number of properties, number of butterfly surveys, and frequency of each with monarchs, Iowa, 2006–2017.


*<sup>a</sup>Note that the total includes properties that were surveyed more than 1 year during the study. The total number of unique properties surveyed was 417.*

species (Clausen et al., 2001; Ries et al., 2001; Pywell et al., 2004; Luppi et al., 2018). In contrast, with the increased use of herbicide-tolerant corn and soybeans, large patches and a high density of agricultural land may negatively affect occupancy and colonization. However, the edges of agricultural fields still harbor milkweeds (Pitman et al., 2018), so we expected the edges of agricultural patches to positively affect monarch occupancy and colonization. Smaller patches of agriculture also have more edge and less interior habitat relative to larger patches, so we hypothesized that agricultural patch size would also influence occupancy and colonization. This may also be true for grassland habitat. The edges of grassland habitat may be more susceptible to disturbance, which could increase the amount of milkweed. Indeed, increased patch size tends to have a negative effect on disturbance-tolerant species, such as the monarch (Davis et al., 2007; Pitman et al., 2018). We expected woodland cover to negatively affect occupancy and colonization and to increase the probability of extinction. Woodlands may not have abundant host plants compared to grasslands or agricultural edges for breeding monarch butterflies (Bhowmik and Bandeen, 1976; Pleasants, 2017). Furthermore, woodland edges tend to decrease monarch crossings when they are adjacent to prairies (Ries and Debinski, 2001), which may indicate that increased woodland within the surrounding landscape decreases colonization.

#### RESULTS

We surveyed for monarchs at 417 sites across Iowa during the 12-year study (**Figure 1**). The total number of site × year visits totaled 787 (**Table 2**; mean = 65, range was 14–126). By year, the percentage of surveyed properties with monarchs ranged from 43 to 100%, but the percentage of surveys that detected one or more monarchs was much lower and ranged from 20 to 56% (**Table 2**).



*This illustrates that of the 417 properties surveyed during this study, 259 were surveyed in only 1 year, 93 were surveyed in 2 years, etc.*

Most properties in this study were visited just 1 year (n = 259) with fewer properties visited more years; two properties were surveyed for 11 years (**Table 3**). Yearly means of the three habitat covariates were variable (**Table 4**) with canopy cover ranging from 32 to 52%, litter depth from 1.19 to 6.10 cm, and milkweed frequency from 0 to 100%. Estimates (mean, SD, and range) for six landscape-level covariates at four spatial scales calculated using the 2009 land-cover data for Iowa showed considerable variation between grassland, woodland, and agricultural habitats (**Table 5**).

Occupancy models revealed that monarchs responded differently to landscape features, environmental conditions, and local habitat conditions for site occupancy, extinction, and colonization probabilities. For site occupancy, there were three competitive models (**Table 6**) but some uncertainty about covariate effects. These three models all had additive positive effects of grassland (patch size, percent of landscape, and patch density) at larger spatial scales, a negative effect of the percent of the landscape in woodland at the 200-m spatial scale, and a negative effect of canopy cover. In the best model, the mean patch size of grassland at the 1-km spatial scale had a positive effect (βGrassPS1<sup>K</sup> = 0.94, SE = 0.54) while the percent of the landscape in woodland at the 200-m spatial scale had a negative effect (βWoodPL<sup>200</sup> = −1.68, SE = 0.34). For extinction there were also just three competitive models (**Table 6**), which showed a similar uncertainty in covariate effects to site occupancy. All three models contained an effect of grassland interspersion at the 1-km spatial scale; woodland effects at small spatial scales were in all three models, litter depth was in two models, and an effect of agriculture at the 1-km spatial scale was in one model. In the best model, there were additive effects of the percent of the landscape in woodland at the 100-m spatial scale (βWoodPLAND<sup>100</sup> = 2.70, SE = 0.63), the interspersion of grassland at the 1-km spatial scale (βGrassIJI1<sup>K</sup> = −2.30, SE = 0.63), and litter depth (βLitter = 0.46, SE = 0.13). Finally, there were just two competitive models for colonization and both contained a woodland effect at the 200-m spatial scale (percent of landscape and largest patch index) and an effect for the interspersion of grassland at the 100-m spatial scale (**Table 6**). In the best model, there were negative effects of the percent of the landscape in woodland at the 200-m spatial scale (βWoodPLAND<sup>200</sup> = −4.67, SE = 1.37) and the interspersion of grassland at the 100-m spatial scale (βGrassPS1<sup>K</sup> = −2.02, SE = 0.70). In general, across all models and parameters grassland effects occurred at larger spatial scales than woodland effects.

Detection probability was most affected by the additive effects of canopy cover and monarch density; no other detection model was competitive. In the top site occupancy model there was a positive effect of monarch density (βDensity = 0.28, SE = 0.05) and a negative effect of canopy cover (βCanopy = −0.18, SE = 0.03) on detection probability (**Table 6**). The direction and magnitudes of these effects were almost identical in the top extinction and colonization models. Estimates of detection probability were generally >0.40 for mean covariate values (range was 0.33– 0.62). There was no support for the effect of wind, cloud cover, or temperature on detection probability. The slope parameter estimates for all of these weather variables were small (<0.03) and confidence intervals on the estimates included zero.

Using the best model for each parameter, we developed predictive relationships for key covariates of interest. First, we predicted the probability of site occupancy as a function of the percent of the landscape in woodland at the 200-m spatial scale (**Figure 2**). Next, we predicted the probability of site extinction as a function of the percent of the landscape in woodland at the 100 m spatial scale (**Figure 3**). Finally, we predicted the probability of site colonization as a function of the interspersion of grassland at the 100-m spatial scale (**Figure 4**). These three figures illustrate the primary drivers of occupancy, extinction, and colonization parameters in our study and illustrate how each responds to a particular landscape-level covariate.

#### DISCUSSION

The monarch butterfly responds to a suite of local- and landscape-level habitat features that collectively describe its occupancy and colonization/extinction dynamics at sites in the core of its U.S. breeding range. Our study is the first to investigate these patterns for the monarch in an occupancy modeling framework within a metapopulation context. We believe the results provide further insight into patterns of habitat selection for the monarch, and relate directly to future management and conservation efforts. Below, we place our findings in a larger context with monarch and other butterfly literature and discuss important model assumptions and how deviations from them may have impacted our findings.

Occupancy analyses have not been previously conducted for the monarch across any large portion of its breeding range. Such an analysis requires repeat visits to many sites, preferably selected at random, which is logistically challenging. Citizen Science efforts such as the Monarch Larva Monitoring Project (MLMP; Prysby and Oberhauser, 2004; Nail et al., 2015) offer large datasets with appropriate survey data, but the non-random site selection limits inferences. Occupancy analyses have only rarely been used for other butterfly species, most often single-season TABLE 4 | Yearly estimates (mean, SD, and range) of the average measures of three site-specific covariates (canopy cover [%], litter depth [cm], and milkweed frequency [%]) at all sites surveyed in that year, Iowa, 2006–2017.


TABLE 5 | Estimates (mean, SD, and range) for six landscape-level covariates at four spatial scales calculated in each of three habitat classes for an occupancy analysis of monarchs, Iowa, 2006–2017.


*The covariates were ED, edge density; IJI, interspersion and juxtaposition index; LPI, largest patch index; PD, patch density; PLAND, proportion of the landscape; PS, mean patch size.*

models where only site occupancy and detection probability are estimated (Puntenney and Schorr, 2016). However, logistic regression analyses where presence-absence is correlated with habitat and other covariates is a common butterfly research methodology (recent examples include Powniatowski et al., 2018; Seidel et al., 2018; Zhang and Miyashita, 2018). In these types of analyses, the detection process is ignored and the focus is on the occupancy process and associated correlates. A potential TABLE 6 | Model selection results for site occupancy (ψ), site extinction (ε), site colonization (γ), and detection probability (p) for monarchs surveyed in Iowa, 2006–2017.


*For each model, we show the number of parameters (K), Akaikie's Information Criterion adjust for small sample sizes (AICc), the difference in AIC<sup>c</sup> from the top model (*1*AICc), the model weight (w<sup>i</sup> ), and the model deviance. Landscape covariates are represented by "Grassland," "Woodland," or "Agriculture" along with the landscape metric and the buffer area [100 m, 200 m, 500 m, or 1 km (1K)]. The landscape-scale covariates were ED, edge density; IJI, interspersion and juxtaposition index; LPI, largest patch index; PD, patch density; PLAND, proportion of the landscape; PS, mean patch size. Local-scale covariates included canopy cover (Canopy) and litter depth (Litter). "Density" is the average monarch density for a given year.*

pitfall of using just logistic regression is that the detection probability is either assumed to be perfect (1.0) or constant across all surveys and is not estimated directly from the data. This assumption may work for certain species, in open habitats where butterflies are not obscured, and for surveys where the area being covered is small (e.g., narrow line transects). However, our estimated detection probability for the monarch, a large, visible,

and easily identified butterfly, was still <1.0 suggesting that this is not a good assumption. Patterson (2016) estimated a detection probability of 0.79 for monarchs sampled using Pollard-Yates transects on grasslands in central Iowa, which was similar to many of our estimates. It is important to note that detection probability is seldom estimated for any butterfly species, although all published studies indicate that it is often substantially

<1.0 and few recommend estimating it directly (see **Table 1** in Lindzey and Conner, 2011; Henry and Anderson, 2016; Ribiero et al., 2016).

Our results reveal that monarchs overall had a high probability of occupancy of our sites during the breeding season (>0.90), and that this was correlated with landscape covariates as opposed to local covariates. It was surprising that we found no strong support for the local-scale covariates, which we hypothesized would be useful predictors of these parameters. This is surprising given what is known about habitat preferences of the monarch. The preferred breeding sites are thought to be open areas with a mix of nectar-rich resources for adults to feed along with milkweed for oviposition sites (Zalucki and Kitching, 1982). We found that woodland at relatively small scales tended to decrease occupancy and colonization probabilities and increase extinction probabilities. This could be attributed to the cooler conditions of woodlands as monarchs tend to utilize areas that receive sunlight (Zalucki and Kitching, 1982). Additionally, the potential lack of feeding sources and oviposition sites may explain the low occupancy and colonization probabilities and high extinction probabilities.

While woodlands likely contain some nectar sources, they do not harbor the high densities of milkweeds like grasslands or agricultural areas (Bhowmik and Bandeen, 1976; Hartzler and Buhler, 2000). Indeed, the probability of an area being occupied and colonized by monarchs decreased as woodland cover increased but this never reached zero. Similarly, extinction probability never reached 1.00. We hypothesize that even with 100% woodland cover, monarchs may still be found depending upon the presence of understory forbs and milkweed plants. This hypothesis agrees with Brower's (1995) assessment that the Upper Midwest prior to conversion to agriculture contained abundant milkweed and other diverse nectar sources. Seitz (1924) noted that milkweeds were capable of easily colonizing disturbed forests, suggesting that even heavily forested regions may be occupied by monarchs.

Our findings suggest that as the percent of woodland increases within the 200-m buffer, the probability of an area being occupied by monarchs decreased yet never reached 0. The average percentage of woodland for our study across all sites at the 200-m scale was 43.35% and ranged from 0 to 100%. Similarly, extinction, or the likelihood that a site that contained monarchs in year one would not have monarchs in year two, was also heavily influenced by the percentage of woodland (**Figure 3**), although for this parameter the scale was 100 m. Our site average for woodland at the 100-m scale was 43.37% and ranged from 0 to 100%, and as the percentage increased, the likelihood that the site would not have monarchs also increased, yet again, never reached 100%.

Lastly, we illustrate the impact of the interspersionjuxtaposition index (IJI) of grassland at the 100-m scale on colonization probability. Our findings suggest that as patches of grassland became smaller and more mixed throughout patches of agriculture, woodland or other landcover types, the probability of colonization decreased; as these grassland patches became larger and more clumped together colonization increased. This seems to be in contrast to findings that monarchs tend to use smaller patches of available habitat (Davis et al., 2007) and have higher egg densities in smaller patches than in larger patches (Pitman et al., 2018). Our findings could be due to the scale at which this metric was measured. Monarchs can fly long distances in a single day (up to 15 km/d according to Zalucki et al., 2016), so "large" patches at a 100-m scale may still be comparatively small for this species. Furthermore, greater connectivity among grassland patches, particularly at this small scale, may be beneficial to the monarch for finding more nectar sources and oviposition sites. For example, other landscape covariates, such as woodland edges, can act as barriers to monarch movement (Ries and Debinski, 2001). Monarchs are highly mobile and females in particular travel frequently among patches to find more milkweed (Zalucki and Kitching, 1982; Grant et al., 2018). Thus, more contiguous grassland patches may be preferable to more interspersed habitat. Note, however, that this parameter did not begin to drop from above 90 until the IJI was above 50 and that it had a wide confidence interval. Conversely, it never dropped to zero, indicating that many factors are influencing whether adult monarchs choose to use a patch.

Associations between landscape features and presenceabsence of butterflies have been widely investigated (Dover and Settele, 2009; Bergerot et al., 2011; Öckinger et al., 2012; Luppi et al., 2018). Bergman et al. (2004) noted that large spatial scales were needed for predicting landscape effects on butterfly communities in agricultural regions. Other studies (Luppi et al., 2018) emphasized the importance of examining these effects at multiple spatial scales because of scale-dependent effects. As a taxon, butterflies may be especially sensitive to increasing human presence across a wide range of ecosystems (Gross, 2016; Van Swaay et al., 2016). This is especially relevant in Iowa, where native habitat occurs in small, fragmented patches that are embedded within a primarily agricultural landscape. Indeed, Iowa has <0.01% of its native grasslands remaining, many of them isolated from other grasslands (Samson and Knopf, 1994). Ries and Debinski (2001) noted that at 26 prairies in central Iowa an estimated 50% of their total perimeter of prairie edges consisted of row crops and 38% had a road or treeline as an intersecting boundary feature.

Occupancy models have the assumption of demographic closure for the secondary sampling periods (MacKenzie et al., 2002), which in our study was the main breeding season of the monarch in Iowa. This means that there are assumed to be no population additions (births or immigrants) or losses (deaths or emigrants) during the secondary sampling period. In the strictest sense, this is seldom true for most wildlife studies, and the primary goal shifts to strategies to minimize violating this assumption. The multi-generational life cycle of the monarch makes this especially challenging because (a) some generations overlap, (b) there is no single time period within the broad breeding season when a local population is not subject to population additions or losses, (c) the timing of generations shifts to an unknown and unpredictable extent between years, and (d) there is no accurate field method for identifying an individual monarch to a specific generation. Like any model, occupancy models produce biased parameter estimates if model assumptions are violated, key among them the assumption of closure (MacKenzie et al., 2003). Puntenney and Schorr (2016) discussed similar challenges meeting the closure assumption, especially as they related to flight times and the mobility of their focal species. Rota et al. (2009) noted that estimates of occupancy tended to be biased when detection probability was low (p < 0.30), the initial occupancy rate was low, and extinction was high. In our study, detection probability was higher (p > 0.40), site occupancy was high, and extinction probability was very low. MacKenzie et al. (2006) noted that when there were random changes in occupancy within a season, the occupancy estimator was approximately unbiased. We believe that movements by monarchs within a season are random because this is a wide-ranging, highly mobile species, and thus the estimates of occupancy in our study should be approximately unbiased. Finally, we note that if the movement of monarchs (immigration and emigration) occurring during the secondary sampling period is random, then occupancy can be interpreted simply as use of patches (Bailey et al., 2007).

An important aspect of occupancy models is that they make use of presence-absence data rather than counts or density (MacKenzie et al., 2002, 2003). Thus, an occupied site contains >1 monarch in our study; no distinction is made between sites occupied by a single individual vs. those with more individuals. There are other models that can integrate site occupancy with measures of abundance (Royle and Nichols, 2003). Occupancy

#### REFERENCES

Arnold, R. A. (1983). Ecological studies of six endangered butterflies (Lepidoptera, Lycaenidae): island biogeography, patch dynamics, and the design of nature reserves. Univ. California modeling has become popular for monitoring populations and understanding habitat associations because there is no need to mark and re-encounter individuals. The models require both temporal and spatial replication (Bailey et al., 2007), which creates a trade-off between adding more sites and sampling existing sites more often. The feasibility of conducting a large number of surveys is enhanced when there is no need to capture and mark individuals (as in a mark-recapture study) and this allowed us to have good temporal replication (approximately four surveys per breeding season per site) while still visiting a large number of sites each year. The limitation of using only presence-absence data is that there is no information on abundance, density-dependent effects on parameters can be hard to model, and the ultimate assessment of a site's importance rests on something other than the number of individuals that are present.

We present the first estimates of the meta-population dynamics of the monarch using a 12-year dataset from Iowa. Our findings suggest that monarchs occupy sites using landscape-level habitat cues, but these cues differ for occupancy, colonization, and extinction. Local site variables such as the amount of milkweed were unimportant in our models. Detection probability was a function of monarch density and canopy cover and exceeded 0.40 in our models. Our models were useful for making predictions about the effects of key covariates on these parameters, which in turn may be useful for monarch conservation efforts in the Midwest and elsewhere.

#### ETHICS STATEMENT

This work involved the study of invertebrates, which are not covered by the Iowa State University IACUC committee. Therefore, no IACUC permit was required for this work.

#### AUTHOR CONTRIBUTIONS

KK and SD conceived the idea. SD, RV, KM, KK, and PF collected the data. SD, RV, and KM analyzed the data and took the lead on writing the paper with contributions from all co-authors.

#### ACKNOWLEDGMENTS

This work was funded by various State Wildlife Grants (grant numbers T-6-R-1, T-6-R-2, T-6-R-3, T-6-R-4, T-6-R-5, F15AF00269, and F15AF00257), Iowa State University, the Iowa Department of Natural Resources, the U.S Army Corps of Engineers, and a Landowner Incentive Program Grant under the U.S. Fish and Wildlife Service Wildlife and Sport Fish Restoration Program. This paper is a product of the Iowa Agriculture and Home Economics Experiment Station, Ames, Iowa. Project No. IOW05438 is sponsored by Hatch Act and State of Iowa funds.

Public. Entomol. 99, 1–161. doi: 10.1111/j.1937-2817.2010. tb01236.x

Arnold, T. W. (2010). Uninformative parameters and model selection using Akaike's Information Criterion. J. Wildl. Manag. 74, 1175–1178. doi: 10.2193/2009-367


Available online at the following web site: <http://www.umass.edu/landeco/ research/fragstats/fragstats.html>.


Prior, J. C. (1991). Landforms of Iowa. Iowa City, IA: University of Iowa Press.

Prysby, M., and Oberhauser, K. S. (2004). "Temporal and geographical variation in monarch densities: citizen scientists document monarch population patterns," in The Monarch Butterfly: Biology and Conservation, eds K. S. Oberhauser and M. J. Solensky (Ithaca, NY: Cornell University Press), 9–12.


**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 Dinsmore, Vanausdall, Murphy, Kinkead and Frese. 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.

# Monarch Butterfly Distribution and Breeding Ecology in Idaho and Washington

#### Beth Waterbury <sup>1</sup> \*, Ann Potter <sup>2</sup> and Leona K. Svancara<sup>3</sup>

*1 Idaho Department of Fish and Game, Salmon, ID, United States, <sup>2</sup> Washington Department of Fish and Wildlife, Olympia, WA, United States, <sup>3</sup> Idaho Department of Fish and Game, Moscow, ID, United States*

Studies of monarch butterflies (*Danaus plexippus*) and their milkweed (*Asclepias* spp.) host plants in North America have focused primarily on monarch populations ranging east of the Rocky Mountains. We report the first systematic assessment of monarch butterfly and milkweed populations in the western states of Idaho and Washington, states at the northern tier of western monarch breeding range. Results of our 2-year study (2016–2017) offer new insights into monarch breeding habitat distribution, characteristics, and threat factors in our 2 states. We documented milkweeds and breeding monarchs in all 16 climate divisions in our study area. Milkweed and breeding monarch phenologies were examined with evidence supporting 2, and possibly 3 monarch generations produced in Idaho and Washington. Key monarch breeding habitats were moist-soil sites within matrices of grasslands, wetlands, deciduous forest, and shrub-steppe supporting large, contiguous, and high-density milkweed stands. Co-occurrence of showy milkweed (*A. speciosa*) and swamp milkweed (*A. incarnata*) was an important indicator of productive monarch breeding habitat in Idaho. Nectar plants were generally limited in quantity and richness across the study area, particularly in late summer, and included frequently-used non-native, invasive species. Primary threats at milkweed sites were invasive plant species, herbicide application, and mowing, followed by secondary threats of recreational disturbance, livestock grazing, insecticide application, loss of floodplain function, and wildfire. We provide management recommendations and research needs to address ongoing stressors and knowledge gaps in Idaho and Washington with the goal of conserving monarchs and their habitats in the West.

Keywords: monarch butterfly, Danaus plexippus, milkweed, Asclepias, Idaho, Washington, monarch breeding habitat, milkweed and monarch threats

# INTRODUCTION

Essential to the conservation of migratory species is understanding the full life-cycle ecology of populations across geographically disparate seasonal ranges (Webster et al., 2012; Small-Lorenz et al., 2013; Flockhart et al., 2015). The North American monarch butterfly (Danaus plexippus plexippus) is an iconic migratory insect that exemplifies the challenges of conserving highly mobile species. Two migratory populations of monarchs occur in North America (Urquhart and Urquhart, 1977). The larger eastern population breeds east of the Rocky Mountains and overwinters in high-elevation forests in Central Mexico (Flockhart et al., 2013), while the western population

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*Emily L. Weiser, United States Geological Survey, United States John Pleasants, Iowa State University, United States*

> \*Correspondence: *Beth Waterbury keegan@centurytel.net*

#### Specialty section:

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

Received: *30 December 2018* Accepted: *30 April 2019* Published: *22 May 2019*

#### Citation:

*Waterbury B, Potter A and Svancara LK (2019) Monarch Butterfly Distribution and Breeding Ecology in Idaho and Washington. Front. Ecol. Evol. 7:172. doi: 10.3389/fevo.2019.00172* breeds west of the Rockies and overwinters at low-elevation wooded groves along the California coast (Dingle et al., 2005; Yang et al., 2015). Recent research indicates the boundary between populations is permeable with significant admixing occurring at breeding and overwintering sites (Vandenbosch, 2007; Pyle, 2015). The past decade has seen major advances in knowledge of monarchs, including studies focused on broadscale population trends and factors driving recent and rapid population declines (Flockhart et al., 2015; Oberhauser et al., 2015; Schultz et al., 2017). Research focus has primarily centered on the eastern monarch population. While investigation of the western population has recently increased, knowledge of basic aspects of western monarch breeding biology, migratory connectivity, and threat factors remains rudimentary. As recently as 2015, significant gaps in knowledge of distribution of monarchs and their milkweed (Asclepias spp.) host plants existed for vast areas of the western U.S. (Jepsen et al., 2015). Addressing these key knowledge gaps is a crucial first step for conserving western monarch natal habitats and migratory connectivity (Martin et al., 2007; Webster et al., 2012).

Knowledge of milkweed and monarch breeding occurrence in the Pacific Northwest in general, and Idaho and Washington specifically, is derived from a limited body of research. Pyle (1999, 2015) described severely restricted monarch breeding incidence in the region due to patchy and low-density milkweed distributions, but noted historical occurrence of dense stands of showy milkweed (A. speciosa) in the Snake River Plain in southern Idaho and Columbia Basin in eastern Washington. Stevens and Frey (2010) identified 2 climate divisions corresponding to the lower Snake River Plain and Columbia Basin as sole regions in Idaho and Washington with potential to support western migrant monarchs. They posited in such northern-latitude states, monarch development was likely constrained by cold temperatures, and to a lesser degree, by low milkweed species diversity and bottom-up effects from drought. In 2011, the Xerces Society for Invertebrate Conservation initiated a project to compile milkweed and monarch breeding records in the western U.S. As of 2015, this database amassed >12,000 milkweed records from multiple sources. Of the 700 milkweed records collated for Idaho and Washington, 88% were collected pre-2000 or had low spatial accuracy or ambiguous species identification. A mere 7 high-quality monarch records existed across both states, with most records lacking life-stage information or spatial accuracy.

In 2016, prompted by concerns about western monarch declines and major knowledge gaps in the distribution and status of monarch breeding habitats in their states, the Idaho Department of Fish and Game (IDFG) and Washington Department of Fish and Wildlife (WDFW) initiated a 2-year study with objectives to (1) determine statewide distributions of monarchs and milkweeds, (2) describe characteristics of monarch breeding habitats, (3) identify primary threats to monarchs and their natal habitats, and (4) utilize these data to guide beneficial management and future research of the western monarch. Our study also presented an opportunity to gather information on aspects of monarch breeding ecology poorly understood for the inland Pacific Northwest, such as breeding phenology, important nectar plants, and whether roosting structure is an essential component of summer natal habitats. Here, we report the findings of our bi-state study offering new insights into the regional distribution and ecology of breeding monarchs and milkweed host plants in Idaho and Washington. In this regional context, we suggest management actions and research needs to mitigate further decline of the western monarch butterfly population.

# METHODS

Our study encompassed the state of Idaho and that portion of Washington east of the Cascade Mountain Range coincident with native milkweed distribution (Xerces Society [Xerces], 2012; Hitchcock and Cronquist, 2018) (**Figure 1**). The 2 states share complex physiography dominated by several major mountain ranges, basaltic tablelands, basin and range deserts, and large river systems including the Snake, Salmon, and Columbia rivers. The study area spans 7 ecoregions (Bailey, 1976), with the Middle Rockies-Blue Mountains, Canadian Rocky Mountains, and Columbia-Snake Plateau ecoregions being contiguous between the states, and the latter ecoregion comprising the majority of the study area. Climates are highly variable, possessing both continental and marine characteristics, and temperature regimes are strongly mediated by latitude and altitude. The study area is positioned entirely west of the Continental Divide and in the northern latitudes (42–49◦N) of the western monarch's breeding range. Idaho and Washington are often grouped with other adjacent states and provinces into western monarch subregions variably named "northern inland range" (Yang et al., 2015), "Cascadia" (Pyle, 2015), and "Pacific Northwest" (James, 2016). These terms not only confer geographic location, they infer certain ecological constraints for breeding monarchs comprising the "outermost immigrants and breeders of the entire western phenomenon" (Pyle, 2015). Milkweed species richness in the study area is comparatively low (≤6 species) among western states and milkweed distributions are characterized as patchy and sparse (Pyle, 1999, 2015), though locally dense stands of showy milkweed have been documented in both states (Pyle, 2015; James, 2016).

Implicit in our study was our aim to contribute current high-resolution milkweed and monarch occurrence records to a new modeling effort to map and characterize habitat for the western monarch butterfly (USFWS and Xerces, 2016; Dilts et al., 2018). Accordingly, IDFG and WDFW developed milkweed and monarch survey protocols and field forms using standardized definitions and categories portable to this modeling effort and its future iterations (see **Supplementary Data Sheets S1**–**S3**). We defined a milkweed patch as a discrete grouping of milkweed plants separated from other milkweed patches by ≥50 m, or by dense, tall shrubs or trees, lakes or rivers, buildings, roads, or other anthropogenic land demarcations. Data collected for each milkweed patch included milkweed species, patch structure, plant count, patch area, predominant phenophase, habitat type(s), habitat association(s), management activities, threat(s), and a GPS (Global Positioning System)-derived location.

Management activities were known land management or other action(s) occurring on the site that may positively or negatively affect milkweed plants. Threats were defined as proximate stressors causing destruction, degradation, and/or impairment of milkweeds. Data collected for monarch observations included weather, time, life stage(s) observed and count, sex and behavior of adult(s), nectar species used, habitat type(s), habitat association(s), management activities, threat(s), and GPS location.

#### Idaho Surveys

We incorporated 3 approaches in our survey methods to maximize data collection opportunities. In 2016, we elected to use a spatially-balanced, stratified survey design to allow extrapolation of data across the landscape (Stohlgren et al., 1997). We selected "predicted milkweed habitat suitability" as our sampling strata to further target survey effort and efficiency. Our strata were derived from the Western Milkweed and Monarch Breeding Habitat Suitability Model (Phase I) for showy milkweed developed by the U.S. Fish and Wildlife Service (USFWS) and Xerces Society (USFWS and Xerces, 2016). Within modeled habitat, we created 4 strata to identify high (0.71–0.99), medium (0.21–0.70), low (0.06–0.20), and null (<0.06) probability of showy milkweed occurrence. We then created a grid sample frame of 270 × 270 m cells consistent with model resolution and considered a feasible size for survey effort. We applied a Generalized Random Tessellation Stratified (GRTS) sampling design to draw an ordered master sample and oversample of survey cells within the 3 suitable strata. We drew a sample of 250 primary sites (100 high, 100 medium, 50 low), the number of cells we estimated we could survey in one field season, with an overdraw of 120 sites to account for inaccessible cells (**Figure 2A**).

Standard field survey methods were developed for GRTS cells. Cell centroids and polygons were uploaded to GPS units to georeference and navigate in the field. At each GRTS cell, a team of 2 field technicians walking in tandem ≤15 m apart systematically searched for milkweeds and monarch butterflies

(eggs, larvae, pupae, adults) along parallel and adjacent linear transects covering the entire cell. Milkweed plants encountered at 10-pace intervals were thoroughly searched for presence of eggs, larvae, and pupae. At each cell corner, technicians scanned adjoining cells with binoculars for 30 s to detect presence of milkweeds and monarchs.

In 2017, we conducted milkweed and monarch breeding habitat inventories at several IDFG Wildlife Management Areas (WMA) spanning the Snake River Plain region of southern Idaho. We used photo imagery and milkweed habitat suitability (USFWS and Xerces, 2016) digital layers to identify and delineate survey units at each WMA. Field technicians systematically searched for milkweeds and monarchs along parallel and adjacent transects covering the targeted unit using the same 10-pace protocol described above to search for monarch eggs, larvae, and pupae. At C.J. Strike WMA, 2 observers surveyed the reservoir perimeter by motor boat. Milkweed and monarch data were directly recorded in Collector for ArcGIS (Android) 17.0.2 (ESRI, 2017) supported by smartphone and tablet mobile devices. This application allowed collection of high-accuracy point or polygon data, other site information, and photo attachments. Our data was synchronized to the College of Western Idaho (CWI) ArcGIS geodatabase server.

Incidental observations can capture important data on where and when plants and animals occur, often at high spatial and temporal resolutions [e.g., eBird (Wood et al., 2012; Kelling et al., 2015), iNaturalist (iNaturalist.com., 2019)]. We recorded milkweed and monarch observations when in transit to target survey sites or in follow-up to reported sightings. Site data were recorded on a field form specifically developed for incidental observations as well as on Collector. Citizen scientists were invited to contribute observational data on field forms, via the Western Monarch Milkweed Mapper website, or Collector.

As opportunities presented, we netted wild adult monarchs to deploy tags from Washington State University's (WSU) Pacific Northwest monarch tagging program, and sampled adults for Ophryocystis elektroscirrha (OE), an obligate protozoan parasite of monarchs, in support of a CWI project. These field activities were the extent to which we handled monarchs in our study. Though butterfly taxa are not regulated under animal ethics and welfare guidelines, we practiced voluntary protocols parsimonious with ethical treatment of monarchs in the field.

#### Washington Surveys

We utilized a network of roving field technicians and geographically distinct WDFW District-based wildlife biologists to conduct surveys for milkweed and monarchs in open habitats of eastern Washington; 9 WDFW Districts cover the 21 counties in this region. Citizen scientists supplemented this information in under-surveyed areas, and WDFW Wildlife Area staff assisted with collection of milkweed and monarch data at their respective stations. In addition, we surveyed for milkweeds and monarchs along road-based transects. Potential survey sites included all ownerships except USFWS Refuges, where milkweed and monarch surveys were underway. Efforts emphasized coverage of WDFW-managed Wildlife Areas. Our approach focused on locating and characterizing milkweed and monarch habitat, and capitalized on staff and volunteers' local knowledge of vegetation and environmental conditions.

Biologists selected and visited sites and roadways known or suspected to support milkweed, and systematically searched for milkweed patches. Road-edge transects were conducted using a vehicle, driver, and passenger-observer. Traveling at ≤50 mph, roadsides were surveyed for milkweed patches, and where safe to pull over, observers collected milkweed patch data and surveyed for monarchs. Standardized time-limited surveys for monarch adults and immature stages were conducted in each patch. In areas with abundant milkweed patches (>20 patches), we surveyed a 50% subset of randomly-selected patches. Adult surveys entailed visually searching milkweed patches and immediate-adjacent areas while walking through patches at ≤10 m spacing, 5-min focal observations at flowering milkweed or other flowering plant patches, and observing any adults detected for ≥10 min or until they left the area. Literature review and local expert consultation (D. James, pers. comm.) indicated monarch eggs and larvae occur most often on undersides of leaves, within the upper third of a milkweed plant, and on plants at patch edges or in lower density stands (Zalucki and Suzuki, 1987). Thus, our surveys for immature stages focused on these expected patterns, with ≥3 min allotted to search a sample within 100 milkweed plants. Milkweed and monarch data were recorded on paper field forms and recreation-grade GPS units later entered into Access (2007) and ArcGIS. A subset of milkweed patches were photo documented using georeferencing software.

#### RESULTS

### Milkweed and Monarch Distributions

From 26 May to 3 August 2016, we surveyed 163 GRTS survey cells in Idaho in predicted suitable habitats for showy milkweed and monarchs (65 high, 68 medium, and 30 low probability A. speciosa strata) (**Figure 2B**). Surveys were attempted in an additional 50 cells, but access was impeded by private land ownership. Survey cells were generally located in lower elevation (≤2,000 m), unforested landscapes outside of the central Idaho mountains. Milkweed was detected in 25 (15%) cells. By strata, milkweed was detected in 17 of 65 (26%) high probability cells and 8 of 68 (12%) medium probability cells, with no detections of milkweed in low probability cells. Showy milkweed was the only Asclepias spp. found in this survey effort. Monarchs (adults, eggs, larvae) were detected in 6 (4%) cells, with 3 detections each in high and medium strata cells.

Our combined bi-state survey effort from 26 May to 1 September in 2016, and 3 June to 20 September in 2017, resulted in 3,616 milkweed patch observations across Idaho (n = 2,875) and eastern Washington (n = 741). In Washington, 149 milkweed patches were detected during road-edge surveys. We documented 5 milkweed species in the study area (**Figure 1**), with showy milkweed by far the most commonly reported species (92%). Swamp milkweed (A. incarnata) (6%), narrow-leaved milkweed (A. fascicularis) (1%), pallid milkweed (A. cryptoceras) (0.6%), and spider milkweed (A. asperula) (0.1%) were less commonly reported. All 5 milkweed species were found in Idaho and 3 milkweed species were found in Washington (A. speciosa, A. incarnata, and A. fascicularis). Milkweeds occurred in all 16 climate divisions (**Figure 3**) and 52 of 65 (80%) counties within the study area (**Figure 1**), with largest patches and greatest abundance of milkweeds found in the Columbia Plateau Ecoregion spanning both states. We documented first showy milkweed records for 12 Idaho counties (Blaine, Boise, Bonner, Bonneville, Boundary, Caribou, Clark, Jerome, Lewis, Lincoln, Minidoka, Oneida), and Franklin County, Washington. First swamp milkweed records were documented for Bannock, Bonner, Cassia, and Idaho counties, Idaho; and a first state record for Washington in Okanogan County. We also documented first narrow-leaved milkweed records for Payette County, Idaho; and Chelan, Ferry, Franklin, Skamania, and Stevens counties, Washington. Several localities with sympatric milkweed species were documented, including extensive areas supporting intermixed populations of showy and swamp milkweeds in Idaho's Snake River Plain (**Supplementary Table S1**). Milkweed elevations ranged from 0 to 1,686 m (0–855 m in Washington; 670–1,686 m in Idaho) with showy milkweed exhibiting the broadest elevational and ecological amplitude of milkweed species found in the study area.

Our surveys generated 842 new breeding-season monarch records (n = 615 from Idaho; n = 227 from Washington) for the study area, including observations of monarch eggs (n = 178), larvae (n = 201), pupae (n = 4), and adults (n = 474). Monarchs were distributed across all 16 climate divisions (**Figure 3**) and 44 of 65 (68%) counties within the study area (**Figure 4**). Compared to available monarch records pre-project, our results demonstrated a 167% increase in climate divisions and 529% increase in counties occupied by breeding monarchs in the study area. Monarchs were observed at elevations from 0 to 1,686 m (0–573 m in Washington; 670–1,686 m in Idaho), with monarchs exhibiting lower elevational amplitude in Washington.

Across both states, milkweed observations were more abundant and broadly distributed than monarch observations.

#### Monarch Life History

Monarch records collected during this study helped to fill life history data gaps on monarch breeding phenology (**Figure 5**) and use of nectar resources and roosting sites in Idaho and Washington. Arrival of first adults was typically the first week of June, and intriguingly, largely comprised fresh condition (immaculate, bright colors) migrants in Idaho and worn (torn or missing wing sections, faded colors) individuals in Washington. The former condition indicates recently emerged adults, and the latter, older butterflies that likely traveled a greater distance. A pattern of different wing wear between the two states suggests the geographic origin of newly arrived adults may differ. First eggs were observed in close succession with arrival of first adults, usually by mid-June. Early-instar larvae from this first generation were observed in mid- to late June, with an apparent lull in activity in early to mid-July during the pupal stage. The first locally-produced adults emerged in mid-July. Second local generation adults, considered to be the fall migrant generation, were observed in mid- to late August and commenced migration from mid-August through mid-September.

We observed monarch oviposition, eggs, larvae, and pupae on showy and narrow-leaved milkweeds in both states. Oviposition, eggs, and larvae were reported on swamp milkweed in Idaho. We did not observe immature stages or ovipositing on pallid milkweed, however, S. McKnight (pers. comm.) of the Xerces Society reported late-instar larvae on pallid milkweed at one location in southwest Idaho in 2017. At the spider milkweed locality in Franklin County, Idaho, we did not detect evidence of the species' use as a monarch host plant, but did observe extensive, possibly damaging herbivory of seed pods by a variety of insects, primarily small milkweed bugs (Lygaeus kalmia) (see **Supplementary Presentation S1** for photographs of milkweeds and monarch neonate life stages observed in the study area).

In both years of study, we encountered aggregations of large numbers (100s) of fresh adults at sites in southern Idaho. Adult

massings were reported from 28 July to 24 August on 7 stateand federally-managed natural areas located within modeled high-suitability milkweed habitat. We interpreted the early adult massings as synchronous enclosure events, and those in late-August likely migrating adults.

During Idaho fieldwork, 293 adult monarchs were tagged through the WSU monarch tagging program. Overall, sex ratios were male-biased, with males accounting for 63% of tagged monarchs in 2016 (n = 63), and 60% in 2017 (n = 113). No recoveries of our Idaho-tagged monarchs were reported (James et al., 2018). We sampled 170 adult monarchs from 6 WMAs for OE, of which 5 (3%) tested positive. All OE positive adults were collected at the Roswell Habitat Area of Fort Boise WMA (Canyon County) on 28 July 2017 (n = 4) and 9 September 2017 (n = 1). Prevalence of OE in 89 adult monarchs sampled at Roswell was 5.6% (84 negative; 5 positive), at the low range of OE infection rates (5–30%) estimated in the western monarch population (Altizer and de Roode, 2015).

We compiled 448 observations of breeding season nectaring use by adult monarchs on 32 plant species (**Supplementary Table S2**). Asclepias spp. (showy, swamp, and narrow-leaved milkweeds) were the primary nectar plants used by monarchs in the study area (53%; n = 237). Of 29 non-Asclepias nectar plants identified, 16 (55%) are native to the study area and accounted for 27% of nectaring observations (n = 120). Of native plants, common sunflower (Helianthus annuus) and goldenrods (Solidago spp., Euthania spp.) were frequently visited by nectaring monarchs and had particular value as late-season forage for migratory generation monarchs and other native pollinators, notably bumble bees (Bombus spp.). We documented monarchs nectaring from 13 non-native plants, with 20% of total observations (n = 91) on 3 species: bull thistle (Cirsium vulgare) in Idaho, purple loosestrife (Lythrum salicaria) in Washington, and Canada thistle (Cirsium arvense) in both states. Diversity of nectar plants used by monarchs was highest in August, although, this may simply reflect peak abundance of monarchs on the landscape or monarch opportunism as primary nectar plants (i.e., showy milkweed) senesce.

Occasional observations of roosting monarchs were reported during our study. In Washington, adults were observed

day-roosting in Russian olive (Elaeagnus angustifolia) and other small trees and shrubs on multiple days and sites when ambient temperatures exceeded 32◦C. At several survey sites in Idaho, adult monarchs were observed day-roosting in herbaceous vegetation, including hardstem bulrush, broadleaf cattail, basin wild rye (Leymus cinereus), and Nuttall's sunflower (Helianthus nuttallii), though trees (native and non-native) were available in close proximity. An observation of an adult monarch day-roosting in Wyoming big sagebrush (Artemesia tridentata wyomingensi) at an eastern Idaho locale was notable in that timing was late in the season (14 September 2017) and use of sagebrush as a day-roost has not been previously reported for the study area. We observed night-roosting monarchs on a few occasions in late August-early September, consisting of 1–3 monarchs roosting at a height of 2–3 m in Russian olive.

# Characteristics of Monarch Breeding Habitat

Habitat types were determined for 3,429 milkweed occurrence records (2,875 in Idaho; 554 in Washington) collected within the study area in 2016–2017. We identified 6 primary habitat types (**Figure 6**) after excluding habitat types with ≤2% of all milkweed records. Types occurring with highest relative frequency were grassland-herbaceous and emergent herbaceous wetland habitats, followed by deciduous forest, shrub-scrub, irrigation canal, and woody wetland habitats. Of these types, native or naturalized grassland-wetland habitats managed as IDFG and WDFW WMAs, USFWS National Wildlife Refuges, and U.S. Forest Service (USFS) National Grasslands supported the largest, most contiguous, and highest density milkweed stands. Cottonwood (Populus spp.) riparian forests within grassland-wetland habitats also supported abundant stands of showy milkweed, as did agricultural lands in the Columbia Basin of Washington and Snake River Plain of Idaho. These irrigation landscapes contain extensive networks of canals used to deliver water for crop irrigation, including sites with regular accumulations of runoff water. Combined season-long availability of water and intermittent disturbance from canal maintenance, mowing, or tilling facilitates rapid colonization by showy milkweed (**Figure 7**). Notable differences between states were higher occurrence of grassland-herbaceous and irrigation canal habitats in Washington, and higher occurrence of emergent herbaceous wetland and deciduous forest habitats in Idaho.

Although shrub-scrub habitat was identified at about 18% of milkweed patches in our study area, it was infrequently the sole or dominant habitat type, and frequently co-occurred with grassland, riparian, and wetland habitat types. Notably, 32 of 163 (20%) randomly selected GRTS cells surveyed in Idaho with sagebrush-dominant shrub-scrub as the primary habitat did not contain A. speciosa or other milkweeds. Milkweed was rarely found in cultivated cropland, bare rock-gravel, developed, pasture-hay, garden, mixed forest, or evergreen forest habitat types.

Milkweed patch area (m<sup>2</sup> ) was reported for 1,232 milkweed occurrence records (n = 653 in Idaho; n = 579 in Washington) and aggregated into 4 size classes (≤400, >400–4,000, >4,000– 8,000, >8,000 m<sup>2</sup> ; **Figure 8**). Nearly half (47%) of total milkweed patches were in the smallest size class (≤400 m<sup>2</sup> ) and 39% fell within the next largest size class (>400–4,000 m<sup>2</sup> ). The balance of milkweed patches (14%) was about equally aggregated between the 2 largest size classes. Milkweed patch areas were fairly consistent between states, with the exception of milkweed patches >8,000 m<sup>2</sup> , which although rare in both states, were more frequently reported in Idaho (n = 69) than in Washington (n = 16).

Milkweed species native to the study area are short-lived herbaceous perennials that senesce in late summer-early fall and are winter dormant, with new stems emerging in spring from established root systems. We sought to better describe milkweed phenology in the study area, not only to understand the changing availability of host plant resources for breeding monarchs, but to inform habitat management windows with least risk and greatest benefit potential to monarchs, and guide selection of milkweed species for restoration project success (Buisson et al., 2016). We found pallid and spider milkweeds to be the earliest phenology milkweed species, emerging in April, flowering in mid-May, fruiting in mid-June, and senescing by mid- to late July. Showy and narrowleaved milkweed foliage began developing in early to mid-May, flowered over a prolonged period from late May to July, fruited in July to August, and senesced in September. However, milkweed phenology is also plastic and capable of response to environmental conditions; in 2017 we observed a narrow-leaved milkweed population (Kootenai County, Idaho) delayed ∼5 weeks due to submergence by high flows in the Spokane River (**Supplementary Presentation S1**). Swamp milkweed exhibited the latest phenology of milkweeds in our study area, emerging in early to mid-June, with prolonged flowering in July-August, fruiting in August-September, and senescing in September-October. We documented multiple areas with sympatric milkweed species, but none more extensive and productive as monarch natal habitat than mixed showy and swamp milkweed stands in Idaho's Snake River Plain. Over the 2 years of our study, we observed phenological synchrony between adult monarch arrival (i.e., egg-laying) and bud burst/young expanding foliage of showy milkweed. Research on eastern monarchs has shown that phenological asynchrony with milkweed host plants can lead to high mortality of early instars and increased predation or poor nutrition in later instars (Zalucki et al., 2011), though how this mortality contributes to overall population dynamics remains unclear (Despland, 2017).

Washington, 2016.

# Management Activities and Threats in Monarch Habitats

We collected data on management activities and threats at milkweed sites where these factors were discernable. Management activities were identified for 644 milkweed occurrences in our study area (n = 270 in Idaho; n = 374 in Washington). Threat factors were recorded for 808 milkweed occurrences (n = 321 in Idaho; n = 487 in Washington) (**Figure 9**). Most management activity categories had corresponding threat categories (e.g., herbicide application was both a management activity and threat category). A commonly encountered management activity, indirect watering, described milkweed patches receiving supplemental watering from agricultural runoff, sprinkler systems, irrigation canals, roadside ditches, or agricultural ponds. Indirect watering was observed at 13% of Idaho milkweed occurrences (n = 34) and 49% of Washington occurrences (n = 183). All other reported management activity categories were also represented by threat categories.

We documented 1,625 threats at 808 milkweed patches. Primary threats were invasive plant species (n = 443), herbicide application (n = 348), and mowing (n = 259). Of primary threats, invasive plant species was more prevalent in Washington (64%) than in Idaho (40%), whereas herbicide application was considerably more frequent in Idaho (62%) than Washington (31%). Mowing occurred at 32% of milkweed occurrences in Washington and Idaho, usually for control of road-edge vegetation, or for harvest of hay or other crops. Secondary threats were recreational disturbance (n = 142), livestock grazing (n = 116), insecticide application (n = 88), flooding regimes (i.e., loss of floodplain function) (n = 68), and wildfire (n = 64). Disturbance of milkweed from recreation, typically from trampling or off-road vehicle use, was reported at 18% of milkweed patches across the study area. Livestock grazing occurred at 14% of sites, but was more common in Washington (20%) than Idaho (6%). Though livestock rarely consumed milkweed, they often trampled milkweed plants and grazed available nectar plants, thereby reducing or eliminating those resources. Washington reported notably higher frequencies of insecticide application, flooding regimes, and wildfire threats than Idaho. Other threat categories, such as irrigation canal maintenance, vegetation encroachment, and development, were less common in the study area (range 18–42 records).

# DISCUSSION

# Milkweed and Monarch Distributions

Our study presents the first statewide inventories of milkweeds and breeding monarch distributions within the western monarch range. These data signify a major advancement over prior understanding of monarch breeding habitat extent and characteristics in Idaho and Washington, states at the northern tier of western monarch range. Our study documented much broader distribution of milkweed host plants and breeding monarchs than previously hypothesized based on suitable thermal regimes for monarch reproduction (Stevens and Frey, 2010). We increased our pre-project dataset of ∼700 milkweed records by >400% and documented a first Washington state record for swamp milkweed, and first county records for showy milkweed (13 counties), swamp milkweed (5 counties), and narrow-leaved milkweed (5 counties). Of the 5 milkweed species documented in the study area, showy milkweed was most ubiquitous and wide-ranging owing to its adaptation to a wide range of soil types, moisture regimes, and disturbance

agents (Stevens, 2000). Our surveys documented first records for breeding monarchs in 37 counties within the study area. Although we did not demonstrate range expansion for the western monarch population, we did produce a high-resolution baseline distribution for breeding monarchs scaled to the Idaho-Washington region.

Geographic and elevation ranges for the 40+ species of western milkweeds are highly variable, owing to the integrated effects of latitude and altitude, and their influence on temperature, precipitation, humidity, heat, and illumination (Xu et al., 2017). The 5 milkweed species native to our study area have geographic ranges extending a few 100 to 2,000+ km south of Idaho and Washington and occur at elevations up to ∼2,700 m in southwestern states (e.g., Arizona, Nevada, Utah). The elevation range of the 5 milkweed species in our study area (0–1,686 m) is at the low to mid-range for these species across the western states, indicating that relatively high latitudes (42–49◦N) and mountain habitats may be important determinants of milkweed distributions in Idaho and Washington. This is born out in the Western Monarch and Milkweed Habitat Suitability Modeling Project, Version 2 (Dilts et al., 2018), which ascribes low habitat suitability for milkweeds and breeding monarchs in mountainous regions of our study area. By refining elevational distributions of milkweeds (and by extension, breeding monarchs) in our study area, monarch conservation work can be appropriately targeted in areas with highest potential for success.

#### Monarch Life History

Although our research was not designed to examine detailed breeding phenology of monarchs, our surveys refined a temporal window for monarch breeding and life stages in Idaho and Washington (**Figure 5**; **Supplementary Table S1**) where few spatial or temporal data previously existed. An important caveat to these results is sampling efforts to identify monarch life stages were not equally distributed in time and space across the study area and thus have limitations regarding spatiotemporal precision. In addition, a larger sample of monarch life stage observations from Idaho may impart a geographic bias to our phenology dataset. With these caveats in mind, we sampled continuously during monarch breeding seasons across a heterogeneous geography within and between states and found strong correspondence of dates for adult arrival (early June), peak egg observation (mid-June), and peak adult observation (late July). James (2016) reported adult arrival time and systematically collected adult numbers over 3 breeding seasons (2013–2015) at a single central Washington study site. Our findings of regional adult arrival time in early June were harmonious with James's observations, and our observations of peak adult numbers in late-July were consistent with 2 of his 3 study years. Further work is needed to determine finer spatio-temporal resolution of monarch breeding phenology and identify effective windows to minimize risk to monarchs in Idaho and Washington.

Our observations of robust monarch production at numerous milkweed-abundant sites in Idaho and eastern Washington counter previous studies suggesting minor recruitment of Pacific Northwest migrant-generation monarchs to the western population (Stevens and Frey, 2010; Pyle, 2015). Rather, our study builds upon evidence of substantial natal contributions of monarchs from interior western states (i.e., "northern inland range") to the California overwintering population (Yang et al., 2015). Furthermore, major pulses of monarch production documented in our study were consistent with similarly large populations of monarchs observed at a milkweed-rich study site in Central Washington, which James (2016) stated were "remarkable and challenge our concepts of summer breeding of D. plexippus in the Pacific Northwest."

Whereas monarch larvae are specialists with respect to host plant use, adult monarchs are nectar generalists, feeding on a wide assortment of flowering plants (Brower et al., 2006). We found nectar species to be limited in quantity and richness in the study area, particularly in late summer. Inadequate nectar resources can reduce fecundity and lipid accumulations needed by monarchs to fuel the fall migration, overwintering period, and subsequent northward flight in spring (Brower, 1985; Alonso-Mejia et al., 1997; Brower et al., 2006). Although we disproportionately sampled predicted milkweed sites, 53% of nectar uses by monarchs were on milkweeds; slightly <60% reported by Xerces (2018) for the 11 western states. Our results underscore the importance of milkweeds not only as monarch host plants but as extended-season nectar resources for adults. Monarchs also nectared on non-native, invasive species, such as Canada and bull thistles and purple loosestrife, and in some locations these non-natives were the only nectar resources available after milkweeds flowered. James (2016) found purple loosestrife to be a principal late-season nectar resource at a monarch breeding site in eastern Washington and a key factor in the site's suitability as monarch natal habitat. In some cases, invasive species may provide significant nectar sources for migrating monarchs (Brower et al., 2006).

Although targeted surveys for roosting monarchs were not part of our study, we recorded roosting behavior when incidentally observed in the field. Previous studies and observations from eastern Washington (Pyle, 1999; James, 2016), Oregon (Cheryl Schultz, pers. comm.), Utah (Utah Lepidopterists' Society, pers. comm.), and Idaho (Rose Lehman, pers. comm.) suggest that tree and shrub roosting structure may be important to western monarch breeding and migration ecology. The dominant tree species noted in these reports is the introduced Russian olive, variably used by adult monarchs for daytime shade (James, 2016) and nighttime roosts (Pyle, 1999). Russian olive is a Class C noxious weed in Washington and considered an invasive plant species in Idaho and several other western states for its propensity to displace and hinder recruitment of native climax species in many waterways of the interior western U.S. (Lesica and Miles, 1999, 2001; Pearce and Smith, 2001). Given the plethora of negative ecological impacts linked to Russian olive, it is often targeted for control efforts by land managers. Research is needed to address whether roosting habitat is an essential component in western monarch breeding range, particularly if Russian olive control could result in unintentional but potentially harmful consequences for breeding and migrating monarchs.

# Characteristics of Monarch Breeding Habitat

Although monarch breeding habitat is delimited by distributions of its obligate milkweed host plants, not all milkweed sites support breeding monarchs (Grant et al., 2018; Pitman et al., 2018). The relatively coarse scale of our study did not allow inferences about microsite attributes or preferred spatial configurations of habitat at monarch natal habitats. However, we did identify some common key characteristics of monarch breeding habitat in the study area. Highly productive monarch breeding habitats were moist-soil sites within a matrix of grasslands, wetlands, deciduous forest, and shrub-steppe habitats supporting large, contiguous, and high-density milkweed stands. These habitats were most often located on public lands managed for wildlife conservation or multiple uses. Common to these sites were presence of naturally-occurring or anthropogenicsourced surface or ground water that resulted in increased soil moisture relative to surrounding landscapes, and maximum daytime temperatures agreeable with the thermal optimum for monarch life stage development (∼28◦C) (Zalucki and Kitching, 1982; York and Oberhauser, 2002). Most of the WMAs and refuges are located within irrigation landscapes and directly or indirectly rely on water delivery systems allocated by legal water right to maintain habitat conditions. WMAs and refuges in Idaho occur in the Snake River Plain where ∼85% of total state water withdrawals support irrigated agriculture (Murray, 2018). Similarly, important monarch breeding habitats in the Columbia Basin of Washington spatially overlap the state's most concentrated region of irrigated cropland, which uses ∼80% of state water withdrawals (McLain et al., 2017). The waterscapes of southern Idaho and eastern Washington are presently at risk of increased water deficits due to population growth, landuse change, and changes in cropping systems and commodities grown (Ryu et al., 2012; Hall et al., 2016; Kliskey et al., 2019). Projected hydroclimatic changes across the region in the next 50 years include a substantial warming (Rupp et al., 2017), decreased snowpack, shorter snow accumulation season, earlier snowmelt, and increased evapotranspiration leading to likely water and soil moisture deficits during summer months (Vano et al., 2015; Gergel et al., 2017). Such scenarios point to the inherent vulnerability of monarch breeding habitats in both states. The persistence and viability of these habitats will rely on adaptive, long-term water plans that recognize and value monarch and other wildlife habitat to proactively address the region's complex water challenges (Kliskey et al., 2019).

The Idaho GRTS survey helped to identify milkweed habitat suitability across a range of habitat types and indicated that sagebrush-steppe habitats in Idaho are generally unsuitable for showy milkweed. This result was unsurprising given seasonal aridity of sagebrush-steppe habitats and showy milkweed affinity for moist-soil sites. A key assumption of this finding is the dataset underlying the showy milkweed habitat suitability model (USFWS and Xerces, 2016) used in our GRTS survey framework was adequately robust in its Phase I iteration. In Washington, shrub-scrub habitats consist of shrub-steppe plant communities dominated by big sagebrush (A. tridentata), bitterbrush (Purshia tridentata), and rabbitbrush (Ericameria nauseosa, Chrysothamnus viscidiflorus) with perennial bunchgrass understory. Portions of Columbia Plateau shrub-steppe are underlain by deep alluvial and eolian sand deposits formed during Pleistocene deglaciation (Hallock et al., 2007). Washington's most productive milkweed stands occur in these mosaics of deep, sandy soils where supplemental irrigation water and resulting raised water tables, as well as naturally occurring lakes, ponds, and rivers, have facilitated establishment of herbaceous and woody vegetation.

While we located several large, high density milkweed patches during our surveys, over half of patches contained relatively few individuals (i.e., 1–50 plants). Our findings were consistent with Pyle (2015) and James (2016) who described milkweed distributions in Pacific Northwest states as patchy and low density. Whether this type of distribution pattern is characteristic across the West and how such patterns affect carrying capacity of monarch breeding habitat in terms of available milkweed resource or reproductive success remains unclear. Monarch females in eastern North America sought out smaller milkweed patches in agricultural, roadside, and non-agricultural areas and oviposited more heavily there (Zalucki, 1981; Zalucki and Kitching, 1982). These results were variably attributed to use of fertilizer, ability of females to detect milkweed in monocultures, and higher quality plants due to reduced competition for resources (Pleasants and Oberhauser, 2012; Pitman et al., 2018). We are uncertain whether a similar pattern would hold true for monarchs in western ecosystems with a different complement of milkweed species, cropping systems, and precipitation patterns.

In southwest Idaho, a key attribute of productive monarch habitat was sympatric occurrence of showy and swamp milkweeds, typically at the patch level. The mix of milkweed species with asynchronous phenologies extended the vegetative stage required for egg, larvae, and pupae development, and the bloom period for nectaring monarchs and other pollinator taxa. Mixed milkweed sites typically had dense, complex vegetative structure with abundant cover for immature monarch stages. Whether this structure or combination of milkweed species influences monarch vital rates (i.e., survival, individual growth, recruitment) are research questions meriting investigation in western monarchs.

# Management Activities and Threats in Monarch Habitats

Though the western U.S. abounds with large natural areas and wilderness, we found milkweeds and monarchs in Idaho and Washington persist primarily in landscapes impacted by high human activity. Threats (e.g., factors that jeopardize persistence of milkweed and monarchs) were commonly observed during our study and most often directly or indirectly human-induced. In addition to sharing threats faced by eastern milkweeds and breeding monarch populations, the butterfly-host system in the West experiences unique threats, likely because occurrence is often restricted to moist-soil conditions within an otherwise arid landscape. Our research is the first assessment and body of data on milkweed and breeding monarch threats collected in western states.

Non-native and invasive grasses, shrubs, and trees assessed as likely milkweed competitors were documented at 55% of sites. Though not all invasive plants were identified to species, Russian olive and perennial and annual grasses (including cheatgrass [Bromus tectorum]) were commonly encountered. The abundance of invasive species at surveyed sites was likely a byproduct of milkweed occurrence in frequently disturbed and moist-soil habitats prone to invasion by non-native plants. In some situations, invasive plants benefit monarchs by providing essential habitat features (e.g., non-native thistles and purple loosestrife providing nectar sources). In some areas Russian olive provided day and night roosting sites and may benefit milkweed and monarchs by creating limited, partially-shaded microhabitats (Pyle, 1999; James, 2016). Because control of invasive plants is desirable from an ecosystem standpoint, herbicide application, mowing, or other practices targeting invasives have the potential to collaterally destroy host and nectar plants and immature monarch life stages (Xerces, 2018).

A primary threat to milkweeds and monarchs documented in both states was herbicide use (43% of all milkweed sites; **Figures 9**, **10**). Herbicide use was likely more widespread, as detection was reliant on survey timing relative to applications. We regularly encountered evidence of herbicide use in milkweed habitat and direct impacts to milkweed plants. Discussions with several land managers and landowners confirmed milkweed is a situational target for control and eradication on some lands. However, much observed herbicide use was conducted for general vegetation control, along roadsides, railroad rightsof-way (ROW), parking areas, etc., and not specifically to affect milkweed. Monarchs rely on milkweed and floral nectar sources, and herbicide applications affecting these resources essentially destroys breeding habitat, even when it may not be effective in controlling or eradicating targeted plant species. Widespread use of herbicide to target or collaterally damage milkweeds illustrates a prevailing regional perspective that these native plants are considered "weeds."

We documented insecticide application at 18% of Washington milkweed patches, but less frequently in Idaho. Primary insecticide treatment observed in both states was for adult mosquito control. In Washington, extensive stands of milkweed fall within regions regularly treated with mosquito adulticides by local mosquito control districts, including on lands managed by WDFW and the U.S. Bureau of Reclamation near Moses Lake (Grant County Mosquito Control District #1., 2015). In Idaho, insecticide application to control adult mosquitos occurs in a milkweed- and monarch-rich area within the Boise River Greenbelt. Mosquito adulticides, including pyrethrin- or permethrin-based pesticides used in Washington, were found toxic to butterfly larvae and adults (Hoang et al., 2011), and specifically monarchs (Oberhauser et al., 2006).

Showy milkweed was commonly found colonizing areas with indirect supplemental watering, including transportation ROW and irrigation waterway edges. Impervious surfaces of roadways serve to harvest and channel rainwater to roadside verges where plant growth is often profuse (Wojcik and Buchmann, 2012). Likewise, season-long availability of irrigation water can produce a hedgerow effect of milkweed along irrigation canals (**Figure 7**). In Idaho, 13% of milkweed patches received supplemental water from paved road and agricultural sources (i.e., paved roads, roadside ditches, sprinkler irrigation, irrigation canals, irrigation runoff, and agricultural ponds). Nearly one-half of milkweed patches in Washington received supplemental watering from

FIGURE 10 | Herbicide use targeting milkweed observed in (A) Lemhi County, Idaho, and (B) Franklin County, Washington. Applications occurred in July during peak monarch breeding activity in the study area.

these sources. These locations are within zones of high human activity maintained for user safety, visibility, accessibility, and, in the case of irrigation systems, efficient water delivery, thereby making milkweeds and monarchs occurring in these areas inherently vulnerable to loss and degradation.

Wildfire and human-caused fire frequency, size, and intensity has increased throughout western states, including Washington and Idaho (Abatzoglou and Williams, 2016). Monarchs and milkweeds can be directly threatened by fires during breeding season, and wildfire smoke may be an additive stressor to fallmigrating western monarchs (Pelton et al., 2016). In 2016, wild and human-caused fires burned milkweed habitat occupied by monarchs on 2 WDFW Wildlife Areas (Lower Crab Creek and Sinlahekin). Anthropogenic climate change is expected to continue driving increased wildfire activity in the West while fuels remain abundant on the landscape (Abatzoglou and Williams, 2016).

Comparison of threats to monarchs and their habitat in the arid western states of Idaho and eastern Washington to those in the Midwest shows commonalities and differences. Herbicide and insecticide use have been implicated in the loss of milkweed and monarchs in the Midwest (Pleasants and Oberhauser, 2012; Krischik et al., 2015). Threats related to herbicide use are being addressed in part by growing participation of state and county Departments of Transportation in Integrated Roadside Vegetation Management programs. Such programs recognize ROW landscapes offer important and overlooked conservation opportunities for monarchs and other pollinators (Hopwood et al., 2015; Xerces, 2018).

#### Management Recommendations and Research Needs

Key results of our study lead us to recommend management actions to abate threats to monarchs in Idaho and Washington. Paramount to monarch persistence in our states is continued protection and beneficial management of known monarch breeding habitats. Our study identified many of these high quality habitats, but other similarly important natal areas likely exist in Idaho and Washington and should be inventoried and assessed for long-term protection.

Managing quality monarch habitat often requires addressing invasive non-native plants and noxious weeds as part of public policy and ecosystem health directives. This can place land managers in a difficult position of navigating between conflicting resource management objectives (e.g., invasive plant nectar availability vs. ecosystem integrity). Where those conflicts exist, we suggest approaches that promote control vs. eradication of invasive plants providing nectar and roosting benefits to breeding monarchs and other pollinators.

The substantial level of herbicide use at milkweed sites in our study area highlights a pressing need for expanded communication with key sectors within our states (e.g., transportation departments, utility companies, farmers, ranchers, irrigation districts, private landowners) to reframe pervasive negative perceptions about native milkweeds and encourage management practices that conserve monarch habitat. Potential actions could entail development of effective messaging for different audiences, promoting financial incentive and technical assistance programs, and developing Integrated Vegetation Management programs that achieve more cost-effective and environmentally-sustainable management of undesirable plants while considering monarch needs.

Our study to delineate milkweed and breeding monarch distributions in Idaho and Washington fortuitously interfaced with development of the Western Monarch and Milkweed Habitat Suitability Modeling Project, Version 2 (Dilts et al., 2018). We recommend regular updating of these west-wide models as well as development of new models at finer spatial scales. Efforts to enhance and restore monarch habitat should consult both milkweed and monarch breeding models to assess suitability of any site. Such analyses are particularly relevant for USFS and Bureau of Land Management (BLM) lands in sagebrush-steppe and forested land cover types of our study area. Sagebrush-steppe habitats in Idaho (with the exception of Curlew National Grassland, Oneida and Power counties) and forested habitats in Idaho and Washington exhibit low suitability as milkweed or breeding monarch habitats. Instead, we suggest these agencies focus conservation work on protection and restoration of migratory habitat and connectivity. Monarch migration in the West is tied to riparian corridors, which provide crucial nectaring habitats, particularly in years of drought (Dingle et al., 2005; Brower et al., 2006, 2015). Management directed to reducing threats (**Figure 9**) to these spatially limited, but highly productive riparian communities could improve quantity and richness of floral nectar sources and roosting structure for spring and fall migrating monarchs.

Finally, we provide suggestions for future research to improve our knowledge base of the western monarch life cycle, breeding habitat requirements, and threat factors:


Based on the new body of knowledge acquired from our study and evidence of continuing declines of the western monarch overwintering population, we recommend the monarch butterfly retain its status as "Species of Greatest Conservation Need" (SGCN) in Idaho and Washington. Moreover, we encourage our western state partners to evaluate monarchs for SGCN designation and consider similar survey approaches to address regional data gaps with the goal of conserving monarch habitats and migratory connectivity in the West.

### ETHICS STATEMENT

Though butterfly taxa are not regulated under animal ethics and welfare guidelines, we observed voluntary protocols parsimonious with ethical treatment of monarchs in the field.

# AUTHOR CONTRIBUTIONS

BW and AP conceptualized the study, coordinated and conducted field data collection, conducted data management and analyses, and wrote the initial drafts of the manuscript. LS designed the GRTS survey framework, developed ArcGIS products, and conducted data management and analyses. All authors contributed to manuscript revisions and read and approved the final manuscript.

# FUNDING

Funding was provided by the USFWS State Wildlife Grant Program (grant number F16AP00020), Idaho Department of Fish and Game Nongame Trust Fund, and Washington Personalized License Plate Fund.

#### ACKNOWLEDGMENTS

We are grateful for collegiality and partnership with S. Jepsen, E. Pelton, C. Fallon, S. McKnight, and S. Hoffman Black of the Xerces Society for Invertebrate Conservation. We thank C. Schultz, D. Perkins, D. James, T. Dilts, and M. Forister for their research collaborations. We thank our IDFG and WDFW colleagues and field crews, numerous citizen scientists, and Spokane Chapter of the Washington Butterfly Association for their contributions in data collection. We are grateful to J. Jenkerson for WDFW database management, D. Smith of IDFG for grant facilitation, and T. Keegan and two reviewers for insightful discussion and comments on the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00172/full#supplementary-material

# REFERENCES


Bailey, R. G. (1976). Ecoregions of the United States (map). Ogden, UT: USDA Forest Service, Intermountain Region. 1:7,500,000.


Available online at: https://xerces.org/wp-content/uploads/2018/04/18-009\_ 01-Monarch\_BMPs\_Final\_Web.pdf


**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 Waterbury, Potter and Svancara. 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.

# Host Plants and Climate Structure Habitat Associations of the Western Monarch Butterfly

Thomas E. Dilts <sup>1</sup> , Madeline O. Steele<sup>2</sup> , Joseph D. Engler <sup>2</sup> , Emma M. Pelton<sup>3</sup> , Sarina J. Jepsen<sup>3</sup> , Stephanie J. McKnight <sup>3</sup> , Ashley R. Taylor <sup>3</sup> , Candace E. Fallon<sup>3</sup> , Scott H. Black <sup>3</sup> , Elizabeth E. Cruz <sup>2</sup> , Daniel R. Craver <sup>2</sup> and Matthew L. Forister <sup>4</sup> \*

<sup>1</sup> Department of Natural Resources and Environmental Science, University of Nevada Reno, Reno, NV, United States, <sup>2</sup> U. S. Fish and Wildlife Service, Portland, OR, United States, <sup>3</sup> Xerces Society for Invertebrate Conservation, Portland, OR, United States, <sup>4</sup> Program in Ecology, Evolution, and Conservation Biology, Department of Biology, University of Nevada Reno, Reno, NV, United States

#### Edited by:

Jay E. Diffendorfer, United States Geological Survey, United States

#### Reviewed by:

Vijay Barve, Florida Museum of Natural History, United States Attila D. Sándor, University of Agricultural Sciences and Veterinary Medicine of Cluj-Napoca, Romania Xiao Feng, University of Arizona, United States

> \*Correspondence: Matthew L. Forister mforister@unr.edu

#### Specialty section:

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

Received: 05 December 2018 Accepted: 08 May 2019 Published: 29 May 2019

#### Citation:

Dilts TE, Steele MO, Engler JD, Pelton EM, Jepsen SJ, McKnight SJ, Taylor AR, Fallon CE, Black SH, Cruz EE, Craver DR and Forister ML (2019) Host Plants and Climate Structure Habitat Associations of the Western Monarch Butterfly. Front. Ecol. Evol. 7:188. doi: 10.3389/fevo.2019.00188 The monarch butterfly is one of the most easily recognized and frequently studied insects in the world, and has recently come into the spotlight of public attention and conservation concern because of declining numbers of individuals associated with both the eastern and western migrations. Historically, the larger eastern migration has received the most scientific attention, but this has been changing in recent years, and here we report the largest-ever attempt to map and characterize non-overwintering habitat for the western monarch butterfly. Across the environmentally and topographically complex western landscape, we include 8,427 observations of adults and juvenile monarchs, as well as 20,696 records from 13 milkweed host plant species. We find high heterogeneity of suitable habitats across the geographic range, with extensive concentrations in the California floristic province in particular. We also find habitat suitability for the butterfly to be structured primarily by host plant habitat associations, which are in turn structured by a diverse suite of climatic variables. These results add to our knowledge of range and occupancy determinants for migratory species and provide a tool that can be used by conservation biologists and researchers interested in interactions among climate, hosts and host-specific animals, and by managers for prioritizing future conservation actions at regional to watershed scales.

Keywords: monarch butterfly, Danaus plexippus, milkweed, Asclepias, habitat, species distribution model, MAXENT, citizen-science

# INTRODUCTION

Species with exceptionally large geographic ranges are not often the focus of conservation and management attention, simply because large ranges typically encompass many populations or subpopulations, which buffer against stressors that are of concern in species with more restricted distributions (Brook et al., 2008). However, in at least some cases, migratory species are an exception in that they can be composed of a single population that covers a large geographic area, and thus may be uniquely exposed to stressors but without the resilience afforded by metapopulation structure (Drechsler et al., 2003). With such species, one of the central challenges is to understand habitat requirements or associations in different parts of the range so that conservation and management effort can be focused appropriately (Martin et al., 2007). Depending on requirements or preferences at different life history stages or at different times of the year, a widespread migratory species might have relatively simple or complex habitat associations. This may be of applied interest if the goal is to target certain habitats for protection (Guisan and Thuiller, 2005). The issue is of basic interest as well, as it bears on the structure and limits of broad geographic ranges. Here we examine habitat and host associations for the western migration of the monarch butterfly, Danaus plexippus plexippus.

The monarch is a relatively large butterfly in the family Nymphalidae that specializes on milkweed larval hosts in the genus Asclepias. The geographic range of the monarch encompasses a large portion of the North American continent and Caribbean islands, with disjunct populations in other areas including northern South America, Australia, the Iberian Peninsula, and islands in the Pacific. In North America, the monarch range is primarily composed of two subpopulations with mostly independent migratory phenomena that are genetically very similar (Zhan et al., 2014). In the eastern migration, monarchs overwinter in Mexico and move north during the summer, expanding across more than half of the continent. In the western migration, the overwintering sites are along the Pacific coast of North America, while the breeding ground is roughly characterized as west of the Rocky Mountains, but has not been described with greater precision [an unknown proportion of monarchs breed west of the Rocky Mountains but overwinter in Mexico (Morris et al., 2015)]. In recent years, both the eastern and western migrations have shown dramatic declines in numbers of individuals: in the East, this has been most evident at the overwintering grounds (Brower et al., 2012; Inamine et al., 2016; Semmens et al., 2016); in the West, the decline has been notable both in overwintering numbers (Schultz et al., 2017) and in observations during the breeding season (Espeset et al., 2016). The decline in the western population has been estimated at >95% since the 1980s (Schultz et al., 2017), which is arguably more severe than the magnitude of the decline in the larger eastern population (Semmens et al., 2016). These declines have triggered conservation concern (Oberhauser et al., 2017), and the monarch butterfly is currently under consideration for listing as threatened under the Endangered Species Act (79 FR 78775). Though smaller, the western population is significant to the species' viability as a whole in that it adds resiliency, or ability to withstand environmental stochasticity, and redundancy, or ability to withstand environmental catastrophes (Shaffer and Stein, 2000) by having healthy populations distributed across multiple heterogenous geographic regions. Monarchs have also experienced heightened interest within the general public in recent years and have long been one of the most widelyrecognized and appreciated insects (Gustafsson et al., 2015).

Despite an increase in professional and amateur interest in monarch butterflies in recent years, very little has been quantified with respect to habitat associations for the western migration, especially during the breeding season. Informally, the monarch is often considered a host specialist but a habitat generalist, since adults travel great distances and are expected to find host plants under a wide range of conditions, but this assumption has not been tested. One of the challenges of working with a widespread species in the western U.S. is the occupancy of large stretches of unpopulated, inaccessible, and infrequently studied areas. The state of Nevada, for example, has over 150 individual mountain ranges, many of which are not easily accessed by roads. Given that challenge, we have crowd sourced a diversity of records and monitoring efforts with the goal of developing a better understanding of habitat and climate associations and requirements for the western monarch and its major host plant species. Specifically, we address the following questions (and note that habitat associations include climate variables in all cases). (1) What are the environmental habitat associations of 13 western Asclepias species (**Figure 1**, **Table 1**) that are known to be larval hosts of monarch butterflies? These species include both widespread, abundant plants, and species with more restricted distributions for which we had sufficient data. (2) What are the non-overwintering habitat requirements of the western monarch? How do habitat requirements change when considering different life history stages? In particular, we consider adult observations as distinct from breeding records (observations of eggs and caterpillars), and we develop a model including all three life history stages (eggs, caterpillars, and adults). (3) Finally, we ask: how do non-overwintering habitat requirements or characteristics differ across the monarch models, from host plants to breeding records to adults and to all monarch observations combined? The first two questions are directly relevant to conservation and management of the monarch, while the last question is of general interest to the extent that it gives insight into the spatial scale of habitat specialization and generalization in a widespread species. In particular, it is of interest to know if monarch habitat associations are structured primarily by host plants alone or by host plants as well as climatic factors (either directly or indirectly through climatic effects on hosts). The distinction between those possibilities bears on our expectations as to how the monarch will respond to climate change and shifting abiotic conditions in the western U.S.

### MATERIALS AND METHODS

#### Occurrence Data and Study Extent

We used occurrence records from 11 U.S. states whose boundaries lay entirely or at least partially west of the Continental Divide (California, Oregon, Washington, Idaho, Nevada, Arizona, Utah, Montana, Wyoming, Colorado, and New Mexico) to create four models partitioned by life history stages, using only non-overwintering records (i.e., occurrences that were recorded April through October). The first model used all occurrence records regardless of life history stage (adults, eggs, and caterpillars). We refer to this as the "all" model. The second model used only those occurrence records that were recorded as "adult," thus omitting records that were only of larvae and eggs. The third model included only records that indicate breeding (eggs and caterpillars) which we refer to as the "breeding" model. The breeding records necessarily involve observations on milkweed hosts, while the adult records include both observations of adults in association with host plants and in other settings (e.g., nectaring on other flowering plants).

Finally, since tropical milkweed (Asclepias curassavica)—a nonnative species commonly planted in gardens, which persists yearround in areas with mild winters—has been implicated as a contributing factor for parasitism by the protozoan parasite, Ophryocystis elektroscirrha, and in disruption of reproductive behavior (Satterfield et al., 2016, 2018; Malcolm, 2018), we considered a fourth model of only breeding records that occur in areas >3.6 km from known occurrences of A. curassavica (based on a 3,600 m grid), which we refer to as our "non-tropical breeding" model.

In addition to modeling habitat suitability for the monarch butterfly, we created models for its host plants in the Asclepias (milkweed) genus (**Figure 1**). We selected 13 species with enough occurrence records to generate models with good predictive



In addition to the 13 milkweeds that were modeled there were 35 additional milkweed species in the database that were not included in the model due to small samples sizes which are shown in Supplementary Table 1.

performance based upon area-under-the-curve of the receiveroperator characteristic plot (validation AUC > 0.7). AUC is a threshold independent measure of model performance that has been commonly used in many fields including ecology (Fielding and Bell, 1997). Similar to the monarch models, we used data from the 11 western states as training and validation for milkweed models (more details on training and validation are given below).

Although the combined monarch and milkweed database contained 39,327 occurrence points, these data were collected from a variety of sources, including museum databases such as CalFlora and the Global Biodiversity Information Facility, citizen science efforts, such as the Southwest Monarch Study, Journey North, and iNaturalist, and targeted field sampling by a wide variety of state and federal agencies, nongovernmental organizations, and university groups, and were often collected opportunistically (see **Supplementary Table 1** for details on databases). Data used in this model and additional data can be accessed via the Western Monarch Milkweed Mapper website (www.monarchmilkweedmapper.org) (Western Monarch Milkweed Occurrence Database, 2018) and included records through January of 2017. To minimize the effects of sampling bias in our data we used geographic thinning, which has been shown to be one of the more effective approaches for reducing sampling bias in occurrence data (Kramer-Schadt et al., 2013; Fourcade et al., 2014). To implement geographic thinning, we first removed records with a known accuracy coarser than 270 m, leaving mostly data from the GPS era (starting in the mid 1990's), then we applied a 3,600 m grid over the study area and retained a single occurrence point closest to the centroid of each cell. The 3,600 m grid was identified by Steele et al. (2016) because it minimized spatial clustering of occurrence points that typically occur due to biased sampling (for example along roads or near urban areas) and was found to result in more generalizable models in that study. The resulting occurrence dataset contained 4,569 records among the 13 milkweed species and four categories of monarch models (all records, adult-only records, breeding records, and breeding records without A. curassavica; **Table 1**). We did not apply other methods of bias correction, such as target group sampling, because that approach assumes an equal probability of detection for all species (Ponder et al., 2001). We could not assume that observers would necessarily have observed or recorded any milkweed had it been present, nor could we be certain that evidence of monarchs and lack of milkweeds represents a true absence of host plants.

#### Environmental Covariates

Since the goal of our work was to assess the general characteristics of habitat associates between monarch butterflies, milkweeds, and the abiotic environment, we tested a wide range of environmental variables in the models that represented topographic, climatic, edaphic, hydrologic, and land use gradients (**Table 2**). Variables assessed in the model included actual evapotranspiration, climatic water deficit, number of degree days, annual precipitation, precipitation of the coldest season, precipitation seasonality, precipitation of the warmest season, temperature range, temperature seasonality, maximum temperature of the warmest season, soil bulk density, clay content, sand content, silt content, pH, aspect, slope, compound topographic index (topographic wetness index), distance to intermittent stream, distance to perennial stream, and land cover types. All variables are on a continuous scale except for land cover, which is a categorical variable, and it included classes such as urban, suburban, agriculture, shrubland, coniferous forest, and deciduous forest (further discussion of the land cover variables are in **Supplementary Materials—**additional methods and **Supplementary Table 3**). All covariates were resampled to a common resolution of 270 m. Details about the environmental covariates and how they were resampled to a single resolution for modeling are included in the **Supplementary Material** and the native resolution of each covariate is shown in **Table 2**.

#### Background Selection

Our habitat modeling approach builds off of the previous monarch habitat modeling effort by Steele et al. (2016) in that it takes a four step approach that includes optimal background selection, optimization of Maxent parameters (feature types and regularization), variable reduction, and finally map projection. We applied a presence-background habitat modeling approach known as maximum entropy (Maxent) to create relative habitat suitability models (Phillips et al., 2006). In contrast to presenceabsence modeling methods, presence-background approaches, such as Maxent, randomly select background points from a frame of pixels that may include both presence and absence locations. TABLE 2 | Environmental covariates used in modeling, their data source, and original cell size of the raster data. Further details about the variables are in Supplementary Table 2.


We selected restricted background areas on a species-specific basis for each of the milkweed species (VanDerWal et al., 2009), with the exception of Asclepias speciosa, for which we used the entire 11 western states. Likewise, all four Danaus plexippus plexippus models used the entire 11 states. For the remaining milkweed species we used the approach presented by Iturbide et al. (2015), in which species occurrence points are buffered by increasingly large buffers, models are run using default Maxent parameters, and selected based upon their inflection curves. We calculated a Michaelis-Menten function (Iturbide et al., 2015) to identify the saturation point and selected the first model that was within the 95% confidence interval for that point. Buffer distances ranged from 175 to 300 km depending upon the species and are shown in more detail in **Supplementary Table 4**.

#### Optimization of Regularization and Feature Types in Maxent (Model Selection)

Several recent studies recommend species-specific tuning of the default settings for the regularization parameter and feature types (i.e., linear, quadratic, product, threshold, and hinge) (Anderson and Gonzalez, 2011; Warren and Seifert, 2011; Radosavljevic and Anderson, 2014). To perform species-specific tuning for the 17 milkweed and monarch models, we tested a combination of five regularization levels (1–5) and five combinations of feature types (linear, linear + quadratic, linear + quadratic + product, linear + quadratic + product + threshold, hinge), representing a gradient of models ranging from simple to complex. Determining the best model that optimizes both fit and generalizability is an ongoing area of research in species distribution modeling and no clear consensus has yet emerged as to how to create an optimal model (Merow et al., 2013; Radosavljevic and Anderson, 2014; Warren et al., 2014). We derived validation AUC withholding 50% of records as a measure of model fit and the difference between training and validation AUC (AUCdiff) as a measure of model overfit (Warren and Seifert, 2011). Combining these two approaches, we define in this paper a new metric that we term "penalized AUC" (pAUC), which combines validation AUC and AUCdiff order to identify models that maximize model fit while minimizing overfit. We calculate

pAUC = validation AUC − (AUCdiff) or

pAUC = validation AUC − (AUCtraining − validation AUC).

For each species, we chose the model with the highest pAUC and identified the combination of regularization values and feature types that produced that model (**Supplementary Table 2**).

#### Variable Reduction

Although Maxent can accommodate a large number of predictor variables, to aid in the interpretation of the models and to reduce multicollinearity we used a variable reduction approach with the following steps. First, models were run using all covariates present (25 for the milkweed models and 38 for the monarch models; the latter have more variables because they include milkweeds as covariates). Maxent software allows variables to be ranked based upon a measure that the authors call permutation importance. Second, we ranked variables based on their permutation importance and variables with <3% contribution were removed from the model. Permutation importance, a standard output from Maxent software, is calculated by randomly permuting the values of each predictor variable among the training points and measuring the decrease in training AUC (Phillips et al., 2006). Three percent was chosen as a threshold for dropping variables because it allowed us to assess the impact that removing a handful of variables at a time has on the importance of the other variables. Third, variables that remained but had a TABLE 3 | Number of samples used in model building, training AUC, validation AUC, AUCdiff, and pAUC for the final habitat models.


correlation of > 0.7 with another variable were removed keeping the variable with the higher permutated importance value. This was repeated until all variables had a minimum contribution of 10% or greater.

#### Model Evaluation and Mapping

After selecting a parsimonious set of covariates for each model, we performed cross validation using 25 replicates withholding 20% of the sample points for validation in each iteration. Creating replicate maps allowed us to calculate the standard deviation of the predictive maps which we used as a measure of model uncertainty. We projected habitat models to the seven westernmost contiguous states of the United States using raster covariates at 270 m resolution and the logistic transformation in Maxent software to scale the suitability values from 0 to 1. While data for the states of Colorado, Montana, New Mexico, and Wyoming were used for model-building, limited fieldwork and outreach in those states resulted in few occurrences, thus we chose to limit the final predictive maps to California, Oregon, Washington, Idaho, Nevada, Utah, and Arizona. Finally, we used the equal sensitivity and specificity method to divide the continuous relative habitat suitability into binary maps of suitable and unsuitable areas for each species and monarch life history stage (**Supplementary Table 7**). These areas are depicted as gray lines on the full page supplemental maps (**Supplementary Figure 2**).

# RESULTS

#### Model Performance

Thirteen milkweed species and four life history stages of monarch butterflies produced models with validation AUC > 0.7 (**Table 3**). Monarch models ranged in validation AUC from 0.81 to 0.87. Milkweed models had validation AUC values ranging from 0.75 to 0.91 with A. eriocarpa, A. tuberosa, and A. viridiflora showing the best performance and A. erosa and A. subverticillata showing the worst performance. A. speciosa and A. fascicularis showed mid-range performance with validation AUC values of 0.86 and 0.83.

#### Associations Between the Monarch Butterfly, Milkweeds, and Abiotic Gradients

In general, the milkweed species were associated with a diversity of abiotic variables (**Figure 2**), with land cover being the most common (seven out of 13 species), followed by precipitation of the warmest season (five out 13 species), climatic water deficit (four out of 13 species), slope (three out of 13 species), actual evapotranspiration (three out of 13 species), and minimum temperature of the coldest month (three out of 13 species). Mean temperature of the warmest quarter was only chosen for three species, but it constituted 81% of the permuted importance for A. subulata. While land cover was the most common variable selected across the models, due to the categorical nature of land cover, the association between milkweed and land cover varied by species. Urban land cover types were positively associated with A. speciosa, A. subulata, A. asperula, and A. tuberosa. Riparian vegetation was positively associated with A. subulata and A. asperula. Oak woodland was positively associated with A. speciosa and A. tuberosa.

In contrast, all four of the monarch models showed similarity among the covariates chosen with the milkweed covariates consistently the most important predictors (four out of four models) followed by land cover (adult and breeding models), climatic water deficit (adult model), mean temperature of the warmest quarter (breeding model), minimum temperature of the coldest month (breeding without tropical milkweed), and precipitation of the warmest season (breeding without tropical milkweed) (**Figure 2**). The adult monarch model and the

FIGURE 2 | Variable importance for the monarch and milkweed models from the Maxent permutation importance. Hotters colors, such as pink and orange, indicatethat a variable was more important for the particular model whereas cooler colors such as green or blue indicates that it was less important. Gray indicates that the variable was not important at all. Due to multicollinearity (particularly among climatic variables) we cannot rule out a functional relationship between covariates that were not selected by the models. Cells in the lower right are empty (but filled in the lower left) because milkweeds were only used as predictors in the monarch models. breeding monarch model both showed associations with urban land cover types with the breeding model showing additional associations with riparian land cover types. Of the milkweed covariates, A. fascicularis was selected in all four monarch models and constituted between 23 and 40% of permuted importance. A. speciosa was selected in the "all model" and the "breeding without tropical milkweed model" and ranged from 34 to 38% permuted importance. A. subverticillata was the next most important species selected in both versions of the breeding model and ranged from 12 to 22% of permuted importance. Finally, A. subulata was selected in the "all model" and was 32% of permuted importance.

Relationships between milkweed occurrence and abiotic gradients included a diversity of functional forms, including linear responses with thresholds, as well as smooth Gaussian-like curves, consistent with ecological niche theory. In contrast, the relationships between monarchs and their milkweed covariates showed largely linear relationships with monarch habitat suitability increasing with milkweed suitability (see **Figure 3** for monarch response curves, and **Figure 4** for milkweed response curves). The selection of particular milkweed covariates (in the monarch models) may reflect the fact that two of these species (A. speciosa and A. fascicularis) are the most geographically widespread species that we studied and the two remaining species (A. subverticillata and A. subulata) occur in areas where monarchs are known to exist but A. speciosa and A. fascicularis are limited. Our findings do not suggest that milkweed species not selected by models are unsuitable for breeding but rather that their geographic range coincides less with the current documented range of D. plexippus plexippus in the West.

#### Geographic Patterns of Habitat Suitability

The habitat suitability models highlighted broad areas of suitability across the seven states, with notable concentrations in California, the Snake River Plain in southern Idaho, the Columbia Basin in eastern Washington, as well as areas of eastern Oregon, northern Nevada and Utah, and southern Arizona (**Figure 5**). The influence of the tropical milkweed A. curassavica in the Los Angeles Basin of Southern California can be seen in the comparison between **Figures 5C**,**D**. In general, suitability for milkweed species tended to be very species-specific with the ranges of the different milkweed species typically showing unique geographic patterns. The area with the highest diversity of habitat suitable for multiple milkweed species is in the California floristic province which had seven of the 13 milkweed species. In contrast, the Pacific Northwest has far fewer species of milkweed, and predicted suitability was generally consistent with the known ranges of each milkweed species. Only A. speciosa, A. fascicularis, and A. cordifolia were predicted to be suitable in parts of western Oregon. Arizona was predicted to have high suitability for a number of species (A. subulata, A. asperula, A. tuberosa, A. subverticillata, and A. erosa) but low suitability for the two most widely distributed species, A. speciosa and A. fascicularis. For monarchs, model uncertainty varied depending upon the model being used. For example, the models that used all records, adults, and breeding records omitting A. curassavica all showed high uncertainty in parts of Sonoran and Mojave Deserts whereas the model using breeding records (including A. curassavia) indicated higher uncertainty in the Central Valley of California (**Figure 6**). Similar to habitat suitability, maps of habitat uncertainty were characterized by species-specific differences. Correlations amongst predictive maps show high similarity for the monarch models with one another (R = 0.72), low correlations among the milkweed species (R = 0.16), and moderate correlations between the monarch and milkweed maps (R = 0.24) being mostly driven by the overlap between monarch models and A. fascicularis (R = 0.67) and A. speciosa (R = 0.52) (**Figure 7**).

# DISCUSSION

Our study is the first peer-reviewed paper to apply maximum entropy habitat modeling to the western monarch and a majority of the important milkweed hosts in the western U.S. (but see Lemoine, 2015 for a continental study of eight different milkweed species). The extensive spatial coverage and broad suite of species makes this a useful tool for land management planning, including identification and prioritization of key monarch breeding habitat where active management and protection efforts may be most efficiently directed, as well as identification of regions and sites where restoration may be an appropriate monarch conservation strategy or where additional monitoring is needed. Within key geographic areas where habitat restoration may be an appropriate conservation strategy, these models can inform which species of milkweed to plant. The geographic distribution of western monarch butterflies is characterized by high heterogeneity of suitable habitat (**Figure 5**), as has been seen with other wideranging species, particularly in western North America (e.g., Lozier et al., 2013). The greatest continuous expanse of suitable habitat encompasses much of California, both the agricultural Central Valley, parts of the Coast Ranges, the western foothills of the Sierra Nevada Mountains and the southern deserts, all of which highlight the general importance of the region for the western monarch butterfly. This concentration of suitable habitat is consistent with a recent isotopic analysis of natal origins for monarchs collected at overwintering sites (Yang et al., 2016), as well as a previous spatial analysis that found precipitation within the core of the California floristic province to be most closely associated with annual fluctuations in monarch densities at the overwintering sites (Stevens and Frey, 2010). Vast stretches of the Intermountain West also contain pockets of high-quality monarch and milkweed habitat, which could provide ecologists and conservation biologists numerous opportunities for further exploration of habitat-host-monarch interactions. The Columbia River Basin in Washington State and the Snake River Plain in Idaho are also notable concentrations of suitable habitat that have historically been near the northern edge of the monarch breeding range (Dingle et al., 2005; Pyle, 2015) but could potentially be more heavily utilized if migration patterns shift under warming climatic conditions.

Four of the milkweed species that we modeled for the western U.S. were also modeled by Lemoine (2015). These included A. speciosa (showy milkweed), A. fascicularis (narrowleaf

to solitary response curves that consider only the relationship between the variable and habitat suitability). Land cover is not shown for clarity's sake because it is a categorical variable. (A–C) Danaus plexippus–all records, (D,E) D. plexippus–adult records, (F–H) D. plexippus–breeding records, (I–M) D. plexippus–breeding records w/o Ascelpias curassavica.

FIGURE 5 | Maps of monarch and milkweed habitat suitability with red indicating higher habitat suitability and blue indicating lower habitat suitability. (A) Danaus D. plexippus–all records, (B) D. plexippus–adults, (C) Danaus plexippus–breeding, (D) D. plexippus–breeding without A. curassavica, (E) A. speciosa, (F) A. fascicularis, (G) A. subulata, (H) A. eriocarpa, (I) A. californica, (J) A. asperula, (K) A. tuberosa, (L) A. virdiflora, (M) A.erosa, (N) A. subverticillata, (O) A. cordifolia, (P) A. cryptoceras. For simplicity, A. incarnata is not shown. Full page maps for each species and monarch life history stage are included in Supplementary Figure 2.

milkweed), A. tuberosa (butterfly weed), and A. incarnata (swamp milkweed). In general, our maps bear great resemblance to those of Lemoine (2015), despite the fact that they were developed at different spatial scales using different covariates and different numbers of occurrence records. A. speciosa was the most widely distributed milkweed in the western U.S. with suitable areas distributed in all seven of the states that we modeled. Compared to Lemoine (2015), our habitat models show more gradation in habitat suitability, particularly in the cold desert regions of eastern Oregon, northern Nevada, and southern Idaho. This is likely due to the inclusion of land cover as a variable in our model, which may reflect the fact that A. speciosa is commonly found in or around agricultural areas and other disturbed sites. The next most widely distributed milkweed species in the western U.S. was A. fascicularis. Like A. speciosa, our habitat maps for A. fascicularis resembled Lemoine (2015) but also show more complexity, likely because distance to perennial water was found to be an important variable in our models. A. tuberosa also generally agrees with Lemoine (2015), showing suitable habitat in the mountainous regions of Arizona. A. incarnata, which was relatively uncommon in the western U. S., showed some scattered occurrences in the Salt Lake Valley of Utah and the Snake River Plain of Idaho, areas that are shown as being moderately suitable by Lemoine (2015).

The U.S. Southwest (Arizona, Utah, Nevada, and the California deserts) had large areas that were suitable for monarchs despite the lack of the two most widespread milkweed species, A. speciosa and A. fascicularis. In states such as Arizona, less geographically-extensive milkweed species, such as A. subverticillata (whorled milkweed), A. subulata (rush milkweed), A. asperula (spider milkweed), A. tuberosa (butterfly weed), and A. erosa (desert milkweed), are the primary host plants, yet these species appear to occur in different habitat types. For example, A. subverticillata, A. tuberosa, and A. asperula tend to occupy higher elevation areas, compared to A. erosa and A. subulata which appear to favor lower elevation, more arid areas. These areas may benefit from further studies that explore more fine-scale associations between environmental associations and milkweed occurrence, in order to guide the selection of potential restoration species.

The area that had the highest diversity of milkweed species was clearly the California floristic province, which contained both A. speciosa and A. fascicularis as well as a number of milkweed species that were less common throughout the range such as A. eriocarpa, A. californica (California milkweed), and A. cordifolia (heartleaf milkweed). Areas north of San Francisco Bay and in the mountain ranges of southern California were predicted to be suitable for as many as four or five milkweed species. Areas inland from the central coast and parts of the Sacramento Valley were also projected to have high species richness with suitability for three to four species of milkweed. Restoration of milkweed in the California floristic province should take into account the high species diversity, and California, like Arizona, may benefit from finer-scale studies of environmental associations and milkweed occurrence, in order to identify which milkweed species may be suitable for restoration in particular locations.

These results can serve as a baseline against which future habitat models for the western monarch and milkweeds can be compared, and could be used to guide future sampling efforts, in order to maximize knowledge gained relative to effort expended. Compared to a similarly large-extent habitat modeling effort that used occurrence records prior to 2016 (Steele et al., 2016), we were able to model additional milkweed species (13 rather than five) and had better validation statistics for both monarchs (validation AUC increased from 0.7 to 0.834 for the breeding model) and milkweeds (validation AUC increased 0.157 on average for the five species in common between the studies). Because the methods in the two studies (Steele et al., 2016 and the present study) were similar, we attribute the improved validation statistics to the larger-scale and geographically more widespread efforts to capture data on monarch occurrences. These efforts coordinated and focused field collection techniques and data sharing between the Xerces Society, the U. S. Fish and Wildlife Service, the Washington Department of Fish and Wildlife and the Idaho Department of Fish and Game, which enhanced data collection throughout Nevada, Washington, and Idaho, and specifically on wildlife refuges in Oregon, Washington, and Idaho.

One of our chief biological findings is that the monarch larval host plants of the genus Asclepias varied greatly in their abiotic habitat associations, while monarch habitat models were more consistent (among adult, breeding, and all-record models) and largely shaped by host plant associations (**Figure 2**). Thus, the monarch butterfly in the western U. S. is indeed a habitat generalist, a condition which can be considered an epiphenomenon of host specialization in a region where the host plants are diverse and include species with specific and divergent habitat requirements. The relatively minimal direct influence of climatic factors in monarch models (**Figure 2**) is generally in agreement with a recent analysis of inter-annual variation in observations across the Northern California breeding range by Espeset et al. (2016), who found a positive influence of warmer temperatures and (to a lesser extent) spring precipitation on the probability of observing monarchs during the breeding season, but with most of the inter-annual variation left unexplained by climatic factors. It is also in agreement with the findings of Lemoine (2015), who found that milkweed distributions appear to be a much stronger predictor of monarch distributions than climate alone.

With respect to determinants of milkweed habitat, different species had idiosyncratic climatic associations while generally sharing a common response to land cover that was an important variable in seven out of 13 milkweed models. Somewhat surprisingly, topographic variables were only important for four species and soil variables were only important for one milkweed species. The role of climate in shaping plant species distribution has been well-documented in ecology since the time of Merriam (1895). Indeed, climatic variables are typically among the most predictive variables when it comes to species distribution models conducted at broad spatial scales. However, the lack of importance for soil and topographic variables was especially surprising for the milkweed species, especially given that there have been increasingly strong calls for the inclusion of soil variables into species distribution models for plants (Bertrand et al., 2012; Diekmann et al., 2015; Velazco et al., 2017), and the recent improvements in spatial prediction of soil variables (Chaney et al., 2016; Ramcharan et al., 2018). It is possible that, although important, the soils and topographic variables may be more predictive at both finer spatial scales and more limited spatial extents.

Further research is needed to assess how different species of milkweeds may promote or hinder different life history stages of the monarch butterfly. In any event, the lack of geographic overlap among many of the milkweed species may offer restoration practitioners an opportunity to plant or restore milkweed species that are climatically suited to their area in regions where planting milkweed is an appropriate conservation strategy. It should be noted that none of our results either support or refute the idea that milkweed availability is a limiting factor for the western monarch or that a lack of milkweed has led to the recent decline in monarch numbers in the West (for comparison, see Inamine et al., 2016; Thogmartin et al., 2017 for discussions of milkweed limitation and other factors affecting the eastern monarch migration). The results of the breeding model identify geographic areas, which are suitable and important for monarch breeding, and which could be prioritized for protection and targeted habitat management efforts that consider the needs of monarchs. In areas which are highly suitable for milkweed and monarch breeding, but which may have lost habitat, habitat restoration including planting milkweed and nectar plants, may be appropriate. For example, restoration might be particularly impactful in the Central Valley of California, which this study identified as highly suitable for monarch breeding and Asclepias fascicularis (as well as multiple other milkweed species within the Sacramento Valley) but which has undergone major changes in land use due to conversion of grassland and shrubland to agriculture and urban or ex-urban development (Lark et al., 2015; Sleeter, 2016) and intensification of farming with increased pesticide use. It is also important to note that areas which show low suitability for specific milkweed species or monarchs, but which also have high uncertainty values, may indeed be more suitable if additional data was collected and added to future iterations of modeling. These areas should not be considered low priority for protection or restoration necessarily; they are simply data deficient.

Other conservation uses for these models include planning and monitoring. Landowners and managers can identify which milkweed species that they might expect on their lands and develop appropriate actions to identify and conserve those milkweeds. Conservation actions may include abstaining from mowing or treating milkweeds with herbicides. Our models can also be used to identify data gaps. For example, areas that are projected to have high suitability may be ideal locations to identify new populations of milkweeds and areas of high model uncertainty are areas that may benefit from additional data collection. The maps resulting from this work may also be useful for identifying potential sites for long-term monitoring. These maps may be important tools for incorporation into other conservation plans, such as multi-species connectivity plans, where the benefits of monarch habitat conservation may be augmented by benefits accrued from conserving other species. Finally, the data and habitat modeling techniques used in this study can be adopted to finer spatial or temporal scales using higher-resolution covariates (such land cover models derived from high resolution imagery ∼1 m cell size) and occurrence data that is local to that area of interest. These scaled-down habitat models may be useful for restoration planning and provide insights regarding local patterns of habitat selection within a portion of the range or within a temporal subset.

Along with the tremendous advantages of using citizenscience and museum databases (Ries and Oberhauser, 2015), our study highlights a number of challenges associated with these types of data. To minimize spatial bias in occurrences, we applied geographic thinning of the occurrence data, a widely practiced, and effective method for dealing with sampling bias (Kramer-Schadt et al., 2013). However, this resulted in a large number of occurrence points that could not be used because they were spatially redundant, usually between 2 and 40 times the number of occurrences used in modeling. The Western Monarch and Milkweed Mapper website and database (Xerces Society, 2018)—which contains all the data used in this study and more recently contributed records—could be used to coordinate efforts to conduct a more geographically representative sampling of milkweed and monarch populations across the western U. S., potentially reducing the costs and maximizing the benefits obtained from sampling. Additional survey efforts in areas west of the Continental Divide in New Mexico, Colorado, Wyoming, and Montana will be particularly important in understanding the relative importance of these habitats to the western monarch population.

Another challenge in working with museum databases in particular, and in many cases with citizen-science databases, is the lack of true absence data. Although methods exist to help deal with these issues, such as target-group sampling (Ponder et al., 2001), they rely on the assumption that the lack of a presence of one species can be treated as an absence of another species. This assumes that occurrence points are being contributed by trained observers who are equally likely to report other species if present. With our citizen-science and museum databases and varied taxa (different butterfly life history stages and multiple host plants) this assumption could not be met. Modification of the Western Monarch Milkweed Mapper and other citizen-science databases to incorporate information regarding search effort across species could allow the database to be used to perform presence-absence habitat modeling rather than just presence-background modeling. Where available, such as within portions of the National Wildlife Refuge System, presence-absence data can also be used to perform local calibration of the models to probability of occurrence. This would be valuable given the strong assumptions that the default logistic transformation in Maxent software makes (Merow et al., 2013).

Finally, we should note that our modeling approach is correlative and not intended for the inference of direct causation, although basic ecological knowledge suggests that causality undoubtedly underlies many of the patterns including the connection between milkweed habitat and monarch presence. The habitat associations reported here could be considered a starting point for common garden experiments designed to assess germination and growth of milkweed species under different abiotic conditions, and to assess monarch preference for and development on different hosts and under different conditions (Robertson et al., 2015; Hoang et al., 2017). The latter issue (monarch performance on different host species) is particularly important given the diversity of hosts used by the western monarch, and the extent to which different hosts appear to have distinct associations with climate. Such field studies could be combined with gridded environmental covariates to make predictive maps that could be compared with the maps produced from this study. Future efforts could also incorporate more complete habitat associations for the most widespread milkweeds (extending beyond our focal western states), as well as the mapping of habitat suitability for the California overwintering sites, which was beyond the scope of our present efforts (but see Fisher et al., 2018). In summary, it is our hope that the results presented here both advance our basic understanding of the biology of a widespread and relatively well-studied insect, as well as provide an important tool for conservation biologists and land managers.

# AUTHOR CONTRIBUTIONS

TD, MS, JE, EP, SJ, SB, and MF conceived and designed the study. TD, MS, EP, SM, AT, CF, EC, and DC designed and built the database. TD, MS, and MF performed the analysis. TD, MS, EP, SJ, and MF wrote sections of the manuscript. All authors provided edits and comments for the manuscript.

#### FUNDING

Financial support for this work came from the U.S. Fish and Wildlife Service and the Xerces Society for Invertebrate Conservation. MF was additionally supported by a Trevor James McMinn professorship.

#### ACKNOWLEDGMENTS

We thank all of the scientists, both citizen scientists and professionals, who have contributed monarch observations over the years, and we thank those that contributed ideas to the earlydevelopment of the analyses and models

#### REFERENCES


presented here, especially Aaron Jones, Erin Stockenberg, and Samantha Marcum.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00188/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 Dilts, Steele, Engler, Pelton, Jepsen, McKnight, Taylor, Fallon, Black, Cruz, Craver and Forister. 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 Importance of Shifting Disturbance Regimes in Monarch Butterfly Decline and Recovery

Nathan L. Haan\* and Douglas A. Landis

*Department of Entomology and Great Lakes Bioenergy Research Center, East Lansing, MI, United States*

The Eastern migratory monarch butterfly has declined in recent decades, partly because widespread adoption of herbicide-resistant corn and soybean has nearly eliminated common milkweed from crop fields in the US Midwest. We argue that in addition to milkweed loss, monarch declines were likely exacerbated by shifting disturbance regimes within their summer breeding range. Monarchs prefer to lay eggs on younger, vegetative milkweed stems. They also benefit from enemy-free space, as most eggs and early-instar larvae succumb to predators. Historically, ecological disturbances during the growing season could have provided these conditions. During most of the 19th and 20th centuries, milkweed was abundant in crop fields where manual weeding and mechanical cultivation set milkweed back, but rather than killing it would often stimulate regrowth later in the summer. Before European settlement, large mammals and fires (natural and anthropogenic) perturbed grasslands during the summer and could have had similar effects. However, presently most remaining milkweed stems in the Midwest are in perennial grasslands like roadsides, old-fields, parks, and conservation reserves, which often lack growing season disturbance. As a result, monarchs may be left with limited options for oviposition as the summer progresses and could have lower survival in grasslands where predation pressure is high. Our recent work has shown that targeted disturbance during the growing season produces milkweed stems that are attractive to ovipositing monarchs and harbor fewer arthropod predators. Targeted disturbance in perennial grasslands could improve habitat heterogeneity and phenologic diversity of milkweeds, and should be explored as a monarch conservation strategy.

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*Stephen Baillie Malcolm, Western Michigan University, United States Julia Leone, University of Minnesota Twin Cities, United States*

> \*Correspondence: *Nathan L. Haan haannath@msu.edu*

#### Specialty section:

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

Received: *19 December 2018* Accepted: *13 May 2019* Published: *29 May 2019*

#### Citation:

*Haan NL and Landis DA (2019) The Importance of Shifting Disturbance Regimes in Monarch Butterfly Decline and Recovery. Front. Ecol. Evol. 7:191. doi: 10.3389/fevo.2019.00191* Keywords: disturbance, predation, monarch butterfly, butterfly conservation, agricultural landscapes

# BACKGROUND

The Eastern North American migratory population of monarch butterflies is in a decades-long decline and the migratory phenomenon is considered at risk (Brower et al., 2012; Vidal and Rendón-Salinas, 2014). The overwintering population in Mexico is estimated to have declined more than 80% from the 1990s to 2014 (Semmens et al., 2016), and monarchs are under review for listing under the US Endangered Species Act (CBD, 2014). Potential causes of this decline include logging of overwintering habitat in Mexico, increased pathogen loads, lost nectar resources, exposure to insecticides, climate change, and loss of breeding habitat [see reviews by Inamine et al. (2016); Thogmartin et al. (2017a); Malcolm (2018); Stenoien et al. (2018)]. While the relative importance of these factors is a topic of ongoing research, there is increasing evidence that a major contributor to the recent decline is the loss of milkweed host plants from breeding habitat in the US Midwest (Flockhart et al., 2013, 2015; Pleasants and Oberhauser, 2013; Oberhauser et al., 2017; Saunders et al., 2017; Thogmartin et al., 2017a; Stenoien et al., 2018).

The majority of monarchs arriving in Mexico each year for overwintering are born in the Midwest and North Central US, where they lay most of their eggs on common milkweed, Asclepias syriaca L. (Malcolm et al., 1993; Flockhart et al., 2017). This species, which we will hereafter refer to simply as milkweed, is considered an agricultural weed. Until recently, milkweed stems growing in annual crop fields, mostly corn and soybeans, supported more monarch eggs and larvae than stems growing in other habitat types (Oberhauser et al., 2001). However, since the 1990s, more than 90% of corn and soybean production has switched to transgenic herbicide-resistant varieties (USDA Economic Research Service, 2018). Now, fields are sprayed with broad-spectrum herbicides that effectively control milkweed, resulting in a ∼40% loss of milkweed stems from midwestern landscapes (Hartzler, 2010; Pleasants, 2016). In response to this loss, recent research has focused on how to rebuild milkweed populations in the US, including calls to restore 1.3– 1.6 billion additional stems in the Midwest (Pleasants, 2016; Thogmartin et al., 2017b).

## HABITAT SHIFT FROM CROPLANDS TO GRASSLANDS

In addition to habitat loss, monarchs have also undergone a habitat shift. Until recent decades, a large proportion of milkweed stems in the Midwest were found in annual crop field interiors. Today, however, remaining milkweeds are predominantly located in perennial grasslands. Thus, a greater proportion of monarchs now rely on grassland habitats. These habitats include ditches, old-fields, pastures, transportation rights-of-way, conservation reserve program (CRP) lands, parks, and reserves (Thogmartin et al., 2017b). Perennial grasslands differ from agricultural fields in multiple respects; understanding these differences and managing in light of them may be key to stabilizing monarch populations.

#### Agricultural Disturbance and Milkweed Suitability

Crop fields are a distinct type of ecosystem, and the phenologic, nutritional, and chemical characteristics of milkweed growing in crop fields may differ from those in grasslands. Crop fields are nutrient-enriched, may be irrigated, and in the case of corn, become shaded as summer progresses. Before the widespread use of effective herbicides, they also would have been mechanically disturbed during the growing season.

Farmers controlled weeds with hand tools or draft animals until at least the 1930s, after which time tractor-based mechanical cultivation became the norm (Swinton and Van Deynze, 2017). In following decades, each year fields were cultivated until late June or early July, when the soybean canopy closes and corn grows too tall (Curran, 2004; Specht et al., 2012). For soybeans, additional manual control often continued later into the summer (Horlyk, 2013; Eller, 2014). Mechanical control was only moderately effective against milkweed; while aboveground tissue was easy to remove, milkweed's modular growth form made it resilient. Plants tended to survive and send up new shoots following cultivation, and equipment often spread roots to new areas (Bhowmik and Bandeen, 1976). Similarly, many herbicides applied to crop fields beginning in the 1960s killed aboveground milkweed growth but left the roots unscathed (Bhowmik, 1994). Monarchs prefer to lay eggs on very young and vegetative stems (Urquhart, 1987; Bergström et al., 1994), so frequent disturbance could have benefitted monarchs by supplying attractive new milkweed growth for oviposition as summer progressed. These patterns stand in contrast to perennial grasslands, where milkweeds often flower by mid-summer and afterwards can begin to senesce. Therefore, we suspect monarchs relying on milkweeds in perennial grasslands are left with increasingly poor options for oviposition as summer progresses.

As a rule, host plants for herbivorous insects vary in suitability. Conservation managers working toward butterfly recovery often need to differentiate between host plants that are suitable and ones that are not (Thomas et al., 2011). Plant suitability to herbivores can change with phenology: newly grown tissues are often replete with water and N, while older tissues tend to be tougher and lacking in these resources (Thomas and Stoddart, 1980; Scriber and Slansky, 1981; Slansky, 1993; Lim et al., 2007). Consequently, many caterpillars perform better on younger tissues compared to senescent or near-senescent ones (Scriber and Slansky, 1981; Slansky, 1993). This pattern is evident in multiple butterflies of conservation concern that avoid senescing plants or fare poorly on them (e.g., Singer, 1972; Grundel et al., 1998; Lane and Andow, 2003; Haan et al., 2018). In the case of monarchs, while it is clear they prefer to oviposit on younger milkweed stems (Urquhart, 1987; Bergström et al., 1994), we know less about how milkweed senescence affects survival and growth of larvae. This should be an area of future research. Along the same lines, milkweed in cornfields would have been shaded in late summer. While we do not know if ovipositing monarchs favored heavily-shaded milkweeds in cornfields, their caterpillars grow larger on shade-grown stems, which are less-defended and have lower leaf toughness and C:N ratios (Agrawal et al., 2012a).

Finally, nutrient enrichment in crop field soils could provide monarchs with more nutrient-dense milkweed tissues to eat. Monarch growth rates can increase with foliar N concentrations in common milkweed (Tao et al., 2014, but see Schroeder, 1976; Lavoie and Oberhauser, 2004). Nutrient enrichment can also cause milkweeds to produce less toxic cardenolides (Agrawal et al., 2012b), which limit the growth and survival of monarch caterpillars (Rasmann et al., 2009). While recent work in our lab suggests nutrients alone do not drive oviposition patterns (Myers et al., 2019), details of how crop field nutrient enrivonments could influence monarch nutrition and response to cardenolides need further investigation.

#### Differential Predation Pressure

Enemy-free space where predation pressure is minimized can be an important component of a species' niche (Jeffries and Lawton, 1984). For monarchs, enemy-free space may be much more limited in perennial grasslands than in crop fields, as predatory arthropods are more diverse and abundant in grasslands (Werling et al., 2014) and predation rates on invertebrates are consistently higher (Werling et al., 2011). Like many Lepidoptera, survival of monarch eggs and first instars is low, with a large fraction of immature monarchs succumbing to predators. For example, in a restored prairie <2% of eggs survived to third instar, with lower survival when plants contained spiders or aphids, the latter of which attract ants and other predators (De Anda and Oberhauser, 2015). Similarly, monarch eggs in an oldfield had a 2% survival rate after 7 d, with ants predating eggs and larvae (Prysby, 2004). Finally, Myers et al. (2019) found >80% mortality of monarch eggs over 72 h periods in grasslands, with lower mortality in corn.

Disturbance during the growing season could have historically reduced predation risk to monarchs through multiple mechanisms. First, milkweed stems with aphids attract more predators (Haan and Landis, 2019), so disturbances that remove aphid-infested stems should reduce predators as well. Second, vegetative milkweed stems host lower predator densities than other stages (Haan and Landis, 2019), so disturbances that reset milkweed phenology could also serve to reduce predator abundance on the regenerating stems. Finally, disturbance temporarily simplifies the structure of surrounding vegetation, which could limit habitat suitability for some predators, particularly spiders (Rypstra et al., 1999).

# DISTURBANCE: A KEYSTONE PROCESS INFLUENCING MONARCH HABITAT SUITABILITY?

Ecological disturbance may be a key factor determining the quality of breeding habitat for Eastern monarchs, as it potentially provides both phenologically-attractive host plants and enemyfree space. Disturbance during the growing season was a defining characteristic of annual crop fields, but it occurs much less often in most perennial grasslands that monarchs now rely on. Positive recovery efforts for several rare butterfly species have depended on whether managers reinstated historical disturbance regimes (Thomas, 1980; Schultz and Crone, 1998, 2015; Thomas et al., 2009; Dunwiddie et al., 2016; Haddad, 2018). In contrast to many rare butterfly species, monarchs breed in landscapes that have been transformed by humans, and thus may have come to depend on agricultural disturbance regimes during the 20th century. Interestingly, habitat management recommendations for monarchs currently discourage disturbances during the breeding season (MJV [Monarch Joint Venture], 2018).

Multiple studies have documented that as milkweed stems regenerate after fields are mowed, they can support large numbers of monarch eggs and larvae (Marsh, 1888; Borkin, 1982; Fischer et al., 2015; Alcock et al., 2016). Building on these observations, we conducted a field experiment in Michigan to determine if strategically-timed disturbance can enhance monarch habitat in perennial grasslands (Haan and Landis, 2019). Monarchs laid more eggs on milkweeds that regenerated after being cut back compared to those we left undisturbed, and predators took 2–4 weeks to recolonize the regenerating milkweed stems, potentially providing a window of enemy-free space. We believe these results suggest disturbance is an important process influencing monarch habitat suitability in the Midwest, and that some types of habitat could be enhanced with strategically-timed disturbance during the growing season (**Figure 1**).

Growing season disturbance may have been common before Euro-American settlement of the Midwest, although its historic effects on monarchs are left to speculation. Native Americans farmed corn, at times quite extensively (Riley et al., 1994; Benson et al., 2009), and managed Midwest and Great Plains ecosystems with fire for thousands of years. In contrast to current prairie restoration practices which concentrate burns during spring and fall, evidence suggests fires were historically set almost any time of year, including summer (Bragg, 1982; Higgins, 1986). Similarly, lightning-ignited fires are most common during summer (Komarek, 1964; Bragg, 1982; Higgins, 1984). Milkweed readily regenerates with new stems after summer fire; in Oklahoma, prescribed fire in July produced regenerating milkweeds (Asclepias viridis) which were used by monarchs in late summer and early fall (Baum and Scharber, 2012). Large mammals could have also been an important historical source of disturbance. Bison were ubiquitous in grasslands of the Midwest and Great Plains until the late 19th century and could have produced regenerating milkweed stems through grazing, trampling, or wallows (Knapp et al., 1999). Similarly, prior to humans' arrival in North America, grasslands hosted diverse and abundant megafauna which would have caused a variety of year-round disturbances (Mack and Thompson, 1982; Milchunas et al., 1988).

On the other hand, contemporary monarch ecology could differ markedly from past centuries. Common milkweed may have proliferated in agricultural landscapes precisely because it tolerates mechanical disturbances, and monarchs may have historically relied more heavily on the several other milkweed species native to the Midwest, as these could have been much more abundant before the destruction of North American prairie (Gray et al., 1889). It has also been hypothesized that the migratory phenomenon in its current form is itself anthropogenic; that it only came about because deforestation of the Eastern US in the 19th century caused a population explosion of milkweeds and monarchs as they colonized newly-available habitat (Vane-Wright, 1993).

# REFINING THE CONCEPT OF MILKWEED LIMITATION

The idea that milkweed shortage in the Midwest underlies monarch declines has been met with controversy (Davis and Dyer, 2015; Dyer and Forister, 2016; Inamine et al., 2016; Oberhauser et al., 2017; Thogmartin et al., 2017a; Zaya et al., 2017; Stenoien et al., 2018; Boyle et al., 2019). Even the casual observer will notice that milkweed is a common sight outside of crop fields in the Midwest, but that most stems contain no monarch eggs or larvae. It follows that milkweed abundance

per se does not limit monarchs. This line of thinking parallels that of Hairston et al. (1960). In their landmark paper they proposed that in contrast to other consumers, herbivores are not generally limited by the availability of food—if they were, the earth would not be covered in such an excess of plants. Therefore, herbivore populations must be limited by something other than plant abundance. This generated two competing, although not mutually-exclusive, hypotheses: first, herbivores could be limited because some plant material is unsuitable, e.g., if some plants are poisonous or nutritionally inferior (Murdoch, 1966). Second, herbivores could be limited by predators. This latter position was the one espoused by Hairston et al., and became part of the basis for the trophic cascade concept.

It is interesting to apply the same logical framework to monarchs and milkweeds. If monarchs were limited by milkweed stem quantity per se, we would expect to find competition for milkweed stems. However, most milkweed stems are not used by monarchs; on undisturbed milkweeds often < 1 monarch egg is found per ten stems (e.g., Fischer et al., 2015; Pitman et al., 2018; Haan and Landis, 2019). If the supposition is that monarchs are limited by milkweed quantity per se, then these simple observations disprove it. However, we believe this apparent discrepancy can be solved by one or both of the following hypotheses, which correspond to the ones generated by Hairston et al. (1960): First, monarchs are limited not by milkweed quantity per se, but rather in terms of the quality and suitability of extant milkweed stems for oviposition and larval feeding. Second, monarch populations are limited by enemies like predators and parasitoids. An important process underpinning both possible mechanisms is disturbance.

# DESIGNING MONARCH-FRIENDLY LANDSCAPES

If the migratory monarch phenomenon is to persist, we need to design and manage agricultural landscapes with abundant, phenologically-diverse milkweeds and associated windows of enemy-free space. This will require disturbance regimes that are coordinated and carried out at a regional scale. Our purpose here is not to be prescriptive about the type or timing of disturbance, as these are likely to be context-dependent, but practices could include fire, grazing, haying, mowing, or others. In our recent work (Haan and Landis, 2019) we focused on occasional mowing, which is an appropriate grassland management technique for some, although certainly not all, contexts: effects of mowing and hay harvesting on biodiversity can be negative, neutral, or positive depending on timing and technique (Dale et al., 1997; Johst et al., 2005; Roth et al., 2005; Humbert et al., 2009; Cizek et al., 2012). Large areas of perennial grasslands in Midwest landscapes are already mowed for agricultural, safety, and aesthetic reasons. Redirecting or adjusting even a fraction of that annual effort into strategically-timed disturbance of milkweed could create a mosaic of phenologically-diverse milkweed stems and patches of enemy free space for monarchs. We emphasize that we are not advocating for wholesale increase in mowing frequency or extent in the Midwest. Instead, we envision a heterogeneous landscape in which some chronicallydisturbed areas are left alone to allow for growth of milkweeds and other plants, while some currently-undisturbed grasslands could be intentionally perturbed mid-season. We believe this would significantly improve the productivity of the current stock of milkweeds, as well as those that are added to the landscape as part of conservation efforts.

In the short-term, more research is needed on how to effectively utilize strategic disturbance for monarch conservation as many questions remain. For example, should we disturb (e.g.) one in three milkweed patches, or a third of each patch? What is the opportunity-cost associated with disturbance, since milkweeds require 1–3 weeks to regenerate? Could disturbance create ecological traps by concentrating oviposition effort in certain areas, increasing natural enemy effectiveness or disease transmission? Will the prevalence of young milkweed tissues in late summer cause butterflies to skip reproductive diapause and fail to migrate south? What are long term effects of repeated disturbance on milkweed persistence? Can we ensure disturbance regimes are compatible with other conservation objectives (e.g., pollinators and grassland nesting birds)? Future work predicting effects of disturbance at the regional level could interface with existing models designed to help us understand where and how to improve monarch habitat (Thogmartin et al., 2017a) and predict the resulting monarch population response (Oberhauser et al., 2017).

In the long term, we also need to consider how to create and maintain heterogeneity in agricultural landscapes to benefit biodiversity and ecosystem services more broadly. Intensified agricultural practices often lead to landscape simplification

#### REFERENCES

Agrawal, A. A., Kearney, E. A., Hastings, A. P., and Ramsey, T. E. (2012a). Attenuation of the jasmonate burst, plant defensive traits and resistance to specialist monarch caterpillars on shaded common milkweed (Asclepias syriaca). J. Chem. Ecol. 38, 893–901. doi: 10.1007/s10886-012-0145-3

and loss of biodiversity services (Landis, 2017) but including perennial elements within agricultural landscapes can have many potential benefits. For example, incorporating prairie strips into corn and soybean fields can reduce erosion and nutrient loss while increasing biodiversity (Schulte et al., 2017). Future production of bioenergy crops on marginal soils also could produce multiple benefits for biodiversity and ecosystem services depending on the crops selected (Landis et al., 2018). Perennial polycultures based on prairie systems support a wide array of biodiversity and are highly compatible with monarch conservation (Werling et al., 2014).

#### CONCLUSION

In conclusion, we believe recent shifts in disturbance regimes across the Midwest US have caused not just a reduction in milkweed quantity, but also reductions in the suitability of extant plants, and in enemy-free space. While we support calls to introduce more milkweed to grasslands in the Midwest, evidence suggests we should also examine ways to improve productivity of existing milkweeds and reduce pressure from natural enemies. More work is needed to understand how the type, timing, and frequency of disturbance could influence monarchs and their complex interactions with milkweeds and other arthropods. More broadly, habitat manipulations to support monarchs must be integrated into the landscape in ways that support other conservation goals (e.g., pollinators and grassland birds) as well as contribute to soil and water quality and the aesthetic aspects of agricultural landscapes.

#### AUTHOR CONTRIBUTIONS

NH and DL generated ideas for the manuscript. NH led the writing effort.

#### FUNDING

This material is based upon work supported in part by USDA National Institute for Food and Agriculture (2017-68004-26323), the Great Lakes Bioenergy Research Center, U.S. Department of Energy, Office of Science, Office of Biological and Environmental Research under Award Numbers DE-SC0018409 and DE-FC02- 07ER64494, and MSU AgBioResearch.

#### ACKNOWLEDGMENTS

We thank A. Myers, S. Hermann, and C. Gratton for helpful discussions on this topic.


northern Virginia. J. Lepid. Soc. 70, 177–181. doi: 10.18473/107.07 0.0302


abundance of monarch butterflies Danaus plexippus. Ecography 41, 278–290. doi: 10.1111/ecog.02719


**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 Haan and Landis. 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 Integrated Monarch Monitoring Program: From Design to Implementation

Alison B. Cariveau<sup>1</sup> \*, Holly L. Holt <sup>1</sup> , James P. Ward<sup>2</sup> , Laura Lukens <sup>1</sup> , Kyle Kasten<sup>1</sup> , Jennifer Thieme<sup>1</sup> , Wendy Caldwell <sup>1</sup> , Karen Tuerk <sup>1</sup> , Kristen A. Baum<sup>3</sup> , Pauline Drobney <sup>4</sup> , Ryan G. Drum<sup>5</sup> , Ralph Grundel <sup>6</sup> , Keith Hamilton<sup>2</sup> , Cindy Hoang<sup>2</sup> , Karen Kinkead<sup>7</sup> , Julie McIntyre<sup>8</sup> , Wayne E. Thogmartin<sup>9</sup> , Tenlea Turner <sup>4</sup> , Emily L. Weiser <sup>9</sup> and Karen Oberhauser 1,10

<sup>1</sup> Monarch Joint Venture, Department of Fisheries, Wildlife, and Conservation Biology, University of Minnesota, St. Paul, MN, United States, <sup>2</sup> U.S. Fish and Wildlife Service, Inventory and Monitoring, National Wildlife Refuge System, Fort Collins, CO, United States, <sup>3</sup> Department Integrative Biology, Oklahoma State University, Stillwater, OK, United States, <sup>4</sup> U.S. Fish and Wildlife Service, Neal Smith National Wildlife Refuge, Prairie City, IA, United States, <sup>5</sup> U.S. Fish and Wildlife Service, Bloomington, MN, United States, <sup>6</sup> U.S. Geological Survey, Great Lakes Science Center, Chesterton, IN, United States, 7 Iowa Department of Natural Resources, Des Moines, IA, United States, <sup>8</sup> U.S. Fish and Wildlife Service, Ecological Services, Tucson, AZ, United States, <sup>9</sup> U.S. Geological Survey, Upper Midwest Environmental Sciences Center, La Crosse, WI, United States, <sup>10</sup> University of Wisconsin-Madison Arboretum, University of Wisconsin, Madison, WI, United States

#### Edited by:

Laurentiu Rozylowicz, University of Bucharest, Romania

#### Reviewed by:

Adam Korösi, MTA-ELTE-MTM Ecology Research Group, Hungary Arthur M. Shapiro, University of California, Davis, United States

#### \*Correspondence:

Alison B. Cariveau alison.cariveau@gmail.com

#### Specialty section:

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

> Received: 28 February 2019 Accepted: 25 April 2019 Published: 29 May 2019

#### Citation:

Cariveau AB, Holt HL, Ward JP, Lukens L, Kasten K, Thieme J, Caldwell W, Tuerk K, Baum KA, Drobney P, Drum RG, Grundel R, Hamilton K, Hoang C, Kinkead K, McIntyre J, Thogmartin WE, Turner T, Weiser EL and Oberhauser K (2019) The Integrated Monarch Monitoring Program: From Design to Implementation. Front. Ecol. Evol. 7:167. doi: 10.3389/fevo.2019.00167 Steep declines in North American monarch butterfly (Danaus plexippus) populations have prompted continent-wide conservation efforts. While monarch monitoring efforts have existed for years, we lack a comprehensive approach to monitoring population vital rates integrated with habitat quality to inform adaptive management and effective conservation strategies. Building a geographically and ecologically representative dataset of monarchs and their habitat will improve these efforts. These data will help track long-term changes in the distribution and abundance of monarchs and their habitats, refine population and habitat models, and illuminate how conservation activities affect monarchs and their habitats. The Monarch Conservation Science Partnership developed the Integrated Monarch Monitoring Program (IMMP) to profile breeding habitats and their use by monarchs in North America. A spatially balanced random sampling framework guides site selection, while also allowing opportunistic inclusion of sites chosen by participants, such as conservation areas. The IMMP weaves new protocols together with those from existing monitoring programs to improve data compatibility for assessing milkweed (Asclepias spp.) density, nectar resources, monarch reproduction and survival, and adult monarch habitat use. Participants may select a protocol subset according to interests or local monitoring objectives, thereby maximizing contributions. Conservation partners, including public and private land managers, academic researchers, and citizen scientists contribute data to a national dataset available for analyses at multiple scales. We describe the program and its development, implementation elements that make the program robust and feasible, participation to date, and how IMMP data can advance research and conservation for monarchs, pollinators, and their habitats.

Keywords: butterfly counts, citizen science, conservation effectiveness, habitat assessment, monarch butterflies (Danaus plexippus), cooperative monitoring, milkweed, nectar plants

# PURPOSE AND RATIONALE

Monarch butterflies (Danaus plexippus) exhibit one of the most spectacular animal migrations (Urquhart, 1976; Brower, 1977). East of the Rocky Mountains in North America, monarchs migrate up to 4,500 km each fall to overwinter in high-altitude fir forests in central Mexico; west of the Rockies, monarchs overwinter in groves along the California coast. In spring, monarchs return to their breeding grounds; several generations move and breed across most of North America throughout the summer. Migrating and breeding butterflies rely on nectar sources for food; to reproduce monarchs depend solely on larval host plants in the milkweed subfamily (primarily Asclepias spp.).

Like many pollinator species (Biesmeijer et al., 2006; Potts et al., 2010; Powney et al., 2019), North American monarch populations have declined over the past two decades (Brower et al., 2012; Vidal and Rendón-Salinas, 2014; Semmens et al., 2016; Schultz et al., 2017), motivating range-wide conservation efforts. Breeding range conservation has focused on enhancing milkweed and nectar availability, as reduction of these resources is implicated in monarch population declines (Pleasants and Oberhauser, 2013; Pleasants, 2017; Thogmartin et al., 2017a; Zaya et al., 2017; Malcolm, 2018; Stenoien et al., 2018). Conservation efforts are driven by population targets, e.g., those in the national pollinator strategy (Pollinator Health Task Force, 2015) and related national habitat goals (Thogmartin et al., 2017b).

Monarch conservation goals are generally based on models of monarch population viability (Semmens et al., 2016; Schultz et al., 2017), geographic prioritization (Flockhart et al., 2015; Oberhauser et al., 2017), threats (Saunders et al., 2017; Thogmartin et al., 2017a), and habitat (Thogmartin et al., 2017b) developed using limited datasets and expert opinion. While some studies have examined breeding habitat use (e.g., Stenoien et al., 2015; Kasten et al., 2016; Pitman et al., 2018; Kantola et al., 2019), they are limited in scope and geography. Citizen science program (e.g., Journey North<sup>1</sup> , Monarch Watch<sup>2</sup> , Monarch Larva Monitoring Project (MLMP)<sup>3</sup> ) data have been instrumental to modeling efforts and expanding general knowledge of monarchs (Oberhauser et al., 2015; Ries and Oberhauser, 2015; Tracy et al., 2019), but are often concentrated near population centers and lack geographical balance (Bird et al., 2014; Nail et al., 2015). Furthermore, use of largely selfselected monitoring locations that often contain high-quality habitat (e.g., butterfly gardens or butterfly monitoring sites) prevents robust statistical inference about average conditions or extrapolation to other land-use types (Bird et al., 2014). Lastly, many programs record monarch locations opportunistically, without measured and repeated effort, making it difficult to identify long-term trends. A monitoring scheme that overcomes these limitations is needed (National Research Council, 2007) to accurately track progress toward habitat and population goals, identify habitat deficiencies, and assess the success of conservation actions.

The Monarch Conservation Science Partnership (MCSP), a collaborative group of scientists addressing information gaps in monarch conservation and ecology, developed an integrated strategy for monitoring conservation progress, starting with the end goal and working backward to determine the details (Thogmartin et al., 2015; Reynolds et al., 2016). Through review of existing programs (Oberhauser et al., 2009), 3 years of design meetings, and pilot testing, the strategy became the Integrated Monarch Monitoring Program (IMMP). The IMMP collects geographically and ecologically representative data using a stratified randomized sampling framework. Data from conservation sites, such as private lands enrolled in Farm Bill conservation programs, are included to provide insight into the effectiveness of management actions. The sampling framework optimizes statistical robustness while minimizing the number of samples needed by prioritizing sites where collecting information will be most valuable.

The IMMP has three primary objectives: to (1) track longterm changes in the distribution and abundance of monarchs and their habitats (2) provide geographically and ecologically representative information to fill data gaps and update current population and habitat models, and (3) acquire information about how habitat conservation actions affect monarchs and their habitat. Metrics include milkweed density, indices of blooming plant abundance, adult monarch abundance, egg and larval abundance, egg and larval survival estimates, and fire ant occurrence.

Below, we highlight design elements of the newly implemented IMMP that make it robust, efficient, and feasible for large-scale, multi-partner data collection and use. We discuss the benefits of the program to researchers, land managers, and citizen scientists, as well as the benefit of compatible and representative long-term data generated over a broad geography.

#### SPATIALLY BALANCED RANDOM SAMPLING

A key IMMP element is its proactive sampling design that obtains geographically and ecologically representative data throughout the monarch's breeding range. Geographically distributed data allow evaluation of how monarch habitat and its use vary across ecoregions, latitude, elevation, and climatic conditions. Ecologically representative sampling emphasizes all habitats that may be suitable for monarchs rather than just easily accessible sites or known habitat locations (Bird et al., 2014).

To establish representative sampling locations for the IMMP, we used a Generalized Random Tessellation Stratified (GRTS) sampling design (Stevens and Olsen, 1999, 2003, 2004). GRTS provides a spatially balanced set of sample units with a randomized component for unbiased representation and can represent multiple strata to reduce variability of parameter estimates. GRTS produces a hierarchical sample list such that for any sample size or geographic subset, the sample will be spatially balanced if the sample list is followed in order (Loeb et al., 2015).

We applied GRTS to rank, and thereby prioritize for sampling, each 10 × 10-km cell within a grid of cells ("blocks" hereafter)

<sup>1</sup>https://journeynorth.org/

<sup>2</sup>https://www.monarchwatch.org/

<sup>3</sup>www.mlmp.org

showing three strata, areas modeled as low-, medium-, and high-suitability for milkweed, per the western milkweed habitat suitability project (Dilts and Forister, 2017). Inset 2: an example monarch block from the eastern population depicting top random site locations for each of the five major land-use strata for sampling (Thogmartin et al., 2017b).

superimposed over the contiguous United States (**Figure 1**). Within each block, a second-stage GRTS draw ranked points for unbiased plot location within each sampling stratum in each block. For the eastern population, the strata comprise five land types associated with milkweed: agriculture, protected grasslandshrubland, unclassified grassland-shrubland, rights-of-way, and developed areas (Thogmartin et al., 2017b). In the west, a model of habitat suitability for milkweed was built from environmental variables (Dilts and Forister, 2017), so western strata are high, medium, and low expected suitability for milkweed.

GRTS can also incorporate data from non-random locations, such as legacy or volunteer-selected sites (Overton and Stehman, 1993; Olsen et al., 1999). Non-random sites may not represent the full landscape (Williams et al., 2001; Kinkead et al., in review), but can provide data from spatially rare land-use types that are poorly represented in the random draw but might be of particular interest (e.g., state parks, Conservation Reserve Program). During analysis, data from non-random sites can be down-weighted to reduce bias while improving statistical power for the entire dataset (Austin, 2011).

# DATA COLLECTION

Integrated Monarch Monitoring Program (IMMP) surveys are modular; data are valuable regardless of whether all protocols are completed at each site. Participants collect data relevant to their interests using Survey123 (data survey application, Esri, Redlands, California) on a mobile device or paper with online data entry. Below, we give a brief overview of the primary field surveys; a complete guide (Monarch Joint Venture, 2019) is posted on the IMMP webpage<sup>4</sup> . A U.S. Fish and Wildlife Service (USFWS) protocol will provide full documentation of the purpose, rationale, and monitoring procedures from design to reporting, to be posted in ServCat<sup>5</sup> (USFWS information repository).

#### Plot Description

A site is sampled by a 1-hectare (ha) rectangle (200 × 50 m) or square (100 × 100 m) monitoring plot originating from

<sup>4</sup>https://www.monarchjointventure.org/IMMP/

<sup>5</sup>https://ecos.fws.gov/ServCat/Reference/Profile/109175

Cariveau et al. The Integrated Monarch Monitoring Program

a random starting point (**Figure 2A**). Longer, narrower plots, (400–500 m long) are used in linear areas (e.g., rights-of-way); alternative configurations fit irregularly shaped sites. Consistency in monitoring plot size reduces variation from area effects and differential effort. Participants collect data regarding site characteristics and management practices in consultation with project managers and/or landowners. Continuity in monitoring sites across years is preferred for trend detection, but shorterterm inventories can inform regional and sector comparisons.

#### Milkweed and Blooming Plant Survey

Participants survey milkweed and blooming plants in 100 quadrats placed every 5 m along transects (0.5 × 1-m frames are placed to each side of transects equaling 1-m<sup>2</sup> area per quadrat). Transects run the length of the plot, 25-m apart (**Figure 2B**), with variation for small, linear, and irregularly shaped sites. Three nested sections within quadrats aid in frequency sampling (Elzinga et al., 1998). To estimate milkweed density, milkweed plants and stems are counted by species within quadrats. To generate an index of nectar availability, all blooming plants are either identified or their presence simply noted and assigned to the smallest quadrat section in which they occur to generate frequency scores (when species are identified, richness and diversity are also calculated). Additional blooming species are recorded during a meandering walk through the plot (following Szigeti et al., 2016a). Surveys average 2.5 h; the recommended interval is monthly during the season of monarch use.

#### Egg and Larva Survey

Egg and larva data are used to examine how immature monarch densities (monarchs per plant) vary spatially, within seasons, and among years. Surveyors examine up to 100 milkweed plants within the monitoring plot, recording the number of monarch eggs and larvae per milkweed plant observed, and identifying larvae to stadium (instar) using visual cues. To representatively sample (and account for aggregations; Zalucki and Kitching, 1982; Pitman et al., 2018), surveyors search milkweed within quadrats. If milkweed is sparse, surveyors also search within 1 m of transects. If milkweed is abundant, only every second, third or fifth plant is searched. This protocol was adapted from the MLMP, allowing data to be combined for analysis. Survey time averages 1 h, and weekly surveys are recommended.

#### Adult Monarch Survey

Adult monarch surveys provide data on the abundance and phenology of monarchs throughout breeding and migration periods. Participants conduct a modified Pollard walk (Pollard, 1977), counting adult monarchs within 5 m on each side of a 500 m transect (**Figure 2C**) and documenting monarch behaviors (e.g., nectaring and associated plant species). Surveys produce a time-specific index of adult monarch abundance (number/ha) compatible with existing butterfly monitoring programs (e.g., North American Butterfly Monitoring Network<sup>6</sup> ). Surveys average 25 min to complete; bi-weekly surveys are recommended during the season of monarch use. Nectar plant selection can be

encountered in the field. From a random point, we anchor a rectangle within patch of particular land-use type (must include <10% non-target land-use type).

(Continued)

<sup>6</sup>https://www.thebutterflynetwork.org/

FIGURE 2 | If this does not fit, we rotate the plot clockwise, shift the rectangle while still encompassing the point, use a square, or delineate an irregular shape to fit within the patch (in decreasing order of preference). For non-random sites, plots are centered within the field or management unit of interest, following the same guidelines. Within plots, nectar plants and milkweed are surveyed along two 200-m transects and one 100-m transect (indicated by yellow lines with black hash marks), and adult monarch surveys are conducted around the plot perimeter (blue solid line). (B) Rectangular quadrats are placed first to the left and then to the right of transects, every 5 m, for a total of 100 quadrats per plot. (C) To count adult monarchs, surveyors move along the perimeter of the rectangle, using a modified Pollard walk with a moving data recording window of 5 m on both sides.

quantified by combining nectaring adult data and the relative abundance of nectar plant species at the same site and date (Manly et al., 2002).

positive program experiences and intent to participate in the next season.

#### Survival and Parasitism

To estimate larval survival and measure spatiotemporal variation in mortality, participants collect fourth or fifth instar larvae from just outside the monitoring plots and rear them indoors to track outcomes (e.g., adult monarch, parasitism by tachinid fly, mortality due to other causes). Before releasing the newly emerged monarchs, participants use a sticker to screen for a protozoan parasite, Ophryocystis elektroscirrha; stickers are sent to Project Monarch Health (PMH)<sup>7</sup> . While daily monitoring of known cohorts would provide more complete survival data, rearing late instars with ample exposure to disease, parasites, and parasitoids provides a broad-scale relative index of larval outcomes across time, regions, dates, and land use types. Rearing and parasite testing protocols were adapted from the MLMP and PMH, yielding compatible data sets.

### PILOT TESTING AND PROTOCOL REFINEMENT

Field testing and protocol refinement spanned 2016–2018, on 97 sites surveyed by USFWS technicians, 82 by Monarch Joint Venture (MJV) staff, 60 by University of Wisconsin technicians, and 127 by MJV-trained volunteers. During 2017–2018, the MJV trained 171 citizen scientists, biologists, and conservation staff representing 25 organizations.

A power analysis was conducted on 2016 and 2017 data to estimate the sampling effort needed to detect trends in densities of milkweed, eggs, and adult monarchs and to compare densities across strata or regions (insufficient pilot nectar data were available; Weiser et al, in revision). The consequences of survey frequency (surveys per year), numbers of quadrats for milkweed, and the number of sites and years were examined. Based on limited pilot data, the numbers of sites and years contributed more than the number of quadrats or visits per year to the statistical power to detect trends or differences, indicating the importance of repeatedly sampling large numbers of sites through time.

The power analysis and feedback from participants led to protocol revisions to improve ease of data collection, including simplification of the transect placement process, reductions in the number of quadrats, and capping the number of milkweed plants examined for eggs and larvae. These changes reduced the time required to collect data and improved participant experiences. In 2018, 86% of participant survey respondents (n = 43) reported

#### DATA MANAGEMENT

A centralized database and GIS platform readily available to participants and partners is hosted by the MJV<sup>8</sup> ; USFWS maintains a database for their staff. Data are documented according to the Darwin core standards as described in (Wieczorek et al., 2012) and are shareable with efforts such as the Trinational Monarch Knowledge Network<sup>9</sup> . Data sharing agreements enable land owners to specify the level of geographic precision for data sharing (e.g., at the scale of the monarch block rather than at a specific point location). Data are available to participants and researchers upon request. Web-hosted data summaries and visualization tools (e.g., graphs and maps) for milkweed densities, monarch distributions, and nectar plant species composition are in development.

#### PARTICIPATION

Involvement from a broad array of partners is essential for implementing a successful monitoring program for such a widely distributed species. Integration with existing naturalist networks has been a successful strategy for spurring participation in the IMMP. Collaboratives (e.g., Monarchs Across Georgia10), nature centers, government agencies, or volunteer groups serve as IMMP "hubs." These entities connect IMMP methods with local conservation goals and expand implementation by recruiting and training local participants. Outreach and training workshops held with these groups have amplified data collection in new localities and mobilized larger audiences.

Participation is fostered by a number of tools hosted by MJV on the IMMP webpage, including activity instructions, training resources (including video), mapping tools, and data entry portal. USFWS hosts guidance documents, site-screening and mapping tools, data collected by their staff, and associated products on ServCat<sup>11</sup> .

By participating in the IMMP, citizen scientists and private landowners deepen their connection with monarchs and appreciation for their conservation challenges. Their participation can broaden civic engagement within local communities (Lewandowski and Oberhauser, 2017), and contribute more representative data than is possible by agencies working alone.

<sup>8</sup>https://monarchjointventure.org/IMMP/

<sup>9</sup>https://birdscanada.org/birdmon/tmkn/

<sup>10</sup>https://www.eealliance.org/monarchs-across-ga

<sup>11</sup>https://ecos.fws.gov/ServCat/Reference/Profile/109175

<sup>7</sup>www.monarchparasites.org

# RELEVANCE

Monitoring is a key element of adaptive management and strategic habitat conservation (National Ecological Assessment Team, 2006). The IMMP will provide conservation professionals with an enhanced understanding of the dynamics of monarch populations, their habitats, and their response to conservation efforts. IMMP results can readily be entered to the USFWS's Monarch Conservation Database12, which tracks monarch conservation efforts and informs decisions about the butterfly's status. The IMMP may be used to achieve research and monitoring objectives within monarch conservation plans (e.g., Midwest Association of Fish and Wildlife Agencies, 2018), including monitoring monarch habitat, estimating milkweed distribution across different land-use sectors, monarch distribution, vital rates, nectar resource selection, and understanding effects of disease and pathogens.

The IMMP is already instrumental to local, regional, and national conservation assessment and research programs. Several county conservation boards in Iowa (K. Kinkead personal communication) and dozens of private landowners in the eastern U.S. are using it to evaluate the quality of monarch habitat on their conservation lands. Statewide collaboratives such as Missourians for Monarchs<sup>13</sup> employ it to track progress toward achieving their statewide milkweed stem goals. IMMP data were used to parameterize a milkweed density index in a model of monarch reproductive use on an Iowa landscape (Grant et al., 2018; Grant and Bradbury, 2019). Regionally, IMMP protocols were used by (Lukens et al., in review) to evaluate conservation projects in the Upper Midwest and in landscape-scale studies of habitat quality and monarch survivorship at University of Wisconsin. Federal programs also use the IMMP, for example, to assess monarch habitat and use on USFWS refuges, and to compare with the U.S. Department of Agriculture's Natural Resources Conservation Service assessment of monarch habitat (i.e., Wildlife Habitat Evaluation Guide).

#### IMPACT

The scale and potential impact of the IMMP compare to other large-scale programs such as the United Kingdom Butterfly Monitoring Scheme (Roy et al., 2001; Brereton et al., 2006), North American Bat Monitoring Program (Loeb et al., 2015), and the Breeding Bird Survey, which have influenced conservation policy (Hudson et al., 2017). In three pilot years, the IMMP has already been implemented at hundreds of sites across Georgia, Iowa, Illinois, Kentucky, Michigan, Minnesota, Missouri, North Dakota, Ohio, Oklahoma, Texas, and Wisconsin, showing strong potential to reach the scale and impact of other successful largescale monitoring programs.

The IMMP will greatly improve our knowledge of monarch biology, particularly in historically under-surveyed geographies and land-use types. The multi-dimensionality of IMMP data, which pairs quantitative habitat data with monarch use, provides an opportunity to assess how monarchs in several life stages interact with a variety of spatially and temporally explicit habitat characteristics. IMMP protocols can also be used to address priority research questions such as the location of gaps in nectar resources along migration routes, or how proximity to fields routinely treated with pesticides affects monarch recruitment and survival (Midwest Association of Fish and Wildlife Agencies, 2018).

IMMP nectar plant information can benefit broader pollinator conservation efforts and efforts for other declining species that rely on flowering plants (e.g., Rusty-patched Bumblebee). Data on nectar plant species richness and frequency can help land managers gauge progress toward habitat goals, such as establishing plants with staggered bloom times recommended by many pollinator plans. While more frequent visits may better characterize nectar availability for pollinators (Szigeti et al., 2016b, 2018), IMMP data can contribute to larger phenology databases (e.g., USA National Phenology Network14) and ultimately contribute to our understanding of habitat availability in a changing climate.

Broad and diverse participation is necessary to achieve the desired breadth and depth of sampling and to ensure the IMMP's long-term sustainability. Success will depend on mobilizing partners across government, academia, and NGOs, alongside a cadre of citizen scientists. These efforts are only just beginning, and the potential for long-term scientific payoff is enormous. Ultimately, monarch conservation relies on the cooperation of all stakeholders not only in protecting and restoring habitat, but also in understanding and evaluating this species and the habitats on which it relies.

#### DATA AVAILABILITY

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

#### AUTHOR CONTRIBUTIONS

AC led refinements to the protocol and writing the manuscript. HH, JW, and KO led initial stages of design. JW is the USFWS lead in documenting the full protocol. LL and JT worked on revisions and lead current implementation. KT developed tools for data entry, GIS, and data management. WT and EW provided the sampling framework and statistical analyses. KB, WC, PD, RD, RG, KH, CH, KyK, KaK, JM, and TT contributed to program development, documentation, and implementation.

#### FUNDING

Funding support has been provided by U.S. Fish and Wildlife Service, the U.S. Forest Service, and the National Fish and Wildlife Foundation.

<sup>12</sup>https://www.fws.gov/savethemonarch/mcd.html

<sup>13</sup>http://www.moformonarchs.org/

<sup>14</sup>https://www.usanpn.org/usa-national-phenology-network

#### ACKNOWLEDGMENTS

The Monarch Conservation Science Partnership developed the IMMP over the course of several years, and we acknowledge (in addition to the authors) participants in several meetings through which the strategy was developed [including but not limited to Greg Butcher (USFS), Melissa Martin (NRCS), Richard Easterbrook, Allen Gilbert, Dan Konzek, Jana Newman, (USFWS), Diane Larson (USGS), and John Pleasants (Iowa State University)]. We also thank the multitude of participants of the

#### REFERENCES


IMMP, including private land owners who provided access to their land and volunteers and technicians who collected data. Funding support has been provided by USFWS, the U.S. Forest Service, and the National Fish and Wildlife Foundation. The findings and conclusions in this article are those of the authors and do not necessarily represent the views of the U.S. Fish and Wildlife Service or other coordinating NGOs, local or state agencies. Any use of trade, product, or firm names are for descriptive purposes only and do not imply endorsement by the U.S. Government.


**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 Cariveau, Holt, Ward, Lukens, Kasten, Thieme, Caldwell, Tuerk, Baum, Drobney, Drum, Grundel, Hamilton, Hoang, Kinkead, McIntyre, Thogmartin, Turner, Weiser and Oberhauser. 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.

# Design Implications for Surveys to Monitor Monarch Butterfly Population Trends

Karen E. Kinkead<sup>1</sup> \*, Tyler M. Harms <sup>1</sup> , Stephen J. Dinsmore<sup>2</sup> , Paul W. Frese<sup>1</sup> and Kevin T. Murphy <sup>3</sup>

*1 Iowa Department of Natural Resources, Boone, IA, United States, <sup>2</sup> Department of Natural Resource Ecology and Management, Iowa State University, Ames, IA, United States, <sup>3</sup> Center for Survey Statistics and Methodology, Iowa State University, Ames, IA, United States*

The Iowa Multiple Species Inventory and Monitoring (MSIM) Program includes a protocol for monitoring butterfly density on conservation lands using transects. Most data are collected from sites chosen randomly; additional sites are chosen non-randomly for other reasons. We analyzed a 12-year dataset for monarchs to address how density (per 50 m<sup>2</sup> transect section) responded to site selection (random vs. non-random), latitude, and measures of the amount of milkweed and canopy cover on survey transects. Between 2006 and 2017, we conducted 2,328 surveys on 420 sites and detected a total of 2,757 adult monarchs. Monarch densities peaked in 2010 for random sites and 2012 for non-random sites, but densities were lowest in 2013 for both site types. The density of monarchs at non-random transects (0.047, 95% CI = 0.031, 0.062) was higher than that at random transects (0.029, 95% CI = 0.019, 0.044) and the temporal trends of density for random and non-random sites were significantly different. Monarch density was positively correlated with UTM northing, suggesting that monarch density increases from south to north in Iowa. The percent of plots containing milkweed was positively correlated with monarch density whereas percent tree canopy cover was negatively correlated with monarch density. Our results show that non-random transects had more monarchs, which may be a concern when interpreting findings from some citizen science efforts that used non-probabilistic sampling designs. Collectively, the MSIM program data provide a comprehensive assessment of monarch densities statewide as well as the first empirically-derived density estimates for monarchs on the breeding grounds and may prove helpful when refining future monitoring efforts.

Keywords: density, Iowa, monarch, transect, survey, butterfly

# INTRODUCTION

The monarch butterfly (Danaus plexippus [L.]) has become a species of interest recently due to declines seen in the overwintering territories in Mexico (Brower et al., 2012; Vidal and Rendón-Salinas, 2014; Monarch Watch Blog, 2018) and the risk of quasi-extinction of 11-57% in the next 20 years (Semmens et al., 2016). This decline led to a petition of the U.S. Fish and Wildlife Service in 2014 to list the monarch as a threatened species (Center for Biological Diversity, 2014) as well as a presidential memorandum calling for the

#### Edited by:

*Ryan G. Drum, United States Fish and Wildlife Service (USFWS), United States*

#### Reviewed by:

*David Zaya, Illinois Natural History Survey (INHS), United States Emily L. Weiser, United States Geological Survey, United States*

> \*Correspondence: *Karen E. Kinkead Karen.kinkead@dnr.iowa.gov*

#### Specialty section:

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

> Received: *01 January 2019* Accepted: *13 May 2019* Published: *31 May 2019*

#### Citation:

*Kinkead KE, Harms TM, Dinsmore SJ, Frese PW and Murphy KT (2019) Design Implications for Surveys to Monitor Monarch Butterfly Population Trends. Front. Ecol. Evol. 7:195. doi: 10.3389/fevo.2019.00195* restoration of pollinator habitat in the U.S. to benefit monarchs (Office of the Press Secretary, 2014). The attention stimulated research across the U.S. to evaluate the status of migratory monarch populations and to investigate potential factors leading to monarch population declines (Vidal and Rendón-Salinas, 2014; Badgett and Davis, 2015; Flockhart et al., 2015; Ries et al., 2015; Schultz et al., 2017; Thogmartin et al., 2017).

The Upper Midwest region of the U.S. has long been considered the primary breeding ground for the monarch (Wassenaar and Hobson, 1998; Flockhart et al., 2017). Various projects have tracked monarch numbers through transect based surveys during fall migration in Michigan (Badgett and Davis, 2015), plot based surveys of 15-mile diameter during the 4th of July Butterfly counts (Swengel, 1990, 1995), and a metaanalysis of the 4th of July data and transect based surveys in Illinois (Ries et al., 2015), although few of those studies utilized randomly chosen locations. Randomization, however, is critical if the study objective involves providing the best estimate of a population trend over a large, variable space (Thompson, 1992; Pollack et al., 2002).

Many authors have suggested that breeding habitat in the Upper Midwest is limiting (Flockhart et al., 2015; Thogmartin et al., 2017). Recent habitat loss in this area has been directly linked to declines in the eastern migratory population of the monarch (Pleasants and Oberhauser, 2013; Flockhart et al., 2015). This habitat loss is characterized mostly by a decrease in milkweed plant stems (in the Asclepiadoideae subfamily), the only larval host plant for monarchs. In Iowa, the number of agricultural fields occupied by Common Milkweed (Asclepias syriaca) declined by 90% from 1999 to 2009 (Hartzler, 2010) and more recent estimates from throughout the Upper Midwestern area of the U.S. suggest milkweed declines of nearly 40% (Pleasants, 2017). Despite studies establishing the loss of milkweed as a significant contributor to monarch population declines in the U.S., there are still some (e.g., Davis and Dyer, 2015 [although see (Pleasants et al., 2016) for rebuttal], Inamine et al., 2016 [but see (Pleasants et al., 2017) for rebuttal]) that feel current monitoring programs do not illuminate the need for additional habitat in the breeding areas. These differences have led many organizations (e.g., State Fish and Game Agencies) to express the need for large scale monitoring within the breeding zones to complement efforts to increase breeding habitat.

While various efforts have tracked monarch numbers at large spatial scales (Swengel, 1995; Badgett and Davis, 2015; Ries et al., 2015), these projects relied on data collected by citizen scientists on targeted survey sites selected using a non-probabilistic sampling design. Several studies have documented the value of opportunistic citizen science data for a variety of purposes, which include tracking migration patterns (Oberhauser et al., 2015; Supp et al., 2015) and monitoring distribution trends using occupancy models (Van Strien et al., 2013). At issue is whether data collected on sites using a non-probabilistic sampling design are appropriate for making inferences on a broad issue such as declining monarch population trends in the U.S. Nonprobabilistic, or targeted, sampling introduces subjectivity into the study design, which can lead to biased results and incorrect inference (Williams et al., 2001). For example, when allowed to choose sites for surveying butterflies, participants often gravitate toward parks, preserves, and other non-randomly selected natural areas likely to have higher densities of the species of interest because of expectations about finding the target species more often. This situation becomes problematic for a generalist species like the monarch, which has long used a variety of "marginal" habitats such as the edges of agricultural fields and rights-of-ways that are less likely to be surveyed in a targeted effort (Pleasants et al., 2017). Monarchs may be more abundant and more likely to persist in natural areas because these areas are less likely to lose milkweed and other native nectar sources important to monarchs compared to more marginal habitats listed above. Therefore, patterns observed on surveys of targeted sites may not be representative of those occurring more broadly, and this introduces risk of making the wrong decisions about conservation actions for a declining species.

In addition to examining the possible differences between targeted and probabilistic sampling locations, our dataset can provide information on the annual variation in adult monarchs on breeding grounds in Iowa. These densities are also compared to the overwintering numbers. Using covariates, we can look at possible influences of tree canopy cover, milkweed presence, and latitude on adult monarch densities. Our study was not designed for adult monarchs in particular, but monarchs are one of the many species we have tracked.

# MATERIALS AND METHODS

#### Study Area and Multiple Species Inventory and Monitoring Program

Iowa's Multiple Species Inventory and Monitoring program (MSIM) was designed to record data on taxonomic groups which have species designated as those of greatest conservation need (SGCN) within the Iowa Wildlife Action Plan (Zohrer, 2006; Iowa Department of Natural Resources, 2015). Although target organisms are those SGCN, we record data on all species of the designated taxon observed as we are not able to predict which common species may become rare in the future and vice versa. When Iowa's MSIM program began, monarchs were not SGCN. When the IWAP was revised in 2015, however, monarchs were added as SGCN (Iowa Department of Natural Resources, 2015). Our data collection began in 2006 and continues through today.

While the MSIM program surveys properties chosen using a stratified random sampling approach, our protocols are available to everyone wishing to complete surveys on properties of interest (Iowa Department of Natural Resources, 2016). As long as the field methodology is followed and data are collected in Iowa, we will accept data from projects where the properties were not chosen randomly. For example, some of our property managers have implemented habitat management practices specifically for non-game birds and butterflies on public grassland areas. Others may be interested in developing management plans and have provided additional staff or funding to document species prior to changes in habitat management. These additional "non-random" properties have allowed our program to examine potential differences seen when properties are chosen randomly

for monitoring broad trends as opposed to chosen non-randomly to answer a specific local question (e.g., "I've done specific habitat management, what species are here now?"). While our dataset encompasses both public and privately-owned properties, all but two are associated with conservation in some manner and therefore are not representative of the larger landscape in Iowa. Our privately-owned properties have been enrolled in some form of conservation program (e.g., Landowner Incentive Program, Conservation Reserve Program, Wetland Reserve Program, etc.). While some of the tracts on public property include corn or soybean food plots planted either for wildlife or to prep the field for a grassland planting, these practices are not representative of high production farm practices in Iowa. Our two non-conservation oriented sites were fields on an Iowa State University Research Farm (one corn and one soybean) surveyed in 2016. These two fields were also part of the non-randomly chosen sites as their purpose was to assist us in assessing potential damage to crops should the MSIM Program incorporate true agricultural lands in the future.

#### Site Selection

Since the MSIM program surveys for 9 taxonomic groups across all habitat types, the majority of properties (n = 333) surveyed for these analyses were chosen by following the methodology described in Harms et al. (2014). Additional properties have been chosen since that time using a straight random selection process without regard to habitat classification as the revised IWAP (Iowa Department of Natural Resources, 2015) utilized different habitat classifications.

An additional 87 sites were included in the analyses as the MSIM field protocols were followed. These sites represent a mix of federal, state, county, and privately owned lands where the property managers were interested in learning more about the species that occur on those lands due to either the property being a recent acquisition or in the process of having new management plans developed. None of these areas were chosen randomly, but rather, were targeted by the property manager as areas they expected to have a high amount of wildlife diversity of all taxa of interest. A subset of the randomly selected properties are surveyed every year but the majority of both property types have been surveyed 1-2 years only.

#### Butterfly Surveys

Our butterfly protocol consists of a modified Pollard walk (Pollard and Yates, 1993) that allows the estimation of density (number of monarchs per 50-meter<sup>2</sup> ). We also conduct timed Visual Encounter Surveys where the technician walks through, in their opinion, the best quality butterfly habitat on the property in order to document rarer butterflies. Visual Encounter Survey data will not be reported in this manuscript. The modified Pollard walk entails recording data within different segments of the transect of known length, thereby allowing for the calculation of densities and the associated variation.

Every effort was made to place the center of the sampling area within the habitat for which the property was originally chosen, not necessarily the best butterfly habitat. A 200 m transect was then extended straight north and an additional 200 m transect was extended straight south from this point at most properties for a total transect length of 400 m. In some cases, properties features (e.g., lakes, rivers, etc.) prevented this placement and the transect was moved accordingly, always maintaining a total length of 400 m (perhaps broken into 2 segments) on the property. Transects are flagged in 10 m sections to assist the observers with knowing where they are in the transect. This flagging system creates a 5 meter-wide transect for the observer to follow.

Transects are walked up to four times per year at a pace of approximately 10 m per min. All butterfly species encountered within the transect are recorded, along with which transect segment the butterfly is in at the time of detection. Most transects are walked between June 1 and August 31, with at least 1 visit per month but transects may be walked earlier in May or later into September. Additional information about this protocol can be found in Chapter 12 of the MSIM Technical Manual (Iowa Department of Natural Resources, 2016). In a separate study that contributed to this dataset (Patterson, 2016), detection probability of monarchs on modified Pollard walks was very close to 1.00. Therefore, we did not estimate detection probability directly in this study.

# Habitat Covariates

Local microhabitat covariates were collected on the property once during late summer and include estimates of tree canopy cover and the percentage of plots containing milkweed of any species. These data were collected in the field following the protocols outlined in Chapter 19 (Terrestrial Habitat Classification) of the MSIM Technical Manual (Iowa Department of Natural Resources, 2016). Canopy cover of trees is measured by taking four canopy cover estimates ("1" if present, "0" if absent) around the perimeter of each of the 0.017 ha vegetation plots in each of the four cardinal directions in the larger MSIM survey area. These values are then averaged together to give an overall percentage estimate for the property. Milkweed presence was recorded in the 1-m<sup>2</sup> quadrats or the 5-min plant search. As such, these data were collapsed into "1" (milkweed species present) or "0" (no milkweed present) within each of the larger habitat plots on the property. The percent of the habitat survey plots which had at least 1 stem of milkweed of any genus on the property became our estimate of milkweed for that survey location.

# Data Analyses

Prior to analyses, we truncated transects to those surveyed between May 16 and August 20 to decrease the likelihood of including migrating monarchs. These dates were chosen based on Journey North (2018) First Monarch reports in the spring and the table documenting peak migration by latitude produced by Monarch Watch (2018). While arrival and departure dates can change annually due to weather conditions, we chose these dates based on the majority of reports from Iowa to the listed organizations across the years. The median survey date for both random and non-random site types was within 2 weeks of 15 July each year and, due to our standardized sampling methodology, we do not expect survey timing to vary spatially.

After truncating our data set to the primary breeding season, we then split the breeding season into 3 monthly periods (16 May−15 June, 16 June−15 July, and 16 July−20 August) to allow for evaluation of seasonal densities throughout the breeding season. We treated transect section (n = 40) within each property (n = 420) as our primary sampling unit, and if surveyed in a given year, most transects were surveyed at least once during each time period. For transects that were surveyed more than once during a given time period, we aggregated monarch observations for each transect section within each transect by taking the maximum number of monarchs observed across all survey visits within the time period. We used the maximum count (as opposed to mean or minimum counts) for several reasons. First, the mean is not appropriate because our Poisson distribution (described later) requires the use of integers, so the mean would need to be rounded and this could introduce bias. Second, using the minimum count would result in many more zero counts, thereby missing non-zero counts on many sites. This resulted in a maximum of three survey visits to each transect within the primary breeding season for a given year. Because monarchs are multi-generational and are continuously migrating throughout the primary breeding season (Brower, 1996), we could not assume population closure at transects between time periods and therefore treated the survey visit within each time period in any given year as independent. Our final data set, consequently, included 69,560 transect section by time period combinations accounting for some properties having multiple transects and some transects being surveyed multiple years. Each sampling unit was a fixed area of 10 × 5 m. Henceforth we consider our response variable to be a maximum density of monarchs per 50 m<sup>2</sup> survey area (monarch density).

Next, we developed a candidate set of models that evaluated the influence of various survey design factors and habitat variables on monarch density to inform both long-term monitoring and habitat management objectives. While several models could have been considered, we decided to focus our model development to estimate monarch density as a function of the covariates that directly addressed our hypotheses:


we included the Universal Transect Mercator (i.e., UTM) northing as a covariate to capture potential variation in monarch density from south to north in Iowa.


We first developed a model to estimate the overall average monarch density throughout Iowa across all years (i.e., null model). We then modeled monarch density as a function of random vs. non-random site type. Next, we built two models to evaluate annual differences in monarch density, the first to estimate the overall average monarch density each year and the second to estimate monarch density on both random and nonrandom transects by year (e.g., Site Type∗Year interaction). We treated year as a factor variable in both the aforementioned models in order to allow for non-linear variation in density across years, and treated site type as a factor variable in order to obtain individual estimates for each site type. Realizing that monarch density varied considerably by site type and year after evaluating the Site Type∗Year model, we modeled all subsequent covariates as additive effects in the Site Type∗Year model to capture any additional variability induced by those covariates. We first modeled the additive effect of UTM northing to determine if a south-to-north gradient in monarch density existed in Iowa (K. Oberhauser, pers. comm. and Prysby and Oberhauser, 2004). Next, we modeled an effect of season as a categorical covariate to assess differences, if any, in monarch density throughout the primary breeding season. We then modeled both percentage of plots with milkweed and percent tree canopy cover as separate additive effects in the Site Type∗Year model to evaluate the influence of habitat covariates on monarch density. Lastly, to test for a difference in the temporal trend in density between random and non-random site types, we converted the year variable from factor to numeric and created a dummy binary variable to represent whether or not a site was randomly selected (i.e., 1 = randomly selected, 0 = not randomly selected). We then modeled monarch density against the interaction of the numeric year variable and the random dummy variable and evaluated the confidence interval of the interaction to determine if the temporal trend across years changed when sites were randomly selected. If the confidence interval did not include zero, we concluded that the temporal trend across years between

random and non-random site types was different. We examined the correlation between point estimates of annual density from random and non-random sites using Pearson's product moment correlation and considered estimates significantly correlated if P ≤ 0.05.

We wanted to account for variability both among and within survey transects in addition to evaluating variability in monarch density in response to the above fixed effects. Therefore, we included a random intercept for site and a random intercept for sampling plot nested within site in each of the above models to capture the among- and within-transect variability, respectively. All models were fit using the "glmer" function within the package lme4 (Bates et al., 2015) in Program R (ver. 3.5.1; R Core Team, 2018). We modeled monarch density as a Poisson random variable accounting for over-dispersion and assumed a log-normal distribution for both random effects. We evaluated how well our top model addressed over-dispersion in the data by comparing the sum of the squared Pearson residuals to the residual degrees of freedom using a Chi-square test (see https://bbolker.github.io/mixedmodels-misc/glmmFAQ.html# testing-for-overdispersioncomputing-overdispersion-factor).

We evaluated each model relative to one another using Akaike's Information Criterion (AIC) and considered the model with the lowest AIC value to have the strongest support (Burnham and Anderson, 2002). Importantly, we made inferences from multiple models because we chose the simplest model with a particular effect to address each of our original hypotheses. A table of overall model selection results is provided to indicate the relative importance of effects considered in our hypotheses. We scaled all covariates to aid in model convergence and interpretation of the regression coefficients. We considered fixed effects to have a significant influence on monarch density if their respective 95% confidence intervals did not include zero. We then predicted monarch density and associated 95% confidence interval as a function of fixed effects by conducting 1,000 simulations using the "bootMer" function within the package lme4 in Program R (R Core Team, 2018). We used the median of all simulated predictions as the density estimate, the 2.5th-percentile of all predictions as the lower confidence limit, and the 97.5th-percentile as the upper confidence limit.

# RESULTS

We surveyed a total of 420 sites throughout Iowa from 2006 to 2017 (**Figure 1**), 333 of which were randomly chosen and 87 were non-randomly selected (**Table 1**). A total of 1,904 and 424 surveys were conducted on random and non-random transects, respectively, resulting in a total of 2,757 adult monarch observations across all years. Percentage of sampled plots at a site that had milkweed averaged 11.4% (SE = 0.016) on nonrandom sites and 12.6% (SE = 0.008) on random sites. Percent tree canopy cover averaged 36.4% (SE = 0.029) on non-random sites and 41.5% (SE = 0.015) on random sites.

The ratio of the sum of squared Pearson residuals to the residual degrees of freedom from our top model was significantly < 1 (ratio = 0.723, P = 1.0), indicating our model adequately addressed over-dispersion in the data. The overall mean estimate of monarch density across all transects from 2006 to 2017 was 0.033 per 50-m<sup>2</sup> (95% CI = 0.023, 0.049). Adult monarch density on random sites was lower than on non-random sites overall (0.029 per 50-m<sup>2</sup> vs. 0.047 per 50-m<sup>2</sup> ; β = −0.455, 95% CI = −0.794,−0.122; **Figure 2**) as well as site type by year (**Figure 3**). Monarch density significantly increased from south to north in Iowa (β = 0.363, 95% CI = 0.237, 0.489) and



TABLE 2 | Model selection results comparing the influence of different fixed effects on monarch density in Iowa, 2006-2017.


*The "Null" model is an intercept-only model and includes no fixed effects. "w" represents the AIC weight of the particular model.*

throughout the breeding season (β = 0.949, 95% CI = 0.878, 1.02). Finally, monarch density was negatively correlated with percent tree canopy cover (β = −0.999, 95% CI = −1.289, −0.709; **Figure 4A**) and positively correlated with percent cover of milkweed (β = 0.582, 95% CI = 0.128, 1.04; **Figure 4B**). The top model (AIC = 19351.7) included the interaction Site Type ∗ Year interaction and the additive effect of season (**Table 2**).

Our model of monarch density as a function of the interaction between the numeric year variable and the dummy variable representing randomly selected sites indicated a significant difference in temporal trend across years between random and non-random site types (β = 0.25, 95% CI = 0.120, 0.384; **Figure 5**). Additionally, the correlation between annual density estimates from random and non-random sites was relatively low and not significant (r = 0.235, P = 0.514).

Including a random intercept for both transect and sampling unit within transect illustrated significant variability both among and within survey transects. Variability among the different survey transects throughout Iowa (SD = 1.272) was greater than the variability within each individual survey transect (SD = 0.523).

#### DISCUSSION

Our study is the first to estimate annual monarch density on the breeding grounds using a long-term monitoring program with a randomized study design. Similar estimates have proven critical for evaluating monarch population viability and response to various habitat management and restoration actions in other studies (Flockhart et al., 2015; Oberhauser et al., 2017). We found considerable annual variation in monarch density on the Iowa breeding grounds from 2006 to 2017, which is not surprising given our current knowledge of the dynamic nature of monarch populations on the overwintering areas in Mexico (Vidal and Rendón-Salinas, 2014).

#### Random vs. Non-random Site Type

For random site types, monarch density was highest in 2010 and lowest in 2013; and a linear trend fit across all years showed a decline in monarch density (β = −0.110, 95% CI = −0.175, −0.048). While a decline in monarch populations on the breeding grounds has been suggested by other studies investigating factors leading to overall monarch population declines (Flockhart et al., 2015; Pleasants et al., 2017), our study is the first to demonstrate such a decline using empiricallyderived estimates of density. Our result, coupled with the continual decline in monarch populations on overwintering grounds (Vidal and Rendón-Salinas, 2014), emphasizes the importance for continued monitoring on the breeding grounds. Using standardized monitoring programs such as the MSIM program in Iowa and the Integrated Monarch Monitoring Program (Cariveau et al., 2019), overseen by the Monarch Joint Venture, combined with efforts to increase appropriate

breeding habitat in the Upper Midwest, should provide a better understanding of the importance of habitat work in the breeding grounds. The MSIM Program collects data only on adult butterflies, whereas additional programs such as the IMMP, similar to the MLMP, collect needed data on egg and larval density to further inform models such as those in Oberhauser et al. (2017).

Other studies have failed to document a declining trend on the breeding grounds and argue that their findings suggest that factors leading to successful fall migration and survival on the overwintering grounds are more critical to conserving monarch populations rather than habitat restoration on the breeding grounds (Ries et al., 2015; Inamine et al., 2016). Both studies, however, used data from citizen science monitoring programs collected on targeted survey sites that were not located randomly on the landscape. Pleasants et al. (2016, 2017) challenged these studies, indicating that population trend estimates based on survey sites that are not representative of all possible survey sites can differ from estimates based on randomly-located and representative survey sites. Similarly, Saunders et al. (2019) highlights the inability to determine whether the lack of trends on the breeding grounds occurred due to a true absence of trend or a result of bias in data collection. Our study corroborates Pleasants et al. (2017) claim with empirically-derived density estimates, showing that monarch density on the breeding grounds was lower on average and by year on random site types compared to non-random site types.

# Year

In addition, we found a significant difference in temporal trend of monarch density between random and non-random site types. Both results have important implications for future monitoring of monarch populations throughout their breeding range. First, some policy for threatened and endangered species is based on estimates of population size, and it is critical these estimates both accurately and precisely represent population dynamics through time. This is also true for indices used to track population trends, which often require strict standardization and assumptions about the ability of the index to approximate true population size which if not satisfied can result in misleading population trends. We know that broad inference based on non-random or targeted sampling strategies is problematic for several reasons, often leading to biased parameter estimates and challenges associated with standardization of data collection. The need for rigorous, random site selection has been wellestablished in the ecological literature (Williams et al., 2001; Johnson, 2002; Mazzocchi, 2007; Nusser et al., 2008). Our results showed poor correlation between annual density estimates for random and non-random site types, indicating that density estimates from one site type cannot reliably predict those for the other site type.

Therefore, based on this previous work and our results, we emphasize caution when interpreting results from studies based on targeted sampling strategies and argue for randomization in future studies of long-term monarch population trends to ensure unbiased inference at appropriate spatial scales. It

may be worth adding that we included non-random sites in a robust design occupancy analysis (Dinsmore et al., 2019) and argued that the inclusion of these sites was of less concern for presence/absence data than for density or count data (as in this study). This is due to the high occupancy rate of monarchs (often > 0.90), independent of actual counts.

#### Spatial Stratification

We documented significant variation both among and within survey transects and found that monarch density increases from south to north in Iowa. Both of these findings have implications for the design of future surveys to track monarch population trends. Monarch density varied more among different survey transects across the state than among sampling units within a single transect. Variation among different survey transects is driven primarily by spatial heterogeneity in the distribution of animals caused by factors such as changes in habitat quantity and quality or climatic variables. Factors influencing the counting of animals during a survey are the primary drivers of within-transect variability which include temporal variability in the number of animals present on a sampling unit, differences in habitat characteristics among sampling units, or differences in conditions that affect an observer's ability to detect animals (i.e., measurement error; Skalski, 1994). Accounting for both sources of variability in long-term surveys improves the ability to detect population trends as illustrated in other long-term monitoring programs such as the North American Breeding Bird Survey (BBS; Link et al., 1994).

#### Season

In addition to demonstrating that monarch density increases from south to north in Iowa, our results also suggested monarch density increases throughout the breeding season. Given results from the MLMP (K. Oberhauser pers. comm., Prysby and Oberhauser, 2004), it appears that monarchs will breed in southern Iowa in June and early July, but late July and August may be too hot and monarchs may move further north. While further analyses could show a more pronounced difference in density from south to north throughout the breeding season, the significant individual effects illustrate the need to consider repeated surveys throughout the breeding season along with spatial stratification. Furthermore, annual variation in climate could lead to changes in the magnitude of the spatial effect on density. For example, we might expect the south-to-north increase in monarch density to be less pronounced or perhaps disappear in breeding seasons with below-average temperatures. Nonetheless, for future large-scale monitoring of monarch population trends, we suggest researchers consider stratification if prior knowledge suggests potential spatial variation in counts in order to increase precision of estimates (Skalski, 1994; Williams et al., 2001; Johnson, 2002).

# Milkweed

We found that percent cover of milkweed was positively correlated with monarch density, which confirms the results of other studies and adds to the growing evidence that restoring milkweed on the breeding grounds should be a priority for conserving monarch populations in North America. Milkweed should be embedded, however, within a matrix of additional nectar sources to provide food for the adults in addition to the milkweed host plant for the larvae (Bull et al., 1985; Suzuki and Zalucki, 1986; Brower, 1995; Zalucki and Lammers, 2010; Pleasants and Oberhauser, 2013; Thogmartin et al., 2017).

# Tree Canopy

Our study found that forest canopy cover is negatively correlated with monarch density, although we did document monarchs at properties with tree canopy cover. Kaul et al. (1991) documented the tallest specimen of Common Milkweed (Asclepias syriaca) along forest borders and other habitat edges in Nebraska. Although it has long been assumed that open habitats, such as grasslands, are more important for monarchs (Thogmartin et al., 2017; Midwest Association of Fish Wildlife Agencies, 2018), we were not able to find empirical studies confirming that monarchs are less likely to be found in areas with a closed tree canopy suggesting this may be a topic for further investigation. The species description in Seitz (1924) states that monarchs will move into previously forested areas quickly after clearings are created, which may include small tree clearings thereby creating disturbance needed for Common Milkweed to thrive (Kaul et al., 1991). The ability of the monarch to use small openings in forested habitats is confirmed by our study; even small disturbances within a closed tree canopy habitat that allows for the establishment of milkweed could benefit monarchs.

#### Study Implications

Evaluating diagnostics of our models indicated some bias in model fit, suggesting our models were over-simplified and did not capture all existing variability in our data. We utilized a hypothesis-driven modeling approach, which allowed us to build simple models that allowed for direct comparison of different covariate effects that satisfied our objectives. It's uncertain whether a more complex modeling approach, for example one that employed multiple models with additive covariate effects, would reduce model bias, but it would almost certainly complicate the interpretation of the results relative to our original objectives. Although our models may be too simplistic, they adequately address over-dispersion in our data and produce reasonable parameter estimates that can be used to inform future monitoring efforts for monarchs on the breeding grounds.

We are in no way criticizing the use of trained citizen scientists to collect data on monarchs and their habitats. We have many observers in our program, making it challenging to account for an observer effect in our models. While we provide training on sampling methods and identification to ensure consistency in data collection, as well as a certain level of rigor common to all observers, we still have variability among the skill levels. Oberhauser et al. (2015) provides an excellent summary of the major advances citizen scientists have contributed to the knowledge of monarch conservation. Our critique is that those citizens should receive guidance about where to survey and should not be allowed to choose where to collect the data. While not every volunteer will be willing to be directed to locations, other existing programs (e.g., the North American BBS or North American Amphibian Monitoring Program) demonstrate that some volunteers will be willing to monitor suggested locations. This highlights the tradeoff between fewer data collected in a standardized manner vs. more data that are unstandardized.

Data from our program show clear differences in conservation lands chosen randomly vs. those chosen non-randomly by property managers. We expect these differences would be even more pronounced if the sampling frame were expanded to include properties from other land-use types (e.g., rights-of-way, agricultural, urban, and privately owned/working grasslands) beyond conservation lands. Despite good intentions, people often want to collect data where they will see the animals of interest. Therefore, when allowed to choose, they will often pick the best quality habitat within a reasonable distance from their home. Our study demonstrates the importance of spatial and temporal patterns in monarch densities in Iowa and reinforces the need to randomly select sites for long-term monitoring efforts.

#### ETHICS STATEMENT

While butterflies are animals, they are not covered under Iowa State University's IACUC and therefore we do not need an IACUC permit for these surveys.

# AUTHOR CONTRIBUTIONS

KK is responsible for the original design and funding acquisition with input on analysis questions and additional funding from SD. TH performed the analyses. KK and TH wrote the manuscript with input from SD and review from KM and PF. PF oversaw field data collection.

#### FUNDING

Funding for the study was primarily provided through the U.S. Fish and Wildlife Service's State Wildlife Grants Program (T-6-R-1, T-6-R-2, T-6-R-3, T-6-R-4, T-6-R-5, F15AF00269, and F15AF00257), the Iowa DNR, and Iowa State University. Additional funding supported data collection on specific properties and was provided through the U.S. Army Corps of Engineers and the U.S. Fish and Wildlife Service Landowner Incentive Program. This paper is partially a product of the

# REFERENCES


Iowa Agriculture and Home Economics Experiment Station, Ames, Iowa. Project No. IOW05438 is sponsored by Hatch Act and State of Iowa funds. USFWS WSFR SWG Funds are formula funds provided to State Agencies with approved Wildlife Action Plans. The USACE and Iowa DNR Funds were provided in order to collect information on what species were found on specific areas in Iowa. Hatch Funds are formula funds to ISU from USDA which support research at land grant institutions.

#### ACKNOWLEDGMENTS

We are thankful for the many, many seasonal technicians that have assisted with data collection each year. P. Dixon and W. Thogmartin provided advice and guidance on analyses. S. Shepherd assisted with training of seasonal technicians. We also thank two reviewers for suggestions that improved this manuscript.


**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 Kinkead, Harms, Dinsmore, Frese and Murphy. 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.

# The Role of Modeling in Monarch Butterfly Research and Conservation

Tyler J. Grant <sup>1</sup> \* and Steven P. Bradbury 1,2

*<sup>1</sup> Department of Natural Resource Ecology and Management, Iowa State University, Ames, IA, United States, <sup>2</sup> Department of Entomology, Iowa State University, Ames, IA, United States*

Models are an integral part of the scientific endeavor, whether they be conceptual, mathematical, statistical, or simulation models. Models of appropriate complexity facilitate comprehension and improve understanding of the variables driving system processes. In the context of conservation planning decision-making or research efforts, a useful model can aid interpretation and avoid overfitting by including only essential elements. Models can serve two related, but different purposes: understanding and prediction of future system behavior. Predictive models can require several iterations of refinement and empirical data gathering to be useful for conservation planning. Models with less predictive ability can be used to enhance understanding of system function and generate hypotheses for empirical evaluation. Modeling monarch butterfly systems, whether it be landscape-scale movement in breeding habitats, migratory behavior, or population dynamics at monthly or yearly timeframes, is challenging because the systems encompass complex spatial and temporal interactions across nested scales that are difficult, if not impossible, to empirically observe or comprehend without simplification. We review mathematical, statistical, and simulation models that have provided insights into monarch butterfly systems. Mathematical models have provided understanding of underlying processes that may be driving monarch systems. Statistical models have provided understanding of patterns in empirical data, which may represent underlying mechanisms. Simulations models have provided understanding of mechanisms driving systems and provide the potential to link mechanisms with data to build more predictive models. As an example, recently published agent-based models of non-migratory eastern North American monarch butterfly movement and egg-laying may provide the means to explore how different spatial patterns of habitat, habitat quality, and the interaction of stressors can influence future adult recruitment. The migratory process, however, has not been addressed with agent-based modeling. Using western monarch migration as an example, we describe how modeling could be used to provide insights into migratory dynamics. Future integration of migratory models with non-migratory and population dynamics models may provide better understanding and ultimately prediction of monarch butterfly movement and population dynamics at a continental scale.

#### Edited by:

*Wayne E. Thogmartin, United States Geological Survey, United States*

#### Reviewed by:

*Matthew Forister, University of Nevada, Reno, United States Francisco Botello, National Autonomous University of Mexico, Mexico*

> \*Correspondence: *Tyler J. Grant tgrant@iastate.edu*

#### Specialty section:

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

> Received: *19 February 2019* Accepted: *13 May 2019* Published: *31 May 2019*

#### Citation:

*Grant TJ and Bradbury SP (2019) The Role of Modeling in Monarch Butterfly Research and Conservation. Front. Ecol. Evol. 7:197. doi: 10.3389/fevo.2019.00197*

Keywords: modeling, monarch butterfly, agent-based modeling, population dynamics, migration

# INTRODUCTION TO MODELING

Models are an integral part of the scientific process. Models have been described as "a representation of reality" that serves a specified purpose (Webb, 2001) or a "description of a system" (Haefner, 2005). Models can be used for several purposes. A useful distinction is between models that help improve understanding and those that provide predictive capability. Better understanding of a system provides insights on key drivers and processes in a system that may otherwise be difficult to observe. Prediction is possible when drivers and processes are quantified so that future system states can be estimated. All models are simplified versions of reality that allow us to make inferences and conclusions about a system without unnecessary complexity. The art of modeling is to construct a model simple enough to make conclusions about a system but not so simple that all connection to reality is lost. This "sweet spot" has been termed the Medawar Zone (**Figure 1**; Grimm et al., 2005).

Classes of models include conceptual, mathematical, statistical, and simulation/algorithmic models. Conceptual models are ideas or hypotheses of how a system functions. All research is based explicitly or implicitly upon conceptual models of how systems work and may be explicitly represented with flowcharts or other tools to communicate working hypotheses of a system (Heemskerk et al., 2003). Mathematical models represent a conceptual model of a system using analytical expressions and equations. They tend to be highly simplified expressions of system processes and are often used to explore general theory. Statistical modeling seeks to separate statistical noise from signal by modeling the probabilistic processes that are thought to have generated the observed data. Simulation models (which are a type of algorithmic models) use a series of rules to describe system behavior. Typically, they use programming code to simulate a system. While other authors describe more detailed taxonomies (see, for example, Hilborn and Mangel, 1997), and some models may not fit readily into one of these categories, all models simplify reality to make generalizable or comprehensible conclusions, and will be more or less appropriate for different purposes.

Conceptual and mathematical models typically increase our understanding of systems, but do not have the detail necessary to predict future system states. Mathematical models typically have poor predictive abilities in temporally and spatially explicit settings because they are simplified and designed to explore fundamental dynamics. Statistical and simulation models are often more suitable for prediction. In creating predictive models, one may be tempted to simply fit outputs to empirical data, with limited mechanistic rationale. Turchin (1998) p. 41–45 reviewed model development in the field of movement ecology and described curve-fitting models as empirical or phenomological models, which only consider the observed phenomenon, and no underlying processes. However, without a mechanistic rationale of system processes, no comparison can be made between predictive capability of different phenomological models. Over time empirical models have been abandoned in favor of models that can provide insight into system processes (Turchin, 1998). The goal of developing a predictive model is to balance internal

mechanistic structure with fit of output to empirical data, consistent with the model's intended application.

Evaluating complex, predictive models may not be as simple as comparing outputs to field observations or experimental data (Rykiel, 1996; Batty and Torrens, 2005; Bennett et al., 2013). A complex model built from "the bottom up" may have many opportunities for error propagation that leads to invalid results. Consequently, complex simulation models may benefit from calibration against empirical data (Thiele et al., 2014), especially when used for predictive purposes. If empirical data is abundant, testing multiple endpoints over a range of variability can provide comprehensive model evaluation. Such evaluations may require complex analysis techniques (Marino et al., 2008).

Models are used extensively in monarch butterfly research and conservation planning. Modeled monarch butterfly systems include migration, population size, movement and egg-laying of non-migratory monarch butterflies, and developmental process from egg to adult. Of particular interest are population trends and causes of trends. Understanding or predicting movement of millions of individual monarchs at continental scales requires simplification of system processes. Determining the magnitude of population trends and causes of trends requires interpretation of variation in observed population measures. Many modeling approaches have been employed toward this end. Mathematical models such as matrix models and island chain models have been employed, as have statistical models such as Bayesian statespace models. In this review we describe the strengths and limitations of modeling approaches applied to monarch butterfly research and conservation issues. We synthesize findings of modeling efforts to date and provide suggestions to improve our understanding of monarch butterfly systems.

# APPROACH

We reviewed the literature to provide an update on the state of the art in modeling monarch butterfly systems, with a focus on population dynamics and their unique migratory system. We reviewed papers covering eastern and western North American monarch butterfly populations. To find an illustrative suite of modeling papers, we searched Web of Science for the topic "monarch model," which returned 570 results. From among these 570 results, we searched for papers that used a variety of modeling approaches to answer questions about monarch biology. We selected papers that provided insight into the history of modeling monarch systems, the variety of methods available, and future directions for monarch modeling. While we did not constrain our search to the eastern monarch population, there are comparatively fewer modeling papers for the western monarch population. In this review we also describe and evaluate model assumptions to help identify opportunities to link empirical studies with model development to further understanding and the means to predict future states of monarch systems (Restif et al., 2012).

We describe three classes of models: mathematical, statistical, and simulation. In addition to describing the models and their assumptions, we attempt to identify common conclusions across the modeling efforts. We highlight instances where similar conclusion are reached by different research teams using different models, which suggests higher confidence in understanding and/or predicting future conditions. We synthesize model findings in the areas of conceptual frameworks, population fluctuations, causes of population declines, empirical data needs, and migratory processes. We close by making recommendations for moving forward.

#### MODELING EASTERN NORTH AMERICAN MONARCH POPULATIONS

While a substantial body of research results is available for monarch butterflies, some authors have noted a need for modeling to help improve understanding and inform conservation options (Oberhauser, 2004; Dyer and Forister, 2016). Most field research is undertaken at patch scales, with conclusions extrapolated to larger spatial and temporal scales based on conceptual models of system function. Much of the literature makes implicit assumptions about how monarch systems function, but often those assumptions have not been explicitly stated and tested.

# Mathematical Models

Yakubu et al. (2004) developed a spatially discrete advection model, which can also be characterized as an island chain model, where "islands" are generations of monarch butterflies. Heuristically, the events (births and death) that occur in one generation (island) are passed on to the next generation. The objective of the model was to understand the effects of migration and intraspecific competition on population dynamics. The model estimates population levels of four generations each year with a system of non-linear difference equations incorporating survival, migration, and fecundity. Since the fecundity function is not empirically known, Yakubu et al. (2004) tested compensatory and over-compensatory density dependence and documented the possibility of complex oscillatory behavior and or even chaos (unpredictable population fluctuations) under some conditions, which was not predicted in the modeling framework of (Flockhart et al., 2015; see below). Such dynamics could partly be responsible for the observed fluctuations in the overwintering monarch populations. This modeling effort does not include weather effects, spatial heterogeneity (except in the sense that different generations occupy different locations), or individual and spatio-temporal variation in survival or fecundity. Weather effects could potentially dampen or amplify oscillatory or chaotic behavior. This model advances a hypothesis that the system has the potential to demonstrate unpredictable oscillatory or chaotic behavior, even without drivers such as weather.

Periodic matrix projection models are very common in ecology (Caswell, 2001) and are based on several key assumptions. Matrix models require partitioning populations into separate spatio-temporal groups for which survival and fecundity are considered constant. Partitioning the population into many groups requires substantial effort to empirically estimate survival and fecundity for each group. Williams et al. (2002) note this as an important limiting factor in the development of complex matrix models. (Newman et al. (2014), pp.39) has some excellent discussion on the pros and cons of matrix models. The important distinction between matrix models and other models is that matrix models project the population size, they do not estimate the population size; thus matrix models are highly sensitive to the quality of the data used in their development. Except for matrix models, mathematical models of monarch systems are scarce.

Flockhart et al. (2015) created a matrix model incorporating stochasticity in parameters and density dependence. The model was spatially explicit in that the different migratory and nonmigratory generations over the season were associated with different locations on the North American continent (a type of island chain model similar to that of Yakubu et al., 2004). Three hypotheses for the cause of the population decline were tested: habitat loss in the breeding grounds, habitat loss on overwintering grounds, and climate change on the overwintering grounds. The model predictions were compared with 19 years of adult overwintering monitoring data. While the projected population sizes were not significantly different from the monitored overwintering population size estimates, the projections did not exhibit observed overwintering population fluctuations. Insights from this model include an estimated decline of 14% in population size over the next 100 years, with the reduction in milkweed in the summer breeding range as the most significant driver of decline. As the model currently stands, nearly constant survival rates are used across years, which results in an exponential model with some variation from added stochasticity and density dependence [see Figure 3a of Flockhart et al. (2015)]. Because the variation in survival rates and fecundity that lead to the overwintering population size in Mexico are unknown, the matrix model is unable to capture the full range of population variability. If there was a desire to understand or predict population responses at finer temporal or spatial scales using a similar model, the population could be divided into smaller geographic regions and time periods. However, survival, fecundity, and migration would then need to be estimated for each region and time period.

Oberhauser et al. (2017) also created a matrix model using a Bayesian framework. In this framework the authors used overwintering population monitoring data to help estimate parameters in the matrix model. Because parameters were estimated, the model has statistical and mathematical facets, but is included here for comparison to the model developed by Flockhart et al. (2015). Model parameters were given informative priors developed from published data and expert elicitation. The data supplied to the model to estimate the posterior distributions was the overwintering population size and the Midwest USA egg production index (Pleasants and Oberhauser, 2013). The eastern North America population was divided into four regions and four generations (i.e., an island chain model). The mean of the population growth rate posterior distribution was 0.957. Sensitivity and elasticity analysis showed that, because the system is very complex, many parameters were important to population growth. The most important parameters were spring migration survival from overwintering sites in Mexico to the southern U.S., fecundity, and spring immature survival in the southern U.S. Among seven management scenarios, increasing breeding habitat in the south and north central regions were most important for increasing population growth rate. This matrix model assumed a stable-age distribution with density independent growth and did not include environmental factors. Because this model was tuned to the overwintering population data, it is likely much more realistic than other matrix models, though it still assumes constant survival rates.

### Statistical Models

Many statistical models have increased our understanding of monarch population dynamics. Pleasants and Oberhauser (2013) regression model predicts the overwintering population, based on a Midwest egg production index, with an r <sup>2</sup> = 0.47. The Midwest egg production index is based on surveys in Iowa agricultural fields as reported by the authors, Monarch Larval Monitoring Program data (MLMP; Prysby and Oberhauser, 2004; Stenoien et al., 2015) from non-agricultural locations in the Midwest, and surveys of milkweed in Iowa (Hartzler, 2010). Data from these sources is extrapolated to estimate egg production in the Midwest. The authors note four factors not incorporated into the model that may be associated with the unexplained variance: variability in survival from egg to adult; variability of contributions to the Mexico overwintering population from geographic regions other than the Midwest; variability in fall migration survival; and variability in the conversion factor used to calculate the number of eggs in the Midwest.

Zipkin et al. (2012) used a complex Bayesian Poisson regression model with 17 parameters to model monarch butterfly counts at 90 sites in Ohio from 1996 to 2008. The model's purpose was to further understanding of climatic effects on monarch butterflies during spring migration. This model involved a complex analysis of the correlation between abundance on breeding grounds and climactic factors. Four climate effects were included in the model (spring precipitation, spring growing degree days, summer growing degree days, and summer Palmer Drought Index), along with several interactions and quadratic effects. Wet Texas springs and average Texas spring temperatures produced the greatest monarch abundance in Ohio.

Saunders et al. (2016) advanced the approach used by Zipkin et al. (2012) to predict monarch abundance in Ohio and Illinois based on spring climatic conditions in Texas by testing model predictive ability. Of 16 years of data, they used subsets of 8–15 years of data to generate models and compare model outputs to the empirical data from years not included in model development. The difference between predicted and observed monarch counts was quantified using Bayesian p-values, where a p-value of 0.5 indicated good model fit and p-values < 0.3 or > 0.7 indicated poor fit (i.e., the model is either overestimating or underestimating parameters). The model had good predictive ability for years that had near average spring precipitation and temperatures, or when there was a year with similar spring precipitation and temperature values in the dataset. Model prediction was poor for years with weather parameters that were not similar to any other years in the dataset. The large numbers of parameters may have overfit the model to some degree, causing less predictive ability particularly for years with unusual weather. The model suggested spatial synchrony in Ohio and Illinois, with monarch abundance in the two states more dependent on climatic conditions in Texas than local Midwest conditions. While the models developed by Saunders et al. (2016) and Zipkin et al. (2012) may provide our best statistical evidence of climate effects on monarch butterfly population responses, these efforts also highlight the difficulty of modeling a continental-scale migratory system. Additional models could be tested for their predictive ability as in Saunders et al. (2016).

Semmens et al. (2016) fit a Bayesian state-space statistical model to egg counts and overwintering population size for the purpose of estimating extinction risk. The model was first-order auto-regressive, meaning the overwintering population from 1 year was estimated using the overwintering population from the previous year. The observed overwintering population and Midwest egg production index (Pleasants and Oberhauser, 2013) informed estimation of the true value of the overwintering population. This model quantified the variation and trends in observed overwintering population size using a normal distribution and no underlying mechanisms, and then projected trends into the future. This model estimated a mean population growth rate of 0.94, with a 66% probability of the average annual growth being below 1.0 and an 11–57% probability of quasi-extinction over 20 years. This model accounted for yearly variation in overwintering population size using a random deviate each year that was assumed to be normally distributed. No environmental covariates were used and the model did not include any demographic mechanics such as survival and fecundity.

Inamine et al. (2016) compared North American Butterfly Association<sup>1</sup> (NABA) citizen science monitoring data from the eastern U.S. and historical counts from Cape May, New Jersey, and Peninsula Point, Michigan over different time periods and regions to link population dynamics across the annual

<sup>1</sup>www.naba.org

migratory and breeding cycles to help elucidate causes of the population decline. They divided the yearly cycle into successive regions that represent the location of the monarchs over the year: Spring Mexico, Spring South, Midwest and Northeast, Fall South, and Fall Mexico. A string of regression analyses between each successive area were undertaken (i.e., an assumed donor-recipient relationship, similar to the island-chain model of Yakubu et al., 2004) to determine if the population counts in a region can be predicted from the counts in the previous spatiotemporal area. The Midwest summer NABA counts showed no statistical relationship with Mexico overwintering counts over the period of 1993–2014. Inamine et al. (2016) concluded that unknown factors must be increasing mortality rates during fall migration. Agrawal and Inamine (2018) restated the arguments in Inamine et al. (2016) that factors during fall migration may be partly driving population decline.

Ries et al. (2015) conducted an analysis similar to Inamine et al. (2016) using NABA data and linear regression between successive spatio-temporal stages of the annual monarch cycle. Their findings were similar to those of Inamine et al. (2016); however, Ries et al. (2015) and Pleasants et al. (2017) cautioned that bias in population count data could be causing the discontinuity between the U.S. Midwest summer population counts and the overwintering population size. While the modeling results of Inamine et al. (2016) and Ries et al. (2015) help form the basis of a migratory failure hypothesis (Agrawal and Inamine, 2018), other modeling efforts indicate loss of milkweed in the summer breeding range as the primary stressor, due to either Roundup Ready crops (e.g., Pleasants and Oberhauser, 2013; Flockhart et al., 2015) or modern agricultural practices since the 1950's (Boyle et al., 2019; but see Ries et al., 2019; Wepperich, 2019). Collectively, these models highlight uncertainty in data used to develop and evaluate models (e.g., the ability of current monitoring designs to quantify monarch migratory patterns across the eastern United States), as well as uncertainty in the independent or interacting roles of reduced nectar sources (Brower et al., 2015), road side mortality (Kantola et al., 2019), spatial-temporal climatic variability (Zalucki and Rochester, 2004), and reduced milkweed (Lemoine, 2015), in monarch population trends.

The environmental niche model of Batalden et al. (2007) used MLMP monitoring data on site occupancy status (whether a site had ≥ one egg or no eggs) to estimate the area of the monarch ecological niche in the eastern U.S. over successive generations. Geographic Information System raster data layers of climatic and topographic parameters were tested for correlation with occupied sites. The variables of maximum, minimum, and mean monthly temperatures; precipitation; elevation; and slope were correlated with occupied sites to predict month-specific models of the area in the eastern U.S. suitable for monarch occupation. The monthly niche models indicate that monarchs follow their preferred climatic niche during the breeding season, but switch climatic preferences during the winter months. Batalden et al. (2007) then looked at changes in the environmental niche using models of future climatic conditions. These models indicate that by 2055, more niche area will be available from March to June. The area of the monarch niche will be similar in July to August, but will extend farther north into Canada, nearly to Hudson Bay. The potential consequences for spatial discontinuity in monarch and milkweed range during July and August remains an unresolved issue.

# Simulation Models

Simulation models attempt to create a simplified representation of a monarch butterfly system using algorithms (i.e., a set of rules). By manipulating the algorithms, researchers hope to gain insights on natural processes; e.g., movement behavior of adult monarchs. Models have been used to simulate movement at landscape scales and monarch development and colonization at continental scales. Adult butterflies are well-suited for agentbased modeling (also known as individual-based modeling), which is a type of simulation model. Adult butterflies were the subjects of some of the first agent-based models (because they move freely across the landscape, unlike terrestrial species whose movement paths are highly restricted by terrain; Jones, 1977; Jones et al., 1980). Insects are also thought to have limited capacity for memory (Collett et al., 2013); hence incorporating memory does not significantly increase model complexity. As a consequence, butterfly movement can often be simulated using random walk assumptions (Codling et al., 2008).

Pioneering efforts to simulate monarch butterfly movement and egg-laying in a spatially explicit environment were reported by Zalucki (1983). Zalucki's model incorporated movement, egg-laying, and immature survival to the adult stage in a spatially explicit environment. The environment was a 3.1 km<sup>2</sup> circular area with large and small patches of milkweed with interspersed individual milkweed plants. Using a 1 s time step, the model included many biological processes, from male-female interactions to the time needed to lay eggs. Key findings were that low directionality of flight increased fitness when milkweed was spread out, while high directionality increased fitness when milkweed patches were more clumped.

More recently, Zalucki and Lammers (2010) and Zalucki et al. (2016) used agent-based models to explore the effect of reduced milkweed in the matrix (i.e., corn and soybean fields in the U.S. Midwest) between patches of milkweed. Zalucki and Lammers (2010) predicted that clearing milkweed from the matrix would result in ∼20% reduction of eggs laid. Zalucki et al. (2016) presented a more detailed model to advance understanding of movement and egg laying in hypothetical 11.2 km<sup>2</sup> landscape scenarios with the same area of milkweed distributed in different sizes of milkweed patches in varying spatial patterns. This model suggested a 30% decline in monarch lifetime egg production in a scenario when large milkweed patches were consolidated in space, as compared to a scenario where small patches were uniformly distributed in the landscape.

Grant et al. (2018) subsequently developed an agent-based model to predict movement of non-migratory, summer breeding female butterflies and egg-laying patterns in a spatially explicit Iowa USA agricultural landscape. Movement decisions were based on assumptions of monarch perceptual range for detecting milkweed and spatial memory. Monarchs probabilistically choose habitat patches with more milkweed and have a lower probability of returning to patches they have already visited. This movement

FIGURE 2 | The rural road in Boone County in which egg density data was collected by Blader (2018). Red squares (50m X 50m) outline patches in the agent-based model developed by Grant et al. (2018). Empirically-measured egg density for each patch is compared to predicted egg density to calibrate the model.

algorithm was more complex than Zalucki et al. (2016) in that monarch agents could choose between multiple habitat types within their perceptual range. In assessing performance of the model algorithms, patterns that needed to be reflected in model outputs (sensu Grimm et al., 2005) included: female monarchs moving long distances (10 km/day); expressing vagile behavior, i.e., not remaining in a selected habitat patch; and laying eggs widely across the landscape. By tying monarch movement to habitat within their perceptual range substantial realism was added to the model. Random walk models are commonly used in biology (Codling et al., 2008), but in previous research, model agents have not responded to habitat heterogeneity (see review by Wallentin, 2017). The use of a spatial memory algorithm by Grant et al. (2018) was also a unique addition to agent-based modeling and contributed to the vagile behavior of the monarch agents. The model does include several assumptions and constraints. First, the minimum size of habitat patches is 50 × 50 m, consequently individual milkweed or very small patches of milkweed are not explicitly included in the model. Rather, milkweed stems are assumed to be homogeneous within habitat patches, though adjusting egg-laying probabilities can help account for habitat heterogeneity. Second, monarch movement is modeled as a fixed-length step of 20–50 m. The model could be improved and further evaluated with empirical research data on adult movement behavior, which could enhance the mechanistic basis for the movement algorithms. Survey data on milkweed and monarch egg density in different, adjoining landcover classes could also improve model calibration, as discussed below.

To calibrate the model described by Grant et al. (2018) requires empirical data that is difficult to obtain because it requires intensive and large-scale sampling. More specifically, egg densities in milkweed patches within adjoining landcover categories is needed. Subsequent to publication of Grant et al. (2018), information provided in Blader (2018) provided the means to calibrate model parameters to field data. Blader (2018) geolocated milkweed and monarch eggs in two 1.6-km rural roadsides in rural Story and Boone counties Iowa in 2017 (**Figure 2**). This data was converted to milkweed and eggs per ha to be compatible with the model input and output for estimates, respectively, for roadsides adjacent to corn and soybean fields. The calibration process first involved running the model using parameters in Grant et al. (2018) for these spatially explicit roadsides and comparing the predicted eggs laid to the observed eggs laid in the same roadsides. A model parameter that establishes a probability for an agent to lay an egg within a landcover type, which is based on the assumed milkweed density, was then adjusted such that subsequent model predictions of eggs laid in these landcover types are similar to the eggs observed in the monitored roadsides. Calibration substantially improved the predictive ability of the model (**Figure 3**). Prior to calibration,

model predictions of egg densities when milkweed density was 20–240 milkweed/ha were approximately 2 to 5 times higher than reported by Blader (2018). After calibration, predicted egg densities were within a factor of 2 compared to the empirical densities. A more robust calibration data set would include measured milkweed and egg densities in the adjoining landcover types (e.g., crop fields, pastures, and restored prairies), as well as the roadsides.

Zalucki and Rochester (2004) presented a Monte Carlo model of eastern North American monarch dynamics from spring colonization of the eastern U.S. until fall migration. This was the first model to explicitly include multiple generations, weather, and developmental dynamics in a single model. Probabilities were assigned to describe components of the system, including colonization probabilities, egg-laying probabilities, and probabilities of developing from egg, through 1st−5th instar, and adult, for each of the generations. Populations were simulated at 187 weather station locations where temperature and other climatic data defined colonization probabilities and developmental rates. A single simulated monarch was then run using a Monte Carlo approach. The proportion of times the monarch was found in a particular state was assumed to represent the proportion of the monarch population that would be at that location. The model explained only a very modest portion of the variation in the observed data (r <sup>2</sup> = 0.12, p < 0.001), but provided key findings that climate is a powerful driver of yearly population dynamics and that breeding generations quickly overlapped, due to individual variation in stage duration.

Feddema et al. (2004) developed perhaps the most comprehensive model of spring migration and development. The eastern U.S. was divided into 50 square mile grids. Fifty miles was chosen to match the assumed daily movement of migrating monarchs. Occupancy status (occupied or not) and population size was calculated for each grid over the course of the year. The model queried National Climate Data Center<sup>2</sup> databases from weather stations nearest to each grid cell. The model makes several assumptions. First, arrival time is defined by solar angle rather than distance from the overwintering site. Consequently, arrival date is determined by latitude. Second, once a grid cell becomes occupied, it stays occupied until fall migration. Third, daily movement is assumed to be 50 miles throughout the year; the authors note the movement step is not realistic. Fourth, only mortality due to temperature is included. Fifth, one new cohort is laid in a grid cell each day after the cell becomes occupied. While this is a unique modeling approach, the model failed to capture many patterns observed in the empirical data. Comparison of model results to Journey North<sup>3</sup> datasets showed discrepancies between predictions and first observations of spring-migrating monarchs. Monarch production estimates from the model are not significantly correlated to empirical data. The model did show that natural variation in temperature can lead to different arrival times at more northerly latitudes.

# MODELING WESTERN NORTH AMERICAN MONARCH POPULATIONS

There are comparatively fewer published population models of the western population. Espeset et al. (2016) used Bayesian hierarchical modeling and path analysis to estimate trends in western monarch populations and the effects of weather on those trends. The data for the analysis was from sites along a transect across Northern California which has been surveyed biweekly for monarch butterfly presence during flight season for 27–42 years, and publicly available overwintering population numbers. They included climate data as predictor variables to explore their effect on the population trend. The model estimated a continuing downward trend in western monarch populations on the Central Valley breeding grounds. A sliding-window regression analysis indicated that the largest decline in numbers occurred in spring months. The authors concluded that factors were mostly strongly affecting western monarchs during overwintering and spring, but that populations tended to rebound to normal levels during the summer. Warmer winter and spring temperatures and higher spring precipitation were correlated with more monarchs observed at the survey sites. However, the path analysis revealed that climate effects were not the strongest factor in the declining population. In other words, after accounting for climate effects, populations still trended downward. The cause of this downward trend is poorly understood, but because the downward trend is most pronounced in the spring, the authors recommend looking at overwintering sites for causes of losses. Poor years for western monarchs were not correlated with poor years for eastern monarchs, prompting the conclusion that the two populations are influenced by different factors.

<sup>2</sup>www.ncdc.noaa.gov

<sup>3</sup>http://journeynorth.org

Schultz et al. (2017) developed a multivariate auto-regressive state-space model using data collected on western monarch overwintering populations, which is similar in approach to that used by Semmens et al. (2016). Survey efforts varied substantially across overwintering sites. Some sites were surveyed every year for many years while some sites were surveyed intermittently over the years. The statistical framework employed in the model addressed data obtained through varying levels of monitoring effort for individual overwintering sites. Average population growth rate from 1981 to 2016 was estimated to be 0.927 and the estimated quasi-extinction risk within 20 years was 0.72. Schultz et al. (2017) noted high variation in annual population growth rates may be due to climate factors, which were not included in their models. They also note that vital rates, such as survival, are poorly understood for eastern and western monarchs.

#### SYNTHESIS OF MONARCH POPULATION MODELS

Significant effort has been invested in developing monarch butterfly models that can improve our understanding and means to predict population dynamics and trends in eastern North America. Nevertheless, there is no standard model framework of monarch population dynamics. Authors have repeatedly stated the need for a full model of the monarch annual demographic cycle that includes weather effects. As Zipkin et al. (2012) stated: "No modeling approach has yet captured the full complexity of how climate interacts with all the potential factors that influence monarch population growth, including the condition and number of incoming migrants from Mexico, milkweed growth and congruence with monarch arrivals, natural enemies, and appropriate climatic environments for activity and growth throughout each phase of their migratory cycle." Inamine et al. (2016) stated: "Understanding the complex population dynamics of monarchs over space and time therefore remains an important ecological as well as conservation challenge." Malcolm (2018), in the context of determining the risk monarchs suffer from anthropogenic factors, stated: "Risk for this highly mobile species has to be put into the context of a complex life history across relevant time and space for both eastern and western populations of monarchs in North America."

Consistent with these observations, we attempt to synthesize several themes from the models summarized in this review.

# Conceptual Frameworks Underlying Mathematical, Statistical, and Simulation Models

Conceptual models of monarch population dynamics implicitly or explicitly acknowledge that weather patterns influence continental to local patterns in system responses, yet few models have included weather effects. Many authors have noted that weather is needed to better predict observed population patterns (Zalucki and Rochester, 2004; Zipkin et al., 2012; Stenoien et al., 2015; Espeset et al., 2016; Semmens et al., 2016; Schultz et al., 2017). This indicates a significant discrepancy between what may be a critical driver of population dynamics and what we can currently model to improve understanding and predictions.

Monarch models tend to have a "top-down" perspective in exploring large-scale trends with uncertain empirical data. Topdown models look at broad scales, either national or continental and tend to be correlative and not mechanistic. For example, Oberhauser et al. (2017) is essentially a correlation between overwintering sites and breeding ground data, with matrix population mechanics constraining the relationship. Zipkin et al. (2012) and Saunders et al. (2016) are also correlative analyses between abundance on the breeding grounds and Texas climate data. Saunders et al. (2016) laudably tested the predictability of their model. These models do provide important advances to understanding the system, but the capacity to predict system responses remains challenging. The complexity of the system necessitates a correlative approach to begin improving understanding; however, development of simulation models including system mechanics may provide the best return on investment. The Zalucki and Rochester (2004) model included developmental mechanics in a national scale model. Zalucki et al. (2016) took the critical step of incorporating movement mechanics in their agent-based model, followed by inclusion of improved movement mechanics into Grant et al. (2018). Improving the mechanistic basis of our models, instead of using only a correlative approach, will likely improve understanding and prediction.

Most mathematical and statistical models use an "island chain" approach in which different generations and areas of the U.S. are modeled sequentially. After the first generation, however, monarch generations overlap and may become indistinguishable in late July and August (Zalucki and Rochester, 2004). Consequently, spatial areas in an "island chain" approach are most appropriate at large, regional scales (e.g., Midwest, Northeast). However, there is no standard definition of regions and different modelers have divided the eastern U.S. into "islands" of different areas. In theory, the island chain approach could provide greater insights using shorter timeframes and smaller geographical divisions; however, as islands become smaller, they become more artificial.

With increasing availability of high-performance computing, simulation modeling (e.g., agent-based modeling) may be a viable alternative to the island model approach to advance understanding and prediction. The process of parameterizing a simulation model would also help inform the lack of system understanding and identify high-priority research needs. A good example of this process in monarch butterfly research can be seen in the work of Drury and Dwyer (2005), who developed models and generated data to understand mechanisms behind spring monarch colonization of habitat patches. Developing a model for the entire eastern monarch population would be a challenging undertaking, spanning multiple nested spatio-temporal scales. To efficiently build the capability to model the annual monarch cycle with agent-based modeling, different modules would need to be developed that can be used concurrently. For example, migratory and non-migratory modules could be developed with the migratory module outputs serving as inputs for the nonmigratory module.

# Modeling Population Fluctuations

To date, published models have limited ability to predict annual fluctuations in overwintering population counts. Matrix models have not predicted population fluctuations because variability in survival rates from year to year is unknown (e.g., see Flockhart et al., 2015). Yakubu et al. (2004) showed that the system has the potential to generate fluctuations and chaotic trends without any explanatory variables. Semmens et al. (2016) modeled population dynamics using a first-order auto-regressive model to investigate population growth rate but modeled annual fluctuations as normally distributed with no mechanistic rationale. Weather patterns have been shown to influence the spring migration and define the monarch's environmental niche (e.g., see previous summaries of Zalucki and Rochester, 2004; Batalden et al., 2007; Zipkin et al., 2012; Saunders et al., 2016). Population fluctuations are likely driven by changes in survival and fecundity under different weather conditions. Improved methods for estimating demographic rates—survival, in particular—are needed to adequately model monarch population dynamics. It seems likely that weather patterns contribute to the overwintering population fluctuations as well; however, to date no attempts have been made to use weather patterns over the entire annual cycle to better understand or predict overwintering population levels. Without an improved understanding how weather drives fluctuations in the overwintering population, it will be difficult to predict how climate change will affect future monarch butterfly population dynamics.

#### Causes of Population Decline

Some modeling approaches support the hypothesis that milkweed decline within the summer breeding range is the primary cause for the decline in the eastern population (e.g., Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Thogmartin et al., 2017; Malcolm, 2018; Stenoien et al., 2018). Other modeling efforts suggest that adverse effects during the fall migration have driven the decline (Ries et al., 2015; Inamine et al., 2016; Agrawal and Inamine, 2018; Saunders et al., 2019). Uncertainties in underlying empirical data and assumptions, combined with different approaches for including uncertainties in the modeling efforts, contribute to varying interpretations of the causes for the population decline.

The NABA data used by Inamine et al. (2016) and Ries et al. (2015) may be biased (see Ries et al. for discussion of potential bias). The NABA data are counts of adult monarchs at sites selected by citizen scientists and have two potential limitations. First, because detection probability is not estimated, the amount of bias in the counts is not known. Second, the sites are not chosen using a probabilistic sampling design; consequently, the relationship of the counts to the true monarch population is unknown. The limitations of using counts with no estimates of detection probability has been noted for numerous species (MacKenzie and Kendall, 2002), including butterflies (Kery and Plattner, 2007; Nowicki et al., 2008; Pellet, 2008; Pellet et al., 2012; Kral et al., 2018). One likely mechanism that would bias detection probability is monarch movement becoming more confined to NABA count locations as milkweed in agricultural fields decreased (Pleasants et al., 2017). There may be additional factors biasing counts; however, without empirical data on monarch densities, which currently seems difficult or impossible to obtain, it may be difficult to determine the presence and causes of bias. Estimating detection probability for adult monarchs in the near term seems unlikely because of their vagile behavior, though statistical methods to address this issue continue to advance (Rossman et al., 2016). The monarch research community is confronted with the difficult task of resolving the extent to which uncertainty in the NABA counts undermines the migration failure hypothesis. Exploration of hypothesized causes of mortality during migration, such as nectar source declines or roadkill mortality, could provide independent support for the migration failure hypothesis.

Other research groups have concluded that milkweed limitation is the primary driver of monarch decline (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Thogmartin et al., 2017; Malcolm, 2018; Stenoien et al., 2018). Additional potential factors for the decline, while acknowledged, have largely been considered insignificant. The Pleasants and Oberhauser (2013) model estimates that the Midwest egg production index accounts for 47% of the observed variation in overwintering population size. These researchers acknowledge that their model does not account for variability in survival from egg to adult; variability of contributions to the Mexico overwintering population from other geographic regions; variability in fall migration survival; and variability in the conversion factor used to calculate the number of eggs in the Midwest.

Ideally a model constructed to quantify variation in overwintering population due to these other factors would be helpful; however, there are limitations in current knowledge to support such a modeling effort:


reached the overwintering sites is unknown. These papers suggest that nectar source availability on breeding grounds and the migration route has the potential to have a negative effect on monarch populations.


Perhaps a model could be constructed utilizing different data (e.g., NABA adult counts, MLMP egg and larval counts, milkweed density, roadkill mortality rates, etc.); however, such a model is more likely to be successful using monitoring data derived through a probabilistic survey design.

### Empirical Data

Empirical survey data on monarch butterflies of all stages is limited and the inferential value of some of that data has been questioned (Pleasants et al., 2017). Most inferences on monarch butterfly populations come from a few sources: overwintering population size measurements, citizen science counts of adults (e.g., NABA, Illinois and Ohio programs), and citizen science counts of larval stages (e.g., MLMP).

Much of the monarch data available is from citizen science programs and it is difficult to imagine how substantial data could be collected on monarch populations without these efforts. Data from citizen science programs may have limitations due to spatial sampling bias, because volunteers choose the survey locations, and there are no estimates of detection probability. Other citizen science programs have recognized these limitations. eBird, perhaps the largest citizen science program in the world (Sullivan et al., 2009), has recognized the imperfect detection issue, among other potential limitations (Sullivan et al., 2014). However, it is encouraging that studies have shown that eBird data can provide very similar results to standardized shorebird surveys (Callaghan and Gawlik, 2015). Detection probability for monarch eggs and larvae could be determined experimentally. Detection probability for adult monarchs is a difficult problem because of their vagile behavior. Monarch butterfly modeling research would be well served by data from monitoring programs with probabilistic sampling. A monitoring program under development by Monarch Joint Venture, USGS, and USFWS will hopefully fill this need (Cariveau et al., 2019). Analysis of MonarchWatch<sup>4</sup> mark-recapture data, another citizen science program, would also be a welcome addition to the monarch data available for modeling efforts.

More data would also be useful for refining and calibrating algorithms in simulation models. Data on monarch movement patterns and behavior could help improve the movement algorithm in Grant et al. (2018), which is the mechanistic basis for the model. Data on milkweed density and egg density in a diversity of adjoining landcover types would improve calibration of landscape-scale simulation models. Milkweed and egg survey design and model frameworks need to be reconciled to ensure spatial and temporal scale concordance.

#### Modeling Monarch Migration

Modeling monarch migration has generally proceeded with an island chain approach (e.g., Flockhart et al., 2015; Oberhauser et al., 2017). Zalucki and Rochester (2004) modeled colonization probability over time in at weather station locations in the eastern U.S. These approaches omit the mechanics of movement (Zalucki et al., 2016). An agent-based model of monarch migration could include the important mechanics of movement and treat monarchs as individuals rather than very large "islands" in which each individual is assumed to be the same. Modeling migration at continental scales is a non-trivial task. Perhaps modeling the migration of the smaller western North American monarch population may be a more tractable starting point.

Western monarchs are declining (Espeset et al., 2016; Schultz et al., 2017) and the 2018 Thanksgiving counts found only ∼28,429 monarchs<sup>5</sup> With so few monarchs spread over the entire western U.S., modeling may help interpret current information and predict future conditions because a limited amount of survey data can be obtained. Pattern-oriented modeling could be a viable option. Grimm et al. (2005) describe pattern-oriented modeling as an approach that incorporates important system patterns; patterns exhibited by systems are likely the result of underlying mechanisms. Grimm et al. (2005) give an example of patterns used to model spatio-temporal patterns in European beech forests. Important patterns were: (1) the spatial mosaic of forest patches at different successional patterns, (2) different successional stages had different forest structure, and (3) when large individual trees fell, they created canopy openings. After incorporating these patterns, the model of beech forest dynamics predicted age structure and distribution of large old trees well and led to new understanding and hypotheses about the system. Incorporating important patterns gives a model robustness under varying conditions, sometimes even without in-depth calibration or evaluation. Some considerations for a potential pattern-orientated model of the western monarch migration that could be used to assess different hypotheses of annual monarch movements are summarized below.

<sup>4</sup>www.monarchwatch.org

<sup>5</sup>https://xerces.org/2019/01/17/record-low-overwintering-monarchs-incalifornia/

The western monarch population overwinters along central and southern coastal California and spreads throughout the U.S. west of the continental divide during the summer (Reppert and de Roode, 2018). The most important breeding habitat has not been identified (Jepsen and Black, 2015), though recent efforts to map milkweed in the west have been undertaken (Dilts et al., (In Prep)). Western monarchs are genetically very similar to eastern monarchs (Lyons et al., 2012) and may have navigation strategies similar to eastern monarchs, or may have strategies adapted to the west. The mountains and deserts of the western U.S. strongly constrain movement patterns (Dingle et al., 2005). Overwintering monarchs originate from a variety of places in the western U.S. Stable isotope ratios demonstrate that western monarchs originate from four isoscapes in the western U.S.: 40% originate from the "northern inland range," 30% from the "southern coastal range," 16% from the "central range," and 12% from the "northern coast and southern inland range" (Yang et al., 2016).

Following Grimm et al. (2005), several patterns that are important in the yearly cycle of western monarch migration should be reflected in a pattern-oriented model:


We hypothesize that fine-scale monarch movement constrained by landscape and weather factors results in the observed patterns. Annual migration strategies are the primary driver of observed patterns. Testing different conceptual models of monarch movement and migration strategies for consistency with these patterns can determine which conceptual models of monarch movement are most accurate, providing inference on monarch migration strategies and improving model robustness. We propose several conceptual models of western monarch migration to test against these patterns. Some models are based on eastern monarch research, under the assumption that western monarchs may use the same migration strategies as eastern monarchs:


In a simulation model, western monarch agents could be programmed with directionality according to the four conceptual models highlighted above. For example, monarch agents following strategy 4 would have a strong northeast directionality during the spring months, but would have a higher probability of staying at the same elevation instead of gaining altitude to maintain a base directionality. Individual variation in directionality (Froy et al., 2003) and probabilistic movement decisions would result in monarchs spreading over the landscape. For each movement step, agents would have a probability of laying eggs depending on the habitat. Each egg would be a new agent, with developmental rate dictated by temperature and survival rate. Eggs that survive would become adult monarch agents and begin moving across the landscape. As directionality fades over the summer, simple diffusion along riparian corridors would occur.

An agent-based model could test the proposed models of the western monarch annual movement cycle. Once a model is found that matches observed patterns, the model could identify which areas are most important to monarch migration and breeding. Modeling movement and demographics of monarchs over the year could increase understanding of important movement and migration corridors and increase understanding of important breeding habitat. Such a model could also address questions about the annual movements of western monarchs. Model results could be tested with empirical surveys and conservation steps could be prioritized accordingly.

# ADVANCING MONARCH BUTTERFLY MODELING

While monarch modeling using mathematical, statistical and simulation approaches has greatly increased understanding of monarch butterfly systems, improved approaches are needed to understand the causes of monarch population trends, including the effects of climate change and other stressors, and to predict population responses as a function of alternate conservation plans. Most statistical modeling to date has been largely correlative with little mechanistic basis. Mathematical and statistical models have used an island chain conceptual framework, but new understanding and prediction may require an agent-based approach. Few models have incorporated weather effects, though most authors acknowledge that they believe weather to be a primary driver of population fluctuations. Agent-based modeling with built-in demographics and weather effects could address this need. Incorporating movement patterns and behavior in agent-based models provide an important mechanistic basis (Zalucki et al., 2016). A model of western monarch migration could serve as an important developmental step to advancing agent-based modeling of eastern monarch migration. A model of eastern monarch migration could incorporate existing agent-based models for non-migratory generations.

What are the challenges to advancing such a modeling effort? Mechanisms that need to be incorporated include movement behavior, developmental dynamics, density dependent effects, effects of nectar resources on movement behavior, spatial memory effects, behavioral interactions between individuals, predation dynamics, and land-use stressors, such as pesticides and mowing. It is not clear that all of these effects are possible or necessary to include in agent-based models, but a conceptual model could begin to address these factors systematically. Models could be developed with nested modules to address these issues and their interactions. Moving forward on collecting data to refine model algorithms and calibrate and evaluate models is a significant challenge. Advancing understanding and predictive capabilities of the monarch system will require well-planned

REFERENCES


collaboration between laboratory and field-based research and modeling teams. The continued existence of monarch butterflies as a migratory phenomenon will likely require new approaches to develop the necessary understanding and predictive capability to determine those conservation actions necessary to reverse population declines in eastern and western North America. International teams with a diversity of backgrounds are well positioned to develop new conceptual models of monarch systems to address these challenges.

### AUTHOR CONTRIBUTIONS

Many of the ideas in this paper have been developed over time by discussions between the authors. TG had primary responsibility reviewing the cited journal articles and wrote the initial draft. SB provided secondary reviews of articles, guidance, editing, and additional discussion and development of ideas for the manuscript.

### FUNDING

This work was supported, in part, by the Agriculture and Food Research Initiative Pollinator Health Program (Grant No. 2018-67013-27541) from the USDA National Institute of Food and Agriculture. The authors also acknowledge support from the College of Agriculture and Life Sciences at Iowa State University and the Iowa Monarch Conservation Consortium.

#### ACKNOWLEDGMENTS

Discussions with the Monarch Butterfly Working Group at Iowa State University and many monarch butterfly researchers aided in development of concepts in the manuscript.

in Mexico," in Monarchs in a Changing World: Biology and Conservation of an Iconic Butterfly, eds K. S. Oberhauser, K. R. Nail, and S. Altizer (Ithaca, NY: Cornell University Press), 117–129.


Pyle, R. M. (1999). Chasing Monarchs. Boston, MA: Houghton Mifflin Company.


**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 Grant and Bradbury. 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.

# Quantifying Pesticide Exposure Risk for Monarch Caterpillars on Milkweeds Bordering Agricultural Land

#### Paola Olaya-Arenas\* and Ian Kaplan

*Department of Entomology, Purdue University, West Lafayette, IN, United States*

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*Christina Mogren, University of Hawaii at Manoa, United States Philip Neil Smith, Texas Tech University, United States*

> \*Correspondence: *Paola Olaya-Arenas polayaar@purdue.edu*

#### Specialty section:

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

Received: *31 December 2018* Accepted: *29 May 2019* Published: *14 June 2019*

#### Citation:

*Olaya-Arenas P and Kaplan I (2019) Quantifying Pesticide Exposure Risk for Monarch Caterpillars on Milkweeds Bordering Agricultural Land. Front. Ecol. Evol. 7:223. doi: 10.3389/fevo.2019.00223* Monarch butterflies are undergoing a long-term population decline, which has led to a search for potential causes underlying this pattern. One poorly studied factor is exposure to non-target pesticides on their primary host-plant, the common milkweed *Asclepias syriaca*, during larval development. This species frequently grows near agricultural fields in the Midwestern U.S., but the spectrum of pesticides encountered by monarch caterpillars on milkweed leaves is unknown. Further, it is unclear whether pesticide exposure can be avoided by isolating restored milkweed patches at sites far from cropland. Over 2 years, we analyzed 1,543 milkweed leaves across seven sites in northwestern Indiana for the presence and concentration of a range of commonly used agricultural insecticides, fungicides, and herbicides. Additionally, we tested the ability of local (i.e., nearest linear distance to crop field) and landscape-level (i.e., % of corn/soybean in 1 km radius) variables to predict the presence of pesticides on focal milkweeds. Overall, we detected 14 pesticides−4 insecticides, 4 herbicides, 6 fungicides—on milkweeds that varied widely in their prevalence and concentration. The neonicotinoid clothianidin, the only pesticide for which toxicity data are available in monarchs, was detected in 15–25% of plants in June with nearly 60% of milkweeds at some sites testing positive (mean conc. = 0.71 and 0.48 ng/g in 2015 and 2016, respectively); however, no samples from July or August contained clothianidin. The related neonicotinoid thiamethoxam and the pyrethroid deltamethrin were detected in most (>75%) samples throughout the season, but only in the second year of the study. For thiamethoxam, isolating milkweeds 50–100 m from the nearest corn or soybean field tended to decrease the concentration and likelihood of detecting residues, whereas landscape composition surrounding milkweed sites had comparatively weak predictive power. These data suggest that monarch caterpillars frequently consume a diversity of pesticides in their diet; the lethal or sublethal impacts of this exposure remain to be tested.

Keywords: milkweeds, agriculture, pesticides, monarchs, landscape

# INTRODUCTION

Since 1960, agricultural intensification and a corresponding rise in pesticide use has been an environmental concern due to contamination of soil-water-air and movement of chemicals through the trophic chain (Carson, 1962; Krupke et al., 2007; Epstein, 2014; Douglas et al., 2015). Because broad-spectrum pesticides are, by nature, not specific to focal pests, they can affect non-target beneficial organisms (i.e., pollinators, parasitoids, predators) inhabiting crops, as well as unmanaged habitats neighboring agricultural land (Longley and Sotherton, 1997; Aktar et al., 2009). Routes of exposure are varied and challenging to track, but include direct contact with contaminated surfaces or spray droplets, residues remaining on the soil, and consumption via food resources such as leaves, nectar or pollen (Cilgi and Jepson, 1995; Longley and Stark, 1996). In many cases, only a small fraction of active ingredient makes contact with target pests, while the remainder is absorbed by the greater ecosystem. Pesticides applied by aircraft, for example, can reach as little as 50% of the target crop with the remainder moving to surrounding areas as far as 30 km downwind (Pimentel and Levitan, 1986; Pimentel, 1995). As a result, a range of insect pests, from aphids to caterpillars, are estimated to contact <0.1% of insecticides applied for their control (Pimentel and Levitan, 1986). Even newer, more targeted technologies are vulnerable to this pesticide 'loss'; namely, seed treatments that were once touted for their limited off-site drift (Jeschke and Nauen, 2008). New data estimate that only 1.3% of initial seed treatment is recovered from corn plants exposed to the neonicotinoid clothianidin, with the remaining 98–99% of material leached into the environment (Alford and Krupke, 2017).

Off-site exposure to mobile insecticides is particularly a concern for pollinators, many of which inhabit agricultural landscapes and are undergoing long-term population declines. Several studies provide evidence of lower abundance and/or diversity of butterflies in the field margins of insecticide-treated crops compared with unsprayed controls (Rands and Sotherton, 1986; Dover et al., 1990; De Snoo et al., 1998). In most cases, it is unknown whether effects are caused by exposure to adults nectaring on flowering plants or larvae developing on contaminated leaves. However, a field experiment exposing Pieris brassicae caterpillars at different distances downwind to spray drift from the insecticide diflubenzuron, showed higher mortality when developing on leaves of their host-plant up to 16 m from the field edge (Davis et al., 1991). Similarly, several studies illustrate that the nectar and/or pollen of wild flowering plants on crop field edges contain residues of neonicotinoid insecticides among other agrochemicals (Krupke et al., 2012; Botías et al., 2015, 2016; David et al., 2016; Mogren and Lundgren, 2016). Indeed, much of the recent focus of nontarget impacts on pollinators centers on the neonicotinoids, due in large part to their widespread adoption in global agriculture (Douglas and Tooker, 2015). Although this work has primarily targeted bees, increasing evidence suggests that butterflies are also affected. Two recent time-scale analyses of reductions in butterfly diversity over the past several decades link these changes with the introduction and rise of neonicotinoids in the UK (Gilburn et al., 2015) and California (Forister et al., 2016). These correlative analyses were complimented by a few experimental lab studies showing strong negative effects on larval development for butterflies reared at fieldrealistic exposure levels for clothianidin (Pecenka and Lundgren, 2015) and imidacloprid (Whitehorn et al., 2018). Yet, due to the strong research emphasis on bees and pollen/nectar composition, we still lack field data on dietary exposure to pesticides for butterfly larvae developing on leaves of host-plants bordering cropland.

The common milkweed Asclepias syriaca L. is an abundant and opportunistic herbaceous plant growing in disturbed agricultural areas throughout the eastern United States (Woodson, 1954). It is notorious as being the primary larval food plant for the migratory monarch butterfly (Danaus plexippus L.) throughout its summer breeding range (Seiber et al., 1986; Wassenaar and Hobson, 1998). While Asclepias is a relatively diverse genus in North America and monarchs are capable of feeding on most, if not all, of these species, A. syriaca is by far the most widely available and used by monarchs in the Midwestern U.S. (Hartzler and Buhler, 2000; Zaya et al., 2017). Because A. syriaca grows in close proximity to corn and soybean fields and monarchs specialize on milkweed, this system offers a unique opportunity to examine the links between crop management, pesticide leaf concentrations, and butterfly development. Importantly, monarch populations have declined sharply over the last 20 years with censuses in overwintering sites reporting an 82% decrease in population size (Inamine et al., 2016; Semmens et al., 2016; Malcolm, 2018). Hypothesized contributors to this decline include: loss of overwintering forests in Mexico (Brower et al., 2012); reductions in milkweed host-plants due to widespread use of the herbicide glyphosate (Hartzler, 2010; Pleasants and Oberhauser, 2013; Stenoien et al., 2016; Thogmartin et al., 2017a); urban development (Brower et al., 2012); severe weather events (Swengel, 1995; Brower et al., 2012); climate change (Oberhauser and Peterson, 2003; Flockhart et al., 2015; Saunders et al., 2017); and parasites (Altizer and Oberhauser, 1999; Altizer et al., 2004, 2015).

TABLE 1 | Area of corn and soybean, expressed as percent of total land use, planted in a 1 km radius around milkweed sites sampled in 2015 and 2016.



TABLE 2 | Local site characteristics and number of plant replicates for milkweeds sampled in 2015 and 2016.

Although pesticides have been considered as a factor underlying the monarch decline (see Oberhauser et al., 2006; Krischik et al., 2015; Pecenka and Lundgren, 2015; Thogmartin et al., 2017a), it is difficult to evaluate this hypothesis because we lack data on field exposure during larval development. Interestingly, monarch declines have temporally coincided with the increase in use of neonicotinoids throughout agricultural regions in their summer breeding habitat, leading some to speculate whether this is a correlative or causal relationship (Stone, 2013). A recent petition by the U.S. Fish & Wildlife Service to protect monarchs under the endangered species act highlights this point: "It is notable that the monarch decline has occurred during the same time period that the use of neonicotinoid insecticides in the key monarch breeding areas has dramatically increased, although, to our knowledge no one has tested the hypothesis that neonicotinoid use is a significant driver of monarch population dynamics."

A lab toxicity assay of monarch larvae exposed to different concentrations of clothianidin—the main neonicotinoid seed treatment applied to corn—showed lethal effects with an LC<sup>50</sup> at 15.6 ng/g and sub-lethal effects at as little as 1 ng/g (Pecenka and Lundgren, 2015). Despite the lack of data on realistic field exposure, some have taken proactive measures to protect monarchs against potential harm. In 2014, for instance, the U.S. Fish & Wildlife Service phased out neonicotinoid insecticides on crops grown on National Wildlife Refuge System lands. Further, the U.S. Department of Agriculture developed a wildlife habitat evaluation guide and decision support tool for monarch butterfly restoration in which a 125-foot-wide pesticide-free buffer around restored milkweed habitat is advocated (USDA-NRCS 2016). To our knowledge, these buffers have not been "ground truthed" by quantifying actual pesticide residues on milkweed plants varying in their distance from the edge of agricultural fields. Such data are critical for defining the validity of nearest-distance thresholds used by land managers creating monarch habitat. Given that recent monarch population models estimate that 1.6 billion milkweed stems need to be added to the Midwestern region to achieve future conservation goals, close proximity to agricultural land is unavoidable (Pleasants, 2017; Thogmartin et al., 2017b).

With this in mind, our primary aim in this study was to define and quantify the spectrum of pesticides exposed to potential consumption by monarch caterpillars on their hostplant, A. syriaca, in agricultural landscapes. Secondarily, we assessed how pesticide presence varies with linear distance between focal milkweeds and cropland. This was done to test the degree to which pesticide residues diminish with increasing spatial isolation at a local-scale, which is most relevant to land managers who often have some amount of flexibility over local habitat placement on their property. Pesticide-free buffers assume a proximity threshold beyond which exposure is minimal to non-existent. Last, we compared the effectiveness of nearest-distance buffer models with broader landscapescale analyses of land use to determine which better predicts monarch exposure.

#### METHODS

#### Study Areas

In 2015 and 2016, we sampled A. syriaca at seven sites across two counties—Tippecanoe and Newton—in northwestern Indiana, USA. Each site was separated from the nearest site by at least 2 km with the farthest two sites ca. 100 km apart. A site consisted of a patch of at least 30 milkweed plants growing in an area adjacent to a corn or soybean field. Although all milkweed patches were embedded within agricultural landscapes dominated by corn and soybean production (see **Table 1** for land use data and SI Appendix, **Figures S1** and **S2** for reference GIS land use maps to visualize surrounding habitats for a representative agricultural and natural site, respectively), the local habitat varied widely from unmanaged crop field edges to large prairies used in restoration or conservation. As a result, the degree of isolation separating milkweeds from the nearest crop field varied widely, from 0 to >2 km; however, most

were within a 100 m buffer zone of the field edge. Because we were constrained by the location of existing milkweeds and site configuration, we had little control over min/max distances, as well as other factors that could affect pesticide movement, e.g., soil type, direction of milkweed patch relative to crop field (upwind vs. downwind). Data on number of plants sampled per site/year, distance range separating milkweeds from crop, size of neighboring crop field, and direction of milkweeds compared to crop are provided in **Table 2**. Sites included:


# Field Sampling Milkweed Leaf Samples

During June, July, and August in both years, leaf tissue was collected from milkweed plants for chemical analysis. On average, we sampled 48 plants per site each year, with 524 total milkweed plants sampled over the 2-year period across all sites. Within a given site, sampled plants were semi-randomly chosen to span a distance gradient along a transect extending out from the crop field edge. Each month, two leaves were removed to provide at least one-gram of tissue for analysis. The two leaves were located in the central portion of the plant, avoiding the new growth in the apical meristem and older senescent leaves at the bottom of the stem. This leaf position roughly coincides with where we often observe monarch larvae feeding in the field. Leaves were sealed in plastic bags and kept in a cooler with ice before they were transferred to a −80◦C freezer in the laboratory. Because we collected whole-leaf samples, we do not know whether residues were on plant surfaces or inside of plant tissues. Similarly, due to the large number of pesticides measured and logistical challenges with sampling from multiple field sites over time, we did not attempt to control for variation in other factors that undoubtedly impact pesticide detection, e.g., rainfall, time since application, half-life. However, our sampling design over two years with several samples at different time points within a given year, using multiple sites, and a relatively large number of plant replicates per site, was in part intended to account for this inevitable background "noise" and provide a reasonable estimate for average exposure at a given time and place.

Plants were labeled with colored flagging tape to sample the same individuals in subsequent months and georeferenced to calculate the linear distance between focal milkweeds and the nearest corn/soybean field in the study area. To calculate the distance of each individual plant to the crop fields we used an ArcGIS model for each individual site. The tools used in the model include: "Project" that converts data from one coordinate system (WGS\_1984) to another (NAD\_1983\_UTM\_Zone\_16N); "Near" which calculates the distance between the input feature (milkweeds) and the near feature (crop field); "Add field" that adds a new field to a table, in our case the distance from milkweeds to the crop; and "Calculate field" which calculates the values within the new field in the table (SI Appendix, **Figure S3**).

#### Soil Samples

We collected five soil samples per site during June, July and August 2016, resulting in 15 total samples per site. Soil was collected from random locations in the same approximate area where milkweeds were growing at different distances from the crop (SI Appendix, **Figure S4**). To do so, we used a soil core (2 cm diameter), sampling the top ca. 18 cm, although the sampling depth varied with soil compaction across sites. Because soil type plays an important role in the retention or degradation of pesticides, we identified the types of soils at each site using the USDA Soil Survey Geographic Database (SSURGO) map data, which contains information for 3,200 soil surveys (SI Appendix, **Table S1**).

TABLE 3 | Summary data for pesticides detected in milkweed leaf samples across both years of the study.


*"% detection"* = *percent of leaf samples with measurable levels (i.e., above the limit of detection, LOD) for a given pesticide. n.d.* = *not detected.*

are calculated from samples summed across all study sites. Months with different letters, by pesticide, indicate significant differences (*P* < 0.05).



*"% detection"* = *percent of samples with detectable levels (i.e., above the limit of detection, LOD) for a given pesticide. LOD and % recovery data are approximately the same as those reported in* Table 3 *for leaf values.*

#### Land Use Analysis

Although we measured the linear distance of each plant to a specific crop field within our study areas, distance alone may be a poor predictor of variation in pesticide residues. Thus, we quantified the area of corn and soybean in a 1 km radius buffer around the milkweed sampling sites since most of the pesticide inputs are compounds applied to these two crops, which dominate land use in our region. To do so, we used the ArcGIS buffer geoprocessing tool with a 1 km radius, extracted by mask to obtain the crops just within the buffer and tabulated area to calculate the percent of corn, soybean and other crops as a fraction of total land use. Land use data were obtained from USDA NASS Cropland Data Layer for Indiana (www. indianamap.org).

We also estimated corn and soybean pesticide use at a broader geographical scale (county-level) to assess the relationship between pesticide inputs in those crops and the residues associated with our plants. We used the USGS pesticide database, which estimates pesticide applications per crop per state; we used Epest-low values, which are more conservative and tend to better match other estimates. To quantify the amount of pesticides applied in the two counties (Tippecanoe and Newton) where milkweeds were sampled, we first divided the total amount of each corn or soybean pesticide applied at the state-level (i.e., for Indiana only) by state-wide acreage to provide a per area use rate in each year. This approach assumes that statewide averages are reflected in local grower practices, which may not always be the case. The per-unit rate was then multiplied by the area of corn or soybean planted per county in that year to estimate how much of each pesticide was applied near milkweed sites (SI Appendix, **Table S2**). Because USGS datasets stopped including seed treatments in their pesticide surveys after 2014, we unfortunately could not include neonicotinoids and some fungicides using this approach. However, virtually all corn (>90%) in our area is seed treated with clothianidin and thus total corn acreage is a good proxy for neonicotinoid input (Douglas and Tooker, 2015). In addition to the information provided by USGS, a list of the pesticides applied during our sampling to corn or soybean close to the milkweeds was provided by the staff managers at the different sites (SI Appendix, **Table S3**).

#### Laboratory Analysis Leaf Pesticide Residue Analysis

QuEChERS (Quick-Easy-Cheap-Effective-Rugged-Safe) extraction method was used to identify pesticide residues associated with milkweed samples. We screened 65 commonly used pesticides following the approach by Long and Krupke (2016). Multiple leaves within a sampled plant/date were combined and chopped with scissors to obtain a roughly homogenized 1 g sample. All plant tissues were processed with scissors and forceps, cleaned in a 70% alcohol solution before processing each sample and latex gloves were used to avoid contamination between samples. Each 1 g sample was transferred into a 7 ml homogenizer tube (Bertin-technologies) with 2 g of zirconium oxide beads (2 mm diameter; Bertin-technologies). To homogenize the tissue, 2 ml of double deionized (dd) water was added to each tube, after which tubes were set in a Precellys 24 lysis homogenizer, which processed samples using four cycles at 5,000 rpm. Homogenized samples were transferred to 15 ml tubes, and 2 ml dd water and 4 ml of the extraction solvent acetonitrile were added. The 15 ml tubes contained the 1 g plant tissue, 4 ml dd water and 4 ml acetonitrile. Ten µl of an isotopically labeled internal standard mix containing the pesticides screened was added to the 15 ml tubes. The standards help in the quantification of the pesticides in the samples, because a calibration curve is then created to assign a concentration value to peaks obtained from the processed samples.

The anhydrous salts magnesium sulfate (1.2 g) and sodium acetate (0.3 g) were added to enhance the extraction efficiency and induce phase separation with acetonitrile. Each 15 ml tube was agitated for 1 min with a S8220 Deluxe Mixer Vortex (Scientific Products) and shaken on a VWR W-150 Waver Orbital Shaker at speed 10 for 10 min. The tubes were centrifuged at 4◦C, 2,500 rpm for 10 min, for phase separation. One ml of supernatant was added to 2 ml Agilent dispersive Solid Phase Extraction tubes (part no: 5982-5321), containing 25 mg PSA, 7.5 mg GCB and 150 mg MgSO4, cleaning up the samples before the analysis by liquid chromatography. The dispersive SPE tubes with the 1 ml supernatant were spun in a vortex (Labnet VX100) for 10 min and centrifuged at 15,000 rpm for 5 min in an Eppendorf Centrifuge 5424. The supernatant was then transferred into 2 ml Eppendorf tubes, which evaporated overnight in a speed vacuum (SC250EXP, ThermoFisher Scientific). The dry residue at the bottom of the tubes was mixed with 100 µl of acetonitrile, spun for 10 min in a vortex, centrifuged for 5 min, and the supernatant was transferred to liquid chromatography mass spectrometry (LC-MS) autosampler vials. The identification, quantification and separation of the pesticide residues were carried out in an Agilent 1200 rapid resolution liquid chromatography with a triple quadrupole mass spectrometry (Agilent 6460 series) and an Agilent Zorbax SB-Phenyl 4.6 × 150 mm, 5µm column (Agilent technologies, Santa Clara, CA). Both the QuEChERS method

modification and LC-MS analysis were performed at the Bindley Bioscience Center at Purdue University.

#### Soil Pesticide Residue Analysis

QuEChERS extraction method was modified and used to identify pesticide residues in soil, similar to the above-described protocol. Seven grams of wet soil were weighed on a scale (Mettler Toledo model MS3001S). The samples were dried for 2 days at 105◦C in individual aluminum baking cups. The dry weight of each individual sample was recorded to calculate the pesticide concentration in ng/g per sample; dry weight varied between 5.04 and 6.96 g. Dry soil was sieved and slowly added and mixed to avoid clumping with 5 ml dd water in a 50 ml falcon tube. The 50 ml tubes were agitated for 1 min, then 5 ml of acetonitrile (ACN) at 99% and acetic acid at 1% were added, followed by 10 µl of an isotopically labeled internal standards mix containing the 65 pesticides targeted for screening. The tubes were agitated in a vortex for 7 min and then 4 g of magnesium sulfate (MgSO4) and 1 g of sodium acetate (NaOAC) were added slowly, shaking regularly in a vortex to facilitate the incorporation of the salts with the soil and avoid clumps. Upon adding salts, tubes were agitated in a vortex for another 2 min to dissolve any clumps. The samples were centrifuged for 5 min at 4,000 rpm and 1.4 ml of supernatant was transferred into dispersive Solid Phase Extraction tubes (part no: 5982-5122), containing 50 mg PSA, 50 mg C18EC and 150 mg MgSO4, to clean up the samples before the analysis by liquid chromatography. The dispersive SPE tubes were spun for 5 min and centrifuged at 5,000 rpm for 3 min; 1 ml of supernatant was then transferred to 2 ml Eppendorf tubes and left to dry overnight in a speed vacuum (SC250EXP, ThermoFisher Scientific). The next day, samples were resuspended in 100 µl of acetonitrile, spun for 5 min and centrifuged for 7 min at 13,000 rpm before transferring the supernatant into an LC-MS vial. Pesticide identification and quantification were carried out as described above for leaf samples.

### Statistical Analysis

We only targeted pesticides for statistical analysis and figures if they were detected in >1% of milkweed samples with overall concentrations >1 ng/g. Pesticides that fell below these thresholds were considered either too sporadic or diffuse to cause significant ecological impacts on monarch populations.

The effects of year, month and site on pesticide presence in milkweed tissue were evaluated with a mixed model logistic regression, with binary data (SI Appendix, **Table S4**). When pesticide residues were found in association with milkweed tissue we assigned a value of 1 and when pesticide residues were below the detection limit we gave a value of 0. Site was considered as a random factor, and year and month were fixed factors. For this analysis, we only used 0/1 data, rather than the actual TABLE 5 | The effects of month and distance separating focal plants from the nearest cropland on thiamethoxam detection associated with milkweed leaves at three sites (A–C) sampled in 2016.


*Data from 2015 were not used due to the overall low detection rates of thiamethoxam (see* Figure 1A*). Significant (P* < *0.05) effects bolded for emphasis.*

concentrations due to the large number of samples below the detection limit.

We used a correlation analysis to test the relationship between pesticide concentrations found in soil vs. corresponding values in milkweed leaves. To do so, we created a 5 m buffer around the points where soil samples were collected and selected the plants inside the buffer (SI Appendix, **Figure S4**). These soil-plant samples were paired together as spatially cooccurring to test for a correlative pattern. In cases where multiple plants were within the soil buffer we averaged the plant data to create a single mean value for each pesticide at that location.

To evaluate the effects of land use on pesticide residues associated with milkweed leaves we used a three-tiered approach, starting with local habitat placement and ending with landscapescale crop pesticide use. For local habitat placement, we used a two-part hurdle model with logistic regression using binary data based on detection frequency, followed by a secondary analysis using the continuous concentration data with non-detections removed. For this analysis, we focused on the three insecticides thiamethoxam, clothianidin, deltamethrin—since the impacts of fungicides/herbicides on monarchs at this point are unknown. Because distance to field is confounded with site, we were unable to include both factors in the model. In working with naturally occurring milkweed patches we were constrained by existing plant distributional patterns (see **Table 2**), resulting in some sites with all milkweeds clustered relatively close to the field margin (0–30 m for 2015 TPAC) and other sites that were far further away (2,300–2,400 m for 2015 Kankakee far). Thus, we developed site-specific models that include the factors year (when appropriate; some pesticides were mostly detected one of the 2 years), month, and distance separating milkweed plants from the nearest agricultural field. This allows us to test the effects of spatial isolation, while controlling for temporal variation. We only analyzed sites in which the distance gradient spanned the 125 ft distance threshold proposed for milkweed restoration. Several of our sites (see **Table 2**) included milkweeds that far exceeded this distance threshold, even at the closest proximity, and, consequently, distance from nearest crop field is biologically less relevant in these cases.

Next, simple linear regressions per year and pesticide active ingredient were used to quantify the relationship between percent of corn and soybean planted in a 1 km radius around milkweed habitats and the frequency of milkweed leaves with pesticide residues. For this analysis, we took advantage of natural variation in land use surrounding our sites, which varied widely from no agriculture to ca. 80% cropland (see **Table 1**). Last, we used correlations to determine whether corn or soybean pesticides applied at the county-level reflected the frequency of residues associated with milkweed leaves. This analysis used Tippecanoe as the focal county since this housed the majority of our milkweed sites and has a similar agricultural backdrop to the other county (Newton) surveyed. Also, we focused this county analysis only on fungicides for two reasons: one, given the chemical and application differences across pesticide classes, we wanted to avoid directly comparing, for example, insecticides and herbicides; and two, fungicides had the most active ingredients- −6 compounds—which allowed us to make this comparison (i.e., we were unable to use a correlation with only 2 or 3 data points in the case of insecticides and herbicides).

All statistical analyses were conducted with R software 3.5.1 using the packages car, ggplot2, lmer4, and multicomp, except for local habitat use (i.e., distance from crop), for which we employed the Proc Genmod and Proc GLM functions in SAS, V. 9.4.

### RESULTS

#### Leaf Pesticides

Across both years of the study, 14 pesticides commonly used in crops in Indiana were detected on milkweed leaves (**Table 3**). It is important to note, however, that this is not a comprehensive list. While we screened a relatively large number of pesticides, focusing on ones that we know are ubiquitous components of row crop pest management in our region, some compounds are difficult to detect due to factors such as high volatility

(e.g., dicamba) or require a different, more specialized analytical approach for quantification (e.g., glyphosate).

Clothianidin, the insecticide that to date has received the most attention for potential non-target impacts on monarchs, was only detected in 4–8% of total samples; however, those values are somewhat misleading since it averages across all sites and dates. As a general pattern for both sampling years, we almost exclusively detected clothianidin in June, but not in July or August (**Figures 1A,D**). During these early season samples, clothianidin was detected in ca. 15–25% of plants with nearly 60% of milkweeds at some sites testing positive. Interestingly, both thiamethoxam (neonicotinoid) and deltamethrin (pyrethroid) varied dramatically in their detection rates across years (SI Appendix, **Table S4**), with both compounds occurring at high frequencies in 2016 (75–99%) while being virtually absent from samples in 2015 (**Figures 1A,D**). Imidacloprid (neonicotinoid) was only found in a small number of plants (0.2%) in the first year of this study.

Atrazine was the most commonly detected (80–87% on average, although some months approached 100% of samples) and occurred at the highest mean concentrations (6.84 and 37.0 ng/g) of any pesticide surveyed in either year, followed by s-metolachlor and acetochlor among the herbicides (**Table 3**). Notably, s-metolachlor displayed consistent withinseason patterns in both sampling years whereby detection rates were several-fold higher early in the season before gradually declining in July and August (**Figures 1B,E**; SI Appendix, **Table S4**).

Overall, fungicides were the most omnipresent of pesticides detected on milkweed with 6 compounds consistently occurring on leaves. Several fungicides, most notably propiconazole (98% detection rate in 2016), were somewhat commonly detected, but only at trace (<1 ng/g) amounts. The compounds that combined relatively high concentrations and detection rates included pyraclostrobin (31–55%) and trifloxystrobin (27–40%; **Table 3**). In contrast with the herbicide s-metolachlor, which decreased throughout the season, the two strobilurin fungicides showed the opposite pattern, gradually increasing from June to August in both years (**Figures 1C,F**; SI Appendix, **Table S4**). Propiconazole detection frequency displayed a nearly 3-fold increase between years one and two, from 34 to 98% of samples.

#### Soil Pesticides

We found 7 pesticides in soil across the sites sampled in 2016 (**Table 4**), which were a subset of the 14 pesticides recorded from milkweed leaves. Clothianidin was the only insecticide detected and it was found in all samples consistently throughout the summer (**Figure 2A**), in contrast with leaf presence, which was restricted to only June. Thus, clothianidin was far more ubiquitous in the soil than leaves. Importantly,

soil concentrations of clothianidin were highly correlated with levels in co-occurring milkweed leaves (**Figure 3**; r = 0.763, p < 0.0001). This was the only pesticide showing a soilplant association.

We detected three herbicides—atrazine, s-metolachlor, acetochlor—and three fungicides—azyoxystrobin, pyraclostrobin, metalaxyl. Similar to clothianidin, soil concentrations of these compounds tended to be far more stable over time, i.e., leaf values fluctuated dramatically across months when the same soil values remained relatively constant (compare **Figure 1** vs. **Figure 2**) even though half-life of pesticides vary with soil physical and chemical characteristics and our plants grow under different soil types (SI Appendix, **Table S1**).

#### Land Use

Linear distance separating milkweed plants from agricultural fields was a strong predictor of thiamethoxam detection frequency at all of the sites evaluated (**Table 5**). However, distance frequently interacted with sampling month, resulting in variation in the nature of the relationship over time. In 6 of 9 cases (3 sites × 3 months), detection rates declined with increasing distance separating milkweed from crop field up to 150 m, although the shape of this relationship varied (**Figure 4**). The other two insecticides either showed no spatial patterning (clothianidin; no significant main or interactive effects of distance from crop) or were detected in nearly 100% of samples and thus did not have sufficient variation in detection frequency to statistically evaluate using binary data (deltamethrin in 2016, **Table 3**). When continuous concentration data were used after removing samples below the detection threshold, one of the three sites also showed a distance relationship involving thiamethoxam (**Figure 5**; distance x month, F = 11.35, P < 0.0001). Similar to detection data, concentrations were higher in milkweeds growing closer to field edges. As with binary data, no relationships were observed for clothianidin or deltamethrin.

Although we found substantial site-level variation in pesticide presence on milkweed, landscape composition—namely, amount of corn and soybean—within a 1 km radius surrounding focal sites was a poor predictor of our data. Across both 2015 (**Figure 6**) and 2016 (**Figure 7**), only one pesticide pyraclostrobin in 2015 (**Figure 6D**; F = 8.61, P < 0.05) showed a relationship between land use and detection frequency (SI Appendix, **Table S5**). In this case, percent of plants with measurable amounts of pyraclostrobin increased from ca. 40 to 70% when comparing the least to most agricultural sites.

Finally, at the county-level, which encompasses the broadest spatial scale employed (for reference, Tippecanoe Co. is ca. 1,300 km<sup>2</sup> ), the total amount of fungicides applied to soybean had a marginally significant (r = 0.85, P = 0.06) effect on the percent detection frequency of fungicides for milkweed leaves in 2015 (**Figure 8B**). However, other relationships were not significant (corn 2015, r = 0.38, P = 0.28; corn 2016, r = 0.18, P = 0.77).

# DISCUSSION

Our study clearly shows that the foliage of milkweed growing in prairies and unmanaged habitats neighboring cropland contains residues from a wide variety of agricultural pesticides, primarily those applied to corn and soybean. The actual risk of these pesticides, however, depends on how frequently milkweeds contain those levels in the field. Our data reveal strong spatiotemporal variation in pesticide occurrence across sites, months, and years, which means that the threat posed by these chemicals depends on if, when, and where they coincide with monarch colonization and phenology. Below we highlight the implications of these findings for each of the three major pesticide classes and discuss whether pesticide exposure can be avoided based on local and landscape-level habitat placement.

#### Insecticides

Contamination of non-target plants by neonicotinoids used in agriculture is widely reported, but almost exclusively for pollen or nectar samples taken from flowers (Greatti et al., 2003; Krupke et al., 2012; Bonmatin et al., 2014; Botías et al., 2015, 2016; Mogren and Lundgren, 2016). Consistent with this literature, our study found neonicotinoid residues associated with milkweed leaves around farmland, specifically the active ingredients clothianidin and thiamethoxam. Although seed treatment data are no longer reported for U.S. row crops, most corn in our region is seed treated, primarily with clothianidin, and much of the soybean acreage also employs a seed treatment, mainly thiamethoxam (Douglas and Tooker, 2015). Corn and soybean dominate land use in the areas surrounding each of our milkweed sites, and thus it is not surprising that these two insecticides were among the ones most commonly detected.

Importantly, the leaf concentrations we recorded (up to 56.5 and 151.3 ng/g for clothianidin and thiamethoxam, respectively) are within the range previously reported from other studies. For example, a recent analysis of clothianidin on the leaves of plants used in pollinator strips bordering seed-treated corn fields

reported values that were comparable to our milkweed data (Mogren and Lundgren, 2016), including sunflower (max. 81 ng/g), buckwheat (max. 54 ng/g), and phacelia (max. 33 ng/g). Interestingly, some milkweed concentrations were also roughly similar to those reported from the leaves of seed-treated crops such as corn (7–86 ng/g at 20–34 days post planting; Alford and Krupke, 2017) and soybean (105 ng/g in V1 stage and 1.7 ng/g in V4 stage after 17 and 56 days; Magalhaes et al., 2009). Perhaps most relevant to our study, Pecenka and Lundgren (2015) documented clothianidin in 36–64% of milkweed leaves surveyed in South Dakota at mean concentrations of 1.24 and 1.11 ng/g. By comparison, we detected clothianidin at a far lower rate (4.6 and 8.1%, overall, for the 2 years), but with comparable mean values (0.71 and 0.48 ng/g). Pecenka and Lundgren (2015) used dose-response curves for monarch larvae to clothianidin, which revealed the LC<sup>50</sup> at 15.63 ng/g and sublethal effects at as little as 1 ng/g. Based on extrapolating these calculations to our field data, sublethal effects should be observed for monarchs on 5–8% of leaves surveyed (averaged across all sites, months, and years; risk varies seasonally), whereas lethal effects (i.e., >LC50) are limited to 1.4% of samples. It is important to note that our assessment is based solely on clothianidin, for which data exist on monarch growth and survival. Our second sampling year revealed that thiamethoxam can be much more prevalent detected in 75% of samples—but its toxicity to monarchs is unknown at present.

Another critical aspect of our neonicotinoid data is that during both years of the study, residues diminished dramatically over the course of the summer. We virtually only detected clothianidin in June, and thiamethoxam detection in year 2 dropped by ∼50% from June-July to August. This within-season decline would be consistent with pesticide degradation from the putative time of exposure (i.e., when seed-treated fields are planted in late spring) to the timing of when milkweeds were sampled. More importantly, the data suggest that early-season monarchs are at greater risk from neonicotinoid exposure than subsequent generations occurring later in the season. Similarly, our data suggest strong annual fluctuations in risk, indicating that monarchs likely encounter a different suite of pesticides each year. Thiamethoxam and deltamethrin, for example, were

more prevalent in the second sample year. This is likely a result of local or regional differences in pest management approaches employed by farms. Active ingredients for foliar sprays such as deltamethrin can vary greatly across years, depending on factors such as price and availability. Thiamethoxam is more likely to be a reflection of seed treatments, which vary with the relative acreage of corn vs. soybean in the landscape. Further, in corn/soy rotations, the insecticides used will change on an alternate year basis. Overlaying temporal variation in pesticide presence with the timing of non-target insect colonization and development is a key component to risk assessment that, to our knowledge, is rarely incorporated into such studies.

While we did not document the mechanism by which neonicotinoids moved from cropland to milkweeds in this study, for clothianidin we found a strong positive relationship between soil and leaf concentrations (**Figure 3**). This could be simply correlative (i.e., areas with high neonicotinoid deposition result in correspondingly higher concentrations both in soil and on plant surfaces), or indicative of systemic uptake from soil into nearby plants. In all cases, we analyzed whole tissue samples so, unfortunately, do not know whether pesticides are on the leaf surface or inside the plant, for systemic compounds. Overall, the clothianidin concentrations in our soil samples (range: 0.88–8.59 ng/g; mean: 1.75 ng/g) were comparable with those reported in other studies of agricultural soils, i.e., 6.57 ng/g (Botías et al., 2015), 7.0 ng/g (Xu et al., 2016), 2.1 and 6.3 ng/g (Krupke et al., 2012).

Last, in 2016 we frequently detected the pyrethroid deltamethrin in milkweed samples. Although pyrethroids are considered highly toxic to lepidopterans in general, nothing is known specifically about the deltamethrin-monarch relationship. A few studies have found negative non-target effects of the related pyrethroids, permethrin, and resmethrin, used in mosquito control on monarch caterpillars (Oberhauser et al., 2006, 2009). Similarly, field applications of deltamethrin in the UK increased mortality of Pieris butterfly larvae developing in hedgerows bordering cereals (Cilgi and Jepson, 1995). Topical application of 20 ng was sufficient to kill 50% of P. brassicae individuals after 2 weeks of exposure (Cilgi and Jepson, 1995); however, host plants influence caterpillar susceptibility to deltamethrin (Tan and Guo, 1996), and thus it is difficult to extrapolate these values for milkweed.

# Herbicides and Fungicides

Monarch decline is often attributed to an indirect effect from glyphosate reducing milkweed abundance (Hartzler, 2010; Pleasants and Oberhauser, 2013; Pleasants, 2017). Yet, the direct effects of herbicides on monarchs (i.e., those not merely caused by a reduction in milkweed availability) are unknown, and likely dismissed since herbicides are considered non-toxic to insects (but see Russell and Schultz, 2009; Stark et al., 2012). Potential non-target pathways could occur via herbicide exposure, either topically or orally, changing some aspect of caterpillar physiology or altering the milkweed-monarch interaction by e.g., interfering with or amplifying the induced defense pathways employed by milkweeds (Boutin et al., 2004, 2014). For instance, the herbicide 2,4D functions as a plant defense elicitor, resulting in resistance to herbivorous insects on plants exposed to low doses (Xin et al., 2012). Also, drift of the herbicide dicamba into field margins reduced pollinator visitation rates (Bohnenblust et al., 2016), impacted the abundance of several arthropods in the community (Egan et al., 2014), and decreased caterpillar development (Bohnenblust et al., 2016). Herbicides such as glyphosate can even act directly on pollinators by disrupting their gut microbiome (Motta et al., 2018).

Of the herbicides sampled, atrazine was the most commonly detected and at the highest concentrations. Much of what is known about atrazine's impacts on invertebrates comes from aquatic food webs where run-off into streams or lakes alters community structure (Dewey, 1986; Gruessner and Watzin, 1996). While these are mostly indirect effects via reductions in the population of algae or related macrophytes, direct effects of atrazine on insects are documented (Miota et al., 2000; Graymore et al., 2001), as well as their role in synergizing insecticides such as organophosphates (Anderson and Lydy, 2002).

For fungicides, the compounds we detected in milkweed leaves largely match those reported from pollen, honey, nectar, wax, and foliage of wildflowers or crops (Krupke et al., 2012; Sanchez-Bayo and Goka, 2014; David et al., 2016; Long and Krupke, 2016). Fungicides inhibiting ergosterol biosynthesis, like propiconazole, act as synergists for neonicotinoid insecticides, increasing their toxicity to bees by inhibiting cytochrome P450s that function in detoxification (Pilling and Jepson, 1993; Pilling et al., 1995; Johnson et al., 2013). There is also the potential for additive toxicity when insects are exposed to mixtures of pesticides. Propiconazole was detected in 98% of milkweed samples in 2016, in many cases co-occurring with insecticides like deltamethrin and thiamethoxam. The high frequency of the fungicides propiconazole, pyraclostrobin, and trifloxystrobin compared with metalaxyl and azoxystrobin could be related to the high use of these fungicides to increase yield in hybrid corn and soybean (Paul et al., 2011; Mahoney et al., 2015).

# Land Use

For the neonicotinoid thiamethoxam, we found that detection frequency and concentrations tended to be higher on milkweeds growing in closer proximity to agricultural land. This suggests that spatially isolating milkweed restoration sites from crop fields could be an effective approach to reduce risk. To our knowledge, the proposed 125 ft buffer distance is a somewhat arbitrary value that is not based on specific criteria; however, our data nevertheless suggest that milkweed habitat restoration abiding by this rule would likely result in fewer plants containing thiamethoxam and at lower concentrations (see values >38.1 m on **Figures 4**, **5**). What remains unclear is the degree to which these reductions result in enhanced survival and/or performance of monarch caterpillars, which is the ultimate goal. This would require experimentally rearing larvae on plants in the field along a distance transect extending from a crop field edge. In fact, bypassing high quality monarch habitat on land that is relatively close to a corn or soybean field could have a net detrimental effect on monarch conservation if the benefits of additional milkweed stems exceed the detrimental impact of higher pesticide load; a scenario that is entirely plausible, depending on the factor(s) most limiting monarch fitness. This is particularly true for pesticides such as clothianidin that already occur at relatively low frequencies. Deltamethrin also occurred at frequencies and concentrations that were independent of distance to crop. This could be due to the fact that this insecticide is likely applied via aerially spraying, which may result in greater propensity for drift beyond the immediate surrounding of crop fields.

At the landscape level, amount of row crop production in the 1 km radius around milkweed sites was generally a poor predictor of pesticide presence on milkweeds. Only one of the pesticides tested—pyraclostrobin—showed a significant relationship whereby prevalence increased on milkweeds with increasing agricultural intensity. That being said, several of the pesticides, including clothianidin in both years, were most prevalent at the most heavily agricultural site while showing the lowest occurrence at the least agricultural site. We suspect that the lack of statistical power due to low site replication (n = 6 and 5 sites in 2015 and 2016, respectively) played a role in these outcomes, especially for a coarse predictor variable like total crop acres that does not account for variation in local site factors. A similar conclusion was drawn from a recent study of pesticide residues on native bees; despite trends, land cover in a 1 km radius around sites was non-significant, likely due to low site replication (Hladik et al., 2018). Our county-level analysis led to an analogous conclusion. Correlations suggested that greater use by farmers at the regional scale increased prevalence of fungicides on milkweeds, but statistical effects were equivocal (i.e., marginal significance) due to low replication (**Figure 8**).

### CONCLUSIONS

Risk assessment evaluating the potential impacts of pesticides on monarchs entails a two-step process; first, documenting the chemicals that larvae and/or adults are exposed to in the environment, and second, experimentally testing those chemicals most commonly encountered to assess lethal and sub-lethal effects. Here, we take the first step in this process, documenting the spectrum of pesticides encountered by monarch larvae on the most critical host-plant in their summer breeding range, A. syriaca. We strongly emphasize, however, that pesticide presence does not necessarily translate into impact. Unlike honeybees, for which LD<sup>50</sup> data are widely available on most compounds, at present such information is only available for clothianidin in the monarch system. Clearly, a major emphasis of future research efforts should be to close this knowledge gap by quantifying monarch larval responses to a range of pesticides under controlled lab settings. Based on our field data, obvious starting points for these trials would be insecticides such as thiamethoxam and several of the ubiquitous fungicides that occur on milkweed leaves.

Assuming pesticide presence is undesirable for land managers focused on restoring milkweed for monarch conservation, our data secondarily point to local habitat placement—namely, site isolation—as an effective tool for reducing non-target exposure. Additional work to help refine these recommendations could focus on site-specific factors that contribute to off-site pesticide drift beyond simple linear distance, e.g., wind direction, slope, soil type.

#### AUTHOR CONTRIBUTIONS

PO-A wrote the manuscript, collected, and analyzed the data. IK guided the research, reviewed, and edited the manuscript.

# FUNDING

This research was supported by a dissertation fellowship from the Colombian government-Colciencias, and partial grants from the Nature Conservancy, Xerces Society, Experiment.com, and Purdue AgSeed.

## ACKNOWLEDGMENTS

We are grateful to the many collaborators who made this project possible. Specifically, we thank those students involved in lab/field assistance for sample collection and analysis (Ivy Widick, Ashley Kirtman, Carolina Zamorano, Javier Lenzi, Rosario Uribe, Sylvia Bonilla, Rachel Thomas, Andrés Sandoval, Megan McCarty, Kayleigh Hauri); ArcGIS professional guidance (Mayra Rodriguez and Mariam Valladares); staff members at each field location for their help with site logistics (Pete Illingworth, Jay Young, Nate Linder, Brian Baheler, John Shuey, James Beaty, Jason Getz, Kile Westerman); personnel who assisted with pesticide residue analysis at the Metabolite Profiling Facility of the Bindley BioScience Center at Purdue (Amber Hopf Jannasch, Bruce Cooper, Jackeline Franco, Elizabeth Long, Christian Krupke, Mercedes Laland, Spencer Bockover); Laura Cruz for statistical guidance; and Maggie Douglas for help with interpreting state- and county-level pesticide records. A special thanks to Larry Bledsoe who provided detailed information related to corn and soybean at the beginning of this project and tools necessary for the success of the data collection in the field. Finally, thanks to thesis committee members—Cliff Sadof, Michael Scharf, Jen Zaspel, and Kevin Gibson—and students/postdocs in the Kaplan Lab—Wadih Ghanem, Laura Ingwell, Jacob Pecenka, Christie Shee, John Ternest—for their valuable comments on the manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00223/full#supplementary-material

# REFERENCES


organophosphate insecticide. Bull Environ. Contam. Toxicol. 57, 683–690. doi: 10.1007/s001289900244


**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 Olaya-Arenas and Kaplan. 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.

# Expanding the Isotopic Toolbox to Track Monarch Butterfly (Danaus plexippus) Origins and Migration: On the Utility of Stable Oxygen Isotope (δ <sup>18</sup>O) Measurements

#### Keith A. Hobson1,2 \*, Kevin J. Kardynal <sup>2</sup> and Geoff Koehler <sup>2</sup>

*<sup>1</sup> Department of Biology, University of Western Ontario, London, ON, Canada, <sup>2</sup> NHRC Stable Isotope Laboratory, Environment and Climate Change Canada, Saskatoon, SK, Canada*

The measurement of naturally occurring stable hydrogen (δ <sup>2</sup>H) and carbon (δ <sup>13</sup>C) isotopes in wings of the eastern North American monarch butterflies (*Danaus plexippus*) have proven useful to infer natal origins of individuals overwintering in Mexico. This approach has provided a breakthrough for monarch conservation because it is the only viable means of inferring origins at continental scales. Recently, routine simultaneous analyses of tissue δ <sup>2</sup>H and δ <sup>18</sup>O of organic materials has emerged leading to questions of whether the dual measurement of these isotopes could be used to more accurately infer spatial origins even though the two isotopes are expected to be coupled due to the meteoric relationship. Such refinement would potentially increase the accuracy of isotopic assignment of wintering monarchs to natal origin. We measured a sample of 150 known natal-origin monarchs from throughout their eastern range simultaneously for both δ <sup>2</sup>H and δ <sup>18</sup>O wing values. Wing δ <sup>2</sup>H and δ <sup>18</sup>O values were correlated (*r* <sup>2</sup> = 0.42). We found that wing δ <sup>2</sup>H values were more closely correlated with amount-weighted growing season average precipitation δ <sup>2</sup>H values predicted for natal sites (*r* <sup>2</sup> = 0.61) compared to the relationship between wing δ <sup>18</sup>O values and amount-weighted growing season average precipitation δ <sup>18</sup>O values (*r* <sup>2</sup> = 0.30). This suggests that monarch wing δ <sup>2</sup>H values will be generally more useful in natal assignments than δ <sup>18</sup>O values. Spatial information related to the use of deuterium excess in environmental waters was similarly found to be not useful when applied to monarch wings likely due to the considerable variance in wing δ <sup>18</sup>O values. Nonetheless, we recommend further testing of monarch wing δ <sup>2</sup>H and δ <sup>18</sup>O values from known natal sites with an emphasis on field data across a strong gradient in precipitation deuterium excess.

Keywords: deuterium, stable isotopes, oxygen-18, natal origins, wing chitin, assignment

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*David Nelson, University of Maryland Center for Environmental Science (UMCES), United States Sarah Magozzi, The University of Utah, United States*

> \*Correspondence: *Keith A. Hobson khobson6@uwo.ca*

#### Specialty section:

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

> Received: *27 February 2019* Accepted: *29 May 2019* Published: *18 June 2019*

#### Citation:

*Hobson KA, Kardynal KJ and Koehler G (2019) Expanding the Isotopic Toolbox to Track Monarch Butterfly (Danaus plexippus) Origins and Migration: On the Utility of Stable Oxygen Isotope (*δ *<sup>18</sup>O) Measurements. Front. Ecol. Evol. 7:224. doi: 10.3389/fevo.2019.00224*

# INTRODUCTION

The monarch butterfly (Danaus plexippus) is an iconic migratory insect that navigates over thousands of kilometers between natal sites in the eastern USA and Canada and overwintering sites in the Oyamel fir (Abies religiosa) forests of central Mexico (Urquhart and Urquhart, 1976; Ries et al., 2015; but see Vander Zanden et al., 2018). This outstanding migration involves multiple generations whereby those individuals returning to wellestablished, long-term overwintering sites do so without any previous experience of their locations. Despite being a highprofile conservation issue among Mexico, Canada and the USA, monarch butterflies have declined considerably over the last two decades (Semmens et al., 2016). Potential causes for the decline are many but likely factors include loss of milkweed (Asclepia spp.) on the breeding grounds, declining extent and quality of overwintering sites, climate change and the myriad of challenges faced during the migratory passage (Flockhart et al., 2015; Thogmartin et al., 2017). Key to understanding declines of this and other migratory species is the establishment of migratory connections that link breeding, stopover and wintering sites (Hobson and Wassenaar, 2019). However, this is challenging for small insects. Previous attempts have used an impressive mark recapture effort involving tagging Monarchs on breeding and stopover sites with labels affixed to wings (Monarch Watch; https://www.monarchwatch.org/tagmig/index.htm), an approach requiring recovery en route and at the few publicly accessible wintering sites in Mexico. However, during the mid-1990s Hobson et al. (1999) and Wassenaar and Hobson (1998) developed an approach using ratios of naturally occurring stable isotopes in Monarch wings to infer natal origins. Those studies marked a turning point in the conservation of migratory monarchs since it readily identified the US Midwest as the center of production of Monarchs recruited into the overwintering population in Mexico. Since then, the approach has been used by other researchers interested in Monarch patterns of spring recolonization in North America (Miller et al., 2012; Flockhart et al., 2013), the effects of natal origin on parasite loads (Altizer et al., 2015). Other applications have involved the role of wing coloration in flight distance (Hanley et al., 2013) and general conservation concerns related to where most individuals are being produced (Flockhart et al., 2017).

The stable isotope approach to tracking migrant animals is based on the fact that naturally occurring stable isotopes of several elements in nature can provide information on provenance and/or habitat. Spatial patterns of stable isotopes (i.e., "isoscapes") are ultimately transferred to animal tissues through local foodwebs and so spatial information related to the period of tissue growth can be inferred providing we know the nature of such isoscapes and how stable isotope values change or discriminate from the source to the tissue of interest (Hobson and Wassenaar, 2019). The early studies on stable isotope patterns in eastern monarchs indicated that both stable–carbon (13C/12C expressed as δ <sup>13</sup>C) and stable-hydrogen (2H/1H, δ <sup>2</sup>H) isotope ratio measurements of wings could provide information on monarch natal origins in Canada and the USA. However, δ 2H measurements provided a more powerful means of inferring origins than δ <sup>13</sup>C measurements. Hydrogen in foodwebs is ultimately derived from precipitation. The amount-weighted average precipitation δ <sup>2</sup>H (expressed as δ <sup>2</sup>Hp) varies across continents according to well-established principles (Clark and Fritz, 1997) and these isoscape patterns are transferred to plants and higher trophic-level organisms (Bowen and West, 2019). Metabolically inert tissues such as animal keratins and chitins are related to local growing season averaged δ <sup>2</sup>H<sup>p</sup> and can be later measured to infer natal origins once the relationship between δ <sup>2</sup>H<sup>p</sup> and tissue δ <sup>2</sup>H is established (Hobson, 2019). This principle has been used effectively to infer origins of numerous taxa including insects, birds and mammals on several continents (Hobson et al., 2018a; Hobson and Wassenaar, 2019).

Water contains both hydrogen and oxygen but measurement of the stable isotope ratios in oxygen (18O/16O; δ <sup>18</sup>O) have not been used extensively to study animal movement (but see Bryant and Froehlich, 1995; Kohn, 1996; Hobson et al., 2004, 2009a, 2012; Bowen et al., 2005; Ehleringer et al., 2008; Chesson et al., 2013; Hobson and Koehler, 2015; Pekarsky et al., 2015). This derives from the fact that oxygen has a compressed isotopic scale compared to hydrogen and that δ 2H and δ <sup>18</sup>O values in environmental waters are highly correlated via the global meteoric water relationship (Craig, 1961). Also, previously, measurement of δ <sup>18</sup>O in organic materials has been challenging analytically (Hobson and Wassenaar, 2019). Oxygen in animal tissues can be derived from air, drinking water, and diet whereas hydrogen is derived from only diet and drinking water. However there is potential for δ <sup>18</sup>O when combined with δ <sup>2</sup>H measurements to provide additional information on provenance of animals because environmental waters can record the degree of evapotranspiration through the evaluation of deuterium excess (defined as δ <sup>2</sup>H-8<sup>∗</sup> δ <sup>18</sup>O; Dansgaard, 1964; Rozanski et al., 1993). In highly evaporative systems, we expect departures from the global meteoric water line whereby deuterium excess values are higher than the global average of 10‰, a phenomenon highly associated with low relative humidity. So, again, the relationship between δ 2H and δ <sup>18</sup>O values in insect wing chitin (δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>Ow) has the potential to contain important environmental and climate information.

The evaluation of both δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> values of monarch butterflies was first reported by Fourel et al. (1998) as a proof of principle for the measurement of these isotopes via continuous-flow isotope ratio mass spectrometry. That study showed a positive correlation between the two isotopes. However, since then, no studies have further investigated the relationship between these isotopes in monarch wings in particular, and only a few researchers have reported on δ <sup>2</sup>H and δ <sup>18</sup>O values among higher-order taxa. Hobson and Koehler (2015) reported that while feathers of American Redstart (Setophaga ruticilla) showed a good relationship between δ <sup>2</sup>H and mean growing-season precipitation δ 2H (r <sup>2</sup> = 0.77), the same was not true for feather δ <sup>18</sup>O (r 2 = 0.32). However, for dragonflies, insects that form wings from nutrients derived from an aquatic larval stage, Hobson et al. (2012) found an excellent relationship between δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> values (r <sup>2</sup> = 0.92). The question remains, then,


TABLE 1 | Mean isotope values (± 1 SD ‰) of wild reared monarch butterflies from eastern North America and mean growing-season δ <sup>18</sup>O<sup>p</sup> (GSO) and δ <sup>2</sup>H<sup>p</sup> (GSD) at each location plus the 95% CI shown in parentheses from Bowen et al. (2005).

*All data shown in* Table S1*.*

whether δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> values of terrestrial insects are also well-correlated and indeed reflect underlying relationships in environmental waters that could be used to increase the accuracy of assignment to origin compared to the use of δ 2H measurements alone. Here, we report measurements of δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> values of monarch butterflies sampled at known origin natal sites in eastern North America. Our objective was to determine if a dual isotope approach could provide additional environmental information related to regional patterns in evaporative conditions that could ultimately contribute to more accurate delineation of natal origins of monarchs and other terrestrial insects.

#### MATERIALS AND METHODS

#### Monarch Sample

Monarch wing material was obtained from samples reported in Wassenaar and Hobson (1998) and Hobson et al. (1999). Those were monarchs raised on milkweed exposed only to precipitation at 31 known sites throughout the range of the eastern monarch population and which formed the calibration relationship between mean growing season δ <sup>2</sup>H<sup>p</sup> and δ <sup>2</sup>H<sup>w</sup> reported by Hobson et al. (1999) (**Table 1**). Additionally, monarchs collected from roadkill mortality (n = 92) at a single site in northeast Mexico during the autumn migration of 2015–16 were used to investigate assignment to natal origins using both δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> values.

#### Stable Isotope Analyses

All monarch wing samples were soaked and rinsed in a 2:1 chloroform:methanol solution and air dried. Subsamples were cut from the same region of the hindwing to reduce intersample variance due to isotopic effects from pigmentation (Hobson et al., 2017) and weighed (0.35 ± 0.02 mg) into silver capsules. All samples were prepared for δ <sup>2</sup>H and δ <sup>18</sup>O analysis at the Stable Isotope Laboratory of Environment Canada, Saskatoon, Canada. Our approach involved the analysis of both δ <sup>2</sup>H and δ <sup>18</sup>O on the same analytical run (i.e., both H<sup>2</sup> and CO gases were analyzed from the same pyrolysis) from samples and standards. All measurements were performed on a HTC system (Thermo Finnigan, Bremen, Germany) equipped with a Costech Zero-Blank autosampler. The helium carrier gas rate was set to 120 ml/min through a 0.6 m ¼-inch 5- Å molecular sieve (80–100 mesh) GC column. The HTC reactor was operated at a temperature of 1,400◦C and the GC column temperature was set to 90◦C. After separation, the gases were introduced into a Delta V plus isotope-ratio mass spectrometer via a ConFlo IV interface (Thermo Finnigan, Bremen, Germany). The eluted N<sup>2</sup> was flushed to waste by

FIGURE 1 | Relationship between δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> for monarchs raised outdoors at 31 known sites (Table 1) throughout their range in 1996. Sample originally reported for δ <sup>2</sup>H<sup>w</sup> and δ <sup>13</sup>C<sup>w</sup> by Hobson et al. (1999) but these are new δ <sup>2</sup>H<sup>w</sup> values measured simultaneously with δ <sup>18</sup>Ow.

withdrawing the CF capillary from the ConFlo interface. We used keratin reference standards CBS and KHS to calibrate sample δ <sup>2</sup>H (−197 and −54.1‰, respectively) and δ <sup>18</sup>O (+2.50 and +21.46‰, respectively; Qi and Coplen, 2011). These standard values were used vs. the newly reported values of Soto et al. (2017) simply to maintain consistency with our earlier published work for monarch assignments and we note that this will not affect the assignment per se (providing appropriate calibration algorithms are used). Based on replicate (n = 5) withinrun measurements of keratin standards, sample measurement error was estimated at ±2‰ for δ <sup>2</sup>H and ±0.4‰ for δ <sup>18</sup>O. All H results are reported for non-exchangeable H and for both H and O in typical delta notation, in units of per mil (‰), and normalized on the Vienna Standard Mean Ocean Water – Standard Light Antarctic Precipitation (VSMOW-SLAP) standard scale.

#### Isotopic Assignments

We created δ <sup>18</sup>O<sup>w</sup> and δ <sup>2</sup>H<sup>w</sup> isoscapes based on derived transfer functions using wing stable isotope values of known origin monarchs found in this study. This was accomplished by calibrating amount-weighted growing-season average precipitation δ <sup>18</sup>O and δ <sup>2</sup>H (δ <sup>18</sup>O<sup>p</sup> and δ <sup>2</sup>Hp) (Terzer et al., 2013; IAEA/WMO, 2015) surfaces into separate δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> isoscapes. We then depicted potential origins of a sample of roadkilled monarchs salvaged during their autumn migration through northeastern Mexico. That depiction was done using techniques described previously (Hobson et al., 2018b). Briefly, we used a likelihood-based assignment method (Hobson et al., 2009b; Wunder, 2010; Van Wilgenburg et al., 2012) to assign individual monarchs using δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> to each calibrated wing isoscape (see results), separately. We used the residual SD error of 9.33‰ for δ <sup>2</sup>H<sup>w</sup> and 1.50‰ for δ <sup>18</sup>O<sup>w</sup> from regressions in our assignments. We limited assignments to the current known geographic range for the eastern monarch

population and used this as a spatial mask (i.e., clip) to limit our analysis. We estimated the likelihood that a cell (pixel) within the δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> isoscapes represented a potential origin for a sample by using a normal probability density function (pdf) based on the observed δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>Ow, and thus depicted the likely origins of each monarch by assigning individuals to the δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> isoscapes, separately. We arbitrarily chose a 2:1 odds ratio to include only those pixels (coded 1) with at least a 67% probability of origin vs. all others (coded 0). This resulted in a binary map per assigned individual of presence and absence. We then summed the results of individual assignments by stacking the surfaces for a final depiction. We conducted geographic assignments to origin using functions within the R statistical computing environment (R Core Team., 2016) employing the "raster" (Hijmans, 2016) and "maptools" (Bivand and Lewin-Koh, 2016) packages. Thus, the final assignment surface depicted the number of individuals co-assigned at each pixel based on the odds criteria. We also conducted assignment to origin analysis for individual Monarch samples using a dual-isotope (δ <sup>2</sup>Hw, δ <sup>18</sup>Ow) approach applied with a multivariate normal probability density function (mvnpdf; see Hobson et al., 2014 for details). Similar to the univariate pdf approach, the mvnpdf method calculates the probability that a particular spatially referenced cell represents a potential origin in calibrated raster isoscape space.

FIGURE 4 | Depiction of (A) the expected monarch δ <sup>18</sup>O<sup>w</sup> isoscape based on the relationship derived between δ <sup>18</sup>O<sup>w</sup> and amount-weighted mean growing season average δ <sup>18</sup>O<sup>p</sup> (Bowen et al., 2005) shown in Figure 2 and (B) the expected monarch δ <sup>2</sup>H<sup>w</sup> isoscape based on the relationship derived between δ <sup>2</sup>H<sup>w</sup> and amount-weighted mean growing-season average δ <sup>2</sup>H<sup>p</sup> (Bowen et al., 2005) shown in Figure 3. Legend is the number of individuals assigned to a pixel based on the odds ratio criterion used.

Frontiers in Ecology and Evolution | www.frontiersin.org

ratio criterion used.

## RESULTS

#### Monarch Calibration

We found a positive relationship between monarch δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> values for known-origin, outdoor-raised individuals (δ <sup>2</sup>H<sup>w</sup> <sup>=</sup> 5.21 <sup>∗</sup> δ <sup>18</sup>O<sup>w</sup> – 211, r <sup>2</sup> = 0.42; **Figure 1**). The relationship between δ <sup>18</sup>O<sup>w</sup> and δ <sup>18</sup>O<sup>p</sup> was positive (δ <sup>18</sup>O<sup>w</sup> = 0.54 <sup>∗</sup> δ <sup>18</sup>O<sup>p</sup> + 22.5; **Figure 2**) but relatively weak (r <sup>2</sup> = 0.30, p < 0.001). In contrast, the relationship between δ <sup>2</sup>H<sup>w</sup> and δ <sup>2</sup>H<sup>p</sup> was positive (δ <sup>2</sup>H<sup>w</sup> <sup>=</sup> 0.78 <sup>∗</sup> δ <sup>2</sup>H<sup>p</sup> – 77.42; **Figure 3**) and stronger (r <sup>2</sup> = 0.61, p < 0.001).

#### Assignment Evaluation

Using the relationships established between δ <sup>18</sup>O<sup>w</sup> and δ <sup>2</sup>H<sup>w</sup> values and δ <sup>18</sup>O<sup>p</sup> and δ <sup>2</sup>H<sup>p</sup> values, respectively, we created predicted monarch wing isoscapes for each isotope (**Figures 4A,B**). These surfaces are superficially similar due to the correlation between isotopes. We then used these surfaces to assign origins of a sample of monarchs from roadkill mortality during their fall migration. That sample included monarchs from across their range and we contrasted depictions of origins of these individuals using only δ <sup>2</sup>H<sup>w</sup> values, only δ <sup>18</sup>O<sup>w</sup> values and both δ <sup>18</sup>O<sup>w</sup> and δ <sup>2</sup>H<sup>w</sup> values in a multivariate normal assignment. These depictions were similar in that they showed highest probability in the southwestern portion of the range but differences were also apparent (**Figures 5A–C**). The dual isotope approach constrained the assignment compared to the use of δ <sup>2</sup>H<sup>w</sup> values alone.

#### Deuterium Excess

We contrasted deuterium excess values calculated for individual monarch wings and compared these to the growing-season average deuterium excess calculated for each natal site. No correlation was found between these values (**Figure 6**, r 2 = 0.02). This suggests that little information associated with

precipitation environmental deuterium excess was transferred to monarchs per se.

# DISCUSSION

Our results confirm a strong relationship between wing chitin δ <sup>2</sup>H values and precipitation δ <sup>2</sup>H values at known natal sites. This result underlines the strong assignment power of using Monarch wing δ <sup>2</sup>H values to infer origins and confirms that even though the earlier investigations of Monarch δ <sup>2</sup>H values to infer origins used different analytical techniques (i.e., offline zinc reduction and steam equilibration, Wassenaar and Hobson, 1998; Hobson et al., 1999), the power of that isotopic assignment holds. The different analytical approaches resulted in different calibrations between wing chitin δ <sup>2</sup>H (δ <sup>2</sup>Hw) and predicted amount-weighted growing season δ <sup>2</sup>H<sup>p</sup> (earlier method: δ <sup>2</sup>H<sup>w</sup> <sup>=</sup> 0.62 <sup>∗</sup> δ <sup>2</sup>H<sup>p</sup> – 79; r <sup>2</sup> = 0.69; recent method: δ <sup>2</sup>H<sup>w</sup> <sup>=</sup> 0.78 <sup>∗</sup> δ <sup>2</sup>H<sup>p</sup> – 77.4; r <sup>2</sup> = 0.61) and we recommend that from now on authors use the recent calibration relationship reported here. The δ <sup>2</sup>H<sup>w</sup> results contrasts with the much weaker relationship we found between δ <sup>18</sup>O<sup>w</sup> and δ <sup>18</sup>Op. Thus, while the measurement of δ <sup>18</sup>O<sup>w</sup> possibly confers additional information on origin, the strong relationship between δ <sup>2</sup>H<sup>w</sup> and δ <sup>18</sup>O<sup>w</sup> values likely precludes the effective use of a dual isotope approach in most cases. This result has been seen previously in assignment exercises involving several taxa (reviewed by Hobson and Koehler, 2015). Our comparison of deuterium excess values calculated for monarch wings and those of the mean growing season average deuterium excess for natal sites confirms that little environmental information related to evaporative conditions was available from our sample by running δ <sup>18</sup>O analyses. Nonetheless, we recognize the potential utility of using both δ <sup>18</sup>O<sup>w</sup> and δ <sup>2</sup>H<sup>w</sup> values, especially for areas corresponding to potential origins involving high deuterium excess values that were not included in our study. It may be possible then, to identify migrant individuals raised in xeric vs. humid environments and we encourage further work in this area for migrant insects in general. Clearly, further dual isotope analyses of a large sample of monarchs from known natal origins that spans large isotopic gradients in North America will be useful in testing situations where this more extensive analytical approach might be justified.

It is interesting to speculate why tissue δ <sup>18</sup>O values do not seem to add much additional information on origins of Monarch Butterflies and other terrestrial taxa based on δ 2H analyses alone. The typical response to this question is that oxygen enters metabolic pathways from more diverse origins than hydrogen (Hobson and Koehler, 2015). Fundamentally, oxygen is derived metabolically from diet, drinking water and air whereas hydrogen is derived only from diet and drinking water. Atmospheric oxygen is added to the body water pool during metabolism and respiration and has a relatively constant isotopic composition (Luz and Barkan, 2011). Therefore, the oxygen isotopic coupling between tissues and environmental waters is dependent on the relative contribution of respirative processes to the oxygen isotopic composition of body water. Additionally, with insects, atmospheric oxygen for respiration is obtained primarily by diffusion through the spiricules and tracheal pathways (Chown et al., 2006) which may further alter its isotopic composition. It is worth noting that while the hydrogen isotopic compositions of chitin are broadly similar to those of keratins and other tissues of birds or mammals (Ehleringer et al., 2008; Hobson and Koehler, 2015), the δ <sup>18</sup>O values of chitin are generally more positive. This, along with the relatively poor fit between oxygen isotopic compositions of wing chitin and precipitation, likely indicates that slightly different sources, processes or pathways are utilized for oxygen in terrestrial insects than in other taxa. This would not necessarily preclude the use of both isotopes in typical assignments to origin but such assignment models require isotopes to be relatively orthogonal. For oxygen isotopes, current analytical methods result in a much higher relative measurement error than for hydrogen isotopes (0.4 vs. 2‰ for hydrogen). To use both isotopes to determine an inferred deuterium excess of environmental waters requires the Meteoric Water Line multiplicative factor of 8, so that currently we can only determine the deuterium excess values to precisions of about 3.2‰. Precipitation varies in deuterium excess from ∼0 to +10‰ which means that such poor precisions will require large numbers of samples depending on which populations are being examined. This is not to say that measurement of both isotopes in monarch tissue will not be useful, but that, in general, the use of wing chitin δ <sup>2</sup>H is the most preferred for single isotope assignments.

In addition to the earlier proof of concept papers on the use of both wing δ <sup>18</sup>O and δ <sup>2</sup>H values to investigate natal origins of monarch butterflies (Fourel et al., 1998), our investigation has underlined the fact that there is much future work required to move the field forward. For example, we stress again that monarchs deriving from natal sites in areas of high deuterium excess (e.g., southern states) might still produce wings which reflect these relationships and so to improve our statistical power more sampling in those areas would be desirable.

#### REFERENCES


# DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the **Supplementary Files**.

# ETHICS STATEMENT

Only dead monarch butterflies were examined in this study. These were derived from collections, natural overwinter mortality and roadkills, so no animal care or ethics approval was necessary.

# AUTHOR CONTRIBUTIONS

KH and GK conceived of the idea and performed stable isotope measurements. KK performed isotope assignments and assisted with figures. All authors contributed to the writing of the manuscript.

# FUNDING

This study was funded by Environment and Climate Change Canada operating funds to KH and GK and by an NSERC Discovery Grant to KH.

# ACKNOWLEDGMENTS

We thank Blanca X. Mora Alvarez for assistance with preparing samples for stable isotope analyses. Two reviewers provided useful comments on an earlier draft of the paper.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00224/full#supplementary-material


of spring re-colonization in eastern North America. PLoS ONE 7:e31891. doi: 10.1371/journal.pone.0031891


**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 Hobson, Kardynal and Koehler. 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.

# Does Nature Need Cities? Pollinators Reveal a Role for Cities in Wildlife Conservation

Abigail Derby Lewis <sup>1</sup> \*, Mark J. Bouman<sup>1</sup> , Alexis M. Winter <sup>1</sup> , Erika A. Hasle<sup>1</sup> , Douglas F. Stotz <sup>1</sup> , Mark K. Johnston<sup>1</sup> , Karen R. Klinger <sup>1</sup> , Amy Rosenthal <sup>1</sup> and Craig A. Czarnecki <sup>2</sup>

*<sup>1</sup> Keller Science Action Center, Field Museum, Chicago, IL, United States, <sup>2</sup> United States Fish and Wildlife Service, Minneapolis, MN, United States*

It is well-established that cities need nature for critical ecosystem services—from storing carbon, to reducing temperatures, to mitigating stormwater—and there is growing momentum to seek out strategies for how these services can intersect with urban design and planning efforts. Social scientists and conservation planners increasingly point to urban residents' need to breathe fresh air, encounter the natural world, and have room to play. It is less obvious, perhaps, whether nature needs cities in order to thrive. The evidence from both urban planning and conservation planning is increasingly "yes." As changes in land use and land cover sweep the planet, cities are becoming important refugia for certain wildlife populations. In recent years, urban planning has embraced the concept of "green infrastructure" as a way to embed green space across metropolitan landscapes to draw on the inherent benefits nature provides to cities, as well as to create habitat for wildlife. We explore this evolving view of cities and nature in the fields of urban and conservation planning. We argue the time is ripe to bring these worlds together, and, using our empirical work, establish that cities matter for monarch butterflies, other pollinators, and at-risk wildlife species.

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*Scott Black, Xerces Society for Invertebrate Conservation, United States Mohammad Imam Hasan Reza, Presidency International School, Bangladesh*

\*Correspondence:

*Abigail Derby Lewis aderby@fieldmuseum.org*

#### Specialty section:

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

Received: *29 December 2018* Accepted: *24 May 2019* Published: *21 June 2019*

#### Citation:

*Derby Lewis A, Bouman MJ, Winter AM, Hasle EA, Stotz DF, Johnston MK, Klinger KR, Rosenthal A and Czarnecki CA (2019) Does Nature Need Cities? Pollinators Reveal a Role for Cities in Wildlife Conservation. Front. Ecol. Evol. 7:220. doi: 10.3389/fevo.2019.00220* Keywords: urban, ecology, wildlife, conservation, culture, monarch, pollinators

# WHY FOCUS CONSERVATION EFFORTS IN CITIES?

More than 80% of Americans live in urban areas<sup>1</sup> , as does over half the world's population (UN DESA, 2018). In contrast, in 1960 twice as many people in the world lived in rural areas (2 billion) as urban (1 billion)<sup>2</sup> . This trend is expected to continue, with nearly 70% of the world's population living in urban areas by 2050 (UN DESA, 2018). However, a striking 60% of the additional land projected to become urban by 2030 is yet to be built (GFDRR World Bank, 2015). Each day in the US more than 4,000 acres of open space are lost to development, the equivalent of more than three acres per minute (Williams, 1975).

As changes in land use and land cover sweep the planet, converting grasslands, forests, wetlands, and other available habitat into agricultural fields and developed landscapes, cities are becoming increasingly important refugia for an array of wildlife populations, including threatened and endangered species (Aronson et al., 2014; Ives et al., 2016). This pattern reflects, in part,

<sup>1</sup>U.S. Census Bureau. (2010). Available online at: https://www.census.gov/geo/reference/ua/urban-rural-2010.html <sup>2</sup>Our World in Data. Available online at: https://ourworldindata.org/urbanization

the propensity to locate urban development in biologically diverse areas such as coastal and riparian locations (Luck, 2007). Indeed, populations of many species are reappearing in force across urban spaces—from fishers (LaPoint et al., 2015) and coyotes (Morey et al., 2007) to bullfinches (Audet et al., 2016) and peregrines (Caballero et al., 2016). Other urban wildlife dwellers include migratory species of birds, dragonflies and butterflies that rely on habitat patches in cities to move through landscapes dominated by large-scale agriculture (Seewagen et al., 2011; Goertzen and Suhling, 2013; Tam and Bonebrake, 2016). Significantly, several American cities support a higher diversity of native bee species—including the endangered rusty patched bumble bee (Bombus affinis)—than do adjacent rural areas (Hall et al., 2017; U.S. Fish Wildlife Service, 2017). These examples of wildlife species utilizing urban habitat illustrate that developed areas can be important in the conservation of species of high concern.

Given these trends, we have a small but critical window of time to develop and implement strategies that create highly functional urban landscapes with benefits for both people and nature (Intergovernmental Science-Policy Platform on Biodiversity Ecosystem Services (IPBES), 2019). Understanding how habitat can best be embedded in urban landscapes is important to help curb a potential "sixth mass extinction" (Ceballos et al., 2015, 2017). This situation cannot be overstated: recent studies reveal that the number of mammals, birds, fish and reptiles on Earth has been reduced by 60% in <50 years (World Wildlife Fund, 2018). In Germany, flying insect populations have plunged by 75% in the last 25 years (Hallmann et al., 2017), and a similar trend has been observed in Puerto Rico (Lister and Garcia, 2018).

Powerful urbanization trends have understandably been accompanied by a sense that nature has been displaced in urban landscapes and can only be found where cities don't exist (Hartig and Kahn, 2016). On the one hand, urban life has been characterized as "distanced from nature" (Tuan, 1978) accompanied by an "extinction of experience" (Pyle, 1978, 1993) as people move to urban settings (Miller and Hobbs, 2002; Turner et al., 2004; Zhang et al., 2014; Soga and Gaston, 2016). On the other hand, the conservation community has achieved huge victories in places far from the urban world, and a side effect has been to reify the notion of "wilderness" in the American mind (Nash, 1967). Large protected areas have "increasingly become the means by which many people see, understand, experience, and use the parts of the world that are often called nature and the environment" (West et al., 2006, p. 255).

As we'll discuss, our work on monarch conservation surfaces new ways to bring the potential for urban conservation into sharper focus. This new way of "seeing" cities includes: valuing new potential partners for nature, many of them historically excluded from the conservation narrative (Finney, 2014; Taylor, 2016); applying finer scales of analysis, with the aid of new data and geospatial tools; and recognizing other practices now being adopted to create sustainable cities for people. A main takeaway from our research is that the places called "urban" in all their size, density, and heterogeneity (Wirth, 1938) do contain powerful voices, activities, and opportunities for conservation. In spite of the perception that racial minorities and low-income Americans—who are often well-represented in urban regions—are considered to have little concern for nature, a recent study reports their higher concern for nature than white and higher-income respondents (Pearson et al., 2018). Activating that concern for conservation may entail folding "nature" into the broader set of priorities that residents and community-based organizations have.

From backyards to rooftops to parks, urban residents have seen to it that nature has a place in the city, from the ground up. Improvements in technology have made this activity increasingly visible from the sky down. Wherever possible, our monarch research employed high-resolution imagery, enhanced by technology such as LiDAR<sup>3</sup> , which enabled precise characterization of land cover at the sub-meter scale. This helps us to visualize what is happening in the urban area with greater precision than the commonly used National Land Cover Database (NLCD), a 30-meter resolution dataset most appropriate to use when studying county-level units or larger (Wickman et al., 2014). When NLCD is applied to highly heterogeneous metropolitan landscapes, large swaths of land are classified as high, medium and low intensity developed land cover classes. While "low intensity developed" and "medium intensity developed" indicate a moderate proportion (20–79%) of the land cover is impervious, higher resolution data are needed to quantify and visualize the remaining land that is permeable and usable as green space. **Figure 1** demonstrates a sample area in Chicago's urban core viewed using NLCD, aerial imagery with LiDAR, and plantable space. Finer grain analysis can support the growing recognition that cities are ripe with opportunity and interest to create spaces where both people and wildlife benefit (Rosenzweig, 2003). This new perspective helps us to appreciate nature abounding in a multitude of contexts that intersect with how people live, work and play in urban areas, including in churchyards and school yards, along boulevards, and amidst corporate campuses, residential yards and community gardens (Beatley, 2011; Van Horn and Aftandilian, 2015; Johnston et al., 2019).

Much of what can be seen, and what exists as an opportunity to enhance urban biodiversity going forward, comes in the context of burgeoning "green infrastructure" efforts (Benedict and McMahon, 2006; Hostetler et al., 2011; Ahern, 2013). The uptick of interest in green infrastructure relates to how well it supports a number of needs in urban areas, including: enforcement efforts to bring municipalities into compliance with the pollution control provisions of the federal Clean Water Act<sup>4</sup> ; nature-based solutions to reduce climate impacts such as flooding in urban landscapes (Derby Lewis et al., 2015); public interest in native landscaping (McMahan, 2006); the growing sector of

<sup>3</sup>LIDAR, or Light Detection and Ranging, is a "remote sensing method that uses light in the form of a pulsed laser to measure ranges (variable distances) to the Earth." Available online at: https://oceanservice.noaa.gov/facts/lidar.html

<sup>4</sup>Technical Report: Tools, Strategies and Lessons Learned from EPA Green Infrastructure Technical Assistance Projects. Available online at: https://www. epa.gov/green-infrastructure/tools-strategies-and-lessons-learned-epa-greeninfrastructure-technical

FIGURE 1 | Comparison of land cover data as classified when applying (A) the National Land Cover Dataset derived from 30-m spatial resolution Landsat satellite data, (B) higher spatial resolution land cover data derived using 2-foot spatial resolution multispectral aerial imagery and LiDAR data, and (C) grass-shrub land cover class in isolation (i.e., plantable space). This scale reveals the many opportunities that exist even under highly developed conditions. Sources: Multi-Resolution Land Characteristics Consortium (U.S.). "National Land Cover Dataset (NLCD)" and University of Vermont Spatial Analysis Lab "Chicago Urban Tree Canopy."

urban farming (Jarosz, 2008; Lovell, 2010); and increased access to nature, particularly in marginalized neighborhoods (Wolch et al., 2014).

We propose that pollinator-focused efforts can help to find alignment between conservation goals and concerns important to cities and urban dwellers. Pollinators are small organisms that interact with households and neighborhoods, but operate in the larger landscape scale—and their associated habitats can leverage a variety of design and management activities that are underway in cities and offer pathways to connect nature and cultural heritage in urban communities. In short, monarchs (and other pollinators) point to new ways to "see" the city as a space for conservation, with new partners, new tools, and new practices.

#### HOW CAN THE MONARCH BUTTERFLY HELP US TO UNDERSTAND THE ROLE CITIES CAN PLAY IN CONSERVATION?

While many pollinators in general have wide public appeal, the monarch butterfly (Danaus plexippus) is arguably an ideal ambassador to engage the public on conservation issues. Monarchs are an iconic animal with a well-known migration and striking orange and black coloration. They also represent a powerful cultural symbol that facilitates people talking about conservation—and to one another (Gustafsson et al., 2015). Monarchs have been referred to as a convener: a species able to connect people across a continent who witness the stunning migration in their own backyard. Currently, there is heightened public awareness that monarchs, like many pollinators locally and globally, are declining rapidly. Over the last two decades, the eastern monarch population has decreased by more than 80% (Semmens et al., 2016), while the western population has declined by a staggering 97% (Schultz et al., 2017). The public interest in monarchs and a growing awareness of their plight create an opportunity to translate attitudes into practices that can help a range of pollinators across the urban landscape. Through our efforts and those of others, we are beginning to discover how monarchs are relevant to the future of conservation, as well as different entry points—from social justice and cultural history to sustainable food initiatives—for engaging people in creating urban habitat (Gustafsson et al., 2015; Derby Lewis et al., 2018).

# A MONARCH'S VIEW OF THE CITY

A combination of efforts such as the creation of a Federal Strategy to Promote the Health of Pollinators<sup>5</sup> and an assessment to determine whether monarchs need Endangered Species Act protection<sup>6</sup> , along with a variety of current urban monarch initiatives (e.g., Mayors' Monarch Pledge<sup>7</sup> and Monarch Watch<sup>8</sup> ), led the U.S. Fish and Wildlife Service and the Field Museum in Chicago to collaborate in assessing the role cities could play in helping to increase the amount of habitat available to support eastern monarch butterflies. While this research does not apply to the Western monarch population, whose numbers have declined so dramatically (Schultz et al., 2017) that individuals are not frequently observed in the urban landscape, lessons can be drawn to support broader pollinator habitat efforts in metropolitan landscapes throughout the United States.

The aim of this effort was (1) to evaluate how much the urban sector, which is the second largest land use sector in the eastern monarch's Midwestern breeding range (Thogmartin et al., 2017), could contribute to the national goal of adding 1.8 billion milkweed stems (Semmens et al., 2016) and (2) to identify best practices for engaging a diversity of urban stakeholders in the creation of monarch habitat. Our team of ecological and social scientists then translated this information into a suite of spatial and social planning tools to help decision makers identify where the biggest opportunities exist to increase monarch habitat, and guidance on how to turn potential into reality. Details of the methods used to estimate the number of existing and potential milkweed stems occurring in urban landscapes, and to identify strategies to engage stakeholder groups to create habitat, have been published elsewhere (Johnston et al., 2019). For the purposes of this perspective, we provide a high-level summary of the approach and key findings, and discuss implications of this work in a broader context.

To understand the ecological landscape from a monarch's perspective, we conducted field sampling to estimate how much milkweed is currently on the ground and quantified the potential space for planting additional monarch habitat (i.e., the amount of grass/shrub land cover identified using high-resolution imagery and LiDAR data<sup>9</sup> ) in different land use classes within four major metropolitan regions: St. Paul-Minneapolis, Chicago, Kansas City and Austin. For comparison purposes, land use types were consolidated into 16 classes based on land management (e.g., residential-single family, community and cultural, open space conservation, etc.). Milkweed stems were counted in three ways: (1) randomly sampled census blocks located at staggered distances along transects extending through the metropolitan area, (2) targeted sampling of open space conservation and nonconservation areas, and (3) targeted sampling of locations where milkweed was intentionally planted. Based on the density of milkweed present in each of the land use classes, we extrapolated the amount of milkweed that is currently present and the potential to add additional milkweed stems in metropolitan areas across the US eastern range of monarchs. Our findings indicate the collective impact of this potential contribution could provide

<sup>5</sup>Pollinator Partnership Action Plan. Available online at: https://www.whitehouse. gov/sites/whitehouse.gov/files/images/Blog/PPAP\_2016.pdf

<sup>6</sup>Assessing the Status of the Monarch Butterfly. Available online at: https://www.fws. gov/savethemonarch/SSA.html

<sup>7</sup>Mayors' Monarch Pledge is a program designed to support U.S. cities, municipalities, and other communities committing to create habitat for the monarchs and other pollinators. Available online at: https://www.nwf.org/Garden-For-Wildlife/About/National-Initiatives/Mayors-Monarch-Pledge.aspx

<sup>8</sup>Monarch Watch is "a nonprofit education, conservation, and research program based at the University of Kansas". Available online at: https://monarchwatch.org/ <sup>9</sup>Urban Tree Canopy Assessment. Available online at: https://www.nrs.fs.fed.us/ urban/utc/

nearly a third of the additional 1.8 billion milkweed stems needed in the Midwest to stabilize the eastern monarch's population (Johnston et al., 2019).

Additionally, we looked at the potential plantable space across each land use class to provide a more detailed characterization of the urban landscape. For example, in the Chicago region, we found that residential land had one of the highest amounts of potential plantable space. Using a land use lens allowed us to link high-potential areas with the stakeholders that would need to be engaged to increase monarch habitat. We were then able to pair those stakeholders with evidence-based approaches to enhance uptake (**Figure 2**).

To identify appropriate approaches for different stakeholder groups in an urban setting, we conducted social science research to assess the motivations, concerns, interests, challenges, and strategies of those both directly and indirectly involved in making their city's landscape more hospitable to monarchs. We surveyed people engaged in different environmental practices (e.g., planting/managing land, designing landscapes, monitoring the natural environment) and within the different land use classes laid out by the team's geospatial analysts. With people who had extensive knowledge or experience relevant to monarch conservation, we conducted semi-structured interviews, as they make efficient use of the participant's time and are well-suited to the exploratory phase of research (Schensul and LeCompte, 2013). For participants drawn from the "interested public," we used an online survey, which allowed us to reach more people. We collected and analyzed 734 online surveys and 75 semistructured interviews in the four pilot metropolitan areas and found that interest in creating monarch habitat was present to varying degrees across all groups, but it took different forms. For example, while some stakeholder groups are singularly focused on the monarch, others may be more interested in broader habitat creation and/or wary of the regulations that singlespecies conservation can bring. This information was used to highlight best practices for engaging urban stakeholders and to develop approaches that connect to community interests and assets (e.g., social justice initiatives, green infrastructure planning, urban farming efforts, public art) in engaging a wide cross-section of urban residents to take actions aligned with wildlife conservation goals.

#### DOES NATURE NEED CITIES?

Our results add to a growing body of literature showing that metropolitan areas matter for wildlife conservation (Morey et al., 2007; LaPoint et al., 2015; Caballero et al., 2016). Despite being developed, these landscapes have high potential to maintain functional habitat for a variety of species, including migratory and threatened endemic species. Habitat within and between US

<sup>10</sup>www.fieldmuseum.org/monarchs

cities can help connect the dots for monarchs, other pollinators, and birds along migratory pathways from Mexico to Canada and back.

The importance of cities for maintaining insect pollinators is particularly noticeable, given the relatively small spatial and temporal requirements for functional pollinator habitat that can be satisfied in urban green spaces. Although urban habitats are highly heterogeneous, with habitat often occurring in isolated patches, evidence suggests there is sufficient opportunity for pollinators to use these spaces (Tommasi et al., 2004; Glaum et al., 2017; Hall et al., 2017)—sometimes even greater opportunity than in surrounding rural areas.

As is the case with any land use sector, however, there are considerations that need to be addressed. For example, the widescale use of pesticide and herbicide in urban landscapes (Hladik and Kolpin, 2015) by public entities and private landowners poses a threat to insect population health. To ensure a net gain for pollinator populations utilizing urban habitat, approaches that limit insecticide exposure in urban areas are recommended.

Interdisciplinary methods that bring together the insights of social, natural and spatial sciences can shed light on the conservation approaches with the most ecological and social potential to scale effective solutions. Our work suggests that the collective impact of conservation-related actions by urban stakeholder groups can play a fundamental role in supporting wildlife—including nearly a third of the milkweed needed for the eastern monarch (Johnston et al., 2019). By identifying the ecological potential and understanding the social perspectives and interests of different stakeholder groups, it is possible to enhance the uptake of conservation strategies within urban areas, where these practices are important for threatened species.

Metropolitan areas also offer the opportunity to engage millions of people in conservation efforts. Despite urban areas' representing only 3% of the total landmass in the US, these areas have a disproportionate influence on the landscape, and investments must be made to turn the urban conservation potential into a reality. Expanding the functional habitat within these urban centers and increasing the commitment of urban stakeholder groups to conservation goals could greatly contribute to the achievement of those goals.

This means we must identify the different entry points where conservation goals can include input from urban partners and overlap with community values and concerns. Embracing community values as assets in conservation planning creates more opportunity for habitat and fosters meaningful new partnerships that are essential in highlighting conservation relevance in a rapidly expanding urban world.

A broader vision of what conservation is, what nature looks like beyond protected lands, and who is included in the conservation community is long overdue. Acknowledging that there are different ways that heritage and history shape how people experience the natural world, or see nature as a part of their lives, is an important first step in broadening the conservation community (Campbell, 2015). Our research indicates that cities can play a critical role in species and habitat conservation and that interdisciplinary approaches that engage urban stakeholders can have an outsize impact on wildlife conservation.

### AUTHOR CONTRIBUTIONS

ADL, MB, AW, EH, DS, MJ, and CZ contributed to the conception of the design and the study. ADL wrote the first draft of the manuscript. ADL, MB, and AW wrote sections of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

# FUNDING

A cooperative agreement between the United States Fish and Wildlife Service, Chicago Wilderness, and the Field Museum. Grant Award number CW FMNH 103115. Publication fees for open access were paid for by the Field Museum.

#### ACKNOWLEDGMENTS

We would like to thank the deeply committed and creative team of individuals whose work is reflected in this study, including Tim Bodeen, Katie Boyer, Brittany Buckles, Wendy Caldwell, Louise Clemency, Jill Erickson, Caitlin Dix, Ryan Drum, Nigel Golden, Adriana Fernandez, Jessica Hellman, Kyle Kasten, Marc Lambruschi, Laura Lukens, Cora Lund-Preston, Patrick Martin, Katie Maxwell, Tom Melius, Kelley Myers, Rosie Nguyen, Karen Oberhauser, John Rogner, Jason Rowader, Glen Salmon, Kristin Shaw, Wayne Thogmartin, Chuck Traxler, Kristin Voorhies, Gwen White, and Barbara Willy. We also thank Debra Moskovits, Ellen Woodward, Iza Redlinski, and Jacob Campbell for their review and comments on this paper.

#### REFERENCES


Audet, J., Ducatez, S., and Lefebvre, L. (2016). The town bird and the country bird: problem solving and immunocompetence vary with urbanization. Behav. Ecol. 27, 637–644. doi: 10.1093/beheco/arv201

Beatley, T. (2011). Biophilic Cities: Integrating Nature into Urban Design and Planning. Washington, DC: Island Press.

Caballero, I. C., Bates, J. M., Hennen, M., and Ashley, M. V. (2016). Sex in the city: breeding behavior of urban peregrine falcons in the Midwestern US. PLoS ONE 11:e0159054. doi: 10.1371/journal.pone.0159054

Benedict, M. E., and McMahon, E. T. (2006). Green Infrastructure: Linking Landscapes and Communities. Washington, DC: Island Press.


Zhang, W., Goodale, E., and Chen, J. (2014). How contact with nature affects children's biophilia, 420 biophobia and conservation attitude in China. Biol. Conserv. 177, 109–116. doi: 10.1016/j.biocon.2014.06.011

**Disclaimer:** 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.

**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 Derby Lewis, Bouman, Winter, Hasle, Stotz, Johnston, Klinger, Rosenthal and Czarnecki. 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.

# Estimating Milkweed Abundance in Metropolitan Areas Under Existing and User-Defined Scenarios

Mark K. Johnston\*, Erika M. Hasle, Karen R. Klinger, Marc P. Lambruschi, Abigail Derby Lewis, Douglas F. Stotz, Alexis M. Winter, Mark J. Bouman and Izabella Redlinski

*Keller Science Action Center, Field Museum, Chicago, IL, United States*

#### Edited by:

*Wayne E. Thogmartin, United States Geological Survey, United States*

#### Reviewed by:

*Carl Stenoien, University of Minnesota Twin Cities, United States Karen Tuerk, Monarch Joint Venture, United States*

> \*Correspondence: *Mark K. Johnston mjohnston@fieldmuseum.edu*

#### Specialty section:

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

> Received: *01 February 2019* Accepted: *21 May 2019* Published: *21 June 2019*

#### Citation:

*Johnston MK, Hasle EM, Klinger KR, Lambruschi MP, Derby Lewis A, Stotz DF, Winter AM, Bouman MJ and Redlinski I (2019) Estimating Milkweed Abundance in Metropolitan Areas Under Existing and User-Defined Scenarios. Front. Ecol. Evol. 7:210. doi: 10.3389/fevo.2019.00210* Metropolitan areas play an undetermined role in supporting migratory monarch butterfly (*Danaus plexippus*) populations despite providing habitat areas rich with milkweed (*Asclepias* spp.), the obligate host plants for monarch larvae. Researchers from the US Geological Survey and collaborating institutions have called for an "all hands on deck" approach to establishing monarch butterfly habitat by focusing on potential contributions from all land use sectors at levels necessary to sustain the eastern migratory monarch butterfly population. To understand the current and potential contribution of milkweed stems in metropolitan areas, our research teams surveyed milkweed densities using a new "metro-transect" protocol and conducted interviews and surveys across a diverse set of stakeholder groups in four major metropolitan areas (Chicago, Minneapolis-St. Paul, Kansas City, and Austin). We developed Geographic Information System (GIS) tools that use these data to model existing milkweed stems in metropolitan areas, and to estimate the potential to add additional milkweed stems with the adoption of milkweed-friendly planting practices across different land use classes (e.g., residential, institutional, and commercial). By extrapolating metropolitan Chicago milkweed densities across US Census urbanized areas in the northern US range of the eastern monarch butterflies, we estimate that approximately 29.8 million stems of milkweed can be added under modest "enhanced" milkweed densities, and up to 271 million stems may be added under "exemplary" milkweed densities. Both estimates are derived from a two percent "adoption rate," or landowner conversion of green spaces. These findings show that metropolitan areas provide important habitat opportunities and should be included prominently in monarch conservation strategies when working toward national goals to increase the amount of milkweed stems and monarch habitat across the Midwest. Municipal decision-makers and planners can estimate their capacity to add stems across the metropolitan landscape by identifying where the biggest opportunities exist with help from our Urban Monarch Conservation Planning Tools.

Keywords: milkweed, monarch butterfly, urban, metropolitan, habitat sampling, GIS tools, conservation

# INTRODUCTION

Monarch butterflies (Danaus plexippus) east of the Rocky Mountains migrate annually between central Mexico and Canada (Flockhart et al., 2013). The overwintering population has decreased more than 80% over two decades (Brower et al., 2012; Semmens et al., 2016). One prevailing theory for this decline is the milkweed-limitation hypothesis (Pleasants and Oberhauser, 2012; Pleasants et al., 2017; but see Dyer and Forister, 2016). This hypothesis posits that increased efficiency in agricultural practices (i.e., universal use of glyphosate and Roundup Ready crops) and widespread conversion of grasslands to other land uses have resulted in a precipitous decline in milkweed stems (Asclepias spp.) that monarch larvae depend on as their obligate host plants (Zaya et al., 2017). The US Fish and Wildlife Service (USFWS) is in the process of assessing whether monarch butterflies should be listed as threatened or endangered<sup>1</sup> . This determination will be based in part on a national Species Status Assessment (SSA) underway by the USFWS, which takes into consideration current and pledged habitat conservation and restoration efforts across all sectors. Research at the national scale suggests that a fivefold increase in milkweed stems is needed to address extinction risks associated with the eastern monarch population (Pleasants, 2017; Thogmartin et al., 2017a). To accomplish this increase, Thogmartin et al. (2017b) recommend an "all hands on deck" approach, with participation in milkweed restoration efforts by five land-cover sectors including "perennial herbaceous vegetation on protected lands, land enrolled in Conservation Reserve Program (CRP), lands in rights-of-way status, land associated with agricultural practices, and the urban/suburban sector." The USFWS and Monarch Joint Venture<sup>2</sup> have expanded the "all hands on deck" initiative to recommend a collaborative partnership effort by organizations and individuals to increase pollinator habitat.

"Urban/suburban" areas are classified as "developed" areas in the National Land Cover Dataset and Cropland Data Layer. These lands make up the second largest footprint (45,313 km<sup>2</sup> ) of the five land-cover sectors in the Midwest, but their importance and potential for breeding monarchs are not well understood. Efforts at the national scale are now underway to improve our understanding of milkweed and monarch butterfly densities. Some examples include the Integrated Monarch Monitoring Program (IMMP) (Cariveau et al., 2019), the Monarch Larva Monitoring Project (Prysby and Oberhauser, 2004; Kountoupes and Oberhauser, 2008), and sampling along roadside rights-ofway (Kasten et al., 2016). Based on our work in urban landscapes, we know milkweed is present in developed areas (**Figure 1**), and some studies have shown increased egg loading in urban area gardens (Cutting and Tallamy, 2015; Stenoien et al., 2015) and high potential in urban rights-of-way (Leston and Koper, 2016). However, published estimates of milkweed densities in urban landscapes are currently limited to the survey results compiled by Thogmartin et al. (2017b), which estimate 0.1 to 1.0 stems/acre in urban areas. Supplemental results from this US Geological Survey (USGS) publication are presently the sole data source used to inform the Monarch Conservation Database (MCD),<sup>3</sup> a USFWS database system that tracks the current and anticipated monarch habitat contributions of participating organizations across the US to inform the SSA. If a main limiting factor in rebounding migratory monarch populations is the decreased availability of milkweed plants, then it is imperative to have accurate estimates about the existing density and the habitat potential of milkweed in urban landscapes, and, because of the urgency of the issue, it is necessary to have them soon.

This paper addresses milkweed abundance across the various land use types that characterize US metropolitan areas—from their historic urban cores through suburbia and into the urbanrural fringe typical of federally-defined Metropolitan Areas4 and provides tools for estimating the potential for adding more habitat in these areas. In this study, we centered our analysis on "metropolitan areas" and "urbanized areas" as defined by the US Census, which provide established countybased boundaries for examining developed areas, rather than using more broadly defined terms such as "cities," "urban," and "suburban." Our first goal was to estimate the current amount of milkweed available to monarch butterflies in Chicago, Minneapolis-St. Paul, Kansas City, and Austin as a basis for understanding the relative contribution of major metropolitan areas; our second goal was to understand the capacity of these metropolitan areas for increasing the amount of milkweed available for breeding monarch butterflies; our third goal was to provide geospatial planning tools that clarify opportunities for habitat expansion and support the development of conservation strategies in metropolitan areas; and our fourth goal was to estimate the potential for adding milkweed across other urbanized areas in the northern range of monarch butterflies in the United States. Based on our past work, we hypothesized that existing milkweed densities in metropolitan areas are higher than previously published by Thogmartin et al. (2017b), and that these areas have the potential to contribute a sizeable portion of the milkweed needed to support the eastern population of monarch butterflies.

#### METHODS

Our study was conducted with support from local partners across four major metropolitan areas along the monarch's migratory route: Chicago, Minneapolis-St. Paul, Kansas City, and Austin (**Figure 2**). Selections were based on: their geographic location along the monarch flyway, organizations interested in partnering, geospatial data availability, and variation in environmental conditions (e.g., land use proportions and growing conditions).

<sup>1</sup>Assessing the status of the monarch butterfly https://www.fws.gov/ savethemonarch/SSA.html

<sup>2</sup> 2018 Monarch Conservation Implementation Plan https://monarchjoint venture.org/images/uploads/documents/2018\_Monarch\_Conservation \_Implementation\_Plan\_FINAL\_1.pdf

<sup>3</sup>The Monarch Conservation Database https://www.fws.gov/savethemonarch/ mcd.html

<sup>4</sup>As deployed by the US Census Bureau using the 2010 OMB definitions found in the June 28, 2010 Federal Register; https://www.census.gov/programs-surveys/ metro-micro.html

FIGURE 2 | Analysis extent of four metropolitan areas: Chicago, Minneapolis-Saint Paul, Kansas City, and Austin, where field research was conducted to study ecological and social factors related to monarch butterfly habitat in urban/suburban areas. Shaded areas represent urbanized areas (US Census) within the metropolitan analysis extent (heavy outlined boundaries). Metro-transect lines shown in blue run from densely populated urban centers to rural sparsely populated areas. Evenly spaced sampling points (red) show where field teams were directed for sampling.

Our research methods are organized into three sections briefly summarized here and further described below (**Figure 3**). The Primary Data Sources and Data Calculations section includes both data sources and density calculations, which are used as inputs for the models and tools. In the Estimation Methods section, we walk through our Baseline Estimation Methods, which describe our geospatial analyses and tools used for estimating existing stems of milkweed on the landscape that are used as input for the User-Defined Scenario Estimation Methods, which describes the methods and tool used for estimating habitat potential. Outputs from the baseline and scenario planning tools were then used in the Extrapolation Methods section for estimating baseline, enhanced, and exemplary milkweed counts across urbanized areas in the eastern US monarch butterfly range.

#### Primary Data Sources and Data Calculations Geospatial Data

In developing our methods for sampling across metropolitan areas, we used precise land use data and high-resolution land cover data, typically provided by Metropolitan Planning Organizations (MPOs)—although data coverage, classification, and resolution can vary substantially across regions. For all four metropolitan areas, we sought data to cover the entire metropolitan area at the finest scale possible. In Austin, the regional planning agency serves Travis, Williamson, Bastrop, Caldwell, and Hays counties. Since available data only included Travis County, where Austin is located, we reduced our analysis extent to this central county. A list of key data sources used is found in **Supplementary Table 1**.

The National Land Cover Database covers the nation with 30 meter resolution cells sorted into 20 land cover classes, including low, medium, and high density developed land. Increasingly, higher resolution data are available for major urban areas, though typically with fewer land cover classes. For Chicago, Minneapolis-St. Paul, and Austin we had high-resolution land cover data (two-foot in Chicago, one-meter in Minneapolis-St. Paul, and three-foot in Austin), which included seven land cover classes for Chicago and Minneapolis-St. Paul (bare soil, buildings, grass/shrub, other paved surfaces, roads/railroads, tree canopy, and water) and only one land cover class (tree canopy) for Austin, which we combined with a "remaining pervious" GIS layer. Kansas City presented a significant data challenge, as the land cover layer was at a much coarser (eight-foot) resolution.

We created a dataset called "plantable space," to describe both where pollinator habitat exists, and where non-forested green space exists that could potentially be converted to pollinator habitat. This layer was primarily derived from the grass/shrub land cover class. Although many green spaces may have a land use that is not compatible with conversion to pollinator habitat (e.g., recreational sport fields and airports), throughout much of metropolitan areas, these existing non-forested green spaces are where most of the opportunities to plant milkweed exist. While there are some exceptions where built up (hardscape) areas may be converted to habitat areas (e.g., the conversion of a rail line), the vast majority of opportunities are in existing vegetated areas

includes everything from turf grass and shrubs to high quality prairie sites. In many cases only a fraction of plantable space might be reasonably converted into pollinator habitat due to land use restrictions (e.g., sports fields), but by eliminating built up areas (e.g., roads) we can improve our density estimates.

due to lower conversion costs. We excluded the majority of tree canopy, because densely forested areas do not have enough light penetration to support the growth of most milkweed species, although Chaplin and Walker (1982) note that fourleaf milkweed (Asclepias quadrifolia) is commonly found in low-light forest understory. From expert opinion and estimates based on the proportion of our occurrence data located under tree canopy, we estimate that approximately 20% of forested areas provide suitable habitat for milkweed species capable of growing under the partial shade of solitary trees and along the woodland edge. In densely forested areas, plantable space calculations included only grass/shrub land cover, while in all other green spaces, plantable space calculations included grass/shrub as well as 20% of tree canopy. To examine how plantable space is distributed across land use categories, we isolated the grass/shrub land cover class and then reclassified it by the land use class where it occurs. This combined data layer is the basis for our modeling tools (**Figure 4**).

Inventories of existing land use are commonly divided into parent categories such as residential, commercial, and industrial. These datasets are developed by MPOs and—for the four metropolitan areas—originally contained between 22 and 60 land use classes. To facilitate comparisons across our study areas, we consolidated land use classes into 16 standardized categories, bearing in mind as we did so that land use in this case had much to do with how land users might manage their land for monarch conservation (**Table 1**). Both the creation of our 16 land use classes and the consolidation of each city's land use into these classes was done by our multidisciplinary team of ecologists, social scientists, and geospatial analysts. We relied heavily on the metadata provided with each dataset and consulted with local partners as we combined classes together. For more details see **Supplementary Section 1**.

The way metropolitan areas classify open space within their land use is an important factor influencing both our plantable space and our final baseline calculations. In Austin and Chicago, the land use data sets differentiated between recreational open



*For more details on land use category descriptions, see* Supplementary Table 2*.*

space and open space managed for conservation, whereas in Minneapolis-St. Paul and Kansas City, the land use data did not differentiate between these categories. For Minneapolis-St. Paul we migrated their combined "preserves and parks" land use into Open space conservation (OS-C) because it contained a large wildlife refuge, and in Kansas City we migrated their "parks and open space" land use into Open space non-conservation (OS-NC) in consultation with local partners.

Of the 16 defined land use classes, three encompass the land along rights-of-way, which makes up the third largest sector (43,148 km<sup>2</sup> ) of the five examined by Thogmartin et al. (2017b). We extracted minor road rights-of-way to incorporate the strip of land or "greenway" between streets and sidewalks or drainage ditches, typically owned by the city or county in more rural areas. Often Minor right-of-way areas are not separated from Major rights-of-way. We isolated Minor right-of-way areas by applying a series of steps including buffering and extracting intersecting major roads, highways and ramps. For definitions of all land use categories see **Supplementary Table 2**.

To assist field teams conducting natural areas sampling in each metropolitan area, we used various geospatial protected area layers from the local landowner authority, or state/national protected land datasets<sup>5</sup> to aid in planning field sampling locations. These boundaries were also used in calculating milkweed densities. US census blocks were used as geographic analysis units for aggregating land use and plantable space model inputs and outputs to/from our geospatial tools. Census blocks vary in size based on population density and allow for comparison of land use areas by demographic characteristics.

For extrapolation when estimating baseline and scenario estimates for urban areas at regional scales, we used a USGSdeveloped habitat raster developed for a national milkweed calculator which estimates the amount of milkweed present at the county scale. This tool uses as its basis for extrapolation a combined raster layer comprised of the Cropland Data Layer (CDL), the National Land Cover Database (NLCD), and US Census Bureau right-of-way data<sup>6</sup> .

#### Field Sampling and Density Calculation Methods

We used three field sampling methods to acquire information about current milkweed densities across different land use classes including: (1) randomized metro-transects for baseline density estimates, (2) targeted sampling of natural areas, and (3) targeted sampling of other "enhanced sites." Both methods of targeted sampling were used for enhanced and exemplary density estimates. Sampling occurred throughout the 2016 growing season, approximately June through September of 2016 (and 2017 in Chicago). To estimate the baseline density of milkweed in each metropolitan land use area, milkweed stems were quantified using randomly sampled "metro-transects" running from more developed to less developed lands. While this approach allowed us to quickly and efficiently examine existing milkweed densities in most land use classes, this method was not practical for examining natural areas in Open space conservation and Open space non-conservation lands. Because sampling in natural areas required concentrated surveying in appropriate areas, advanced planning, and ample survey time, we conducted targeted rather than randomized sampling of natural areas. Since this approach was biased toward sites with higher milkweed densities, baseline densities for Open space conservation and Open space nonconservation areas were calculated using alternative methods described below. In addition to targeted sampling of natural areas, targeted "enhanced sites" sampling was conducted for all land use classes to get an understanding of the typical density of milkweed at sites where it is intentionally planted. The top ten percent of these enhanced sites were used to define "exemplary sites," which set our upper limit on the amount of greenspace landowners might convert to habitat across different land uses.

Baseline densities were calculated on both a plantable space basis as well as on a total land use (LU) basis. For densely forested areas, we calculated plantable space on a grass/shrub (GS) basis and for all other green spaces, we calculated plantable space on a grass/shrub plus 20% tree canopy (GST) basis. To illustrate these density calculations, suppose 10 milkweed stems were counted in the agricultural component of a sampled site. For density calculations on an LU basis, we calculate the baseline density as the number of milkweed (10) divided by the acres of the site's agricultural component. For density calculations on a GST basis, we calculate the baseline density as the number of milkweed (10) divided by the acres of grass/shrub plus 20% of the acres of tree canopy that occur in that agricultural component as calculated from the plantable space land cover data reclassified by land use. Density calculations on a plantable space basis result in a higher number of milkweed stems per acre than calculations on an LU basis. This is because the total stem count is divided by a smaller area representing plantable space as opposed to dividing by the entire land use area.

#### **Metro-transects sampling and milkweed density calculations**

In the random sampling protocol called "metro-transects," we sampled milkweed densities and habitat information across the metropolitan population gradient in Chicago, Kansas City, and Austin. Due to limitations in the number and time available of field crew members, Minneapolis-St. Paul restricted their sampling to 38 randomly selected neighborhood blocks inside the city limits. Metro-transects were run from highly urban areas through suburban and into rural areas at the outermost edge of the counties that comprise US Census-defined metropolitan areas. Transects used randomly staggered start distances between zero to four miles and were followed by evenly spaced sampling points every five miles in Chicago, and every four miles in Kansas City and Austin (**Figure 2**). The larger span between sampling points in Chicago was to accommodate the longer transect distances originating from the populated lakefront (on the eastern edge), as opposed to Kansas City and Austin where the urban cores are more centrally located within the metropolitan area. In all cases, we purposely angled our transect lines to avoid aligning with highly rectangular road networks which could otherwise have systematically skewed our sampling. **Figure 5** shows examples of sampling locations along metro-transect lines in the Chicago metropolitan area.

"Sampling clusters" were delineated around each sampling point by taking the intersection of all US census blocks within a 100-meter buffer from the sampling point. By using census block clusters, boundaries were clearly defined for the field team since the clusters align well with streets and property lines. Field teams conducted both walking and driving surveys. In Kansas City, 54 random sampling clusters were visited; in Austin, 66 sampling clusters were visited; and in Chicago, 65 were visited.

<sup>5</sup> e.g., Protected Areas Database of the US (PADUS) https://gapanalysis.usgs.gov/ padus

<sup>6</sup>A detailed methodology for how this raster was developed can be obtained by downloading the user manual for the desktop Monarch Conservation Tools: http://www.umesc.usgs.gov/management/dss/monarch/ desktop\_monarch\_conservation\_planning\_tools.html

Shading color indicates the land use, points are where habitat was encountered (red point contain milkweed), and hatching indicates areas that could not be sampled.

Within each sampling cluster, the field teams counted the number of stems of milkweed by species. Patches of land with plants in bloom were noted, and depending on the size of the patch, were recorded as either individual point data (typically small patches less than 25 square feet), or as a polygon within which plants were counted. Additional collected data included estimated patch size, percent native plants, percent volunteer/weed patch, percent ornamental plants, percent food garden, and number of native and blooming species; these categories were not mutually exclusive. We reserved the "native plants category" for areas that appeared to be intentional plantings and used the "volunteer/weed patch" category for unmanaged areas. These additional data attributes were not utilized for the present study. For field data collection, we used Android-based tablets with a built-in GPS and ESRI Collector application (version 18.0.1) pre-loaded with sampling locations and data templates. Field teams captured data for all land use classes that intersected the sampling cluster with the exception of Open space conservation and Open space non-conservation areas, since separate sampling methods were used for capturing stem densities across natural areas (see below).

Field visits were limited to a maximum of two to three hours for each sampling cluster. Areas that could not be observed from the public right-of-way were excluded from the analysis unless permission was granted by the landowner. This excluded most backyards in residential areas. Since residents may have planted or allowed milkweed plants to "volunteer" in their backyards, we recognize this may underestimate residential milkweed densities. Field technicians marked areas not sampled due to time limitations or inaccessibility and provided a reason why sampling did not occur. These areas were removed from our density calculations.

Each sampling cluster was divided into "components" based on land use type. For example, a single cluster could be made up of an Agricultural component, an Industrial-small component, etc. Within each survey cluster, if a recorded milkweed polygon crossed multiple land use types, the percentage of the polygon within each land use was calculated and the number of milkweed stems counted for the polygon was applied proportionally to each land use component based on that area percentage. Thus, the total stem count for a land use component within a sampling cluster is comprised of the total stem count from individual points that fall within the land use component in addition to the proportion of milkweed stems contributed by polygon sampling areas.

For more accurate estimates of the existing baseline and distribution of milkweed stems, the milkweed density for each land use component was calculated in stems per acre on a plantable space (or GST) basis for use with our geospatial tools, and on a total land use basis to align with other research standards in the literature. To calculate the average milkweed density for each land use class, we took all of the metro-transect components

that shared the same land use and calculated the mean density (on both a GST and LU basis) along with the standard deviation and standard error. Because there was a large variation in component size within a given land use class (e.g., agricultural components ranged in land use size from 0.03 acres to 341 acres), we calculated both standard and weighted means. The latter allows larger areas to contribute more to the mean. We calculated the weight that would be applied to each component's GST and LU density by taking the area sampled for each component and dividing it by the sum of all land use components sampled. The density for each component was then multiplied by each component's weight. The mean weighted density was calculated for each land use class.

#### **Targeted sampling of natural areas and milkweed density calculations**

Milkweed densities from metro-transect surveys were calculated for all land use classes except natural areas. For this study, natural areas were divided into two categories: Open space conservation and Open space non-conservation. OS-C refers to lands which are managed primarily for conservation (such as a state park or a wilderness area) while OS-NC refers to lands with a primarily recreation or other land use, but which could have natural areas as a secondary land use (such as a city park or a municipal golf course).

To sample natural areas, we carried out a "targeted" sampling strategy distinct from metro-transects for several reasons: (1) natural areas are made up of many vegetation communities, some of which are not milkweed habitat, and we wanted to ensure that our field teams were collecting data within areas appropriate for milkweed habitat; (2) natural areas require advanced planning with randomized site level transects and landowner permits for scientific research; and (3) natural areas sampling often takes several hours, disrupting the workflow of metro-transect sampling. For both OS-NC and OS-C sites, scientific research permits were obtained from the land-owning agencies and land managers were consulted about the sites and units most likely to have milkweed habitat. This process allowed us to target our sampling to areas where we would likely find milkweed stems. Because we were unable to obtain site histories for many of our sites, we were not able to control for the restoration age of the places we sampled. In Austin, 51 natural areas were visited; in Chicago, 27 natural areas were visited. Natural areas in Minneapolis-St. Paul were not sampled due to time constraints, and in Kansas City, available GIS data was inadequate to follow our sampling protocol. The full sampling protocol for measuring milkweed density in natural areas can be found in **Supplementary Section 2** and on the Field Museum's website<sup>7</sup> .

The milkweed density for each natural area sampling site was calculated based on the area surveyed. Due to the variable densities of milkweed within and among sites and our method of targeted sampling, we removed the statistical outliers using the 1.5 x Interquartile Range (IQR) Rule for both OS-NC and OS-C lands. After removing values above the third quartile + 1.5 IQR as sampling site outliers, the mean density of the remaining areas within each land use class was calculated along with the standard deviation and standard error. We termed these our "hotspot" densities, which represented the average density that was found in enhanced sites.

We know that not all OS-NC and OS-C lands will have the same milkweed density as their respective hotspot areas. These lands are managed for multiple uses that are not compatible with pollinator habitat, including developed areas and grassy areas that are regularly mowed. As a result, we sought to determine the proportion of open space lands that are maintained as nonforested natural areas and thus are compatible for milkweed habitat. We designated this compatible habitat as plantable noncultural (PNC) land. Below we present equations used to calculate the percent of plantable non-cultural land, and the total milkweed stems within OS-C and OS-NC. We used GIS data and numbers provided by major landowning agencies in the Chicago area to determine what percent of OS-NC and OS-C land is typically managed as natural areas (defined as non-cultural land), as opposed to areas maintained for recreational activities such as sports fields and picnic areas (defined as cultural land). For OS-NC, the Chicago Park District reports that of its 8,832 acres, 1,850 acres are managed as natural areas<sup>8</sup> . To get this on a plantable space basis, we calculated the acreage of grass/shrub area and added 20% of the tree canopy area within the park district's natural areas. Based on consultation with land managers, we characterize trees in parks as being dispersed, allowing enough light penetration for milkweed habitat. We used GIS analysis to determine the percentage of natural areas within OS-NC lands that contain plantable space for a known sample of natural areas (defined as GST in Sampled Natural Areas), and applied that proportion to the 1,850 acres of park land managed for natural areas to get a more accurate estimate of plantable space in all OS-NC natural areas (GST in NA). The result, when divided by the calculated acres of plantable space within all OS-NC lands (Total GST in LU), produced the percentage of plantable space within OS-NC lands that are within natural areas (defined as %PNC land).

$$\frac{\text{GST in NA}}{\text{Total GST in LU}} \times 100 = \% \text{PNC land}$$

The percent of plantable non-cultural land was applied to the amount of land in a given land use that is plantable space to derive the amount of land that is available for milkweed (Total PNC area in LU).

%PNC land × Acres of plantable space in LU = Total PNC area in LU

This was then multiplied by different milkweed densities (for baseline, enhanced, and exemplary estimations) to

<sup>7</sup>http://www.fieldmuseum.org/monarchs

<sup>8</sup>https://assets.chicagoparkdistrict.com/s3fs-public/documents/departments/ budget/2019%20Budget%20Summary.pdf

calculate the total number of milkweed stems in the land use:

Total PNC area in LU × Milkweed Density = Total Milkweed stems per LU

To apply this equation for baseline estimates, we used the milkweed density from Thogmartin et al. (2017b, Supplement table 3.1) for Conservation Reserve Program-Non-wet (CRP-NW, 112.14 milkweed stems/acre), which we estimate as the closest baseline approximation for natural areas. For enhanced density estimates, we used the hotspot milkweed density from our targeted surveys. For exemplary density estimates, we applied the mean milkweed density observed in the top ten percent of OS-NC sites, including outliers.

We were unable to calculate the percent of plantable non-cultural land for OS-C using the same methodology as OS-NC, because the GIS data we had for lands within the Forest Preserve District of Cook County (FPDCC) was not recommended for this purpose. Thus, for our density calculations, we used their published information. In their Natural and Cultural Resources Master Plan, the FPDCC describes 27.6% of their land as "cultural," which includes developed and heavily altered vegetation (including mowing)<sup>9</sup> . From this, we derived that 72.4% of their land is managed as natural areas (Non-cultural land). We estimate that this percentage approximates the proportion of plantable space managed in natural areas (Percent of Plantable non-cultural land). As a check to this percentage, we also used the GIS data provided by FPDCC to calculate the percentage of natural areas that were within OS-C plantable space. We calculated the acreage of natural areas that could be considered suitable milkweed habitat, which included Eurasian meadow, prairie, savanna, sedge meadow, and shrubland, and divided it by the acres of grass/shrub land within OS-C. The result (73.4%) was similar; therefore, we decided to apply the more established 72.4% estimate. Using the same formula above, we applied the 72.4% (%PNC land) to the acres of plantable space in OS-C land to derive the total plantable non-cultural area in OS-C. Plantable space was calculated on a grass/shrub basis to account for the densely forested nature in this land use category, which would not provide adequate sunlight for milkweed. To apply this equation for baseline estimates, we used the national milkweed density from Thogmartin et al. (2017b, Supplement table 3.1) for CRP-NW (112.14 milkweed stems/acre). For enhanced density estimates, we used the hotspot milkweed density from our targeted surveys. For exemplary density estimates, we applied the mean milkweed density observed in the top ten percent of OS-C sites, including outliers.

The baseline density estimates used for OS-C and OS-NC in Chicago were also used in Minneapolis-St. Paul and Kansas City. For Austin's baseline, we applied the sample percentages of plantable non-cultural land to the milkweed densities recorded in Austin for OS-C and OS-NC hotspots, since we used the same sampling approach as in Chicago. While this assumes that the percent of natural areas in OS-C and OS-NC in Austin are about the same as they are in Chicago, we lack the GIS data on natural areas in Austin to test this.

#### **Enhanced and exemplary sites sampling and milkweed density calculation**

To run the scenario planning tool, we used estimated milkweed densities at sites with intentionally planted milkweed and sampled densities from OS-C and OS-NC lands. We divided those sites into two categories: enhanced sites, which refer to the average density of milkweed at sites with over five total stems of intentionally planted milkweed, along with exemplary sites, which refers to the average of the milkweed densities observed in the top ten percent of sites in each land use category, including outliers. As previously mentioned, the purpose of the targeted sampling of exemplary sites is to understand the upper threshold of what people are willing to plant on their land. These sites were located based on a snowball sampling method (Schensul and LeCompte, 2010), in which we contacted people through our networks and through participants in our other monarch work to find patches of intentionally planted milkweed across the 16 land use categories. For every site with at least five milkweed stems, the milkweed density was calculated on a grass/shrub and 20% tree canopy basis for the surveyed area, and the mean density, standard deviation, and standard error were calculated for each land use class. For land use classes where we were unable to find sites with more than five stems of milkweed, we used parcels with over five stems of milkweed collected through our metro-transect surveys. We used the average density found at enhanced sites to estimate the milkweed density likely to be found at sites where milkweed is intentionally planted, and we use the exemplary site density for the realistic maximum that landowners will likely implement. We assume that targeted future installations and restorations will on average have similar milkweed densities as those measured at these sites. Enhanced and exemplary densities are used in combination with "adoption rates" which estimate the percent land area that landowners will convert into similar milkweed densities for a given geographic extent.

#### **Municipal case study sampling**

Municipalities have increasingly taken steps to enhance monarch and pollinator habitat within their jurisdiction, such as signing the Mayors' Monarch Pledge10. Such places offer a glimpse into what "enhanced" and "exemplary" scenarios might look like, should a wide variety of actions be taken. In the Chicago region, we worked with two local municipalities, Glenview and Schaumburg, to estimate milkweed densities at localized scales for comparison with regional estimates, and as case studies for

<sup>9</sup>http://fpdcc.com/downloads/plans/FPCC-Natural-Cultural-Resources-Master-Plan\_3-9-15\_WEB.pdf.

<sup>10</sup>https://www.nwf.org/Garden-For-Wildlife/About/National-Initiatives/Mayors-Monarch-Pledge.aspx

testing the application of our monarch conservation planning tools. We partnered with the Natural Resources Manager for the Village of Glenview to deploy a field team in summer 2017 to comprehensively survey all of Glenview visible from the public right-of-way, using similar data gathering techniques deployed in metro-transect sampling. As with metro-transect sampling, areas that were not accessible/visible (e.g., most backyards) were digitally captured by the field team and later removed from density calculations. In the Village of Schaumburg, we partnered with the Landscape and Sustainability Planner for the Village to survey a random selection of census blocks. In all other aspects, the field sampling methods were the same.

#### Social Science Surveys and Interview Methods

Our research team included social scientists, who conducted interviews and surveys to understand the current and potential contribution of cities from a more qualitative approach. Based on our prior ethnographic work with Chicago communities, the team's social scientists began this project knowing that many people in and around cities care about monarchs and/or nature more broadly11. The goal of the social science research was to understand what this enthusiasm and interest adds up to; how these activities connect to other issues; and which engagement strategies are most effective with which groups. To this end, social scientists collected 734 online surveys and conducted 76 phone or in-person semi-structured interviews across the four pilot metropolitan areas, asking participants about their conservation beliefs and practices with extra attention given to monarchs and milkweed. Study participants included members of faith-based organizations, universities, elementary schools, community gardens, parks and recreation departments, utility companies, conservancies, departments of transportation, as well as citizen scientists, home gardeners, land managers, landscape designers, and others. Researchers employed snowball sampling to recruit participants through their organization's conservation community connections, and thus reached a population more conservation-oriented than the general population. Of particular relevance to the geospatial estimates discussed in this paper are the questions we asked participants about the land they manage, such as: how much of the total plantable open space on this site is made up of native plants?; how much milkweed is there at this site?; and, in the next 5 years, would you be willing and able to convert more of the plantable open space of this site to native plants and/or native milkweed?

Social science surveys and interviews were used in part to validate our ecological methods and model assumptions (e.g., enhanced and exemplary concepts and methods), and helped to inform our understanding about adoption rates used in our scenario planning tool. Surveys and interviews were also used to identify landowners with enhanced and exemplary sites across land use sectors. Knowing the landowners helped our field teams locate enhanced sites to visit.

# Estimation Methods

#### Baseline Estimation Methods

We developed an Urban Milkweed Baseline Tool for estimating the total amount of existing milkweed and the average stem density for each census block in the metropolitan area. This tool, along with our Scenario-Based Planning Tool, Urban Monarch Guidebook, and Urban Monarch Conservation Planning Toolsets, is freely available via our website12. The baseline and scenario-based planning tools are designed to allow municipal and regional planning agencies, major landowners, and federal, state, and non-profit conservation organizations to estimate their current milkweed contribution in support of monarchs, and to assist in goal-setting for monarch butterfly conservation planning. The tools also allow users to examine potential co-benefits and opportunities available through the combined application of establishing pollinator habitat areas while addressing other goals and infrastructure issues such as stormwater runoff, flooding, and compliance issues. These geospatial tools were developed in collaboration with USGS staff by modifying a national Milkweed Calculator Tool developed by the Monarch Conservation Science Partnership13. GIS users are encouraged to download and utilize the tools, manual, and guidebook for these and other planning purposes.

Our baseline milkweed estimation tool uses US census blocks with a modified attribute table (for the four metropolitan areas where we sampled). This modified table includes, for every census block, the total amount of plantable space by each of our 16 consolidated land use classes. These values were calculated for each census block by summing the total amount of grass/shrub and 20% tree canopy using the Tabulate Area and Spatial Join tools (ArcMap Spatial Analyst extension version 10.6). We recorded these values in square feet both to be consistent with the unit of measurement from the layer (Illinois State Plane, US survey feet), and to use a more accurate measure than fractions of acres for the many small patches surveyed. Ultimately, our geospatial tools convert all measures in square feet to acres. All densities are calculated on a stems per acre basis.

Using the weighted mean milkweed densities by land use class, the calculator applies these densities to each census block and provides both a numeric and graphic output showing the estimated number of milkweed stems occurring in each census block and the average stem density. We used this tool for estimating the total stem count for each of the four metropolitan areas. Using the attribute tables from these model outputs, it is possible to summarize stem contributions by land use class.

#### User-Defined Scenario Estimation Methods

We use the output from the Urban Milkweed Baseline Tool along with user-supplied adoption rates as inputs for running our Scenario-Based Planning Tool. This tool estimates milkweed stem counts and density by census blocks after applying user scenarios across land use types. More specifically, users apply an estimated adoption rate (see below) for one or more land use class(es) and

<sup>12</sup>http://www.fieldmuseum.org/monarchs

<sup>13</sup>https://www.umesc.usgs.gov/management/dss/monarch/

desktop\_monarch\_conservation\_planning\_tools.html

<sup>11</sup>http://climatechicago.fieldmuseum.org/pilsen

can apply an "enhanced," "exemplary," or user-supplied milkweed density for the land use classes of interest. Since these tools use small geographic units (census blocks) as their basis, they can be combined with other data layers (e.g., public parks, vacant lots, utility corridors, or planning project areas) to give area and project-specific milkweed stem density and total count estimates. For more information about the tool see our Urban Monarch Guidebook, or for information on how to access and use the tool itself, see the Urban Monarch Conservation Planning Toolsets manual on our website<sup>14</sup> .

#### **Estimating Adoption Rates**

Adoption rate is defined as the rate at which landowners convert their plantable space to habitat with a milkweed density akin to what we have observed and calculated from either enhanced or exemplary sites. In referring to a two percent adoption rate, for example, for a given land use class (e.g., Residential-single family), we are estimating that two percent of all grass/shrub area within that land use class will have a stem density that is equivalent to the target density specified by the user (e.g., enhanced or exemplary stem density) when running the scenario tool. Since our exemplary stem densities are meant to be aspirational, these can be used for setting the upper limit of goalsetting over a longer term, while enhanced densities are meant to reflect a starting point for goal-setting over a shorter term.

When using our tool, the user inputs the adoption rate for their analysis extent. For the purpose of exploring what may be possible at local and regional scales, we use two, five, and ten percent adoption rates at enhanced densities, as well as a two percent adoption rate at exemplary densities. To determine appropriate adoption rates, we would ideally conduct a longitudinal study on the conversion of green space into habitat over time while also tracking changes in milkweed density. However, since our study did not directly research adoption rates, we used a space-for-time comparison<sup>15</sup> to examine reasonable adoption rates. To do this we compared case studies in Glenview and Schaumburg where pollinator work has been recently focused, to the Chicago region where much less pollinator work has been done. Thus, we consider the larger regional area as time zero (t0), and the local density in Glenview and Schaumburg as time one (t1). For each land use class, we then used a pairwise comparison of our baseline milkweed density from the Chicago region (Rd) with the local-baseline milkweed density (Ld) over a period of time (t1-t0) in years of focused monarch activity. The following equation shows the relationship between the calculated adoption rate (AR) and the enhanced or exemplary milkweed density (Ed):

$$AR = \frac{Ld - Rd}{(t\_1 - t\_0) \times Ed}$$

where the adoption rate (AR) is equal to the difference in local density (at t1) and regional densities (at t0), divided by time in years (t1-t0) over which this transformation in density took place, multiplied by the enhanced or exemplary density (Ed). The adoption rate for enhanced and exemplary densities are calculated separately. The assumption is that the main driver of higher observed densities is known and is not due to environmental or geographic factors. Our sampling in Glenview and Schaumburg indicates that a two percent adoption rate at exemplary densities is a reasonable scenario, given aggressive local action (see **Supplementary Section 3**).

#### Extrapolation Methods

The USGS has delineated several core monarch model regions (Rohweder and Thogmartin, 2016). Based on our findings for the Chicago metropolitan area, we extrapolate across the North Central and Northeast monarch regions, and these regions combined with the South monarch region, to estimate what a two percent adoption rate might look like for all urbanized areas in these core areas using our baseline, enhanced, and exemplary milkweed densities (**Supplementary Figure 1**). The analysis areas within these core regions are US Census urbanized areas, which are defined as urban areas with a population of 50,000 or more people<sup>16</sup> .

There are significant model assumptions in this extrapolation, including: environmental growing ranges and conditions of milkweeds, similar socioeconomic conditions to Chicago, social momentum and desire to support monarchs, accuracy of our milkweed density estimates, and translation of our model onto the USGS habitat raster<sup>17</sup> and its applicability across the eastern range. In addition, when considering extrapolations to other metropolitan areas, it is worth knowing that the Chicago region has long been engaged in open space preservation and ecological restoration that may not be as prevalent elsewhere and is among the many local circumstances that should be borne in mind when making broad generalizations across land cover categories (Heneghan et al., 2012; Crane et al., 2014; Watkins et al., 2015). Therefore, we consider these extrapolation results as useful in goal-setting and as aspirational particularly for our exemplary results. This approach relies on an areas-of-overlap matrix between our 16 consolidated land use classes and the 35 land cover classes used in the USGS habitat raster (see **Supplementary Section 4** for detailed methods). We applied this matrix to our baseline milkweed densities (calculated on a total land use basis) to translate our densities into coefficients for use with the USGS habitat raster. This process was then repeated for both enhanced and exemplary densities. For our Open space conservation and non-conservation baseline densities, we used the CRP-NW values from Thogmartin et al. (2017b, Supplement), which is further explained in the discussion.

After calculating all of the USGS habitat coefficients for baseline, enhanced, and exemplary densities, we summarized the total acres for the 35 USGS habitat land cover classes within all urbanized areas across the US eastern range. We then multiplied each land use coefficient by the acreage of each land use class

<sup>14</sup>www.fieldmuseum.org/monarchs

<sup>15</sup>Substituting space for time is a common practice when studying ecological systems with a component of time [e.g., climate change (Blois et al., 2013), and ecological forecasting (Banet and Trexler, 2013)].

<sup>16</sup>https://www.census.gov/geo/reference/urban-rural.html

<sup>17</sup>https://www.umesc.usgs.gov/management/dss/monarch/

desktop\_monarch\_conservation\_planning\_tools.html

and totaled the number of estimated stems for the baseline. The process for enhanced and exemplary estimates used the following formula:

$$\mathcal{S} = (L \times AR)(E - B)$$

Where the estimated total number of additional milkweed stems (S) for a given land use class, is equal to the total land use area (L) for a given class, multiplied by the adoption rate (AR) this gives us the proportion of land that is being converted times the difference between the enhanced or exemplary density (E) minus the baseline density (B). Note that enhanced and exemplary are calculated separately, and the total estimate of additional stems is the sum of estimated stems across all land use classes.

#### RESULTS

#### Plantable Space

Across the four metropolitan areas, we found that about half of all plantable space is found in agricultural areas. The proportion was lower in Austin because our study area was limited to the central county in the 5-county area. After Agricultural land, Residential-single family was the second largest land use category by area for three of the four metropolitan areas, and the third largest for Minneapolis-St. Paul (**Figure 6**). In Minneapolis-St. Paul, Vacant lots made up the second largest land use category because it included land with buildings, whereas in other metropolitan areas Vacant lots did not include land with buildings. Other land use classes making up significant portions within metropolitan areas included Open space conservation, Rights-of-way categories, and Vacant lots (without buildings).

# Milkweed Densities

#### Chicago Baseline Milkweed Density Estimates

Metro-transect sampling methods were effective at capturing milkweed densities across major metropolitan areas despite the low detectability of milkweed across such a vast area. Of the 54 random sampling clusters visited in Kansas City, 25 contained milkweed; in Austin, 50 of the 66 clusters visited contained milkweed; and in Chicago, 53 of the 66 visited clusters contained milkweed. In Chicago, Residential-common space and multifamily (Res-C) had the highest stem density (weighted mean = 18.9 stems/acre), followed by Minor road rights-of-way (Rd, 7.4), Restricted use rights-of-way (RU, 5.4), Vacant lots (Vac, 4.8), and Industrial-small (Ind-S, 3.9) (**Figure 7**). Due to variability in milkweed densities among Res-C and Ind-S sites, these land use classes had large standard errors. It is also worth mentioning that Residential-single family (Res-S, 1.4), while lower than many land use classes, carries a lot of influence on the final milkweed stem counts due to the sizable extent occupied by residential properties in metropolitan areas. Several of the remaining land use classes were either not observed within our randomized sampling clusters due to their scarcity across the landscape, or milkweed was rarely or never observed. These classes include Commercial (Comr), Industrial-large (Ind-L), Transitional and restricted use (TR), and Water (W). Consequently, these land use categories have an existing milkweed stem density that is approximated as zero stems per acre. Milkweed density values for all land use classes from both metro-transect sampling and natural area sampling are combined in **Table 2**. This table is also available with calculations on a total land use basis rather than plantable space in **Supplementary Table 3**. To examine whether there is a difference in milkweed densities in more densely vs. less densely populated parts of the metropolitan area, we compared milkweed densities in two population density size classes. These results are in **Supplementary Section 5**.

#### Enhanced and Exemplary Sites in Chicago

Through targeted sampling, we were able to capture enhanced and exemplary milkweed densities for most land use classes. In **Figure 8**, the highest enhanced density average was found in the OS-C land uses class (365.6 stems/acre). The second highest enhanced density average was 195.4 stems/acre for Ind-S based on three highly variable site densities ranging from 2.0 to 321.7 stems/acre. For exemplary sites, the highest density average was found in OS-C at 4,109 stems/acre, followed by Res-S (330.4 stems/acre) and OS-NC (304 stems/acre). In the Res-S category, we sampled 50 sites, where the lowest density was 1.3 stems/acre and the top five densities (ten percent of samples) ranged from 225 to 419.5.

#### Comparing Baseline Densities Across Metropolitan Areas

Using metro-transects, we evaluated milkweed densities randomly in Chicago, Kansas City, and Austin (**Figure 9**). In general, average milkweed density values were low in both Res-S (1.4 in Chicago, 0.01 in Kansas City, and 0.6 in Austin) and Res-C (0.0 in Kansas City and 1.8 in Austin). However, in Chicago, there was a high density in Res-C (18.9) and in Minneapolis-St. Paul, the Res-S density was 30.0 stems per acre. This is over 20 times higher than the other metropolitan areas studied. While this density may be associated with the different sampling protocol used in Minneapolis-St. Paul that targeted neighborhood blocks inside the city limits, results from our social science survey also found that Minneapolis-St. Paul had higher milkweed stem counts. Also notable are elevated baseline densities in Ind-S (in Chicago and Kansas City), Rd (especially in Chicago), and Vac (in Chicago and Kansas City).

When we applied national density values for Open space land use classes as described above, areas such as nature preserves and city parks contributed the greatest number of milkweed stems compared with all other land use classes. We used national milkweed densities (for Open space in Chicago, Minneapolis-St. Paul, and Kansas City) and metro-transect milkweed densities (for all others) to run our baseline calculator for the four metropolitan areas. The resulting maps show areas of high to low densities (**Figure 10**). The corresponding total estimated stem counts by land use are shown in **Table 3**.

Results from our baseline geospatial tool estimate that Chicago has 15.3 million stems of existing milkweed. Approximately 66% of these stems are from OS-C areas, followed by 10% from OS-NC areas. Our findings in other cities were similar, in that OS-C dominated all other land use classes (36% in Minneapolis-St. Paul, 51% in Kansas City, and 81% in Austin). These were followed by Res-S in Minneapolis-St. Paul (35%), Agriculture in Kansas City (34%), and OS-NC in Austin (15%). In Minneapolis-St. Paul, we estimate a baseline of 11.7 million stems, in Kansas City we estimate 5.2 million, and in Austin we estimate 1.3 million stems. As noted above, the geographic extent of Austin only reflects the central county (Travis) or this estimate would be much higher. These data should also be interpreted with some caution since we draw from national densities for the Open space baseline density estimates in Chicago, Kansas City and Minneapolis-St. Paul.

#### Municipal Case Studies

When looking at our results from sampling at the municipalscale in Glenview and Schaumburg, it is helpful to compare these localized milkweed densities with our Chicago regional densities. We found that the density for some land use classes were higher, while others were similar or lower (see **Supplementary Section 3** for more details). Average density values for Glenview differed from the greater Chicago region especially for Community and cultural (CC, +9.1 stems/ acre), Major rights-of-way and landfill (ROW, +5.5 stems/acre), and Res-S (+0.7). In Schaumburg the higher density values were much less pronounced; ROW was slightly greater (+0.4), as was Corporate and medical (Corp, +0.3), and Res-S (+0.3).

## Social Science Research Results

The social science research covered a wide range of topics relevant to monarch conservation, including both open-ended and closed-ended questions, and it generated both qualitative and quantitative data. Of the most relevance here are the quantitative online survey results having to do with how many people already are growing milkweed, how much milkweed is on their property, and how much more they would be willing to add in the future. Of the 734 people surveyed in the four pilot metropolitan areas, 226 indicated that they "manage plantable open space on one or multiple sites" (most, but not all, were residential sites, i.e., home gardens). Of those 226 respondents, 184 or 81% answered that they had milkweed growing at their site. Of the 184 with milkweed, almost half had 1–10 milkweed plants at their site and a little over a third had 11–50 milkweed plants. Just 17% said that all or almost all of their site's plantable space was already taken up by native plants, i.e., many indicated that there was more plantable space at their site. When asked what they would be willing to plant in the next five years, 63% said they would be willing to convert more of the plantable space at their site to native milkweed. These additional data, while not representative of the general population, did function to provide a check on the baseline and scenario estimates the geospatial team generated.

#### Scenario-Modeling Results by Metro Area

After running the baseline tool, the total estimated number of milkweed stems for each metropolitan area ranged from 1.3 million in Austin to 15.3 million in Chicago (**Table 3**). With a two percent adoption rate across all sectors, at enhanced densities, Chicago would add an estimated 1.4 million stems to the 15.3

FIGURE 7 | Chicago baseline weighted mean milkweed densities (calculated on a grass/shrub and 20% tree basis) for the metro-transect sampling results. Standard error around the mean are indicated by lines. Metro-transect sampling was used for all land use classes except for open space which used density values provided by Thogmartin et al. (2017b).

TABLE 2 | Chicago milkweed density survey results by land use class calculated on a "plantable space" basis, which leverages high-resolution land cover.


*The weighted means provided the baseline density for all land use classes except OS-C and OS-NC. For an explanation of how the densities for OS-C and OS-NC were calculated, see Targeted Sampling of Natural Areas and Milkweed Density Calculations. The mean OS-C and OS-NC values below were used as enhanced densities.*

million current stems for a total of about 16.7 million stems (**Supplementary Table 4**). We also applied a five and ten percent adoption rate across all land use classes for Chicago, based on enhanced densities, which result in an estimated 3.6 million additional stems and 7.2 million additional stems, respectively. When a two percent adoption rate across all sectors is applied at exemplary densities, Chicago would add an estimated 13.4 million stems to the 15.3 million current stems for a total of 28.7 million stems (**Table 4**).

## Estimates for Baseline and Projected Stem Counts When Extrapolating Across the Eastern Range of the Monarch Butterfly

Bearing in mind the previously identified assumptions, when extrapolating across urbanized areas, we estimate that 312 million stems currently exist on the landscape in the North Central and Northeast monarch regions as delineated by the USGS (**Supplementary Figure 1**) utilized by Oberhauser et al. (2017,

Figure 1). We also estimate that an additional 29.8 million stems could be added in this area with modest effort, based on a two percent adoption rate using our Chicago enhanced milkweed densities. Our more ambitious estimate, based on exemplary densities, adds 271 million stems of milkweed in urbanized areas for this same geography at a two percent adoption rate. It is important to note that common milkweed (Asclepias syriaca), which comprises approximately 74% of milkweed occurrence along our metro-transect lines (**Supplementary Section 6**), has a species distribution that roughly matches the USGS North Central and Northeast monarch regions (see Lemoine, 2015). We therefore have more confidence in extrapolating across this region, since the species distribution and ecotype have a greater affinity to the Chicago region. However, for regional comparison purposes, if we extrapolate across the USGS North Central, Northeast, and South monarch regions, a total of 56.1

million stems may be added under an enhanced scenario, and 517 million stems may be added under an exemplary scenario. For the entire region east of the Rockies, the total contribution reaches 31%, or nearly one third of the national goal. For all of the above estimates, we use baseline densities from the literature for Open space conservation and non-conservation land use classes, and Chicago densities for all other land use classes.

calculated milkweed densities, shown in gray.

TABLE 3 | Baseline milkweed densities (stems/acre) used as input for the Urban Milkweed Baseline Tool for each metropolitan area and the resulting outputs including acres of plantable space and estimated count of current milkweed stems.


*Chicago baseline densities were also used for Minneapolis-St. Paul (for all but Res-S, where field data was collected) and for Kansas City (for OS-C and OS-NC) where data was insufficient for determining baseline densities.*

TABLE 4 | Estimated number of milkweed stems that would be added to the baseline density for each metropolitan area based on using the Scenario-Based Planning Tool with Chicago's exemplary milkweed density and a two percent adoption rate.


\**Value of 0.1 used to represent land use classes where enhanced density appeared to be 0, but milkweed populations are known to occur for that land use class.*

# DISCUSSION

This research is the first in-depth examination into understanding the role that metropolitan areas play in supporting monarch butterflies, and what their capacity may be for supporting breeding populations. Our three main goals of this study were to (1) estimate the current and potential contribution that metropolitan areas provide for monarch butterfly habitat, (2) provide geospatial planning tools to support the development of conservation strategies, and (3) estimate the potential for adding milkweed within urbanized areas across the monarch butterfly's eastern range. Our study offers several key findings and outcomes about metropolitan areas and monarch conservation.

One key finding is that a large presence of milkweed occurs across most land use classes in metropolitan areas, especially in Open space conservation, Open space non-conservation, Residential-single family, Residential-common space and multifamily, Vacant lots, and the Rights-of-way land use classes. In the metropolitan areas where we sampled both open space and residential areas, we found that these land use categories had the highest potential for increasing the number of estimated milkweed stems. Targeting people and organizations in these classes may result in some of the best opportunities for focused engagement strategies aimed at bolstering additional habitat. However, excellent large-scale opportunities may also exist across other land use classes particularly where cumulative land areas have few or a single owner such as rights-of-way and vacant lots owned by a city or municipality.

Another important outcome from our research was the development of geospatial tools that allow landowners and planners to evaluate the potential for adding pollinator habitat to large land areas and to measure the success of planting efforts by comparing baseline estimates over time. These tools are also useful for evaluating the co-benefits that exist when combining green infrastructure improvements with conservation planning for monarchs. Our tools are dependent on the accuracy of field sampling results. We found that sampling milkweed across vast metropolitan areas is difficult, but possible. Our metro-transect sampling method was successful in detecting milkweed across the diverse mix and patterns of land use classes, even at low densities. Sampling across open space natural areas was challenging but was successful in locating milkweed stems through targeted sampling. These sampling methods were effective for understanding more about the density of milkweed where it occurs across the landscape, but further study is needed for establishing more precise metropolitan-scale baseline density estimates particularly for Open space conservation and nonconservation land use classes in urban areas (see Methodological Considerations below).

Our geospatial extrapolations indicate that if all urbanized areas in northern and northeastern US were engaged, it may be possible to add an estimated 271 million stems of milkweed, or over 15% of the projected milkweed stems needed to rebound the eastern monarch population. This projection increases to 31% if we extend this extrapolation across all urbanized areas in the entire US eastern range. These findings suggest that urbanized areas should figure prominently in monarch conservation planning (see Implications for Monarch Conservation below).

# Implications for Monarch Conservation

Although agricultural land had the largest amount of plantable space in all four metropolitan areas, the majority of this land is not available as plantable space for milkweed stems. Agricultural land is primarily found in the outer fringes of the metropolitan areas studied. Most of this land is composed of intensively farmed row crops (cereals and soybeans primarily). The loss of milkweeds stems from this habitat resulting from GMO crops followed by extensive herbicide use is a big part of the monarchs' plight (Brower et al., 2012; Pleasants et al., 2017). Community gardens and other small-scale agriculture in the urban and suburban core are treated as part of the land use category in which they are imbedded, rather than in the agricultural land use class. After Agriculture, Residential-single family land use made up the second largest amount of plantable space in three of the four metropolitan areas.

While existing milkweed densities were typically low (<2 stems/acre) in Residential-single family land use, because this land use category is so massive in total size, low densities can still have a big contribution. Our results likely underestimate the density of milkweed in this land use, because we were unable to survey in backyards. Results also show that densities within city limits can be 20–30 fold higher, as was observed in Minneapolis-St. Paul. Residential-single family also had the third highest enhanced density and the second highest exemplary density (**Figure 8**). Further, results from our online surveys of the interested public suggest over 60% of people surveyed in residential areas would be willing to plant more milkweed in the next 5 years. With the large amount of plantable space, pronounced interest of residents, and high measured densities at enhanced and exemplary sites, Residential-single family areas have the potential to add a considerable number of stems even at a two or five percent adoption rate.

In achieving a two or five percent adoption rate, the highly fractured ownership of residential land poses both a challenge and an opportunity. While the acreage-to-landowner ratio is generally low, the high level of enthusiasm and capacity among monarch-friendly gardeners (Derby Lewis et al., 2018) makes the task of converting additional residential land to monarch habitat less daunting and demonstrates real conversion potential in one of the largest land use classes in metropolitan and urbanized areas. While working with a greater number of landowners, each with a relatively small amount of land, may intuitively seem like an inefficient approach, our social science research shows this is not necessarily the case. The higher population density in cities and their surroundings tends to bring people in contact with one another, encouraging the adoption and spread of new practices. One's neighbors (and their gardens) are often more visible, and organizations can achieve a critical mass of involvement to move conservation initiatives forward.

Within the public sector, agencies own or manage a high proportion of land in rights-of-way, open space, and vacant lots and have the potential to improve and convert large areas through management and policy changes (Anderson and Minor, 2017). We estimate, for example, that right-of-way areas, which occupy 128,000 acres of plantable space in the Chicago region, could add approximately 577,000 stems by converting 20% of the green space to habitat at exemplary levels in the Chicago region alone. Similarly, converting 20% of vacant lots, which occupy over 67,000 acres of plantable space in the Chicago region, could add approximately 678,000 stems. By default, our projections use a two percent adoption rate. However, a higher adoption rate may be reasonable for vacant lots, rights-of-way, and open space natural areas, since existing green space on these lands can more often be augmented with habitat without changing how the land is currently utilized. In addition, public land owners can couple management practices and funding opportunities, such as stormwater and green infrastructure improvement programs and grants, with pollinator improvements to further enhance milkweed populations.

OS-C and OS-NC are another example of publicly owned land with a relatively low landowner-to-acre ratio. Stem densities for OS-C and OS-NC in metropolitan areas where our teams sampled (Chicago and Austin) were much higher when compared with other land use classes. Much of this is due to our use of targeted sampling, however our results suggest that milkweed can be present at very high densities in these land use classes. These high densities coupled with the large amount of plantable space means these areas can be important opportunities for adding stems even with relatively low adoption rates.

While OS-C and OS-NC provided some of our highest sampled milkweed densities, some land use classes in metropolitan areas appear to have low milkweed densities and lower potential for adoption, based on current conditions. Of our 16 consolidated land use classes, several had no milkweed occurrence at all. Some of these classes were scarce in our study areas (e.g., Transitional and restricted use, Industrial-large, and land classified as Water). Commercial land, on the other hand, was well represented in our sampling, but had zero random milkweed occurrences. In fact, we actively tried to locate commercial sites with milkweed during our targeted sampling of enhanced sites and were only able to find one site in Chicago. Future research could assess the potential for increasing commercial adoption rates with different engagement practices.

#### Methodological Considerations

Our findings, comparisons, and extrapolations rely more heavily on Chicago data, as it was more complete and consistent. Our local knowledge of Chicago and that of our partner organizations allowed us to conduct "targeted" sampling of known milkweed locations. This was an important part of the effort to determine the typical milkweed densities where milkweed occurs, which was key to estimating the potential capacity to add habitat in metropolitan areas.

Fieldwork across the four metropolitan areas highlights the difficulty of sampling the urban environment. Urban areas are highly heterogeneous landscapes, and there is a complex matrix of landowners and policies. Nearly all of our land use field data, other than the Open space categories, comes from randomized sampling of what is visible from the public right-of-way (i.e., roads and sidewalks) except when our team was explicitly invited onto private land (e.g., targeted sampling on enhanced sites). In contrast, for OS-C and OS-NC, our field teams obtained permits and conferred with land managers to conduct more intensive sampling of known milkweed sites. This was necessary because little was known about milkweed densities in urban natural areas. Both OS-C and OS-NC are comprised of many plant communities, including large areas devoid of milkweed. However, we knew these land use classes would be important in determining densities where milkweed is found (enhanced sites) and their upper threshold (exemplary sites). Using targeted sampling ensured we would collect data on these important land use categories.

For Open space baseline densities, we used values taken from the supplement of Thogmartin et al. (2017b) as it is the only source estimating milkweed densities across a wide geographic extent and across a full range of land use types. Because our land use classes and the land cover classes used in their supplement are not the same, we chose classes that were the best representation of the management approaches within OS-C and OS-NC. For OS-C, we selected CRP-NW, with a value of 112.14 stems/acre, as the best match. Although CRP land is based on a very different program than are lands like state parks, national wildlife refuges, forest preserves and other protected lands with a primary conservation goal, they share a management goal of biodiversity conservation. Since this value was provided on a land use basis, we applied our estimated proportion of the amount of OS-C land that is non-cultural plantable space (72.4%) to this amount, resulting in a baseline of 81.19 stems/acre. We also applied the CRP-NW value of 112.14 stems/acre to the proportion of OS-NC land that is non-cultural plantable space (20.2%), resulting in a baseline of 22.65 stems/acre. This category may not appear to be the best match and possibly overestimates the baseline density for OS-NC; an alternative option would be to use the Protected Grasslands category, however, in applying this value (3.09) to the proportion of natural areas in OS-NC (20.2%), the result is a density of 0.62 stems per acre, lower than nearly all measured land use densities in Chicago, which we believe would greatly under-represent the presence of milkweed in this land use class.

We expect that the actual milkweed density present across all Open space conservation and Open space non-conservation land is somewhere between the estimate provided by Thogmartin et al. (2017b) and the densities observed in the targeted sampling areas. More research is needed to better approximate a baseline density of milkweed stems in this land use category. In addition, another of our biggest challenges was determining how much of the plantable space within land in OS-C and OS-NC is actually plantable space as opposed to land set aside for recreation or other uses. As a future project we would like to better delineate those lands.

Our methods may need to be adapted for other metropolitan areas, towns, and municipalities that wish to utilize our geospatial tools, or to replicate this study. In particular, we leverage highresolution land cover data to estimate stem densities on a plantable space basis. Using estimations on a total land use basis is an acceptable alternative when these data are not available. We also recognize that when extrapolating from our findings, the milkweed species and growing conditions present in Chicago differ from other parts of the country. For example, common milkweed (Asclepias syriaca) is found throughout the north and northeast but is not common in the south.

Due to qualitative differences in the data available and in the methodological approach, inferences are challenging; however, some clear distinctions are apparent. Minneapolis-St. Paul, for example, achieved a much higher residential milkweed density as compared to its peers, and in Schaumburg, where they are actively engaging residents through community planting programs, the milkweed density was moderately higher as compared to the citywide density. In Glenview where they have a 10-year history of pollinator friendly institutional projects, there was a large boost over the regional density. These are encouraging results that may be indicators of effective social and institutional engagement; however, we also caution that it is difficult to tease apart cultural and environmental factors that may be at play.

Lastly, it's important to note that our study only looked at the current and potential contributions of milkweed stems in metropolitan areas, and not at actual monarch productivity. While our approach in Chicago reveals both a higher baseline where milkweed was found and a higher potential for milkweed stems in metropolitan areas than previously reported, it is not known whether larval survivorship is similar to what has been observed in other sectors such as rights-of-way or agriculture. With such high public awareness and interest in creating pollinator habitat in metropolitan areas, it is important to have a better understanding of what influences monarch productivity (e.g., patch size, milkweed density, floral diversity, distance to green space, etc.) (Nail et al., 2015) in this landscape in order to inform best planting practices for a variety of stakeholder groups who we see clamoring for this information.

#### Recommendations and Next Steps

Although urban areas cover only three percent of land in the United States, they are home to 80% of the country's population18. We show that there is more milkweed on the ground in metropolitan areas than previously published by Thogmartin et al. (2017b) (0.1–1.0 milkweed stems/acre in developed areas). If there is a higher baseline density of milkweed in urban areas than was previously thought, this could mean that the existing goal of 1.8 billion stems is not adequate to boost overwintering populations to sustainable levels. However, much of our data is drawn from the Midwest and all of the cities in our study had populations in excess of one million people. More study of small to mid-size cities and those outside of the Midwest is needed before conclusions can be drawn. Our findings clearly indicate that the urban sector can make important contributions to monarch recovery and that the diverse landscape of urban areas requires careful attention to both ecological and social differences across land use classes and the engagement strategies employed for getting additional habitat on the ground. This includes establishing planning objectives that prioritize appropriate engagement strategies for key decision makers to harness social momentum for milkweed adoption. Some of the biggest potential occurs in Residential land use classes, which require successful engagement of residents (Derby Lewis et al., 2018). Researchers from the US Geological Survey and collaborating institutions have called for an "all hands on deck" approach across all sectors, including urban and suburban areas (Thogmartin et al., 2017b). We suggest a targeting and engagement approach in metropolitan and urbanized areas to complement the "all hands on deck" strategy for fulfilling the goals of increasing planted milkweed by 1.8 billion stems to support monarch butterflies (Thogmartin et al., 2017a).

Our Urban Monarch Conservation Planning Tools are designed to help municipal decision-makers and planners estimate their capacity to add stems across the metropolitan landscape by identifying where the biggest opportunities exist and what the best practices are to engage different stakeholder groups. Our tools can be applied to a land use class, for example, estimating how many stems could be added if five percent of homeowners adopted our enhanced site numbers; and they can be used spatially to estimate the potential of particular neighborhoods, municipalities, or parts of a city. One active area of research has been applying our tools to estimate the potential addition of milkweed stems within areas set aside for stormwater management, thus capturing the co-benefits of increased habitat along with water infiltration.

While our research has largely been successful at accomplishing our objectives, we acknowledge that further study is needed. Our research is based on metropolitan areas ranging in size from 1.3 million people in Austin to 9.7 million in Chicago. We recommend further study into small and mediumsized metropolitan areas across Middle America to understand differences and similarities to these findings, and to test the replicability of our methods. Smaller cities and towns may show different social and ecological trends, which could impact both regional assumptions and social engagement strategies. Studies into how social networks affect the transfer and spread of information resulting in on-the-ground habitat are also critical to driving future restoration efforts at larger scales. Also, due to the urgency needed to support monarch butterflies with on-the-ground resources, we are aware that we are addressing an ecological problem with an "engineering solution" focused on getting milkweed stems into the ground vs. what would be a more nuanced approach of producing healthy diversified ecosystems that can support pollinator networks capable of resisting disturbances such as localized effects of climate change. We recommend research into several topics that would increase our understanding of how monarchs and other pollinators perform under different environmental conditions such as patch

<sup>18</sup>United States Census Bureau 2010 https://www2.census.gov/library/ publications/decennial/2010/cph-2/cph-2-1.pdf

size, habitat diversity needs and efficacy at producing monarchs, connectivity to other patches and resources, and the effect all these factors may have on monarch butterfly fecundity and predation (especially at the egg and larval stages).

#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of The Field Museum Institutional Review Board Policy and Procedures. The IRB granted the team a waiver of written informed consent based on the fact that participation in the study presented very minimal risk of harm to subjects. Subjects were informed of study confidentiality procedures and told how their answers would be used before they took part in an interview or survey. The protocol was approved by The Field Museum Institutional Review Board.

#### AUTHOR CONTRIBUTIONS

MJ: primary contributing author, geospatial team manager, primary on research and tool design. EH: contributed content, editing, research advice, GIS analysis and field coordination. KK: contributed GIS analysis, statistics, content, and editing. ML: GIS analysis and research advice. AD: contributed content, editing, research advice and project management. DS: contributed research advice, editing, and content. AW: contributed social research design, analysis, and team coordination, content, and graphic design. MB: contributed content, editing, and research advice. IR: contributed editing and research advice.

### REFERENCES


#### FUNDING

This research was funded by a US FWS Cooperative Agreement No. F11AC01062 awarded for the period September 1, 2015 to May 31, 2017; and US FWS Cooperative Agreement No. F17AC01018 awarded for the period October 1, 2017 to September 30, 2018.

#### ACKNOWLEDGMENTS

Funding for this project came from the US Fish and Wildlife Service. We would like to thank the deeply committed and creative team of individuals whose work is reflected in this study, including Tim Bodeen, Katie Boyer, Brittany Buckles, Wendy Caldwell, Louise Clemency, Caitlin Dix, Martha Dooley, Ryan Drum, Jill Erickson, Adriana Fernandez, Robyn Flakne, Nigel Golden, Jessica Hellman, Kyle Kasten, Susan Lenz, Laura Lukens, Cora Lund-Preston, Patrick Martin, Katie Maxwell, Tom Melius, Kelley Myers, Rosie Nguyen, Karen Oberhauser, Zach Paolillo, John Rogner, Jason Rowader, Glen Salmon, Kristin Shaw, Wayne Thogmartin, Michelle Thompson, Chuck Traxler, Kristin Voorhies, Gwen White, and Barbara Willy. We also thank our reviewers and Amy Rosenthal, Ellen Woodward, Katherine Moore Powell, and Nigel Pitman for their review and comments on this paper.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00210/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 Johnston, Hasle, Klinger, Lambruschi, Derby Lewis, Stotz, Winter, Bouman and Redlinski. 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.

# Western Monarch Population Plummets: Status, Probable Causes, and Recommended Conservation Actions

Emma M. Pelton<sup>1</sup> , Cheryl B. Schultz <sup>2</sup> , Sarina J. Jepsen<sup>1</sup> , Scott Hoffman Black <sup>1</sup> and Elizabeth E. Crone<sup>3</sup> \*

*<sup>1</sup> The Xerces Society for Invertebrate Conservation, Portland, OR, United States, <sup>2</sup> Department of Biological Sciences, Washington State University, Vancouver, WA, United States, <sup>3</sup> Department of Biology, Tufts University, Medford, MA, United States*

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey, United States*

#### Reviewed by:

*Tyler Flockhart, University of Maryland Center for Environmental Science (UMCES), United States Ralph Grundel, U.S. Geological Survey, Great Lakes Science Center, Chesterton, IN, United States*

> \*Correspondence: *Elizabeth E. Crone elizabeth.crone@tufts.edu*

#### Specialty section:

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

> Received: *05 March 2019* Accepted: *18 June 2019* Published: *03 July 2019*

#### Citation:

*Pelton EM, Schultz CB, Jepsen SJ, Black SH and Crone EE (2019) Western Monarch Population Plummets: Status, Probable Causes, and Recommended Conservation Actions. Front. Ecol. Evol. 7:258. doi: 10.3389/fevo.2019.00258* Western monarch butterflies dropped by ∼97% of their average historic abundance between the 1980s and mid-2010s. In winter 2018–2019, the population plummeted even farther, to fewer than 30,000 monarchs, which represents a single year drop of 86% and a drop of >99% since the 1980s. The population may now be hovering at its quasi-extinction threshold. In this Perspectives piece, we: (1) Place the current status in context, (2) Highlight the most likely window during the annual life cycle when the population declined, (3) Review probable causes of long-term declines, and (4) Recommend steps that the public, policy makers, and land managers can take to recover western monarchs. The available studies reinforce the hypotheses that overwintering habitat loss and loss of central California breeding habitat, as well as pesticide use, are likely important contributors to the western monarch's long-term decline. The most limiting part of the migratory cycle appears to be concentrated during the overwintering stage and/or in early spring. If western monarchs are in fact entering an extinction vortex, they need extraordinary efforts—focused on the most vulnerable periods of the annual cycle— to save the migration. Critical short-term conservation priorities are to (1) Protect, manage and restore overwintering habitat, (2) Protect monarchs and their habitat from pesticides, (3) Restore breeding and migratory habitat in California, (4) Protect, manage, and restore summer breeding and fall migration monarch habitat throughout the western monarch's range, and (5) Fill research gaps to inform western monarch recovery strategies.

Keywords: Danaus plexippus plexippus, western monarchs, quasi-extinction, conservation, population trends

# INTRODUCTION

Monarch butterflies (Danaus plexippus plexippus) across North America have been undergoing a multi-decade decline (Semmens et al., 2016; Schultz et al., 2017). Nonetheless, the crash of the western population (**Figure 1**) in winter 2018–2019 was particularly stunning. In 2017, we estimated that the overwintering population had dropped by 97% of its average historic abundance, from ∼3 to 10 million to ∼200–300 thousand butterflies (Schultz et al., 2017). In winter 2018-2019, the population plummeted to fewer than 30,000 monarchs,

which represents a single year drop of 86%, and a >99% drop since the 1980s (**Figure 2A**).

In this Perspective, we: (1) Place the current status in context, both how trends compare to the eastern population and potential implications of dropping to unprecedentedly low abundance in the West, (2) Highlight the most likely window during the annual life cycle when the population declined, (3) Review probable causes of long-term declines, and (4) Use our understanding of drivers of declines to recommend steps that the public, policy makers, and land managers can take including identifying knowledge gaps for which focused mechanistic studies could contribute to developing more effective and efficient conservation actions.

#### STATUS OF WESTERN MONARCHS IN WINTER 2018–2019

Since 1997, volunteers have estimated the overwintering population in California each fall at coastal groves (Xerces Society Western Monarch Thanksgiving Count, 2019). The 2018 Xerces Thanksgiving Count revealed a new low—only 28,429 monarchs were tallied—<1% of the historic population (**Figure 2A**). The current trend in western monarchs is in contrast to eastern monarchs, which hit the highest estimated population size in the last decade in winter 2018–2019 with 6.05 hectares occupied (Rendón-Salinas et al., 2019).

We know from our past analyses that a western population of <30,000 butterflies is unprecedented. The 2018 Thanksgiving count mirrors a textbook extinction vortex (Gilpin and Soule, 1986), in the sense that fluctuations in abundance—which have been happening throughout the past 30 years—become riskier as the population becomes smaller. As populations become smaller, "ordinary" environmental variation can cause a population to drop below a point from which extinction is inevitable, unless extraordinary measures are taken. We call this point the quasiextinction threshold. In 2016, a group of experts proposed 30,000 butterflies as the quasi-extinction threshold for western monarchs (Schultz et al., 2017). Now, it is suddenly imperative to know if the experts were correct, and, if so, what extraordinary measures need to be taken to preserve the population.

In general, we know very little about what happens when formerly large populations become small. Individuals in small populations may have reduced mating success, suffer increased predation, and lose other benefits of schooling or flocking (Courchamp et al., 1999). These effects due to small population size are known as "Allee effects" and are difficult to estimate in wild populations because they are only expressed after a population has begun to decline to extinction (Liermann and Hilborn, 1997). Therefore, setting quasi-extinction thresholds is one of the most subjective steps of population viability analysis (e.g., Frick et al., 2010; McGowan et al., 2017). If the published quasi-extinction threshold is correct, then positive density-dependent processes associated with Allee effects could lead to further rapid decline. If the quasi-extinction threshold is incorrect, we will see the western monarch recover to a larger population size. Regardless, this serves as a call to intensify efforts to boost abundance to healthy enough numbers in the wild for the population to be able to sustain itself through normal ups and downs in the population size.

# ENVIRONMENTAL DRIVERS

### Causes of Rapid Decline From 2017 to 2018

Given the large drop in western monarchs from 2017 to 2018, some are tempted to blame the weather for the low numbers. Late rainy season storms swept across California in March. There was a severe and extended wildfire season in the West and smoke was widespread at times. California is still recovering from a historic drought. Large amplitude inter-annual fluctuations are an intrinsic aspect of butterfly population dynamics, and causes of year-to-year variation are not necessarily the same as the causes of long-term declines. Nonetheless, it is important to try to understand western monarch abundance throughout the year from winter 2017–2018 through winter 2018–2019, when the decline occurred.

Starting in winter 2016–2017, the Xerces Society and volunteers began a second count at overwintering sites, the New

monarch butterfly 1981–2018 estimates for overwintering abundance during the Thanksgiving Count time period in coastal California. Estimates for 1981–2017 were calculated with state space models (Schultz et al., 2017), scaled to be comparable to the raw count from 2018 (shown). (B) Monarch egg and larva counts per stem at all 12 monitoring sites (shown in Figure 1) throughout the season in 2017 and 2018. Curves were fitted with generalized additive models (Wood, 2011) to show general trends in abundance. The fact that the two curves are parallel suggests that densities were lower by the time monarchs arrived in 2018; the decline does not appear to be due to different dynamics during breeding. Note the log scale and 10-fold difference among years.

Year's count (centered around New Year's Day, to complement the Thanksgiving Count 6 weeks earlier). Monarch abundance at the New Year's Count had declined by 43% on average in 2017 (n = 44 sites), 49% on average in 2018, (n = 115 sites) and 36% in 2019 (n = 130 sites), when compared to monarch abundance at those same sites during the Thanksgiving Count. These data suggest that monarch butterflies did not have exceptionally low survival between November 2017 and January 2018, compared to the previous year.

In addition to counts at overwintering sites, we started monitoring summer breeding of western monarchs in 2017 at 12 sites throughout the West (**Figure 1**). Across these 2 years, the density of monarch eggs and larvae was consistently lower in 2018 than 2017 (**Figure 2B**), with about a 10-fold decline between the 2 years (average immature monarchs/stem = 0.0273 [95% CI = 0.0025, 0.2953] in 2017 and 0.0022 [95% CI = 0.0001, 0.0429] in 2018; paired t-test of site averages between years: t = −2.53, df = 10, P = 0.030). We therefore suggest that the drop measured at Thanksgiving 2018 originated before the beginning of the 2018 breeding season, either late during the overwintering season or very early in the breeding season.

This inference is consistent with Espeset et al. (2016) who concluded that western monarch declines were concentrated in early spring. Of the environmental events that seemed "unusual" in 2017–2018, this pattern points to the possible negative effects of unusually heavy rains in March 2018 with the caveat that many other factors may have caused the population drop, including the interaction of weather with habitat quality at overwintering sites, and habitat inland from the coast in California, where the first generation breeds.

#### Causes of Long-Term Declines

In the larger eastern population, declines have largely been attributed to overwintering habitat loss (Brower et al., 2012; Vidal et al., 2013) and breeding habitat loss, especially through the use of herbicides (e.g., Pleasants and Oberhauser, 2012; Flockhart et al., 2014). We (Crone et al., in press) recently evaluated climate and land use factors simultaneously as potential drivers of western monarch abundance. Trends in abundance were more strongly associated with land use variables including coastal development in overwintering areas and pesticide use (glyphosate and neonicotinoid insecticides) in breeding areas than climate variables in both overwintering and breeding areas (Crone et al., in press). These results are consistent with the hypotheses that overwintering habitat loss and loss of central California breeding habitat are important for western monarchs (see Espeset et al., 2016) and that trends in pesticide use likely contribute to declining monarch populations as well as declines in other butterfly taxa (see also Forister et al., 2016).

In addition to this broad scale analysis, we estimated daily survival using data from Tuskes and Brower (1978), for comparison with population declines estimated from Thanksgiving and New Year's counts. Daily survival at Natural Bridges near Santa Cruz was 0.995 (95% CI 0.988, 0.997) and at Santa Barbara was 0.991 (0.989, 0.993). Over 6 weeks (the approximate time between Thanksgiving to New Year's counts), this historical estimate translates into a 29% drop (95% CI 12–40%) using estimates from Santa Cruz and a 32% drop (95% CI 26–37%) using estimates from Santa Barbara. Hence, based on the best available evidence, apparent survival during winter in recent years (36–49% drop) has been lower than it was in the past. This change reinforces the importance of overwintering habitat quality on the long-term decline of the western monarch population. At the present time, we have not found comparable data to evaluate whether breeding season survival or reproduction have changed in western monarchs.

#### URGENT STEPS FOR CONSERVATION

To date, western monarchs have received far less conservation attention and financial resources than the larger eastern population. Nonetheless, the western monarch breeds across most of the US west of the Rocky Mountains, a significant portion of the monarch's overall North American range. It makes an important contribution to the resilience, redundancy, and representation of the species as a whole (see definition in Shaffer and Stein, 2000).

While the precise causes of the recent dramatic drop in the western population, as well as the longer term decline, remain unknown, this knowledge gap should not prevent conservation action. We suggest that a precautionary approach be taken to remediate potential causes of decline. Specifically we recommend efforts (1) to protect, enhance, and actively manage overwintering sites; (2) to protect monarch habitat from pesticides, particularly systemic insecticides (including neonicotinoids); (3) to supplement larval and adult resourcesespecially in the early spring-in California; (4) to identify, protect, and enhance monarch habitats throughout the West, and (5) to prioritize research efforts to answer questions critical to developing an effective and efficient recovery strategy. Here, we briefly explain our recommendations, and their relationship to the causes of western monarch declines, described above. These recommendations and relevant resources are expanded in in our "Western Monarch Call to Action."<sup>1</sup>

### Protect, Manage, and Restore Overwintering Habitat

Our analyses ("Environmental drivers" above) point to the importance of monarch habitat in winter and early spring, prior to the breeding season. Conservation biologists have long known that efforts focused only on one stage of a species' life cycle (e.g., breeding) may not be sufficient if populations are limited by another life stage [e.g., overwintering (Brown et al., 2017)]. Despite the importance of monarchs to Californians and the state's tourism economy, few overwintering sites are meaningfully protected (International Environmental Law Project and the Xerces Society, 2012) and sites continue to be destroyed—indeed, from 2017 to present, over one dozen sites have either been newly destroyed or are reported to be threatened by inappropriate tree trimming, removal, and/or development (Xerces Society Western Monarch Overwintering Sites Database 2019, unpublished). To protect remaining habitat, overwintering sites could be designated as Environmentally Sensitive Habitat Areas (ESHAs) by the California Coastal Commission, protected as Critical Habitat if monarchs were listed under the federal Endangered Species Act, protected by California Department of Fish and Wildlife if monarchs were listed as endangered under the California Endangered Species Act, or a new law could be created by the California state legislature to protect overwintering sites from destruction.

To address the need for active management of overwintering sites, the majority of which occur on publicly owned land, a greater financial investment is needed. The Monarch Butterfly and Pollinator Rescue Program (California Assembly Bill 2421), was signed into law in 2018, and \$3 million was allocated to this program. An additional \$3.9 million was recently allocated for restoration of overwintering sites owned by the City of Goleta. While these represent important steps forward, more resources are needed to restore and manage the over 200 actively used overwintering sites. While there are no published estimates, restoring a significant number of overwintering sites would easily require tens of millions of dollars and, more importantly, would benefit from sustained funding to continue to manage the groves for monarchs in the long-term. Of the Top 50 priority sites identified by Pelton et al. (2016) many of the most important sites are owned by the California Department of Parks and Recreation, followed by cities, U. S. Department of Defense, East Bay Regional Parks District, and county, university, and other state and federal agencies as well as private entities. Some of these owners do not encourage or permit the planting of eucalyptus (the dominant tree used by monarchs in California during overwintering), nor are these land managers necessarily focused on managing for monarch overwintering habitat—and, in some cases, may be unaware of the full extent of overwintering habitat within their jurisdiction.

### Protect Monarchs and Their Habitat From Pesticides

In our analyses of long-term trends, insecticide and herbicide use were almost as tightly associated with monarch declines as overwinter habitat loss. Restricting insecticide and herbicide use increases adult Lepidoptera abundance (Frampton and Dorne, 2007). Broadcast herbicide use can kill host and nectar plants and have non-target effects on butterflies (Stark et al., 2012). We advise protecting the most important monarch breeding and overwintering habitats from insecticide and herbicide use. Specifically, we recommend avoiding herbicide applications that damage monarch breeding and migratory habitat such as milkweed and wildflowers. These recommendations apply to home gardens and lawns, as well as lands used for agriculture and other purposes. If herbicides are used, we advise using targeted application methods, avoiding large-scale broadcast applications of herbicides, and taking precautions to limit off-site movement of herbicides. Neonicotinoid insecticides, in particular, should be avoided at all times in monarch habitat due to their persistence, systemic nature, and toxicity. When purchasing milkweeds or wildflowers from nurseries, we recommend ensuring that they have not been treated with neonicotinoids or other systemic insecticides.

# Restore Breeding and Migratory Habitat in California

Enhancing monarch breeding habitat may be able to partly mitigate reductions in overwintering habitat quality because larger populations at the end of the summer can potentially withstand higher mortality. Numerous studies have quantified the importance of host and nectar plants for butterfly populations (Dennis et al., 2006; Dennis, 2010), and restoration efforts which enhance host and nectar have been effective approaches for the conservation of rare butterflies (Carleton and Schultz, 2013). We recommend planting native milkweeds in areas where they historically grew in California, and, in particular,

<sup>1</sup>www.savewesternmonarchs.org

in the Coast Range, Central Valley, and the foothills of the Sierra Nevada, areas where the first generation of monarchs are produced each spring. Early emerging native species that may be particularly important in spring include woollypod (Asclepias eriocarpa), California (A. californica), and heartleaf milkweed (A. cordifolia). However, commercial availability of these species is limited. Later-emerging native California milkweed species that are more readily available, and may also help, include narrowleaf (A. fascicularis) and showy milkweed (A. speciosa). In the desert southwest of California, we recommend rush (A. subulata) and desert milkweed (A. erosa). We recommend only planting milkweed >5 miles inland from overwintering sites, as milkweed does not naturally grow close to the coast north of Santa Barbara and milkweed at overwintering sites can interrupt natural overwintering behavior. Tropical milkweed (A. curassavica) is exotic to California, disrupts the monarch's migratory cycle, and serves as a reservoir for monarch pathogens (Satterfield et al., 2016). As such we recommend against planting tropical milkweed. In places where tropical milkweed already exists, we recommend cutting it back to the ground in the fall (October/November) and repeatedly throughout the winter to mimic native milkweed phenology and break the disease cycle; ideally, it should be replaced by native milkweed.

In addition, we recommend planting nectar-rich wildflowers, especially those that bloom early in the spring (February–April) and fall (September-October). If located close to the coast, plants which bloom in the winter (November-January) may also be useful.

## Protect, Manage, and Restore Summer Breeding and Fall Migration Monarch Habitat Throughout the Western Monarch's Range

Identifying key areas of breeding and migrating habitat for monarchs in the West remains a knowledge gap. Some geographic regions contribute disproportionately to the eastern monarch overwintering population in Mexico (e.g., Flockhart et al., 2017), and it is important to know whether the same is true for western monarchs. No data exist from which we could meaningfully evaluate their importance for short- or long-term population declines. Thus, while some of the most important monarch habitat within its western breeding (Yang et al., 2016; Dilts et al., 2019) and overwintering (Pelton et al., 2016) range has already been identified, additional work is needed to identify and rank these areas. We recommend identifying existing monarch habitat, ensuring that it is managed to protect monarchs (Xerces Society, 2018) and in some regions and landscape types, we recommend habitat enhancement or restoration. Habitat restoration in regions where monarch habitat historically occurred, but have likely been lost (such as the Columbia Plateau and Snake River Plain), as well as riparian areas, are high priority areas outside of California. Such restoration would likely benefit from habitat elements beyond milkweed and nectar, such as shrubs or trees for roosting and shade.

# Fill Research Gaps to Inform Western Monarch Recovery Strategies

Breeding and migrating habitat are only a few of the gaps in our knowledge of western monarchs. We especially need observations of monarch biology in places where human populations are low (e.g., the Great Basin desert) and at times of year when monarch butterflies are sparse (e.g., early spring in western California, just as they leave the overwintering grounds). We urge volunteers across the West to collect observations of monarchs and milkweeds, especially in the early spring (February–April), the period in which monarchs typically leave the overwintering sites. Together these observations will help answer questions about monarch breeding phenology. In this year, when numbers are low in the West and high in the East, targeted observations of monarch adults and larvae may also tell us whether the West sees an influx of monarchs arriving from Mexico (see Pyle, 2015). Monarch adult, larva, egg, nectaring, and milkweed sightings can be reported to the Western Monarch Milkweed Mapper<sup>2</sup> and first adults observed can be reported to Journey North<sup>3</sup> as well. More robust monitoring may be achieved through increased western participation in the Integrated Monarch Monitoring Program<sup>4</sup> .

We urge academic ecologists to conduct targeted experimental and observational studies to complement large-scale observations like the ones described above. In both the eastern and western monarch populations, filling knowledge gaps about demography throughout the life cycle would allow us to design quantitative thresholds for conservation and restoration. For example, it may be possible for targeted actions at one point in the life cycle to make up for stresses at other points. If climate change is making the landscape less favorable, can we make up for this with improved breeding or overwintering habitat quality and/or area? Can more breeding habitat in the outer parts of the breeding distribution make up for habitat loss at breeding or overwintering sites in California? Intuitively, the answer is probably "yes, but only partly." To answer this in a more quantitative way, we need a better understanding of how the life cycle pieces fit together.

# CONCLUSION

In closing, western monarchs are currently in peril. Their status reflects a long-term decline due to some combination of habitat loss and degradation in their overwintering and breeding range, increased pesticide use, and possibly climate change. The recent dramatic drop reflects conditions when the least is known about western monarchs—where they are, what habitat they are using, and what they need to survive, migrate and reproduce. In spite of their current status, monarchs are resilient; we believe that rapid conservation actions can recover the population. This recovery will require the protection of monarchs and their habitat, as well as targeted research to understand the unique life cycle of western

<sup>2</sup>www.monarchmilkweedmapper.org

<sup>3</sup>https://journeynorth.org/monarchs

<sup>4</sup>https://monarchjointventure.org/immp

monarch butterflies. If we are going to take these actions, the time is now.

#### DATA AVAILABILITY

The datasets for this study will not be made publicly available because restrictions apply to some of the datasets. Some of the datasets are in a publicly accessible repository:

The Xerces Society Western Monarch Thanksgiving and New Year's Counts analyzed in this study can be found at www. westernmonarchcount.org/data.

Restrictions apply to some of the datasets:

The Xerces Society Western Monarch Overwintering Sites Database 2019 is not publicly available because of privacy concerns with a subset of the information. Requests to access the database should be directed to Emma Pelton, monarchs@xerces.org.

The western monarch and milkweed phenology dataset summarized in this manuscript are not publicly available because it is part of a study currently in-progress. Requests to access the datasets should be directed to Cheryl Schultz, schultzc@wsu.edu.

#### AUTHOR CONTRIBUTIONS

EP, SJ, and SB (along with others—see Acknowledgments) oversee Thanksgiving and New Year's Counts and maintain the overwintering sites database. All authors contributed to funding

#### REFERENCES


and implementing the 2017–2018 surveys in the breeding range. EC conceived and ran all analyses with input from CS and EP. All authors wrote and revised the manuscript.

## FUNDING

Funds for the 2017–2018 breeding and phenology surveys and analysis were provided by Department of Defense Legacy Natural Resources Program (NR 16 Western Monarch) and the U.S. Fish & Wildlife Service Coastal Program. Authors were supported by their institutions (WSU, Tufts and Xerces Society) and EC, SJ, EP, and CS were partly supported by the National Science Foundation (NSF DEB 1920834).

#### ACKNOWLEDGMENTS

Thank you to the Western Monarch Thanksgiving Count volunteers, particularly our regional coordinators, Mia Monroe, and Katie Hietala-Henschell of the Xerces Society; Stephanie McKnight of the Xerces Society and Cameron Thomas of Washington State University for conducting the fieldwork for the breeding and milkweed phenology project; fellow western monarch researchers and conservation practitioners for conversations that led to the development to the Western Monarch Call to Action; US Fish and Wildlife Service Coastal Program, Department of Defense, National Science Foundation, and Xerces Society members and other funders for supporting the work presented in this Perspective.

revealed by four decades of intensive monitoring. Oecologia 181, 819–830. doi: 10.1007/s00442-016-3600-y


Lange's metalmark. Environ. Pollut. 164, 24–27. doi: 10.1016/j.envpol.2012.0 1.011


**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 Pelton, Schultz, Jepsen, Black and Crone. 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.

# Mortality of Monarch Butterflies (Danaus plexippus) at Two Highway Crossing "Hotspots" During Autumn Migration in Northeast Mexico

#### Blanca Xiomara Mora Alvarez <sup>1</sup> , Rogelio Carrera-Treviño<sup>2</sup> and Keith A. Hobson1,3 \*

<sup>1</sup> Department of Biology, University of Western Ontario, London, ON, Canada, <sup>2</sup> Facultad de Medicina Veterinaria y Zootecnia, Universidad Autónoma de Nuevo León, Escobedo, Mexico, <sup>3</sup> Environment and Climate Change Canada, Saskatoon, SK, Canada

The contribution to annual mortality of migrating monarch butterflies (Danaus plexippus)

due to collisions with vehicles is poorly understood but likely significant. Recent estimates based on a study in Texas suggests that mortality during autumn migration may be of the order of 2 million per year or about 3% of the population. However, MaxEnt models used in that study are not well suited to quantifying mortality at hotspots where monarchs are concentrated by topography such as canyons when crossing highways. Potentially catastrophic mortality could occur at such sites if timing of migration and weather conditions conspire to force a large proportion of the migrating population across highways at low altitude. We investigated monarch mortality 15 October to 11 November, 2018 at two highway crossings in northeastern Mexico known for their frequent and extensive collisions (La Muralla and Santa Catarina). During a 15–19 day period of migration, we collected dead and injured monarchs along a series of 500 m roadside transects. We estimated a minimum total mortality during fall migration at just these sites of about 196,560 individuals. Monarchs exhibited a diurnal pattern of passage at Santa Catarina of peaks in late morning and late afternoon. Average vehicle speeds exceeded posted 60 km/h limits designed to protect monarchs, ranging from 75.1 to 99.6 km/h at La Muralla and 86.6 to 106.8 km/h at Santa Catarina. We recommend finer-scale documentation of migration pathways and an inventory of significant highway crossing hotspots for monarchs during fall migration in northeast Mexico. Mitigative measures could include better enforced vehicle speeds at least during the short period of migration, deflection structures to raise the height of crossing monarchs, and/or manipulation of habitat to lower the potential for monarchs descending to roost near key crossing points.

Keywords: roadkills, mortality, vehicles, migration, mitigation

# INTRODUCTION

The eastern North American population of the monarch butterfly (Danaus plexippus) migrates annually from natal sites primarily in the eastern United States and southeast Canada to overwintering sites in the highlands of central Mexico. This journey represents an iconic example of long-distance insect migration and is the focus of tremendous public and scientific interest.

#### Edited by:

Ryan G. Drum, United States Fish and Wildlife Service (USFWS), United States

#### Reviewed by:

Kelly R. Nail, United States Department of the Interior, United States Carl Stenoien, University of Minnesota Twin Cities, United States

> \*Correspondence: Keith A. Hobson Khobson6@uwo.ca

#### Specialty section:

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

> Received: 28 February 2019 Accepted: 02 July 2019 Published: 16 July 2019

#### Citation:

Mora Alvarez BX, Carrera-Treviño R and Hobson KA (2019) Mortality of Monarch Butterflies (Danaus plexippus) at Two Highway Crossing "Hotspots" During Autumn Migration in Northeast Mexico. Front. Ecol. Evol. 7:273. doi: 10.3389/fevo.2019.00273 Currently, conservation of this population is of considerable concern due to long-term declines in the population of migratory individuals (Vidal and Rendon-Salinas, 2014; Thogmartin et al., 2017). Causes of the decline are not well understood, but the disappearance of milkweed host plants on the breeding range, loss or reduction in quality of overwintering sites in Mexico and long-term factors associated with increased use of pesticides and global climate change are all areas of current research (Flockhart et al., 2017; Pleasants et al., 2017; Agrawal and Inamine, 2018; Tracy et al., 2019). Less attention has been placed on factors operating during the migratory phase which can be up to 4,000 km for some individuals. During migration, monarchs must fuel their journey through nectaring at stopover sites en route. They must also secure safe roost sites each evening and ultimately capitalize on suitable winds to aid their migration. Collisions with vehicles has been raised as a potentially important mortality factor for migrating Lepidoptera including monarchs (McKenna et al., 2001; Rao and Girish, 2007; Her, 2008; Skórda et al., 2013; Bennett, 2017; Tracy et al., 2019). However, this topic has been largely ignored apart from anecdotal accounts of roadkill hotspots, especially in south Texas and northeastern Mexico (Correo Real, 2015; Journey North, 2017). Recently, Kantola et al. (2019) used a MaxEnt niche model to estimate monarch mortality due to roadkills during Autumn migration through the Central Funnel, or the constricted flyway from Oklahoma to the Mexican overwinter sites. Their model was based on field data collected in Texas 2016–2017 and extrapolated to highways that occur in the rest of the region. These authors estimated that an average of 2.1 million monarchs are killed annually in the funnel representing about 3% of the overwintering monarch population. Their model provided a useful means of estimating mortality but lacked data from Mexico. Furthermore, the model was less suited to dealing with modeling mortality at acute hotspots where a large percentage of the autumn migration may be funneled by local topography across major highways as occurs in northeast Mexico (Correo Real, 2015).

The general autumn route taken by monarchs from their natal sites to overwintering sites in Mexico are generally known, but specific details on local routes are still lacking. From nectaring and staging sites in Texas, monarchs migrate in large numbers through and along the Sierra Madre Oriental mountains in Coahuila and Nuevo Leon of northeast México and along the eastern cordillera before crossing westward to their high altitude wintering sites in Michoacáan and México states (**Figure 1**). This route in Mexico is characterized by acute concentrations whereby a large percentage of the total migratory population funnels through canyons, river valleys, and other topographical features.

It is at those concentrations or bottlenecks where the population is most susceptible to local conditions. During our own research in northeast Mexico in the vicinity of Monterrey, we became aware of two highway crossings where it was well known among local people that considerable monarch mortality occurs annually during the autumn migration. The nongovernmental organization, Protección de la Fauna Mexicana (PROFAUNA) as part of their Correo Real initiative, have reported significant roadkill at six highways: Highway MEX 054, Saltillo-Concha del Oro, Coahuila; Highway MEX 030, Cuatro Ciénegas, Coahuila; Highway MEX 057, La Muralla, Coahuila; Highway MEX 057, Jaguey de Ferniza, Saltillo, Coahuila; Highway MEX 040, Saltillo-Monterrey; toll Highway MEX 040- D Saltillo-Monterrey, Santa Catarina, Nuevo Leon. One observer counted 115 dead monarch butterflies in the ditch of the toll Highway MEX 040-D in Santa Catarina, Nuevo Leon (Correo Real, 2015) and we believe this to be the first quantitative report of roadkilled monarchs in Mexico. These anecdotal accounts prompted us to attempt the first quantitative estimates of monarch mortality due to vehicle collisions at key points where migratory routes intercept major highways in northeast Mexico.

Our objectives were to focus on two sites well known for their highway collisions with monarchs and to quantify the mortality rates at these sites during the complete passage period. In addition to counting and collecting killed monarchs, we also monitored overall diurnal patterns of movement through these sites. Here, we wished to identify if there were key periods during the day when monarchs were more susceptible to collisions as such information would inform potential mitigation measures. Height of migrating monarchs was also estimated in order to evaluate the percentage of migrants more at risk due to collisions. Our own observations indicated that monarchs flying under 6 m were most at risk with collisions or disturbance from the taller transport trucks so we quantified numbers migrating above (to 20 m) and below that threshold. We additionally aimed to quantify traffic speed at these crossings using hand-held radar and to compare average vehicle speed and type with legislated speed limits. Local authorities have posted signs to alert motorists to crossing butterflies and have specified 60 kilometers/h limits in these regions to reduce monarch mortality but it is not clear if motorists pay any attention to these measures.

Our investigation also complements the recent MaxEnt based model of Kantola et al. (2019) because that study was based entirely on a sample (n = 546) of roadkilled monarchs in Texas and while these authors considered hotspots separately from chronic background mortality rates, conditions leading to hotspot mortality areas can be diverse and highly dependent on local conditions which are hard to model. In addition, hotspots identified in Texas by Kantola et al. (2019) were more diffuse occurring over tens of km whereas the hotspots we identified and studied were much more narrow, corresponding to a funneling of the migration primarily over < 5 km. While mortality at these sites is expected to differ considerably among years, depending on local weather conditions and the nature of the migration, they also represent sites where potentially catastrophic mortality can occur.

#### MATERIALS AND METHODS

#### Monarch Mortality

Field sites were chosen based on our consultation with local people and observers from Correo Real. During the autumn of 2018, we focused on two highways along the Sierra Madre Oriental in the state of Coahuila and Nuevo Leon where monarch butterflies funnel during southward migration (**Figure 1**). One site was on Highway MEX 057, Saltillo-Monclova (landmark

km 128, 26◦ 36′ 72.15′′N, 101◦ 35′ 53.17′′ W) in La Muralla, Coahuila. Along a 10 km extent, the highway follows the base of the mountains with both narrow and wide areas. Two deep major canyons cross the road where high mortality occurs. We focused on a 5 km section of the highway centered on the main canyon crossing (**Figure S1**). The second site was on toll-Highway MEX 40-D, Saltillo-Monterrey (land mark km 58, 25◦ 39′ 18.54′′N, 100 ◦ 27′′ 12.24′′W) in Santa Catarina, Nuevo León. Mountain ridges at this site concentrate monarchs during migration and force a cross-highway passage (**Figure S2**). This toll highway is parallel to a much busier free road (Mex-040) but we were able to sample only on the toll highway. At both study sites, the highways dissect natural habitat of the region with little to no human habitation. The La Muralla site has single lane traffic in each direction and the Santa Catarina highway had two lanes in each direction with a parallel single lane free highway (Mex-040) adjacent.

Transects (500 m) were established along sections of highways in order to quantify monarch mortality. Transects were placed to sample sections of highway and also before and after bridges crossing canyons as these were the most likely sites for monarch passage as well as open areas (e.g., Morris et al., 2015). However, transects were spaced as uniformly as possible in order to capture the total crossing width of the migration at each site. These transects included the shoulder pavement and the ditch, a width which varied depending on the highway and region of the highway sampled. Very few dead or injured monarchs remained on the actual highway and were typically blown to the side. In addition, the main highways were too dangerous to attempt retrieval from those locations. Surveys were performed on one side of the highways and all dead and dying monarchs were collected and placed in plastic bags (**Figure S3**) and archived at the Wildlife Laboratory at the University of Nuevo León. All transects were conducted on consecutive days to ensure each sample represented mortality over a 24 h period. In La Muralla, we established six 500 m transects (i.e., 3,000 m) of 2.9 m width over a total distance of 6 km. In Santa Catarina, we had five transects of 500 m length and 3.4 m width over 14 km. Variable widths of transects reflected local conditions of ditch width and pavement verge. We consider these widths to accurately reflect the sampling regions of the two sites (i.e., where dead monarchs accumulated).

#### Observations of Migrating Monarchs

Daily counts of flying monarchs were conducted at chosen transect points. Observers were stationed at different crossing points and monarchs counted for two 20 min periods each hour (i.e., a total of 40 min/h). We separated flying monarchs into two height categories, 0–6 and 6–20 m. The 6 m threshold corresponded to our estimate of the influence of the taller transport truck vehicles and the 20 m upper limit was based on the fact that this range proved to be manageable for observers. In La Muralla, counts of flying monarchs were conducted at two locations (2 observers each concentrating on different height categories) for 12 days between 09:00 h until 16:30 h (26.355 N, 101.355 W; 26.367 N, 101.355 W) and at Santa Catarina by one observer (recording both height categories) at one location for 16 days between 10:30 and 16:30 h (25.653 N, 100.453 W).

#### Vehicles

We estimated vehicle speed across random assays with a hand-held radar system (SPEEDSTER III- Bushnell), delineated according to the categories passenger vehicle (including SUVs), light truck, and heavy (transport) truck. These categories were used because we realized that the threat to flying monarchs will depend on a combination of the speed and size of the vehicle but we had no a priori expectation of how these factors would interact and direct monarch mortality associated with vehicle type could not be recorded. We did not record which lane target vehicles were using.

#### RESULTS

We were able to judge the start of migration at both sites by driving the routes and looking for dead monarchs on the roads or flying. Rain prevented migration for periods of several days and this dictated largely our attendance at sites during the migration period. At the La Muralla site, during 84 transects over 15 days (15 October to 7 November 2018), we collected a total of 601 roadkilled monarchs (0.14 monarchs/m/d; **Figure 2A**). Unfortunately, we were unable to sample the complete monarch migration period at La Muralla which extended from 22 October through 14 November. At Santa Catarina, during 95 transects over 19 days (24 October to 11 November) we collected 11,280 monarchs (0.24 monarchs/m/d; **Figure 2B**). Again, we were forced to terminate our work before the end of the migration period which lasted until 24 November. These counts represented about half of the actual mortality on each transect because only one side of the highway was sampled and so should be doubled (La Muralla: 0.28 monarchs/m/d; Santa Catarina 0.48 monarchs/m/d). This allowed us to estimate the total mortality over the actual crossing distances and duration of our study assuming our transects were representative of each crossing (La Muralla: 0.28<sup>∗</sup> 6000<sup>∗</sup> 14 = 23,520; Santa Catarina: 0.48<sup>∗</sup> 14000<sup>∗</sup> 19 = 127,680). Because we terminated our fieldwork before the end of the migration period we consulted with Correo Real, who continued to conduct migration monitoring at both locations. Based on those consultations, we feel that each estimate of the actual monarch mortality could be increased by 30% at both sites. That decision was based on the fact that at least one third of the total migration at each location was not documented but the intensity of the migration was also diminishing. So, we

suggest a reasonably conservative estimate of total mortality at La Muralla to have been 30,576 and at Santa Catarina 165,984. Our Santa Catarina estimate was also considered to remain an underestimate because we were unable to sample at the parallel and busier free highway (Mex-040).

At La Muralla, peak movements of monarchs tended to be in the morning and evening (**Figure 3A**). At the Santa Catarina crossing, we observed a more distinct bimodal pattern of diurnal migration (**Figure 3B**) with peak movements during 1,100– 1,200 h and from 1,430 to 1,600 h. During 16 days of regular counts at Santa Catarina, we recorded 203,240 flying butterflies and, of these, 41.4% were crossing the highway below 6 m. At La Muralla, we recorded 25,060 crossing monarchs and of these, 24.8% crossed below 6 m.

The average speed of cars, light and heavy (transport) trucks all exceeded local speed limits (**Table 1**). Daily traffic average volumes (including both directions) were available from Secretaría de Comunicaciones y Transportes for Santa

Catarina in 2018<sup>1</sup> and these were 14,330 vehicles for La Muralla and 8,862 vehicles for the Santa Catarina toll highway (Mex-040D) and 41,377 (both ways) for the parallel free highway (Mex-040; SCT 2019).

#### DISCUSSION

We provide for the first time quantitative estimates for monarch mortality during autumn migration at two highway crossings in northeast Mexico. We estimated that a minimum of 196,560 monarchs were killed by collisions with vehicles during their short and concentrated crossings at these two sites in 2018. We consider this estimate to be conservative, in part, because we did not sample at the busier parallel highway (Mex-040) at Santa Catarina. We stress that these estimates are for only two sites but are significant because they represent points where a large proportion of the entire eastern population of North American monarchs concentrate en route to wintering grounds. Lower but more chronic mortality rates are expected in regions where monarchs are much less concentrated (Kantola et al., 2019). However, there are numerous highways in Mexico which involve monarch crossings and mortality rates there are yet to be estimated. We have provided descriptions of the six sites in northeast Mexico provided by Correo Real and these would be a useful starting point.

Monarchs crossing the La Muralla site were primarily in transit and we observed little structure in their diurnal movements. At Santa Catarina, however, monarchs appeared to look for roost sites in the vicinity of the highway at the end of the day and this typically brought them to lower altitudes where they were more vulnerable to collisions. This observation is important because it suggests that mitigative measures may be appropriate in terms of roadside habitat manipulations (Skórda et al., 2013) in the vicinity of the Santa Catarina hotspot.

We argue that potentially catastrophic mortality could occur at these sites where canyons concentrate a vast proportion of the migratory population. Should weather conditions force monarchs to pass below 6 m over these roadways and traffic passes at high speed and at high volumes, then substantially higher mortality rates than we recorded are expected. We observed that rain halted migration and that local winds clearly can influence vulnerability to road crossing mortality. Future work at hotspots should measure these conditions. While mitigative measures to prevent butterfly mortality have been largely unexplored, we contend that traffic speed is an issue and one which can presumably be enforced at least for the short period of monarch autumn migration. Our observers noted many butterflies injured simply due to wind vortices (vs. actual collisions) caused by high speed vehicles with effects apparent to 6 m, especially for large transport trucks. Presumably, the concentrated nature of these crossings temporally and spatially lend themselves to tailored actions for short periods. This could involve increased use of temporary but more effective speed restrictions through radar traps or simple police presence. Apparently, signage warning motorists to slow down for butterfly crossings (**Figure 4**; https://www.milenio.com/cultura/caminospasa-monarca-limite-60-km) is ineffective.

Alternatively, more ambitious solutions could involve structures that would deflect monarchs over these crossings. For example, in Taiwan, the double-banded crow butterfly (Euploea sylvester) migration crosses major highways and mitigative measures taken to protect this species have included the construction of over a kilometer of four-meter high protective netting. Those measures have apparently reduced the mortality from 20–30 per thousand in 2007 to 4.7 per thousand in 2009 (Taiwan Environment Protection Administration, 2010). Certainly, close monitoring of the autumn migration of monarchs in the areas we studied could provide predictions of timing of movements when direct and effective action could be applied for relatively short periods (in the case of traffic speed restrictions) or more permanently (in the case of erected netting).

We consider our work preliminary and not yet at a level where absolute mortality estimates are possible for the fall migration. Future studies could provide an estimate of the proportion of

<sup>1</sup> Secretaria de Comunicaciones y Transportes http://www.sct.gob.mx/carreteras/ direccion-general-de-servicios-tecnicos/datos-viales/2019/


Posted speed limits were 60 km/h at both sites.

highway crossing. Highway 57 at the La Muralla study site, Coahuila. Photo by Omar Franco Reyes.

monarchs killed to those flying over but this would require several observers recording continuously throughout the day. The potential for high altitude flight (beyond observer range for counting) also suggests that radar may assist in estimating monarch movements at such concentrations (Ovaskainena et al., 2008). So, it is currently unknown if highway mortality in Mexico and elsewhere is a significant and additive mortality factor and could be contributing to population declines. On the other hand, it is clear that traffic volumes are increasing and monarchs are facing more and more sites where collisions occur and are increasing. Monterrey is the third largest city in Mexico (4.8 million) and continues to grow at about 1.2% per year (https:// populationstat.com/mexico/monterrey). This city intercepts a major portion of the eastern continental monarch migration. Future studies should attempt to refine mortality estimates annually at key sites in Mexico where monarch mortality has been identified. We need to know more about specific routes taken by monarchs and identify where these sites intercept significant highway crossings. Mitigative measures need to be explored but we suggest that the provision of suitable and extensive roost sites away from highway crossings may be a way of inducing monarchs to travel on to these "safe" sites to roost and so avoid low altitudes around highways, at least later in the day. The same principle could apply to nectar lure crops to again attract monarchs away from vulnerable highway sites. At other sites, the only feasible solution appears to be the construction of high (∼8 m) netting to deflect monarchs over the highway.

We recommend that estimates of monarch mortality due to collisions with vehicles be conducted throughout the Mexican portion of the autumn flyway. This would allow direct testing and refinement of the MaxEnt model derived by Kantola et al. (2019) for this critical component of the migratory funnel. In addition, mortality estimates need to be continued at known hotspots as done here. Mitigative measures should be developed and experiments conducted to evaluate their effectiveness. Finally, we noted that monarch mortality at hotspots is apparently absent during the spring northward migration in the region where we worked. It is not clear if monarchs take alternate return routes at that time that are "safer" in terms of highway collisions. However, the magnitude of the spring migration in northeastern Mexico will be considerably less than that in autumn due to overwinter mortality.

# DATA AVAILABILITY

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

# ETHICS STATEMENT

Only roadkilled monarchs were used in this study and so no ethical considerations or animal care permits required. All Mexican permits issues to RC-T.

# AUTHOR CONTRIBUTIONS

BM and KH conceived of the project and all authors contributed to the study design. BM conducted most of the fieldwork. All authors helped to write the manuscript.

# FUNDING

Funding was provided by the University of Western Ontario and an NSERC Discovery Grant to KH and by Universidad Autónoma de Nuevo León to RC-T.

# ACKNOWLEDGMENTS

We thank Rocío Treviño Ulloa and all her Correo Real observers who first addressed the monarch roadkill mortality issue in northeastern México. Carlos Carrera Treviño provided important insights and discussion. We greatly value all the assistance in the field from the Wildlife Lab volunteers (Facultad de Medicina Veterinaria y Zootecnia, Universidad Autónoma de Nuevo León) and special thanks to Victoria González Ledezma for her assistances during all the field work. Many thanks to Angel Balbuena Serrano and Zuleyma Zarco González for creating **Figure 1**. The manuscript benefitted from the reviews of Ryan Drum and two reviewers.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00273/full#supplementary-material

#### REFERENCES


Supplementary Figure S1 | Highway 57 at the La Muralla study site, Coahuila. Photo by B. Xiomara Mora Alvarez.

Supplementary Figure S2 | Toll highway 40-D at the Santa Catarina study site in Nuevo León. Photo by Omar Franco Reyes.

Supplementary Figure S3 | Bags of dead monarchs collected on one transect day from the Toll Highway 40-D in Santa Catarina, Nuevo León. Photo taken by Rogelio Carrera Treviño.

monarch butterfly population: pitfalls and prospects. PLoS ONE 12:e0181245. doi: 10.1371/journal.pone.0181245


**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 Mora Alvarez, Carrera-Treviño and Hobson. 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.

# Estimating Overwintering Monarch Butterfly Populations Using Terrestrial LiDAR Scanning

Nickolay I. Hristov 1,2 \*, Dionysios Nikolaidis 1,3, Tatjana Y. Hubel <sup>4</sup> and Louise C. Allen1,2

*1 Information Design Studio, Center for Design Innovation, Winston-Salem, NC, United States, <sup>2</sup> Department of Biological Sciences, Winston-Salem State University, Winston-Salem, NC, United States, <sup>3</sup> Animusing Productions, Winston-Salem, NC, United States, <sup>4</sup> Structure and Motion Laboratory, Royal Veterinary College, London, United Kingdom*

Concerns about the state of decline of the North American monarch butterfly (*Danaus plexippus*) have prompted their consideration for listing under the Endangered Species Act. Data suggest a substantial decline (> 80%) in overwintering numbers for both eastern and western monarch populations. Making an accurate status assessment is difficult due to highly variable density estimates in the eastern monarch overwintering sites. We have developed a novel application of terrestrial LiDAR scanning (TLS) which creates a scene using millions of LASER-based distance measurements in the landscape. In this technology report we discuss the use of TLS and development of Subtractive Volume Estimation (SVE) methodology for estimating overwintering monarch butterfly populations. The principle proposition of the SVE method is to compare volumetric differences between two TLS surveys, a *reference* scan that records roosting monarch butterflies in their overwintering environment and a *derivative* scan, that records the same site without butterflies. Using paired long-range laser scanners, we collected data from four overwintering sites; two in California and two in central Mexico. To help estimate the number of butterflies, we developed an accurate 3D model of an individual monarch. To test the SVE method, we created digital 3D models of bare tree trunks and distal branches, based on laser scans at two sites and combined them with our monarch model to create virtual reference and derivative point clouds. To convert from volume to number of butterflies, we introduce a scaling factor, n, which represents the estimated volume occupied by one butterfly and a correction factor, f, which accounts for variation in clustering behavior and scanner position. While work is ongoing, we confirm that TLS combined with SVE is a suitable technique for surveying clusters of overwintering monarchs at overwintering sites in Mexico and the US.

Keywords: LiDAR, monarch butterfly, population estimates, overwintering counts, technology

#### INTRODUCTION

The globally recognized monarch butterfly (Danaus plexippus) is imperiled, threatened by a fast-changing world across much of its distribution (Thogmartin et al., 2017b; Wilcox et al., 2019). To conserve the iconic species, a growing community of scientists, policymakers and the public must select the right tools and act fast. Novel technologies have long been an indispensable part of the scientific enterprise and a catalyst for new discoveries

#### Edited by:

*Wayne E. Thogmartin, United States Geological Survey (USGS), United States*

#### Reviewed by:

*Kim Calders, Ghent University, Belgium Jason John Rohweder, United States Geological Survey (USGS), United States*

> \*Correspondence: *Nickolay I. Hristov hristovn@cdiunc.org*

#### Specialty section:

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

> Received: *01 January 2019* Accepted: *25 June 2019* Published: *23 July 2019*

#### Citation:

*Hristov NI, Nikolaidis D, Hubel TY and Allen LC (2019) Estimating Overwintering Monarch Butterfly Populations Using Terrestrial LiDAR Scanning. Front. Ecol. Evol. 7:266. doi: 10.3389/fevo.2019.00266* (Hristov et al., 2008; Chilson et al., 2012). Human history is rich with examples of the impact and contribution of such tools as vehicles for the exploration of new frontiers. Population size is an important variable for understanding the biology, ecology, and conservation of any organism. However, estimating the size of large group-living organisms, like the monarch butterfly, under field conditions has been notoriously challenging (Hristov et al., 2010). LiDAR technology, a 3D measurement technique, and its associated computational tools offer a promising method to help tackle the challenge and assess the abundance and density measures needed to build on decades-long efforts to monitor monarch populations in Mexico and the US.

Monarch butterflies exhibit one of the most impressive displays of animal migration, spanning multiple generations and migrating over thousands of kilometers across the continent. Monarchs east of the Rocky Mountains overwinter in dense aggregations in high-elevation Oyamel fir (Abies religiosa) forests in Mexico, while western monarchs overwinter more diffusely at numerous sites in trees on the California coast (Urquhart and Urquhart, 1976; Malcolm, 2018). Regular monitoring of the overwintering population in Mexico began in 1994 by World Wildlife Fund-Mexico in collaboration with a number of other organizations (including the Comisión Nacional De Áreas Naturales Protegidas and Secretaría De Medio Ambiente Y Recursos Naturales; Thogmartin et al., 2017a). The Mexican overwintering population is mapped as total forest area occupied in hectares which is used as an index of population size (Rendón-Salinas et al., 2017). Annual western overwintering population estimates in California began in 1997 (Malcolm, 2018) and records are kept by the Xerces Society<sup>1</sup> , but estimates, via markrecapture methods, were made as early as the 1970s (Tuskes and Brower, 1978). Currently, the California overwintering estimates are produced via direct visual counting of clusters using an "area method" in groves along the coast<sup>2</sup> .

While the methods of estimation differ, data show a consistent decrease in overwintering numbers for both eastern and western monarch populations (Brower et al., 2012; Semmens et al., 2016; Rendón-Salinas et al., 2017; Schultz et al., 2017; Malcolm, 2018). By some measures, monarch abundance may have declined more than 80% over the past 25 years, leading to a petition in 2014 for listing under the Endangered Species Act<sup>3</sup> . US Fish and Wildlife Service (USFWS) is leading a monarch status assessment, to be finalized by 2020, to determine whether the monarch should be listed. The accurate assessment of the size and seasonal dynamics of overwintering monarch butterflies, therefore, is increasingly urgent.

One challenge for completing an accurate status assessment is difficulty determining accurate numbers of the eastern monarch population, due to the size of the task. While monarchs settle in up to 300 sites in coastal California, estimated millions concentrate in just a few locations in Central Mexico (Urquhart and Urquhart, 1976; Malcolm, 2018). As mentioned above, occupied area estimates have been used as a proxy for monitoring the eastern monarch population (Rendón-Salinas and Tavera-Alonso, 2014; Vidal and Rendón-Salinas, 2014; Vidal et al., 2014). There is considerable variation in density estimates at these sites, from 6.9 to 60.9 million butterflies per hectare, making it difficult to obtain accurate numbers–numbers which are critical for conservation efforts (reviewed in Thogmartin et al., 2017a). Regular monitoring has produced evidence of population decline; however, with monarchs now clustered on fewer and fewer trees, more precise and accurate estimates of their population at the individual level are critical. One of the most promising methods of estimating density is the capturerecapture method replicated over multiple years, (Calvert, 2004; Thogmartin et al., 2017a), however the method is disruptive to overwintering butterflies and is currently not allowed. To overcome this limitation, we look to technology to help build upon the decades of work to more accurately and precisely estimate the density and numbers of overwintering monarchs in a non-invasive way (Hristov et al., 2008). The solution, while more critical for use in the eastern overwintering sites, due to their concentration over a smaller area, will be useful for the entire North American population of the species.

To address this urgent need, we propose a novel approach for monarch population estimation using terrestrial LiDAR scanning (TLS) and associated digital tools. TLS creates a detailed digital survey using millions of LASER-based distance measurements from a scanning device to objects in the landscape. We have created innovative workflows and analytical algorithms for obtaining and converting TLS scans and high-resolution imagery of overwintering monarch aggregations into cumulative volumes and estimates of monarch abundance. Terrestrial LiDAR scanning and other photographic techniques have recently been used by biologists to visually count individual bats (Azmy et al., 2012; Shazali et al., 2017), and butterflies (Leong et al., 2017) in clusters. The technology has also been used in other large-scale applications like estimating tree biomass (Momo Takoudjou et al., 2018). In contrast, we expand the use of TLS to generate high-resolution 3D models and use the volumetric data in the models to estimate the number of butterflies in these overwintering sites.

In this technology report we discuss the use of TLS and imaging methods and the development and application of an innovative Subtractive Volume Estimation technique for estimating overwintering monarch butterfly populations. The Subtractive Volume Estimation (SVE) method works by comparing the detailed geometry of a site with and without the animals of interest, deriving as a result an estimate of the overall bio-volume and eventually an estimate for the number of individuals and density. The SVE method has shown great promise to enable the estimation of large clusters of animals and can help increase population count precision and accuracy through technology.

#### MATERIALS AND METHODS

#### Overwintering Sites

We collected data from four monarch overwintering sites; two in California, USA (Santa Cruz Lighthouse Field State

<sup>1</sup>www.Xerces.org/monarchs/

<sup>2</sup>www.pgmuseum.org/blog/2014/10/17/how-do-you-count-monarch-butterflies <sup>3</sup>www.biologicaldiversity.org/species/invertebrates/pdfs/Monarch\_ESA\_Petition. pdf

FIGURE 1 | The analysis of Terrestrial LiDAR Scanning (TLS) data follows a numerical model based on shape analysis. Subtractive Volume Estimation (SVE) compares a scan with butterflies, referred to as the *reference scan,* to a scan without butterflies, known as the *derivative scan* and calculates the difference between the two shapes. That difference represents the volume of butterflies present. The final estimated colony size is based on dividing that volume by a scaling factor, *n*, which represents the volume of an individual butterfly multiplied by a correction factor f that accounts for characteristics of the TLS survey.

Park and Pismo State Beach Butterfly Grove) representing the western population of the species and two sites in central Mexico (Carpinteros and San Antonio) representing the eastern population<sup>4</sup> . The data was collected over 3 years from 2017–2019.

# Terrestrial LiDAR Scanning

To collect TLS data, we used a pair of FARO S70 long-range laser scanners (Faro Technologies, Lake Mary, Florida, USA). This model has a maximum scanning resolution of 700 megapixels and is capable of scanning objects up to 70 m away with a ranging accuracy of ± 1 mm at that distance. The S70 also has the ability to generate a 350 megapixel, composite, color image based on 64 pictures captured with its on-board camera. The composite image can then be overlaid on top of the point cloud during postprocessing of the data to give a realistic representation of the color space of the surveyed environment. Individual scan settings ranged from ½ to ¼ of the maximum resolution, based on the complexity of the surveyed environment, with lower settings used for more cluttered environments and higher-resolution scans used in more open, uncluttered environments. Scanning quality, the number of successful readings before a distance-estimate is recorded, was set at 2x; therefore, we collected approximately 175 million and 44 million distance measurements per scan at ½ and ¼, respectively. The beam diameter at exit for the S70 was 2.12 mm (beam divergence 0.3 mrad). The vertical and horizontal step size of the scanner was 0.009◦ . On average, we positioned the scanners 5.39 meters away from butterfly clusters of interest resulting in a minimum resolvable feature diameter of 1.53 and 3.07 mm at ½ and ¼ resolutions respectively; well below the dimensions of the roosting butterflies. Each scan took between 3 and 12 min so an entire survey generally took < 2 h. We used a pair of scanners to speed up the surveys and worked early in the morning, between sunrise and 10:00 local time. Any measurements of overwintering groves must be taken before ambient temperatures reach ≈16◦C, which is the flight threshold for monarchs (Masters et al., 1988). Above that temperature, the butterflies grow increasingly active and begin to leave their clusters. TLS surveys were composed of 8– 25 scan positions depending on the size and complexity of the

<sup>4</sup>www.xerces.org/wp-content/uploads/2015/10/MonarchMap-NatureServe-10. 20.png

surveyed site. Our goal was to capture most of the groves where butterflies were roosting; from 2–10 trees on approx. 0.2–0.4 hectares at California sites and 0.6–1.7 hectares with dozens of trees in central Mexico. The resulting point-cloud datasets were represented by over 352 million points in California and 4.4 billion points in Mexico. We were able to capture every portion of the habitat where butterflies were roosting at the California sites and large sections of the habitat at the sites in Mexico. From these composite datasets we selected clusters with different densities of butterflies for further analysis.

#### Subtractive Volume Estimation

The principle proposition of our Subtractive Volume Estimation (SVE) method is to compare the volumetric differences between two TLS surveys, a reference scan that records the roosting monarch butterflies in their overwintering environment and a derivative scan, that records the same site without the butterflies. The subtraction of the two geometric volumes results in an expression of the biovolume of the clustering insects. To convert from total volume of butterflies to number of individuals, we introduce a scaling factor, **n**, which represents a species-specific volumetric estimate of the space occupied by one individual. Monarch butterflies arrange themselves differently depending on the ambient temperature, characteristics of their environment and overall size and density of the cluster (Masters et al., 1988; Vidal and Rendón-Salinas, 2014). In addition, differences in the precision of the laser survey due to distance from the scanner, angle of incidence of the laser beam and surface reflectivity require an additional variable, correction factor **f**, that accounts for these characteristics of the TLS instrument and survey process. The result is an estimate for the number of butterflies in a subset cluster of the overall group or the entire colony (**Figure 1**).

#### Data Processing

The individual, raw LiDAR scans were processed and registered together into a composite point cloud using FARO Scene<sup>5</sup> , a proprietary software provided by the scanner manufacturer. Additionally, we applied outlier, edge and stray-point filters in FARO Scene that removed any noise in the data. The point cloud data was then exported for further manipulation by other computational tools.

Few analytical tools exist for point cloud data, therefore, to make use of the richness of LiDAR data, the point clouds must first be converted to polygonal geometry—a rather computationally intensive process (**Figure 2**). Additionally, the polygonal 3D models of these clusters must be enveloped in complete surfaces to be suitable for 3D metrology. The subtraction of volume calculation can be done in either of two ways: (1) direct shape subtraction where the surveyed objects or scenes do not change between the reference and derivative scan (e.g., tree trunk; see **Figure 11A** and **Supplementary Video 2** for example) and the subtracted metrics have a spatial meaning and (2) an arithmetic subtraction of the volumes when organic structures, such as branches, change their shape or position between the reference and derivative scans due to the weight of the butterflies (**Figure 11B** and **Supplementary Video 3**).

To analyze the clustering shapes, we categorize the roosting aggregations of butterflies into three types based on features of the clusters—(A) Type 1, tree-trunk clusters, (B) Type 2, distal branch clusters, and (C) Type 3, canopy clusters (**Figure 3** and **Supplementary Video 1**). Overwintering sites along the California coast feature mostly Type 2 clusters while sites in central Mexico display all three types.

To estimate the individual number of butterflies in the total subtracted volume of a cluster, we calculated both the scaling factor, **n**, and correction factor, **f**, as described above. To derive **n** we developed a physically accurate 3D model of an individual monarch butterfly using scaled reference images of naturally dead monarch butterflies found at the survey

<sup>5</sup>www.faro.com/products/construction-bim-cim/faro-scene

sites. There are no substantial size differences between male and female butterflies and both sexes overwinter at the same sites. The model allowed us to calculate the maximum speciesspecific volume of an individual butterfly. The scaling factor, **n,** for Danaus plexippus, is 49 cm<sup>3</sup> (**Figure 4**). Simulations described below showed that butterfly volume **n** derived from virtual pointclouds were not constant and therefore needed additional modification. We derive the correction factor **f** based on two approaches that both require starting with known numbers of butterflies: (1) The 3D model was used as an instancing reference—a 3D modeling technique that allows the efficient generation of large, repetitive geometries based on a single reference (see below; **Figure 4**), and (2) we identified instances in the LiDAR surveys where we could count individual butterflies or small clusters of them roosting on tree surfaces and extracted their volumes. While (1) initially indicated that **n** is not constant in clusters with different densities, necessitating the formulation of **f**, (2) provided a more realistic measure based on the performance of the LiDAR scanners in natural clusters.

# Testing and Modeling

Validation is an important component of computational and simulation methods as such approaches generally produce a numerical result; but, it is important to understand if such estimates are representative of biological reality. Since invasive or destructive methods of testing and validation that require the handling of butterflies were not an option for this project, to assess the accuracy of the estimation process, we developed a virtual, physically accurate, 3D model of roosting butterflies and arranged them on (1) tree-trunks and (2) distal branches. At present, because of the size, threedimensional complexity, inability of LiDAR to see through opaque surfaces and need for additional proofing data, Type 3, canopy clusters remain a challenge and are the focus of ongoing efforts.

We describe the validation process for the tree-trunk models, but a similar procedure is used for distal branch models. We created digital 3D models of bare tree trunks and distal branches of Oyamel and Eucalyptus trees (Eucalyptus globulus), based on laser scans in California and Mexico (**Figure 5**) and combined these models with the 3D model of a monarch butterfly (described above) to create virtually simulated reference and derivative point clouds. In this virtual environment we were able to "place" known numbers of butterflies in the scene to validate the performance of the SVE algorithm. We used an adapted, virtual photogrammetric method to create point-cloud representations of these virtual 3D structures. Photogrammetry, also known as structure from motion, stitches highly overlapped digital images into an accurate 3-dimensional representation of the scene (Edward et al., 2001). The photogrammetry program, Agisoft Photoscan<sup>6</sup> , analyzes the position and rotation difference of specific patterns between different photos and derives the shape of objects on those photos, as well as the position of the camera that took the photo in relation to those objects. Through experimentation we found that 1 degree of separation between renders was adequate to accurately and effectively calculate a point cloud. We rotated the camera so that it is always facing the center of the trunk, and we took a render at each 1◦ interval. We processed the entire image-sequence in Agisoft Photoscan to calculate a collective point cloud representation of the photographed geometry. A similar process captured the tree trunk without butterflies. The two point clouds were imported into a point-cloud editing program called MeshLab<sup>7</sup> where we used a Screened Poisson Surface Reconstruction algorithm (Kazhdan and Hoppe, 2013) to generate a polygonal surface from adjacent points in the point cloud. The result was two cylindrical 3D models representing the bare tree, and the tree populated with butterflies. Next, the 3D models were closed to make them "watertight," having no holes in the geometry, a requirement for volumetric calculations, and the resulting objects were imported into 3D Coat<sup>8</sup> , a digital sculpting program that treats 3D objects as voxel-based volumes. In this program there are tools that allow the accurate subtraction of one volume from another, and the export of resulting volume as a new object. The subtracted volume difference between the bare trunk and trunk with butterfly models was imported once again into MeshLab, where we calculated the volume of the end shape (**Figure 6**). The process was repeated for a variety of butterfly clusters (with varying known numbers of virtual butterflies)

<sup>6</sup>www.agisoft.com

<sup>7</sup>www.meshlab.net

<sup>8</sup> 3dcoat.com/home

reference scan, including the branch with butterflies and derivative scan of the branch without butterflies. These quantitatively accurate models are used to validate the results from the TLS estimation of real-world butterfly aggregations.

to derive a direct connection between changes in volume and butterfly numbers.

The workflow sequence presented here uses generally available software tools which make them more accessible to potential users and facilitate adoption of this new methodology in practice. Other computational and metrology environments exist such as the industry-standard, Innovmetric Polyworks<sup>9</sup> which provide integrated and more consistent and robust editing and analytical capabilities for higher-order and larger-scale computations when larger clusters or entire sites must be estimated.

#### RESULTS

In this technology report we describe the development of an analytical workflow that begins with the acquisition of threedimensional TLS data in the field and ends with an estimate of the number of butterflies in a section of the roosting habitat.

#### Derivation of Correction Factors

When comparing the simulated 3D models of clusters with an increasing number of butterflies with their surveys derived from virtual photogrammetry, we observed, as expected, that the total volume of the subtractive volume envelopes increased from 503 cm<sup>3</sup> for 10 butterflies to 177,780 cm<sup>3</sup> for 2,000 butterflies (**Figures 6**, **7**). However, when dividing the total subtractive

volume for each cluster by the number of butterflies, known exactly from the simulation engine, to arrive at a representative volume for an individual butterfly, the estimated volume per butterfly was not constant as might be expected (reference values from **Figure 8**). Instead, it changed as a function of the number of butterflies in the simulated clusters in a non-linear way.

the subtracted volume shapes for butterfly clusters of 100–2,000 individuals,

indicating increasing density of the resulting shapes.

If the virtual photogrammetric surveys recorded correctly the geometry of the simulated roosting butterflies, we would have had a 1:1 relationship between the number of butterflies entered in the simulation and the number of butterflies estimated through the SVE process. However, this is not the case. This inconsistency is explained by the relationship between the number of butterflies and the distance among them and how TLS and photogrammetry represent this inter-individual spacing. Noting that when butterflies are fewer (10–100), thus

<sup>9</sup>www.innovmetric.com/en

further apart, the laser and photogrammetric processes correctly detect individual butterflies as separate geometric objects and accurately represent their volumes. On the other end of the spectrum, in dense clusters of many butterflies (≥ 2,000), individuals are arranged closely together and their collective mass is captured accurately by TLS/Photogrammetry as a single envelope. However, during intermediate densities (300–1,000), the spacing among butterflies falls into a range where the virtual photogrammetric process cannot distinguish individual butterflies as separate objects and fuses them into one, filling the intermediate gaps between neighboring butterflies–a product that we call phantom volume. As a result of this additional volume, the estimate for the calculated volume of an individual butterfly (i.e., division of total envelope by correct number of butterflies) incorrectly increases (**Figures 9**, **10**) at these intermediate values (300–1,000 butterflies). Because of these observations we computed the correction factor **f** that is included as part of the scaling factor **n** used for estimating the number of individual butterflies in the total subtracted volume from simulated and natural clusters.

## Estimation of Natural Clusters

Based on these calculations and using values of 49 cm<sup>3</sup> and 0.29 for **n** and **f** respectively, values derived from a small patch of butterflies on a twin-trunk tree at the San Antonio site, we estimate that 192 butterflies occupy a volume of 0.002657 m<sup>3</sup> in that cluster. Thus, 2,641 butterflies occupy 0.0365 m<sup>3</sup> of volume for every 1-meter length of trunk with Type 1 clusters exclusively. Given the approximately 12 meter section covered by butterflies

on each trunk of the tree (24 meters combined for both trunks), we estimate that 63,378 butterflies, occupying 0.8770 m<sup>3</sup> roosted on the entire tree (**Figure 11A** and **Supplementary Video 2**). Similarly, for an SVE analysis of a tree from the Carpinteros site, consisting of Type 1 and Type 2 clusters, we estimate that the total volume of the clusters is 0.3917 cubic meters, resulting in an estimated 36,995 butterflies for the entire tree (**Figure 11B** and **Supplementary Video 3**).

spaced butterflies are not detected correctly and phantom volume is generated, necessitating the introduction of a correction factor (see text and

#### DISCUSSION

Figure 11).

The overwintering population size of monarch butterflies in Mexico, expressed as hectares of forest, has become the most commonly cited measure of yearly monarch abundance. Millions to hundreds of millions of monarch adults may inhabit these few hectares of central Mexican forest in the winter. There is a great variation in published estimates of overwintering density (Brower et al., 2004; Calvert, 2004) making extrapolation

to the entire population difficult. Thogmartin et al. (2017a) used 6 published density estimates to develop a probability distribution for monarch overwintering density in Mexico. While they were able to calculate a median density of 21.1 million per hectare, they acknowledge that considerable uncertainty in the estimated densities remain. Such uncertainty in overwintering densities (and population counts) is a problem when building conservation interventions and particularly when trying to assess the effectiveness of such efforts.

volume of a single individual and *f* represents the derived correction factor for

each monarch density.

In this technical report, we share progress and early achievements in the development of a novel workflow and analytical methods for extracting monarch abundance and density information from TLS. We confirm the suitability and exciting potential of TLS in combination with digital analytical techniques such as SVE, to provide a reliable and non-disruptive method for more accurately estimating overwintering monarch butterfly densities and population size. We demonstrate that, 3D, polygonal reconstructions based on TLS surveys have ample resolution and accuracy to enable the separation of butterflies from the background vegetation and the estimation of butterfly clusters from a few individuals in a simple arrangement to thousands in intricate clusters. Furthermore, we have developed a unique, non-destructive method for validating our estimates, via quantitative 3D modeling and virtual environments.

With these achievements, we deliver on the primary goal of this initiative–to provide an alternative approach to monarch estimation at the individual level that can expand on the decades of monitoring work in North America. Such efforts are well suited for establishing a repeatable standard for calculating overwintering density per unit areas. The work presented here can significantly reduce the uncertainty in density estimates that are still very much part of the conversation about monarch conservation (Thogmartin et al., 2017a). Therefore,

this novel technological approach could be combined with the current area estimate methodology (Rendón-Salinas et al., 2017) for practical monitoring efforts that use the two techniques hand-in-hand.

Despite these promising results, important work remains as we continue to expand the application of the approach to larger sections of forest and clusters with more complex structures. Improved speed, accuracy, portability and new developments in analytical capabilities will help the process and fine-tune the workflows. Still we recognize that, despite its promise and important capabilities, novel application and great promise, TLS/SVE have several shortcomings: At present, we have not attempted to estimate the number of butterflies in Type 3, canopy clusters as we do not know enough about their internal structure. TLS is a surface survey tool, thus dense, canopy clusters in Mexico will require further data collection and analysis to understand the intricate structure of the interior of these massive aggregations. For example, it may be equally possible that these clusters are densely packed with butterflies or largely hollow and thus any assumption, one way or the other, would significantly affect the outcome and estimates in each case (**Figure 3A**). Additional observations, scanning techniques and detailed analyses are underway to inform the application of this method with regards to these Type 3 clusters. Similarly, TLS scanners are expensive and current computational methods for analysis of the data are

FIGURE 12 | A reflective IR photograph of a cluster of monarch butterflies at Pismo Beach State Park Monarch Butterfly Grove, California, showing the absorption of monarchs in the IR part of the electromagnetic spectrum. This effect has potential to be used to identify visually clusters of monarchs in complex environments and thus aid computer vision analysis.

tedious, time-consuming and requiring trained staff to perform both the data collection and analytics.

Nevertheless, TLS combined with SVE brings important and promising capabilities to establish the foundation for future development of individual-based whole-colony assessment. With its high inherent accuracy and smooth workflow, TLS/SVE sets the groundwork for moving this agenda forward. Perhaps the greatest promise of such future developments is the use of spectral imaging in combination with photogrammetry. Just as we described in the validation work discussed here, photogrammetric (aka Structure from Motion) reconstructions from serial sequences of images can produce detailed pointcloud data that can be used for 3D models to analyze for census estimates. This capability, combined with emerging evidence that spectral imaging, particularly IR photography, provides strong separation of monarch butterflies from background vegetation a challenging computer vision problem (**Figure 12**). The two technologies combined, could offer an even more efficient and affordable technique for monarch butterfly enumeration. Further technical developments could provide important new proxies that could develop in a similar way to bat census work over the last two decades (Betke et al., 2008; Kloepper et al., 2016; Hristov pers. obs.). In that work initial efforts were intensive and expensive (multiple years and ∼\$500K) while current work is quicker and cheaper (a few evenings and \$1–2K) which offers agencies, field workers, policy managers, and others accessible, effective tools for managing and conserving resources. As this technology matures, and more user-friendly and capable processing environments for stitching reconstructions emerge, the early work described by TLS will be superseded by these more accessible technologies and tools.

In the meantime, TLS and associated analytics offer an important formalization of the methods and practices to collect, organize and archive critical temporal and spatial information on monarch butterfly distribution and population trends. TLS not only provides important survey data that can serve as the starting point for further analysis and estimation, it serves as a detailed record in space and time to archive the overwintering behavior of the species, similar to efforts around the world to protect and archive sites of historic and cultural importance by groups like CYARK<sup>10</sup> This is particularly important in the face of unprecedented change in distribution and population numbers for the monarch butterfly.

#### CONCLUSSION

The Species Status Assessment for the monarch is currently underway by FWS with completion required by 2020. Improved abundance estimates are important for use in monarch extinction risk models that have been created for the Assessment. An ultimate goal is to vastly improve our ability to quantify the number and density of butterflies at overwintering sites, particularly in the highly localized eastern monarch sites in central Mexico. This paper describing the establishment of a robust TLS workflow and the development of SVE-based analytical algorithms improves our chances of making that goal a reality and these efforts stand to benefit the field of animal enumeration tremendously. Ultimately, this method is one of several early steps to understanding the dynamic aggregation of monarchs at a finer scale that will allow for more precise and accurate density and populations estimations.

# AUTHOR CONTRIBUTIONS

NH conceived and designed the study. LA and NH collected the data. NH, LA, DN, and TH analyzed the data and contributed to technology adaptations. NH and LA wrote the manuscript.

#### ACKNOWLEDGMENTS

We are extremely grateful to our colleagues who helped facilitate access and work at field sites, including Ronnie Glick, Danielle Bronson, with California State Parks, and Samantha Markum with USFWS and our collaborators with CONANP, WWF-Mexico, and the stewards of the Monarch Biosphere Preserve, Mexico. We are thankful to undergraduate assistants at Winston-Salem State University, Georgina Dzikunu, Zaria Jean, and Timothy Nixon, for help with TLS processing and analysis and to Raunak Kapoor and Trent Spivey for help with data collection. We also thank Ryan Lebar for insight into spectral photography. We thank Ryan Drum and Shauna Marquardt from USFWS and Ralph Grundel and Wayne Thogmartin from

<sup>10</sup>https://www.cyark.org

#### REFERENCES

Azmy, S. N., Sah, S. A. M., Shafie, N. J., Ariffin, A., Majid, Z., Ismail, M. N. A., et al. (2012). Counting in the dark: non-intrusive laser scanning for population counting and identifying roosting bats. Sci. Rep. 2:524. doi: 10.1038/srep00524

USGS for constructive conversations that helped shape this work. We wish to thank Lisa Stack, without whose logistical expertise this work would not have been accomplished and Chris White, Tommy Maddox, and Chris Morabito from Faro Technologies for technical support with LiDAR. Two reviewers (KC and JR) improved greatly the quality and clarity of the manuscript for which we are especially grateful. This work was supported by the US Department of the Interior via DOI US Fish and Wildlife Service awards (F18AC00004 and 0040294142) and the Commission for Environmental Cooperation, with additional support provided by National Science Foundation DRL-1514776. Publication costs were supported by a Winston-Salem State University Faculty Professional Development Committee research grant to LA. Funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00266/full#supplementary-material

Supplementary Video 1 | A high-resolution, three-dimensional, animated model of an Oyamel Fir tree in central Mexico covered in monarch butterflies. The model is generated based on TB LiDAR which is then converted into high-resolution, polygonal mesh. The intricate detail of the model allows the analysis of elaborate clusters and roosting patterns of the butterflies. We identify three clustering patterns on these trees: *Type 1*, tree trunk clusters; *Type 2*, distal branch clusters and *Type 3*, canopy clusters (see also Figure 3).

Supplementary Video 2 | A high-resolution, three-dimensional, animated model of a pair of Oyamel Fir trees in central Mexico (San Antonio site) covered in monarch butterflies with Type 1 clusters. The model is generated based on TB LiDAR which is then converted into high-resolution, polygonal mesh. The intricate detail of the model allows the analysis of elaborate clusters and roosting patterns of the butterflies. Segment at 00:00–00:07 shows the *point cloud without butterflies*; segment at 00:07–00:13 shows the high-resolution *polygonal mesh without butterflies*; segment at 00:13–00:19 shows the *point cloud with butterflies*; segment at 00:19–00:24 shows high-resolution *polygonal mesh with butterflies*; segment at 00:24–00:30 shows the *combined*, with and without butterflies, mesh models; segment at 00:30–00:36 shows the *subtracted volume* of the butterflies used for the number of individuals estimate (see also Figure 11A).

Supplementary Video 3 | A high-resolution, three-dimensional, animated model of an Oyamel Fir tree in central Mexico (Carpinteros site) covered in monarch butterflies with Type 1 and Type 2 clusters. The model is generated based on TB LiDAR which is then converted into high-resolution, polygonal mesh. The intricate detail of the model allows the analysis of elaborate clusters and roosting patterns of the butterflies. Segment at 00:00–00:06 shows the *point cloud with butterflies*; segment at 00:06–00:12 shows the high-resolution *polygonal mesh with butterflies*; segment at 00:12–00:18 shows the *point cloud without butterflies*; segment at 00:18–00:24 shows the high-resolution *polygonal mesh without butterflies*; segment at 00:24–00:30 shows the *combined*, with and without butterflies, mesh models; segment at 00:30–00:36 shows the *subtracted volume* of the butterflies used for the number of individuals estimate (see also Figure 11B).

Betke, M., Hirsh, D., Makris, N. C., McCracken, G. F., Procopio, M., Hristov, N. I., et al. (2008). Thermal imaging reveals significantly smaller brazilian free-tailed bat colonies than previously estimated. J. Mammal. 89, 18–24. doi: 10.1644/07-MAMM-A-011.1


Available online at: http://assets.panda.org/downloads/monitoreo\_mariposa\_ monarca\_en\_mexico\_2013\_2014.pdf (accessed December, 2018).


**Conflict of Interest Statement:** DN is the owner of Animusing Productions.

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.

Copyright © 2019 Hristov, Nikolaidis, Hubel and Allen. 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.

# Size of the Canadian Breeding Population of Monarch Butterflies Is Driven by Factors Acting During Spring Migration and Recolonization

Tara L. Crewe<sup>1</sup> \* † , Greg W. Mitchell <sup>2</sup> and Maxim Larrivée<sup>3</sup>

<sup>1</sup> National Data Center, Bird Studies Canada, Port Rowan, ON, Canada, <sup>2</sup> Wildlife Research Division, Environment and Climate Change Canada, Ottawa, ON, Canada, <sup>3</sup> Insectarium de Montréal, Montréal, QC, Canada

#### Edited by:

Jay E. Diffendorfer, United States Geological Survey (USGS), United States

#### Reviewed by:

Nathan Lemoine, Colorado State University, United States Chip Taylor, University of Kansas, United States

#### \*Correspondence:

Tara L. Crewe tlcrewe@gmail.com

#### †Present address:

Tara L. Crewe, College of Engineering, IT and Environment, Charles Darwin University, Darwin, NT, Australia

#### Specialty section:

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

> Received: 15 January 2019 Accepted: 31 July 2019 Published: 14 August 2019

#### Citation:

Crewe TL, Mitchell GW and Larrivée M (2019) Size of the Canadian Breeding Population of Monarch Butterflies Is Driven by Factors Acting During Spring Migration and Recolonization. Front. Ecol. Evol. 7:308. doi: 10.3389/fevo.2019.00308 The eastern North American monarch butterfly population shows a long-term population decline. While it is hypothesized that forest loss on the wintering grounds and milkweed loss throughout the breeding range are responsible for the observed decline, there is much less certainty regarding the factors driving year-to-year variation around the current population level. Using 15 years of butterfly count data, we used a community-based approach to delineate the stage of the annual cycle during which population limiting factors are most strongly acting. We compared annual fluctuations in size of the breeding population of monarch butterflies in Canada to fluctuations in 13 additional butterfly species which either migrate long distances to Canada or are resident but breed in similar habitats to the monarch. We show that the breeding population of monarchs in southern Canada shows a higher degree of synchrony with other long-distance migrants than with breeding residents, and that annual fluctuations of all migrant butterflies show a positive correlation with the number of 21◦C days during spring migration and re-colonization. Further, we found that size of the monarch breeding population shows a higher degree of synchrony with the size of the following winter population than with the size of the previous winter population. Combined, our results suggest that the monarch population in Canada is limited by factors acting during spring migration, and that weather plays an important role in the ability of the monarch to successfully re-colonize and breed in the northern portion of their summer range each year. A predicted increase in temperature in the early spring, combined with continued loss of breeding and wintering habitat, has the potential to limit the reproductive capacity of monarchs and their ability to recover from population lows.

Keywords: monarch butterfly, Danaus plexippus, population limitation, community ecology, temporal synchrony, weather

# INTRODUCTION

Monarch butterfly (Danaus plexippus) populations are declining throughout their eastern range, during all phases of the annual cycle (fall migration: Crewe and McCracken, 2015, but see Badgett and Davis, 2015; over-wintering: Thogmartin et al., 2017a; breeding: Pleasants and Oberhauser, 2013), and its migration is considered by some to be an endangered biological phenomenon (Brower et al., 2012). While it is generally hypothesized that milkweed loss throughout the monarch's breeding range and forest loss on the monarch's wintering grounds are largely responsible for the 84% population decline observed since the mid-1990s (Stenoien et al., 2016; Malcolm, 2017; Thogmartin et al., 2017b), there is much less certainty regarding the primary factors driving year-to-year variation around the overall negative population trend. For example, Thogmartin et al. (2017b) found that early warm temperatures in the northern breeding range negatively affected population size on the wintering grounds the following winter, but that warmer temperatures later in the season had a positive effect, likely due to effects on the growing conditions for milkweed. Conversely, Badgett and Davis (2015), Ries et al. (2015), and Inamine et al. (2016) suggested a lack of correlation between fall population indices in the US and population size on the wintering grounds is evidence that limiting factors are occurring during fall migration. Given these competing, yet not mutually exclusive hypotheses, there is a need to better understand when, and where limiting factors are occurring in the annual cycle of the monarch butterfly.

In Canada, the monarch butterfly is listed as "Special Concern" but is being considered for uplisting to the status of "Endangered" under the Species at Risk Act (SARA), and occurs in greatest numbers in southern and eastern Ontario, where they are at the northern limit of their breeding range (COSEWIC, 2016; Environment Climate Change Canada, 2016). The population size of monarchs in Canada may be limited by factors acting throughout the monarch's annual cycle. For example, in the north central US, summer breeding populations are best predicted by higher amounts of precipitation, and cooler temperatures in Texas during early migration/breeding, likely reflecting ideal growing conditions for milkweed in the south (Zipkin et al., 2012; Saunders et al., 2016, 2018). Indeed, correspondence between changes in egg density and milkweed abundance in the upper mid-western United States with the number of individuals migrating in fall in Canada (Crewe and McCracken, 2015) suggests that breeding habitat and/or weather conditions experienced at breeding locations south of Canada are important. Northern breeding populations could also be limited by habitat availability and/or weather conditions as well as local density dependence on the breeding grounds (e.g., Marini and Zalucki, 2017; Thogmartin et al., 2017b). Alternatively, previous winter's population size in Mexico has been found to be positively correlated with the following winter's population size (Thogmartin et al., 2017b), suggesting a positive ripple effect through successive breeding generations that may affect breeding numbers reaching the north. Last, conditions during fall migration such as nectar availability and suitable weather conditions for migration may influence migration success and the ability of monarchs to reach the wintering grounds (Badgett and Davis, 2015; Ries et al., 2015, and Inamine et al., 2016), potentially affecting Canada's overall contribution to the winter population.

Until now, research addressing monarch population limitation has generally been conducted in isolation from the broader butterfly community. However, if weather and habitat change are acting not just on monarchs, but on the entire butterfly community, then the amount of temporal synchrony among species with similar life history strategies has the potential to reveal broad-scale relationships between observed patterns of change, and the processes driving those changes (Michel et al., 2016). Monarchs are among several long-distance migratory butterfly species that breed in southern Canada; several migratory species overwinter in the southern United States, north of the monarch's Mexican over-wintering grounds. Because these additional migrants do not overwinter in the same region as monarchs, temporal synchrony among those migrants on their Canadian summer breeding grounds would lend support to the hypothesis that migrant butterflies, including monarchs, are most limited by factors influencing the success of the spring migration and recolonization period. In contrast, many species that breed in similar habitats to the monarch are non-migratory residents in southern Canada. Greater temporal synchrony between monarchs and resident species would support the hypothesis that breeding habitat quantity or quality and/or weather conditions directly affecting survival and reproduction during breeding limit southern Canada's breeding monarch population.

We use monarch overwintering density estimates (Semmens et al., 2016; Thogmartin et al., 2017a) and butterfly survey data collected in the form of checklists by the Ontario Butterfly Atlas (2003–2017; Macnaughton et al., 2017), and eButterfly (2012–2017; Prudic et al., 2017) to test four competing hypotheses on whether monarchs breeding in southern Canada are most strongly limited by factors acting during fall migration, winter, spring migration and recolonization, or summer breeding (**Table 1**). We predicted that if monarchs are limited by factors acting during fall migration ("fall limitation hypothesis"), there will be weak correlation between relative abundance on the breeding grounds and subsequent wintering density; and if monarchs are limited by factors acting on their overwintering grounds ("winter limitation hypothesis"), there should be a strong correlation between previous winter density and relative abundance on the breeding grounds during the following summer. Alternatively, we predicted that if monarchs are limited by conditions during spring migration and re-colonization ("spring limitation hypothesis"), there will be (a) weak correlation between previous winter population density and breeding abundance, and (b) stronger temporal synchrony between monarchs and other migratory butterflies, than between monarchs and resident butterfly species. If monarchs are limited by conditions on their southern Canada breeding grounds ("summer limitation hypothesis"), we predicted that there should be (a) strong correlation between summer relative abundance and monarch densities during the following winter, and (b) stronger synchrony between monarchs and resident butterfly species than between monarchs and species which undergo long-distance migrations. By incorporating community-level comparisons of a 15 year time series dataset and comparing indices of summer and winter population densities, we differentiate between these competing limitation hypotheses, and test when during the annual cycle monarchs breeding in Ontario are most likely limited.

TABLE 1 | Competing hypotheses testing when during the annual cycle monarch butterfly populations in Canada are most limited.


Predictions for correlations between wintering and breeding populations are not mutually exclusive; by also testing predictions for temporal synchrony among migrant and resident butterfly species, we can better distinguish support for or against the competing hypotheses.

# METHODS

#### Data Collection Overwintering Population Size

Estimates of the overwintering density of monarchs (2002–2014) were obtained from Thogmartin et al. (2017a), which provides the raw observed hectares collected by the World Wildlife Fund Mexico, fitted hectares as estimated by Semmens et al. (2016), and associated predicted population size in millions of individuals (see Thogmartin et al., 2017a for details).

#### Ontario Breeding Population Size

Annual indices of population size for the monarch butterfly and additional migratory and non-migratory butterfly species that share similar breeding habitats (**Table 2**) were estimated using butterfly checklist data from the Ontario Butterfly Atlas (hereafter "Atlas") for the years 2003–2017. The Atlas includes occurrence and abundance data for Ontario from several sources including museum collections, eButterfly, Butterflies and Moths of North America (BAMONA), and iNaturalist (Macnaughton et al., 2017). eButterfly is a crowd-sourced checklist-based webplatform for gathering presence/presumed absence data from across North America (Prudic et al., 2017). Regional experts verify the validity of submitted observations on both Atlas and eButterfly platforms. Records submitted to the Atlas were excluded when: (1) they had missing day, month, or year, (2) "x" was listed as the count, or (3) "road-kill" or "specimen" were listed as the record type.

Because Atlas data are submitted as independent observations for each species, a "checklist" is defined here as observations submitted by an observer or group of observers on a particular date for a particular location (latitude/longitude, rounded to the nearest 2 decimal places, or ∼ 1-km accuracy); i.e., each unique combination of date, latitude, longitude, and observer(s) is considered a checklist. Prior to calculating total number of species observed (hereafter "list length") on a checklist, we TABLE 2 | Species included in community level analyses of temporal synchrony.


Species were classified as either long-distance migrants ("Migratory") or resident; resident species were further classified as either univoltine, two-generation, or multi-generation.

aligned the butterfly taxonomy where it differed among databases represented in the Atlas (**Supplementary Table 1**). Records of unknown or very rare species were excluded from the data, and records for individuals with unknown species but known genus (e.g., Colias sp.) were included in calculations of list length only when the Genus was not already present on a list. In total, 160 species were detected by the Ontario Atlas and included in the calculation of list length (**Supplementary Table 1**).

After calculating list length, we used the open GIS software QGIS (v.2.18.16) to select lists that fell within the predicted breeding range of the monarch, based on known milkweed and monarch distributions (Larrivée, unpubl. data; **Supplementary Figure 1**). Lists with fewer than 4 species were also removed to avoid including rare species reports (Breed et al., 2013). Because list length declined with an increase in latitude, we controlled for the effect of latitude on list length by using the residuals from a linear regression of the log of list length by latitude as a surrogate for list length (**Supplementary Figure 2**). We further filtered data for each species by excluding lists collected before or after the minimum and maximum observation date for that species across years.

#### Weather Data

We used the NCEP.gather function of the RNCEP package (Kemp et al., 2012) in R (R Core Team, 2017) to extract surface-level weather variables for the lower-, mid-, and upper United States in April, May and June, respectively, from the National Centers for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) Reanalysis data set (Kalnay et al., 1996). These data have a spatial resolution of 2.5◦ × 2.5◦ and a temporal resolution of 6 h (00, 06, 12, 18 h UTC; Kemp et al., 2012). For the month of April, weather data were extracted for the lower U.S., which we defined as the area encompassed by latitudes in the 30 to 35 degree range, and within −104 to −83 degrees longitude. In May, we extracted weather data from the mid-U.S., which we defined as the area

encompassed by latitudes in the 35 to 40 degree range, and within −104 to −83 degrees longitude. In June, we extracted data for the upper-U.S., defined as the area encompassed by latitudes in the 40 to 45 degree range, and within −104 to −83 degrees longitude. For each month, we calculated the total number of ≥ 21◦C days, and for April and May, total number of freezing (≤ 0 ◦C) days. The mean number of 21◦C days and freezing days across locations in a region was used as an index of annual "21◦ days" and "0 days," respectively. We chose the total number of ≥ 21◦C days because Thogmartin et al. (2017b) found that this number for early May was negatively correlated with proceeding winter population sizes. Data for May 0◦C days were not complete in 2009, and 2009 was excluded from those analyses.

#### Breeding Population Estimates

We estimated annual indices of population size for each butterfly species (**Table 2**) by fitting a modification of the list length analysis for checklist data described in Breed et al. (2013) and Szabo et al. (2011) to the Ontario Butterfly Atlas data. Because monarchs are large and conspicuous, probability of detection nears 0.75 or greater at the end of the breeding season during both low and high abundance years (**Supplementary Figure 3**). We therefore used abundance and assumed a negative binomial distribution for overdispersed counts in place of an occurrence model with a binomial data distribution as in Breed et al. (2013) and Szabo et al. (2011). For each species we used hierarchical linear regression to model the number of individuals observed on a list as a function of fixed effects for list length (residuals of the linear regression of log list length by latitude, as described above) and year. The parameter for list length was assumed to control for all factors that influence detection, including effort, observer skill, weather, time of day, etc. (Szabo et al., 2011; Breed et al., 2013). We re-parameterized the model by removing the intercept for year; in doing so, an effect size for each year, after accounting for random variability among years, was output for the model. These year effects provided a year-specific estimate of abundance, which we used as an index of relative abundance in the correlation and synchrony analyses described below. Random first-order autoregressive random effects for year and day of year nested within year were also included to account for temporal autocorrelation of counts among years and days of the year, respectively. All models were fit in a Bayesian framework using integrated nested Laplace approximation with the INLA package (Rue et al., 2009; Martins et al., 2013) in the R statistical programming language (R version 3.4.0; R Core Team, 2017).

# Temporal Synchrony

To tease apart which of the four competing hypotheses (**Table 1**) was best supported by the data, we tested (1) the correspondence of monarch overwintering and breeding annual indices, (2) community-level correspondence of indices among migrating and resident butterfly species, and between monarchs and other migrant or resident species that breed in similar habitats; and (3) correspondence of butterfly and weather indices for the time period during the annual life cycle deemed to be driving population fluctuations based on results from (1) and (2).

#### Correspondence of Breeding Monarch Butterflies and Overwintering Densities

We tested whether size of the breeding population was correlated with the size of the previous and following overwintering population (raw observed hectares and fitted hectares, as described above) using Spearman rank correlation for both the raw and de-trended breeding and wintering annual population indices. De-trending removes any underlying linear trend from a time series by taking the residuals of a linear model fit to the indices and allows fluctuations in indices to be compared after accounting for any linear trend over time. Indices were detrended using the detrend function of the RSEIS R package (Lees and Harris, 2008). Because temporal autocorrelation of annual indices can bias cross-correlation estimates, we adjusted the p- and t- values of the r estimate by correcting for the loss of degrees of freedom due to temporal autocorrelation, following the methods described in Kirchner (2001), Michel et al. (2016). Next, we used the peaks function of the synchrony R package (Gouhier and Guichard, 2014; R Core Team, 2019) to determine the proportion of maxima and minima that corresponded between the breeding and wintering population time series, for both raw and de-trended annual indices of abundance.

#### Synchrony Among Resident and Migratory Butterfly Species

For community level correspondence of annual indices, we first tested the correlation of de-trended annual indices among species-species pairs using Spearman rank correlation, correcting for any temporal autocorrelation among annual indices (Kirchner, 2001; Michel et al., 2016). We then used a hierarchical linear model with the correlation coefficient (r) as the dependent variable, comparison type (migrant to migrant [M-M]; migrant-resident [M-R]; and resident to resident [R-R]) as an explanatory factor, and species 1 and 2 as random effects to account for repeated measures across species in the species-species comparisons. For the 5 migrant and 9 resident species (**Table 2**), this resulted in 10 M-M, 36 R-R, and 45 M-R correlations. Next, we used the community.sync function of the synchrony R package (Gouhier and Guichard, 2014) to estimate the community-level synchrony across (1) all species, (2) migrant butterfly species, and (3) resident butterfly species. If monarchs are limited by factors acting during spring migration and recolonization, migrant species should show a higher degree of synchrony amongst each other, than with southern Canada resident butterfly species. Alternatively, if monarchs are limited by factors acting during the breeding season, there should be greater synchrony between monarchs and resident butterfly species. Finally, we compared the correspondence of maxima and minima in the time series of species-species pairs using the peaks function in the synchrony R package. Using the same modeling structure described above, we also fit a hierarchical linear model with proportion of corresponding peaks as the dependent variable. Last, we repeated the same analysis with the proportion of corresponding peaks as the dependent variable, but only considering comparisons between the monarch and each other species.

#### Correspondence of Weather Variables With Breeding Population Indices

We tested the correlation of detrended weather and breeding annual indices for migrant butterflies using Spearman rank correlation, with t- and p-values adjusted for temporal autocorrelation of counts as described above and in Kirchner (2001), Michel et al. (2016). We also used the peaks function in the synchrony R package to estimate the proportion of corresponding maxima and minima between the de-trended annual indices of each migrant and each weather variable (mean number of 21◦C days and 0◦C days in April in the lower U.S., May in the mid-U.S., and June in the upper mid-west).

# RESULTS

Between 2003 and 2017, total number of lists submitted to the Ontario Butterfly Atlas increased significantly (linear regression: DF = 1, 13, F = 13.57, p = 0.003), but mean list length did not (linear regression: DF = 1, 17,352, F = 0.28, p = 0.59). Across years, mean and median list length were 8 and 7

the Ontario Butterfly Atlas (2003–2017). Indices were estimated using hierarchical linear regression that assumed a negative binomial data distribution, with count as the dependent variable, and with residuals of list length by latitude (a measure of effort), and year (factor) as explanatory variables. Temporal autocorrelation among

years and among days within years were accounted for through the specification of random year and day effects.

TABLE 3 | Spearman correlation (N.Eff, effective sample size after accounting for temporal autocorrelation of indices; r, correlation coefficient; t, t-value; p, p-value) and proportion of corresponding maxima and minima (Prop. peaks) of breeding annual indices estimated using Ontario Butterfly Atlas data with overwintering population estimates (density, fitted Hectares, and observed hectares) for the winter previous and following the breeding season, and using raw and de-trended annual indices.


Values in bold indicate correlations that were statistically significant at p ≤ 0.05.

species, respectively. Because list length declined with increasing latitude, we used residuals from the linear regression of log list length by latitude (DF = 1, 17,352, F = 145.3, p < 0.0001; **Supplementary Figure 2**) in our trend analyses, to model the effect of list length after accounting for variation due to latitude. Estimated annual indices for the 5 migrant and 9 resident butterfly species are shown in **Figure 1**.

#### Temporal Synchrony Correspondence of Wintering and Breeding Population Estimates

Annual indices for the southern Canada monarch breeding population had a significant positive correlation only with the following winter's fitted hectares and density estimates (**Table 3**), which lends support to the summer limitation hypothesis and suggests a lack of support for the fall limitation hypothesis (**Table 1**). Weak correlation with the previous winter's estimates (**Table 3**) lends support to the spring limitation hypothesis and suggests a lack of support for the winter limitation hypothesis (**Table 1**). Proportion of corresponding maxima and minima ("Prop. Peaks," **Table 3**) in the time series was also greater between the summer breeding and following winter populations, than between the breeding and previous winter populations (**Figure 2**), again supporting the summer and spring limitation hypotheses, respectively. De-trending the time series did not increase the proportion of corresponding peaks, but did result in an increase in the Spearman correlation coefficient from 0.28 to 0.66 (p = 0.02) and 0.70 (p = 0.01) for the fitted hectares and density overwintering estimates, respectively (**Table 3**; **Figure 2**), suggesting that breeding and wintering populations show a greater correspondence in the direction of annual fluctuations than among linear trends in counts over time.

#### Correspondence Between Migrant and Resident Butterfly Species

Mean correlation of annual indices for migrant-migrant species comparisons was greater than mean correlation among migrant-resident and resident-resident comparisons, and 95% confidence intervals of migrant-migrant and migrant-resident groups did not overlap. This result suggests that the annual indices of long-distance migrants are more strongly correlated with the annual indices of other migrants, than they are with annual indices of resident butterfly species (**Table 4**, **Figure 3A**; for raw correlation estimates see **Supplementary Table 2**) and suggests stronger support for the spring limitation hypothesis than for the summer limitation hypothesis. The stronger correlation between migrant-migrant pairs compared with migrant-resident pairs was maintained when residents were classified into groups according to number of generations in a year (univoltine, 2-generation, or multi-generation; **Supplementary Figure 4**). Spearman correlation coefficients were also significantly greater when monarch annual indices were compared to those of other migrants, than when compared with Ontario breeding residents (linear regression: intercept = 0.38, p < 0.001; group (M-R) = −0.27, p = 0.01; **Figure 3B**).

For the butterfly species compared, community-wide synchrony was estimated to be 0.44 (p = 0.01) across all 14 migrant and resident butterfly species, 0.48 (p = 0.01) across the 9 resident species, and 0.62 (0.04) across the 5 migrant species. When residents were broken down into groups based on number of breeding generations, community-wide synchrony increased to 0.64 (p = 0.04) among multi-generational residents, 0.89 (p = 0.01) among 2-generational residents, and 0.49 (p = 0.04) among univoltine residents.

The spring limitation hypothesis was also supported by results of proportion of corresponding peaks, which was greatest among migrant-migrant comparisons. Confidence limits (95%) of the M-M estimate did not overlap with those of the migrant-resident and resident-resident estimates, which supports that migrants are showing more similar fluctuations in annual indices as a group than when migrants are compared with resident butterfly species (**Table 4**, **Figure 4A**; for raw proportion of corresponding peaks for each species-species pair, see **Supplementary Table 3**). As with correlation estimates, the stronger relationship between migrant-migrant pairs than migrant-resident pairs was maintained when residents were further broken down into groups according to number of annual generations (**Supplementary Figure 5**). When comparisons were restricted to those that included the monarch butterfly, proportion of corresponding maxima and minima between the monarch's annual indices and indices of other migrants was significantly greater than when monarch annual indices were compared to those of resident butterfly species (linear regression: intercept = 0.68, p < 0.0001; group (M-R) = −0.42, p < 0.0001; **Figure 4B**).

#### Correspondence of Weather Variables With Breeding Population Indices

Annual indices of migrant butterfly species tended to show greater correspondence with the number of 21◦C days in April

not shown).

TABLE 4 | Estimates for mean Spearman correlation (r) and proportion of corresponding peaks for species-species comparisons between migrant species (M-M, n = 10), between migrant and resident species (M-R; n = 45), and between resident species pairs (R-R; n = 36).


Estimates were derived from a hierarchical linear model where r or proportion of corresponding peaks was the response variables, group was a predictor variable, and species 1 and species 2 were included as random effects to account for repeated measures on each species across species-species pairs.

and May than with 21◦C days in June and the number of 0◦C days in April or May (**Figure 5A**), though the only statistically significant relationships were for the correlation between annual indices of the monarch (r = 0.68, p = 0.01) and question mark butterflies (0.56, p = 0.05) and the number of 21◦C days in April in the lower U.S. (**Figure 6**; raw correlation coefficients in **Supplementary Table 4**). Correspondence of maxima and minima in the time series was greatest for all migrant species with the number of 21◦C days in May in the mid-U.S. and June in the upper mid-west (**Figures 5B**, **7**; raw proportion of peaks in **Supplementary Table 4**). Proportion of corresponding peaks was significant between annual indices and May 21◦C days for monarch (0.70, p = 0.02), painted lady (0.73, p = 0.03) and American painted lady (0.64, 0.04) butterflies, and between annual indices and June 21◦C days for monarch (0.70, p = 0.04) and American painted lady butterflies (0.73, p = 0.02).

#### DISCUSSION

Our unique butterfly community analysis allowed us to assess and disentangle the relative importance of migration and summer breeding factors as drivers of annual and interannual variability in monarch population size in southern Canada. We show that weak correspondence between the size of the overwintering population and subsequent breeding population, along with stronger temporal population synchrony among migrants than among resident butterflies, supports the hypothesis that monarchs breeding in Canada are most

FIGURE 3 | Model derived estimates (black circles) and associated standard errors of the spearman rank correlation of species-species annual population indices for models that tested for a difference among groups (M, migrant; R, resident), for (A) a model that looked at all species-species comparisons across the 14 migrant and resident species (M-M: n = 10; M-R, n = 45; R-R, n = 36), and (B) a model that included only species-species comparisons that included the monarch butterfly (M-M, n = 4; M-R, n = 7). In both cases, the raw spearman correlation coefficient for each species-species comparison is also plotted for each group (open circles).

models that tested for a difference among groups, for (A) a model that looked at all species-species comparisons across the 14 migrant (M) and resident (R) species (M-M: n = 10; M-R, n = 45; R-R, n = 36), and (B) a model that included only species-species comparisons that included the monarch butterfly (M-M, n = 4; M-R, n = 7). In both cases, the proportion of corresponding peaks for each species-species comparison is also plotted for each group (open circles).

strongly limited by factors acting during spring migration and recolonization given contemporary levels of milkweed availability in the US (see Thogmartin et al., 2017c). The lack of synchrony between annual numbers of monarchs and resident co-occurring butterfly species suggests that the size of the monarch breeding population is not driven by factors acting during the breeding season, such as direct and indirect effects of weather (e.g., effects on nectar availability or milkweed quality). Instead, interannual synchrony of all migratory butterfly species breeding in southern Canada over the 15 year time series examined in this study suggests that factors acting during spring migration and recolonization are driving observed patterns in the year-to-year variability in monarch annual population indices.

Weak correspondence between the size of the overwintering population and subsequent southern Canada breeding population did not support the hypothesis that the size of the Ontario breeding population is predominantly limited by factors acting on the overwintering grounds. This hypothesis is also not supported by our result that monarch numbers fluctuate in sync with all long distance migrant butterflies on their southern Canada breeding grounds, but not with 9 common co-occurring resident species of butterflies that breed in the same habitats as monarchs in southern Canada. Therefore, the relationship between overwintering and subsequent breeding population sizes likely breaks down as a result of variation in reproductive potential with weather conditions during spring migration and recolonization (Zipkin et al., 2012; Saunders et al., 2016, 2018).

The correspondence of monarch (and other migrant) annual indices with the mean number of 21◦ days during spring migration is in agreement with Thogmartin et al. (2017b) and suggests that given favorable breeding conditions, monarchs have the potential to recover from low winter population sizes.

Indeed, while the pattern of ups and downs were strongly correlated between the breeding and following winter population estimates, correspondence was poor when the time series were not de-trended; i.e., the strong decline in overwintering population size, observed prior to more recent increases since population lows in 2012–2014, was not reflected in the breeding population. However, Crewe and McCracken (2015) found that the magnitude of change between peaks and lows in the number of monarchs migrating out of Ontario each fall has declined over time, which might be indicative of a reduction in reproductive and re-colonization potential. If so, the monarch population may become more susceptible to further declines over time, particularly if spring weather conditions become increasingly warm with climate change (e.g., Schwartz and Reiter, 2000; Zipkin et al., 2012; Saunders et al., 2016, 2018).

Strong correspondence between breeding and following winter population sizes did not support the hypothesis that the current breeding population of monarchs in Canada is predominantly limited by mortality during fall migration (Badgett and Davis, 2015; Ries et al., 2015; Inamine et al., 2016). Also, observed shifts in the phenology of monarch butterflies, such that monarchs are staying in southern Canada later in the year (Prudic et al., 2017, e.g., Zipf et al., 2017), could have an impact on fall migration mortality, if later migration results in a higher probability of encountering extreme weather events on the breeding grounds or during migration. Research is required to determine whether a shifting phenology on the breeding grounds is resulting in additional breeding generations each year or delayed migration, and whether individuals delaying their migration, or a new late summer or early fall generation, are likely to survive a later fall migration.

Migratory monarch butterfly populations are currently extremely low and vulnerable (Semmens et al., 2016). Loss of milkweed breeding habitat as a result of glyphosate application to genetically modified crops is thought to be largely responsible for current population levels (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Stenoien et al., 2016; Pleasants, 2017). Historical forest loss on the wintering grounds in Mexico due to logging has also been hypothesized to affect population size and has been responsible for the disappearance of a wintering colony (Brower et al., 2002; Vidal et al., 2014). Further, nectar resource availability has likely declined through time (Potts et al., 2010), potentially limiting the availability of nectar resources for monarchs to build fat reserves late in the fall migration in the southern US and Mexico, which are required to sustain them throughout the winter (Brower et al., 2006). While our study suggests that fluctuations in the number of breeding monarchs counted in southern Canada since 2003 are driven by factors acting during spring migration and recolonization, our results do not imply that factors acting during alternative phases of the annual cycle, including the availability of milkweed during the summer breeding season, nectar resources during fall migration, and forest habitat during the overwintering period, have not contributed to historical rates of population decline, or are not limiting the potential of monarchs to recover to previous population levels. Further research is needed on historical and contemporary milkweed availability in Canada to better understand how land management practices affect current breeding populations of monarchs and other pollinators. Research should also address whether increasing rates and severity of extreme weather events in the fall since the turn of the 21st century, such as hurricanes and extended drought periods, are influencing monarch survival during fall migration. Finally, while overwintering forest loss has sharply declined since 2008 (Vidal et al., 2014), increasing overwintering habitat could help increase population numbers observed in Canada, particularly in years where growing conditions for milkweed in the southern US are favorable. It should also be noted that severe weather events on the wintering grounds, which were not explicitly accounted for in our analysis, have the potential to limit next summer breeding success in northeastern North-America in any given year. For example, large die-offs of monarchs on the wintering grounds owing to severe storm events (e.g., Brower et al., 2017) might affect the number of fall migrants returning to Mexico the following winter. We suggest that further research is needed on the interactions between events on the wintering grounds and the subsequent spring migration and recolonization to fully understand contemporary population limiting processes of the migratory monarch population.

In conclusion, we show that our list-length approach using data that was largely collected by citizen scientists to develop population indices for 14 butterfly species breeding in Canada, speaks to the value of community science data and butterfly

FIGURE 6 | Detrended annual indices for breeding monarch butterflies and detrended mean number of 21◦ days in April in the lower United States, by year and, inset, showing linear relationship between detrended monarch and detrended mean April 21◦ day indices. A significant Spearman correlation of 0.71 (<sup>p</sup> <sup>=</sup> 0.01) was detected.

mid-eastern United States. Significant correlations between detrended breeding annual indices and detrended mean 21◦ days were not detected, though proportion of corresponding maxima and minima ranged between 0.60 and 0.82.

checklists and suggests our approach could easily be extended to other parts of the monarch range. Our community level analysis which allowed us to compare inter-annual population indices of several co-occurring migratory and resident butterfly species in southern Canada also revealed for the first time that contemporary populations of monarchs in Canada are most strongly limited by weather events occurring in the US during spring, a result supported by multiple analyses in the US (Zipkin et al., 2012; Saunders et al., 2016, 2018). Moving forward, it will be important to understand how conservation actions taken by all three countries throughout the annual cycle of the monarch interact with stochastic weather effects, and where those interactions are strongest, so that conservation planners can better prioritize where remedial actions may be most effective.

### AUTHOR CONTRIBUTIONS

ML, GM, and TC conceived and designed the experiments. ML, TC, and GM contributed data and analysis tools. TC performed the experiments and analyzed the data. TC wrote the manuscript, with input from GM and ML.

#### REFERENCES


# FUNDING

Funding for this work was provided by Environment and Climate Change Canada.

#### ACKNOWLEDGMENTS

We are grateful to the many individuals whose butterfly observations are incorporated into the Ontario Butterfly Atlas each year.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00308/full#supplementary-material


of monarch butterflies overwintering in Central Mexico. PeerJ. 5:e3221. doi: 10.7717/peerj.3221


**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 Crewe, Mitchell and Larrivée. 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.

# Monarch Butterfly Conservation Through the Social Lens: Eliciting Public Preferences for Management Strategies Across Transboundary Nations

Rodrigo Solis-Sosa<sup>1</sup> \*, Christina A. D. Semeniuk <sup>2</sup> , Sergio Fernandez-Lozada<sup>1</sup> , Kornelia Dabrowska<sup>1</sup> , Sean Cox <sup>1</sup> and Wolfgang Haider 1†

*<sup>1</sup> School of Resource and Environmental Management, Simon Fraser University, Burnaby, BC, Canada, <sup>2</sup> Great Lakes Institute for Environmental Research, University of Windsor, Windsor, ON, Canada*

#### Edited by:

*Jay E. Diffendorfer, United States Geological Survey (USGS), United States*

#### Reviewed by:

*Paul Cross, Bangor University, United Kingdom Jose Roberto Soto, University of Arizona, United States*

> \*Correspondence: *Rodrigo Solis-Sosa rsolis@sfu.ca*

> > *† In memoriam*

#### Specialty section:

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

Received: *23 December 2018* Accepted: *06 August 2019* Published: *22 August 2019*

#### Citation:

*Solis-Sosa R, Semeniuk CAD, Fernandez-Lozada S, Dabrowska K, Cox S and Haider W (2019) Monarch Butterfly Conservation Through the Social Lens: Eliciting Public Preferences for Management Strategies Across Transboundary Nations. Front. Ecol. Evol. 7:316. doi: 10.3389/fevo.2019.00316* The monarch butterfly *(Danaus plexippus)*, an iconic species that migrates annually across North America, has steeply declined in numbers over the past decade. Across the species' range, public, private, and non-profit organizations aim to reverse the monarch decline by engaging in conservation activities such as habitat restoration, larvae monitoring, and butterfly tagging. Urban residents can actively participate in these activities, yet their contribution can also be realized as an electorate body able to influence the design of conservation programs according to their interests. Little is known, however about their preferences toward the objectives and design of international monarch conservation policies. In this paper, we investigate these preferences via a survey design using Discrete Choice Experiments (DCEs) and Latent Class Analysis (LC) of urban residents across the main eastern migratory flyway in Ontario, Canada, and the eastern United States. Attributes in the DCE included the size and trend of overwintering butterfly colonies, the type of institution leading the conservation program, international allocation of funds, and the percentage of funds dedicated to research. From the general populace, we isolated respondents already engaged in monarch conservation activities to explore how they compare. We sent a smaller set of surveys deliberately withholding the expected-success forecast of the monarch recovery program to assess the value of information for urban residents within a conservation context. The LC distinguished three groups of respondents among urban residents: (1) the main group, labeled "Eager," accounting for 72.4% of the sample, that showed a high potential for supporting conservation policies and had remarkable similarities with the monarch enthusiasts' sample; (2) a "Pro Nation" group (18.4%) marked by their increased willingness to support conservation initiatives solely focused within their country of residence; and (3) an "Opinionated" segment (9.23%), that was highly reactive to changes of the leading institution, resources allocation, and economic contribution proposed. Key findings from this research reveal that to maximize potential support amongst urban residents in the monarch's breeding range, a conservation strategy for the monarch butterfly should be led by not-for-profit organizations, should strive for transboundary cooperation, and should include the communication of anticipated ecological outcomes.

Keywords: monarch butterfly, citizen science, choice experiment, latent class, conservation, public preferences, international cooperation, transboundary conservation

#### INTRODUCTION

The design of conservation strategies for transboundary migratory species has proven to be a challenging topic for decision makers, partly due to the presence of multiple institutions, groups of interest, administrative barriers, and political and cultural differences (Grant and Quinn, 2007). The monarch butterfly (Danaus plexippus; henceforth referred to as "monarch") is a highly migratory and globally distributed butterfly species (Oberhauser et al., 2008). Its eastern North American population has the longest migration (Brower and Missrie, 1999)—up to 4,000 km—in which butterflies across the eastern states/provinces of the US and Canada establish overwintering colonies within a few specific forest patches in Mexico (Urquhart and Urquhart, 1976). The area occupied by monarchs in these overwintering sites has decreased from an average area of 5.71 ha in 1993 to an all-time-low area of 0.67 ha during the 2013–2014 season (Vidal and Rendón-Salinas, 2014). Its current estimate is at 6.05 ha (Rendón-Salinas et al., 2019).

Habitat destruction in both overwintering and breeding areas is currently the most plausible hypothesis for the population decline (Brower et al., 2012). Overwintering monarch colonies rely on the forest canopy for protection against freezing temperatures, precipitation, and wind (Anderson and Brower, 1996). During the breeding season, monarchs depend on milkweed (Asclepias spp) for larvae feeding across the breeding range from Northern Mexico to the northeastern US and eastern Canadian provinces (Zalucki et al., 2001). Here, agricultural land transformation combined with the introduction of transgenic-specific herbicides for crop management, to which only genetically modified crops can resist, have caused a general decline in milkweed abundance across the eastern states of the US over the last decade (Pleasants and Oberhauser, 2012).

In addition to those multiple stressors, the heterogeneous sociopolitical backdrop of the monarch's decline is a considerable challenge as well. Different resource-extraction activities, socioeconomic differences (Lopez-Hoffmann et al., 2009), and distinct legislative tools and processes for its protection (Waples et al., 2013) can hinder the effectiveness and coherence of joint strategies (Scott and Collins, 1997). Moreover, the limited resources available for conservation, from governments and NGOs alike, are allocated based on national priorities, which may significantly differ between countries. For example, while the monarch butterfly is a top priority for WWF-Mexico with more than 25 scientific monarch-related reports (WWF- Mexico, 2018), its Canadian office 2017 annual report has no mention of the monarch butterfly (Miller, 2017). Additionally, since political institutions tend to be responsible for internalizing environmental externalities, with their mandates focusing on local issues, externalities at an international level are frequently overlooked (Perrings and Halkos, 2012). One example of such an externality is the potential loss of revenue that Mexican communities incur from monarch-reserve tourism caused by extensive use of herbicides in the North (Esquivel-Rios et al., 2014).

Despite this intricate sociopolitical backdrop, the monarch's decline and its widespread appeal have spurred people's interest in its protection across the migratory flyway. For example, The Monarch Waystation program, an initiative seeking to stimulate the public to provide habitat for monarchs and other pollinators, is continually increasing its presence every year, with 21,946 registered waystations up to date (Lovett, 2018). Likewise, Journey North, an entry-level citizen science platform, received 1,574 reports of egg sightings and 14,381 adult sightings during fall 2017, contrasted with 193 eggs and 3,310 adults reported in 2012. Several other citizen science hubs have witnessed that same surge of interest by the general public such as the Monarch Watch Tagging Program, eButterfly, iNaturalist, and the Monarch Larvae Monitoring Program. Such participation of private residents in citizen science and ecologically-related activities provide scientists with an extraordinary capacity of having useful, cost-effective data collected and analyzed. Ries and Oberhauser (2015) estimated that 17% of 503 papers related to the monarch since 1940 have relied on citizen science data. Moreover, Lewandowski and Oberhauser (2017) found that individuals engaged in citizen science activities are more likely to provide and protect critical habitat as well.

However, the role of the general public in protecting the monarch, as well as any other imperiled species, can go beyond data gathering and habitat provision—at which farmers could be substantially more effective (Thogmartin et al., 2017). Instead, when a conservation target is embedded within a complex network of economic and cultural interests at a transboundary level as described above, the involvement of the general public is especially needed. Agnone (2007) studied how the general public's opinion and protests have impacted the passage of environmental laws in the United States between 1960 and 1998. Several national conservation policies have been successful when the public is engaged. For example, Lutrin and Settle (1975) documented the success of passing California's Coastal Zone Conservation Act due to the active engagement of the Coastal

**Abbreviations:** ASC, Alternative Specific Constant; DCE, Discrete Choice Experiment; LC, Latent Class or Latent Class Analysis; MNL, Multinomial Logit Model; NEP, New Environmental Paradigm Scale; RI, Relative Importance; WTP, Willingness to Pay; mWTP, Marginal Willingness to Pay.

Alliance with the public and contrasted it with the failure on passing the Clean Air Initiative that same year due, most likely to a lack of public engagement. More recently, Nicholls (2017) documented the crucial role the general public had for the introduction of neonicotinoid regulations in Ontario. We expect that, just as in the national context, at the transboundary level, finding the specific levers to promote the support of the general public for conservation policies could strongly influence the design, reach, and ultimately, success of conservation policies.

The present paper seeks to improve the understanding of public preferences for transboundary conservation strategies for the monarch butterfly conservation. Its main goal is to determine whether inherent heterogeneity exists in public preferences for strategic-level characteristics of a recovery-management strategy that includes institutional leadership, international cooperation, and support for citizen science and research activities. It also aims to evidence the effect that providing a projection of the conservation program's success has on the overall willingness of people to support such a program. We used Discrete Choice Experiments (DCE) with a Latent Class Analysis (LC) to achieve such objectives.

#### METHODS

#### Sampling

The sampling area included the 35 eastern-most states of the US and Canada (Ontario), representing all areas where there is more than a 50% probability that monarch populations are present (Galindo-Leal et al., unpublished). Geographically speaking, the US-Mexican Border, parallel 49, meridian 102, and the east coast constitute the southern, northern, western and eastern limits of the sampling area, respectively (**Figure 1**). Based on the study objectives, we surveyed three different respondent samples: (1) main urban residents, (2) sample of monarch enthusiasts, and (3) modified urban residents' sample with a modified version of the survey to investigate the value of knowledge.

The surveying tool was delivered through the Survey Sampling International marketing company (Teel and Manfredo, 2010), targeting urban residents<sup>1</sup> within the sampling area. Respondents were obtained from the panellists' database of the marketing company and were contacted directly by them based on our target demographics via email with an invitation link. The survey was sent in batches of 100, and only after analyzing their geographic and gender representativeness, the next batch of surveys was sent adjusting the target demographics to obtain a representative sample.

An invitation link was also sent through the Monarch Watch DPLEX mailing list<sup>2</sup> , which contains subscribers, mostly citizen scientists, dedicated to the conservation of the monarch. This list is maintained by Monarch Watch, a non-profit organization hosted at the University of Kansas and dedicated to the monarch butterfly conservation (Lovett, 2018). We additionally isolated responses of individuals self-reported as engaged in monarch conservation activities from the main urban resident's sample and pooled their responses with the ones from the DPLEX mailing list to obtain a monarch enthusiasts' sample.

The use of an online survey through a marketing company, instead of a mailed or in-person survey, was due to the geographical and numeric extension of the sample. Online internet surveys have many other advantages, such as reduced cost and higher design flexibility. However, they also introduce new potential sources of bias that have to be accounted when analyzing the results such as a potential increase of self-selection processes (Olsen, 2009) and the risk of introducing "professional respondents" to the sample (Dennis, 2001).

The presence of "professional respondents" is one of the main risks associated with using marketing companies for delivering an online surveying tool. Such respondents tend to click through the survey without paying proper attention and potentially adding unwanted noise to the results (Dennis, 2001). To control for this, following Malhotra (2008), we removed individuals with a time-to-completion of two standard deviations away from the mean (individuals that averaged their responses in <5 s or above 22 min per question, including the choice experiment). This range was chosen since we could not find any evidence of primacy (Belson, 1966) or recency (Kalton et al., 1978) effects within those outliers. Most of the outliers did not complete the demographics section of the survey, for the few that did answer that section, we tested their demographics and attitudinal responses against the rest of the sample and did not find any significant differences (Malhotra, 2008). Lastly, an instructional manipulation check question was embedded (Oppenheimer et al., 2009) within the survey which read: "This question is intended to filter respondents that are not reading every question thoroughly. Please select the option 'Very Little' as your answer. Another question like this one will be placed further in the survey."

We sent out the survey to 5,750 people in Canada and the US from which we received 2,557 responses with an overall completion rate<sup>3</sup> of 40.13%. The main sample included 916 individuals from Canada and 943 from the US, from which 302 self-reported as being monarch enthusiasts. Twentynine additional surveys were obtained through the Monarch Watch mailing list. We pooled the monarch enthusiasts from those two samples into a Monarch enthusiasts' group of 331 respondents. Finally, we sent 1,104 surveys with a variation of the DCE obtaining 625 completed surveys for the modified urban resident's sample with a completion rate of 56.51%. Completion rates varied according to the source of the respondents (marketing company = 46.4%, Monarch watch = 66.26%), and their nationality (Canada = 38.49%, US = 49.4%). Although our sample was mostly similar to the demographics of the target population, there were some significant differences, e.g., respondents from the sample with no high school diploma were 30% whereas the target population was 52%. Such demographic differences may be an effect of the way we defined urban residents in comparison to how it is stated in the census data, also, being

<sup>1</sup>We defined as Urban Resident a person that does not derive their main source of income from agriculture and owns a non-rural postal code.

<sup>2</sup>The survey was sent as an open link; however, we did not observe any duplicate IP addresses in the responses.

<sup>3</sup>Here and elsewhere, completion rate is defined as the number of surveys filled out and submitted divided by the number of surveys started.

this the reason to not balance the sample with an iterative proportional fitting or other raking procedure (Kolenikov, 2014); nevertheless, broad generalities to the target population can still be inferred. Respondents' demographics from the main urban residents and monarch enthusiasts' samples are summarized in **Tables 1**, **2**, respectively.

#### Survey Overview

Choice-experiments data were collected using a web-based survey conducted during November 2016 across Canada (Ontario) and the eastern US. The survey consisted of the following sections: (1) assessment of the individual's knowledge about the monarch, (2) video introduction for the survey and essential terminology, (3) choice experiment, (4) followup questions, (5) demographics, and (6) New Environmental Paradigm Scale (NEP) Statements. The survey also included questions on the allocation of resources and level of involvement of different organizations, which were not analyzed here but will be revisited in subsequent manuscripts.

The survey design and delivery were developed following Salant and Dillman (1994) and Dillman et al. (2014) design principles. Before giving any information about the monarch, we elicited the individual's knowledge of the monarch through three Likert-scale questions: (1) awareness of the monarch's decline, (2) level of concern about the current monarch's situation, and (3) awareness of the importance of milkweed for the monarch's survival and conservation. A short introductory video (2:32 min) followed explaining the purpose of the survey, the current decline of the monarch's population, and the definition of each DCE attribute. We used a video instead of text to avoid cognitive fatigue and to ensure respondents had a better understanding of the survey elements (Mendelson et al., 2017). Although we were unable to confirm that all respondents watched the video, they were unable to skip forward through the video to continue with the survey before it ended. The full survey and the video can be found in the **Supplementary Material**.

A demographics section was included after the DCE and, finally, the respondent was presented with the NEP Scale for the assessment of their environmental attitudes (Dunlap, 2008). The NEP scale consists of 15 environmentally-related statements to which the respondent must choose their level of agreement/disagreement. The totalled result is a score between 0 and 150, where the higher the score, the more ecologically oriented the mindset of the respondent (Dunlap and Van Liere, 1978).

# Discrete Choice Experiment (DCE)

The DCE is a stated preference valuation method that forces the respondent to make trade-offs between distinct levels and attributes ideally resembling the context in which individuals make real-life decisions. The DCE assumes that respondents' decisions follow the Random Utility Model, which states that an individual will strive to maximize utility while making choices (Manski, 1977). Under this assumption, it is possible to estimate the proportion of the sample, market share, that would choose any given program configuration (Landauer et al., 2012). By including a contribution attribute, the marginal economic value



*"Obs" is the observed percentage from the sample and "Exp" is the expected percentage of people based on census data (Statistics Canada, 2016 Census of Population; https:// www.census.gov/). The definition of urban resident of the US Census and Statistics Canada differed from ours. While their definition is based on population density, presence on urban clusters or urbanized areas, land use, distance, and population thresholds, our definition is based on main source of income and zip code. Note that some items do not add up to the total sample size due to missing data from incomplete responses.*

of the attributes can be estimated as well (Kuhfeld, 2006). The ability to explore hypothetical non-existent scenarios is another advantage of this method (Vega and Alpízar, 2011).

A DCE consists of a list of key characteristics, or attributes, describing an alternative. Each of these attributes has different values, or levels, defining the configuration of that alternative. Several alternatives, 2 or 3 at a time, are presented at the same time to respondents in a choice set. Then, respondents are asked to analyze and choose their preferred one from each choice set (Louviere et al., 2000). An orthogonal experimental design ensures that each choice set is presented to respondents enough times, allowing researchers to estimate respondents' preferences for the attributes and all the levels that defined those alternatives.

The DCE estimates the utility<sup>4</sup> , or satisfaction that respondents derive from a choice, which, in this case, is a potential management scenario. Also, the DCE allows valuing not only the resource as a whole but also the incremental worth of its components—i.e., the marginal part-worth utility<sup>5</sup> of its attributes (Birol et al., 2006) and their Relative Importance, RI (Vermunt and Magidson, 2005). DCEs are commonly used to forecast likely changes in behavior as a reaction to changed circumstances or to the hypothetical availability of certain goods (Louviere et al., 2000). The utility estimates from the DCE represent the utility that a level or unit of an attribute provides. When the attribute is categorical, this is measured as utility relative to the mean of the other levels from the same attribute. When the attribute is numerical, the interpretation of utility is on a "per unit" basis. The RI of an attribute, also known as Relative Maximum Effect, is the proportion of the overall utility explained by a change of one unit of that attribute when numeric, or from the difference between the least and most preferred levels of that attribute when categorical (Crouch and Louviere, 2004; Casini et al., 2016). The higher the RI value of one attribute, the

<sup>4</sup>Utility is defined as the weight of outcomes in making a decision (Ariely et al., 2003). It can also be explained as the level of short-term happiness derived from a specific material or immaterial good (Kimball and Willis, 2006). DCEs quantify utility by a mean-centered dimensionless value representing the preference associated with a particular level of an attribute compared with the reference level.

<sup>5</sup>Marginal part-worth utility is a measure of welfare that the respondent derives from a one-unit increment (all else being equal) of one attribute from the choice set (Steinke and Van Etten, 2017).

#### TABLE 2 | Demographics from the monarch enthusiasts' sample (*n* = 331).


more such attribute influences the preference of the respondent (Crouch and Louviere, 2004; Casini et al., 2016).

By including a contribution attribute within the experimental design, it is also possible to estimate a marginal willingness to pay (WTP) for each attribute (Kerr and Sharp, 2009). Taking advantage of this possibility, the estimates reported within this paper are in USD value. These estimates reflect the economic value of changing any attribute by one unit while leaving the remaining attributes fixed. The WTP presented here is a marginal WTP estimate on a per-unit basis from the baseline, which is different from the total WTP provided by other methods such as Contingent Valuation (Diffendorfer et al., 2013). Instead, the marginal WTP provided here denotes the difference in the contribution that the respondent would be willing to pay from the unweighted average of all the levels, for categorical variables (Daly et al., 2016). For numerical variables, it describes the difference of the respondent's WTP to increase one unit of a particular attribute while leaving the rest of the attributes fixed (Kerr and Sharp, 2009). This manuscript explores for the first time the marginal WTP for the monarch conservation.

#### Discrete Choice Experiment Design

We constructed choice alternatives describing potential management scenarios for the conservation of the monarch using a list of attributes that described a hypothetical ecological status of the monarch, and the strategic-level characteristics of a proposed conservation initiative. These attributes were refined using input from interviews with academics with expertise in human dimensions, conservation biology, or both. The final alternatives were made up of nine<sup>6</sup> attributes, three of them as context and the other six as program attributes (**Table 3**). Values for the levels of each attribute were selected based on feedback from academics, two focus groups, and a pilot study with 200 respondents (100 Canadians and 100 US citizens) 2 months prior to the final version release. The first focus group (n = 8) consisted of experts on this method, with the primary objective of finding technical deficiencies. The second focus group (n = 13) was composed of graduate students of the authors' universities with differing levels of familiarity with choice experiments or the monarch and sought feedback about the size and complexity of the survey. Finally, the pilot study was directed to the same demographics as the target population of the main survey and sought to detect cognitive fatigue, such as positive WTP estimates, lack of

<sup>6</sup>The context attributes that appeared in the survey were "Trend" and "Area-Trend." The attribute "Area" did not appear in the survey, but it was used to calculate the "Area-Trend" attribute (which is an interaction between "Area" and "Trend"). Also, the "Payment Vehicle" and "Leader" are part of a single attribute in the experimental design but appear separately in the survey. See **Table 3** for details.

TABLE 3 | Attributes and levels used in the choice experiment exercise.


significance of the utility estimates, or extensive skipping of optional screens.

Each choice set (**Figure 2**) consisted of an ecological context scenario with three attributes, and three options: two alternative conservation programs, and one status quo option. Context attributes established the scenario under which the respondents would be making their choice (Tversky and Simonson, 1993; Haegeli et al., 2012). Here, the context attributes set a hypothetical situation of the overwintering colonies to investigate the change of respondents' preferences with the assumption that respondent preferences were context-dependent (Mazar et al., 2014). These context attributes remained the same for all options of the choice set and only changed between choice sets.

The program attributes included international allocation of funds, probability of success of the program, institution leading the program, monetary contribution to the described program, fund-raising mode, and percentage of funds dedicated to research and citizen science activities. These attributes varied their levels independently from each alternative so that the respondent could perceive a contrast between the options. The "status quo" option as a base alternative consisted of abstaining from contributing to any program and maintaining the current trend shown in the specific scenario. Most literature agrees that a base alternative has to be included to estimate the welfare change associated with the other alternatives (Bateman et al., 2004; Train, 2009). If the respondent chose the base alternative in any of the presented choice sets, they were asked to provide a rationale for their choice.

The experimental design for the main urban residents' survey was a 4<sup>6</sup> × 8 <sup>1</sup> orthogonal fractional factorial design with two of those factors entered as context variables. For the modified urban

resident's survey, a new design with the same characteristics was generated but with one factor removed (4<sup>5</sup> × 8 1 ). Both designs were generated with the SAS "%MktEx" Macro (Kuhfeld, 2001) and had a D-efficiency of 100% as a measure of the design's goodness (efficiency), and orthogonality (Kuhfeld et al., 1994).

#### Statistical Analyses

All the numeric levels were standardized and centered before analyzing the DCE model. The data were analyzed using conditional logit and latent class regression with Latent Gold 3.0 software (Vermunt and Magidson, 2005), obtaining Relative Importance (RI), latent class segmentation outputs, and model performance metrics.

Latent Class Analysis (LC) was used to identify and segment heterogeneity in utility estimates among urban residents. The LC assumes that the sample constitutes a finite number of groups of individuals, also known as classes, with relatively similar preferences within their group and considerably different from each other (Birol et al., 2006). Random Parameters Logit can also identify the heterogeneity of preferences within a sample (McConnell and Tseng, 1999); however, Random Parameters Logit elicits the individual differences amongst the sample rather than grouping them (as LC does). The latter scale of analysis is considered more convenient for the design of management strategies (Boxall and Adamowicz, 2002).

The non-significantly different attributes across classes in preliminary models were constrained to be the same across all classes to prioritize the delineation of classes by the most highly variable attributes (**Table 5**). That model restriction reduced the number of parameters and improved the fit of the model (Vermunt and Magidson, 2000).

Embedding a DCE within a comprehensive survey allows descriptive data, as covariates or predictors, to define individuals by linking these with their preferences. Covariates are a posteriori explanatory variables that describe class membership and can inform the policymaker about which demographic strata can be targeted with specific actions (Boxall and Adamowicz, 2002). Covariates included in the model were the pre-survey knowledge about milkweed and the monarch's status, whether the respondent was engaged in any ecological/citizen science activity, and the age group of the respondent.

Alternatively, predictors are characteristics of the choice replication or the person and have the same value across alternatives. Predictors are part of the regression model, just like attributes, and are therefore considered a priori explanatory variables (Vermunt and Magidson, 2005). As a result, covariates can predict class membership, whereas predictors contribute to its creation. Here, the model included the level of concern about the monarch's situation as a predictor.

For the three respondent samples (main urban residents, monarch enthusiasts, and the modified sample of urban residents), we also conducted a Multinomial Logit Model (MNL) analysis to obtain a one-class model for each. These types of models are suitable for observing the main trends of the sample without accounting for heterogeneity. The MNL was used to compare the three samples and qualitatively detect any differences between the general preferences of people engaged—or not—in ecological activities (urban residents vs. monarch enthusiast's sample), or between people provided with an expected probability of success—or not—of the proposed program (main urban residents' vs. modified urban residents' sample).

To control for the uneven spacing of some of the numericvariable attributes and to achieve more interpretable results, we linearized all our numeric attributes (Kohlhardt et al., 2018). All the categorical attributes were effects coded for the interpretation of their estimates (Daly et al., 2016). Numeric data were analyzed with one-way ANOVA and a post hoc Tukey's Honest Significant Tests. For categorical data, a Pearson's chi-squared test was used. All statistical treatments were done with JMP 13 (SAS Institute Inc, 2016), and R 3.51 (R Core Team, 2013) was used to plot the results.

# RESULTS

### Latent Class Analysis of Main Urban Residents

#### Description of Classes

Preliminary models with different number of classes, covariates, predictors, and constraints (**Table 4**) were defined and evaluated using Bayesian Information Criteria (Burnham and Anderson, 2004). We also built a preliminary 2-known-class model based on nationality, and no significant differences were found between the classes regarding their preferences for the attributes presented; we pooled the data as a result. The final model was a three-class model with significantly different preferences for the geographical allocation of the resources, sensitivity toward the allocation of funds across classes, and the Alternative Specific Constant (ASC), which can be described as the utility derived from selecting any choice different from the status quo without accounting for the specific levels of the rest of the attributes. Each class was labeled based on those differences as "Eager," "Pro-Nation," and "Opinionated." The final model had the "Leader," and "Area" attributes constrained between class "Eager" and "Pro-Nation," "Research" across "Eager" and "Opinionated," and "Trend" across the three classes (**Table 5**). The "Eager" group was the largest, making up 72.4% of the overall sample. The "Pro-Nation" class was second in size (18.37%) and "Opinionated" was the smallest (9.23%)<sup>7</sup> .

Individuals from the class "Eager" showed a large estimate for the ASC, which represents a strong motivation to support conservation initiatives regardless of the configuration of the choice set (**Table 6**). In contrast, the other two classes denoted an unwillingness to participate in any management program. People from the "Pro-Nation" class strongly based their decisions on the allocation of funds across countries. When the choice task indicated that the allocation of funds would favor the respondent's country of residence, their utility markedly rose. In contrast, when funds were allocated only to Mexico or to the "other country," i.e., the US for Canadians, or Canada for US citizens, their utility considerably decreased in comparison to the other two classes. This class had a difference between the highest and lowest valued estimates 34.4% larger than that of "Eager." Finally, the third and smallest class was labeled "Opinionated" due to the large estimates associated with the leading institution, resources allocation, and especially the economic contribution. This class also had the most negative ASC, implying that they are the most reluctant to participate in any management program.

Respondents in the "Eager" group displayed the highest NEP score, indicating that these individuals possess largely pro-environmental attitudes. They tended to be younger and had a higher level of education, where 82.6% obtained at least a bachelor's degree, furthermore, 17.1% had a graduate certificate. Their income level was also higher than the other two classes, where 62.4% of the group earned at least \$50,000 per annum and also had the largest household size. The "Eager" class had the most considerable share of people contributing to ecologically oriented NGOs and actively participating in ecological conservation meetings, protests, and lectures. However, 58.5% of the people participating in those activities did not contribute economically to any ecologically oriented NGO (**Tables 7**, **8**).

The "Pro-Nation" and "Opinionated" classes were similar in attitudinal preferences and demographics, except in the percentage of individuals contributing to environmentally related activities and in age. Also, a higher proportion of the "Pro-Nation" class contributed to ecologically oriented organizations in comparison with people from the "Opinionated" class.

Only the level of concern about the monarch situation was included as a predictor of choice in the definition of the model as it significantly improved model fit. The overall utility estimates for "Eager" and "Pro-Nation," which add up to 91% of the overall sample, were positively affected when respondents had a higher level of concern about the monarch's situation. The reaction of "Opinionated" was counterintuitive, where its overall utility was negatively affected by an increase in their level of concern.

#### Context Attributes' Estimates

Further interpretation of the classes can be made by considering the attributes themselves and their levels (for a full list of estimates refer to **Table 6** and **Figure 3**). Respondents reacted to the percentage change of the overwintering monarch colonies' size over the last 5 years, in relation to the current area, similarly negative across the three classes, and all respondents' interest in supporting management programs decreased when the monarch population trend increased.

For the current area of the overwintering colonies, the "Eager" and "Pro-Nation" classes reacted similarly. They both were significantly affected negatively by the increase of the area of the overwintering colonies, i.e., their interest in supporting management programs decreased when the current colony population was higher. For the "Opinionated" class, we found the opposite effect. All the "Area" estimates were significant only at the 10% level.

<sup>7</sup>A LC provides the posterior probability that an individual belongs to a certain class (McCutcheon, 1987). We assumed that the class membership of a respondent was dictated by the class that gave them the highest posterior probability (Pacifico and Yoo, 2013).

TABLE 4 | Model selection for the main urban resident sample (*n* = 1,859).


*The base models have no restrictions, whereas subsequent models (with the same number of classes) are variations of that first model with different combinations of constraints, covariates, and predictors. Model selection was based on the best (lowest) BIC and smaller classification error (Class. Err).* \*\*\**1% significance level with two-tailed tests.*

TABLE 5 | Definition of constraints for the 3-latent class model of the main urban resident's sample (*n* = 1,859).


*Classes with similar preferences on preliminary models for a particular attribute were assumed to be the same in the final model, so other attributes with higher variance could drive the splitting of classes. Classes with the same letter denote that they have the same estimate for that specific attribute.*

As described in **Table 3**, the "Change" attribute was an interaction attribute between the overwintering colonies' Trend and Area. Respondents from the "Eager" class derived a positive utility from this attribute, i.e., the more substantial the increase, the higher the interest in supporting management programs. "Pro-Nation" respondents derived a negative utility, and "Opinionated" respondents were not significantly affected by this attribute.

#### Program Attributes' Estimates

The estimates for the institution leading the program were equal across "Eager" and "Pro-Nation." For these two classes, International NGOs and Educational institutions were significantly positive. Alternatively, "Opinionated" respondents showed a preference for local NGOs as leaders of the program. In all cases, the least preferred leading institution was the federal government.

When the allocation of resources was distributed to the respondent's own country, the utility estimates were the highest for the "Pro-Nation" class. The utility of the "Pro-Nation" and "Eager" classes became negative when either Mexico or the counterpart North American country were the receivers of those resources. Respondents from the "Opinionated" class were only significantly negatively affected when the counterpart country was the beneficiary of the resources. When the resources were distributed equitably across the three countries, the attribute's estimates were the highest for the "Eager" and "Opinionated" classes.

Regarding the percentage of funds dedicated to research and citizen science activities, the utility was similarly negative across the "Eager" and "Opinionated" classes and not significant for "Pro-Nation." For the probability of reaching the conservation goal of a minimum size of 6 ha for the overwintering colonies in 10 years, the utility estimates for "Eager" and "Pro-Nation" were significant and positive but being the first double than the latter; "Opinionated" had no significant preferences.

Finally, the attribute asking for the amount of money that respondents would be willing to donate for supporting the selected management strategy was negative and highly significant for all three classes. However, the "Opinionated" class estimate was almost double than that of "Pro-Nation" and almost 10-fold than that of "Eager" respondents.

#### Monarch Enthusiast's Estimates

The monarch enthusiasts sample (n = 331) consisted of individuals from the main urban residents' sample that selfreported as being monarch enthusiasts, and people from the DPLEX Monarch Watch mailing list. The primary objective of this sample was to identify differences between this group TABLE 6 | Latent class (3 classes) estimates and Marginal Willingness to Pay (mWTP) for the main urban residents' sample.


Monarch Conservation Through the Social Lens

*See text for the definition of Relative Importance (RI).* \*\*\**1% significance level,* \*\**5% significance level,* \**10% significance level with two-tailed tests. The Attribute "Area-Trend" is an interaction attribute between "Area" and "Trend".*


*We tested the differences between classes with a one-way ANOVA test and post hoc Tukey's test and the class "Eager" was significantly different (P* < *0.001) to the other two classes, which were no significantly different between each other.*



\*\*\**1% significance level,* \*\**5% significance level.*

and the main urban residents' sample. For this sample, the estimates obtained from the MNL (**Table 9**) closely resembled the estimates from the "Eager" class of the main urban residents' sample with the following exceptions: this sample showed a positive utility for the type of institution leading the program only when it was an educational institution. The remaining levels did not significantly affect the monarch enthusiasts' choice, unlike "Eager" respondents that had significantly positive estimates for both international NGOs and educational institutions. Also, while in the main urban residents' sample each of the classes had significant estimates for at least one of the context attributes ("Area," "Trend," or "Area-Trend"), the monarch enthusiasts did not exhibit significant preferences for any of them. Lastly, the estimate for the monetary contribution to support the program was negative (just as with the main urban residents' sample), but the value of the attribute was noticeably smaller in magnitude.

The ANOVA test shows that the demographics of this sample were significantly different from the main sample and each one of the three classes. A more substantial proportion of monarch enthusiasts were engaged in ecologically-related activities (p < 0.001) as well as the percentage of them who contributed to ecologically-oriented NGOs (p < 0.001). The percentage of enthusiasts that were Canadian was significantly lower than the share of Canadians from the urban residents' sample (p < 0.001). Respondents from the monarch enthusiasts' sample also had a higher level of education (p < 0.001), although the income level was not significantly different. Unlike the main sample that had more females than males, the citizen scientists' sample had a significantly higher proportion of males (p = 0.007). Finally, the average age of the enthusiasts' sample averaged significantly (p < 0.001) lower than the urban resident's sample (**Table 2**).

#### Modified Urban Resident's Estimates (Success Omitted)

The attribute most influenced by the inclusion/exclusion of a success probability was the percentage of resources dedicated to research. When included, the utility estimate of contributing funds to research was negative, i.e., respondents from the main urban sample were less willing to provide funds toward research when the program specified an expected success. Conversely, with the removal of this attribute, the estimate for research became positive; i.e., contribution-support increased in the absence of knowing success. However, amongst the respondents from the modified sample, the ASC value was negative, denoting a decrease of willingness to support conservation measures overall (**Table 9** and **Figure 4**).

#### Willingness to Pay

The marginal willingness to pay (mWTP) for each of the attributes was calculated and is shown in **Table 6**. The mWTP is defined as the difference in the contribution that the respondent would be willing to pay from the mean of all the levels, for categorical variables (Daly et al., 2016) and the difference of the respondent's WTP to increase one unit of a particular attribute while leaving the rest of the attributes fixed (Kerr and Sharp, 2009). Finally, the total WTP to support a conservation program for the monarch was also estimated. The WTP was contingent on the configuration of the program<sup>8</sup> and followed the utility estimates described in previous sections and, based on the

<sup>8</sup>The configuration of the "Best program" was defined as a program with the levels that obtained the higher utility estimate for each of the categorical attributes, with 90% success, and 20% of funds dedicated to research. Conversely the "Worst program" used the levels with lower utility, had 70% success, and also dedicated 20% of funds to research.

current area and trend of the overwintering colonies (Rendón-Salinas et al., 2019), it ranged between \$100.41 and \$141.01 for the worst and best program configurations, respectively. When analyzing each of the classes, the average WTP was \$161.76, \$76.85, and \$-5.04 for the classes "Eager," "Pro-Nation," and "Opinionated," respectively.

# DISCUSSION

The monarch butterfly is an iconic species for people from the US, Canada, and Mexico alike (Guiney and Oberhauser, 2008). As such, its conservation provides an excellent opportunity to find common points of interest and strengthen, or create, institutions of tri-national cooperation for the recovery of the monarch and other transboundary migratory species as well (Lopez-Hoffmann et al., 2009). Moreover, the monarch's plight has mobilized a considerable number of urban residents across the three countries to participate in habitat restoration and citizen science efforts to protect it (Ries and Oberhauser, 2015). The role of small habitat providers and citizen scientists that urban residents play in this context also extend to conservation-policy support. Conservation practitioners should strive to find the most effective ways to funnel this potential capacity, with that objective, and this study aimed to determine urban-resident preferences toward strategiclevel characteristics of a management strategy for monarch conservation that would generate the highest amount of support from urban residents.

We found that people across the main eastern breeding range of the monarch, represented by the eastern United States and the province of Ontario, share preferences concerning their inclination for non-governmental leadership in conservation programs, and joint international cooperation. Nonetheless, within-respondent sample heterogeneity was identified. Additionally, people currently engaged and non-engaged in ecological activities had marked differences over the identity of leaders of a conservation program, as well with their sensitivity toward ecological issues. Lastly, the knowledge about the success of a conservation program proved to also play an influential role in guiding people's preferences, albeit we acknowledge the challenge in ascribing a probability of success for conservation actions. All these findings, discussed below, have direct and relevant policy implications that can affect the adoption and support of conservation programs for the monarch and other migrating North American species.

#### Institutional Leadership

There was a clear tendency across the three classes for choosing any other alternative as a leader before the federal government. Previous research directly compared people's perception about different types of institutions spearheading conservation programs (Wells, 1998), exploring the distrust of people toward the federal government in the United States (Brook et al., 2003), Canada (Parkins et al., 2017), and elsewhere (Chen and Hua, 2015) within a conservation context. A common finding was that distrust was mainly credited to the perception of a lack of accountability and effectiveness with regards to the exercise of conservation funds by the government (Chen and Hua, 2015). Similarly, studies have found distrust with non-government organizations as well, mainly due to



\*\*\**1% significance level,* \**10% significance level with two-tailed tests.*

discrepancies between their mission statements and on-theground actions, combined with the perception of being profitdriven organizations (Arenas et al., 2009). As such, the sense of trust, respect, and credit people have for conservation institutions, whether NGO or government-related, can vary widely (Jepson, 2005). However, there is a general trend of respondents preferring NGOs and educational institutions over the federal government as leaders of monarch conservation programs. Considering that urban residents are a substantial majority in Canada and the US, and reflected in the present study, we concur with the recommendations of Amano et al. (2018) on effective governance. Specifically, that governments should continue to decentralize their decision-making and community engagement processes while also encouraging broader and more coordinated participation of non-government actors in the conservation of the monarch and other species across North America.

The preference for NGO leadership within the monarch conservation context may be explained by the extensive and meaningful contributions of NGOs across the overwintering sites (Carlos Galindo-Leal, 2005; Oberhauser et al., 2008; Valera-Bermejo, 2009; Solís, 2012), migratory flyway (Urquhart and Urquhart, 1976), and breeding grounds (Ries and Oberhauser, 2015). While further research would be needed to verify the awareness of monarch-related NGOs amongst urban residents, NGOs dedicated to monarch conservation have provided valuable opportunities for public engagement through citizen science activities, although at a smaller scale educational institutions, zoos and aquariums, and governments, work with citizen scientists as well. Indeed, 17% of 503 monarchrelated research published over the last 74 years has relied to a certain extent on citizen science (Ries and Oberhauser, 2015). Our findings, along with similar outcomes in birds (Horns et al., 2018) and pollinators (Kleinke et al., 2018) suggest that the engagement practices of monarch-related NGOs could serve as a template for other NGOs dedicated to other multinational conservation issues to foster trust and support in their fields.

#### International Implications

Most monarch research focuses on the overwintering sites in Mexico and the breeding grounds across the mid-west of the US. Although those are considered the most sensitive areas of the migratory cycle (Flockhart et al., 2015), the northern range of the migratory flyway also plays an important role, especially when the mid-western states of the US have lost much breeding habitat (Pleasants and Oberhauser, 2012), and the Canadian sites are presumably increasing in relative habitat availability (Lemoine, 2015). Furthermore, the northern range may become crucial with the potential northward range shift in light of climate change (Batalden et al., 2007).

The success of transboundary conservation programs increases in difficulty depending on the amount of sociocultural differences between the parties involved (Kark et al., 2015). As such, it is crucial to document whether Canadians react to management strategies the same way as US citizens do, which had not been explicitly examined until now. Previous research shows considerable differences between Canadians and US citizens regarding their interaction with the environment (Leech et al., 2002) and their attitudes toward environmental investment (Lachapelle et al., 2012). However, at a finer grain of analysis, the heterogeneity of preferences, common to each country, make it very difficult to assume different attitudinal trends for Canadians and US citizens (Alston et al., 1996). Similar heterogeneity was found in the preferences across the two countries and revealed demographic and attitudinal variables such as age, level of education, and income could explain such heterogeneity better than nationality does. This finding will be essential to consider, not only for the design of new management strategies for the monarch, and presumably other North American transboundary migratory species, but also can help facilitate international institutions to improve their coordination efforts between their national offices.

The preferences for the attributes presented in the choice experiment between Canadian and US citizens yielded no significant differences, which anticipates a positive outcome for the design and success of transnational conservation strategies for the monarch. However, it is essential to note the absence of Mexico in this study, which should be the next stage of analysis. We acknowledge the presence of international institutions currently working in the monarch conservation context, such as the Commission for Environmental Cooperation (CEC), but further involvement is needed from governments, NGOs, and academia to promote efforts at the international scale. The relevance of the results presented here, aside from contributing to the available knowledge of Canadian/US behavioral traits, validates previous monarch research that assumes that preferences of Canadians and US citizens are similar (Flockhart et al., 2015; Oberhauser et al., 2017).

Conservation of transboundary migratory species requires not only the understanding of preference heterogeneity of the multiple actors involved but also needs to achieve cooperation amongst those actors to attain a common goal (Kark et al., 2015). Possible avenues for achieving such agreement were explored here by eliciting the respondent's preferences for the allocation of conservation funds either nationally or internationally. The two largest classes, accounting for 81.62% of the sample, derived almost twice the utility when the conservation funds were distributed across the three countries in comparison to when the funds stayed local. Such predilection for international allocation of funds is contrary to a case in foreign aid where the utility tended to be higher when a proposed program would fund local efforts (Okten and Osili, 2007). The social construct<sup>9</sup> that the monarch has become might well explain this discrepancy (Gustafsson et al., 2015), which has mobilized international conversations and policy development (Gustafsson et al., 2015). In light of these findings, the monarch's plight can be used to catapult it as a flagship species for other conservation efforts of migratory pollinator species in peril throughout North America, by designing multi-species conservation strategies for the protection of shared habitat as well as to provide nectar sources for many pollinator types across their range at the appropriate times (Guiney and Oberhauser, 2008).

# Citizen Science and Public Engagement

The demographic, lifestyle, and attitudinal variables describing each of the classes provide insights into the willingness of people to participate in conservation programs. Individuals from the main residents' urban sample that self-reported as participants of conservation efforts had a higher sensitivity to environmental topics overall and were more likely to invest their resources in conservation efforts. Johnson et al. (2014) explain that these highly motivated individuals tend to turn into skilled leaders, transmitting skills and motivations to the rest of their social network. Congruently, people identified here to be already engaged in citizen science and environmental activities had a smaller utility overall for the economic contribution to the selected program in comparison with people not engaged in conservation. This finding suggests that ecologically engaged people "suffer" less for every dollar they invest in conservation. Interestingly, 60.6% of monarch enthusiasts reported not contributing economically to any environmental organization, implying that a lack of monetary contribution does

<sup>9</sup>Virtue ascribed to a subject by the general public (Czech et al., 1998).

not necessarily mean a lack of interest or absence of participation via other means. Therefore, providing opportunities to capture those types of non-monetary contributions such as community engagement, citizen science activities, and lobbying, may provide significant momentum to environmental causes.

When asked about funds dedicated to research and citizen science activities, this attribute had a negative estimate for monarch enthusiasts (indeed, for all respondents). This trait along with low estimates for the economic contribution for the selected program and high values for supporting monarch conservation in general, suggests that monarch enthusiasts are not resource-driven individuals, and place a high value on active participation instead of a monetary donation. When comparing the demographics of both, the main urban residents and monarch enthusiasts' samples, the latter tended to be from a higher income level, which could also help explain why citizen scientists are less motivated in their monetary preferences. This result is an example of income effect, a change in demand of a good or service in relation to a modification of an individual's income (Horowitz and McConnell, 2003), which has proven to be more than just an artifact from the valuation method (Roy et al., 1990), and can have important implications for designing a public engagement strategy (Hardy, 2013). For example, if highincome areas are almost self-driven toward ecologically-related activities, a certain proportion of economic resources invested could be diverted into low-income areas without losing too much participation. At the same time, this could provide broader support for conservation policies from other demographics more sensitive to financial incentives, e.g., low-income strata, farmers, other countries, and demographics that would be more sensitive to modifying their preferences with financial incentives such as participation rebates.

### Value of Knowledge

Participatory approaches for conservation have increased over the last few decades (Fritsch and Newig, 2012), not only as a data-gathering tool but to acknowledge the importance that communities have within the conservation dialogue (Roberts and Jones, 2013). All else being equal, a program that engages and informs the community will have higher chances of success than a program that does not follow this path (Andrade and Rhodes, 2012). Here, we explored two vital elements of the most basic level of community knowledge: sharing a forecast of a program's success, and level of concern about the current situation of the monarch.

Community-based conservation is a viable method for bridging sociopolitical barriers for transboundary conservation (Berkes, 2007) but can have considerable struggle in achieving the involvement of the community. In particular, behavioral engagement (Sutton and Tobin, 2011) can be constrained by a lack of knowledge, in addition to other factors such as other competing priorities, and a lack of enabling initiatives (Lorenzoni et al., 2007). Here, we tested the effect of knowing the success of a program on the willingness to support monarch recovery. Firstly, we did not find any evidence of overshadowing (Huber, 1997) due to the high similarities among the estimates for most of the attributes between the two resident samples, particularly the sign of the estimates, and the relatively low RI estimates of this attribute from the main urban resident's sample.

The differences that did arise are, arguably, explained by factors unrelated to overshadowing. Overall, we detected that the sample without knowledge about the probability of success of the program showed a smaller willingness to support conservation measures in comparison to the one that was informed about the level of success. By telling the respondent about the expected success of the conservation program, a considerable objective constraint was presumably abated, motivating the increased support for the conservation program. Although we are cautious about the impacts of this finding given the difficulty in providing a reliable expected success estimate for conservation actions, we recommend that institutions should strive to synthesize available knowledge in a systematic, rational, and transparent way (Addison et al., 2013). Moreover, they must acknowledge the inherent uncertainties in their work to provide the relevant information necessary to aid the decision-making process (Peterson et al., 2003).

Furthermore, our research demonstrated, in support of findings from Best (2010), that the respondent's level of concern about the current status of the monarch strongly influenced the respondent's level of support for conservation actions. When respondents were aware of the current situation of the monarch and were concerned about it, they showed an increase in their willingness to support monarch conservation. Taken together, these utility shifts in relation to the amount of information provided is termed "information as a commodity" (Bucy, 2002), meaning people tend to place a significant value on being informed about the expected success of their decision making (Herian et al., 2012), even if that information has a certain level of uncertainty given by a percentage probability of success. This finding underscores the need for organizations to increase the information they provide to the public. Indeed, the ecological and population models of the monarch developed by several research teams (Yakubu et al., 2004; Batalden, 2011; Flockhart et al., 2015; Oberhauser et al., 2017) are not only a tool for better decision-making (Schmolke et al., 2010), but can be used as a tool for community engagement, if properly broadcasted by the institution leading the program. Lockwood (2010) proposes transparency and accountability of a management program as keystone elements for the effective governance of protected areas, and arguably, we can generalize those results into broader conservation objectives not confined within the borders of a protected area such as is the case of the monarch. This reliance on transparency for improving the support of a conservation program was evident in our results as well. Moreover, we were able to demonstrate that if the community perceives an information deficiency about the expected success of the program, they are more likely to endorse the use of resources for funding that research. Further studies should focus on linking this kind of behavior with management, policy development, and public engagement implications.

#### Willingness to Pay

The WTP of a hypothetical conservation program is calculated by summing the utility derived from the levels that comprise the program's configuration and dividing it by the utility of the contribution attribute. Here, the WTP of the whole sample, estimated with the MNL, ranged between \$100.41 and \$141.01. Previously, Diffendorfer et al. (2013) estimated through a contingent valuation method a WTP per respondent ranging from \$53.89 to \$74.04. The difference between that study and our findings can be explained by a number of reasons. First, that study surveyed all U.S households whereas our study focused only in urban residents. Previous ecological studies have also found that respondents from rural areas have a lower WTP when compared to urban residents (Bandara and Tisdell, 2003). However, this should not be considered as indicative of a lower ecological interest from rural residents, rather it can be an evidence of an income effect (Train, 2009). Also, it is important to consider that the survey from Diffendorfer et al. (2013) was released in 2012, a time when most lay people were not aware about the role that milkweed had as a main driver of the monarch's plight.

# CONCLUSION

The results of this research provide significant findings for understanding not only the social system surrounding the monarch butterfly, but also the general trends in preferences for transboundary conservation. Policy-makers and program managers need to understand the motivations of urban residents for supporting conservation strategies, acknowledging them not only as resource users but as a dynamic part of the system that acts and reacts to the rest of the system's elements (Berkes, 2004). As a response to that need, the most significant conclusion of this research is that the bulk of society places a higher value on international programs led by NGOs for the conservation of the monarch, even though the allocation of resources would be split amongst the participant countries instead of staying in their own country.

Without diminishing the importance of local programs, an international coordination body can play a pivotal role in the monarch conservation. The CEC, the environmental branch of the North American Free Trade Agreement (NAFTA), facilitates collaboration and public participation to foster conservation, protection and enhancement of the monarch and several other North American migratory species. We recommend to continue with the coordination efforts of the CEC's "Science for Monarch Butterfly and Pollinator Conservation" project and to include a new objective into that program aimed to strengthen outreach campaigns for urban residents across the three countries. However, recent political unrest across North America, particularly the dissolving the NAFTA (Stevenson, 2018), calls for alternative institutions that could be a surrogate or partner for the CEC.

The need for alternative non-governmental institutions to support the CEC on its coordination responsibilities brings us to the next key finding of this research. We observed that all else equal, most respondents prefer an international nongovernmental organization to lead the monarch's conservation efforts. Currently, several organizations could serve this role. In the US, the Monarch Joint Venture has brought together a substantial number of institutions (government and nongovernment) proving to be an essential agent of change for US conservation policies (Oberhauser et al., 2015). However, the mandate of this coordinating body<sup>10</sup> bounds it to US-based institutions only and, unless a new mandate is created, it keeps it from scaling up to an international stage. An organization already participating at a worldwide-scale and playing a central role in conservation is the World Wildlife Fund which has been involved with the monarch butterfly almost since the discovery of the overwintering sites in Mexico (Brower and Missrie, 1999). Notwithstanding the vast contributions this institution has given to the conservation of the monarch, there are areas of opportunity that could increase its effectiveness, such as a higher involvement of the US and Canadian WWF offices. We, therefore, recommend improving the communication of these units, the same with other NGOs, and the coordination with other organizations alike.

Lastly, the strength of this study relies on its ability to be integrated with a population-ecology model of the monarch to create a coupled social-ecological system (CSES) model to increase the realism and applicability of the results. Within the context of natural resource management, previous empirical research has demonstrated the applicability and advantages of a CSES approach by incorporating societal responses as another dynamic element of the ecological system, e.g., Semeniuk et al. (2010) and Bodin et al. (2016), and is increasingly being evaluated as a useful transdisciplinary tool (Holzer et al., 2018). In the case of the monarch, such a coupled socio-ecological model can be used as a scenario forecasting tool for the design of conservation strategies (Peterson et al., 2003). By capitalizing on the support of urban residents for conservation initiatives, and additionally accounting for active participation of urban residents, citizen scientists, and other key stakeholders to increase habitat production, one could model the consequent impacts on monarch population and trends. That information and knowledge could then be used to feedback into a change of resident-level support dynamically; this is the focus of ongoing research.

# ETHICS STATEMENT

This study was carried out in accordance with the recommendations of Tri Council Policy Statement: Ethical Conduct of Research Involving Humans (TCPS 2) and the Simon Fraser University Ethics Research Board's Policy R20.01; with written informed consent from all subjects. All subjects gave

<sup>10</sup>The Monarch Joint Venture Website (https://monarchjointventure.org/aboutus; accessed on August, 2019) states that "Our mission is to protect monarchs and their migration by collaborating with partners to deliver habitat conservation, education, and science across the United States."

written informed consent in accordance with the Declaration of Helsinki. The protocol was approved by the Simon Fraser University's Research Ethics Board.

# AUTHOR CONTRIBUTIONS

RS-S, CS, and WH conceived the presented idea. RS-S, KD, CS, and SF-L designed the surveying tool. RS-S and KD generated the experimental design. RS-S analyzed the data and developed the models, with contributions from SF-L. Manuscript written by RS-S, revised by CS with contributions of SC.

#### FUNDING

The Social Sciences and Humanities Research Council (SSHRC) Insight Grant (No. 31-639962), the National Council of Science

# REFERENCES


and Technology (CONACyT), and the Simon Fraser University Open Access Fund provided the funding for this study.

#### ACKNOWLEDGMENTS

We want to thank Eduardo Rendon, Karen Oberhauser, and Scott Black for their valuable comments and feedback. Special thanks to Paulus Mau for programming and helping to design and implement the online survey. In loving memory of Wolfgang Haider.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00316/full#supplementary-material


Kerr, G. N., and Sharp, B. M. (2009). Efficient Design for Willingness to Pay in Choice Experiments: Evidence From the Field. Nelson: New Zealand Agricultural and Resource Economics Society.

Kimball, M., and Willis, R. (2006). Utility and Happiness. University of Michigan.


**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 Solis-Sosa, Semeniuk, Fernandez-Lozada, Dabrowska, Cox and Haider. 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.

# Monarch Habitat as a Component of Multifunctional Landscape Restoration Using Continuous Riparian Buffers

Darius Semmens\* and Zachary Ancona

*U.S. Geological Survey, Geosciences and Environmental Change Science Center, Denver, CO, United States*

Stabilizing the eastern, migratory population of monarch butterflies (*Danaus plexippus*) is expected to require substantial habitat restoration on agricultural land in their core breeding area, the Upper Midwestern United States. Previous research has considered the potential to utilize marginal land for this purpose because of its low productivity, erodible soils, and high nutrient input requirements. This strategy has strong potential for restoring milkweed (*Asclepias spp.*), but may be limited in terms of its ability to generate additional biophysical, and socioeconomic benefits for local communities. Here we explore the possibility of restoring milkweed via the creation of continuous riparian buffer strips around rivers and streams throughout the region. We use a GIS-based analysis to consider the potential of several different buffer-width scenarios to meet milkweed restoration targets. We further estimate the ability of these habitat areas to provide additional functionality in the form of crop pollination and water quality regulation across the entire region. Finally, we conduct a cost-benefit analysis comparing the conservative economic value of these ecosystem services with the lost value of crops for each scenario. Results suggest that riparian buffers could be used to meet 10–43% of the total milkweed restoration target of 1.3 billion new stems with moderate management. The value of water quality and pollination benefits provided by buffers is estimated to exceed costs only for our narrowest buffer scenario, with a cost-benefit ratio of 1:2. Larger buffer widths provide more milkweed, but costs to farmers exceed the benefits we were able to quantify. The restoration of narrow multifunctional riparian corridors thus has the potential to be a win-win scenario, adding milkweed stems while also providing a variety of other valuable benefits. This suggests the potential to leverage monarch habitat restoration efforts for the benefit of a wider variety of species and broader coalition of beneficiaries.

Keywords: ecosystem services, water quality, pollination, wild pollinators, geospatial analysis, monarch butterfly

# INTRODUCTION

The migration of monarch butterflies throughout eastern North America is celebrated across the continent, from festivals to back yards to school yards. Americans have expressed a one-time willingness to pay of US\$ 4.78–6.64 billion for monarch conservation via a national survey (Diffendorfer et al., 2014), and Mexicans and Canadians are willing to pay at the same rate, adjusted

#### Edited by:

*Ryan G. Drum, United States Fish and Wildlife Service (USFWS), United States*

#### Reviewed by:

*Steven Bradbury, Iowa State University, United States Douglas Landis, Michigan State University, United States*

> \*Correspondence: *Darius Semmens dsemmens@usgs.gov*

#### Specialty section:

*This article was submitted to Conservation, a section of the journal Frontiers in Environmental Science*

> Received: *08 March 2019* Accepted: *16 August 2019* Published: *30 August 2019*

#### Citation:

*Semmens D and Ancona Z (2019) Monarch Habitat as a Component of Multifunctional Landscape Restoration Using Continuous Riparian Buffers. Front. Environ. Sci. 7:126. doi: 10.3389/fenvs.2019.00126*

**257**

for income (Haefele et al., 2018). Despite their importance to people, however, the monarch population progressively declined over 2 decades to its lowest recorded level in 2014 (Vidal and Rendón-Salinas, 2014) and despite a subsequent rebound it remains at an elevated risk of extinction (Semmens et al., 2016). A population target of 6-ha occupied by overwintering monarchs in Mexico, the easiest way to monitor the size of this population, has been suggested as a near-term conservation goal (Pollinator Health Task Force, 2015), which would reduce the extinction risk over 10–20 years by more than 50% (Semmens et al., 2016).

Habitat loss, particularly the loss of milkweed species that developing monarch larvae require for food, due to changing agricultural practices in the U.S. is thought to be an important cause of monarch population declines (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Oberhauser et al., 2017; Pleasants, 2017; Saunders et al., 2017; Thogmartin et al., 2017b) among many other contributing factors (Ries et al., 2015; Inamine et al., 2016; Thogmartin et al., 2017a). An estimated 1.3–1.6 billion milkweed stems must be restored to the Upper Midwestern landscape to meet the 6-ha conservation goal (Pleasants, 2017; Thogmartin et al., 2017a). A geospatial analysis conducted to explore milkweed restoration scenarios found it impossible to reach this target without participation from the agricultural sector, which occupies 77% of all potential monarch habitat in the Upper Midwest (Thogmartin et al., 2017a). All non-agricultural sectors combined could accommodate up to 62% of the 1.3 billion stem milkweed target, necessitating at least 500 million stems on agricultural land. Thogmartin et al. (2017a) suggested restoring this number of stems could be accomplished by retiring the least productive farmland and/or through changes in agricultural practices that may allow the persistence of milkweed.

Marginal farmland is commonly associated with steeper slopes, highly erodible soils, and/or the need for substantial amounts of fertilizer to maintain crop yields (Kang et al., 2013), all of which can increase sediment and nutrient loading into waterways. By purchasing conservation easements on these lands via the Conservation Reserve Program (CRP), taxpayers receive benefits in return, such as improvements to downstream water quality and habitat for wildlife (Dunn et al., 1993). Johnson et al. (2016) found that the value of these easements for reducing flood damages, improving water and air quality, and contributing to greenhouse gas mitigation exceeded the cost of CRP payments to farmers by a factor of 1.3 to 4.9. If, however, the primary objective of conservation investments was to maximize benefits to the public, then it is likely that alternative sites would also be considered. In addition to quantifying the costs and benefits of conserving the least valuable agricultural land, it is worth considering the costs and benefits of conserving the most valuable land in terms of its ability to provide benefits to the public.

The most important or valuable source areas for the provision of ecosystem services have not yet been systematically investigated via comprehensive, quantitative mapping. Numerous studies, however, suggest that riparian corridors are among the most important source areas in terrestrial landscapes. Riparian corridors are generally defined as the stream channel between the low- and high-water marks plus the terrestrial landscape above the high-water mark where vegetation may be influenced by elevated water tables or extreme flooding and by the ability of the soils to hold water (Naiman et al., 1993). Riparian corridors have long been recognized as hosting an unusually diverse array of species and environmental processes and have been characterized as the most diverse, dynamic, and complex biophysical habitats on the terrestrial portion of the planet (Naiman et al., 1993; Naiman and Decamps, 1997). They directly regulate the flow of water, sediment, and nutrients from land areas to oceans (Aufdenkampe et al., 2011), and between surface waters, and groundwater aquifers (Goodrich et al., 2018). Riparian corridors are also the interfaces between aquatic and terrestrial ecosystems, providing highly dynamic and connected habitats to a wide array of species (Gregory et al., 1991), including wild pollinators, which are also experiencing severe population declines (Colla and Packer, 2008; Cameron et al., 2011; Cole et al., 2015). This diversity, coupled with the aesthetic amenities found along rivers and streams, also make riparian corridors an important resource for cultural ecosystem services such as recreation (Loomis et al., 2000; Sherrouse et al., 2014; Darvill and Lindo, 2015).

Despite the abundance of ecosystem services produced by riparian corridors, we found no studies that have attempted to quantify them at a regional, landscape scale. A likely reason for this is that the processes and functions producing many riparian ecosystem services operate at small spatial scales, making them difficult to model with accuracy over large areas. For example, Tomer et al. (2013, 2015) demonstrated how processbased, watershed-scale modeling can be used to anticipate reductions in nutrient loading into waterways from a variety of conservation practices, including grassed waterways, nutrientremoval wetlands, saturated buffers, and others. Their work utilized sub-meter digital elevation models (DEMs) derived from light detection and ranging (LIDAR) data, soil data from the Natural Resources Conservation Service (NRCS) Web Soil Survey, and detailed agricultural boundaries with crop rotation information on a field basis (Tomer et al., 2013). At a landscape scale these high-resolution datasets are not available and sophisticated process-based modeling is not practical. Some services, however, are amenable to generalization, such as water quality regulation and the pollination of food crops. The use of riparian buffers to regulate water quality has been a best management practice in agricultural landscapes for decades and numerous studies have quantified performance measures (e.g., Osborne and Kovacic, 1993). Similarly, crop yield increases associated with proximity to natural habitat, and wild pollinators have been documented for a variety of crop species (Garibaldi et al., 2013).

The large body of knowledge on the benefits of naturally vegetated riparian corridors has led to growing interest in policy options that would result in more uniform implementation of this best management practice (Fremier et al., 2015; Merrill, 2015; González et al., 2017). In November of 2017, Minnesota implemented a riparian buffer regulation, the first statewide regulation in the U.S. that mandated natural vegetation within 50 feet (15.24 m) of public waters (MNBWSR, 2019). Many other states have implemented or are considering a variety of buffer protection policies in selected sensitive watersheds, primarily focused on water quality improvement (Gene et al., 2019). Analyses of the costs and benefits of landscape-scale riparian restoration activities are thus needed to inform the political debate about specific policy options.

We explore the question of whether it can be cost effective to retire productive farmland adjacent to rivers and streams in the pursuit of milkweed restoration goals. We use a landscape-scale geospatial analysis of the U.S. Upper Midwest to consider three buffer-width scenarios that could be implemented throughout the region. For each, we identify how much natural vegetation currently exists within the buffer zone and how much milkweed could be added. We further quantify and value a partial set of ecosystem services provided by the habitat scenarios and compare that value with the cost of giving up agricultural production within the buffer zone. We discuss strengths and limitations of the approach and identify opportunities for further research.

# MATERIALS AND METHODS

#### Study Area

The study area consists of Iowa, Minnesota, Michigan, and Wisconsin in their entirety along with the northern portions of Ohio, Indiana, and Illinois in the Upper Midwest region of the United States (**Figure 1**). The total land area for the study area is ∼846,000 km² with 353,000 km² of cropland. Based on the Cropland Data Layer (CDL), corn (47%) and soybeans (38%) are the primary crops produced in this area, comprising ∼85% of agricultural land in production throughout the region (USDA/NASS, 2015). The next largest agricultural commodity grown in this area accounts for 7.5% of the total land area and those crops are alfalfa (6%), and hay (1.5%). The 67 remaining crops accounted for in the CDL comprise just 7.5% of the total land area in agricultural production. The study area represents the agricultural heartland of the U.S. but it is also the primary summer breeding range of monarch butterflies. It is within this region that Pleasants (2017) estimated the potential monarch support capacity (milkweed) loss has been 71% over the last 20 years, and residents have indicated a collective willingness to pay of ∼US\$45 million per year for monarch habitat restoration (Semmens et al., 2018).

#### Scenario Development

We developed three scenarios representing the restoration of natural vegetation along riparian corridors of different widths along perennial and intermittent streams. Our first scenario used 30-m buffers on each side of the waterways. The second scenario used 100-m buffers (200-m total). The third used a variable-width buffer, which included an 80-m first order stream buffer, a 100-m second order stream buffer, and a 120-m buffer for streams with orders of three to five. These scenarios were selected to represent a range of different buffer widths described in previous studies of their effectiveness at sediment and nutrient trapping, although beyond 30 m trapping levels do not increase substantially (Lyons et al., 2000; Mayer et al., 2007; Tiwari et al., 2016). The amount of land available for milkweed and pollinator habitat restoration under each scenario was a driving force in creating the larger buffer scenarios.

To create geospatial representations of the buffer zones for each scenario, we used the National Hydrography Dataset PLUS Version 2, which excludes ephemeral streams (McKay et al., 2012). We subset the NHD to include only waterways with stream orders ranging from 1 to 5. This subset of the NHD was selected after an examination showed that when stream order got above 5 the river width tended to be >30 m, making our analysis with the 30-m CDL difficult to accomplish because NHD does not include a width attribute for streams. Once we had this subset of the NHD perennial and intermittent streams, we created buffers by running the ArcGIS Buffer Tool for our three different buffer-width scenarios.

#### Intersection of CDL and Buffer Zones

Using the newly created riparian buffers, we extracted areas in the CDL that were contained within the buffer areas under each of the three scenarios to determine the total hectares of different crop and land-cover types using ArcGIS. The CDL for this region consists of 91 land-cover types where 22 of the classes correspond to aquaculture, open water, developed-open space, developedlow intensity, developed-medium intensity, developed highintensity, barren lands (e.g., gravel pits), deciduous forest, evergreen forest, mixed forest, shrubland, grassland/pasture, woody wetlands, herbaceous wetlands, and long-lived orchard crops (e.g., apples, cherries, peaches, grapes, pears, plums, Christmas trees, and other tree crops). These classes were excluded from the buffer analysis since they were already natural areas, could not sustain a milkweed population, or were longlived crops that would not be practical to retire. The remaining 69 classes relate to a specific crop type, with 8 classes relating to a double crop of two different crop types. The buffer zone was comprised of 63% natural vegetation (deciduous forest, evergreen forest, mixed forest, shrubland, woody wetlands, and herbaceous wetlands) for the 30-m scenario, which indicates that most streams in the region already have some form of buffer.

# Milkweed Restoration Potential

We used three different levels of milkweed density within riparian buffers to estimate the number of plants that could potentially be added throughout the region, with milkweed densities derived from Thogmartin et al. (2017a). The first level is based on the existing milkweed density of non-prairie grassland at 7.64 stems per hectare and is intended to represent conditions typical of pasture and agricultural grassland, which receive occasional spot treatment to remove milkweed. The second level of 151.65 milkweed stems per hectare is based on densities estimated for CRP lands with persistently wet soils, which are typical of riparian wetlands. The third level is based on CRP lands with dry soils that are estimated to support an average of 277.1 milkweed stems per hectare. Our scenarios only consider the conversion of current agricultural land within the buffer zone, which Thogmartin et al. (2017a) estimated could support an average density of 277.1 milkweed stems per hectare.

riparian buffer scenarios overlaid (Blue: 30-m, Yellow: Variable-Width, Green: 100-m).

The range of milkweed potential we consider is thus intended to represent the possibility that milkweed densities in riparian buffers could vary widely by landowner, from enthusiastic planting to continued spot treatment, and as a function of specific site characteristics such as soil type and moisture content, the encroachment of woody vegetation, herbicide drift, and/or other factors. We did not formally take into account soil composition when estimating milkweed restoration potential because no quantitative relationships have been established. Bowles et al. (2015) noted that well-drained sites on fine-textured, Wisconsinan-aged glacial soils found commonly throughout our study area would enhance the establishment and growth of

milkweed, but quantitative estimates of milkweed density across multiple soil types were not available.

#### Sediment and Nutrient Removal Upland Areas and Headwater Reduction

Riparian buffer strips are a common best management practice (BMP) in agricultural settings because they filter and trap sediment and nutrients from surface runoff before it enters streams. They are limited, however, because they can only filter runoff flowing overland laterally into streams; they cannot filter channelized flow. Our calculations thus necessitated estimating the fraction of the landscape contributing flow laterally into

perennial and intermittent streams and thus subject to water quality improvement from riparian buffer strips. To make this estimate, we started by delineating all watersheds with pour points (outlets) located at the transition to 6th order streams or higher. This represented the maximum potential area that could be serviced by riparian buffers in our analysis, but it needed to be reduced to account for headwater areas contributing flow directly into the origination point of the NHD streams. We then selected one representative watershed for each state and delineated all the headwater watersheds upstream from the upstream end of each 1st-order stream (**Figure 2**). Given the number of 1storder streams in the study area it was not feasible to do this for all watersheds, so we used the sampled percentages of our representative watersheds occupied by headwater watersheds in each state to reduce the potential area to which riparian buffers could provide sediment and nutrient retention benefits. Representative watersheds for each state were selected based on their average size relative to other watersheds in the state and had to have sufficient relief such that the stream network could be delineated cleanly from the DEM. Estimates of upland and headwater areas are presented in **Table 1**.

#### Drainage Tiles

Subsurface drainage through drainage tiles is a major source of nutrient pollution throughout the study area, and one that is not well-mediated by riparian buffers (Osborne and Kovacic, 1993). The lack of information available on the precise locations and


extent of drainage tiles on agricultural land (Ruark et al., 2009) make it difficult to quantify their specific impacts, necessitating a more generalized approach. Using historic estimates provided by the United States Department of Agriculture (USDA) combined with a GIS analysis conducted by Sugg (2007) we were able to estimate of the percentage of agricultural land that uses subsurface drainage for each state within our study area (**Table 2**). We used these percentages to reduce the total amount of sediment eroded from upland areas that could potentially be filtered by riparian buffers, assuming that the water needed to transport that sediment to streams is drained away below ground. It is unlikely that all subsurface drainage in the region falls within

#### TABLE 2 | Drainage tile percentages by state (Sugg, 2007).


the portion of the landscape subject to filtration by our buffer scenarios, making this a highly conservative reduction of the potential sediment retention benefits.

#### Sediment Erosion and Trapping Efficiencies

We applied an average erosion rate of 7.21 metric tons per hectare per year (3.22 tons per acre per year) on agricultural land, which USDA-NRCS (2018) estimates to be representative for this region. A more thorough model-based analysis of erosion accounting for slope, soil type, and management factors was not practical due to the size of the study area, consistency and resolution of available data, and difficulty of delineating the specific areas with potential to benefit from sediment trapping by riparian buffers (as described in section Upland areas and headwater reduction). The average erosion rate was applied to the total area of agricultural land subject to filtration by riparian buffers, following the above described reductions for headwater areas and subsurface drainage. We assume that the resulting total sediment yield is currently making its way into rivers and streams and is thus available to be trapped by riparian buffer strips.

We surveyed published sediment trapping efficiencies for riparian buffer strips to derive the value used in our analysis (**Table 3**). Sediment trapping by riparian buffers is highly dependent on soil type, slope, land use, and other factors which creates discrepancies across studies for filtering capabilities (Hawes and Smith, 2005). From the selected studies and reviews, we used the lowest reported value for sediment trapping efficiency where buffer width was similar: a trapping efficiency of 61% (Meyer et al., 1995) was applied to sediment originating upslope from and thus subject to filtering by a buffer. For agricultural land converted directly to buffer, we did not assume a 100% trapping efficiency but rather applied a value of 97%, consistent with the observed performance of buffers at filtering upslope sediment in several studies (Yuan et al., 2009). These rates were applied uniformly throughout the entire region following the area reductions described in sections Upland areas and headwater reduction and Drainage tiles. Existing natural areas present in the buffer scenarios were also taken into account for the sediment trapping they currently provide by further reducing the total sediment from upland areas by an additional 63% to align with our estimate of existing natural areas within the 30-m buffer zone.

TABLE 3 | Buffer characteristics from multiple studies on width, composition, and trapping efficiency.


#### Valuing Water Quality Regulation by Buffers

The USDA Economic Research Service estimated that for each avoided metric ton of eroded soil entering waterways there are US\$2.51 in benefits for the corn belt states (Iowa, Illinois, Indiana, Ohio) and \$4.25 for the lake states (Minnesota, Wisconsin, Michigan) (Hansen and Ribaudo, 2008). These estimates were based on September 2008 US dollars and were adjusted for inflation using U.S. Bureau of Labor Statistics' (BLS) Consumer Price Index (CPI) Inflation Calculator (BLS, 2015) to September 2016 US dollar equivalents of \$2.78 and \$4.68, respectively, to correspond with our crop and pollination values. All subsequent values presented in this paper are 2016 US dollar equivalents. These estimated values include the benefits of water-quality improvements to irrigation ditch, canal, and road drainage ditch maintenance, municipal water treatment, avoided agricultural flood damages, marine fisheries, freshwater fisheries, industrial water use, steam power-plants, and soil productivity. To the extent that nutrient and sediment effects on benefits are correlated, The Hansen and Ribaudo (2008) benefit estimates include effects of nutrients. Nutrient sorption in sediment is common, making it difficult to distinguish willingness to pay for water quality improvements resulting from sediment vs. nutrient reductions. Hansen and Ribaudo (2008) acknowledge that monetary values derived from their data are likely to be lower-bound estimates and although they lack precision for small-scale value estimates, the values are thought to be detailed enough for national and regional estimates. Such applications have been conducted for the Prairie Pothole region of the north central U.S. (Gascoigne et al., 2011) and to the state of Iowa (Zhou et al., 2009) to evaluate scenarios of land use and conservation practices, respectively.

#### Pollinator Benefits

Expanding natural vegetation along riparian corridors is expected to enhance habitat heterogeneity and ecological connectivity, creating suitable conditions for wild pollinators that can increase agricultural yields for certain crops. Cole et al. (2015), for example, found that riparian buffers supported a greater diversity of insect pollinators than adjacent grassy fields, and that insect abundance increased with buffer width. To quantify this benefit, we needed to estimate three things: the average yield increase for pollinator-dependent crops due to wild pollinators, how far wild pollinators travel from areas of natural vegetation (foraging distance), and how much pollination effectiveness is likely to decline with distance from natural habitat. The average value of crop yield increases from wild pollinators was derived from Kleijn et al. (2015) who synthesized data from 90 studies globally to derive the average contribution of \$3,251/ha to the production of 20 pollinator-dependent crops. We estimated the maximum foraging distance to be ∼2,400-m (∼1.5 miles) from a list of the 12 most effective wild bee pollinator species in Pennsylvania (McGlynn, 2009) by taking the average of the midpoint of flight/forage distances for each species and rounding down (**Table 4**).

A GIS analysis determined that all pollinator-dependent crops in the study area (apples, blueberries, canola, cantaloupes, cherries, cranberries, dry beans, eggplants, flaxseed, gourds, peaches, pears, peas, plums, pumpkins, squash, sunflowers, tomatoes, and watermelons) are located within 2,400 m of at least one pixel of natural land cover. To estimate the value of restoring

TABLE 4 | Bee species, flight/foraging distances (McGlynn, 2009).


additional pollinator habitat conservatively, we assumed that provision of pollination services decreases exponentially with distance from natural habitat. We applied an exponential decay function (e −0.003x ) to the value of pollination such that it decreased from \$3,251/ha adjacent to natural areas down to \$0 beyond 2,400 m. We then calculated the area of pollinatordependent crops within 14 distance bands up to 2,400 m from the baseline natural vegetation (including existing CRP leases) as well as the expanded natural vegetation under each buffer scenario. This analysis is limited by the 30-m cell size of the CDL, so we created our distance bands as multiples of 30 m (**Table 5**), using the "expand" tool in ArcGIS. For each buffer distance we intersected the buffered natural vegetation with the pollinatordependent crop extent, using raster calculator to identify the overlapping cells, and recording their area. To account for the conversion of existing crops to natural area, we subtracted the area of pollinator-dependent crops converted to buffer from the first distance band of the baseline for each scenario so that we only consider the additional pollination value for the remaining crops. These two steps yielded the distribution of pollinatordependent crop area falling within each distance band from the existing and expanded natural vegetation associated with each scenario. To estimate the value associated with each scenario, we multiplied the value (\$/ha) at the midpoint of each distance band by the area of pollinator-dependent crops in that band (ha) to first compute the existing value of wild pollinators to pollinator-dependent crops. We then calculated the total value associated with each scenario and subtracted the existing value to get the value added as a result of the new buffer area. This process accounts for the increase in pollination value as natural area shifts closer to pollinator-dependent crops rather than providing access to new crops.

This process was repeated for soybeans, which are primarily self-pollinated but can benefit from insect pollination. Numerous

TABLE 5 | Value of pollination benefits per hectare of pollinator-dependent crops (PDC) and soybeans within 90-m distance bands from natural vegetation.


*Values decrease exponentially according to the distance decay function.*

studies have documented visitation of soybean flowers by wild pollinators (Rust et al., 1980; Milfont et al., 2013; Gill and O'Neal, 2015; Monasterolo et al., 2015; Wheelock et al., 2016) and yield increases resulting from pollinators (Erickson, 1975; Erickson et al., 1978; Chiari et al., 2005; Milfont et al., 2013; Santos et al., 2013; Blettler et al., 2018). Of these, only Erickson (1975) quantified yield increases from wild pollinators within the study area, finding an average yield increase of ∼5% in open, untreated plots relative to plots treated with insecticide throughout the flowering period. Considerably higher yield increases have been observed in association with managed honey bees (Apis mellifera), but these increases vary widely with weather and soybean variety (Blettler et al., 2018). Wheelock et al. (2016) describe the limitations of using yield increases from studies that focus on honey bees because they comprise a small percentage of the pollinator community found in the primary soybean region of our study. We used the 5% yield increase from Erickson (1975) together with the average yield and price of soybeans (in 2016 US\$) within the study area to estimate the maximum value for soybean yield increases provided by wild pollinators. This maximum value was decreased exponentially over 2,400 m to estimate the total value of increased soybean yield associated with each scenario following the same procedures described above for pollinator-dependent crops. Due to the uncertainty of how pollinators may contribute to soybean yields, the value of pollinator-dependent crop and soybean yield increases are kept separate so the soybean value can be excluded easily from the final cost-benefit analysis.

#### Lost Crop Estimates and Cost-Benefit Analysis

In addition to benefits from restoring habitat and ecological function at a landscape scale, there is also an important cost to private landowners, namely the lost profit from crops currently grown within the buffer zone. An understanding of this cost relative to the value of benefits is needed to inform the debate over viable policy options involving riparian buffers. We used data from the USDA that reports average yields and prices on crops grown in the states throughout our study area (USDA/NASS, 2016). Using this average yield and price information we were able to estimate the value of each specific crop and thus the total value of lost crops within each of our three buffer scenarios (**Supplementary Table 1**). Some crops in the CDL do not have corresponding yield and price information for every state, in which case neighboring state prices and yields were used. In addition, specific crop classes had no yield and price information, such as other crops, miscellaneous vegetables and fruits, fallow/idle cropland, sod/grass seed, clover/wildflowers, herbs, and vetch. These classes are included in the overall analysis but are not reflected in the crop-loss estimates.

The gross value of lost crops overestimates their value to farmers who incur considerable costs in bringing crops to harvest. The net profit margin from farming varies as a function of farm size, crops grown, and other factors. MacDonald et al. (2006) reported profit margins on U.S. farms ranging from −24.8 to 16.4% depending on annual farm sales. A large majority of small family-owned farms in the U.S. have an operating profit margin of <10 percent and these farms make up more than 50% of the total land operated for farming purposes (Hoppe, 2017). In the interest of being conservative in our estimates, we elected to apply a profit margin of 10% to the total value of all crops grown within the buffer zones of our three scenarios. This allows for the possibility that yields are higher adjacent to rivers and streams, perhaps due to fertile floodplain soils. We do not consider the cost of planting or maintaining the natural vegetation, including milkweed, associated with our buffer scenarios, which would be broadly similar for any monarch habitat restoration occurring on agricultural land. If habitat establishment costs were to be included, our expectation is that first year cost-benefit ratios would be very low, but would approach our estimates in the long-term as buffer vegetation becomes fully established.

# RESULTS

# Milkweed Potential

Our GIS analysis of riparian buffer scenarios throughout the Upper Midwest shows that buffer width and management style create a large range in the number of milkweed stems that can be supported. Between 238 million and 1.02 billion stems can be added with the restoration of riparian buffers if densities typical of upland native prairie can be attained (**Table 6**). This would meet between 50 and 200% of the milkweed restoration target Thogmartin et al. (2017a) estimated would be needed from the agricultural sector, assuming maximum restoration on non-agricultural lands. In contrast, if the buffers are managed as agricultural grassland then just 6.6 to 28.1 million stems may be possible. Due to the highly variable soil and moisture conditions within the riparian zone and uncertainty about landowner willingness to accept milkweed, an intermediate density associated with wet CRP land may be more realistic. This would add 130 to 559 million stems, meeting 25–100% of the milkweed target for the agricultural sector, depending on buffer width. These results only estimate the amount of milkweed that could be added on the land converted from agriculture to natural vegetation under each scenario.

# Water Quality Benefits

We estimated the total amount of sediment eroded from agricultural lands annually in each state that are subject to

TABLE 6 | Estimates of milkweed restoration potential using three management styles (CRP dry, 277.1 stems/ha; CRP wet, 151.65 stems/ha; and grassland/pasture, 7.64 stems/ha) for each of the three buffer-width scenarios.


filtration and sediment retention by riparian buffers (**Table 7**). The estimated amount and value of sediment trapped by each buffer scenario (**Table 8**) indicate that the 30-m buffer scenario could produce \$302M in avoided costs as a result of water quality improvements. The variable width and 100-m scenarios are only marginally better, producing \$356 and \$371M in water quality benefits, respectively. These increases result from the retirement of larger land areas rather than the greater trapping efficiency of wider buffers.

#### Pollination Benefits

Pollinator-dependent crops could benefit from increased pollination and higher yields, regardless of whether or not they are presently relying on pollination services from managed honey bees (Garibaldi et al., 2013; Kleijn et al., 2015). The results of our pollination analysis are presented in **Table 9**. The 30-m scenario resulted in ∼\$102 million of increased crop yields. The variable-width and 100-m scenarios add far more natural habitat on the landscape but at the expense of existing crops and do not result in pollinator access to additional crops, resulting in \$50.6 and \$54.8 million in increased crop yields, respectively.



TABLE 8 | Estimated retention of sediment under different buffer scenarios and associated values of improved water quality.


TABLE 9 | Summary of estimated pollination value for each buffer scenario.


# Crop-Loss Estimates and Cost-Benefit Analysis

The estimated value of lost agricultural production resulting from the conversion of cropland to natural habitat is presented in **Table 10** for each scenario. We include two measures of crop loss, the gross annual crop value (average yield/hectare × hectares × price), and the net cost to farmers less inputs (seed, fertilizer, fuel, etc.) assuming a uniform 10% profit margin. A comparison of costs with the aggregated benefits is presented in **Table 11** (state-specific costs and benefits are available at doi: 10.5066/P9DV375U). The 30-m scenario results in ∼\$205M in annual lost profits for the landowners across all crop types, but the benefits (water quality and pollination) are approximately twice that amount. The variable-width and 100-m scenarios are more expensive to implement and benefit gains are less substantial, resulting in cost-benefit ratios below 1. These results are relatively insensitive to the value of yield increases from soybeans. If this value is excluded entirely, the cost-benefit ratio for the 30-m scenario only drops to 1.72. Again, these estimates neglect the initial cost of restoring natural vegetation, which would certainly drive the cost-benefit ratio below 1 for all scenarios during the first year but would be similar to restoration costs on any tilled agricultural land regardless of its productivity.

# DISCUSSION

We have estimated the potential contribution of the retirement of agriculture and restoration of natural vegetation along riparian corridors throughout the Upper Midwestern U.S. toward achieving milkweed restoration goals. Our estimates indicate the possibility of adding between 6.6M and 1B new milkweed stems depending on the width of the buffers and the type of management they receive. This large potential range is because milkweed counts have never been conducted specifically within riparian corridors and there is uncertainty about the potential of these corridors to serve as host habitat. Several studies, however,

TABLE 10 | Area of existing crops converted to natural vegetation under each scenario and associated gross and net value annually.


TABLE 11 | Summary of final benefits and costs in 2016 U.S. dollars per year.


have documented the presence of both common milkweed (Asclepias syriaca) and swamp milkweed (Asclepias incarnata) in riparian corridors within the study area (Paine and Ribic, 2002; Goebel et al., 2003; Benson et al., 2006) and the seeds of other milkweed species have been shown to have high buoyancy and viability after extended contact with water (Edwards et al., 1994) suggesting the potential for hydrochory (seed dispersal by water). The wide range of moisture conditions, vegetative composition, and frequency of disturbance within riparian corridors may increase the number of stems and variety of milkweed species likely to be present, but also makes it difficult to assign one value for milkweed stem density representative of a region as large as the Upper Midwest.

Under our smallest, 30-m buffer scenario with moderate management for milkweed, ∼10% of the overall monarch habitat restoration goal, and 25% of the contribution from the agricultural sector can be met. For comparison, Thogmartin et al. (2017a) estimated that roadside milkweed density could be increased to meet ∼15.6% of the overall monarch habitat restoration goal. The 500M stem goal for the agricultural sector can be met entirely by the retirement of marginal farmland and restoration of native prairie, but this strategy is unlikely to provide an equivalent magnitude and diversity of benefits as the potential gain from restoring a combination of marginal land and riparian corridors throughout the region. The fact that riparian buffers alone may produce up to 6.6M milkweed stems may make them a desirable component in the broader mix of restoration efforts.

We conducted an extensive geospatial analysis to value a subset of the ecosystem services provided by riparian buffers and investigate how they compare to the lost value of crops currently grown within the buffer zone. In the 30-m scenario, the annual value of benefits provided by the buffers is US\$405M, approximately double the lost crop value of US\$205M, but for larger buffer widths the cost increases outweigh the additional benefits. Our cost-benefit calculations do not account for initial restoration costs, the time required for buffer vegetation to become fully established, or changing commodity prices over time. As such, our cost-benefit ratios reflect a static snapshot of selected potential benefits at some point in the future once vegetation is established relative to the current annual opportunity costs to farmers. Restoration costs and establishment times for prairie vegetation on agricultural land would have to be considered carefully prior to any policy implementation.

Empirical parameters obtained at the field and small watershed scale can overestimate the performance of best management practices when applied at larger spatial scales (Liu et al., 2017), so we used conservative numbers for service provisioning and the associated value of benefits. In addition, although the impact of sediment and nutrients on a variety of ecosystem services is incorporated into our valuation of water quality benefits, we were not able to quantify many other services known to be provided by riparian buffers because of a lack of data and models that could be defensibly applied at a regional scale. We were thus unable to account for the value of enhancements to scenic amenities in rural landscapes, flood mitigation to cities and towns, carbon sequestration, recovery of threatened and endangered species, wildlife viewing and hunting opportunities, and other services within the region and downstream. A more complete accounting of these benefits could further increase the return on investments in habitat restoration adjacent to rivers and streams.

There are also potential disservices associated with restoring corridors of natural vegetation throughout agricultural landscapes. It is possible that the large-scale reintroduction of intact riparian habitat could create shelter and dispersal corridors for agricultural pest species (Maisonneuve and Rioux, 2001; Zhang et al., 2007), which could impose an additional cost on farmers that we have not attempted to account for in the present analysis. Such disservices, however, are likely to be outweighed by the cumulative benefits of creating contiguous habitat to species that provide natural control of crop pests (Landis et al., 2000; Marshall and Moonen, 2002; Zhang et al., 2007), and other desired terrestrial and aquatic species, which include ∼70% of the region's threatened and endangered species that rely to some extent on these habitats (USFWS, 2019). Stewart et al. (2001), for example, found that streams dominated by riparian corridors without gaps and with less fragmentation of natural vegetation have healthier fish and macroinvertebrate communities and a greater density of fish for recreational fisheries.

Our approach to estimating pollination benefits implicitly assumes that wild pollinators will recolonize new buffer areas in sufficient numbers to provide the modeled service. However, it may take some time for this to occur and other factors may come into play. Koh et al. (2016) showed the Upper Midwest to have the lowest abundance of wild bees and greatest negative trend in their abundance relative to other regions in the U.S. Continued indiscriminate utilization of neonicotinoid seed treatments, particularly in soybeans, could perpetuate wild pollinator declines (Tooker et al., 2017), and negate the benefits associated with creating new pollinator habitat. Similarly, herbicide drift can impact the diversity of field-edge vegetation (de Snoo and Van der Poll, 1999), flowering in selected species, and arthropod abundance (Egan et al., 2014), though some herbicide use can be managed to avoid damage to milkweed and monarch larvae (Lizotte-Hall and Hartzler, 2019). Olaya-Arenas and Kaplan (2019) document 14 pesticides−4 insecticides, 4 herbicides, 6 fungicides—on milkweed leaves in northwest Indiana suggesting monarch caterpillars consume a diversity of agricultural chemicals, but the lethal or sublethal impacts of this exposure remain unknown. Without careful management these factors may limit both the potential availability of nectar resources for pollinators as well as the density of milkweed in riparian buffers.

Due to data and resolution limitations, we only quantified benefits resulting from increasing natural vegetation around the 1:100,000-scale perennial and intermittent streams within the NHD dataset; ephemeral headwater stream channels were excluded from the analysis, but could be major sources of sediment, and nutrients if not similarly protected. Some of these ephemeral headwater channels are presently buffered with natural vegetation, but we were unable to quantify their extent. Buffering these areas with natural vegetation could add to the potential for milkweed restoration and increase pollination and water quality benefits, among many others. With higher resolution topographic and land-cover information it would be possible, and prudent, to expand our analyses to include these areas. Numerous studies have shown that headwater streams account for a substantial fraction of streamchannel length in the U.S. and are crucial to ecological and biophysical functioning and attendant ecosystem services (Lowe and Likens, 2005; Alexander et al., 2007; Creed et al., 2017; Wohl, 2017).

# CONCLUSIONS

Retiring agricultural land and restoring habitat along riparian corridors could significantly increase the availability of milkweed for monarchs throughout the Upper Midwest. This highly connected habitat is also widely distributed throughout the region and could serve as movement/migration corridors for monarchs and many other species of conservation concern. The habitat could further provide a variety of other valuable benefits throughout the region that are twice their cost in terms of lost agricultural production for a 30-m buffer width. Numerous additional benefits could not be quantified but may further increase the return on investment in riparian buffers. Gustafsson et al. (2015) observed that the monarch has been a powerful communication vehicle and a potent ally in environmental politics. These factors all suggest the strong potential to leverage monarch restoration goals, and the popular momentum for meeting them, toward the restoration of a multifunctional

#### REFERENCES


landscape that benefits a wider range of species, people, and communities.

# DATA AVAILABILITY

The official data release for this study are published on ScienceBase at doi: 10.5066/P9DV375U.

# AUTHOR CONTRIBUTIONS

DS designed the project. ZA conducted the analyses. DS and ZA jointly drafted the manuscript.

#### FUNDING

This work was supported by the U.S. Geological Survey, Land Change Science Program.

#### ACKNOWLEDGMENTS

We would like to thank Laura Norman and two reviewers for their thoughtful comments on this manuscript. Any use of trade, product, or firm names are for descriptive purposes only and does not imply endorsement by the U.S. Government.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fenvs. 2019.00126/full#supplementary-material


and Mexico. Environ. Manage. 62, 229–240. doi: 10.1007/s00267-018- 1046-1


17, 81–90. Available online at: http://www.scielo.edu.uy/scielo.php?pid=S2301- 15482013000100009&script=sci\_arttext&tlng=en


Zhou, X., Helmers, M. J., Al-Kaisi, M., and Hanna, H. M. (2009). Cost-effectiveness and cost-benefit analysis of conservation management practices for sediment reduction in an Iowa agricultural watershed. J. Soil Water Conserv. 64, 314–323. doi: 10.2489/jswc.64.5.314

**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 past co-authorship with one of the authors DS.

Copyright © 2019 Semmens and Ancona. 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.

# Butterflies Across the Globe: A Synthesis of the Current Status and Characteristics of Monarch (Danaus plexippus) Populations Worldwide

#### Kelly R. Nail <sup>1</sup> \*, Lara Drizd<sup>2</sup> and Kristen J. Voorhies <sup>3</sup>

*<sup>1</sup> Minnesota Wisconsin Field Office, U.S. Fish and Wildlife Service, Bloomington, MN, United States, <sup>2</sup> Ventura Field Office, U.S. Fish and Wildlife Service, Ventura, CA, United States, <sup>3</sup> Chicago Field Office, U.S. Fish and Wildlife Service, Chicago, IL, United States*

Recent declines in the migratory North American populations of monarchs (*Danaus plexippus*) have necessitated efforts to evaluate the current status of the species, including worldwide populations. While monarchs originate from North America and may be ancestrally migratory, they have expanded throughout many parts of the world over the past 200 years. Most of these newer populations no longer migrate and face a variety of threats across a wide range of habitats, but we lack a comprehensive review of locations and characteristics of these worldwide populations. We thus delineated the current range of monarchs and their status throughout the world, recording over 90 countries, islands, and island groups where monarchs occur (74 with recent documented sightings) and known features of these populations. We discuss the major differences between these populations, focusing on morphology, migration, overwintering, natural enemies, larval diet, and genetics. The differences documented here provide the species with adaptive capacity, thus better allowing the species to adapt to novel changes in its environment. We end with a discussion of current gaps in our understanding of monarchs worldwide and directions for future research.

Edited by:

*Wayne E. Thogmartin, United States Geological Survey (USGS), United States*

#### Reviewed by:

*Jeffrey M. Marcus, University of Manitoba, Canada Ayse Tenger-Trolander, University of Chicago, United States*

> \*Correspondence: *Kelly R. Nail kelly\_nail@fws.gov*

#### Specialty section:

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

Received: *21 May 2019* Accepted: *11 September 2019* Published: *27 September 2019*

#### Citation:

*Nail KR, Drizd L and Voorhies KJ (2019) Butterflies Across the Globe: A Synthesis of the Current Status and Characteristics of Monarch (Danaus plexippus) Populations Worldwide. Front. Ecol. Evol. 7:362. doi: 10.3389/fevo.2019.00362* Keywords: monarch, Danaus plexippus, migration, worldwide range, natural enemies, morphology, genetics, larval diet

#### INTRODUCTION

Monarchs (Danaus plexippus) are well-known within North America for their long distance migration to overwintering sites along the western coast and in central Mexico. These colorful orange and black butterflies have also expanded to occupy areas throughout the world, from Australia to Spain. With the recent declines in both the eastern and western North American populations, the U.S. Fish and Wildlife Service (USFWS) was petitioned to list monarchs under the Endangered Species Act (ESA) of 1973 (Center for Biological Diversity, 2014). As part of this process, the USFWS is conducting a Species Status Assessment (SSA) to evaluate the status and viability of the species. This SSA requires the species be evaluated as a whole, including in locations outside of the eastern and western North American populations. This process thus necessitated a better understanding of worldwide monarch populations, including where monarchs currently exist, threats faced, and how these populations contribute to the adaptive capacity of the species. Adaptive capacity, or the ability of a species to adjust to novel changes in its physical and biological environment, is important to evaluate to understand the continued success and viability of the species (Nicotra et al., 2015). This mini-review summarizes the findings on monarchs throughout the world, primarily focusing on areas outside of the eastern and western North American populations.

### BACKGROUND AND CLASSIFICATION OF MONARCHS WORLDWIDE

Monarchs were not recorded outside of North America until the mid-nineteenth century, when they colonized areas across both the Pacific and Atlantic Oceans (Zalucki and Clarke, 2004; Fernández-Haeger et al., 2015; although see also Zhan et al., 2014 regarding genetic evidence on potential earlier timing of dispersal events). They established new populations, using available milkweed host plants (primarily Asclepias spp.) and often became non-migratory in the face of year-round suitable temperatures. Monarchs continue to reside in many of these areas worldwide, although North American individuals vastly outnumber the combined numbers of individuals in all the other regions.

We initially limited this review to the migratory subspecies of monarch, D. plexippus plexippus, as this was the subspecies that was petitioned to be listed under the ESA. However, the subspecies categorization is not well-defined for monarchs (e.g., there are non-migratory monarchs that live year-round in areas where others migrate), and most of the literature only refers to monarchs at the species level. Thus, we examined the entire worldwide range of the species, D. plexippus.

To determine where monarchs are located worldwide, we first built a database of all observations of the species published in scientific papers (including Ackery and Vane-Wright, 1984; Zalucki and Clarke, 2004; Patrick and Patrick, 2012; Fernández-Haeger et al., 2015). Observations were removed from locations where we now know it was likely another species [e.g., butterflies classified as monarchs that occur in southern South America are likely Danaus erippus (the southern monarch; Malcolm and Slager, 2015)]. We then looked for current evidence of monarch occupation, which we defined as a sighting in the twenty-first century, by first conducting an extensive literature search to locate countries and islands where monarchs have been recently observed. We also searched iNaturalist, a citizen science platform, and the photo sharing site Flickr for posted monarch sightings with photographic documentation in locations throughout the world. All photos were vetted by the authors, and records were not used if the species could not be verified or if the photo was taken in a butterfly exhibit (as monarchs present might have been imported from other areas). Observations were also excluded that were likely monarchs passing through (e.g., there are occasional sightings of monarchs in England, but no observed breeding). Based on the differences discussed below, we then grouped these countries and islands into eight different geographic regions (**Table 1**). In total, 90 countries, islands, or island groups were identified as having been historically occupied by monarchs (**Figure 1**). Of those, 74 have verified sightings since 2000 (**Table 1**). While the monarch now resides in many worldwide locations, they do not reside in all climatically suitable locations (Zalucki and Rochester, 1999).

We next examined the literature for differences between monarchs throughout the world in morphology, larval diet, natural enemies, migration, overwintering, genetics, and population sizes and trends, to help us better understand the potential worldwide sources of adaptive capacity for the species.

# WORLDWIDE DIFFERENCES

#### Morphology

Morphological differences that contribute to the adaptive capacity of the species include wing structure and coloration differences. Wing length was examined in non-migratory monarchs throughout the western hemisphere (from Costa Rica, Puerto Rico, Hawaii, and southern Florida), as well as eastern and western migratory monarchs (Altizer and Davis, 2010). There were differences in shape and size between non-migratory and migratory populations, and non-migratory populations had relatively smaller wings. Additionally, within the eastern population, long-distance migrants tend to have redder coloration (Davis, 2009). Redder coloration is associated with the ability to fly for longer periods of time, although the mechanism for this correlation is unknown (Davis et al., 2012). Recent research also suggests that long-distance migration is a selective force within populations, with longer migration distances positively correlated with longer and larger wings (Flockhart et al., 2017).

#### Larval Diet

Monarchs rely on milkweed as their host plant, but larval diet still contributes adaptive capacity through variation in species of milkweed consumed. Within North America, there are 108 milkweed species in the genus Asclepias, of which at least 33 are known to be used as larval host plants, as well as at least three species of milkweed vines in the genera Cynanchum and Funastrum (Woodson, 1954; Lynch and Martin, 1993; Yeargan and Allard, 2005). Outside of the eastern and western North American populations, monarchs use Asclepias spp. and closely related species in the subfamily Asclepiadoideae as host plants. In most cases outside of the Americas, these host plants are introduced. Some areas with resident monarchs (e.g., Micronesia and Hawaii) are associated with the common ornamental milkweed, Calotropis gigantea (Buden and Miller, 2003; Buden and Tennent, 2017). Monarchs in Morocco are associated with both A. curassavica and Gomphocarpus fruticosus (also known as A. fruticosa; Fernández-Haeger et al., 2015). Larvae in Australia use C. procera, A. curassavica, and G. fruticosus, of which the latter two have a restricted range due to their inability to tolerate frost and dry conditions (James, 1993; Zalucki, 1993). Monarch larvae in New Zealand and other islands use introduced species

#### TABLE 1 | Locations with occurrences of monarchs.


\**Indicates that the country/island has evidence of historical occupation, but no evidence has been found of monarch occupation since 2000; † Indicates that the country is listed in multiple regions.*

*Monarchs are known to have occupied 90 countries, islands, or island groups, which we grouped into eight worldwide regions, with 74 recent documented sightings.*

including G. fruticosus, Araujia sericifera, and Oxypetalum caeruleum. Larvae in the Azores have been observed consuming Gossypium arboreum and some species of the genus Euphorbia, although these species may not be suitable as host plants (Ramsay, 1964; Neves et al., 2001).

#### Natural Enemies

Predation, parasitism, and disease impact monarchs throughout their range and thus contribute to the species' adaptive capacity. One natural enemy is the tachinid fly, which impacts monarchs in Australia (Gibbs, 1994), Hawaii (Etchegaray and Nishida, 1975), and throughout Central America and into South America (Arnaud, 1978; Toma, 2010). In Hawaii, parasitism rates from tachinid flies ranged from 0 to 42% (Etchegaray and Nishida, 1975), and in Australia, rates fluctuate throughout the year, going from very low to 100% of sampled monarchs in February (Smithers, 1973). For comparison, the largest North American study estimated tachinid fly parasitism at 13% (Oberhauser et al., 2007). Another parasitoid, a wasp in the Pteromalus genus, is also known to attack monarch pupae in other locations (Ramsay, 1964).

The protozoan parasite, Ophryocystis elektroscirrha (OE), infects monarchs throughout Australia, Central and South America (Altizer et al., 2000), and Hawaii (Pierce et al., 2014b). Infection rates averaged 35% in Hawaii (range: 4–85%; Pierce et al., 2014b), with Australian infection rates averaging between under 10 and almost 66% (Altizer et al., 2000; Barriga et al., 2016). These average rates of OE infection are lower than those observed in the non-migratory population in southern Florida (75–100%), but higher than average rates in the eastern (<10%) and western (5–30%) North American monarch populations (Altizer and de Roode, 2015). Sternberg et al. (2013) further determined that in lab settings, monarchs from South Florida had lower OE spore loads (relative to eastern migratory monarchs) and were less likely to become infected, potentially indicating that nonmigratory southern Florida monarchs have increased resistance to OE (however, see also Altizer, 2001). Furthermore, the OE parasites from Florida have been shown to cause higher parasite loads than those from the eastern population (Altizer, 2001). Outside of North America, these high rates of OE infection may not be as detrimental to monarchs. Although the Hawaiian strand of OE is particularly virulent, Hawaiian monarchs are both more resistant to and tolerant of OE (Sternberg et al., 2013).

Monarchs have a number of vertebrate and invertebrate predators that have been studied in North America, and are likely to have many predators outside of North America as well (Oberhauser et al., 2015). There are documented bird predators of monarchs in Australia and Oahu, Hawaii (Smithers, 1973; Stimson and Berman, 1990). Additionally, Australia also has a number of recorded spider and insect predators of monarchs, including mantids, ants, and wasps (Smithers, 1973).

#### Migration and Overwintering

Monarchs worldwide exhibit varying overwintering and migratory behaviors (with migration potentially being an ancestral trait; Zhan et al., 2014). This variation creates a range of behavioral adaptive capacity. Eastern North American monarchs migrate upwards of 4,000 km every fall (Solensky, 2004), to overwinter in mountainous forests, which provide a unique, protective microclimate (Williams and Brower, 2015). Western North American monarchs also migrate in the fall, flying up to hundreds of kilometers to primarily coastal overwintering groves, which provide a slightly different specific microclimate (Jepsen and Black, 2015; Pyle, 2015). There are fewer monarchs in the western population, spread out among hundreds of overwintering sites (compared to fewer than 20 sites in Mexico; Vidal and Rendón-Salinas, 2014; Jepsen and Black, 2015). Western North American overwintering monarchs may also have a shorter diapause compared to those in eastern North America (Herman et al., 1989), and there may be differences in mating behavior at the different overwintering grounds (Brower et al., 1995).

While these long-distance migrations are well-studied, many locations worldwide have non-migratory monarchs and yearround or winter breeding, including Central America (Ackery and Vane-Wright, 1984), southern Florida (Brower, 1961), along the Gulf Coast (Howard et al., 2010), and southern California (Satterfield et al., 2016), as well as throughout many Pacific Islands. Monarchs in Australia employ both migratory and non-migratory strategies concurrently (James, 1993), with monarchs breeding year round in a northeastern coastal area, but overwintering without breeding at two other sites (Smithers, 1977). This strategy of partial migration (where some individuals migrate and others do not) thus seems common throughout the monarchs' worldwide range, although the proportion of migrants to non-migrants varies greatly.

Australian monarchs have been recorded flying as far as 380 km northeast, forming transient autumn roosts and eventually overwintering roosts. These autumn and winter roost sites are similar in configuration and are often adjacent to major breeding grounds (James, 1993). These sites tend to be inland (20–80 km), have some protection from southerly and westerly winds, and have trees and bushes for roosting [primarily Melaleuca styphelioides (a native tree) and Lantana camara (a naturalized, invasive plant that is also a nectar source for migrating monarchs)]. These sites have high numbers of males during autumn and at the end of overwintering, likely reflecting differences in male and female behavior (James, 1993). Relative to North American monarchs, Australian monarchs spend a shorter time in overwintering aggregations (about 2–4 months, compared to 4–5 months in North America; James, 1993).

Monarchs in New Zealand are non-migratory (one study showed that <3% of recaptured butterflies had flown more than 20 km; Wise, 1980). However, these non-migratory monarchs form overwintering clusters, using Quercus spp., Eucalyptus spp., Cedrus libani, and other species of trees in locations that are both sunny and sheltered (Ramsay, 1964). These sites have nectar sources used by adults, and the colonies vary in size from tens of monarchs to thousands (Ramsay, 1964).

#### Genetics

The genetics of monarch populations worldwide reflect the widespread variation in dispersal ability, gene flow, and both genetic and allelic diversity that contribute to the adaptive capacity of the species. Genetic results from genome-wide analyses (Zhan et al., 2014) and microsatellite data (Pierce et al., 2014a, 2015) suggest multiple dispersal events from an ancestral North American population, reflecting the capacity for the species to repeatedly expand its range to take advantage of new geographies and resources. Information on gene flow also reflects the wide adaptive capacity of monarchs, as they differentiate and persist at both low and high levels of gene flow. For example, low genetic differentiation among North American monarchs suggest high levels of gene flow between eastern and western populations, even though they display differing migratory behavior (Pierce et al., 2015). Similarly, monarchs from Pacific and Hawaiian Islands or Spain, Portugal, and Aruba also have the capacity to persist and differentiate in their behaviors despite low levels of gene flow that result in more highly genetically differentiated populations (Pierce et al., 2014a). In addition, monarch populations have been established and have persisted across a wide range of genetic and allelic diversity. North American monarchs have higher allelic and genetic diversity and allelic richness than monarch populations in the Pacific and this disparity increases the further the populations exist away from North America (Pierce et al., 2014a).

The adaptive capacity of monarchs worldwide is reflected through the genetically-based estimates of numbers of distinct populations. Pierce et al. (2015) estimate three genetic populations: North America, island populations, and Ecuador, based on genetic clustering analysis of microsatellite markers. Population structure analyses in Pierce et al. (2014a) support a total of seven worldwide populations: North America (including USA, Mexico, Costa Rica, Belize, Puerto Rico, and Bermuda), South America (Ecuador), Aruba, Spain, Portugal/Morocco, the Hawaiian Islands, and a series of Pacific islands (including Australia and New Zealand). Furthermore, Zhan et al. (2014) found genetic distinction among all geographically sampled locations (17), but also reported structure analyses that resulted in 2–11 populations. Additionally, recent research indicates that there may be genetic differentiation between migratory and nonmigratory Mexican monarchs (Pfeiler et al., 2017). Microsatellite analyses of monarchs in several locations in the Pacific (Australia, New Zealand, New Caledonia, Fiji, and Samoa) indicate that these monarchs are genetically distinct from other areas and have lower allelic diversity than North American monarchs (Shephard et al., 2002; Pierce et al., 2014a). While specific findings vary, these results show that monarchs worldwide reflect complex population structuring based on extensive dispersal and varying levels of gene flow, which create a wide capacity for the species to adapt to novel environmental and ecological changes in the future.

#### Population Size and Trends

Despite limited information on population estimates and population size trends outside of North America, there is potential adaptive capacity for the species resulting from monarchs persisting at a range of population sizes worldwide. James (1993) notes that Australian overwintering colony cluster sizes from 1978 to the 1990s were much smaller than those reported in the 1960s (with maximum numbers of 3,500 between 1978 and the 1990s, and ∼40,000 as a maximum number in the 1960s). There are also reports of Pacific Islands having boom and bust cycles when first colonized, with monarchs quickly becoming very common before defoliating available host plants (and perhaps moving on to colonize nearby islands; Zalucki and Clarke, 2004).

# DISCUSSION

As summarized here, the range of characteristics of worldwide monarchs contribute to the species' adaptive capacity. We know that variation occurs in monarch morphology, with butterflies having varying wing shape, color, and length throughout their range. This variation may be associated with differing migration behavior, which also varies throughout occupied habitat, from the long-distance migrations of North America, to populations exhibiting partial migration, and island monarchs that are now non-migratory. As monarchs have expanded their range, larvae still need suitable milkweed host plants, but the milkweed species (and sometimes genera) can vary from what is common in North America. Monarchs have a suite of natural enemies that they encounter throughout the world. OE is one of the more well-studied natural enemies, with infection and virulence rates varying greatly between populations. The genetics of monarchs throughout their range provide insight into how monarchs expanded and now differ from region to region. By having healthy populations with a variety of these characteristics, the species has more capacity to adapt to future changes in its physical and biological environment.

While there is much that we know about monarchs living outside of the migratory North American populations, many areas of study would benefit from further research. One of the biggest information gaps is the lack of basic information on monarch locations, numbers, and trends throughout much of their range. This information is particularly important with the documented declines of the North American populations. Additionally, without more surveys, it is difficult to delineate range boundaries for many countries. For example, monarchs are presumably not located on every one of Indonesia's over 17,000 islands, but they are probably on multiple islands. Therefore, island nations where we shaded in the entire country likely overrepresent the monarch range (**Figure 1**).

It is also important for us to better understand threats and their impacts for monarchs worldwide. While some threats are known and discussed above, many are not. Threats that impact monarchs in North America (including habitat loss, insecticides, and climate change, among others; Belsky and Joshi, 2018) may also impact non-North American monarchs in similar or different ways. For example, climate change may impact the suitable breeding range of monarchs in ways similar to North America (Lemoine, 2015), but it may also impact monarch-occupied habitat on Pacific Islands through sea-level rise ([IPCC] Intergovernmental Panel on Climate Change, 2014). These future research efforts, both on novel threats and on monarch locations and population trends, can provide a better understanding of monarchs worldwide, including their contribution to the adaptive capacity of the species, and their likelihood to persist into the future.

# AUTHOR'S NOTE

The findings and conclusions in this article are those of the authors and do not necessarily represent the views of the U.S. Fish and Wildlife Service.

# AUTHOR CONTRIBUTIONS

KN, LD, and KV contributed to the initial conception and design of this review. LD and KN worked on worldwide analyses. KV and KN completed literature reviews of the worldwide differences. KN wrote the manuscript. All authors contributed to manuscript revisions and approved the submitted version.

# ACKNOWLEDGMENTS

We thank Erik Olson for his work on the species distribution map. We greatly appreciate the feedback we received from experts on an earlier draft of this paper, including Karen Oberhauser, Stephen Malcolm, Sarina Jepsen, Pablo Jaramillo López, and Chip Taylor, as well as the many reviewers of the SSA draft. We are also extremely grateful for the valuable insight of our SSA team, who helped us define and refine what was

#### REFERENCES


needed for a review of monarchs worldwide. We additionally thank two reviewers Jeffrey Marcus and Ayse Tenger-Trolander for their useful feedback, and to citizen scientists for uploading information on monarchs worldwide.


**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 Nail, Drizd and Voorhies. 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.

# Rapid Assessment of Roadsides as Potential Habitat for Monarchs and Other Pollinators

Alison B. Cariveau<sup>1</sup> \*, Erik Anderson<sup>2</sup> , Kristen A. Baum<sup>3</sup> , Jennifer Hopwood<sup>4</sup> , Eric Lonsdorf <sup>5</sup> , Chris Nootenboom<sup>5</sup> , Karen Tuerk <sup>1</sup> , Karen Oberhauser 1,6 and Emilie Snell-Rood<sup>7</sup>

*<sup>1</sup> Monarch Joint Venture, Department of Fisheries, Wildlife, and Conservation Biology, University of Minnesota, St. Paul, MN, United States, <sup>2</sup> Environmental Incentives, Denver, CO, United States, <sup>3</sup> Department of Integrative Biology, Oklahoma State University, Stillwater, OK, United States, <sup>4</sup> Xerces Society for Invertebrate Conservation, Portland, OR, United States, 5 Institute on the Environment, University of Minnesota, St. Paul, MN, United States, <sup>6</sup> University of Wisconsin-Madison Arboretum, University of Wisconsin, Madison, WI, United States, <sup>7</sup> Department of Ecology, Evolution and Behavior, University of Minnesota, St. Paul, MN, United States*

#### Edited by:

*Cheryl Schultz, Washington State University Vancouver, United States*

#### Reviewed by:

*Nathan L. Haan, Michigan State University, United States Katherine Kral-O'Brien, North Dakota State University, United States*

> \*Correspondence: *Alison B. Cariveau alison.cariveau@gmail.com*

#### Specialty section:

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

Received: *14 January 2019* Accepted: *24 September 2019* Published: *16 October 2019*

#### Citation:

*Cariveau AB, Anderson E, Baum KA, Hopwood J, Lonsdorf E, Nootenboom C, Tuerk K, Oberhauser K and Snell-Rood E (2019) Rapid Assessment of Roadsides as Potential Habitat for Monarchs and Other Pollinators. Front. Ecol. Evol. 7:386. doi: 10.3389/fevo.2019.00386* Sustaining native pollinator populations and reversing declines in species such as the monarch butterfly (*Danaus plexippus*) will require enhancing and maintaining habitats across many regions and land use sectors. Rights-of-way, such as the areas surrounding roads, have long been regarded as important habitat for pollinators due to their ubiquitous nature and management for herbaceous species including nectar plants and larval host plants. With better information regarding the quality of pollinator habitat in roadside rights-of-way, managers can identify the location of potential habitat and evaluate the effects of management activities. We conducted a survey of roadside managers to determine needs and limitations related to assessing and managing rights-of-way as monarch habitat. Survey results indicated that managers are often limited by time, funding, and expertise in plant identification. Based on survey results and consultations with roadside managers, we developed a protocol for rapid assessment of roadside rights-of-way (hereafter, Rapid Assessment) that can be easily implemented by managers and is flexible based on the expertise of the observer and the data needs of the roadside management authority. Using readily available software, the field data are automatically processed through a Roadside Monarch Habitat Evaluator to generate habitat quality scores that may be used by managers to describe the habitat resources and to inform management strategies. We field-tested the protocol at roadsides in Minnesota and compared results with a more intensive protocol for monarch habitat monitoring (the Integrated Monarch Monitoring Program). We found that the Rapid Assessment provided similar data as the more intensive protocol regarding milkweed densities, nectar plant species richness, and monarch use of sites (eggs and larvae, when detection levels were sufficient). Observed high habitat values in roadside rights-of-way confirm the potential of such habitat for pollinator and monarch conservation.

Keywords: rights-of-way, roadside vegetation management, habitat assessment, butterflies, milkweed, nectar, host plants, Danaus plexippus

# INTRODUCTION

Monarch butterflies (Danaus plexippus) are an important flagship species for insect conservation. Monarchs, insect pollinators, and indeed most insect species, have experienced steep population declines in recent decades (National Research Council, 2007; Cameron et al., 2011; Brower et al., 2012; Vidal and Rendón-Salinas, 2014; Goulson et al., 2015; Semmens et al., 2016; Hallmann et al., 2017; Schultz et al., 2017; Sánchez-Bayo and Wyckhuys, 2019). Multiple factors are driving monarch declines (Malcolm, 2018), but habitat loss is primary (Flockhart et al., 2015; Thogmartin et al., 2017a,b) and the United States, Mexico, and Canada have pledged to reverse declines by improving and expanding habitat (CEC, 2008; Pollinator Health Task Force, 2015). Two important components in monarch habitat are nectar sources for adult monarchs, provided by a wide variety of blooming plants that benefit pollinators in general, and plants for larval development, provided by plants in the milkweed subfamily (Apocynaceae: Asclepiadoideae), which are also important nectar plants for many insect pollinators. Demographic models of the North American eastern monarch population indicate that the breeding season is likely the phase of the monarch life cycle that contributes most to population dynamics (Flockhart et al., 2015; Oberhauser et al., 2017) and loss of milkweed in the core of its breeding range is implicated in population declines (Pleasants and Oberhauser, 2013; Pleasants, 2017; Thogmartin et al., 2017a,b; Zaya et al., 2017; Stenoien et al., 2018). This has led to the goal of adding 1.3–1.6 billion stems of milkweed in the United States to increase the monarch population to sustainable levels (Pleasants, 2017; Thogmartin et al., 2017a). To reach this goal, habitat conservation is needed across all land use sectors (e.g., agriculture, developed areas, rights-of-ways), not just in lands set aside for conservation (Thogmartin et al., 2017b).

Rights-of-way may provide suitable pollinator habitat if managed in ways that promote and maintain host and nectar plants (Munguira and Thomas, 1992; Ries et al., 2001; Saarinen et al., 2005; Hopwood, 2008; Skorka et al., 2013; Halbritter et al., 2015), although concerns exist about dangers from roads (Munoz et al., 2015) including collisions (McKenna et al., 2001; Skorka et al., 2013; Keilson et al., 2018) and chemical runoff (Kaspari et al., 2010; Snell-Rood et al., 2014). A growing number of transportation agencies have implemented pollinator habitat programs (e.g., Iowa Living Roadway Trust Fund, Illinois DOT Monarch Program, Monarch Highway, Ohio Pollinator Habitat Initiative), and best management practices have been developed for pollinator habitat in roadside rightsof-way (Hopwood et al., 2015, 2016a,b). However, critical information about the availability of milkweeds and nectar plants within rights-of-way habitats is largely missing (but see Hartzler and Buhler, 2000; Kasten et al., 2016; Pitman et al., 2018), both generally and specifically within roadside management authorities.

Roadside managers need information to decide where to invest limited resources for maintaining and developing additional monarch habitat, and data on how various management actions affect the extent and quality of monarch habitat within their jurisdictions. For example, mowing is needed to maintain safety strips along road margins and is used to control woody and invasive species. However, frequently mowed areas often have fewer species of blooming nectar plants (Halbritter et al., 2015), and mowing can detrimentally impact insects using mowed areas (Johst et al., 2006; Cizek et al., 2012). However, mowing can also stimulate growth of new milkweed leaves preferred by egg-laying monarchs (Baum and Mueller, 2015; Fischer et al., 2015; Alcock et al., 2016; Haan and Landis, 2019; Knight et al., 2019). Many roadside management authorities are implementing reduced mowing practices particularly when monarchs are breeding in their regions to protect habitat for monarchs and other pollinators. These managers are interested in assessing the habitat characteristics created by such programs. In addition, data are needed for landscape-level planning and broad conservation efforts such as the Mid-America Monarch Conservation Strategy (MAFWA, 2018) and the USFWS Monarch Conservation Database<sup>1</sup> .

We developed several tools to help rights-of-way managers develop, assess, and manage monarch habitat. Here we present a rapid field assessment methodology, the Rapid Assessment of Roadside Habitat for Monarchs ("Rapid Assessment"), designed for quick and easy implementation by rights-of-way vegetation managers and maintenance operators. The data from the Rapid Assessment automatically feeds into a habitat calculator that generates a habitat quality score for each site; the package together is the Roadside Monarch Habitat Evaluator.

To guide the design of the Rapid Assessment, we surveyed transportation managers to learn about their interest in pollinator habitat programs, their information needs, and the personnel resources that may be dedicated to habitat assessment. To calibrate the new Rapid Assessment, we collected data from the same roadside sites using both our rapid assessment protocol and a more intensive protocol from the national Integrated Monarch Monitoring Program<sup>2</sup> (CEC, 2017; Cariveau et al., 2019; IMMP). Specifically, we compared results from the Rapid Assessment to those from the IMMP for milkweed densities, nectar plant species richness, indices of nectar plant abundance, and monarch observations (eggs and larvae). We were interested in whether both protocols would yield similar estimates for these key metrics, and in the correlation of measures from the two protocols.

In this paper, we explain features of the Rapid Assessment that facilitate its use by roadside managers in transportation departments. We relate these findings to other studies and discuss the results in the context of managing rights-of-way as pollinator habitat. We additionally provide, as **Supplemental Material**, the Rapid Assessment protocol and datasheet. The User Guide for the Roadside Monarch Habitat Evaluator that enables roadside practitioners to run it with standard Esri products is provided online<sup>3</sup> .

<sup>1</sup>https://www.fws.gov/savethemonarch/MCD.html

<sup>2</sup>https://monarchjointventure.org/immp

<sup>3</sup>https://monarchjointventure.org/roadsidehabitat

# MATERIALS AND METHODS

#### Survey of Managers

We created a 30-question survey about desired management tools in Qualtrics<sup>4</sup> that we distributed to a network of roadside management authority representatives via email. The survey was reviewed by the Institutional Review Board at the University of Minnesota and determined not to constitute human subjects research, therefore not requiring IRB approval. It included questions about existing pollinator habitat programs; what types of information would be helpful for planning or implementing these programs; the availability of data about factors that could influence pollinator habitat quality, including noxious weeds, salt applications, mowing regimes, and herbicide applications; and manager interest in tracking management practices. The survey captured information about personnel resources available for conducting habitat assessments, including the number of people and number of days they could spend assessing habitat, and the expected skill levels of the personnel relative to assessing habitat. Answers were mostly categorical with some free response.

Semi-structured interviews with a subset of survey respondents who indicated that their organizations have established or were considering establishing pollinator habitat programs were held to elicit further input, better understand the context in which roadside managers make decisions, identify barriers to establishing habitat programs, and evaluate the usefulness of tools such as a Rapid Assessment protocol in managing roadside rights-of-way as habitat.

# Design of Roadside Monarch Habitat Evaluator

We designed a Rapid Assessment protocol to assess rightsof-way as pollinator habitat, with an emphasis on monarchs. The protocol includes information on road type, adjacent land use, management practices, forb species richness and percent cover, noxious weed presence and percent cover, and milkweed species richness and abundance (**Table 1**; field data sheet and protocol instructions provided in **Supplemental Material 1**). We developed both a paper data sheet and an electronic data form that could be filled using a tablet or smartphone in the field.

Secondly, we developed a habitat calculator that automatically computes habitat quality scores from the data collected in the Rapid Assessment. Together the Rapid Assessment and Habitat Calculator form the Roadside Monarch Habitat Evaluator (hereafter, "Habitat Evaluator").

When developing the Habitat Evaluator, and in collaboration with the Rights of Way as Habitat Working Group facilitated by the Energy Resources Center of the University of Illinois – Chicago, we reviewed more than a dozen existing assessment tools including the Monarch Habitat Quantification Tool (Environmental Defense Fund, 2017), the Solar Site Pollinator Habitat Planning and Assessment Form (Minnesota Board of Soil Water Resources, 2016), and Bee Better Certified Farm Management Assessment Guide (Xerces Society for Invertebrate Conservation 2015). While none of these tools were created for use by transportation managers, they provided examples of ways in which pollinator habitat attributes were compiled into scores.

We designed the Habitat Evaluator tool in Survey123 for ArcGIS (Esri), a free product that affords several benefits for roadside management authorities. States or other entities can collect, manage, and view their own datasets using their own Esri Enterprise account. The Habitat Evaluator is installed within each agency's ArcGIS Online platform, when it is populated with a plant list for their state. Then managers may customize their assessment by selecting the noxious weeds they wish to track and set default answers regarding herbicide use and mowing practices, if desired. Within their own Survey123 website, transportation managers can view site locations, field data, and monarch habitat quality scores. A User Guide to the Roadside Monarch Habitat Evaluator is online<sup>3</sup> .

The electronic form of the Rapid Assessment provides the field user advantages such as the ability to automatically record the location, date, and time of the assessment. The survey also provides features such as a searchable drop-down list of plant species that enables one to type in letters from either the common name or the Latin name to select the species. It also includes choices based on genera, such as "Solidago/goldenrod species" for many groups. The assessment is flexible in that observers may also tally plant types they cannot identify and choose to estimate milkweed plant abundance in categories rather than count individual plants (e.g., depending on the abundance of the milkweed and time constraints). Observers also specify whether they are assessing the full right-of-way or just the unmowed areas, and whether or not they wish to collect optional data regarding the presence of monarch eggs and larvae. We incorporated several factors identified as important to roadside managers, including the need to assess sites quickly and once per growing or monarch breeding season, the ability to specify weeds of local or state importance, and the ability to specify the width of the area to be surveyed with regard to mowed areas, each of which we describe subsequently.

Given the strong preference of roadside managers for a protocol that could adequately characterize the habitat quality of a site in a single visit per year, we required a proxy for the availability of nectar throughout the growing season. We defined a term "Potentially Blooming Nectar Plants" (hereafter "nectar plants") to describe forbs and shrubs that could provide nectar to pollinators (e.g., excluding grasses), whether or not blooming on the date of assessment. This broad categorization encompasses plants that may provide nectar, regardless of their nativity or the amount or quality of nectar they may provide. The numbers of nectar plant species may be important because a greater number of species may represent a greater variety of bloom times and thereby provision nectar for a greater proportion of a season of monarch use or use by other pollinators. We identified plants to species when possible and also estimated the aerial percent cover of nectar plants as a group. To make the protocol usable for people with varying skills in identifying plants to species, we included an option for tallying unidentified types of plants.

To accommodate variation in the list of invasive species, weeds, or non-native species of management concern from

<sup>4</sup>Qualtrics Version 12/17 © 2107; Available online at: https://umn.ca1.qualtrics. com/



*<sup>a</sup>CROP, cropland, no barrier; HCR, Crop with woody barrier or hedgerow; DEV, Developed, lawn, or paved; HDE, Developed with woody barrier or hedgerow; DIV, Diverse grassland/natural habitat; NDI, Not diverse grassland with few forbs; WOOD, Woody habitat; WET, Wetland habitat.*

*<sup>b</sup>Potential Blooming Nectar Plants (PBNP) are forbs and shrubs that can provide nectar for monarchs or other pollinators, whether or not blooming on the survey date.*

*<sup>c</sup>Weeds we define to be of management interest by the transportation authority; may include noxious weeds and other invasive species under active surveillance or management. <sup>d</sup>ROW, right-of-way.*

state to state, we created a customizable weed list. When transportation managers initially set up the protocol for their organization, they can list the weed species they want to include in the assessment. Observers will then report whenever those species are present on the assessment areas and estimate aerial cover for those species as a group to describe their prevalence.

Our survey of roadside managers indicated that the frequency and widths of mowing in the rights-of-way were highly variable; some routinely mow the full right-of-way width multiple times per growing season, while others mow the full right-of-way only once every several years. Some mow a safety strip (e.g., first 10– 12 feet) monthly during the growing season, while others mow the strip only once per year (and some do not mow from May-July for wildlife and pollinators). Furthermore, some roadside managers expressed interest in using the Rapid Assessment to gain information about the effects of their mowing practices on pollinator habitat. In the first year of the project, we collected data across the entire right-of-way. In the second year, we focused our estimates of cover on the unmowed area for qualitative measures (such as percent cover) and collected milkweed and nectar plant richness in both the mowed and unmowed areas, which we sum in these analyses. The final Rapid Assessment protocol allows surveyors to choose whether to conduct their assessments in full rights-of-way, unmowed areas, or in mowed and unmowed areas separately.

Finally, because some departments of transportation were interested in monarch breeding activity in their roadside areas, we included optional fields for recording monarch eggs, larvae, and adults. This section also includes a place to record the species and number of milkweed plants searched.

We field-tested the Rapid Assessment protocols with representative users from the Illinois, Minnesota, and Wisconsin Departments of Transportation at sites that depicted high quality conditions, such as prairie remnants, as well as sites where restoration activities had been completed, to gain further feedback and refine the protocols and data forms.

#### Design of the Habitat Calculator

The Habitat Calculator is derived from the Monarch Habitat Quantification Tool (Monarch HQT, Anderson et al., 2017). The Monarch HQT is based on a modified Habitat Evaluation Procedure (HEP, see US Fish Wildlife Service, 1981) in which habitat characteristics (e.g., milkweed density) are translated to quality scores using suitability indices. Suitability indices approximate the relationship between a given habitat characteristic at a location and the location's suitability for monarchs. Suitability indices are weighted and summed to develop a Habitat Suitability Index (HSI), or habitat quality score. Habitat characteristics identified for important functional components of monarch habitat include breeding habitat (milkweed), foraging habitat (nectar plants) and factors that influence monarch habitat, including threats such as pesticide drift from agricultural fields.

For the Roadside Monarch Habitat Evaluator, the habitat characteristics evaluated were modified to match the data collected through the Rapid Assessment and expanded to include factors relevant to roadside rights-of-way. For example, the Rapid Assessment uses ocular estimates of cover of potentially blooming nectar plants whereas the Monarch HQT captures frequency of blooming nectar plants. The suitability indices were adapted as necessary based on expert opinion. In addition, the Habitat Evaluator includes additional indices of threats specific to roadside rights-of-way, including risk of collision with vehicles and chemical runoff and invasive weeds that may displace vegetation contributing to habitat quality. Finally, the Habitat Evaluator also incorporates vegetation management, including mowing and herbicide use. Measures of each variable are weighted and summed to produce a habitat quality score (see the online User Guide to the Monarch Roadside Habitat Evaluator<sup>3</sup> ).

#### Rapid Assessment Field Technique

Rapid Assessments are completed for a 45.7 m (150 ft) length of roadway, implemented at random locations or systematically (e.g., every km or ten km) in a road system (see protocol in **Supplemental Material 1**). Upon arrival at a location of interest, the observer walks parallel to the road, toward traffic, pacing the 45.7 m distance (**Figure 1**). Next, the width of the vegetated rightof-way (perpendicular to the road) is measured or estimated (e.g., paced). These two distances bound the rectangular assessment area that extends from the road to the back of the right-of-way. The observer walks back through the right-of-way to the starting point, systematically zigzagging back and forth throughout the roadside habitat, while recording data. The observer records the number of milkweed plants by species, where stems separated by soil are counted as plants regardless of whether they are clonal or genetic individuals (following Kasten et al., 2016; CEC, 2017), the species or number of nectar plants (and notes for each species if it is blooming or not), and the presence of weeds (as defined by their roadside organization). Percent aerial cover is also estimated by classes for potential nectar plant species collectively (regardless of whether currently blooming) and for weeds of concern. In 2018 observers also estimated the percent cover by flowers for comparison to the IMMP blooming plant frequency. The observer records the dominant adjacent land use and mowing and herbicide application information. As an option, observers may examine milkweed plants by species for monarch eggs and larvae, recording the number of plants searched, and number of eggs and larvae detected. To maintain efficiency when milkweed is abundant, observers may choose to monitor every 2nd, 3rd, or 5th milkweed plant encountered to gain a sample size of 50–100 milkweed plants searched per site.

# Integrated Monarch Monitoring Program Methods

IMMP sampling employs a total of 100 quadrats placed along ten transects arrayed diagonally from the road edge to the back of the right-of-way along a 400–500 m length of roadway (see **Figure 2**). Transects are 50 m in length and quadrats are placed every 5 m (however, in 2017, we placed quadrats every 2 m along 25 m transects, with 25 m between each transect). Quadrats consist of a 1.0 m by 0.5 m sampling frame placed to either side of the transect line for a 2.0 m by 0.5 m or 1 m<sup>2</sup> quadrat area. Within each quadrat, observers count milkweed plants (same definition as above) to estimate milkweed density (milkweed plants/ha). All blooming plants are identified to species and assigned to the first subplot (area within the quadrat) in which they occur (first 0.5 x 0.5 m, 1.0 x 0.5 m, or 2.0 x 0.5 m) to generate a frequency score (proportion of subplots occupied). Plants that are not blooming on the date of the assessment are not recorded. The IMMP protocol is available on its website<sup>2</sup> .

# Field Trials to Compare Habitat Assessment Techniques

For 2017 field trials, we chose 14 sites from a set of randomly selected roadside sites in Minnesota that had been surveyed for milkweed and monarchs in 2015 (**Figure 3**; Kasten et al., 2016). We selected sites that contained milkweed in 2015. In 2018, we selected 15 new sites through the IMMP, which uses generalized random tessellated stratified sampling (GRTS) to identify random 10 x 10 km blocks and random point locations within them stratified by land use sector and prioritized to accommodate for variable inclusion probability (Cariveau et al., 2019). Sites in 2018 were randomly selected using the GRTS list of point locations; 13 sites were within the 15 highest ranked blocks in Minnesota (with vegetated roadsides at least 4 m wide) plus two additional sites within the 25 highest ranked blocks, for a total of 15 sites. Sites in 2018 also needed to have a minimum of 4 m width of vegetation in GIS preview for inclusion. Sites in both years represented variation in roadway types (except freeways which were excluded due to safety concerns).

To account for the different sizes of the survey areas for each protocol, at each of these sites we completed one IMMP survey and typically three Rapid Assessments spaced 200–250 m apart within the footprint of the IMMP site (**Figure 2**). One site in 2017 had four Rapid Assessments and one site had only two; in 2018 three sites had only two Rapid Assessments.

### Statistical Analyses

We calculated milkweed plants/ha based on the number of milkweed plants counted (all species combined) and the area searched at each site and converted to hectares. For the IMMP, the area searched was 100 m<sup>2</sup> based on the 100 1-m<sup>2</sup> quadrats. For the Rapid Assessment, the area searched was estimated as 45.7 m (the length of the plot) multiplied by the right-of-way width.

We present monarchs/plant as the sum of all monarch eggs and larvae observed, divided by the number of milkweed plants searched. For the IMMP protocol, the number of milkweed plants searched differed from the number of milkweed plants in

the density estimate, because observers could search additional milkweed plants between the quadrats to look for monarch eggs and larvae. We focused analyses on sites with at least 10 milkweed plants examined by each method to ensure robustness of our monarchs/plant estimates (10 sites in 2017; 11 sites in 2018). We also estimated monarchs/ha by multiplying the average number of monarchs/plant times the average number of milkweed plants/ha using the IMMP method.

To represent nectar resource availability, we compared two indices: species richness and abundance. For species richness, we compared the number of blooming species. For the IMMP protocol, this is a list of all blooming species encountered in the quadrats. For Rapid Assessments, in 2017, we listed all of the blooming plant species encountered; in 2018, we identified all of the potentially blooming nectar plants and noted whether or not plants were blooming. Here we present the blooming subset to compare to the IMMP data. The nectar plant species lists across the several Rapid Assessments (RA) for each IMMP site were combined in two ways. First, the number of blooming species was determined for each RA, and then the number averaged across the several RA for each IMMP survey location; we call this RA averaged. Second, because of known relationships between species richness and area, we also depict the number of blooming species determined when summing the species across the RAs for each IMMP site (removing duplicates), which we call RA summed. For abundance, we compared the frequency of blooming nectar plants from the IMMP (number of quadrats out of 100 in which at least one blooming nectar plant was present) to percent cover by flowers from the Rapid Assessment, for 2018, the only year in which we estimated cover (averaged across the multiple Rapid Assessments per site).

We computed statistics using R version 3.5.1 (R Core Team, 2018). For milkweed plants/ha, and monarchs/plant, we compared the mean of the two to four Rapid Assessments to the IMMP measure for each site. To determine if protocol type had a significant effect on response variables, we ran generalized linear mixed models with year and protocol type as fixed effects and site as a random effect for each of the response variables of milkweed density, monarchs/plant, and number of blooming species ("nlme" package; Pinheiro et al., 2018). We report an interaction term for year and protocol type when significant. The sample size was 113 visits to 29 sites for the plant data; because we found no milkweed plants during 17 visits, the model for monarchs per plant contained 96 visits to 29 sites. For number of blooming species, we compared the estimates by the IMMP protocol to the RA averaged and RA summed in a generalized linear mixed model with year and protocol type as fixed effects, site as a random effect, and a year by protocol type interaction effect. For clarity, we also compare the numbers of blooming species by the IMMP protocol to the RA averaged and RA summed for each year separately. For milkweed density, monarchs/plant, nectar plant richness, and nectar plant abundance, we also compared the mean of the Rapid Assessments per IMMP site to the IMMP measure

with a correlation coefficient. If variables met the Shapiro-Wilk normality test for normality, we computed a Pearson correlation; if they did not, then we used a Kendall rank correlation. We plotted data in Excel and ggplot2 (Wickham, 2016).

# RESULTS

### Manager Survey Results

We received 79 responses to the survey; with respondents representing states (58%), counties (25%), regional or national entities (8%), local entities (9%), and other entities (5%) in19 states: Arizona, Arkansas, California, Illinois, Indiana, Iowa, Kansas, Maryland, Michigan, Minnesota, Nebraska, New Hampshire, Ohio, Oklahoma, South Dakota, Texas, Virginia, Washington, and Wisconsin. Respondents from 14 (74%) of the states from which we received responses indicted that they had a pollinator program. We asked if managers would like guidance about where to install or manage monarch habitat, tools for monitoring that habitat, or both. Of 33 respondents to this question, 39% wanted monitoring methods, 12% indicated that the planning information would be most valuable, and 39% wanted both (9% had other answers). We report their answers to questions about capacity for field work and management practices in **Table 2**.

#### Field Surveys

In 2017, we assessed 14 sites between June 29 and August 22. All sites were located along paved roads, eleven along 2-lane roads, and three along 4-lane roads. Eight sites were adjacent to cropland, with two sites each by woodland, grassland, and developed land. Right-of-way widths from the Rapid Assessments varied from 3 to 21.5 m (mean = 12.35 m, standard deviation (SD) = 3.71); widths were not recorded by the IMMP protocol in 2017.

In 2018, we surveyed 15 sites between July 23 and August 29; all sampled sites were along two-lane roads; 12 were paved; and three were dirt/gravel. In 2018, adjacent land uses included: cropland (7), woodland (3), grassland (2), and wetland (3). The widths of the rights-of-ways by Rapid Assessments varied from 5 to 52 m (mean = 14.07, SD = 12.79). The average width of IMMP rights-of-way in 2018 was 9.43 m (SD = 3.70, range 3.5–19.5 m).

Single Rapid Assessments took an average of 22 min in 2017 (SD = 15 min; range 4–88 min) and 20 min in 2018 (SD = 12 min; range 5–59 min). IMMP visits took 134 min on average (SD = 67 min; range 68–345 min) in 2017 and 167 min in 2018 (SD = 56 min; range 92–274 min). Variation in the duration of visits was affected by the number of nectar plant species present and the number of milkweed plants counted and examined for monarch eggs and larvae.

### Milkweed Density

We detected milkweed at all sites in 2017 and 14 of the 15 sites (93%) in 2018 using the IMMP protocol. The vast majority of milkweed was Asclepias syriaca (common milkweed; 96%); other species were A. incarnata (swamp milkweed, 3%), A. verticillata (whorled milkweed, 0.69%), A. sullivantii (Sullivant's milkweed, 0.2%), and A. tuberosa (butterfly weed, 0.01%). The mean milkweed density for all species of milkweed combined using the IMMP protocol was 1,242 plants/ha (SD = 1,303) in 2017, 2,807 plants/ha (SD = 4,864) in 2018, and for both years combined: 2,052 plants/ha (SD = 3,639; median = 800; range 0–18,000) (**Figure 4A**). Averaging the RAs per site, the mean milkweed density for all species of milkweed across sites in 2017 was 1,508 plants/ha (SD = 2,082), 1,545 plants/ha (SD = 2,377) in 2018, and 1,527 plants/ha for years combined (SD = 2,199; median = 625; range 0–8,966). Milkweed density did not vary with year (t<sup>27</sup> = 0.415, p = 0.681) or survey type (t<sup>83</sup> = −0.639; p = 0.524, df = 83). Milkweed density as estimated by the two protocols was correlated (Kendall's rank correlation tau = 0.568, z = 4.257, df = 27, p < 0.001; see **Figure 5A**).

### Monarch Eggs and Larvae

The mean number of milkweed plants searched for monarch eggs and larvae in 2017 was 40.93 (SD = 47.66) with the IMMP and 76.11 (SD = 91.15) with the RA. In 2018 the mean number of milkweed plants searched for monarch eggs and larvae was 113 (SD = 134.48) with the IMMP and 36.27 (SD = 44.38) with the RA. In 2017, using the IMMP method, monarch eggs or larvae were found at 6 of 14 sites (43%); with the RA monarch eggs or larvae were found at 7 of 14 sites (50%). In 2018, using the IMMP method or the RA, monarch eggs or larvae were found in 11 of 15 sites (73%), or in 11 of 14 sites containing milkweed (79%). If considering RAs independently from one another, then in 2017, monarch eggs or larvae were found in 11 of 42 (26%) RAs or 11 of 37 (30%) sites with milkweed, and in 2018, monarch eggs or



larvae were found in 19 of 42 (45%) RAs or 19 of 30 (63%) sites with milkweed.

For monarchs/plant, year was a significant factor (t<sup>27</sup> = 2.373, p = 0.025) with more eggs and larvae found in 2018 than 2017, but protocol type did not have a significant effect on estimates of monarch density (t<sup>66</sup> = 0.118; p = 0.906; **Figure 4B**).

When restricting analysis to sites with at least ten milkweed plants examined by each protocol, in 2017, monarch egg or larvae were found at 40% of the sites with the IMMP protocol and 50% with the RA protocol (summed per site; 10 sites). At five sites monarchs were found with the RA protocol but not by the IMMP; at three sites monarchs were detected by the IMMP but not by the RA. In 2017, the mean number of monarchs/plant with the IMMP protocol was 0.010 (SD = 0.014) and 0.011 (SD = 0.025) with the RA (**Figure 4B**). In 2017, the monarchs/plant estimated by the two protocols were not correlated (Kendall's rank correlation tau = −0.216; z = −0.762, p = 0.446; **Figure 5B**).

In 2018, monarch eggs or larvae were found at 82% of the sites with the IMMP protocol and 91% with the RA (summed per site; 11 sites); on one site monarchs were found with the RA method but not by the IMMP. In 2018, the mean number of monarchs/plant was 0.099 (SD = 0.105) with the IMMP and 0.153 (SD = 0.173) with the RA (**Figure 4B**). In 2018, monarchs/plant measured with the two protocols were correlated (Kendall's rank correlation tau = 0.661, z = 2.81, p = 0.005; **Figure 5B**).

An estimate of the average number of monarch eggs and larvae per ha, using the overall IMMP mean was 115 monarchs/ha (2,052 plants/ha<sup>∗</sup> 0.056 monarchs/plant) across both years. Separating the 2 years, for 2017, the estimate was 12 monarchs/ha (1242<sup>∗</sup> .010) and for 2018, 253 monarchs/ha (2807<sup>∗</sup> .099). Using RA averages, the overall estimate was 131 monarchs/ha (1527<sup>∗</sup> 0.086); for 2017 it was 17 monarchs/ha (1508<sup>∗</sup> 0.011) and 2018 it was 236 (1545<sup>∗</sup> 0.153).

#### Blooming Nectar Plants

The average number of blooming species per site in 2017 was 6.71 (SD = 4.50, range 1–18) with the IMMP protocol, 6.72 (SD = 2.56, range 1–12.33) with RA averaged, and 12.14 (SD = 4.45, range = 5–19) with RA summed (**Figure 6**). In 2018, the average number of blooming species per site was 10.40 (SD = 6.40, range = 1–23) with the IMMP protocol, 6.57 (SD = 2.85, range 2–11.33) with RA averaged, and 12.00 (SD = 5.35, range = 1–20) with RA summed (**Figure 6**).

Comparing the number of blooming species by IMMP to the RAs (taking each RA independently as in milkweed and monarch analyses), the significance of the factors in the model was as follows: year (t<sup>27</sup> = 2.33, p = 0.027), protocol type (t<sup>82</sup> = −0.047; p = 0.963), and protocol type by year interaction (t<sup>82</sup> = −2.86; p = 0.005). In 2017, the number of blooming species estimated by IMMP did not differ from the RA averaged (t<sup>26</sup> = 0.007, p = 0.995), but was lower than the RA summed (t<sup>26</sup> = 6.247, p < 0.001). In 2018, for the same comparison, IMMP results did not differ from RA summed (t<sup>28</sup> = 1.532, p = 0.136), but were higher than RA averaged (t<sup>28</sup> = −3.463, p = 0.002).

In 2017, the number of blooming species by IMMP protocol was correlated with RA averaged (Pearson's r = 0.706, t<sup>12</sup> = 3.453, p = 0.005) and RA summed (Pearson's r = 0.690, t<sup>12</sup> = 3.302, p = 0.006; **Figure 7A**). In 2018, the number of blooming species by IMMP protocol was correlated with RA averaged (Pearson's r = 0.801, t<sup>13</sup> = 4.829, p < 0.001) and RA summed (Pearson's r = 0.698, t<sup>13</sup> = 3.515, p = 0.004; **Figure 7B**).

The correlation of the estimate of percent cover by blooms in the Rapid Assessment plots (percent cover classes converted

sampling methodologies, the Integrated Monarch Monitoring Program (IMMP) and averaged values for 2–4 Roadside Habitat for Monarchs Rapid Assessment (RA) taken from the same sampling location. Milkweed density did not vary by protocol type (*t*<sup>83</sup> = −0.639; *p* = 0.524, *df* = 83) or by year (*t*<sup>27</sup> = 0.415, *p* = 0.681). Monarchs/plant did not differ by protocol type (*t*<sup>66</sup> = 0.118; *p* = 0.906) but year was a significant factor (*t*<sup>27</sup> = 2.373, *p* = 0.025). Mean values are indicated by the "x"; median values by a horizontal line, boxes indicate 25 and 75% quartiles, bars indicate the upper and lower quartiles, and outliers more than 1.5 the 75% quartile are depicted by dots.

to midpoints, then averaged per site) to the frequency of nectar plants based on the IMMP method was 0.53 (Kendall's tau; n = 14, z = 2.477, p = 0.013; **Figure 8**).

#### DISCUSSION

We designed and tested a Rapid Assessment protocol for monarch habitat within roadside rights-of-way. Observers focus on a small length along the roadway to count milkweed plants and types of nectar plants and estimate cover of nectar plants and noxious weeds. Rapid Assessment data are automatically calculated into habitat quality scores that provide information to

transformed, for sites monitored in 2017 (circle) and 2018 (triangle), using the Integrated Monarch Monitoring Program (IMMP) and Roadside Habitat for Monarchs Rapid Assessment (RA) averaged for each site. 95% confidence interval indicated in gray. (B) Monarch eggs and larvae per milkweed plant searched (monarchs/plant), log<sup>10</sup> transformed, for sites monitored in 2017 (circle) and 2018 (triangle), for the same two methodologies. 95% confidence interval indicated in gray. The correlation between techniques for 2017 was non-significant.

managers about their habitat resources, enable them to compare conditions across sites, and inform their management decisions. This is similar to other applications of simple vegetation assessment methods to support natural resource management. One example is identifying groups of plants of particular interest, such as cool-season grasses, in a 25 m x 0.01 m belt transect (Grant et al., 2004) to provide data inputs for a decision support tool for adaptive management of native prairie (Hunt et al., 2016). Pywell et al. (2011) combined vegetation metrics and management information including seed mix and mowing

practices to accurately predict use by bees and butterflies in the United Kingdom.

This project furthers conservation efforts for monarch butterflies by creating a tool that is tailored to the needs and preferences of state departments of transportation that manage an estimated 17 million acres of potential habitat for monarchs (Ament et al., 2014). Great attention has come to rights-ofways for their ability to provide habitat for monarchs, such as the effort to provide habitat in roadside and energy rightsof-way through a National Candidate Conservation Agreement with Assurances (CCAA; Cardno, 2019). As an indication of the interest in this project, personnel at the Delaware Department of Transportation implemented the Rapid Assessment at nearly 100 locations in the summer of 2018 to learn about monarch habitat along their roadways.

Through a survey and field visits with transportation personnel, we learned that a flexible survey design was needed to meet the departments' wide range of needs. We designed the assessment in Esri software typically used by transportation departments so that states could customize their assessments. For instance, some field staff are knowledgeable about vegetation and would like to quantify not only the number of nectar plant species present but also how many are native. Others are only able to quantify numbers of plants that look distinct from one another; we created a convenient lookup table from which a surveyor can pick plants from their state by either common or Latin names or simply tally unknown types. Departments differ also in the tracking of noxious weeds, from no tracking to extensive lists of species that differ state to state and sometimes by counties or bioregions within states, so we enabled the ability for road managers to specify a list of species they wish to

track. Because managers indicated that only a limited number of days and people would be available for assessments, we designed a survey to be conducted once per growing season. To accommodate the single yearly sample, we created the term "potentially blooming nectar plants" to represent all of the plants that could provide nectar for monarchs and other pollinators, regardless of whether they were blooming on the date of the survey. This is consistent with a pollinator scorecard being designed by the Rights-of-way as Habitat Working Group of the Energy Resource Center at the University of Illinois-Chicago (A. Cariveau personal communication).

This Rapid Assessment fits into a suite of habitat assessment tools for monarch butterflies but is the only one designed for ready application in the roadside context with specific consideration of the needs and constraints of transportation managers. The Natural Resources Conservation Service (NRCS) Monarch Butterfly Wildlife Habitat Evaluation Guides (WHEG) and Decision Support Tools<sup>5</sup> are designed to provide a qualitative rating of current monarch habitat and assess habitat management alternatives for working agricultural lands (USDA NRCS, 2018). The WHEG similarly focuses on milkweed presence and species richness of nectar producing plants, while focusing on specific plants known to be used by monarchs. The Habitat Quantification Tool is used to evaluate the quality of monarch habitat for protection and enhancement in a variety of land-use contexts including roadside rightsof-way (Environmental Defense Fund, 2017), but it is more time consuming to implement than the Rapid Assessment. Programs such as the Western Monarch and Milkweed Mapper<sup>6</sup> use metrics similar to those in our protocol (e.g., milkweed counts and monarch presence) but rather than characterizing particular locations, the goal is enhanced understanding of the distribution of monarchs and their habitats to inform conservation, like the Integrated Monarch Monitoring Program (Cariveau et al., 2019).

Our testing results suggest that the Rapid Assessment provides a standardized and rapid way to describe habitat conditions for monarch butterflies in roadside rights-of-way and produces similar results to those of a more intensive protocol. Outcomes from other rapid assessments are mixed. A rapid technique for characterizing habitat in agricultural fields predicted the overall abundance and richness of butterfly species in Britain, though performed less well for predicting occurrence of some particular species (Pywell et al., 2004). For the Fender's blue, rapid assessment of vegetation did not align with more detailed assessments of host and nectar plant availability in determining habitat suitability for this rare butterfly (Schultz and Dlugosch, 1999). Our study did not focus on relating use by monarchs to the habitat, but rather on habitat availability and the ability for the rapid assessment to concur with a more intensive quantification method, the IMMP. The IMMP is designed to track changes in habitats throughout seasons and across years and to compare monarch habitat quality and use across land use sectors. The IMMP collects additional data, including a quantitative survey for adult monarchs and nectar plant diversity, and could be used to address questions such as whether monarchs prefer particular nectar plant species. However, our results suggest that for assessing and comparing rights-of-way habitat to inform roadside habitat restoration and management for monarchs and other pollinators, the Rapid Assessment produces sufficiently similar results.

It should be noted that we only compared the Rapid Assessment and the IMMP within the range of the eastern monarch population. However, we developed the Rapid Assessment with input from road managers in both western and eastern states. While little is known about how the relationship between monarchs and their habitats may differ between the two main North American populations, we expect that the Rapid Assessment should effectively depict habitat conditions across roadside sites in western states. Individual state managers may adjust and customize the Roadside Monarch Habitat Evaluator tool for appropriate plants and scoring for their bioregions. Given the recent population levels of the western monarch (Pelton et al., 2019), we encourage use and adjustment of these tools to learn more about habitat availability and use by monarchs in western roadsides.

The Rapid Assessment is efficient; our two-person field crew completed assessments in an average of 21 min, including time spent looking for monarch eggs and larvae, which many departments of transportation will elect to skip. It was much faster than the IMMP even when repeating the protocol three times over the footprint of the IMMP (sum of 62 min as compared to an average duration of 2 ½ h). The Rapid Assessment also appears easier to learn and may be spread out to sample from a larger landscape in the same amount of time as the IMMP. In 1 day, a crew can complete 10–15 assessments. Experienced crews, after learning how to identify the plants in the rights-of-way, are likely to be faster than employees who are conducting assessments for the first time, but observers typically become faster through practice.

There may be concerns about whether road crews could effectively collect the data required for the Rapid Assessment. However, volunteers with no formal research training have effectively contributed biological data to a multitude of programs, such as the Breeding Bird Survey, which has produced excellent information about the status and trends of North American

<sup>5</sup>https://www.nrcs.usda.gov/wps/portal/nrcs/detail/national/plantsanimals/ pollinate/?cid=nrcseprd402207

<sup>6</sup>https://www.monarchmilkweedmapper.org/

birds (Hudson et al., 2017). Citizen scientist contributions were instrumental in years of research on butterflies in Britain (Roy et al., 2007) and in building a butterfly database in Florida (Jue and Daniels, 2015). For monarchs, citizen scientists have had a long history of contributing to research (Howard and Davis, 2009; Ries and Oberhauser, 2015), including a recent analysis of the population status of western North American monarchs (Schultz et al., 2017). In some studies, volunteer data were compared to data collected by researchers or by a more rigorous method. A study of stream monitors found high concurrence of data collected by volunteers and paid researchers (Fore et al., 2001), and collection of terrestrial invertebrate diversity data by volunteers and researchers were similarly satisfactory (Lovell et al., 2009). In contrast, volunteers were not very successful in identifying stream macroinvertebrates (Nerbonne and Vondracek, 2003). A comparison of pollinator data from citizen scientists and researchers found similar trends in detection for higher level bee taxa but not for detections of all species (Kremen et al., 2011). Thus, fine-scaled species identification is typically more difficult for non-researcher observers, but this should not pose a problem for roadside assessments that only rely on distinguishing types of plants and optionally identifying one distinctive butterfly. In particular to our protocol, volunteers successfully collect similar data on milkweed, monarchs, and nectar plants in the Monarch Larva Monitoring Project (MLMP)<sup>7</sup> and the IMMP. Furthermore, our field-testing of the protocol with three departments of transportation indicated that their personnel could effectively collect these data.

The Rapid Assessment was effective for measuring milkweed density, nectar plant species richness, and for monarch eggs and larvae per plant in 2018 (the measures were not significantly correlated in 2017 when monarch detections were low). In general, averaging parameter estimates for multiple Rapid Assessments yielded more consistent results than any single Rapid Assessment from sites, suggesting that combining multiple Rapid Assessments to characterize areas is preferred over single samples. This is similar to a comparison of rapid qualitative score to quantitative scores of vegetative condition, where there was broad association in the scores across many sites, but rapid assessments were not reliable at the level of a specific site (Cook et al., 2010). Furthermore, this underscores our recommendations that managers sample multiple sites. In particular, we note that it is important for managers to preselect random or systematic (e.g., every km or 1/2 km) sampling locations to effectively characterize larger areas without bias from sampling in locations where habitat quality is known to be or appears to be high.

When averaging Rapid Assessments across several sites, milkweed density estimates were not statistically different than those derived by the IMMP protocol, and the estimates by the two methods were correlated across sites (tau = 0.568; **Figure 5A**). Numbers of species of blooming nectar plants also were highly correlated between survey protocol types (r = 0.69– 0.80, depending on the comparison; **Figure 7**). Differences between protocols likely reflect the patchiness of common milkweed, which often grows in clonal patches, as well as many nectar plants, rather than undesirable biases in either method. Due to this patchy distribution of milkweed across the landscape, the spatial distribution of quadrats sampled with the IMMP protocol (spread over 500 m) was more reliable for detection of milkweed than a single Rapid Assessment (50 m), although milkweed detection was similar when combining the several Rapid Assessments per site (two or three 50 m widths spread across 500 m). Also, we had predicted that the IMMP likely provided more accurate estimates of milkweed density by focusing observer attention into small areas, but the estimates obtained by the Rapid Assessment were similar. While the two assessments are not perfectly correlated, they result in a similar categorical quality ranking of habitat sites that would be relevant for management decisions. For example, managers could differentiate high-quality sites that would benefit from preservation, moderate sites that could benefit from enhancement, and low-quality sites that would be cost prohibitive to improve or might be good sites for full re-seeding.

The high milkweed density documented in this study in Minnesota (2,052 plants/ha by IMMP (834 plants/ac); 1,527 plants/ha (620 plants/ac) by Rapid Assessment) confirm that roadside rights-of-way can provide significant amounts of breeding habitat for monarchs (Kasten et al., 2016). And, adult monarch numbers are associated with percent cover of milkweed (Kinkead et al., 2019) and milkweed abundance has been associated with adult monarch abundance (e.g., Zalucki and Lammers, 2010; Pleasants and Oberhauser, 2013). Converted into linear miles of interest to road managers, for the average rightof-way width we surveyed (9.43 m), this is 2,316–2,641 milkweed plants/mile (using the range of IMMP and RA estimates). The 2017 milkweed density estimate could have been inflated due to the fact that we selected sites from a set that contained milkweed in a prior study, but the 2018 average milkweed density was higher and these sites were selected through a random process. These milkweed densities are higher than other studies in the upper Midwest (508 plants/ha, Kasten et al., 2016; 141 plants/ha, as converted from Hartzler and Buhler (2000) in Thogmartin et al. (2017b) and used to estimate levels in current roadside rights-of-way). However, our sample size was small and we did not sample all types of roads, such as those in developed areas that do not typically provide habitat or those that appeared to be <4 m wide when previewed online. Overall estimates of habitat availability must take into account different rights-ofway types and potential variation by region; data collected from more locations are needed in ongoing assessments of monarch habitat availability.

The levels of monarch use for reproduction suggest these roadside rights-of-way can serve a significant function for breeding habitat. The per plant density of monarch eggs and larvae ranged from 0.01 monarchs/plant in 2017 to 0.099 in 2018 (IMMP protocol), bracketing the 0.059 reported for roadsides by Kasten et al. (2016) and 0.043 eggs/plant reported by Nail et al. (2015); from Monarch Larva Monitoring Project data from non-roadside areas, primarily gardens.

<sup>7</sup>https://mlmp.org/

We detected a strong difference among years in monarch egg and larval abundance, which is not surprising given high inter-annual variation in monarch numbers (Thogmartin et al., 2017a). In 2017, when monarch numbers were low, the two survey methods did not correlate well. In fact, monarchs were not detected with one or the other of the techniques in eight of ten sites. However, in 2018 when monarchs were detected at a higher rate, the two methods were correlated (tau = 0.661). Our findings suggest that single visits to describe monarch use are unreliable in years with lower monarch numbers. This coincides with recommendations from MLMP and IMMP to conduct monarch use surveys weekly. For roadside managers or others constrained to single visits, monarch use data may be regarded as descriptive rather than quantitative (i.e., eggs or larvae indicate breeding but their absence is not meaningful). Managers must be aware that monarch use data from 1 year may not be representative of other years. Furthermore, as monarch abundance also fluctuates within seasons, surveys should be conducted during similar dates within the season to compare monarch use among sites. If monarch use is a primary focus for a roadside manager, collecting data from repeat surveys within a year and across multiple years would greatly improve information about monarch use.

Similarly, we found inter-annual variation in nectar plant metrics, but only using the IMMP method, which may have been due to several factors. First, different sites were visited each year by different field crews. Secondly, the IMMP method recorded only plants blooming at the time of survey, which varies throughout the season, another factor that was not controlled for. The Rapid Assessment technique will be more resilient to seasonal effects because it includes all potentially blooming plants, including those that have already bloomed or will bloom after the survey. Because it is generally easier to identify plants when they are blooming, we recommend that surveys be conducted in peak blooming season within the period(s) of time when monarchs are present (usually mid- to late-summer), to facilitate identification, or at least differentiation, of plant species. Best practices will be to minimize variation by comparing habitat quality scores from visits to sites within the same year and season. For vegetation, it is likely that surveys may be done periodically, every several years, while for monarch use, extrapolation across years would be less representative.

The presence of late instar larvae indicates that monarchs are developing in these habitats. Providing more milkweed dispersed across the landscape may improve monarch larval survival in lower density patches of milkweed (Zalucki and Kitching, 1982), and having access to milkweed across the landscape should increase the number of eggs females lay (Zalucki and Lammers, 2010; Zalucki et al., 2016; Grant et al., 2018). However, monarch eggs and larvae experience high levels of mortality due to predation, weather, disease, and other factors (Nail et al., 2015). Additionally, milkweed in roadside areas may support lower densities of monarchs than milkweed found in adjoining agricultural habitats (Pitman et al., 2018), although it is not known if these patterns reflect differences in habitat quality or other factors, such as behavioral responses to linear landscape features or opportunistic use of the few milkweed plants remaining in an agricultural matrix dominated by geneticallymodified crop fields treated with glyphosate. Therefore, more information about the survival of monarch eggs and larvae in roadside habitats compared to other habitat types will be important for assessing the relative benefits of roadside habitat for producing monarchs.

The species richness of blooming nectar plants in each small roadside site ranged up to 18 species, suggesting roadside areas could serve an important function in providing foraging resources for pollinators. Flowering plant diversity is associated with greater frequency of visits by pollinators and pollinator diversity (Potts et al., 2003; Ebeling et al., 2008) and increases the likelihood of nectar availability throughout the season of monarch use. Also, the frequency of blooming plants ranged up to 79% of plots occupied, with estimates of the area covered by flowers as high as 46%. Floral display is well known to relate to pollinator use (e.g., Hegland and Totland, 2005; Gunnarsson and Federsel, 2014). In restored mine sites, nectar plant diversity and nectar abundance related to butterfly numbers and diversity (Holl, 1995), and similarly in roadsides in England, abundance of flowering plants was related to butterfly richness (Munguira and Thomas, 1992). While the Fender's blue was associated with the availability of native plant nectar sources (Thomas and Schultz, 2015), in many studies butterflies appear to be generalists, for instance using many nectar sources regardless of sugar content (Pavlik et al., 2018). In an experimental study of pollinator gardens, butterfly use increased with number of flowering plants; monarchs nectared on non-native flowers more than native (Majewska et al., 2018). In particular, monarchs may be limited by access to nectar in the fall that is critical for gaining lipids sufficient for successful overwintering (Brower et al., 2006; Inamine et al., 2016). Indeed, greater numbers of fall migrant monarchs were found in association with greater abundances of flowers on fire-restored pine-grasslands than on control sites or those more than 3 years since burned in Arkansas (Rudolph et al., 2006).

Our approaches to describing nectar availability were limited; practices such as counting and measuring flowers, and measuring nectar quantity and quality in them, would be much more informative (e.g., Denisow et al., 2014; Hicks et al., 2016; Szigeti et al., 2016, 2018). However, these do not fit within the constraints of a rapid assessment. Additional research that relates more intensive measures of nectar availability to simpler indices would also be helpful to many future studies of pollinator habitat. Work has been done on the relative nutritional value (e.g., sugars, amino acids) of different nectar sources (e.g., Gottsberger et al., 1984; Baker and Baker, 1986; Abrahamczyk et al., 2017). However, for monarchs in particular, few quantitative studies investigate relative use or nutrition of different nectar sources (Malcolm, 2018), which could vary among years, locations, and with environmental conditions. The Nectar Plant Guides produced by USDA NRCS and The Xerces Society report species used by monarchs; species reported as of "outstanding value" or mentioned by multiple sources were rated "very high"; species "cited as attractive to monarchs but with less frequency" were rated "high" (USDA NRCS, 2016). Further work on preference and nutritional value (including pyrrolizidine alkaloids used in pheromone production; Boppré, 1990) of various types of nectar sources for monarchs would help to guide conservation efforts.

While our results and a handful of previous studies highlight the promise of roadsides as monarch habitats, these areas also bring a suite of threats to monarchs and other pollinators including collisions with vehicles and chemical inputs (Skorka et al., 2013; Snell-Rood et al., 2014; Keilson et al., 2018; Pitman et al., 2018). However, larger butterflies such as monarchs may sustain a lower rate of mortality from car collisions than smaller butterflies (Skorka et al., 2013). Furthermore, mortality from cars is lower in roadside habitats with certain characteristics, such as greater plant species richness (Ries et al., 2001; Skorka et al., 2013). The width of the right-of-way habitat as well as the composition of adjacent lands also may affect collision mortality rates, such that wider habitats with greater access to adjoining habitats may reduce collision mortality (Munguira and Thomas, 1992; Skorka et al., 2013, but see Saarinen et al., 2005). In addition, collision risk appears greater in areas where monarchs funnel together during migration, such as in southern Texas and northern Mexico (Kantola et al., 2019; Tracy et al., 2019). Chemicals, including sodium and heavy metal run-off from roadways, are incorporated into roadside vegetation (Snell-Rood et al., 2014; Munoz et al., 2015). These chemicals could affect the development of monarch eggs and larvae or even affect adults through contamination of nectar resources. Further study of roadside areas to profile monarch egg and larval survival in relation to chemical or traffic-induced mortality would allow better understanding of how roadside habitats perform as monarch breeding areas.

Roadside management authorities are becoming aware of the impact of management policies on roadside habitat, and exemplary programs with deferred mowing, re-establishment of native plants, control of noxious weeds, and integrated vegetation management occur around the country. Mowing, in particular, is a complex topic. This ubiquitous management action is required to provide safety in roadside rights-of-ways and can be instrumental in the control of invasive plants. However, mowing can harm animals inhabiting mowed areas (Dale et al., 1997; Johst et al., 2006; Cizek et al., 2012) and can reduce floral cover (Halbritter et al., 2015) or cover by native plants (Entsminger et al., 2017). Indeed, reduced mowing during the monarch breeding season is recommended to reduce direct mortality for monarch eggs and caterpillars and to preserve more plant blooms as nectar sources (Monarch Joint Venture, 2019). On the other hand, a single mowing of milkweed in early-mid growing season can increase oviposition by monarchs (Haan and Landis, 2019; Knight et al., 2019). More information about the effects of deferred mowing practices on nectar availability and the prevalence of invasive species is needed. Several of the managers advising this project expressed their interest in recording the effects of new mowing practices on the habitat in their roadside rights-of-ways. The Roadside Monarch Habitat Evaluator allows managers to track milkweed, nectar plants, and monarchs under different management, such as comparing mowed and unmowed portions of their rights-of-way. Data about monarch habitat quality will help managers to make management decisions to benefit monarchs and pollinators generally. Challenges remain in balancing the multiple management needs for rights-of-way and communicating the benefits of native, uncut vegetation to shift public preferences for well-manicured turf grass along roadways. Future research on optimal mowing regimes and effects on milkweed, nectar availability, and use by monarchs continue to be particularly pertinent.

Because of the importance of the breeding season to the monarch annual cycle (Oberhauser et al., 2017), the strong connection between habitat loss in the core of the eastern population's breeding range and low monarch numbers (Thogmartin et al., 2017a), and use of roadsides for monarch breeding (Kasten et al., 2016), roadside restoration and management of existing habitat is promising for monarch conservation. Furthermore, roadside areas managed for monarch habitat provide native plants that could benefit other wildlife, such as small mammals, birds, pollinators and other beneficial insects. Ongoing communication and research around the potential conservation benefits of well-managed roadside rightsof-way will be highly beneficial.

# AUTHOR CONTRIBUTIONS

AC directed the study, conducted data analysis, and led the writing of the manuscript. EA, KB, JH, EL, CN, KO, KT, and ES-R advised on the design of the project. KO was involved with the inception of the project. ES-R later became the primary investigator overseeing the project at the University of Minnesota. KB collected data using the Rapid Assessment for another dataset. KT designed and implemented the data collection system in Survey123.

# FUNDING

This work was funded by the National Cooperative Highway Research Program of the Transportation Research Board, grant 20-119 Evaluating the Suitability of Roadway Corridors for Use by Monarch Butterflies. The Snell-Rood lab is additionally supported by funding through the Environment and Natural Resources Trust Fund as recommended by the Legislative-Citizen Commission on Minnesota Resources.

# ACKNOWLEDGMENTS

The project benefited greatly from contributions by the advisory panel, research advisors, and roadside managers at departments of transportation who assisted us throughout the project. For interviews, we thank Rob Roman (Linn County, Iowa), Kayti Ewing (Arkansas DOT), Dan MacSwain (Washington County, MN), and Stephanie Dobbs (IL DOT). For field visits, we thank Chris Smith and Tina Markeson at MNDOT, Stephanie Dobbs and Susan Hargrove at ILDOT, and Alyssa Barette and Christa Schaefer at WISCDOT. We appreciate Nicholas Haas, Alexandra Grace Haynes, and Patrick Perish for their dedicated work in collecting high quality field data, Daniel Cariveau for statistical assistance, and Holly Holt, Wendy Caldwell, Kyle Kasten, and Laura Lukens for their input into the design and execution of the project. We thank Iris Caldwell and Klaudia Kuklinski at the Energy Resources Center (University of Illinois – Chicago) for their roles in facilitating the review of other pollinator habitat rating systems.

### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00386/full#supplementary-material


and pollen resources of urban flower meadows. PLoS ONE 11:e0158117. doi: 10.1371/journal.pone.0158117


bumblebees and butterflies within intensively farmed landscapes. J. Insect Conserv. 15, 853–864. doi: 10.1007/s10841-011-9383-x


the Midwestern US: all hands on deck. Environ. Res. Lett. 12:74005. doi: 10.1088/1748-9326/aa7637


**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 Cariveau, Anderson, Baum, Hopwood, Lonsdorf, Nootenboom, Tuerk, Oberhauser and Snell-Rood. 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 Method to Project Future Impacts From Threats and Conservation on the Probability of Extinction for North American Migratory Monarch (Danaus plexippus) Populations

#### Kristen J. Voorhies <sup>1</sup> \*, Jennifer Szymanski <sup>2</sup> \*, Kelly R. Nail <sup>3</sup> and Mason Fidino<sup>4</sup>

<sup>1</sup> Chicago Illinois Field Office, U.S. Fish and Wildlife Service, Chicago, IL, United States, <sup>2</sup> Midwest Regional Office, U.S. Fish and Wildlife Service, La Crosse, WI, United States, <sup>3</sup> Minnesota Wisconsin Field Office, U.S. Fish and Wildlife Service, Bloomington, MN, United States, <sup>4</sup> Urban Wildlife Institute, Lincoln Park Zoo, Chicago, IL, United States

#### Edited by:

Wayne E. Thogmartin, United States Geological Survey (USGS), United States

#### Reviewed by:

Tyler Flockhart, University of Maryland Center for Environmental Science (UMCES), United States Ruscena Wiederholt, Everglades Foundation, United States

\*Correspondence:

Kristen J. Voorhies kristen\_voorhies@fws.gov Jennifer Szymanski jennifer\_szymanski@fws.gov

#### Specialty section:

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

Received: 02 May 2019 Accepted: 24 September 2019 Published: 16 October 2019

#### Citation:

Voorhies KJ, Szymanski J, Nail KR and Fidino M (2019) A Method to Project Future Impacts From Threats and Conservation on the Probability of Extinction for North American Migratory Monarch (Danaus plexippus) Populations. Front. Ecol. Evol. 7:384. doi: 10.3389/fevo.2019.00384 The United States Fish and Wildlife Service is currently evaluating the monarch butterfly (Danaus plexippus) for listing under the Endangered Species Act and using the Species Status Assessment (SSA) framework to estimate and forecast drivers that impact the species' risk of extinction. To evaluate eastern and western monarch populations the monarch SSA built on a foundation of published population models and other literature to identify current growth rates and information on threats and conservation efforts. Here we present the resulting methodology, which aimed to explore the magnitude of monarch population responses to the aggregation of multiple drivers under various scenarios. Our methodology differs from previous research by developing a series of functional cause and effect relationships that link monarch population-specific responses to threats or conservation actions. We incorporated these population-specific responses into stochastic geometric growth models for both eastern and western populations to estimate the probability of quasi-extinction in 50 years. Our models were parameterized using previously estimated population-specific trend data (growth rates and environmental variability) and expert elicited estimates of population responses to multiple drivers (i.e., amount of available breeding and overwintering habitat, insecticide use, migration resource availability, and climate change). We explored plausible future scenarios with realistic place-holder data to evaluate how changes in these drivers influenced monarch quasi-extinction risk for each population. In addition, we captured uncertainty in quasi-extinction risk by calculating cumulative quasi-extinction risk over a full range of quasi-extinction threshold values which were sampled from a uniform distribution bounded by expert-elicited estimates. In both populations, our baseline for comparison was the "current" condition defined by population-specific growth rate and environmental stochasticity from previous research. The result of the methodology presented here is a novel and comprehensive tool that incorporates the impact of future stressors into projections of population numbers over time. The approach provides a tractable and updatable tool that includes multiple types of information and the associated uncertainty of drivers, population impacts, and risk of extinction. For monarchs, this tool will be critical for incorporating the best scientific and commercial information available in the upcoming listing decision.

Keywords: monarch butterfly, Danaus plexippus, population viability analysis, quasi-extinction risk, threats, expert-elicitation

#### INTRODUCTION

Migratory monarch (Danaus plexippus) populations in North America are in decline and several population viability analyses (hereafter PVAs) predict the likelihood of monarch extinction or quasi-extinction in the near future (Flockhart et al., 2015; Semmens et al., 2016; Oberhauser et al., 2017; Schultz et al., 2017). Given the decreasing population trend of North American monarchs and subsequent extinction concern, monarchs are being evaluated for listing under the Endangered Species Act by the U.S. Fish and Wildlife Service (The Service). As part of this effort The Service conducts a species status assessment (SSA)–a scientifically rigorous framework designed to evaluate a species' status using the best available science to aid decision makers (U.S. Fish Wildlife Service, 2016). Within the SSA, analyses define a species' viability (the ability to sustain populations into the future given existing and future threats and conservation efforts) while also tracking key uncertainties and assumptions. For monarchs, most information available for evaluating population-level persistence exists for the North American eastern and western migratory populations through a suite of PVAs. Flockhart et al. (2015) focused solely on the eastern migratory population and used matrix modeling to address scenarios for changes in climate change and habitat and their effects on demographic rates into the future. Also using a matrix model, Oberhauser et al. (2017) focused on vital rate differences within geographic sub-regions of the breeding range of the eastern migratory population and future conservation scenarios that would increase population growth above the replacement rate. In a third matrix model for the eastern population, Hunt and Tongen (2017) explored the range in possible growth rates for monarchs under varying hypothetical seasonal and habitat effects (in both breeding areas and overwintering grounds). Semmens et al. (2016) and Schultz et al. (2017) explore quasi-extinction risk in the eastern and western populations, respectively, using threats implicit in the estimated population growth rate. These PVAs partially fill SSA requirements and are the foundation of the framework presented in this manuscript to assess both populations' resiliency (ability to sustain plausible expected changes in their environment and threats into the future).

The SSA must evaluate and incorporate information on current and future threats to accurately estimate future monarch population resiliency. Existing information on the threats to monarchs is mostly based in the eastern monarch population and reconstructs the patterns that lead to current declines (Flockhart et al., 2015; Oberhauser et al., 2017; Thogmartin et al., 2017b). These primary drivers of monarch population decline include the loss of breeding habitat (land conversion and adoption of glyphosate tolerant genetically-modified crops; Thogmartin et al., 2017b), the loss and degradation of overwintering habitat (Sáenz-Romero et al., 2012; Vidal et al., 2014; Honey-Rosés et al., 2018), climate change (Brower et al., 2012; Lemoine et al., 2015), insecticides (Belsky and Joshi, 2018), and threats faced during the annual migration (Inamine et al., 2016). The western population faces additional threats such as loss of overwintering or breeding habitat from climate-related fire and drought (Griffiths and Villablanca, 2015; Pelton et al., 2016). Furthermore, both populations face the threats of road kill mortality (Kantola et al., 2019; Mora Alvarez et al., 2019), diseases, parasitism, and predation (Altizer and de Roode, 2015; Oberhauser et al., 2015). Each of these threats contributes to the current and future resiliency of eastern and western monarch populations through population-specific responses that the SSA endeavors to make explicit.

In addition to threats, population projections of North American migratory monarchs should incorporate future conservation efforts, which may slow or potentially reverse declining population trends. Increasing the number of milkweed stems, for example, may overcome current monarch population declines (Nail et al., 2015; Thogmartin et al., 2017a). Furthermore, improving breeding habitat within specific subregions of migratory monarchs may increase population growth rates above the level of replacement (Oberhauser et al., 2017). In this manuscript, we seek to further define the degree to which conservation actions impact monarch populations by combining conservation efforts with a full suite of current and future threats.

Finally, the SSA endeavors to track key uncertainties when estimating extinction risk. Quasi-extension thresholds the level at which a population is no longer viable—are one source of uncertainty for North American migratory monarchs. Quasi-extinction threshold estimates for the western migratory monarch population vary from as low as 20,000 butterflies (Schultz et al., 2017) to as high as 50,000 butterflies (Wells et al., 1990). Quasi-extinction thresholds for the eastern migratory monarch population range from 1,000 butterflies (Flockhart et al., 2015) to 0.25 hectares of occupied overwintering habitat (without reporting a density estimate for the number of butterflies per hectare; Semmens et al., 2016; Oberhauser et al., 2017). Because of this uncertainty, future monarch population projections should compare against a range of quasiextinction thresholds.

Here, we describe a geometric growth model for eastern and western migratory monarch populations that incorporates population responses to future threats and conservation actions and addresses the uncertainty around quasi-extinction thresholds. We used formal expert elicitation to obtain a list of population-specific threats and quantify how changes in threats or conservation actions influenced monarch populations. We used additional expert elicitation to derive a range of quasiextinction thresholds (highest, lowest, and most likely) that define a distribution of quasi-extinction thresholds. Relying on expert judgment is a common and necessary step to parameterize models (e.g., Canessa et al., 2018; Gerber et al., 2018). We believe this model fulfills the goals of the monarch SSA and provides unique insights into plausible future scenarios of monarch population trends over the next 50 years. This model also allows for the rapid and transparent integration of new information on population threats, growth, or quasi-extinction thresholds for future SSA analyses. We hope the inclusion of this model in peer-reviewed literature will invite feedback and additional testing that improves our SSA analyses and contributes to The Service's commitment to transparent and scientifically rigorous species evaluations.

#### MATERIALS AND METHODS

#### Biology

The biology and migratory behavior of eastern and western monarch populations are comparable enough to allow for a similar modeling approach to project population growth. Adult monarchs in the eastern North American population migrate annually and in the spring move northward from Mexico to breed in the United States and Canada over several successive generations. At the end of the breeding season, the final generation of adults migrate back to Mexico to overwinter before starting the cycle again the next year (Malcolm et al., 1993; Solensky, 2004). The western North American monarch population disperses annually from overwintering sites along the California and northern Mexican coast to breeding grounds that expand as far east as the Rocky Mountains and northward to Canada. Similar to the eastern population, western monarchs have several successive generations before returning to overwinter in California (Solensky, 2004; Stevens and Frey, 2010).

#### Model Development

The base population growth model for both the eastern and western monarch populations is based on the log-scale geometric growth model developed by Semmens et al. (2016) for t in 1, . . . , T discrete time steps:

$$N\_{t+1} \sim \text{Normal}\left(\mu\_t, \varepsilon\right) \tag{1}$$

In this model let Nt+<sup>1</sup> be the log monarch population size in their wintering grounds at time t + 1, which is assumed to be a lognormal random variable such that the mean, µ<sup>t</sup> , is composed of the log monarch population size in the current time-step, N<sup>t</sup> , and their log population growth rate (λ):

$$\log(\mu\_t) = N\_t + \lambda. \tag{2}$$

The variance term in Equation (1), ε, is process noise (i.e., environmental stochasticity) which is assumed to be a gamma random variable with shape k and scale θ:

$$
\varepsilon \sim \Gamma \omega \mu \text{m}(\mathbf{k}, \theta) \tag{3}
$$


Parameters include starting population size (Nt=1) and population growth rate (λ) as well as the shape (k) and scale (θ) parameters used to generate random environmental noise (ε) via a gamma distribution. Western populations are modeled by number of individual monarchs in their wintering grounds while eastern populations are modeled as the number of hectares of space the population occupies in their wintering ground.

Parameter values slightly differ from the associated citations as the datasets were updated with data through 2018–2019 (for the eastern population see Rendón-Salinas et al., 2019 and for the western population see Xerces Society Western Monarch Thanksgiving Count, 2019).

\*2.56 hectare multiplied by a density estimate of 21.1 million monarchs per hectare (Thogmartin et al., 2017c).

The gamma distribution is a continuous probability distribution that generates positive real numbers such as standard deviations for a random normal variable. The gamma distribution has an expected mean of k ∗ θ and variance k ∗ θ 2 .

To incorporate future threats and conservation actions into monarch population projections we modified Equation (2) by adding an additional term, δ, which represents a net change in population size (N) due to the both positive and negative drivers (i.e., δ is the summed effect of α<sup>i</sup> in 1, . . . , I influences on the monarch population):

$$\log(\mu\_t) = N\_t + \lambda + \delta \tag{4}$$

$$\delta = \sum\_{i=1}^{I} \alpha\_i \tag{5}$$

Values for N<sup>t</sup> , λ, k, and θ were derived from previous monarch research (**Table 1**), while the calculation of δ was derived through expert elicitation (**Figure 1**, described in more detail below).

#### Incorporating Threats and Conservation Actions Into the Future, δ

To incorporate future threats and conservation actions, δ, we first updated monarch population data (λ and ε, **Table 1**) from 2015 to 2019 for both the eastern and western populations to represent the "current" state of growth (λ) of the populations (Equation 2). Given this "current" state of growth, we characterized monarch population responses to expected future threats or conservation actions by two factors: (A) the magnitude of change in a threat or conservation action that happens over time and (B) the specific population responses, α (Equation 5), to that change in a threat or conservation action (**Figures 1A,B**).

Factor A was determined by developing a range of plausible future scenarios (see Future Scenarios below) with varying magnitudes of change in a threat or conservation

quasi-extinction values (E) to estimate the cumulative probability of extinction in 50 years.

action. These future scenarios represent the future state conditions and were derived by the SSA Team and based on published data, expert knowledge, and professional judgment (see **Supplemental Material 1**, **Tables S2, S8**). The data sources for each driver represent different time horizons that were converted to a per-year effect. For example, if η is the expected proportion of monarchs lost to a certain threat over 80 years the annual proportion that remained was calculated as (1 – η) 1/80 . To balance variations in the different time horizons associated with different drivers we modeled a 50-year time window into the future.

Factor B was determined by constructing "populationresponse curves" derived through expert elicitation in a series of separate expert elicitation workshops for eastern and western monarch populations (**Figure 1B**). We followed widely-accepted best practices to plan, prepare, elicit, and synthesize expert judgments (Hemming et al., 2017; see **Supplemental Material 2** for information specific to our elicitation meetings). Experts identified the influences they believed drive monarch population dynamics. Using the top drivers (**Table 2**), we elicited the expected population response of monarchs (in terms of proportional change in population size) due to the proportional changes of a driver (e.g., a 2-fold increase or decrease in milkweed availability). This was necessary because relationships between a driver and monarch response may not be linear. For example, a 1.25 proportional increase of a threat does not necessarily represent a reduction of 1.25 in monarch populations as (B) may not have a 1:1 relationship with (A).

We elicited population responses to a full range of potential changes in the drivers. For example, experts described the expected proportional change in population size given a 1.10, 1.25, 1.50, 1.75, or 2.00-fold increase or decrease in nectar resources. Each expert was asked to provide a highest, lowest, and most likely response for each proportional change in the drivers, thereby producing three response curves per driver (these curves are denoted as the "most likely" track, "reasonable best" track, and "reasonable worst" track).

TABLE 2 | Primary drivers of population responses (threats and conservation actions).


The relationship between factor A and B was often nonlinear. To account for varying degrees of non-linearity, a smoothed loess curve (R Core Team, 2013) was applied to the median expert scores for each proportional change in threat or conservation action to generate a populationresponse-curve (See **Supplemental Material 3** for the fitted loess curves to expert elicited population responses). This method allowed population response curves to be as linear or non-linear as necessary. For some drivers (insecticides and climate) population response curves were developed differently by inferring population response from historical information (**Supplemental Material 1**). Collectively, the monarch population response captures biological variability (the expert predicted possible outcomes in numbers of individuals) as well as variability in future state conditions (the scenarios developed by the monarch SSA Core team).

# Specific Model Modifications for the Eastern Population

In our eastern model our expert elicitation yielded an additional nuance—that monarchs respond differently to milkweed and breeding range nectar for r in 1, . . . , 3 sub-regions, which differ in their relative contribution of individuals to the overall population (sub-regions were based on those used in Oberhauser et al., 2017). We incorporated these regionspecific population growth responses in an approach similar to that of the region-specific matrix model projections in Oberhauser et al. (2017). To incorporate sub-regional effects, we introduce a proportional term to the model, ρ<sup>r</sup> , which splits the population into the r sub-regions according to their expert elicited importance such that P ρ = 1. While this subregional splitting does not explicitly account for the fact that the eastern monarch population breeds multiple times in each region, it accounts for the cumulative contribution of each region to the monarch population across the entire breeding season. Therefore, for the three sub-regions the population that arrives in Mexico is:

$$\log|\mu\_t| \ = \lambda + \sum\_{r=1}^{3} (N\_t \ast \rho\_r) + \delta\_r \tag{6}$$

$$\delta\_r = \sum\_{i=1}^{I} \alpha\_{i,r} \tag{7}$$

Ultimately, the drivers used in the eastern population model were similar to those of the west, the drivers or α values were related to habitat availability (milkweed and nectar), exposure to insecticides, climate variability, and overwintering conditions in Mexico where monarchs face threats due to deforestation and climate change. However, unlike the western model, the climate impacts in the eastern population only acted by impacting habitat (milkweed and nectar). Additionally, the eastern model included an additional driver (and response curve) for nectar along the migration route which represented changes in available nectar resources during the late summer and fall for migrating adult monarchs. Experts felt nectar availability for fall migrants was a limiting resource for the eastern population because migratory route and breeding range do not completely overlap (as it does in the Western population) and monarchs funnel through a smaller geographic area where spatially available nectar resources need to be distinguished.

#### Quasi-Extinction Threshold and Carrying Capacity

Our methodology for calculating quasi-extinction deviates from other published monarch models in that the quasi-extinction threshold varies along a range of expert-elicited values. For the eastern population, we elicited estimates of the lowest, highest, and most likely threshold values for quasi-extinction. We defined the minimum value of our quasi-extinction threshold as the median "lowest" estimate across all experts (1 million butterflies; 0.05 ha given 21.1 million density). The maximum value of our quasi-extinction threshold was similarly created using the median "highest" quasi-extinction values (12.8 million butterflies; 0.61 ha given 21.1 million density). For the western population, we used quasi-extinction thresholds reported in the literature with the lowest quasi-extinction threshold of 20,000 individuals (Schultz et al., 2017) and highest value of 50,000 (Wells et al., 1990). Using these ranges in quasiextinction, we calculated the cumulative probability of extinction in three steps. First, we ran the population model for 100,000 simulations. Next, to capture the range of uncertainty around the expert-elicited quasi-extinction thresholds, we approximated a uniform distribution of quasi-extinction values by generating an evenly spaced sequence of 500 numbers between the minimum and maximum values for each population. Following this, we compared the 100,000 simulations per population to each of the 500 quasi-extinction threshold values. Simulated populations went extinct if they fell below the selected quasi-extinction threshold (**Figure 1**). Once a population hit the quasi-extinction threshold it could not recover throughout our simulations (i.e., remained at zero for the rest of the simulation). The resulting 50 million values (100,000 sims <sup>∗</sup> 500 qE values) were then used to calculate the cumulative probability of extinction per year (proportion that were quasi-extinct).

Our initial model lacked an upper bound for population size (i.e., carrying capacity), and consequently, our initial simulations occasionally resulted in unrealistically high population sizes (e.g., >800 million or 38 ha). To address this, we capped yearly population sizes to approximately twice the largest observed population size for eastern (36 hectares) and western (2.4 million individuals) populations. When a population trajectory crossed its carrying capacity the estimate for that year was replaced with the carrying capacity value. The growth rate for the following year was still drawn from the distribution of growth rates defined by lambda (λ) and process noise (ε).

#### Future Projections

Future projections of monarch populations included a baseline scenario (the "current" state of growth based only on updated λ and, Equation 2) and a number of future scenarios. A complete future scenario was a composite of various states of each threat or conservation action. For example, the most optimistic states for each threat or conservation action were combined to create a composite "Best-case" scenario. For the eastern monarch population, we defined five scenarios: Best-case, Worst-case, and three Intermediate scenarios. We defined four scenarios for the western monarch population: Best-case, Worst-case, and two Intermediate scenarios. For a complete list of all scenarios see **Supplemental Material 1**. The baseline comparison for all future scenarios is the population projections under the "current" state of growth. The "current" state of growth is the eastern and western PVAs projected into the future with only the impacts of λ and environmental stochasticity (ε; **Table 1**). This resulted in a total of eleven scenarios across the eastern and western monarch populations (**Figure 1**). For each scenario, we ran a set of simulations (100, 000 simulations) for each of the three separate response curves generated (i.e., most likely, reasonable best, and reasonable worst). Both the uncertainty in monarch response (through the expert-elicited "highest" to "lowest" response) and in the future state of the drivers (through SSA team's range of "best-case" to "worst-case" future scenarios) was captured in the analysis.

FIGURE 2 | The impacts from future drivers in the western monarch population as predicted by our model. The impact is represented by a magnitude change above or below baseline lambda estimates and grouped by scenario along they x-axis. The colors represent the specific impacts and shapes represent the three expert elicited tracks created for each scenario (see legend).

TABLE 3 | Western population scenario specific λ values with 95% confidence intervals after incorporating scenario specific drivers.


The mean expected difference between the "track" lambda from simulations and the "current" or baseline lambda taken from prior research is represented by 1λ.

# RESULTS

#### Population Growth Under Future Scenarios Western Population

#### **Current**

Under the "current" condition, the western population had a λ = 0.839 (0.47–1.51). This estimate served as the "baseline," or zero-change value, for comparisons of future scenario results (**Figure 2**).

#### **Best-case**

The Best-case scenario for the western monarch population included the best plausible estimates for reductions in threats and increases in conservation efforts to combat the current declines in monarch numbers. These scenario inputs resulted in changes to λ that, once aggregated, ranged from a −0.003 decrease from the baseline λ estimate to a −0.193 decrease in λ after 50 years (**Table 3**). The range represents the multiple "tracks" of possible population response curves according to experts: most likely, reasonable best, and reasonable worst. The most likely track yielded a λ of 0.8283 (0.46–1.49; **Table 3**) or a decrease of 0.0107 from baseline λ. The individual drivers of milkweed and nectar caused positive proportional changes in population (α values) under the reasonable best tracks (proportional increases of 0.109 and 0.044, respectively, **Figure 2**) and even a small positive change under the most likely track for milkweed (proportional increase of 0.003, **Figure 2**). Alternatively, the reasonable worst tracks for milkweed and nectar estimated negative α values of −0.108 and −0.079, respectively (**Figure 2**). The largest α values were negative and were driven by overwintering habitat loss which predicted a most likely track value of −0.213 and a reasonable worst track value of −0.336 (**Figure 2**).

#### **Intermediate A**

Intermediate A for the western population represented a moderate reduction in the impact of threats and achieved only partial implementation of conservation efforts over the next 50 years. Intermediate A is less optimistic than the Best-case scenario but more optimistic than Intermediate B. Specifically, Intermediate A had lower rates of habitat loss in breeding and overwintering areas than Intermediate B as well as smaller population loss due to insecticides (**Figure 2**). These scenario inputs resulted in decreases ranging from −0.005 to −0.029 from the baseline λ estimate over 50 years (**Table 3**). The most likely track yielded a λ of 0.8231 (0.45–1.48; **Table 3**) or a −0.016 decrease from baseline λ. The reasonable best track for milkweed and nectar predicted α values of 0.097 and 0.040, respectively (**Figure 2**). However, all other tracks and drivers yielded zero changes in α or negative α values (**Figure 2**). Overwintering habitat loss drove the largest and most negative α values, predicting changes of −0.386 under the most likely track and −0.626 under the reasonable worst track (**Figure 2**).

#### **Intermediate B**

Intermediate B for the western population represented a moderate reduction in the impact of threats and partial implementation of conservation over the next 50 years. Intermediate B is more pessimistic than Intermediate A in estimates of habitat loss and in population loss due to insecticides (**Figure 2**). These scenario inputs for Intermediate B resulted in decreases from the baseline estimate of λ ranging from −0.0073 to −0.0334 over 50 years (**Table 3**). The most likely track yielded a λ of 0.8211 (0.45–1.48; **Table 3**) or a −0.0179 decrease from baseline λ. The reasonable best track for milkweed and nectar predicted positive α values of 0.068 and 0.030, respectively (**Figure 2**). However, all other tracks and drivers yielded negative α values (**Figure 2**). Overwintering habitat loss drove the largest α values in population size predicting changes of −0.379 under the most likely track and −0.617 under the reasonable worst track (**Figure 2**).

monarchs).

#### **Worst-case**

The Worst-case scenario for the western monarch population included the plausible but reasonably pessimistic expectations for threats with minimal help from conservation efforts (**Figure 2**). These reasonably pessimistic expectations included a larger impact on monarch population decline from insecticides and climate (**Figure 2**). These scenario inputs for the Worst Case resulted in decreases from the baseline estimate of λ ranging from −0.0164 to −0.0551 over 50 years (**Table 3**). The most likely track yielded a λ of 0.8089 (0.45–1.46; **Table 3**) or a −0.0301 decrease from baseline λ. The reasonable best track for milkweed and nectar predicted α values of 0.067 and 0.029, respectively (**Figure 2**). However, all other tracks and drivers yielded negative α values (**Figure 2**). Overwintering habitat loss drove the largest α values predicting changes of −0.646 under the most likely track and −0.871 under the reasonable worst track (**Figure 2**).

#### **Influence of quasi-extinction threshold**

Our quasi-extinction threshold collected from the literature ranged between 20,000 and 50,000 butterflies. Across this range, the probability of quasi-extinction for the western monarch population reached 99.99% (99.98–100.0) by 50 years in both the "current" or baseline model and all future scenarios tested (**Figure 3**). The differences between future scenario results are primarily in the times it took to reach 100% quasi-extinction. Relative to the "current" model, all future scenarios took shorter amounts of time to reach 100% probability of quasiextinction. The baseline or "current" model took 33 years to reach 99.9% (99.58–99.94) while the Worst-case scenario under the reasonable worst track reached 99.9% (99.47–99.95) quasiextinction within 20 years (**Figures 3A,E**). Our Best-case scenario under the reasonable best track reached 99.9% (99.51–99.94) quasi-extinction in 31 years while the most likely track reached 99.9% (99.49–99.94) quasi-extinction in 29 years (**Figure 3B**). Our Intermediate A and B scenarios both take 28 years to reach 99.9% (99.54–99.94) and 99.9% (99.59–99.95) probability of quasi-extinction, respectively, under their own most likely tracks (**Figures 3C,D**).

# Eastern Population

# **Current**

Under the "current" condition, growth rates varied slightly across the sub-regions North Central λ = 0.976 (0.24–4.02), North East λ = 0.975 (0.24–4.02), and South λ = 0.976 (0.24–4.02). These estimates served as the "baseline," or zero-change value, for comparisons of future scenario results (**Figures 4A–C**).

#### **Best-case**

The Best-case scenario for the eastern monarch population included the best plausible estimates for reductions in threats and maximum expected increases in conservation efforts to combat current monarch declines. These scenario inputs resulted in increases to λ in each sub-region, ranging from a 0.002 increase over baseline λ (reasonable worst, North East, **Table 4**) to the largest increase of 0.026 over baseline (reasonable best, North Central, **Table 4**). The range of changes in λ values represented the multiple "tracks" of possible population responses according to experts. The most likely tracks in all sub-regions yielded increases in λ of 0.9913 (0.24–4.09) in the North Central, 0.9827 (0.24–4.05) in the North East, and 0.9839 (0.24–4.06) in the South (**Table 4**). Under the Best case scenario, milkweed and nectar were predicted to yield positive α values of 0.93 and 0.54, respectively, in the North Central region over the next 50 years under the reasonable best tracks (**Figure 4A**) and also under the most likely tracks (0.54 for milkweed and 0.29 for nectar). However, plausible negative α values were still expected under the Best-case scenario for overwintering habitat (−0.018 to −0.036, most likely and reasonable worst tracks) and insecticides (−0.050 to −0.063, most likely and reasonable worst tracks; **Figures 4A–C**).

#### **Intermediate A**

Intermediate A for the eastern population moderately reduced the impact of threats and achieved only partial implementation of conservation efforts over the next 50 years. Intermediate A is less reasonably optimistic than the Best-case scenario but more reasonably optimistic than Intermediate B. Notably, Intermediate A assumed no net change in habitat due to climate and balanced gains in habitat due to conservation with losses due to landuse changes (**Figures 4A–C**). Other drivers of insecticides and overwintering habitat continued at the same rate as historical estimates (**Figures 4A–C**). These scenario inputs resulted in either no changes in λ or very small declines from the baseline λ. The largest decline below baseline λ was a drop of 0.007 in the reasonable worst track in the South (**Table 4**). The most likely tracks in all regions yielded declines of only −0.005 to −0.006 in λ from baseline with 0.9706 (0.24–4.00) in the North Central, 0.9700 (0.24–4.00) in the North East, and 0.9709 (0.24– 4.00) in the South (**Table 4**). Under the Intermediate A scenario, milkweed and nectar each were predicted to produce positive α values of 0.034 and 0.025, respectively, in the North Central over the next 50 years under the reasonable best track (**Figure 4A**). Across the reasonable best and worst tracks overwintering habitat changes yielded negative α values ranging from −0.070 to −0.182 and insecticides yielded negative α values of −0.01 to −0.09 (**Figures 4A–C**).

#### **Intermediate B**

Intermediate B for the eastern population represented moderate changes in threats and conservation expectations over the next 50 years. Specifically, Intermediate B assumed two changes:



North Central, NC; North East, NE; South, S. The mean expected difference between the "track" lambda from simulations and the "current" or baseline lambda taken from prior research is represented by 1λ.

(1) conservation efforts overcame continued losses of breeding habitat due to land-use changes and (2) climate change impacts could moderately decrease available habitat. Furthermore, insecticide use increased at a low rate of 5.0–10% per year. These scenario inputs resulted in zero change or very small increases or decreases from the baseline λ. The largest increase above baseline λ was a 0.002 increase in the reasonable best track of the North Central region (**Table 4**). The largest decline below baseline λ was −0.007 in the reasonable worst track of the South region (**Table 4**). The most likely tracks in all regions yielded −0.004 declines in λ from baseline with 0.9718 (0.24–4.00) in the North Central, 0.9718 (0.24–4.00) in the North East, and 0.9711 (0.24– 4.00) in the South (**Table 4**). Under the Intermediate B scenario, milkweed and nectar each contributed positive α values of 0.003– 0.062 for milkweed and 0.017–0.053 for nectar in the North Central over the next 50 years (**Figure 4A**). Plausible negative α values for the Intermediate B scenario under the reasonable best and worst tracks resulted from the drivers of overwintering habitat, a values of −0.07 to −0.182, and insecticides, α values of −0.01 to −0.093 (**Figures 4A–C**).

#### **Intermediate C**

Intermediate C for the eastern population represented a combination of plausible but reasonably optimistic habitat gains and more moderate increases in threats over the next 50 years. Specifically, Intermediate C combined the assumptions of the Best-case scenario for habitat specific drivers (milkweed, nectar, and migration nectar) and the moderate changes in threats from Intermediates A and B for insecticides and overwintering (**Figures 4A–C**). These scenario inputs caused increases in λ across all regions except for track 3 in the North East where there was no change from baseline. Increases ranged from 0.004 to 0.024 over the baseline λ estimate over 50 years (**Table 4**). The most likely tracks in all regions yielded increases of 0.002– 0.012 in λ from baseline with 0.9880 (0.24–4.07) in the North Central, 0.9786 (0.24–4.04) in the North East, and 0.9811 (0.24– 4.04) in the South (**Table 4**). For the Intermediate C scenario under the most likely or reasonable best tracks, milkweed, and nectar yielded the same positive alpha values as the Best case scenario, with α values of 0.54 and 0.93 for milkweed and 0.29 and 0.54 for nectar (**Figure 4C**). Plausible negative α values under the Intermediate C scenario mirror the α values of Intermediate B for overwintering, ranging from of −0.07 to −0.182 (reasonable best to worst), and the α values for Intermediate A for insecticides, ranging from −0.01 to −0.09 (**Figures 4A–C**).

#### **Worst-case**

The Worst-case scenario for the eastern monarch population included reasonable pessimistic expectations for threats with minimal help from conservation efforts. These reasonable pessimistic expectations included monarch population losses from all drivers (no net gains from conservation actions) and a larger impact on monarch population decline from insecticides and overwintering habitat loss (**Figures 4A–C**). These scenario inputs resulted in declines across all regions ranging from −0.061 to −0.0171 below the baseline λ estimate over 50 years (**Table 4**). The most likely tracks in all regions yielded declines of −0.0318 to −0.0472 in λ from baseline with 0.9389 (0.23–3.86) in the North Central, 0.9442 (0.23–3.89) in the North East, and 0.9288 (0.23– 3.82) in the South (**Table 4**). Under the Worst-case scenario,

shaded ribbons are 95% confidence intervals. For all sub-figures the 95% confidence intervals cover the full range of quasi-extinction thresholds (0.05–0.61 hectares).

the driver of milkweed predicted larger negative α values than the drivers of nectar, migration nectar, and insecticides when comparing across all regions. The smallest α value from changes in milkweed was −0.181 under the reasonable best track in the North Central region (**Figure 4A**). The largest α value from change in milkweed was −0.686 under the reasonable worst track in the South region (**Figure 4C**). Nectar and migration nectar related α values were smaller than those of milkweed, but still negative, and ranged from −0.01 to −0.13 over all tracks and regions (**Figures 4A–C**). The loss of overwintering habitat drove the overall largest α values under the Worst-case scenario, up to −0.89 under the reasonable worst track over 50 years (**Figures 4A–C**).

#### **Influence of quasi-extinction threshold**

Our expert elicited quasi-extinction threshold ranged between 0.05 and 0.61 hectares. The "current" or baseline probability of quasi-extinction for the eastern monarch population was 46.7% (17.0–62.2) in 50 years (**Figure 5A**). The inclusion of future scenarios would either increase or decrease this estimate over time depending on the scenario. The Best-case scenario reduced the probability of quasi-extinction estimate by 6.9– 22.2% below the baseline for all three tracks of expert predicted responses (**Figure 5B**): the mostly likely track estimates the probability of quasi-extinction in 50 years as 40.8% (13.4–56.7), the reasonable best track estimates the probability of quasiextinction as 36.3% (22.0–52.2), and reasonable worst track estimates the quasi-extinction probability as 43.5% (15.0–59.4). Our Worst-case scenario increased the baseline quasi-extinction estimate by a much greater magnitude than the reductions from Best-case, increasing the risk of quasi-extinction by 25.2–60.3% (**Figure 5F**). For the Worst-case scenario the probability of quasiextinction in 50 years was, respectively, 66.7% (34.0–79.1), 58.5% (26.1–72.6), and 74.9% (44.2–85.2) for the most likely, reasonable best, and reasonable worst tracks (**Figure 5F**).

#### DISCUSSION

The primary goal of this modeling effort was to create a rigorous, transparent, and re-usable tool that incorporates future threats and conservation actions and quantifies uncertainty around quasi-extinction thresholds for both the eastern and western migratory monarch populations. The challenges presented by the unique biology of migratory monarchs included: the need to represent multiple generations in the eastern migratory population, incorporating a mechanism for density dependence to better reflect population numbers, uncertainty around quasiextinction levels, and a continuum of monarch responses to future state conditions of threats that could incorporate a range of scenarios for future projections. We believe incorporating expert elicitation in this framework allowed us to address many of these challenges through sub-regional growth responses in the eastern population, ranges in quasi-extinction thresholds that were used to test uncertainties around quasi-extinction risk, and population response curves that allowed multiple future state conditions to be tested under varying scenarios. We also addressed density dependence through introducing a carrying capacity to limit false resilience in population sizes. This modeling framework can be easily updated by the monarch SSA team as more information on threats and conservation actions become available. This tool also allows for seamless updates of population growth rates that vary each year with newly reported monarch overwintering numbers that will result in new population estimates and estimates of future quasi-extinction risk. Additionally, as more information becomes available on quasi-extinction thresholds for each population, the thresholds tested by this model can be modified to update future predictions.

These results provide novel insights into the relative magnitude of positive and negative drivers, based on the expert-elicited response curves. In our reasonable scenarios, the outcomes of the model yielded future state conditions where the effects of negative drivers outweighed the effects of positive drivers on population size. In the western population, each driver explored was associated with its own negative population response (**Figure 2**) which, when combined, resulted in growth rates that were anywhere from −0.003 to −0.0551 lower than current estimated growth rates (**Table 3**). Despite including conservation efforts for overwintering areas and breeding grounds, the plausible scenarios still resulted in continued monarch population declines and high risks of quasi-extinction into the future (**Figure 3**). The magnitude of change in driverspecific population responses between scenarios considered for the west were extremely similar (**Figure 2**) resulting in very little variation in population quasi-extinction risk under future scenarios (**Figure 3**).

In the eastern population there were more variable population growth rates and lower risks of quasi-extinction likely due to higher N, higher lambda, and a wider range of driver-specific population responses than in the west (**Figures 4**, **5**). Drivers of milkweed, nectar, and overwintering habitat represented the largest sources of future changes in the eastern population (**Figure 4A**). The Best-case and Intermediates B and C scenarios in the eastern monarch population included large enough changes in habitat to result in larger monarch populations and lower probabilities of quasi-extinction. However, scenarios that did not include large conservation gains in habitat (Intermediate A and Worst-case) yielded probabilities of quasi-extinction equal to or higher than the baseline estimate (**Figures 5C,D**).

These results illustrate the sensitivity of the model to the inputs. Therefore, it is important to construct realistic projections of both threats and conservation actions. Furthermore, these results underscore the need for research to better understand how conservation efforts can be used to reduce or possibly counteract current monarch population declines. Because population drivers and responses are separated in our approach it is possible to consider the manipulation of drivers for the biggest benefit to the species. In the western population, further protecting overwintering grounds and nectar resources could cause a large and positive population response by the species. However, those changes would need to be greater in scope than what our analysis viewed as plausible. In the eastern population, gains in habitat drivers like milkweed and nectar may combat population losses from other drivers but only at high levels (Best-case scenario). Current large-scale, multi-state conservation efforts could be an excellent future test of this model prediction.

It is important to note that our model (similar to Flockhart et al., 2015; Semmens et al., 2016; Schultz et al., 2017) does not include parameters to address the uncertainty around a metapopulation-based link between eastern and western monarch populations. While there is evidence for exchange of individuals between the eastern and western populations (and the southern Florida non-migratory population), the specific rates and consistency of those exchange events are unknown (Brower and Pyle, 2004; Dingle et al., 2005; Knight and Brower, 2009; Morris et al., 2015). The inclusion of emigration and immigration, however, could possibly reduce our quasiextinction estimates if immigration is large enough to allow a population to recover. Thus, further research is necessary to determine the magnitude of monarch immigration and emigration so that these rates may be included in future monarch PVAs.

There is also uncertainty around the accuracy of overwintering density estimates for the eastern monarch population. Because monarch overwintering population size in Mexico is measured in hectares, the density value determines the initial population size estimate, N<sup>t</sup> , in our model. We chose one plausible density estimate—the median density of 21.1 million (Thogmartin et al., 2017c). Prior to Thogmartin et al. (2017c), published estimates of these densities range from 6.9 to 60.9 million monarchs per hectare (Calvert, 2004). In addition, experts who participated in the monarch expert elicitation reported density fluctuations within and among years. Our model did not include an underlying density function to allow for this possible fluctuation, but the effects of such uncertainty could be incorporated into future PVAs. Within the framework we developed, shifting density assumptions would alter the initial starting population size, N<sup>t</sup> , thereby possibly affecting risk of quasi-extinction. Higher density values would equate to larger N<sup>t</sup> , which would provide greater buffer against poor years and lower quasi-extinction risk over time. Lower density values would equate to opposite outcomes in N<sup>t</sup> and quasi-extinction risk over time. Density estimates and assumptions could easily be updated in the future and tested with multiple values to better capture the developing knowledge on how density estimates translate hectare estimates to numbers of individuals.

We believe our model results build on previously published PVAs for the eastern and western migratory monarch populations. In the eastern monarch population, our "current" or baseline results are most equivalent to the future quasi-extinction estimates from models by Flockhart et al. (2015), Semmens et al. (2016), and Oberhauser et al. (2017). Our analysis updated the existing λ and ε for the eastern population from Semmens et al. (2016) (with overwintering data from 2017 to 2018 and 2018 to 2019) and forecasted monarch population trends into the future. In the western monarch population, our "current" or baseline analysis would be most equivalent to Schultz et al. (2017) with similar updates to the existing λ and ε. However, our future scenario results differ from these models as we incorporated population responses to changes in threats and conservation into the future. In addition, the use of a full range of quasiextinction thresholds and a carrying capacity bring our modeling effort closer to the goals for assessing risk and uncertainty. Ultimately, these modifications build on published PVAs while also adding to the collective understanding of monarch risk into the future.

The PVA presented here is not the only factor included in The Service's process for evaluating the monarch for listing under the Endangered Species Act. There are additional influences and analyses for non-migratory monarchs and monarchs outside of North America considered within the SSA framework. Furthermore, this study does not explicitly test assumptions about population response to influences outside of those considered by our expert elicitation process, many of which require further study, nor does it take into account potential catastrophic events outside the scope of historical events (implicitly incorporated into λ). Future studies looking to incorporate threats for monarchs into the future may shed light into which drivers should or should not be included in this model or if the assumptions associated with a geometric growth model are valid. Our results show the potential of incorporating future threats and conservation actions into population projections for migratory monarchs, thereby making it easier to change which threats and conservation actions are included and to what degree they will change into the future. By doing so, we believe we have not only met the goals set by the SSA framework, but we have created a transparent and reproducible tool that will be

#### REFERENCES


repeatedly applied to exploring monarch population responses into the future.

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.

#### DATA AVAILABILITY STATEMENT

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

# AUTHOR CONTRIBUTIONS

KV, KN, and JS contributed to conception and design of the study. KV and MF designed the modeling framework and developed R code for this analysis. KV wrote the first draft of the manuscript. All authors helped with later drafts, manuscript revision, and read and approved the submitted version.

# FUNDING

MF's work was supported by the Abra Prentice Foundation.

#### ACKNOWLEDGMENTS

We would like to recognize the following experts for providing their monarch expertise and for participating in the formal expert elicitation process: Anurag Agrawal, Sonia Altizer, Linda Fink, Matt Forister, Jessica Griffiths, Pablo F. Jaramillo-López, Sarina Jepsen, Vera Krischik, Stephen Malcolm, Dan Meade, Gail Morris, Karen Oberhauser, Ian Pearse, Emma Pelton, John Pleasants, Cheryl Schultz, Chip Taylor, Francis Villablanca, and Louie Yang. Additionally, we would like to thank the following experts for their contributions to our analyses: Brice Semmens and Wayne Thogmartin. We thank all of the FWS biologists on the monarch SSA team for their help with scenario development and population response curves as well as and Nancy Golden, Dave Warburton, Sarah Warner, and Lisa Williams.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00384/full#supplementary-material

and Nectar Corridors in Western North America, ed G. P. Nabhan (Tucson, AZ: University of Arizona Press), 144–166.


emerging fungal pathogen of amphibians. J. Appl. Ecol. 55, 1987–1996. doi: 10.1111/1365-2664.13089


**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 Voorhies, Szymanski, Nail and Fidino. 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.

# Spatio-Temporal Distribution of Monarch Butterflies Along Their Migratory Route

Saul Castañeda1,2, Francisco Botello1,2 \*, Víctor Sánchez-Cordero<sup>1</sup> and Sahotra Sarkar <sup>3</sup>

<sup>1</sup> Departamento de Zoología, Instituto de Biología, Universidad Nacional Autónoma de Mexico, Mexico City, Mexico, <sup>2</sup> Departamento de Monitoreo Biológico y Planeación de Conservación, Conservación Biológica y Desarrollo Social, Mexico City, Mexico, <sup>3</sup> Departament of Integrative Biology and Philosophy, University of Texas at Austin, Austin, TX, United States

#### Edited by:

Wayne E. Thogmartin, United States Geological Survey (USGS), United States

#### Reviewed by:

Thomas Earl Dilts, University of Nevada, Reno, United States Mirko Di Febbraro, University of Molise, Italy

\*Correspondence: Francisco Botello francisco.botello@ib.unam.mx

#### Specialty section:

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

> Received: 28 February 2019 Accepted: 08 October 2019 Published: 23 October 2019

#### Citation:

Castañeda S, Botello F, Sánchez-Cordero V and Sarkar S (2019) Spatio-Temporal Distribution of Monarch Butterflies Along Their Migratory Route. Front. Ecol. Evol. 7:400. doi: 10.3389/fevo.2019.00400 Efforts to conserve the migratory phenomenon of monarch butterflies in eastern North America have increased since a 2013–2014 monitoring report documenting a historical population low at the Monarch Butterfly Biosphere Reserve in Mexico. Surprisingly, there have been few systematic attempts to develop predictive models of monarch butterfly distributions along their migratory route between Mexico, the United States and Canada. Here we produced monthly habitat suitability models for monarch butterflies along their migratory route to identify potential areas for resting, feeding, and reproduction of the population. We compiled a point occurrence database of monarch butterflies for Mexico, USA, and Canada, including georeferenced records from GBIF, the Naturalista platform in Mexico, Correo Real initiative, and the Mexican governmental monitoring network for the monarch butterfly. We produced monthly habitat suitability models (HSM), using the R language and environment for statistical computing, abiotic (WorldClim), edaphic, and topographic variables. A total of 95 HSM were produced for each month. June to September, corresponding to the reproduction months in North America showed the highest geographic extent with suitable habitats; April, corresponding to the reproduction of the first post-migration generation, showed the smallest area. September, October, and November, correspond to the movement of the monarch butterfly southward, showed typical recognized distribution of the phenomenon and the overwintering months. December to February showed the smallest geographic extent in habitat suitability. Edaphic variables ranked high in importance in HSM for 11 of 12 months, indicating the relevance of vegetation and floral resources in the monarch butterfly migration route. Identifying such regions contribute to establish concrete conservation programs accordingly, as reduction of the use of pesticides and herbicides, decrease in the speed of cars in roads, and planting species with high nectary value, among other. Our study provides a first predictive spatio-temporal approximation of the monarch butterfly migratory route annual cycle.

Keywords: citizen science, habitat suitability models, distribution, conservation areas, habitat suitability, monarch butterflies

# INTRODUCTION

The eastern North American monarch butterfly (Danaus plexippus) population undertakes the longest documented insect migration in the world (Agrawal, 2017; Sarkar, 2017). The journey from its overwintering habitat in central Mexico to the Midwest and northeast of the United States and southeastern Canada over 3–4 generations each spring, and back to Mexico in one generation each fall, is up to 4,500 km each way. The behavioral complexity of the long and, in one direction multi-generational, migration both makes the phenomenon unique and of conservation interest, but also difficult to protect because of the extent and diversity of habitats used by the monarch butterflies. There is also a western population of monarchs that was believed to winter in southern California and migrate to northern California and further north and east during the summer. While it was assumed that the two populations were geographically isolated, evidence has recently accumulated of moderate gene flow between them (Yang et al., 2016; Agrawal, 2017).

Overwintering monarch populations of the eastern population at the Monarch Butterfly Biosphere Reserve in the State of Michoacan and the State of Mexico, were estimated to be over 400 million individuals in the early 1990s but only about a hundred million since 2010 with a historical low of about 35 million in 2013–2014. Efforts to conserve the migratory phenomenon of monarch butterflies have increased since that monitoring report. Different threats have been proposed to negatively affect monarch butterflies along their migratory route including (a) the lack of availability of milkweed (Asclepias ssp.) in their breeding areas due to pesticide use along their migratory route, contributing to loss of vegetation (the milkweed limitation hypothesis; Brower et al., 2012); (b) individual mortality during the fall migration (the migration survival hypothesis; Agrawal, 2017); (c) decline in the size of the winter habitat for roosting in Mexico (the winter habitat loss hypothesis; Brower et al., 2012);(d) extreme climatic events in their overwintering area (the climate hypothesis; Brower et al., 2012), and the loss of nectar resources (Malcolm, 2018). Multiple causal mechanisms may be responsible for the decline in monarch numbers and no single hypothesis yet suggested can be excluded. The relative role of each of the proposed mechanisms will require further research.

Thus, it is useful to map the areas used by the monarch butterflies at discrete time steps; HSM models built from records at monthly time intervals allows us to examine how habitat selection changes over the course of the monarch butterfly migration (Batalden et al., 2007; Hayes et al., 2015; Coxen et al., 2017). In the case of the eastern monarch butterfly population, HSM models for the reproductive season (March–September) have been produced for the United States and Canada (Batalden et al., 2007; Lemoine, 2015). Those results reported that monarch butterflies prefer similar environmental features throughout the summer but switch to a very different set of environmental features for the winter (Batalden et al., 2007; Lemoine, 2015).

Since the report showing low values of the overwintering population in 2013–2014 (Vidal and Rendón-Salinas, 2014), several initiatives have been launched to improve our knowledge on the migratory route of the monarch butterfly in Mexico. For example, the establishment of the Monarch Butterfly National Monitoring Network in 2015 includes now detailed monitoring programs in 44 protected areas in 29 states that have produced more than 4,000 new records of monarchs butterflies along their migratory route in Mexico (CEC, 2017). New data demonstrate that there are monarch butterflies crossing southward along western of Mexico in the States of Chihuahua, Durango, Nayarit, and Sinaloa (unpublished data). This means that the western population does not overwinter only in California. The extended southern migration could hypothetically also facilitate interbreeding between the eastern and western populations, which would be in closer spatial proximity during the winter compared to the summer. Moreover, a new wintering site in the same geographical region as those previously known was reported during the 2018–2019 season. Given this context, we aimed to produce spatio-temporal HSM for both the eastern and western populations of the monarch butterfly migratory route. Our study intends to provide a first predictive spatio-temporal approximation of both the eastern and western populations of the monarch butterfly at a monthly resolution.

# MATERIALS AND METHODS

Our methodology involves building monthly HSM using a maximum entropy modeling approach. HSM use two sources of data: occurrence (or presence-pseudo-absence) points (longitude and latitude of observation of an individual) and a set of environmental layers. This methodology does not require presence-absence data for species that can only be obtained from systematic surveys. In the case of monarch butterflies, this is particularly important because most occurrence points consist of observations by citizen scientists. The output of the model consists of the relative occurrence rate of the species in every cell located within the geographical study area over which the model is constructed. This distribution of spatial probabilities can form the basis for further analysis or converted into a binary distribution map using a minimum probability threshold for predicted presence of a species or some other similar method (Phillips et al., 2006).

#### Presence Data

We obtained records of collected specimens and observations of monarch butterflies as follows: Global Biodiversity Information Facility, GBIF (12,806 records of collected specimens and observations, references below; download of records 15 March 2018), the Naturalista platform, CONABIO, Mexico (2,166 records of observations, http://www.naturalista.mx/taxa/ 43155-Danaus-plexipus, download of records 21 December 2017), Correo Real Initiative, Mexico, CRI (11,199 records of observations, unpublished data), the database of the National Monarch Butterfly Monitoring Network in Mexico, NMMN (4,050 records, unpublished data), published records of overwintering colonies in Mexico (Vidal and Rendón-Salinas, 2014; 96 records) and unpublished data CONANP (6 records). We refined this database selecting: (a) only georeferenced records between 1970 and 2018, (b) records where collectors or TABLE 1 | List of climatic and topographic variables used in our study and in previous studies to model monarch butterfly distribution; tavg, average temperature; tmax, maximum temperature; tmin, minimum temperature; prec, precipitation; srad, solar radiation; vapr, water vapor pressure; elev, elevation; slope, slope; aspect, aspect.


observers were specifically mentioned, (c) unique localities, and (d) records with complete dates (day, month, and year), totaling 1,928 records.

#### Predictor Variables

We used climatic, topographic, and edaphic variables as environmental layers to produce the HSM monarch occurrence. These variables represent direct and indirect gradients (Austin, 2002) that are presumed to be ecologically meaningful for monarch butterflies, as floral resources along the migratory route, or in overwintering grounds (**Table 1**). The selected climatic variables were minimum and maximum monthly and average values for temperature, precipitation, solar radiation, wind speed, and water vapor pressure from the WorldClim 2.0 database at the 30 arc-second (∼1 km<sup>2</sup> ) spatial resolution (Fick and Hijmans, 2017). The 19 bioclimatic variables derived from WorldClim data are extensively used in model construction (Booth et al., 2014; Porfirio et al., 2014; Vega et al., 2017). Since they represent annual trends, extreme values and seasonality calculated from temperature and precipitation of more than 1 month (e.g., BIO5 = Max Temperature of Warmest Month), they may not have spatio-temporal coincidence with all the presence records along the migratory route. Thus, these variables were not all used produce HSM; rather, only the climatic variables that matched the month of observation of the records were selected. A similar monthly approach has been used to model the ecological niche of breeding monarch butterfly populations (Batalden et al., 2007) and the distribution of migratory bat species in North America (Hayes et al., 2015).

The topographic variables included were elevation, slope, aspect, and Compound Topographic Index (Moore et al., 1991) from the HYDRO1k Elevation Derivative Database (EROS Center, 2015). These variables represent attributes that are related directly or indirectly to environmental gradients affecting species distributions (Franklin, 2010), they have been used to model plants (Franklin, 1995, 1998) as well as other taxon distributions (Elith et al., 2006; Hasui et al., 2017) and their inclusion can increase the accuracy of the models (Sormunen et al., 2011). Aspect values were transformed from continuous to categorical to reflect the slope direction as cardinal points. The edaphic TABLE 2 | Feature classes (FC), regularization multipliers (RM), AUC based on the test set (tAUC), average difference between training and testing AUC (av.diff.AUC), the difference between the sample-size-adjusted Akaike information criterion value (AICc) of the model, and the model with the lowest AICc value (1AICc), and true skill statistic (TSS) of the selected models by month for the monarch butterfly migratory route.


variables were percent of clay content, bulk density, pH and organic carbon values at 0.05, 0.3, and 2 m soil depth obtained from SoilGrids1km database (Hengl et al., 2014). These are known to affect plant growth and have been used to predict the distribution of shrub species (Hageer et al., 2017). Although they are not very frequently used, edaphic variables could improve the predictive value of distribution models of plant species based solely on climate and topographic predictors (Dubuis et al., 2013; Buri et al., 2017; Hageer et al., 2017; Figueiredo et al., 2018). In sum, we included seven climatic, four topographic and 12 edaphic variables, respectively.

#### Treatment of Occurrence Data

The occurrence points were treated by removing outliers according to the values of the predictor variables using (a) its position with respect to the interquartile range, and (b) with the reverse Jackknife procedure implemented in the R "Biogeo" package (Robertson, 2016; Robertson et al., 2016). Given that there are resident monarch butterflies and presence data do

TABLE 3 | Number of records, suitable area, number of ecoregions intersected by predicted suitable area and percentage of predicted suitable area where model extrapolation occurs for the monarch butterfly migratory route.


not provide enough information to discriminate residents from migrant individuals, occurrence points were eliminated on the basis of expert opinion that these records should be considered as residents according to their geographical position and date. For example, all records from April to August of Mexico and records from November to February of northern Mexico and southern USA, were considered to be from non-migratory monarch butterflies and were removed. **Table 3** summarizes the results of this process by recording the number of occurrence points that remained in the dataset after outliers were removed (See Acknowledgments for the list of experts).

#### Habitat Suitability Models

HSM were constructed using a maximum entropy algorithm (Phillips et al., 2006). Records were grouped by month, and each month was modeled separately. For model calibration, the values of the climatic variables for the corresponding month were selected along with the topographic and edaphic variables, and spatially masked to the polygons of the terrestrial ecoregions of the world (Olson and Dinerstein, 2002) that contained occurrence points. We chose the ecoregions as a delimitation criteria (Soberón and Peterson, 2005) to draw pseudo-absences from Di Febbraro et al. (2016), since they reflect the history of the distributions of particular biotas (Soberón, 2010), and represent suitable areas for species that have been presumably available over a relevant time period (Barve et al., 2011). Thus, there were 12 sets of 23 predictive variables, one set for each month.

Multicollinearity can confound the interpretation of variables driving the spatial distributions derived from species HSM (Elith et al., 2010; Dormann et al., 2013). It is recommended to minimize correlation among them through different methods (Merow et al., 2013). Highly correlated variables were identified and removed using the variance inflation factor (VIF) with the R package "usdm" (Naimi et al., 2014). The algorithm identifies a pair of variables with a correlation coefficient greater than a defined threshold (i.e., 0.7), removing the variable with the highest VIF, and repeats the process until no highly correlated

variables remains. Further, presence-background data records from collections or citizen science projects are typically biased since they often come from opportunistic surveys or accessible sites (Dennis and Thomas, 2000; Syfert et al., 2013; Bird et al., 2014; Fithian et al., 2015). This sampling bias could affect performance and lead to inaccurate models (Phillips et al., 2009; Fourcade et al., 2014). One approach to reduce the effects of sampling bias is thinning the occurrence records in the geographical space, and removing those located at a distance from the nearest neighbor lesser than a threshold distance (NND) (Aiello-Lammens et al., 2015). Although there is a considerable number ofrecords in our dataset, geographical thinning results in a significant reduction at a small NND (**Figure 1**). Therefore, we used a target-group background approach (TGB). TGB approach is a method proposed to deal with sampling bias by choosing background or pseudo-absence data with the same bias as occurrence data (Phillips et al., 2009). We constructed a kernel density map as a bias file for each month with records of the Nymphalidae family obtained from GBIF, masked them with the polygons of the terrestrial ecoregions of the world (Olson and Dinerstein, 2002) that contained occurrence points, and used them as sampling probability surface to draw 10,000 random pseudo-absences (Di Febbraro et al., 2016).

For each of the 12 sets of predictive variables, 95 HSM were constructed with MaxEnt Version 3. 3.3K (Phillips et al., 2006). For each model, a unique combination of 19 feature classes (FC) and five regularization multipliers (RM) were used. These two parameters have influence on model accuracy (Phillips and Dudík, 2008; Merow et al., 2013) and it is recommended to "tune" them since the MaxEnt default settings can lead to overly complex models (Radosavljevic and Anderson, 2014). The FC are transformations of the covariates (i.e., predictor variables)

four bins (different colors) by the "checkerboard2" method to train and test the

models (See Methods for details).

that allows the fitting of non-linear and complex response curves (Elith et al., 2011; Merow et al., 2013), while the RM are constant values that prevent model over-fitting (Phillips and Dudík, 2008; Merow et al., 2013). The FC used were: L = linear; Q = quadratic; P = product T = threshold, H = hinge; and 14 combinations of them (i.e., LQ, LP, LH, LT, QP, QH, QT, PH,PT, HT, LQP, LQPH, LQPT, and LQPHT). The RM values went from 1 to 5 by increments of 1. Parameter tuning and model fitting were performed with the R package "ENMEval" (Muscarella et al., 2014). The data were partitioned into training and testing bins by the "checkerboard2" method, which is a masked geographically structured approach (Radosavljevic and Anderson, 2014). This method divides the presences and pseudo-absences into four bins according to two different checkerboard-like grids based on an

aggregation factor set to 10 (**Figure 2**) (Muscarella et al., 2014). Of the 95 models, the HSM with the best performance was selected with a sequential approximation, by first minimizing the difference between the sample-size-adjusted Akaike information criterion value (AICc) (Warren and Seifert, 2011) of the ith model and the model with the lowest AICc value (1AICc). We minimized the difference between training and testing AUC, averaged across the four bins (avg.diff.AUC) (Muscarella et al., 2014). This approximation allowed to select the optimal model complexity, avoid overfitting (Wisz and Guisan, 2009; Sarkar et al., 2010; Warren and Seifert, 2011) and to use the second criterion (avg.diff.AUC) in case there were several models with the same 1AICc value (Shcheglovitova and Anderson, 2013). The selected models were projected to the geographic space to the same extent of the predictors used to train the model. The continuous suitability projections were transformed to discrete presence-absence maps using the equal training sensitivity and specificity threshold, which is adequate for presence-background models (Cao et al., 2013). Its accuracy was assessed by means of the true skill statistic (TSS) (Allouche et al., 2006). TSS values ranges from −1 to 1, where 1 indicates perfect agreement and values <=0 indicate a performance no better than random.

In order to detect areas of extrapolation due to predictor values non-analogous to those under which the models were calibrated, the extrapolation detection tool (ExDet) (Mesgaran et al., 2014), implemented in the "ecospat" package (Di Cola et al., 2017), was used. This tool measures the similarity between reference and projection domains like the Multivariate Environmental Similarity Surface feature implemented in MaxEnt, but allows to detect novel combinations between covariates, even if these are within the range of univariate variation (Mesgaran et al., 2014). Predictor values corresponding only to the presence and pseudo-absence points were used as reference, while the projection set included all the predictors values. ExDet output consist of continuous values; values below zero indicate novel conditions at the univariate level, values between zero and one indicate analogous conditions and values above one represents new covariable conditions. All analyses were carried out in the R language and environment for statistical computing (R Core Team, 2018).

#### RESULTS

1AICc scores of the selected models were <2, indicating a good fit (Muscarella et al., 2014). Model predictive performance, estimated by the AUC based on the test set (tAUC) and TSS are shown in **Table 2**. tAUC scores had a range from 0.63 to 0.90 (mean = 0.79) while TSS ranged from 0.40 to 0.69 (mean = 0.57). Model projections to the geographical space are shown in **Figure 3**. The month with the largest extent of suitable area was September (2,105,545 km<sup>2</sup> ) and it was also the month with predicted suitable habitat for the largest number of ecoregions (38). Conversely, the month with the lowest suitable area was December (28,305 km<sup>2</sup> ). The number of ecoregions intersected by the habitat suitability predictions for this month were six (**Table 3**).

The differences in the extent of suitable areas between months showed a contraction-expansion pattern when grouped by different coarse defined stages. Southward movement stage (MS), that included September, October and part of November, presented a consistent suitable area reduction, reaching the smallest suitable area at the overwintering stage (OW). The northward movement stage presented an expansion through March and, although there was a contraction for the two first moths of the reproduction stage (REP) (i.e., April and May), the suitable area reached the largest extent at this stage (**Figure 4**).

Predictor importance varied across months (**Figure 5**). According to permutation importance, minimum temperature was the most important for December and it was among the three most important variables for February and August. Precipitation was the first ranked for summer months (June and August), while water vapor pressure was one of the most important variables for January, March, April, July, October, and November. Edaphic variables were the most important for January, April, May, October, and among the three most important for the remaining months except February. The areas of extrapolation in the discrete projections were negligible; according to the performed with ExDet, none of them showed more than 2.5% of their area

represented by non-analogous conditions (**Figure 3**, **Table 3**). This is not surprising since the calibration and projection extents were the same.

# DISCUSSION

Most of the records used in this study were obtained from citizen science. Thousands of individuals across North America participate annually in different initiatives to monitor the presence of monarch butterflies and these efforts allow, year after year, the delineation of a sketch of the movements of monarch butterflies. In the case of Mexico, for more than 25 years there have been several programs such as the Correo Real initiative that facilitate the tracking of the migratory route of the monarch butterfly the east of the country. Since October 2018, with the support of the National Commission of Natural Protected Areas of Mexico, a mobile application was created and has been available to facilitate the use of a monitoring protocol by citizen scientists. This application, which will be operational throughout the year, enables citizen scientists to report a wide suite of potentially relevant variables such as the growth stages of observed individuals and the physical state of the wings of the butterflies by transmitting photographs of the individuals. This application may significantly improve the quality of the data available including an ability to discern whether the recorded individuals are migratory or not. This in turn will allow better modeling and analysis of the migratory phenomenon in the future.

Due to the origin of these records, most of them have only basic information such as locality, date, and time. However, an increasing number of new records include additional relevant information such as measures of visible threats or behavior of monarch butterflies (e.g., perches, in erratic flight, migratory flow, or feeding). Given that there are also resident populations in Mexico, for which adults, caterpillars and eggs have been observed during the same months in which the migration occurs (unpublished data), the construction of monthly HSM for the migratory process may present shortcomings when the locations coincided with resident monarch individuals. Here we tried to minimize this problem by using expert opinions for the data set used for constructing the HSM.

Overall, the models of June to September were those that predicted larger spatial extents with suitable habitats (**Figures 3**, **4**). This result concurs with predictions of the models proposed by Batalden et al. (2007). Conversely, April, corresponding to the reproduction of the first post-migration generation, had the smallest area for that season, with 90,312 km<sup>2</sup> . This result underscores the importance of monitoring and maintaining the critical habitat to reduce threats, so as to allow the population growth and the movement of the next generations, leading to large areas of suitable habitat in the subsequent months. On September, October, and November, corresponding to the movement of the monarch butterfly southward (CEC, 2017), the HSM showed typical recognized distribution of the migratory phenomenon. During the overwintering season (here considered between December to February due to the presence of multiple individuals following the route during November), our HSM resulted in a larger area than previously known to be occupied by monarch butterflies in central Mexico of <0.18 km<sup>2</sup> (CEC, 2017) (**Table 3**; **Figure 3**). This overestimation is not surprising since the HSM did not considered some of the variables affecting their distribution (i. e. biotic variables like predation, parasitism, and food availability) (Soberón and Peterson, 2005).

The importance of the minimum temperature on December was consistent with previous findings (Masters et al., 1988), although not for the same month; Oberhauser and Peterson (2003) mentioned the influence of this variable for the overwintering season. On the other hand, the relevance of the edaphic variables is evident, ranking as important in the HSM for 11 of the 12 months. This result is indicative that vegetation and floral resources play an important role in the monarch butterfly migration route. Water availability is considered critical to the butterfly's survival (Bojórquez-Tapia et al., 2003); predictor variables related with this condition, as water vapor pressure and precipitation (Jones, 1987) showed to be relevant for the four stages. These findings demonstrate the pertinence of including other variables in addition to climatic variables for building HSM (Hageer et al., 2017). However, these interpretations should be taken with caution since the removal of highly correlated predictors poses the risk of leaving out those with ecological relevance for the species (Braunisch et al., 2013).

The HSM showed concordance with the identified overwintering zones in the west, both in California (USA) (Fisher et al., 2018) and in Baja California (Mexico). However, the model for March predicted a migratory route from the overwintering sites to the north of Mexico and south of the USA. It is crucial to document the monarch butterfly migration route followed in northwestern Mexico and citizen monitoring should be promoted in these regions during the spring migration. Our HSM corresponding to Abril apparently showed a smaller area than those resulting from other studies conducted for

content at 200 cm, cti, compound topographic index; elev, elevation (m); orc\_05, organic carbon at 5 cm; org\_200, organic carbon at 200 cm; pH\_05, pH at 5 cm; pH\_200, pH at 200 cm; prec, precipitation (mm), slope; srad, solar radiation (kJ m−<sup>2</sup> day−<sup>1</sup> ); tmax, maximum temperature (◦C); tmin, minimum temperature (◦C); vapr, water vapor pressure (kPa); wind, wind speed (m s−<sup>1</sup> ).

the eastern population (Batalden et al., 2007; Lemoine, 2015). Nonetheless, the HSM for May to August coincided remarkably with these studies, with the peculiarity that our model for September, seemed to adequately describe the migratory movements (**Figure 4**).

One contribution of our analysis is the urgency to establish collaborative agreements between multiple stakeholders to reduce the possible threats of priority sites of the monarch butterfly migratory route, both for the eastern (where many such sites have long been known), and for the western sites (**Figure 3**). Advances in the monitoring of autumn and spring migration will be crucial to determine, also, the proportion of individuals that could be moving between the two currently recognized populations. Our analysis has important conservation implications because it identified regions that have a high priority for monarch butterfly migration and, therefore, should be targeted for protection from the use of pesticides or insecticides. One strategy would be to establish dynamic conservation programs over specific months and regions. For example, targeting one set of areas, in March in Mexico for the beginning of the migration; in April, for the first reproduction event in Texas and California, and other areas for October and November for areas in Texas, California, and in Mexico for the fall migration.

# DATA AVAILABILITY STATEMENT

The habitat suitability models generated for this study are available on request to the corresponding author.

# AUTHOR CONTRIBUTIONS

SC cleaned and organized the data and modeled the potential distribution. FB made the original research design. SC, FB, VS-C, and SS contributed to the writing.

# FUNDING

This work was partially funded by the Commission for Environmental Cooperation: proyect 241-00360 "Strengthening of the National Monarch Butterfly Monitoring Network along the Northwest migratory route in Mexico".

#### ACKNOWLEDGMENTS

We thank Rocío Treviño (Correo Real Initiative), Marco Castro (CONANP) for sharing unpublished data, and the supervision

#### REFERENCES


of the monarch butterfly records. Jay Diffendorfen for revision of the monarch butterfly records. G. Monroy supported in the logistics of this project. The comments and suggestions of two reviewers and the editor greatly improved the presentation of this manuscript. We thank all citizens who, day by day, make an effort to contribute to the monitoring program providing many records of observation of monarch butterflies.

science and satellite tracking data. Global Ecol. Conserv. 11, 298–311. doi: 10.1016/j.gecco.2017.08.001


**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 Castañeda, Botello, Sánchez-Cordero and Sarkar. 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.

# Recent Forest Cover Loss in the Core Zones of the Monarch Butterfly Biosphere Reserve in Mexico

José Juan Flores-Martínez <sup>1</sup> \*, Anuar Martínez-Pacheco<sup>2</sup> , Eduardo Rendón-Salinas <sup>3</sup> , Jorge Rickards <sup>2</sup> , Sahotra Sarkar 1,4 and Víctor Sánchez-Cordero<sup>1</sup> \*

<sup>1</sup> Department of Zoology, Institute of Biology, National Autonomous University of Mexico, Mexico City, Mexico, <sup>2</sup> World Wildlife Fund Inc., Mexico City, Mexico, <sup>3</sup> World Wildlife Fund Inc., Zitácuaro, Mexico, <sup>4</sup> Department of Integrative Biology and Philosophy, University of Texas at Austin, Austin, TX, United States

#### Edited by:

Wayne E. Thogmartin, United States Geological Survey (USGS), United States

#### Reviewed by:

Karen Oberhauser, University of Wisconsin-Madison, United States Matthew D. Cross, University of Colorado Denver, United States

\*Correspondence:

José Juan Flores-Martínez jj@ib.unam.mx Víctor Sánchez-Cordero victor@ib.unam.mx

#### Specialty section:

This article was submitted to Conservation, a section of the journal Frontiers in Environmental Science

Received: 28 February 2019 Accepted: 07 October 2019 Published: 01 November 2019

#### Citation:

Flores-Martínez JJ, Martínez-Pacheco A, Rendón-Salinas E, Rickards J, Sarkar S and Sánchez-Cordero V (2019) Recent Forest Cover Loss in the Core Zones of the Monarch Butterfly Biosphere Reserve in Mexico. Front. Environ. Sci. 7:167. doi: 10.3389/fenvs.2019.00167 The Monarch Butterfly Biosphere Reserve (MBBR) in central Mexico was established in 2000 to protect monarch butterfly (Danaus plexippus) overwintering colonies and contribute to the conservation of the monarch migratory phenomenon. The MBBR has faced forest cover losses due to illegal logging and climate-related factors. Here we report forest cover losses from 2012 to 2018 in the core zones of the MBBR where most monarch overwintering colonies perch. We used aerial ortho-photographs and satellite images complemented with field validation for temporal comparisons. During this period, 163.44 ha of forest cover were affected, 125.44 ha due to climate-related factors (rain and wind), 25.86 ha due to large-scale illegal logging, and 12.14 ha due to small-scale illegal logging. The core zone of the MBBR located in the State of Michoacan showed the highest forest cover loss values with 94.07 ha lost due to climate-related factors, and 38.0 ha lost due to illegal logging. Our study also showed a substantial decrease of ∼98% in large-scale illegal logging in the core zones of the MBBR compared to previous reported forest losses from 2001 to 2012. Forest cover loss was similar, yet the periods of the two studies differed, one 12 years in length, this one 6 years. The decrease of forest cover during the period studied suggests that factors elsewhere rather than forest cover loss in the monarch butterfly's winter habitat have strongly contributed to the dramatic population declines observed in monarch overwintering colonies since 2010.

Keywords: biosphere reserves, climate-related factors, forest cover loss, monarch butterfly, illegal logging, overwintering colonies, population declines

# INTRODUCTION

Protected areas are a cornerstone for conserving biodiversity worldwide (Margules and Sarkar, 2007). Most protected areas were decreed ad hoc for protecting scenic values, as refuges of cultural heritage, or to conserve specific places based on political criteria. Other protected areas have been decreed in areas that are biodiversity hotspots thereby contributing to their conservation (Room et al., 2000; Saura et al., 2018). Protected areas are also important for human well-being when they provide many environmental services, serve as areas of resilience to ameliorate negative impacts of climate change and other global change factors, and serve as refugia for the cultural heritage of local communities (Carey and Stolton, 2000; Hockings, 2003). Forest cover loss and fragmentation of habitats, illegal hunting, extraction of species, and overexploitation of their natural resources threaten the viability of protected areas worldwide (Hockings, 2003; Miranda et al., 2016). An increasing awareness of the importance of conservation and surveillance of protected areas has been recognized in many countries, and supported by national and international agencies, NGOs and academic institutions (Sánchez-Cordero et al., 2009; Saura et al., 2018).

The Monarch Butterfly Biosphere Reserve (MBBR) was decreed in 2000 to protect the monarch butterfly (Danaus plexippus) overwintering colonies in Mexico and to contribute to the conservation of the monarch migratory phenomenon (SEMARNAT, 2001; Vidal and Rendón-Salinas, 2014). The MBBR has been internationally recognized as an essential component of strategies for the conservation of the monarch butterfly migratory phenomenon due to the importance of its forests in which monarch overwintering colonies perch (Calvert et al., 1983; Alonso-Mejía et al., 1997; UNESCO, 2008; Vidal and Rendón-Salinas, 2014). However, the MBBR has faced continuous forest cover losses (Rendón-Salinas et al., 2005; Vidal et al., 2013; Vidal and Rendón-Salinas, 2014; Sarkar, 2017). Specifically, illegal logging, forest fires, and diseases causing damage to trees in the reserve are severe problems directly affecting forest cover that have negative impacts on monarch overwintering colonies. Further, the high demand for legal logging by local communities is also affected by these factors and has created social tensions between stakeholders from local communities and external agents participating in illegal activities (Honey-Rosés, 2009; Vidal et al., 2013; Vidal and Rendón-Salinas, 2014; Ramirez et al., 2015).

Forest cover loss has negative impacts on environmental services, ecotourism, opportunities for legal logging and wood extraction, and soil erosion. In addition, watersheds can be irreversibly harmed, contributing to potentially deleterious changes in microclimate (Rendón-Salinas et al., 2005; Vidal and Rendón-Salinas, 2014). In response to these problems, several conservation initiatives have been implemented with respect to the MBBR. For example, the Monarch Butterfly Conservation Trust (also known as the Monarca Fund) established in 2000, was created by the World Wildlife Fund (WWF) and the Mexican Fund for Nature Conservation (FMCN) with the financial support from the David and Lucile Packard Foundation, the former Mexican Secretariat for the Environment, Natural Resources and Fisheries (SEMARNAP), and the States of Michoacan and Estado de Mexico. It consists of a management tool based on economic incentives for the protection of the MBBR core zone forest habitats and is owned by stakeholders who have accepted restrictions on their exploitation rights and promoted conservation programs and ecotourism (Rendón-Salinas et al., 2005; Vidal and Rendón-Salinas, 2014). In 2016, the Mexican Federal Government established an initiative supporting the conservation and monitoring of the MBBR that included participation by governmental authorities, federal police, NGOs, and academic institutions (Honey-Rosés et al., 2009). Finally, the trinational initiative promoting the conservation of the monarch butterfly migratory phenomena sponsored by the Commission for Environmental Cooperation, was launched in 2015 by the Presidents of Mexico and the United States of America and the Prime Minister of Canada to ensure the conservation of the monarch butterfly migratory phenomenon (Trudeau et al., 2016). This initiative involved the creation of a trinational scientific working group for coordinating academic, governmental, and NGOs activities related to the conservation of the monarch butterfly migratory phenomenon.

In this study, we examined whether forest cover loss decreased since commencement of these initiatives. Specifically, we quantified recent forest cover losses in the core zones of the MBBR from 2012 to 2018 using satellite images and aerial orthophotographs complemented with field validation for temporal comparisons. Our goals were to (1) compare recent forest cover loss due to climate-related factors (wind and rain), and large-scale and small-scale illegal logging between years, and (2) analyze long-term forest cover losses in the core zones of the MBBR by comparing a previous study (2001–2012, Vidal et al., 2013) with our study (2012–2018).

# MATERIALS AND METHODS

### Study Area

The MBBR is located in the border between the States of Michoacan and Mexico along the Transvolcanic Belt in central Mexico. It was decreed in 2000 and consists of 56,259 ha (SEMARNAT, 2001). The MBBR is composed of three core zones in which most monarch overwintering colonies occur (Calvert and Brower, 1986; Rendón-Salinas et al., 2005; Galindo-Leal et al., 2009). A northern core zone is located in Cerro Altamirano (558 ha), a central core zone (9,671 ha) is located in Sierra de Chincua, Sierra El Campanario, and Cerro Chivatí-Huacal, and a southern zone (3,339 ha) includes Cerro Pelón (see Vidal and Rendón-Salinas, 2014). In these core zones use of natural resources is restricted. The MBBR includes two buffer zones in which sustainable use of natural resources is allowed, including supervised legal logging (SEMARNAT, 2001; Galindo-Leal et al., 2009; Vidal and Rendón-Salinas, 2014) (**Figure 1**). This protected area holds a high diversity of habitats, including pine forest (Pinus spp.), oyamel forest (Abies religiosa), pine-oak forest (Quercus spp), oak forest, and cedar forest (Cedrus spp), and has high biodiversity content including 493 species of vascular plants and 198 species of terrestrial vertebrates (SEMARNAT, 2001).

# Forest Cover Loss

Overall, we followed the methods provided by Brower et al. (2002) and Vidal et al. (2013). This analysis was quantified using satellite Quickbird sensor images (for 2012) and orthophoto images from the core zones of the MBBR from 2012 to 2018. These images were ortho-rectified and georeferenced using ArcMap editing tools (see below). A total of 45 images were obtained biennially covering the core zones of the MBBR at a resolution of 30 × 30 cm for comparison with the orthophoto images. The orthophoto images (aerial photographs, with the Argeomatica company) were obtained annually from February 2012, 2013, 2014, 2015, and 2016, May 2017, and March 2018. The image for 2012 that was used came from a previous study

FIGURE 1 | Location of the Monarch Butterfly Biosphere Reserve (MBBR), along the border (thick line) of the States of Michoacan and Mexico. The buffer zones are depicted in light green, and include areas where sustainable use of natural resources is allowed by local communities. The core zones of the MBBR are depicted in dark green, and include areas where use of natural resources is restricted. Monarch overwintering colonies (monarch sanctuaries) are depicted as a red monarch logo. The polygons depicted in the buffer zones represent the agrarian properties owned by the stakeholders [Registro Nacional Agrario, Mexico 2018 (National Agrarian Records, Mexico 2018)].

(Vidal et al., 2013). This overlap of imagery allowed visualization of long-term trends in forest cover loss in core zones of the MBBR in a continuous year sequence (2001–2018) by combining both studies. Forest cover loss was recorded in a shapefile in ArcGIS (v.10.5), allowing calculation of the arithmetic difference of forest cover loss in different habitats (see Vidal et al., 2013; Vidal and Rendón-Salinas, 2014). A buffer of 300 meters surrounding the core zones of the MBBR was established in the GIS platform. We generated a grid of hexagons (16 ha) covering the core zones of the MBBR and buffer area to compare images biennially.

Visual interpretation was performed using ArcMAP editing tools, "Effects-Swipe," which allowed images to be superimposed Flores-Martínez et al. Forest Cover Loss in MBBR

and compared. This tool facilitated the visual comparison of the two basic inputs to quantify forest cover loss. As a rule of interpretation, a screen scale was set to 1:3,000. This scale allowed a complete visualization of each hexagon of the grid, and ensured homogenization for the analyses (Vidal et al., 2013). Forest cover loss was estimated for the local communities and the core zones of the MBBR (**Figure 1**) (Vidal et al., 2013). Given our scale of analyses (1:3,000), other semi-automated tools in ArcMap can produce change detection errors, by the amount of topographic shadows that exist on the ground. We do believe that other ArcMap tools can be implemented in further priority studies of habitat loss at the MBBR. For example, including forest loss analyses of the remaining areas of the reserve.

Once forest cover loss was located and mapped, we proceeded with field validation using a GPS device configured with the exact route and position to the center of each affected polygon. Visits were made at least once to the 39 identified affected areas to record evidence of forest cover loss by climate-related factors (rain and wind) and illegal logging. Specifically, we classified forest cover loss into three categories: (1) climate-related factors due to wind and rain; (2) largescale illegal logging due to massive logging carried out by organized delinquent groups; and (3) small-scale illegal logging due to logging of a few trees by individuals from local communities (see Vidal et al., 2013). Treefall due to illegal logging was easily detected as they showed clear marks of saw or axes in the stumps. Treefall due to climatic factors did not show evidence of human activities. The field trips included previously trained personnel to record tree fall according to our classification from the MBBR, World Wildlife Fund (WWF), Mexican Fund for the Conservation of Nature (FMCN), the National Forestry Commission (CONAFOR), the National Commission of Natural Protected Areas (CONANP), Forest Protector (PROBOSQUE), the Federal Office of Environmental Protection (PROFEPA), Institute of Biology, UNAM (IBUNAM), Science and Community for Conservation AC (CCC), and representatives of agrarian properties.

#### RESULTS

Forest cover loss totaled 163.44 ha from 2012 to 2018 in the core zones of the MBBR (**Figure 1**). Forest cover loss due to climate-related factors was 125.44 ha (77%), large-scale illegal logging, 25.86 ha (15%), and small-scale illegal logging, 12.14 ha (8%) (**Table 1**, **Figure 2**). Forest cover loss due to climate-related factors peaked at 81.75 ha between 2015 and 2017 in our study; in March 2016, strong winds and rains produced a peak in treefall (55.21 ha) in the core zones of the MBBR (**Table 1**, **Figure 3**). The State of Michoacan reached higher forest cover loss values due to climate-related factors (94.07 ha) from 2012 to 2018 compared to the State of Mexico (31.37 ha) (**Table 1**, **Figure 3**). These values represent 0.23 and 0.63% of the total area in the core zone of each State, respectively (**Table 1**).

Illegal logging caused moderate forest cover loss of 38.0 ha from 2012 to 2018, but peaked with 21.61 ha during 2013 to 2015. The State of Michoacan was the most affected with 25.86 ha compared to the State of Mexico, where no evidence of large-scale logging was observed (**Table 1**, **Figure 2**). Large-scale illegal logging decreased after 2015 in both States (**Figure 2**). Small-scale illegal logging resulted in low forest cover loss values during 2012 to 2018, although it reached a peak of 4.47 ha during 2015 to 2017 (**Figure 3**). Small-scale illegal logging was marginally present in the State of Mexico with 1.50 ha affected. Large- and small-scale logging represented 0.30% and 0.04% of forest cover loss in the core zones of Michoacan and the State of Mexico, respectively (**Table 1**). Although small-scale illegal logging continues, it has a minor impact compared to largescale illegal logging in previous years. Over all, forest cover loss represented <1.0% and 0.30% in the core zones of Michoacan and the State of Mexico, respectively, and 1.21% in the core zones of the MBBR (**Table 1**, **Figures 2**, **3**).

# DISCUSSION

Forest cover loss in the core zones of the MBBR showed a decreasing trend from 2012 to 2018. Climate-related factors caused the highest damage to the forest cover, as shown by the high forest cover loss values in the core zones of the MBBR (**Table 1**, **Figures 2**, **3**). Previous studies have suggested the importance and persistence of climate-related factors affecting the monarch overwintering colonies (Brower et al., 2004, 2012; Narayani et al., 2012; Vidal et al., 2013; Vidal and Rendón-Salinas, 2014). Climate-related factors such as rain, wind and low temperatures directly cause high mortality in monarch populations in addition to treefall, as monarchs are highly vulnerable to low temperatures and rain while perched on trees, and usually fall to the ground resulting in high mortality (Brower et al., 2004, 2017; Narayani et al., 2012). For example, we observed a high monarch mortality at El Rosario colony in January due to a heavy storm (pers. obs.). Brower et al. (2017) and Vidal et al. (2013) reported a peak of forest cover loss in 2009– 2011 due to climate-related factors (**Figure 2**). Other natural factors such as disease or forest fires affecting trees can increase treefall. Thus, climate-related factors appear to play a crucial role affecting monarch overwintering colonies either by treefall due to strong wind and rain, diseases and forest fires, and causing high monarch mortality as well by exposure to adverse climatic conditions (Narayani et al., 2012; Vidal et al., 2013; Vidal and Rendón-Salinas, 2014; Ramirez et al., 2015).

Though the period of observations differed (12 years in Vidal et al., 2013, vs. 6 years in this study), the magnitude of forest cover loss from climate-related factors was similar (120 ha vs. 130 ha; **Table 1** and **Figures 2**, **3**) (Vidal et al., 2013). Previous studies have also reported adverse effects of climate-related factors on MBBR forest cover and on monarch overwintering colonies (Narayani et al., 2012). Climate-related factors as heavy wind and rain resulted in extensive forest cover loss and high mortality in monarch overwintering colonies at the MBBR in 1981 (Calvert et al., 1983) and in 1992 (Culotta, 1992). Brower et al. (2004) described a high number of mortality of monarchs due to climate-related factors, and Ramirez et al. (2015) suggested that forest cover loss is due to an additive effect TABLE 1 | Recent forest cover loss (ha) due to climate-related factors (wind and rain), and large-scale and short-scale illegal logging in the core zones of the Monarch Butterfly Biosphere Reserve (MBBR).


Forest cover losses were quantified by comparing annually orthophoto images for the MBBR core zones from 2012 to 2018. The number of hectares is depicted by years and by State, and the percentage of forest cover loss in core zones of each State, and in the core zones of the MBBR, respectively, is included (see Methods for more details).

of a poor land management and illegal logging, and climaterelated factors. Vidal et al. (2013) reported a strong negative impact of climate-related factors on monarch overwintering colonies between 2005 and 2007, and Vidal and Rendón-Salinas (2014) observed the decreased of several overwintering colonies in the MBBR caused by climate-related factors between 2004 and 2007. The most recent damage occurred in March 2016, where MBBR was affected by climate-related factors (rain and snow storms) producing high damage to the Oyamel forests and consequently to the overwintering colonies from the sanctuaries of Sierra Chincua and Cerro Pelón showed a mortality of 31 and 38% respectively, while Sierra Campanario (El Rosario colony) showed a mortality of 40% (Brower et al., 2017). Further studies should also monitor the adverse impact of climate-related factors on trees and feeding plants not only in the MBBR, but along the monarch migratory route, as damage to trees and plants can strongly affect monarch's perching and feeding sites (Oberhauser and Peterson, 2003; Batalden et al., 2007; Lemoine, 2015).

Forest cover loss from large- and small-scale illegal logging was lower than from climate-related factors in the core zones of the MBBR. Moreover, large- and small-scale illegal logging decreased from 2012 to 2018. Large-scale illegal logging has been absent since 2015, and only small illegal logging is still marginally present. The State of Michoacan showed higher forest cover loss rates due to both large and small-scale illegal logging compared to the State of Mexico in the core zones of the MBBR. A decrease in forest cover loss due to illegal logging was observed between an earlier study (2001–2012) (Vidal et al., 2013) and our study (2012–2018). Vidal et al. (2013) reported a total of 2,057 ha forest cover loss by both large-scale and small-scale illegal logging in an 11-year period, while our study reported <50 ha in an 8-year period. These results indicate approximately a 98% reduction in illegal logging in the core zones of the MBBR in recent years. Conversely, forest cover loss due to climate-related factors was similar; 122 ha were reported from 2001 to 2012 (Vidal et al., 2013), and 125.44 ha in our study. Overall, forest cover loss due to

FIGURE 3 | Maps depicting forest cover loss due to climate-related factors (yellow), and large-scale (red), and small-scale (orange) illegal logging in the core zones of the MBBR (light green and dark green). Forest cover loss includes from 2012 to 2018.

both climate-related factors and illegal logging was lower in the cores zone of State of Mexico (0.27%) compared to Michoacan (0.93%), and only 1.21% in the core zones of the MBBR (see Brower et al., 2017) (**Table 1**).

The recent overall decrease of large and small-scale illegal logging is likely explained by a shared effort of stakeholders, Mexican government, NGOs, academic institutions, and philanthropists. Several actions have been implemented that have succeeded in preventing illegal logging in crucial areas of monarch overwintering colonies. For example, in 2016, the Mexican government established a program of a strict surveillance in the core zones of the MBBR involving environmental police in coordination with local communities, NGOs, and other stakeholders. This program is ongoing.

Further, the Mexican government continues to support stakeholders and local communities with specific conservation programs. The Federal Environmental Protection Agency (PROFEPA) continues a monitoring program for natural resources, and the Ministry of Environment and Natural Resources (SEMARNAT) maintains a program of payment for environmental services, which mitigates the overexploitation of the MBBR forests. Specifically, more than 330 programs were established between 2014 and 2018 supporting local communities located inside and in the vicinity of the MBBR. The subsidies granted as the Temporary Employment Program (PET), Conservation Program for Sustainable Development (PROCODES) and the Community Surveillance Programs (PVC) are examples of those that have benefitted local communities; 55% of the projects supported are aimed at the conservation and monitoring of the MBBR forests.

The role of NGOs has also been instrumental for conserving MBBR forests. World Wildlife Fund (WWF) is involved in the conservation of monarch overwintering colonies (Vidal et al., 2013; Vidal and Rendón-Salinas, 2014). It coordinates the development of forest protection programs, reforestation projects, development of production programs, and environmental monitoring with incentives to promote the conservation of the forests in the core zones of the MBBR. Other Mexican NGOs that are involved include the Monarch Fund, which grants payments to local communities for the protection of forests, the Mexican Fund for the Conservation of Nature, AC, Alternare AC, and Monarch Butterfly Fund (MBF) aimed at protecting the forest in the core zones of the MBBR. The coordinated efforts involving the Mexican government, national and international NGO and agencies, and academic institutions have increased the surveillance and monitoring programs aimed at conserving forest in the MBBR, particularly in the core zones of the MBBR, where most monarch overwintering colonies perch and feed (Narayani et al., 2012; Vidal et al., 2013). Our study showed that illegal logging has substantially decreased in recent years in the core zones of the MBBR, indicating that these conservation efforts have been successful and should continue.

Recent studies have documented a dramatic reduction in monarch overwintering colonies of more than 90% in recent years (Vidal and Rendón-Salinas, 2014; Saunders et al., 2019). One question that continues to be debated is whether the decrease in monarch overwintering colonies is mainly due to forest cover loss in the core zones of the MBBR where monarch overwintering colonies occur (Brower et al., 2011). Other alternative proposals suggest that the dramatic decrease of monarch overwintering colonies is a combination of significant reductions of milkweed populations due to herbicides in extensive areas throughout the east and midwest US or that the decline is due to increasing mortality during the fall migration (Pleasants and Oberhauser, 2013; Agrawal, 2017; Sarkar, 2017; Thogmartin et al., 2017a; Saunders et al., 2019).

Our study contributes to understand monarch population declines by showing that, overall, forest cover loss in the core zones of the MBBR has substantially decreased in recent years and that this decrease is due to the prevention of large-scale illegal logging. Thus, problems generated by human activities in winter habitat are unlikely to be a determinative factor in the etiology of the population decline. In this context, it is important to highlight the success of the coordinated efforts of the Mexican government, NGOs, and national and international agencies and academic institutions in implementing the necessary conservation strategies. Forest cover loss in core zones of the MBBR due to climate-related factors needs continued monitoring and integrated to any conservation program that has been and will be undertaken. The conservation of the monarch migratory phenomena is

#### REFERENCES


a complex task and requires an internationally coordinated effort to address the challenges: prevent illegal logging in the core zones of the MBBR; increase milkweed populations and avoid toxic herbicides to ensure nectar availability to monarch butterflies throughout its migratory route and breeding areas, and restoring and maintaining habitats along the fall monarch migration (Brower et al., 2011; Agrawal, 2017; Thogmartin et al., 2017b). The trinational initiative between Mexico, the US and Canada aimed to conserve the monarch migratory phenomena can play an important role in promoting a shared responsibility that requires international coordination and cooperation on a continental scale.

# DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

# AUTHOR CONTRIBUTIONS

JF-M, VS-C, and ER-S conceived and designed the research, and conducted data analyses. All authors wrote the manuscript.

#### ACKNOWLEDGMENTS

We thank the personnel at MBBR, the Alliance WWF Foundation Telmex Telcel, Mexican Fund for Nature Conservation (FMCN), the National Forestry Commission (CONAFOR), The National Commission of Protected Natural Areas (CONANP), the Protector of Forests (PROBOSQUE), the Federal Procurator for Environmental Protection (PROFEPA), the Instituto de Biología [Laboratorio Nacional de Biodiversidad (LANABIO)], Universidad Nacional Autónoma de México, Science and Community for the Conservation AC (CCC), Biological Conservation and Social Development AC (CONBIODES), Conservation Wind AC and the representatives of the local ejidos and communities. We are particularly thankful for the comments of two reviewers and the Editor who made substantial comments that improved the presentation of this work.


march 2016 storm in the monarch butterfly Biosphere reserve in Mexico. Am. Entomol. 63, 151–164. doi: 10.1093/ae/tmx052


**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 Flores-Martínez, Martínez-Pacheco, Rendón-Salinas, Rickards, Sarkar and Sánchez-Cordero. 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 Restoration of Abies religiosa Forests Using Nurse Plants and Assisted Migration in the Monarch Butterfly Biosphere Reserve, Mexico

Aglaen Carbajal-Navarro<sup>1</sup> , Esmeralda Navarro-Miranda<sup>2</sup> , Arnulfo Blanco-García<sup>1</sup> , Ana Laura Cruzado-Vargas <sup>2</sup> , Erika Gómez-Pineda<sup>2</sup> , Cecilia Zamora-Sánchez <sup>2</sup> , Fernando Pineda-García<sup>3</sup> , Greg O'Neill <sup>4</sup> , Mariela Gómez-Romero1,5 , Roberto Lindig-Cisneros <sup>6</sup> , Kurt H. Johnsen<sup>7</sup> , Philippe Lobit <sup>2</sup> , Leonel Lopez-Toledo<sup>8</sup> , Yvonne Herrerías-Diego<sup>1</sup> and Cuauhtémoc Sáenz-Romero<sup>8</sup> \*

#### Edited by:

Jay E. Diffendorfer, United States Geological Survey (USGS), United States

#### Reviewed by:

Carolina Martínez-Ruiz, University of Valladolid, Spain David Gorchov, Miami University, United States

\*Correspondence:

Cuauhtémoc Sáenz-Romero csaenzromero@gmail.com

#### Specialty section:

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

Received: 22 December 2018 Accepted: 18 October 2019 Published: 06 November 2019

#### Citation:

Carbajal-Navarro A, Navarro-Miranda E, Blanco-García A, Cruzado-Vargas AL, Gómez-Pineda E, Zamora-Sánchez C, Pineda-García F, O'Neill G, Gómez-Romero M, Lindig-Cisneros R, Johnsen KH, Lobit P, Lopez-Toledo L, Herrerías-Diego Y and Sáenz-Romero C (2019) Ecological Restoration of Abies religiosa Forests Using Nurse Plants and Assisted Migration in the Monarch Butterfly Biosphere Reserve, Mexico. Front. Ecol. Evol. 7:421. doi: 10.3389/fevo.2019.00421 <sup>1</sup> Facultad de Biología, Universidad Michoacana de San Nicolás de Hidalgo (UMSNH), Morelia, Mexico, <sup>2</sup> Instituto de Investigaciones Agropecuarias y Forestales (IIAF), UMSNH, Morelia, Mexico, <sup>3</sup> Escuela Nacional de Estudios Superiores (ENES), Universidad Nacional Autónoma de México (UNAM), Morelia, Mexico, <sup>4</sup> British Columbia Ministry of Forests, Lands and Natural Resource Operations and Rural Development, Vernon, BC, Canada, <sup>5</sup> Cátedras CONACyT, Facultad de Biología, UMSNH, Morelia, Mexico, <sup>6</sup> Instituto de Investigaciones en Ecosistemas y Sustentabilidad, UNAM, Morelia, Mexico, <sup>7</sup> Southern Research Station, USDA Forest Service, Asheville, NC, United States, <sup>8</sup> Instituto de Investigaciones Sobre los Recursos Naturales (INIRENA), UMSNH, Morelia, Mexico

Increasing disturbance events (forest fires, windstorms, pest outbreaks) associated with climate change are creating new ecological restoration challenges. Here, we examine the utility of assisted migration in combination with naturally established nurse plants in order to improve the success of afforestation with Abies religiosa (sacred fir), the overwintering host of the Monarch butterfly (Danaus plexippus). We established high-elevation assisted migration A. religiosa provenance field tests at two sites (Las Palomas and Los Ailes) in the core overwintering areas of the Monarch Butterfly Biosphere Reserve (MBBR), central Mexico. At each site, 2-year-old seedlings were planted either in the open or under existing nurse plants. Three and a half years after planting at the Las Palomas site (5.5 years from germination), A. religiosa seedling survival was 72% under the shade of nurse plants but only 18% in open areas (without shade). At the Los Ailes site, 1 year and a half after planting in the field (3.5 years from germination), survival was 94% and 10%, respectively. There were not significant differences in seedling height increment among populations at either site. The results of our study and those published elsewhere suggest that A. religiosa benefits from shade protection of nurse plants and that population transfer 400 m upward in elevation (i.e., assisted migration) to compensate for future warmer climates does not appear to have any negative impacts on the seedlings, while potentially conferring closer alignment to future climates. If absent in planting sites, we recommend growing nurse shrub species (such as Baccharis conferta) alongside tree seedlings in forest nurseries so that these shrubs can be transplanted to the reforestation site 2 years before planting the tree seedlings.

Keywords: Abies religiosa, assisted migration, climatic change, nurse plants, reforestation, Monarch butterfly, overwintering sites, ecological restoration

# INTRODUCTION

Abies religiosa (Kunth) Schltdl. & Cham. (sacred fir) forests, which comprise the Monarch butterfly (Danaus plexippus) overwintering sites in the core area of the Monarch Butterfly Biosphere Reserve (MBBR) at the border between the Michoacán and Estado de México states in west-central Mexico, are occasionally subject to serious disturbances, likely related to climate change. For example, in a single night in March 2016, a windstorm felled approximately 20,000 trees and damaged many more (Brower et al., 2017; Fondo Monarca, 2017; **Figure S1**). Such disturbances undoubtedly reduce stand density (**Figure S2**) and, when combined with illegal cutting (Brower et al., 2016; **Figure S3**), are decreasing the umbrella and blanket effect (sensu Anderson and Brower, 1996) that protects overwintering Monarch butterfly colonies. There is an urgent need for ecological restoration in the face of ongoing climate change, as reflected in the declining Monarch butterfly populations (Semmens et al., 2016) and an agreement between the federal governments of Mexico, United States and Canada regarding conservation of habitat targets in the MBBR (Environment Canada, 2014). In this study, we explore the use of nurse plants as facilitators for the establishment of A. religiosa seedlings in "climate-smart" reforestation programs and consider the A. religiosa seed source as a tool with which to improve adaptation to climatic change.

Abies religiosa is a shade tolerant species (Sánchez-Velásquez et al., 1991; Rzedowski, 2006) distributed on moist and cold sites at high elevations with more affinity to northern aspects, mostly along the Trans-Mexican Volcanic Belt, at between 2,800 and 3,500 m above sea level (m a.s.l.) (Rzedowski and Rzedowski, 2005; Benavides-Meza et al., 2011).

The use of local seed sources of A. religiosa when reforesting the MBBR may no longer be appropriate. A. religiosa populations are genetically differentiated for adaptative quantitative traits (seedling height, seedling annual elongation, date of growth cessation, foliage dry weight and frost resistance) along environmental gradients. In particular, populations separated by as little as 350 m in elevation, or 1.2◦C in mean temperature of the coldest month (MTCM), differ genetically; there is a very strong association between A. religiosa seedling frost damage and elevation of the provenance (and its corresponding MTCM), suggesting that climate is the selective driving force in terms of genetic differentiation among populations (Ortiz-Bibian et al., 2017). There is also evidence of genetic differentiation among populations of A. religiosa at the landscape scale, detected with molecular markers [amplified fragment length polymorphisms (AFLPs) and three chloroplast microsatellites (cpSSRs) (Méndez-González et al., 2017)]. There are consistent reports of genetic differentiation for adaptive quantitative traits among conifer populations along environmental gradients, in which a vital trade-off is expressed: higher growth potential and lower frost resistance for populations growing in mild environments, and vice versa in harsh colder environments (see for example: Rehfeldt et al., 1999, 2002, 2003, 2014a,b, 2017). Furthermore, ongoing climate change will act to decouple host (A. religiosa) populations from the relatively narrow climate interval to which they are adapted, with the suitable climatic habitat for A. religiosa inside the MBBR expected to disappear entirely by the end of the century (Sáenz-Romero et al., 2012).

Assisted migration is the practice of humans moving populations (in our case, tree populations by seeding or transplant) to a different habitat. Such practice could be used to alleviate the impacts of increased warming if species′ historic climatic envelope can be matched to the future climate at a transplant location (Rehfeldt et al., 2014c). Some authors do not distinguish between translocating genotypes or populations inside versus outside a species' historic natural distribution (Rehfeldt et al., 1999, 2002, 2012; O'Neill et al., 2008; Marris, 2009; Pedlar et al., 2012; Castellanos-Acuña et al., 2015; Prober et al., 2015; Sáenz-Romero et al., 2016). Other authors (e.g., Hewitt et al., 2011; Commander et al., 2018) consider assisted migration to involve only the translocation to locations outside the historic range of the species. By planting A. religiosa from seed sources collected in locations with historic (e.g., period 1961– 1990) climates that match the MBBR climates anticipated for the near future (e.g., the decade centered on the years 2030 or 2050), populations could be realigned with their suitable climatic habitat (in this case, inside the historic range of the sacred fir) and misalignment may thus be mitigated (Sáenz-Romero et al., 2012).

The climate distance that seed sources are moved (i.e., the migration distance) may be a critical factor in the success of assisted migration efforts. If migration distances are too short (i.e., sites are established with populations whose contemporary climates are only slightly warmer than that of the planting site), planted trees may be maladapted to the prevailing climate by the time the stand reaches late maturity. However, if migration distances are too great (i.e., sites are established with populations whose contemporary climates are much warmer than that of the planting site), planted trees may be susceptible to frost damage during their period of establishment (Loya-Rebollar et al., 2013). Clearly, the target for optimum migration distance lies between the climates expected during these two stand phases (establishment and late maturity). Balancing the two risks, and considering that trees are most sensitive to climate extremes in the establishment phase, we propose a migration target of 2030 in order to maximize adaptation during establishment (Sáenz-Romero et al., 2010; Ortiz-Bibian et al., 2017). Similarly, in British Columbia, Canada, a commercial system of large-scale assisted migration was recently introduced using a migration distance that accounts for climate change in the last 70 years and for the change anticipated in the next 20 years (O'Neill et al., 2017). In Mexico, however, there are no experiences of assisted migration conducted on commercial or large scale conservation programs—only those at small experimental scale, where germplasm migration has generally been successful where it did not exceed 400 m upward elevation shift (e.g., Valle-Díaz et al., 2009; Castellanos-Acuña et al., 2015; García-Hernández et al., 2019; Gómez-Ruiz et al., 2019).

Shrubs can serve as nurse plants, facilitating the establishment of target (due to their economic or ecological value) species, where harsh environmental conditions act as barriers to restoration (Callaway et al., 2002; Filazzola and Lortie, 2014). Such ecological restoration using nurse plants as facilitators has been achieved under extremely harsh conditions, e.g. (a) growing as "krummholz mats" in cold and windy timberline sites, serving as nurse plants for Picea engelmanii and Abies lasiocarpa (Germino et al., 2002; Brodersen et al., 2019); (b) at abandoned coal mine sites in Spain, using native shrubs Genista florida and Cytisus scoparius as nurse plants to facilitate the establishment of Quercus petraea and Q. pyreniaca seedlings (Torroba-Balmori et al., 2015; Alday et al., 2016); (c) in grazing areas in dry Mediterranean environments in Spain (using Cytisus multiflorus as the nurse plant for Quercus pyrenaica and Q. ilex; Costa et al., 2017); and (d) at low altitude, sunny, dry Mediterranean slope sites (Gómez-Aparicio et al., 2004).

Similarly, in restoration ecology efforts in Michoacán state, western Mexico, survival of planted A. religiosa seedlings increases when planted in conjunction with nurse plants. For example, Lupinus elegans seeds were sown together with A. religiosa seedlings in an abandoned farm field at a site originally occupied by A. religiosa and Pinus pseudostrobus forest. When L. elegans covered the A. religiosa seedlings, the mortality of the latter essentially fell to zero (Blanco-García et al., 2011). These results indicate the potential utility of nurse plants to A. religiosa, especially in the context of forest management for climate change.

Assisted migration of A. religiosa genotypes inside the core area of the MBBR would require an upward shift of 350 m in elevation in order to restore populations to their 1961–1990 climatic origin and account for the climatic change projected for the decade centered on the year 2030 (Ortiz-Bibian et al., 2017). Since most of the Monarch butterfly overwintering sites are between 3,000 and 3,300 m asl (García-Serrano et al., 2004), nurse plants must be resistant to severe frost; however, L. elegans is sensitive to the frequent frosts that occur at those elevations (Díaz-Rodríguez et al., 2013). Lupinus montanus, a native species found at the over-wintering elevations, is more frost-tolerant but usually grows to only 40 cm in height, which is too short to provide shade to A. religiosa during the critical establishment phase. Thus, taller native plants, such as the shrub Baccharis conferta Kunt (Snook, 1993; Lara-González et al., 2009; Sánchez-Velásquez et al., 2011), should be explored as possible nurse plants for A. religiosa.

The objective of the present study was to test the feasibility of using naturally established nurse plants to moderate possible stress associated with population migration during seedling establishment in sites within the core area of the MBBR that are disturbed and increasingly hostile as a result of climate change. Our specific research question was: What are the differences in survival and growth for A. religiosa seedlings when planted with and without the protection of existing naturally established shrubs serving as nurse plants, when the A. religiosa seed originates from an elevation lower than that of the planting site?

# MATERIALS AND METHODS

#### Seed Origin and Design of Field Tests

We established two high-elevation A. religiosa provenance field tests: Las Palomas at 3,440 m and Los Ailes at 3,360 m elevation. Both sites are inside the core area of the MBBR, at Ejido La Mesa, Municipality of San José del Rincón, Estado de México (**Table 1**; **Figure 1**), and both were natural pure old A. religiosa stands that had been heavily disturbed: Las Palomas by a severe

TABLE 1 | Geographic coordinates, elevation, mean annual temperature (MAT), mean annual precipitation (MAP) and number of frost-free days per year (NFFD) for the Abies religiosa provenances and for the test sites at Las Palomas and Los Ailes, both within the core area of the MBBR, Mexico.


Provenance climate is the mean for the period 1961–1990. Site climate is the mean for 1981–2010.

crown fire in 1989 and Los Ailes by deforestation and subsequent sheep grazing ∼2,000. We use the term "population" to refer to a group of open-pollinated individuals represented in the tests by their seedlings and "provenance" as the geographic origin of a population. Our operational definition of seasons is illustrated in **Figure S4**.

For the Las Palomas test site, open-pollinated seeds were collected from 11 randomly selected trees (five to ten cones per tree), in each of 10 natural populations from between 3,000 and 3,450 m on the San Andrés mountain, located 35 km northwest of the provenance test site (**Table 1**; **Figure 1**). Seeds from each population were bulked (i.e., seeds from the 11 trees of each population were combined) and sown in a forest nursery in Morelia, Michoacán (1,830 m), then transferred after 1 year to the Ejido Los Remedios communal forest nursery (3,000 m, under a shade-mesh, with herbivores excluded) inside the MBBR for a full second year as a hardening treatment. Since germination was very poor (averaging 13%, but with large differences among populations, and extremely poor at both altitudinal extremes; Ortiz-Bibian et al., 2019) and there was high nursery mortality due to frost and other maintenance problems, the number of seedlings available for field testing was limited.

For the Los Ailes provenance trial, in October 2015, recently germinated seedlings from mossy microsites were collected under six natural stands of A. religiosa, between 2,960 and 3,450 m and up to 20 km from Los Ailes (**Table 1**; **Figure 1**) and transplanted to nursery containers (initially around 200 seedlings per population), where they were

nurse plants; yellow squares blocks planted on open areas. (B) Planting of Abies religiosa seedlings under the shade of nurse plants (Baccharis conferta in this case) and (C) the open treatment without nurse plants. Las Palomas site, 3,440 m, Ejido La Mesa, Municipality of San José del Rincón, Estado de México, core area of the Monarch Butterfly Biosphere Reserve. The site was heavily impacted by a crown fire in 1989. Prior to this fire, the site consisted of a dense Abies religiosa stand and an overwintering site for Monarch butterflies. Francisco Ramírez-Cruz, ejidatario of La Mesa, is pointing to recently planted A. religiosa seedlings. Appropriate informed consent was provided by Francisco Ramírez-Cruz in terms of appearing in this photograph indicating his personal identity.

grown for 2 years at Ejido La Mesa, also as a hardening treatment (3,000 m, inside a shade-house, with herbivores excluded), before being out-planted in July 2017. The number of transplanted seedlings (henceforth, simply "seedlings") that died at the nursery just after transplanting from the field also limited the availability of seedlings for the field tests.

Seedlings were planted at both field sites with two treatments: (a) under the shade of nurse shrubs already existing at the site—B. conferta (the most abundant), Ribes ciliolata, Juniperus monticola, Salix paradoxa, and Senecio cinerarioides—and (b) in open areas without the shade of nurse plants. The few existing shrubs on microsites chosen for the planting in the open at Las Palomas site were removed in order to achieve the condition of full absence of shrubs for the treatment of planting in open areas (**Figures 2**, **3**).

In July 2015, seedlings (aged 2 years from germination) were planted in the rainy season at the Las Palomas test site following a randomized complete block design, where five blocks each were assigned to nurse plant and open field treatments. In each block, the 10 populations were each randomly assigned to a single plot. Populations were represented by three seedlings in a row in each plot. Spacing was 1.5 × 1.5 m in a regular grid (**Figure 2**).

In July 2017, seedlings (aged 2 years from germination) were planted in the rainy season at the Los Ailes site. Seedlings from six

provenances (see **Table 1**) were planted in 50 blocks with nurse plants and in 20 main plots without nurse plants (in open field treatment). With the experience gained in the Las Palomas site 2 years previously, the blocks in the Los Ailes site were composed of one seedling from each of the six populations, randomly selected and planted in a circle around the stem of the nurse plants, similar to the methodology used by Costa et al. (2017) (**Figure 3B**), and around an imaginary point for the blocks without nurse plants (open field; **Figure 3C**). This unorthodox arrangement of seedlings in the blocks helped ensure that all the seedlings under the nurse plant treatment received approximately equal levels of shade. Uniform application of shade was not well-achieved at the Las Palomas site, since a regular grid arrangement was used for the planted seedlings within the existing irregular location of naturally established nurse plants.

#### Field Measurements and Analysis

Seedling height and survival were assessed periodically (approximately every 2 months) for 3.5 years at Las Palomas

(when they reached 5.5 years of age from germination) and 1.5 years at Los Ailes (3.5 years of age from germination). When we measured seedling height, the status of the seedling was examined; if it appeared to be alive, we recorded survival as "1"; if not alive, we recorded survival as "0" and did not record seedling height.

brackets indicate the season of the year (seasons defined as in Figure S4).

To characterize the amount of shade received by the plants, the percentage of plant cover was estimated at the end of the dry and warm season (May) with a digital image analyzer (Winscanopy, Regent Instruments, Canada), by taking a photograph with a hemispheric camera from the center of each block, at ground level. Images were processed with the Winscanopy program (2014).

Temperatures at 40 cm above ground surface (the approximate seedling height 2 years after planting) were monitored hourly at the Las Palomas site using dataloggers (Hobo <sup>R</sup> H01-001-01 Onset Computer Corporation, USA). One sensor was installed in each block for two periods: January 20th to May 31st, 2016, to record the dry season, and December 3rd, 2016 to February 13th, 2017, to record the coldest part of the dry and cold season.

In order to obtain a detailed record of the temperature fluctuations to which the seedlings were exposed, air temperature in the close environment of A. religiosa seedlings grown either in the open field or under a nurse plant was measured continuously during a day/night cycle. Simultaneously, the sky (or sky and nurse plant) brightness temperature (taken as a proxy for downwelling long wave radiation) and the soil brightness temperature (an approximation of soil surface temperature) were measured. For this purpose, a device was built using two infrared (IR) temperature sensors (Melexis <sup>R</sup> MLX 90614, 40◦ view angle) inserted within two horizontal polystyrene plates and installed in the field close to and at the same height as the seedlings. The sensors measured air temperature between the plates, while the IR sensors pointed both up and down to measure the sky and soil brightness temperatures, respectively. The system was connected to an Arduino <sup>R</sup> data-logger, which recorded measurements every 3 min. Two devices were installed and monitored over a 24-h period: one in the open field and the other under a nurse plant (B. conferta shrub) at the Los Ailes site in January (a month of the cold and dry season) and April (a month of the dry and warm season).

A visual stress index was created to evaluate the physiological status of the plants, starting the year after planting. The index values ranged from 1 (shiny dark-green needles, indicating a healthy non-stressed seedling) to 6 (brown decaying foliage, apparently dead seedling) (**Figure S5**). Stress assessments were conducted after planting at both sites from December to February, to represent most of the cold and dry season, and then also from March to early May, months of the warm and dry season, as well as during the rainy season (June to October).

Analysis of variance (ANOVA) was performed using Proc Mixed in SAS (SAS Institute Inc, 2004) to test the main effects of treatment, block and population, and their interactions, on final height increment (final seedling height minus seedling height at the time of field planting) and stress index. Treatment was considered a fixed effect; all remaining effects were considered random effects. Significance was tested using the COVTEST option for the random terms and the F-test for the fixed effects.

Survival data were examined with a General Linear Model, using a binomial distribution, logit link function, with the GLM module of the R program, and testing with Chisquare (Crawley, 2013).

Analyses were performed separately for each site due to the different experimental designs used at the two sites.

TABLE 2 | Significance (P) of the sources of variation in the ANOVA for Abies religiosa seedlings for survival (using a General Linear Model analysis, binomial distribution), height increment and stress index (using a mixed model), at the Las Palomas and Los Ailes assisted migration provenance test sites, 5.5 and 3.5 years from germination (3.5 and 1.5 years after planting in the field), respectively.


Treatments were planting under nurse plants and planting in the open field. Stress Index data from March 2017 for the Las Palomas site and March 2018 for the Los Ailes site. d.f., Degrees of freedom; n.d.f., numerator degrees of freedom.

# RESULTS

Three and a half years after planting at the Las Palomas site (5.5 years old from germination), A. religiosa seedling survival was 72% under the shade of nurse plants, but only 18% when planted in open areas (**Figure 4A**). In the Los Ailes site, one and a half years after planting (3.5 years-old from germination), seedling survival was 94% and 10%, under nurse shrubs and in open areas, respectively (**Figure 4B**). Differences between treatments were highly significant at both sites (P < 0.0001; **Table 2**). The timing of the pulses of mortality for the seedlings planted in open areas differed between sites: for the Las Palomas site, the mortality pulse occurred during the dry and warm season (March to early May; **Figure 4A**), whereas in the Los Ailes site, it occurred during the second half of the cold and dry season (January to February; **Figure 4B**).

Measurement of the coverage using a hemispheric camera indicated a highly significant difference between the treatments in open areas and under nurse plants, at both the Las Palomas (mean percent of coverage ± interval coefficient at 95%, in open areas: 48.9 ± 11.3; under nurse plants: 10.4 ± 3.76%; P < 0.001) and the Los Ailes (85.3 ± 1.7%; 36.2 ± 2.3%; P < 0.001; respectively) sites (**Figure 5**).

The temperatures measured with Hobo dataloggers were cooler under the nurse plants than in open areas (**Figure 6**) and survival was positively related to shade (**Figure 5**).

However, perhaps the greatest difference between the treatments in open areas vs. under nurse plants was not the average daily air temperatures (**Figure 6**), but rather the different patterns of temperature variation experienced by seedlings during the day. There was a very wide variation of temperatures in the open areas, while the temperature variations recorded below the nurse plants were much smaller (**Figure 7**).

Survival differed little among populations in the shade treatment (**Figures 8A,C**). However, the differences in survival among populations in blocks without existing nurse plants appeared to be larger (**Figures 8B,D**), and differences among populations were found to be statistically significant only when the analysis was conducted separately per shade treatment, both at the Las Palomas site (**Table 3**) and at the Los Ailes site (**Table 4**). That is consistent with the significance of the interaction population by treatment at the Las Palomas site (P = 0.0026; **Table 2**). In the open area planting of the Los Ailes site (**Figure 8D**), greater and earlier mortality was found in the population from the second lowest elevation (i.e., transferred from 3,052 m at the seed source to 3,360 m at the test site); in contrast, the lowest mortality occurred in the population that originated at the highest elevation (3,450 m at the seed source, close to the test site). This pattern was not found under the shade of the nurse plants (**Figure 8C**). A similar pattern was found at the Las Palomas site: in the open area planting, the lowest survival was for the population that was migrated the furthest in elevation (from 3,000 m at the seed source to 3,440 m. at the field test site), whereas the highest survival was for the local population (3,450 m at the seed source) (**Figure 8B**). In contrast, these differences were buffered under the shade of the nurse plants (**Figure 8A**). When plotting the average final survival per population against elevation of the provenance (**Figure S6**), it becomes apparent that under the shade of nurse plants, there is not a pronounced patterning of differences among populations associated with the

elevation of origin, whereas when planting on open areas, the best survival is of the populations originated at the highest altitudes, closer to the elevation of the planting sites.

There were no statistically significant differences in seedling height increment among treatments at the Las Palomas site, although there were significant differences at the Los Ailes site (**Table 2**). There were not significant differences among populations for seedling height increment at both sites (**Table 2**), even when the maximum upward shift in elevation was 440 m from seed source to planting site at the Las Palomas site, and of 400 m at the Los Ailes site.

The effect of the nurse plants was reflected in the stress status of the seedlings, as measured by the stress index. Seedlings under the shade of nurse plants had significantly lower stress index values (indicating reduced stress) than those without nurse plants (**Figure 9**; index values explained on **Figure S5**) in both sites (Las Palomas: P = 0.0007; Los Ailes: P < 0.0001; **Table 2**). There were no significant differences among populations for stress index in either test site (**Table 2**).

### DISCUSSION

Our results strongly suggest that nurse plants provide shade protection to recently planted A. religiosa seedlings, significantly increasing their survival regardless of the seedling seed source. Thus, the use of nurse plants in the early stages of a reforestation program with A. religiosa is highly recommended, as was found with planted Lupinus elegans serving as nurse plants in a previous study (Blanco-García et al., 2011) and, in our case, with the existing nurse plants (mostly the shrub Baccaris conferta). Our results are very similar to the patterns of survival found when Quercus petraea, Q. pyrenaica, and Q. ilex were grown under the shade of native shrubs (Genista florida, Cytisus scoparius, and C. multiflorus), and without the nurse effect, in studies conducted in harsh environments (Torroba-Balmori et al., 2015; Alday et al., 2016; Costa et al., 2017). In recent decades, the determinant role of facilitation has been recognized as a positive interaction among species that directly affects their performance, distribution and metabolism (Bruno et al., 2003; Brooker et al., 2008). The role of shade as a modifier of microclimatic conditions is important because this mechanism of improving microclimatic conditions is one of the most recognized forms of facilitation (Callaway et al., 2002; Callaway, 2007; Walker et al., 2007; Torroba-Balmori et al., 2015; Alday et al., 2016; Costa et al., 2017). The effects of nurse plants are of particular importance to plants growing under severe climatic conditions, as predicted by the stress gradient hypothesis (Bertness and Callaway, 1994; Callaway et al., 2002).

At both sites, seedling survival was substantially lower in the open areas than under the nurse shrubs. The nurse plants significantly increased the amount of shade received by the A. religiosa seedlings. A. religiosa is a shade tolerant species (Rzedowski, 2006) and thus greater height increment and survival are to be expected under shade. Differences in survival between sites occurred at different times of the year and under different environmental site conditions. Consequently, this observation is difficult to explain in simple terms; however, a closer look at the data shows that at Las Palomas, the site containing the 5.5-year-old seedlings, survival began to decrease at the end of the warm and dry season of 2016, then declined steadily during the cold and dry season, before presenting a pronounced decrease at the end of the warm and dry season (**Figure 4A**).

The 3.5-year-old seedlings at the Los Ailes site began to present mortality during the cold and dry season (**Figure 4B**). The common trend between both sites is that initial drops in survival began during dry periods when either warm or cold. The reduction in survival of the 3.5-year-old seedlings was less gradual and more dramatic than that of the 5.5-year-old seedlings. The more rapid decline in survival at the Los Ailes site is also likely due to the fact that the site contained transplanted seedlings with disturbed root systems. This is probably due to the different rooting depths of the 3.5- and 5.5-year-old seedlings. Although no data were measured, it can be presumed that 5.5-year-old seedlings would have rooted further and deeper than the 3.5-year-old seedlings, allowing these plants to better buffer the drought (Padilla and Pugnaire, 2007). **Figure 6** indicates temperatures that were frequently high enough to

allow reasonably high rates of photosynthesis and stomatal conductance, so water loss could have been sufficient to kill the seedlings during the dry periods. The lower temperatures under nurse plants (**Figure 6**) would also confer a lower vapor pressure deficit, which acts to decrease transpiration (Will et al., 2013).

The detailed measurements of the temperatures experienced by the A. religiosa seedlings at Los Ailes site, with and without nurse plant protection, indicate that the nurse plants prevented the formation of significant temperature gradients within the bottom 40 cm layer of the atmosphere, resulting in considerably reduced temperature fluctuations in this location during the diurnal cycle. This is due both to the interception of solar radiation during the day and to the emission of IR radiation by the nurse plant canopy during the night. The consequence of this for the seedling is a reduction of heat stress during the day along with a 2.5◦C temperature increase during the night, which probably acts to decrease the risk of frost.

Another factor that might contribute to the greater mortality of A. religiosa seedlings without nurse plant protection during the cold and dry (November-February) and the dry and warm (March to May) seasons (**Figure 4**; **Figure S4**), could be saturation of the photosynthetic apparatus, given that the plants experience drought in both seasons (Yin et al., 2006; Arena et al., 2008; Lambers et al., 2008). In the cold, dry season, subzero temperatures are common at night. Therefore, water is not available for plants for several hours each day. Likewise, insufficient water is available during the dry, warm season. In both cases, in the treatment with no nurse plants, photosynthesis might be water limited (as stated above), promoting oxidative stress. The detailed temperature measurements indicate that the shade provided by nurse plants prevents excess insolation during the cold and dry January-February period and also prevents excess insolation along with elevated temperatures during the late March to May period. In any case, in order to confirm any of the previous possible explanations, it would be necessary to conduct detailed ecophysiological measurements to pinpoint with certainty the mechanism that causes the mortality recorded.

site (3,360 m) with and without the shade of a nurse plant (a Baccharis conferta shrub). The green line represents air temperature (◦C) measured at 40 cm above the ground (as a proxy of the temperatures to which an Abies religiosa seedling would be exposed), while the red and blue lines represent the brightness temperature of the soil and sky, measured by two IR sensors pointing downwards and upwards, respectively.

The lack of differences among populations for seedling height increment suggests that upwards seed transfer of up to 440 m in elevation may be a feasible climate change adaptation strategy. In other words, A. religiosa seedlings seem to have sufficient phenotypic plasticity to survive and prosper under the nurse plant protection when exposed to colder temperatures at a higher elevation.

From our results, we strongly suggest that local communal forest nurseries in the MBBR begin producing local shrubs (such as B. conferta) for planting as nurse plants prior to planting A. religiosa seedlings in disturbed sites that lack naturally established shrubs. Nurse plants are needed because there are many seriously disturbed sites lacking adequate shrub cover inside the MBBR that would require identification of appropriate nurse plant species, and their field planting perhaps 2 years in advance of A. religiosa planting. This strategy, combined with use of seed sources from lower elevations, appears to be an effective approach to restoring disturbed sites for overwintering Monarch butterflies.

An alternative to produce shrubs in the nursery and then planting them as nurse plants, might be to use artificial shading over young recently planted A. religiosa seedlings. That would be placing a shade cloth, a practice that actually is used today to protect avocado young seedlings (the first year after planting a new orchard) in the region (at lower altitude). However, that has not been tested for A. religiosa.

Las Palomas site (B) and the Los Ailes site (D).

The relatively current common practice of removing existing bushes as "site preparation" previous to reforestation with A. religiosa seedlings, seems to be a counterproductive practice that needs to be eradicated. There is the idea among some foresters that shrubs compete with conifer seedlings. That might be true for some shade-intolerant pine species, but not for the shade-tolerant A. religiosa. Also, in our field observations, when A. religiosa matures and becomes a large tree, understory bushes are naturally excluded by light competition when the A. religiosa stand forms a dense closed (and dark) canopy.

#### LIMITATIONS OF THIS STUDY

The possibility that microsites with shrubs might have some sort of pre-condition that facilitates shrub establishment remains unexplored. This could also be beneficial for A. religiosa seedlings (independently of the shade provided by the nurse plants). Similarly, it might be possible that perturbed open areas might have a negative precondition that hampers the establishment of both shrubs and A. religiosa. It would therefore be desirable to conduct an experiment with two treatments: transplantation of shrubs in 1) open areas and 2) microsites where shrubs were TABLE 3 | ANOVA per treatment (planting under nurse plants and planting in the open field) at the Las Palomas site.


Significance (P) of the ANOVA for Abies religiosa seedlings at the Las Palomas site (5.5 years old; 3.5 years after field planting), seedling height increment and stress index, using a mixed model. Survival used a General Linear Model analysis using a binomial distribution. Stress Index data from March 2017.

TABLE 4 | ANOVA per treatment (planting under nurse plants and planting in the open field) at the Los Ailes test site.


Significance (P) of the ANOVA for Abies religiosa seedlings at the Los Ailes site (3.5 years old; 1.5 years after field planting), for seedling height increment and stress index, using a mixed model. Survival used a General Linear Model analysis using a binomial distribution. Stress Index data from March 2018.

naturally established but removed prior to the experiment. This would maximize the efficiency of the strategy proposed above regarding the planting of shrubs in open areas prior to any A. religiosa reforestation.

#### and improved the English writing. LL-T and YH-R provided helpful comments during the development of the project. All co-authors revised and contributed to the manuscript. CS-R led the writing.

# CONCLUSIONS

We conclude that A. religiosa seedlings require shade protection from nurse plants and that shifting populations upwards up to 440 m in elevation in order to compensate for future warmer climates does not appear to have any negative impacts on the seedlings and may confer adaptation to future climates. We recommend reforestation programs that include shrub species in their forest nursery production in order to provide nurse plants to accompany A. religiosa seedlings in those sites without shrubs. In the case of deforested sites with shrubs, it would not be necessary to remove these shrubs before planting the A. religiosa seedlings.

# AUTHOR CONTRIBUTIONS

CS-R, AB-G, and RL-C conceived the research project. AC-N, AC-V, EN-M, AB-G, and CS-R produced the seedlings. AC-N, AC-V, EN-M, EG-P, MG-R, AB-G, and CS-R planted the field experiments. AC-N, EN-M, AC-V, EG-P, CZ-S, AB-G, and CS-R took the measurements. FP-G, AC-N, and CS-R developed and standardized the stress index. RL-C, AC-N, and EN-M designed and conducted the plant cover measurements. AC-N, EN-M, CS-R, and AB-G conducted the statistical analysis. EG-P developed the maps and provided monthly precipitation data. FP-G, KJ, and PL contributed to the ecophysiology discussion. PL conceived, designed and constructed the devices for taking detailed temperature measurements on the field. GO'N provided climate estimates, helpful discussion about assisted migration,

# ACKNOWLEDGMENTS

This paper is an undertaking of the Forest Genetic Resources Working Group (FGRWG)/North American Forest Commission/Food and Agricultural Organization of the United Nations. Financial support was provided to CSR by The Monarch Butterfly Fund (Madison, Wisconsin, USA), the Mexican Fund for the Conservation of Nature and United States Forest Service USDA (funded by US Agency for International Development), the Forest Genetic Resources Working Group of the North American Forestry Commission (FAO), the Joint Fund (Fondo Mixto) CONACyT-Michoacán (FOMIX-2009-127128), and the UMSNH Coordinación de la Investigación Científica; the CONACYT Basic Research Fund (Ciencia Básica-2014- 242985) to AB-G; the PAPIIT program (IN-116218) of UNAM to RALC; and the Southern Research Station of Department of Agriculture US Forest Service to KJ. We thank Francisco Ramírez Cruz and Doña Petra Cruz-Cruz for the provision of numerous facilities for seed collection, seedling nursery production, field test planting, maintenance and measurements at Ejido La Mesa; without their help, this study would not have been possible. Thanks also go to Gabriel Muñoz Montoya (Queréndaro, Michoacán) and Miguel Angel Silva-Farías (Servi-Ambiental El Bosque) for the facilities and assistance provided for seed collection at Cerro de San Andrés; Eduardo Rendón, World Wildlife Mexico, and Jaime and Rogelio Díaz, Ejido Los Remedios, for providing maintenance to the seedlings at the forest nursery at Sierra Chincua, RBMM; Felipe Martínez-Meza

and Gerónimo Mondragón, Director and staff of the MBBR, respectively, provided the official permits and advice to establish the field tests at the core zone of the MBBR; Dante Castellanos-Acuña, Marisol Ortiz-Bibian, Gerardo Guzmán-Aguilar, Nancy Farías-Rivero, Miriam Linares-Rosas, Jorge Herrera-Franco, Rubí Contreras-Bailón, Francisco Loera-Padilla, Ana Beatriz Guerrero-Carmona, Nancy Izquierdo-Calderón and other UMSNH students for assistance in maintenance and measurements of the nursery and field experiments; Isabel Ramírez, CIGA-UNAM, Morelia, for providing data from weather stations within the MBBR; Eligio García-Serrano (Monarch Fund, Zitácuaro, Michoacán) and Xavier Madrigal-Sánchez (School of Biology, UMSNH) for providing helpful comments about the ecology and distribution of A. religiosa and

the overwintering sites of the Monarch butterfly; I. Ramírez and E. García-Serrano also provided photographic evidence of the storm damage. Special thanks go to G. E. Rehfeldt (USDA-Forest Service, Moscow, Idaho) for advice on the experimental design and data analysis. An editor and three reviewers provided valuable comments that helped to significantly improve the manuscript. Keith MacMillan provided assistance with the English writing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00421/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 Carbajal-Navarro, Navarro-Miranda, Blanco-García, Cruzado-Vargas, Gómez-Pineda, Zamora-Sánchez, Pineda-García, O'Neill, Gómez-Romero, Lindig-Cisneros, Johnsen, Lobit, Lopez-Toledo, Herrerías-Diego and Sáenz-Romero. 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.

# Adult Monarch (Danaus plexippus) Abundance Is Higher in Burned Sites Than in Grazed Sites

Julia B. Leone<sup>1</sup> \*, Diane L. Larson<sup>2</sup> , Jennifer L. Larson<sup>3</sup> , Nora Pennarola<sup>4</sup> and Karen Oberhauser <sup>5</sup>

*<sup>1</sup> Department of Fisheries, Wildlife and Conservation Biology, University of Minnesota, St. Paul, MN, United States, <sup>2</sup> U.S. Geological Survey, St. Paul, MN, United States, <sup>3</sup> Polistes Foundation, Inc., St. Paul, MN, United States, <sup>4</sup> Department of Entomology, University of Minnesota, St. Paul, MN, United States, <sup>5</sup> University of Wisconsin Arboretum, University of Wisconsin–Madison, Madison, WI, United States*

#### Edited by:

*Cheryl Schultz, Washington State University Vancouver, United States*

#### Reviewed by:

*Victoria Pocius, Pennsylvania State University (PSU), United States Erica Henry, North Carolina State University, United States*

> \*Correspondence: *Julia B. Leone leone050@umn.edu*

#### Specialty section:

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

Received: *29 December 2018* Accepted: *25 October 2019* Published: *14 November 2019*

#### Citation:

*Leone JB, Larson DL, Larson JL, Pennarola N and Oberhauser K (2019) Adult Monarch (Danaus plexippus) Abundance Is Higher in Burned Sites Than in Grazed Sites. Front. Ecol. Evol. 7:435. doi: 10.3389/fevo.2019.00435* Much of the remaining suitable habitat for monarchs (*Danaus plexippus)* in Minnesota is found in tallgrass prairies. We studied the association of adult monarch abundance with use of fire or grazing to manage prairies. Sites (*n* = 20) ranged in size from 1 to 145 hectares and included land owned and managed by the Minnesota DNR, U.S. Fish and Wildlife Service, The Nature Conservancy, and private landowners. We measured *Asclepias* spp. (milkweeds, monarch host plants) and forb frequency in 0.5 × 2-m plots located along randomly-placed transects that were stratified to sample wet, mesic, and dry prairie types at each site. Adult butterfly surveys took place three times at each site during the summers of 2016 and 2017, using a standardized Pollard Walk (400 m). Data were analyzed using mixed effects models. Monarchs were more abundant at sites managed with prescribed fire than with grazing. We found no difference in milkweed and forb frequency between burned and grazed prairies. There was no relationship between monarch abundance and the other predictor variables tested: milkweed frequency, site area, forb frequency, and percent prairie in a 1.5 km buffer area surrounding each site. Monarch abundance was lowest at grazed sites with high stocking rates. Our findings suggest that milkweed and forb frequency do not vary between burned and grazed sites, although we only considered land management practices for the 12 years before the study and the most recent burns occurred in 2014, 2 years prior to the start of our study. They also suggest that heavy grazing may have negative impacts on monarchs.

Keywords: monarch butterfly, Danaus plexippus, tallgrass prairie, prescribed fire, conservation grazing, milkweeds, Asclepias, prairie management

# INTRODUCTION

The current decline in eastern North American monarch (Danaus plexippus) numbers (Rendón-Salinas et al., 2018), the risk this decline poses for the long-term survival of the population (Semmens et al., 2016), the strong evidence that breeding habitat in the Upper Midwestern U.S. is a key factor driving the decline (Oberhauser et al., 2016; Pleasants et al., 2016; Pleasants, 2017; Thogmartin et al., 2017b), and the interest that people have in preserving monarchs (Diffendorfer et al., 2014), behooves us to understand best management practices for potential monarch breeding habitat. While we focus on management of remnant tallgrass prairie, where the ground has never been plowed, our findings are likely to be applicable to most Midwestern grasslands. Understanding how management affects monarch use of grasslands is important because they have the potential to contribute more to monarch conservation goals than any landcover category except if current cropland is restored to grassland (Thogmartin et al., 2017a).

The tallgrass prairie evolved with natural disturbances such as fire, grazing, and drought, without which woody plant encroachment is inevitable (Axelrod, 1985; Anderson, 2006). Today, many land managers use prescribed burns and grazing, often by cattle, to mimic these historical processes and maintain remnant prairie (Brudvig et al., 2007). Prairies are one of the most critically endangered habitats in North America and tallgrass prairie alone once covered over 100 million acres; <2% of these grasslands remain (Samson et al., 2004; Anderson, 2006). Minnesota, once home to 18 million acres of tallgrass prairie, has suffered comparable losses (Samson et al., 2004). Until recently, the agricultural lands that now dominate the fragmented landscape supported more milkweeds and monarchs than other habitat types (Oberhauser et al., 2001). With the advent and widespread use of transgenic herbicide-resistant crops and increased herbicide use, however, the Midwest has suffered a 40% loss in milkweed stems (Pleasants and Oberhauser, 2013; Pleasants et al., 2016). It is critical that we understand the capacity of our remaining grasslands to support monarchs so that land managers can take this information into consideration when setting conservation and management goals.

There has been considerable attention given to the effects of burning on arthropods, which may be killed directly by fire, or which could be affected positively or negatively by changes to the vegetation or soil. In many cases, prairie dependent, less mobile insect species, and species in less mobile life stages are more likely to exhibit negative post-fire responses in isolated grassland sites, where recolonization is presumably less likely (Panzer, 2002), unless refugia are maintained (Swengel and Swengel, 2007). In Minnesota, however, monarchs are generally not present in any life stage during the early- to mid-spring (Prysby and Oberhauser, 2004), when most prescribed fires occur (Emery and Gross, 2005; Towne and Craine, 2014), so direct impacts are likely to be minimal. Indirect impacts due to ways in which fire affects monarch host plants or nectar plants are more likely. The effects of fire on common milkweed (Asclepias syriaca) are unclear; Towne and Kemp (2008) reported that A. syriaca declined in frequency after summer fire, but a recent study in Kansas found that it increased after fire and decreased with grazing, while seven other species of milkweeds increased with grazing (Ricono et al., 2018). In the southern Great Plains, Asclepias viridis density and re-growth were significantly higher immediately after summer burns, leading to an increase in observed monarch eggs and larvae in burned areas (Baum and Sharber, 2012). Impacts of grazing could include direct consumption of immature monarchs or removal of their host plants or nectar sources. However, milkweed, especially the most toxic species, can harm vertebrate herbivores if they consume it (Holmgren, 1971; Panter et al., 2011). If more desirable forage is available, vertebrates usually do not consume milkweed (personal observations; Holmgren, 1971; Panter et al., 2011).

Conservation grazing, the use of grazing by domestic animals to achieve conservation goals, is seen as an attractive management alternative to reduce potential threats of fire to insect communities (Panzer, 2002). Prior to European settlement, bison were the dominant grazers in the tallgrass prairie. Today, however, conservation grazing is done almost exclusively with domesticated cattle, which preferentially graze different vegetation, prefer wetter areas, and move with different herd patterns than bison (Plumb and Dodd, 1993; Allred et al., 2011; Kohl et al., 2013). The impacts of grazing include the removal of plant material (thus making grazers potential competitors of herbivore arthropods) and effects on plant communities. Grazer impacts also include soil disturbance, nutrient concentration, and direct consumption of arthropods as they consume plant material. Because ungulates are grass specialists, they can increase plant diversity (Collins et al., 1998) and can also increase habitat heterogeneity (Knapp et al., 1999). Grazing can also alter plant quality and abundance (Joern, 2005; Moran, 2014) leading to enhancement of herbivorous arthropod abundance (Moran, 2014). Both herbivorous arthropods and large ungulates can consume considerable amounts of plant biomass, suggesting that interactions between them could be important (e.g., Pringle et al., 2007). However, most work on impacts of grazing has focused on extreme levels (e.g., comparing no grazing to high levels of grazing), and there has been less work on levels between these extremes (van Klink et al., 2015; Neilly et al., 2016).

While there have been several studies that examined the responses of butterflies, including monarchs, to fire, grazing, and patch-burn grazing (e.g., Vogel et al., 2007; Moranz et al., 2012), no individual studies focus on the impacts of these management strategies on both monarchs and their host plants. Vogel et al. (2007) found higher adult monarch abundance at grazed sites than at burned sites, as well as sites with higher floral resources, while Moranz et al. (2012) found adult monarchs in highest abundance at burned sites compared to sites that were burned-and-grazed or patch-burn-grazed. Vogel et al. (2010) found a positive correlation between time since burn and butterfly richness and abundance in Iowa tallgrass prairie; although they did not report specific results for monarch butterflies, they found that habitat generalists [which monarchs are characterized as by Vogel et al. (2010)] were more influenced by the direct effects of fire whereas habitat-specialist butterflies were more influenced by vegetation responses to fire. Migrating monarch and nectar resource abundance in the Ouachita Mountains of Arkansas increased following frequent prescribed fires after an extended period of fire suppression (Rudolph et al., 2006).

Most existing analyses of adult monarch numbers look at regional or population-level trends, even those that use data from individual surveys (e.g., Semmens et al., 2016; Thogmartin et al., 2017b). Few have looked at the importance of sitespecific characteristics (but see Saunders et al., 2018), including management. As part of a larger study of the impacts of vegetation management through fire and grazing on plants, butterflies, and bees, we collected data on adult monarch abundance at 20 remnant prairie sites in western Minnesota, 10 managed with burning, and 10 with grazing.

Our objective was to estimate the effect of prairie management by fire vs. grazing on adult monarchs in Minnesota tallgrass prairie remnants. We hypothesized that if prairie management by fire or grazing positively influenced the frequency of nectar plants (forbs) and monarch host plants (milkweeds), then that management type would positively affect monarch abundance. If prairie management by fire and grazing did not influence nectar and host plant frequency, we expected to find no difference in monarch response to management, although patterns in site area or prairie habitat in the surrounding area may still influence site occupancy.

#### MATERIALS AND METHODS

#### Site Selection

The land-use legacy of fire and grazing management can take years to become apparent and is known to affect prairie plants and butterfly communities (Debinski et al., 2011; Moranz et al., 2012). To account for this, we selected remnant prairies that had never been plowed and that had documented management histories of either grazing or prescribed fire from 2005 through the completion of our study in 2017. We chose 20 sites, all within the Prairie Parkland Province of Minnesota (**Figure 1**). Sites ranged in size from 1.13 to 144.7 hectares with a median of 10.6 hectares (see **Table S1** for additional details). We obtained management histories and permission to survey the sites from owners and land managers: private landowners (5 sites) and multiple agencies including The Nature Conservancy (TNC) (2 sites), US Fish and Wildlife Service (10 sites), and the MN Department of Natural Resources (MN DNR) (3 sites). To minimize variability, the grazed (n = 10) and burned (n = 10) sites in which we surveyed monarchs were selected to represent similar geographical distributions and size ranges. Sites managed with fire were burned during the spring, 1–3 times since 2005; none were burned during 2016 or 2017. All grazed sites were rotationally grazed by domesticated cattle and stocking rates ranged from 0.52 to 2.9 Animal Unit Months/acre (AUM), representing a range of stocking rates used for rotational grazing in the Midwest (McCollum et al., 1999; Derner et al., 2008); privately-owned sites were grazed every year from 2005 to 2017 and public and TNC sites were grazed 2–5 years since 2005. Presence of milkweeds was not considered during site selection because we were interested in adult monarch use of sites and the extent to which this may or may not correlate with larval host plants.

For each site, we determined the percent of prairie in the surrounding landscape by first creating a 1.5 km buffer around each site using ArcMap (v 10.5.1) and overlaying this buffer onto landscape data obtained from the US Department of Agriculture, MN DNR, and South Dakota State University. We then calculated the percentage of the land within the buffer that was classified as prairie.

#### Vegetation Sampling and Analysis

We surveyed vegetation twice at each site, once in 2016 and again in 2017 (May 31 through August 28 in 2016 and June 1 through August 24 in 2017). Vegetation and monarch surveys were not concurrent. Water retention properties of soils were used to stratify sampling within all potential prairie types (wet, mesic, dry) at each site. Within each prairie type, transects were delineated on maps prior to the field season and were parallel to the elevation gradient; vegetation was sampled in each prairie type in proportion to its area at a site. Sites contained 1 to 10 transects, depending on prairie type, distribution, and site shape, and transects ranged in length from 45 to 792 m. We used 0.5 x 2-m plots arrayed equidistantly on transects to estimate the frequency (proportion of plots occupied) of Asclepias and forb species (Elzinga et al., 1998). Asclepias species were included in the measure of forb frequency because milkweeds are both important host plants and nectar resources for monarch butterflies. Frequency was estimated whether or not plants were in bloom. Plots were distributed along transects in proportion to the transect length and were at least 10 m from the ends of transects to avoid potential edge effects. Plots were oriented perpendicular to the transect, and the number of plots per site was proportional to the size of the site, with a maximum of 30 plots and a minimum of 5 plots in any given site. The number of plots at each site was determined using the equation:

$$f = a^\*(1 - \exp(-b^\*x))$$

where a = 30 (the maximum number of plots per site), b = 0.163, and x = site area(ha). See **Table S1** for an index of vegetation plots at each site and **Figure S1** for an example of transect and plot distribution at a site.

We also conducted botanist-directed meandering walks through each prairie type within a site in both 2016 and 2017 to scout for additional species not seen along transect surveys, which is especially relevant for patchy species such as Asclepias. Effort, or time spent searching, was recorded and proportional to the size of the search area. Observations from the meandering walks are included in the summary statistics of milkweed presence at each site and **Table 2** but not in the forb and Asclepias frequency analyses.

We summed the frequency (n occupied plots/total plots) of all forb species combined, and Asclepias species only by year for each study site. Vegetation data were analyzed using analysis of variance models [proc mixed in SAS software, Version 9.4 SAS Institute Inc., 2015]. With management type as the predictor variable of interest, we built two models, one with forb frequency as the response variable, and one with Asclepias frequency as the response variable. Models were developed separately for 2016 and 2017 data because different people conducted the vegetation surveys in each year. Site nested within management type was the random effect in all models.

#### Butterfly Sampling and Analysis

Monarchs were surveyed three times each summer for two summers (June 15 through August 31 in 2016 and May 14 through August 18 in 2017), with each round of surveying beginning in the south and moving north. Sites sampled in the morning during one visit were sampled in the afternoon during the subsequent visit, and vice versa. Prior to surveying,

400 meters of transects were randomly selected from preestablished vegetation transects at each site, with transects selected to proportionally represent the prairie types (wet, mesic, dry) present at each site. If multiple transects were required due to the size and shape of the site, they were at least 20 meters apart to avoid counting redundancy. Monarchs were surveyed concurrently with bees and other butterfly species.

We used a modification of the standardized Pollard Walk for relative abundance (Pollard, 1977; Thomas, 2005; Swengel and

Leone et al. Monarch Abundance Higher in Burned Sites

TABLE 1 | Comparison of the frequency of *Asclepias* and forb species at burned vs. grazed sites in 2016 and 2017.


TABLE 2 | Summary of the total number of sites where each milkweed species was observed (transect surveys + meandering walk), the number of sites where each milkweed species was observed in transect surveys only and the management (B = burned, G = grazed) at sites where milkweed species were observed.


*Total site number* = *20. Only species that occurred in the transect surveys were included in the frequency analysis.*

Swengel, 2013; Smith and Cherry, 2014). During each site visit, the observer walked 400 m of transects at a steady pace of 10 m/min and recorded all individuals seen within 2.5 m on both sides, 5 m ahead, and 5 m above. Monarchs were sampled by sight identification, and sex was not recorded. Surveys were conducted, when possible, between 09:30 h and 18:30 h when temperatures were above 18◦C, sustained winds <17 km/h, and cloud cover was <50% with no precipitation (Shepherd and Debinski, 2005; Moranz et al., 2012). Butterfly surveys were all conducted by the same observer. Adult monarch abundances from three survey visits per site per year were summed separately for summer 2016 and summer 2017 to create an index of monarch abundance for each year, hereafter referred to simply as monarch abundance. One grazed site, G-1, was only surveyed in 2017. Sixty surveys were conducted at burned sites and 57 at grazed sites.

Monarch data were analyzed using Poisson distributed generalized linear mixed effects models (GLMMs). With number of monarchs as the response variable, we used a two-step modeling process. First, we chose five predictor variables of interest as fixed effects to examine based on study design and a priori knowledge: (1) management type, (2) milkweed frequency, (3) forb frequency, (4) site area, and (5) percent prairie in a 1.5 km buffer surrounding each site. Year was included as a sixth predictor. We built six univariate GLMMs to look at the individual effect of each of these predictor variables on monarch abundance. Second, we built a single global GLMM including all six predictor variables listed above and three interaction terms and then used backward elimination, eliminating least significant variables one at a time (alpha = 0.05); the final model was determined to be the one with only significant terms remaining and the lowest AIC value (1AIC > 2, Arnold, 2010). We included the interactions between management type and milkweed frequency, management type and forb frequency, and milkweed frequency and site area (an index of resource availability). Site was included as a random effect. Prior to review, year was analyzed as a nested random effect instead of a fixed effect; final model results did not differ. The variable for site area was log10 transformed to normalize overdispersed size data.

To examine the possibility of management-specific effects on monarchs, we built additional management-specific GLMMs. First, univariate responses to management-specific predictors were modeled for monarch abundance at burned sites (the number of years each site was burned between 2005 and 2017 and time since the last burn) and grazed sites (the number of years each site was grazed between 2005 and 2017, time since last grazing, and stocking rate). Next, we built two additional sets of GLMMs, one with monarch abundance at burned sites as the response variable and one with monarch abundance at grazed sites as the response variable. We replaced management type in these models with management-specific predictors: time since last burn in the burn-only model and stocking rate and the number of years a site had been grazed between 2005 and 2017 in the graze-only model. The number of years each site was burned (2005–2017) was not included in the burn-only models due to its collinearity with time since last burn and time since last grazing was not included in the graze-only models due to its collinearity with the number of years a site had been grazed. All other predictor variables and random effects remained the same.

All monarch analyses were conducted in R 3.4.1 (R Core Team, 2017), RStudio 1.0.153 (R Studio Team, 2016) using the glmmTMB function from the glmmTMB package (Magnusson et al., 2017). Multicollinearity was tested using Pearson's correlation in the ggscatter function from the ggpubr package in R (Kassambara, 2017).

#### RESULTS

#### Vegetation

The frequency of Asclepias did not differ between burned and grazed sites in 2016 or 2017, nor did the frequency of all forbs combined (**Table 1**). Seven species of milkweeds were observed in at least one site during the course of this study: Asclepias incarnata, A. ovalifolia, A. speciosa, A. syriaca, A. tuberosa, A. verticillata, and A. viridiflora. Five of these were observed in transect plots and included in frequency analyses: A. incarnata, A. speciosa, A. syriaca, A. verticillata, and A. viridiflora (**Table 2**).

For transect survey data used in frequency analyses, we only observed one species of Asclepias in eight sites, another eight sites had two species observed, and two sites had three species

observed. At two sites (G-3 and G-5) no Asclepias species were observed in our transect plots during either year of the study, although three species were observed in meandering walks at both sites. At three sites (G-2, B-3, and B-9), Asclepias species were observed in transect surveys during only 1 year of the study, although Asclepias species were observed at G-2 and B-3 in 2016 and 2017 during meandering walks. All milkweed species were observed at both burned and grazed sites with the exception of A. ovalifolia and A. tuberosa, which were seen at only one grazed site each during meandering walks.

#### Monarchs

Adult monarchs were observed at all 20 study sites (**Figure 2A**). One hundred ninety-eight adult monarchs were observed during Pollard transect walks in 2016 and 2017 (99 in 2016 and 99 in 2017). One hundred forty-eight monarchs were observed at sites managed with fire and 50 monarchs were observed at sites managed with grazing. The majority of monarch observations were made during the third round of monarch surveys in late July and August in both years (**Figures 2B,C**).

The number of adult monarchs found at sites that had been burned was approximately three times as high as the number found at grazed sites (z = −2.076, p = 0.0379) (**Figure 2**). **Table 3** shows the univariate responses of predictor variables to monarch abundance and **Figure 3** the correlations between monarch abundance and milkweed frequency, forb frequency, percent prairie, and site area in 2016 and 2017. In univariate models, only management type was a significant predictor of monarch abundance (**Table 3**); monarch abundance was not correlated with milkweed frequency (**Figure 3A**), nor with the other predictor variables tested.

In the global model, none of the interaction terms tested were significant. There was also no relationship between monarch abundance and forb frequency, a surrogate for nectar resource availability, site area, or the percent of prairie habitat in a surrounding 1.5 km buffer around sites, The final model after



\**Denotes statistically significant p-values* < *0.05.*

backward elimination of all non-significant predictors included management type as the only predictor variable and was the same as the univariate model 1 in **Table 3** (est = −0.862, SE = 0.415, z = −2.076, p = 0.0379, AIC = 201.4). The next best-fit model included milkweed frequency as a second, although non-significant, predictor variable (est = 1.036, SE = 0.817, z = 1.269, p = 0.205, AIC = 201.8). Additional predictors did not improve model fit and were removed from the model in the following order: (1) the interaction between forb frequency and management type, (2) the interaction between site area and milkweed frequency, (3) site area, (4) the interaction between management type and milkweed frequency, (5) year, (6) forb frequency, (7) percent prairie, and (8) milkweed frequency. Global model results can be found in **Table S2**. Adult monarchs and milkweeds were observed at all sites during at least one survey year, indicating that sites are potential

breeding habitat or possess some other resource sought after by adult monarchs.

In management specific models, there was no relationship between monarch abundance at grazed sites and the number of years a site had been grazed (**Figure 4A**) or monarch abundance at burned sites and time since last fire (**Figure 4D**). The best-fit model for grazed sites after backwards elimination of all nonsignificant predictors included only stocking rate as a predictor TABLE 4 | Univariate models showing the effects of fire- (B0 – B2) and grazing- (G0 – G3) specific predictor variables on monarch abundance at burned or grazed sites, respectively.


\**Denotes statistically significant p-values* < *0.05.*

variable; monarch abundance was higher at grazed sites with lower average stocking rates (est = −0.923, SE = 0.417, z = −2.212, p = 0.027, AIC = 80.7) (**Figure 4E**). No predictors explained variation in monarch abundance at burned sites; the null model was the best fit (est = 1.625, SE = 0.335, z = 4.852, p = 1.22e-06, AIC = 118.7) Although not included in analyses due to collinearity, time since grazing and the number of years each site was burned are shown in **Figures 4B,C**, respectively. Univariate responses of management-specific predictors are presented in **Table 4**. Full model results for burn-only and graze-only models can be found in **Table S2**.

Data are archived and available (Leone et al., 2019).

#### DISCUSSION

Adult monarchs in the Minnesota tallgrass prairie remnants that we surveyed were significantly more abundant in sites that had been managed with prescribed fire than those managed with grazing. Adult monarchs, milkweeds, or both monarchs and milkweeds were observed at all sites during at least one survey year, indicating that sites are potential breeding habitat or possess some other resource sought after by adult monarchs. Monarch abundance was independent of milkweed and forb frequency, as well as other site variables and the amount of grassland habitat within 1.5 km of the site. None of the variables that we measured allowed us to pinpoint a mechanism for the association between burning or grazing and monarch abundance. Because this association was across all sites on which burning is used as a management tool, and no sites were burned in 2015, 2016, or 2017, we do not think that it is due to qualitative differences in host or nectar plants due to immediate effects of burning.

We had hypothesized that if prairie management by fire or grazing influenced the frequency of forbs and milkweeds, then these management types would affect monarch abundance. We also hypothesized that if prairie management by fire and grazing did not influence nectar and host plant frequency, then we would find no difference in monarch response to management, unless that response was driven by patterns in site area or prairie habitat in the surrounding area. It is surprising, therefore, that we found no difference between forb and milkweed frequency at burned vs. grazed sites and yet observed significantly higher TABLE 5 | Recommendations for further research into the effects of fire and grazing on monarch butterflies.


abundance of monarchs at burned sites. This correlation between monarch abundance and management type is apparently not driven by the indirect effect of management on the vegetation, nor is it the result of an interaction between management type and host plant frequency or forb frequency; host plant resource availability, measured as the interaction between milkweed frequency and site area, also was not predictive. The lack of patterns in the monarch response to the vegetation may be due in part to the scale of sampling (frequency). A study with more exhaustive milkweed sampling methods (for example, density or abundance), including a study of the chemical response of milkweed to fire over multiple years, may help explain this finding. Similarly, nectar plant frequency ignores important variation in flower density and floral traits.

The fact that we found no pattern in monarch response to time since burn or frequency of fire management does not necessarily mean that no such pattern exists. Given the limitations in a sample size of 10 burned sites that were not explicitly selected to encompass a range of additional fire variables, more studies with a larger sample size and a range of times since fire and years of management might provide additional insight. Several studies have found significant postburn butterfly and host plant responses to fire within 1– 2 years of fire (e.g., Fleishman, 2000; Rudolph et al., 2006; Baum and Sharber, 2012); however, few studies examine longterm trends in butterfly or monarch abundance after fire. Vogel et al. (2010) found that floral resource availability was negatively correlated with time since burn and suggests that post-fire recovery may exceed 5 years for some butterfly species. Extrapolation from studies of other insect responses to fire should be approached with caution, since multi-taxa studies have found that different insect taxa respond differently to time since fire (New et al., 2010; Pryke and Samways, 2012a; Yekwayo et al., 2018). For example (Pryke and Samways, 2012b), found that aerial assemblages, including lepidoptera, showed little difference in species composition immediately after fire but significant differences 3 years post-fire, contrary to other insect taxa. There is still much to learn about how time since fire may impact butterflies and monarchs in particular, and more research that encompasses 1–5+ year post-burn responses would help contextualize our findings and guide best management practices for monarchs.

Our observation that there were fewer monarchs at sites with higher stocking rates of cattle suggests that grazing, or at least heavy grazing, may have a negative impact on monarchs. It is possible that monarch eggs, larvae, and pupae may suffer incidental mortality from cattle grazing and trampling. It is also possible that there are fewer nectar resources available at grazed sites. While the frequency of forb plants did not differ between burned and grazed sites, we did observe cattle consuming flowers during our surveys, and also made the anecdotal observation that sites with cattle present tended to have fewer plants in flower. In addition, frequent fire has been shown to increase nectar resources and migrating monarch abundance compared to unburned controls (Rudolph et al., 2006). We did not quantify this impact, which is a limitation of our study. A study that directly measures nectar resources, not just forb frequency, would help elucidate the potential impacts of cattle on monarchs and their nectar plants.

Heavier grazing rates by cattle can also reduce the height of the vegetation, which may be detrimental to monarchs by limiting host plant and nectar plant biomass. Multiple studies have documented positive correlations between butterflies and taller vegetation in grasslands (Poyry et al., 2006; Berg et al., 2013). We made anecdotal observations that the vegetation at many grazed sites was much shorter than at burned sites, however, we are unable to quantify its effect on monarch abundance. This is an area for further study.

Our study found no relationship between the size of our sites and monarch abundance. Other studies examining the relationship between site area and butterfly abundance and diversity have observed similar patterns (e.g., Krämer et al., 2012), although in a study in the United Kingdom, butterfly abundance and diversity increased with an increase in grassland habitat area, with additional effects of surrounding habitat diversity and the larger landscape context (Botham et al., 2015). It is possible that the range of site areas in our study (many small and a few large; **Figure 3C**) was insufficient to capture a pattern in monarch response to site area. Landscape and local variables, including site area, have been found to affect the butterfly community in the fragmented tallgrass prairie (Davis et al., 2007). We know that monarchs can travel fairly long distances, up to 15 km/day (Zalucki et al., 2016). Their mobility is another possible explanation for why larger sites do not correlate with higher abundance of monarchs. Grant et al. (2018), Zalucki and Lammers (2010), and Zalucki et al. (2016) all suggest that monarchs will be more likely to encounter habitat patches when small patches are dispersed at distances within the monarchs' perceptual range than when a few large sites are more dispersed, but empirical data to parameterize and test these models are lacking. Additionally, we have little understanding of the characteristics of sites that are used by monarchs and there are no published studies that selected monitoring sites in a way that would allow comparisons of the relative importance of site and landscape characteristics. We attempted to account for the potential patchiness of host plant distribution within sites by including the interaction between milkweed frequency and site area as a surrogate for host plant resource availability in our analyses and found no effect.

Grazed and burned sites were chosen to represent management history in this part of the monarch breeding range and to be roughly equivalent in size; due to constraints in the number of grazed sites that met these criteria, more of our grazed sites are closer to the eastern edge of the Prairie Parkland Province (**Figure 1**). We do not believe that this has biased our results, as many of these sites (G-10, G-8, G-1) have some of the highest monarch counts compared to grazed sites in closer proximity to other burned sites along the western edge of the Prairie Parkland Province. Our study was set up to test management effects on insects and plants in remnant prairies, so our site selection process did not include consideration of the surrounding habitat. Our finding that monarch abundance was not affected by the amount of other grassland habitat in the surrounding 1.5 km buffer could mean that the surrounding habitat is not important, or that the spatial scale of our buffer was too small to account for monarch perceptual range. More research is needed to detect the mechanism(s) driving these observations.

We chose a retrospective study design using sites with known management history instead of implementing our own experimental design because management can take many years to become apparent on the landscape and our study was constrained to 2 years. While there are benefits to this design, there are also disadvantages and potential for bias; for example, management has not been applied randomly to the landscape and there may have been initial bias in the decisions of managers to burn or graze certain prairies. Because of the inherent challenges in a retrospective study design, given sufficient time for implementation, we recommend experimental fire and grazing management studies to eliminate confounding variables. Long-term experimental research of this kind would provide valuable insight into understanding the mechanisms underlying the patterns we observed. Another challenge in this system is that there are no undisturbed or unmanaged prairies to serve as controls to compare to fire and grazing management; this is inherent in a disturbancedependent ecosystem that evolved with fire, grazing, and drought (Anderson, 2006). Unmanaged prairie succeeds quickly to shrubland and experimental "controls" no longer represent the ecosystem they are designed to study.

Remnant tallgrass prairies and other grasslands in the Midwestern U.S.A. provide important habitat for eastern monarchs during the breeding season (Oberhauser et al., 2016). In the fragmented landscape of our present day, natural disturbances such as unchecked wildfire and roaming herds of bison are no more, and these grasslands would succeed to shrub and woodland without management (Gibson and Hulbert, 1987; McClain and Elzinga, 1994; Anderson, 2006). Although grazing has been suggested as a tool for minimizing potential negative effects of fire on invertebrates, our study suggests that we should treat grazing, or at least grazing at high stocking rates, with caution as a management tool for monarchs, until we can understand a mechanism for lower numbers of monarchs at grazed sites. It is clear that additional research is needed to fully understand the effects of fire and grazing management on monarch butterflies. To this end, we present our recommendations for further research in **Table 5**. Our research is an important step toward understanding the effects of fire and grazing management practices on monarch butterflies and informing additional studies that will guide future conservation and management decisions for monarch butterflies.

# DATA AVAILABILITY STATEMENT

Data and metadata are available on ScienceBase.gov through a U.S. Geological Survey data release, https://doi.org/10.5066/ P940ICLS.

# AUTHOR CONTRIBUTIONS

DL, KO, and JLe conceived the ideas for the study. All authors designed the methodology. JLe collected the monarch data. JLa supervised vegetation data collection and analyzed GIS and landscape data. JLe and DL analyzed monarch and vegetation data, respectively. JLe led manuscript writing and revisions. KO led writing of the introduction and discussion. All authors contributed critically to the drafts and gave final approval for publication.

#### Leone et al. Monarch Abundance Higher in Burned Sites

# FUNDING

Funding for this project came from the Minnesota Environment and Natural Resources Trust Fund (ENRTF), M.L. 2015, Chp. 76, Sec. 2, Subd. 03o; Prairie Biotic Research, Inc.; National Science Foundation Graduate Research Fellowship (GRFP).

### ACKNOWLEDGMENTS

Thank you to scientists and land managers at the U.S. Fish and Wildlife Service, The Minnesota Department of Natural Resources, The Nature Conservancy, and private land owners for sharing resources, land, permits, site access, and management history. Thank you to the botanists and field technicians who helped collect data. We are grateful to Brian Aukema and Deb Buhl for providing valuable advice and feedback on statistical analyses, Erin Treiber for reviewing an earlier version of the manuscript, and Alex Leone for providing database and programming expertise and support. This paper has been improved through conversation with Myron Zalucki. Any use of trade, firm or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

# SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2019.00435/full#supplementary-material

Figure S1 | Example site map with transects and vegetation plots distributed proportionally in each prairie type. Mesic prairie is shown in white (5.9 ha) and wet prairie is shown in gray (1.53 ha). Total site area is 7.43 ha. Transects run perpendicular to elevation gradients and vegetation plots (not to scale) are oriented perpendicular to the transects they are on. The three transects in mesic prairie are M1 (103 m; 6 plots), M2 (236 m; 7 plots) and M3 (170 m; 6 plots). Wet transects are W1 (53 m; 2 plots) and W2 (64 m; 2 plots). Monarchs were surveyed on 400 meters of randomly selected vegetation transects in proportion to wet and mesic prairie types: all of transect M2 and 84 m of M3 were surveyed to equal 320 m in mesic prairie (∼80%) and all of W1 and 27 m of W2 were surveyed to equal 80 m in wet prairie (∼20%).

Table S1 | Index of study sites including forb, *Asclepias,* and monarch survey data.

Table S2 | Full model results for monarch abundance at burned and grazed sites, burned-only sites, and grazed-only sites.

# REFERENCES


Samson, F. B., Knopf, F. L., and Ostlie, W. R. (2004). Great Plains ecosystems: past, present, and future. Wildl. Soc. Bull. 32, 6–15. doi: 10.2193/0091- 7648(2004)32[6:GPEPPA]2.0.CO;2


**Conflict of Interest:** JLa is a research affiliate of the Polistes Foundation, Inc., a 501-c-3 non-profit organization.

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.

At least a portion of this work is authored by Diane L. Larson on behalf of the U.S. Government and, as regards Diane L. Larson and the U.S. Government, is not subject to copyright protection in the United States. Foreign and other copyrights may apply. 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.

SAS Institute Inc. (2015). SAS Software.

# Configuration and Location of Small Urban Gardens Affect Colonization by Monarch Butterflies

Adam M. Baker and Daniel A. Potter\*

Department of Entomology, University of Kentucky, Lexington, KY, United States

Ecological theory predicts that specialist insect herbivores are more likely to locate and colonize host plants growing in relatively sparse or pure stands compared to host plants growing amongst diverse non-host vegetation. We tested the hypothesis that increasing the apparency and accessibility of milkweed (Asclepias spp.) host plants in small polyculture gardens would boost their colonization by the monarch butterfly (Danaus plexippus), an iconic native species of conservation concern. We established replicated gardens containing the identical mix of milkweeds, flowering nectar sources, and non-host ornamental grasses but arranged in three different spatial configurations that were monitored for monarch colonization over two successive growing seasons. Monarch eggs and larvae were 2.5–4 times more abundant in gardens having milkweeds evenly spaced in a 1 m corridor around the perimeter, surrounding the nectar plants and grasses, than in gardens in which milkweeds were surrounded by or intermixed with the other plants. Predator populations were similar in all garden designs. In a corollary open-field experiment, female monarchs laid significantly more eggs on milkweed plants that were fully accessible than on milkweeds surrounded by non-host grasses of equal height. In addition, we monitored monarch usage of 22 citizen-planted gardens containing milkweed and nectar plants in relation to their botanical composition, layout, and surrounding hardscape. Multivariate analysis explained 71% of the variation, with significantly more eggs and larvae found in gardens having milkweeds spatially isolated as opposed to closely intermixed with non-host plants, and in gardens having 100 m north/south access unimpeded by structures. Numerous programs encourage citizens to establish gardens with milkweed and nectar plants to help offset habitat loss across the monarch's breeding range. Our findings suggest guidelines for garden design that can help make the urban sector's contributions to monarch habitat restoration more rewarding for participants, and of greater potential value to monarch recovery.

Keywords: Danaus plexippus, reconciliation ecology, conservation biology, citizen science, Asclepias, garden, urban

#### INTRODUCTION

Reconciliation ecology, "the science of inventing, establishing, and maintaining new habitats to conserve species diversity in places where people live, work, and play" (Rosenzweig, 2003a) aims to modify human-dominated landscapes to support native biota without compromising societal utilization (Rosenzweig, 2003a,b; Francis and Lorimer, 2011). As natural habitats increasingly are

Edited by:

J. Guy Castley, Griffith University, Australia

#### Reviewed by:

Martha Weiss, Georgetown University, United States Douglas Landis, Michigan State University, United States

> \*Correspondence: Daniel A. Potter dapotter@uky.edu

#### Specialty section:

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

Received: 06 September 2019 Accepted: 22 November 2019 Published: 05 December 2019

#### Citation:

Baker AM and Potter DA (2019) Configuration and Location of Small Urban Gardens Affect Colonization by Monarch Butterflies. Front. Ecol. Evol. 7:474. doi: 10.3389/fevo.2019.00474

**355**

cleared, fragmented and degraded by anthropogenic activities, properly designed urban green spaces, including pollinator gardens, can be refuges for native biodiversity, particularly of invertebrates, birds, and other animals able to adapt to human proximity (Goddard et al., 2010; Baldock et al., 2015; Hall, 2016; Aronson et al., 2017). Reconciliation ecology also provides opportunities for urban citizens to connect with nature, helping to foster a wider interest in conservation issues (Goddard et al., 2010; Lepczyk et al., 2017). Among insects of conservation concern, none exceeds the power of the monarch butterfly (Danaus plexippus L.) to inspire public engagement in reconciliation ecology (Gustafsson et al., 2015).

Instantly recognizable by gardeners and nature lovers, the iconic monarch is renowned for its annual migration in which butterflies from discrete overwintering areas in the highlands of central Mexico recolonize breeding grounds across the United States and southern Canada east of the Rocky Mountains over several generations, followed by a single autumn migration back to Mexico (Reppert and de Roode, 2018). The eastern migratory monarch population has declined >80% in the past 25 years (Brower et al., 2011; Vidal and Rendón-Salinas, 2014), fueling concern that it may face extirpation unless habitat conservation and restoration efforts are enacted on a continental scale. The monarch population in western North America is also in sharp decline (Schultz et al., 2017). The US Fish and Wildlife Service (USFWS) is currently assessing the monarch's status in response to a petition to list the species under the Endangered Species Act, while working with a broad range of partners as part of an international initiative to conserve the butterfly across its range1, 2 .

Given that monarch larvae feed exclusively on milkweed (family Apocynaceae, subfamily Asclepiadoideae), and that adults migrate to locate host plants across diverse landscapes, two primary concerns facing monarch populations are shortages of milkweed, and floral nectar to fuel migration (Pleasants and Oberhauser, 2013; Oberhauser et al., 2017; Malcolm, 2018; Saunders et al., 2019). Conserving and restoring monarch habitat, especially planting of milkweeds and nectar resources on public and private lands, has emerged as the central conservation strategy to meet monarch population goals set by the USFWS and adopted by Mexico, Canada, and the United States1,2. Most research on monarch habitat restoration to date has focused on "non-use" land, e.g., publicly owned grasslands, road right-ofways, Conservation Reserve Program land, edges of fields and pastures, and other marginal habitat (e.g., Kasten et al., 2016; Oberhauser et al., 2017; Pitman et al., 2018). However, restoring enough milkweed to ensure a stable monarch population will require an "all hands on deck" strategy involving participation from all land use sectors including urban and suburban areas (Thogmartin et al., 2017; Johnston et al., 2019). In cities and towns, initiatives such as the Million Pollinator Garden Challenge<sup>3</sup> , the Monarch Waystation Program<sup>4</sup> , National Wildlife Federation's Butterfly Heroes program<sup>5</sup> , and Mayor's Monarch Pledge<sup>6</sup> are underway, with myriad gardens being planted in backyards, schoolyards, parks, and other public and private places. As of 2019, >25,000 Monarch Waystation habitats (managed gardens containing milkweeds and nectar plants) had been registered with MonarchWatch<sup>4</sup> and the National Pollinator Garden Network<sup>3</sup> had surpassed its goal of registering >1,000,000 pollinator gardens, many likely containing milkweed.

Guidelines for setting up a certified Monarch Waystation<sup>4</sup> recommend that such gardens should have "at least 10 milkweed plants, made up of two or more species," "should contain several annual, biennial, or perennial plants that provide nectar for butterflies," and that "the plants should be relatively close together" because "all monarch life stages need shelter from predators and the elements." Monarchs find and colonize milkweed in urban gardens (Cutting and Tallamy, 2015; Baker and Potter, 2018; Geest et al., 2019), but little is known about how to configure such gardens to maximize their conservation value.

Ecological theory (e.g., Root, 1973; Andow, 1991) suggests ways to increase monarch use of milkweed gardens. Susceptibility of plants to attack by insect herbivores may be strongly influenced by the structural and taxonomic complexity of surrounding vegetation (Tahvanainen and Root, 1972; Root, 1973; Rausher, 1981). Dietary specialists, in particular, tend to have difficulty locating host plants growing amongst non-host vegetation, and are less likely to remain on hosts grown in polyculture (Root, 1973; Finch and Collier, 2000). Mechanisms proposed for such "associational resistance" (Tahvanainen and Root, 1972) include visual or olfactory masking, repellent odors, physical obstruction or shading, or inappropriate landings on non-hosts triggering herbivores' premature dispersal (Tahvanainen and Root, 1972; Root, 1973; Risch, 1981; Finch and Collier, 2000). Neighboring plants may also provide harborage and food resources for natural enemies (Root, 1973; Risch, 1981). The aim in polyculture agriculture is to discourage host-finding and colonization by specialist herbivores. The goal for monarch conservation gardens is just the opposite.

We hypothesized that the spatial configuration of host and non-host plants within small gardens, particularly the milkweeds' visual apparency and butterflies' access to them, as well as location of gardens relative to surrounding hardscape, would strongly affect their colonization and use by monarchs. Here, we tested those hypotheses by monitoring (1) monarch use of 22 preexisting citizen-planted Monarch Waystations in relation to those gardens' botanical composition, configuration, and surrounding hardscape, (2) colonization of experimental gardens containing an identical mix of milkweeds, nectar sources, and non-host grasses, but planted in different spatial layouts, and (3) oviposition on isolated milkweeds and milkweeds that were visually obstructed by non-host vegetation.

<sup>1</sup>https://www.fws.gov/savethemonarch/

<sup>2</sup>https://monarchjointventure.org/images/uploads/documents/

<sup>5431</sup>\_Monarch\_en.pdf

<sup>3</sup>http://millionpollinatorgardens.org/

<sup>4</sup>https://www.monarchwatch.org/

<sup>5</sup>https://www.nwf.org/Butterfly-Heroes.aspx

<sup>6</sup>https://www.nwf.org/mayorsmonarchpledge

′

#### MATERIALS AND METHODS

#### Monarch Use of Preexisting Waystations

Twenty-two preexisting registered Monarch Waystation gardens were identified via the Monarch Waystation Registry<sup>4</sup> or through the Wild Ones<sup>7</sup> Lexington, Kentucky Chapter, and monitored with permission from landowners or other authorized persons. The Waystations were in residential, commercial, and institutional landscapes, road medians, parks, and nature preserves encompassing a range of anthropogenic settings in and near the cities of Lexington, Richmond, and Berea, in central Kentucky. All of the gardens were mulched, and contained at least three Asclepias species, swamp (A. incarnata), common (A. syriaca), and butterfly (A. tuberosa) milkweeds, as well as a variety of annual and perennial flowering plants. Each Waystation was visited twice per month from 5 July to 20 September 2016. Each time, we inspected all milkweeds for monarch eggs and larvae, which were counted and left in place. Monarch eggs and larvae were observed in 20 of the 22 Monarch Waystations.

The Waystations were further characterized by features of the gardens and their surrounding landscape. Garden configuration was classified into two types: "structured" or "non-structured." In structured gardens (N = 9), the milkweeds had been planted in a relatively uniform array, set off by mulch, and separated from neighboring plants by 0.5 m or more. Non-structured gardens (N = 13) were also mulched, but had the milkweeds haphazardly intermixed with nectar and non-host plants in no particular arrangement, their foliage often touching or partially shaded by nearby plants. Other garden variables included total area, number of ramets of each milkweed species (counted during bloom when the plants were done producing new ramets for the year), and number of nectar plants.

We used satellite images and the Measure Tool feature of Google Earth Pro geospatial software (Microsoft, Palo Alto CA) to quantify the area of buildings and other hardscape within a 100 m radius centered each garden, the ratio of impervious to pervious surfaces, and distance of the garden to nearby structures. Linear transects were drawn from the garden through corners of all buildings to the edge of the circle. We summed the angles defined by those transects, divided by 360◦ , and subtracted from 1 to calculate a "360◦ accessibility index"; i.e., the proportion of access not blocked by buildings if an incoming butterfly approached the garden from 100 m away. Because monarchs fly predominantly northward during their spring migration and south toward their overwintering grounds during fall migration, we hypothesized that unimpeded lines of sight from those directions to resources may be important. Therefore, we determined straight line north/south access by scoring whether or not flight of a butterfly approaching the garden from due north or due south would be blocked by structures.

# Monarch Use of Experimental Gardens of Differing Configurations

Fifteen gardens (5.5 × 5.5 m) were established in spring 2017 in open, non-shaded grassland at the University of Kentucky ′

We used swamp milkweed, A. incarnata, because it grows to a consistent height of about 1 m and does not spread via rhizomes (Baker and Potter, 2018). Two-year old potted plants (30 cm tall) were transplanted (12 per garden) in early May 2017. To increase the structural and taxonomic complexity of the vegetation surrounding the milkweeds, each garden also contained flowering annuals differing in height and form, including Mexican sunflower, Tithonia rotundifolia (12 per garden) and common zinnia, Zinnia elegans "Canary Bird" (12 per garden), which are attractive nectar sources for adult monarchs, and ornamental feather reed grass, Calamagrostis × acutiflora (four per garden). Mexican sunflower grows to 1.2– 1.5 m height and 0.6–0.9 m spread; Z. elegans to 0.6–0.9 m height and 0.2–0.3 m spread, and Calamagrostis reaches 0.9–1.5 m height and 0.45–0.76 m spread<sup>8</sup> . Nectar plants were greenhousegrown from seeds (Applewood Seed, Arvada, CO), whereas the ornamental grasses were purchased in 11.5 liter pots (Baeten's Nursery, Union, KY).

For gardens with perimeter milkweeds, the 12 A. incarnata were planted with even spacing in the 1 m border, 1.5 m apart, and the Tithonia, Zinnia, and Calamagrostis were evenly spaced within the inner block with one grass transplanted at each of the four cardinal directions (**Figure 1A**). For gardens with interior milkweeds (**Figure 1B**), the 12 A. incarnata were spaced 1.1 m apart in the inner block, with the Tithonia and nectar plants alternated evenly around the perimeter in the 1 m border, and for mixed gardens (**Figure 1C**), all plants were assigned to random distribution over the whole plot. Each garden received a 5 cm deep layer of dark-brown mixed hardwood mulch over the entire plot and surrounding all plants. The gardens were watered to maintain plant vigor for a month after planting, but received only natural rainfall for the duration of the study. They were handweeded, and re-mulched at the start of the second (2018) growing season, at which time a few of the less-vigorous milkweeds were replaced with similar-sized healthy 2-year-old plants. The

Spindletop Research Farm (38◦ 07 35.9 ′′N 84◦ 29 58.1 ′′W) in north Lexington, Kentucky. To establish the gardens, plots were sprayed with glyphosate (Roundup ProMax, Monsanto, St. Louis, MO) in April to kill existing vegetation, tilled, and covered with weed barrier cloth. Each garden contained the same mix of swamp milkweed, nectar plants, and ornamental grasses in one of three different spatial configurations, representing treatments: (1) milkweeds evenly spaced in a 1 m wide corridor around the perimeter with nectar plants and grasses in the interior (**Figure 1A**); (2) nectar plants and grasses in a 1 m corridor around the perimeter with milkweed in the interior (**Figure 1B**); or (3) random arrangement of all plants without formal garden structure (**Figure 1C**), hereafter referred to as gardens with "perimeter milkweeds," "interior milkweeds," and "mixed," respectively. Gardens were placed on 300 m transects (100 m spacing between treatments) oriented on an east-west axis within each replicate to minimize bias in their likelihood of being encountered during flight of north or south bound monarch butterflies. Each of the five replicates was separated by at least 300 m.

<sup>7</sup>https://lexington.wildones.org/

<sup>8</sup>http://www.missouribotanicalgarden.org/plantfinder/plantfindersearch.aspx

respectively, as they appeared in 2018.

grass (mostly tall fescue, Festuca arundinacea) surrounding each garden was mowed weekly to 10 cm height.

#### Assessing Monarch Colonization and Use of Gardens

Gardens were inspected for all monarch life stages during the 1st and 3rd week of each month from June to September 2017, and during the 2nd and 4th week of each month beginning 9 April until 23 July 2018, when a severe storm uprooted the taller, mostly Tithonia nectar plants, reducing integrity of the treatments. At each visit we carefully inspected above-ground portions of each milkweed by examining the stems, and the top and bottom of each leaf for monarch eggs, larvae, and pupae which were counted and left in place.

#### Natural Enemy Abundance in Gardens

Two methods were used to assess if garden design influenced abundance of generalist invertebrate predators in the gardens. First, all above-ground portions of the 12 milkweeds in each garden were inspected every 2 weeks from June to September 2017, and April to July 2018 on alternate weeks from when monarch life stages were counted. We recorded numbers of adults and immatures belonging to predominantly predatory taxa on each plant, spot-identifying to order and family and leaving them in place. Predatory wasps seen nectaring on the milkweed umbels were not counted.

Abundance of ground-dwelling predators that monarch larvae might encounter while moving between plants or to pupation sites was assessed using pitfall traps deployed for 48 h from July 19–21 and July 26–28, 2018, during peak monarch activity. Traps consisted of 0.47 liter plastic cups, with 2 cm of ethylene glycol as a killing agent, set into the ground with the brim 2 cm below the surface. There were four traps per garden spaced at least 2 m apart, but within 1 m of the milkweed. Trapped invertebrates were stored in 70% ethanol, and sorted and identified to order and family.

#### Effect of Surrounding Vegetation on Susceptibility of Milkweeds to Oviposition

A supplemental experiment investigated how presence or absence of surrounding non-host vegetation affects a milkweed plant's susceptibility to monarch oviposition. The trial ran from 6 to 21 August 2018 in an open grassy area of the University of Kentucky State Botanical Garden and Arboretum (38◦ 00′ 57.5′′N 84◦ 30′ 15.7′′W), Lexington, KY. Six pairs (replicates) of A. incarnata (about 90 cm tall) in 4 liter pots were sunk into the soil so that the pot rims were even with the ground surface. Plants within replicates were spaced 9 m apart along an east-west transect, with replicates separated by at least 11 m. One randomly-chosen milkweed of each pair was surrounded by three clumps of ornamental grasses, Panicum virgatum "Shenandoah," in 11 liter pots that were placed in a triangular array at 0.6 m distance. The uppermost foliage of the grasses and milkweeds was at similar height, with their foliage separated by about 0.5 m, but the grasses close enough that they might form a visual screen to monarchs flying over the landscape in search of milkweed for oviposition. The milkweeds were inspected daily for monarch eggs, and at each visit, such eggs were removed.

#### Statistical Analysis

Data relating the characteristics of the preexisting Monarch Waystations and total number of monarch eggs and larvae found in those gardens were analyzed by multivariate analysis of variance (ANOVA) using the Statistical Analysis System general linear models procedure (SAS, Version 9.4; SAS Institute, Cary, NY, USA) to test for associations between monarch abundance and garden characteristics including area, milkweed density, nectar plant density, and whether or not the garden configuration was structured or non-structured, as well as surrounding landscape features within a 100 m radius of the garden including % hardscape, number, and total area of buildings, distance to nearest building, 360◦ accessibility index, and north/south accessibility. We used stepwise model selection to omit independent variables not producing a significant Fstatistic and calculate adjusted r 2 values for the full and reduced models.

Counts of monarch life stages on the milkweeds were summed across sample dates, within year, and those totals were compared between garden layouts by two-way (ANOVA) for a randomized complete block design using Statistix 10 (Analytical Software, Boca Raton, FL). Direct counts of predatory invertebrates on the milkweeds, and numbers captured in the pitfall traps, were similarly analyzed for each data set, as were numbers of monarch eggs deposited on milkweeds that were or were not surrounded by ornamental grasses. Log or square root transformations were used if needed to meet normality and homogeneity of variance assumptions. Data are reported as original means ± standard error (SE).

#### RESULTS

#### Monarch Use of Preexisting Waystations

Multivariate analysis of variance for predictors of monarch egg and larval abundance in the 22 citizen-planted Monarch Waystations explained 63 and 71% of the variation with complete and reduced models, respectively (**Table 1**). Stepwise model selection identified three factors: garden configuration, north/south accessibility, and proximity to nearest building as significant sources of variation. Total numbers of monarch TABLE 1 | Summary of analysis of variance for the effects of garden characteristics and landscape features on the number of monarch eggs and larvae observed in gardens.


Adjusted r<sup>2</sup> full model; 0.63, reduced model; 0.71.

<sup>a</sup>Garden area (m<sup>2</sup> ), milkweed ramet density, nectar plant density, plant spacing (use of mulch to achieve plant separation) in garden.

<sup>b</sup>All measurements based on 100 m radius buffer zone around center of gardens. Accessibility index (degrees visually obstructed out of 360◦ ), line of sight north/south (visual obstruction north/south), area occupied by structures (% of buffer zone), % hardscape (includes buildings and any impenetrable surfaces), proximity to nearest structure, number of structures.

Significant variables that were retained from the full model during stepwise model selection indicated by (\*).

eggs and larvae observed in twice-monthly visits to each garden were about five-fold higher in structured gardens with spacing between milkweeds and non-host plants than in non-structured gardens where those plants were closely intermixed (**Figure 2A**), and similarly higher in gardens with unobstructed north-south access compared to ones where such access was obstructed by buildings (**Figure 2B**). There was also a positive relationship between monarch abundance and proximity to the nearest structure. Other features of the gardens themselves (area, density of milkweeds, or nectar plants) or of the surrounding landscape with a 100 m radius did not explain a significant amount of variance in use by monarchs (**Table 1**). The gardens varied with respect to percentage of surrounding area occupied by hardscape (5–78%) and degrees of 360◦ access impeded by buildings or other structures (0–360◦ ).

All 22 gardens contained A. incarnata, A. syriaca, and A. tuberosa which were nearly equally represented (**Figure 2C**). Two gardens also contained one or two plants of A. verticillata (whorled milkweed), but no other milkweed species were represented. Total milkweed ramets per garden averaged 54 ± 8.7 (range 10–198). Total numbers of eggs and larvae found in the six, twice-monthly inspections averaged 13.3 ± 3.9 per garden, with high variability (range 0–61) between garden sites. Across all gardens, we found a total of 137, 134, and 11 monarch eggs and larvae on 380, 437, and 312 ramets of A. incarnata, A. syriaca, and A. tuberosa, respectively, with proportionately more on A. incarnata and A. syriaca than on A. tuberosa (χ <sup>2</sup> = 109.0, P < 0.001; **Figure 2D**). Monarch abundance (total for all garden counts) built up over the growing season, peaking in September.

FIGURE 2 | Summary data from season-long survey of citizen-planted Monarch Waystations (N = 22): (A) Mean total monarchs (eggs and larvae) in structured gardens (milkweeds in uniform array, separated from other plants by ≥0.5 m) or non-structured gardens (milkweeds closely intermixed with non-host plants); (B) Mean total monarchs (eggs and larvae) in gardens with or without unimpeded north-south access to 100 m: (C) Mean total ramets per garden of the three predominant milkweed species; (D) Mean total monarch eggs and larvae per 100 ramets of each milkweed species. Asterisk denotes significant difference. See text and Table 1 for statistical comparisons.

# Monarch Use of Experimental Gardens of Differing Configurations

In both 2017 and 2018 monarch eggs and larvae were 2.5–4 times more abundant in gardens in which the milkweeds were planted around the perimeter, surrounding the nectar plants and grasses, than when the layout was reversed, with milkweeds in the garden interior, or when the milkweeds were randomly intermixed with the other plants (**Figure 3**).

All three garden configurations harbored similar communities of predatory arthropods. Lady beetle adults and larvae (Coccinellidae), lacewings (Chrysopidae), and spiders (Araneae) were the most abundant predators observed on the milkweed plants (**Figures 4A,B**) with smaller numbers of ants, predatory Hemiptera (Pentatomidae, Reduviidae, and Nabidae) and others. Direct counts on the milkweeds did not differ among garden types for any predator group [**Figures 4A,B**; F(2,8) ≤ 1.7 for all individual taxa; all P ≥ 0.24]. Ground-dwelling predators captured in pitfall traps included ants, spiders, ground beetles (Carabidae), rove beetles (Staphylinidae), harvestmen (Opiliones), and other groups (**Figure 4C**). Garden design had no effect on activity-density of any of those groups [F(2,8) ≤ 1.5 for all individual taxa; all P ≥ 0.27].

## Effect of Surrounding Vegetation on Susceptibility of Milkweeds to Oviposition

Female monarchs foraging in an open-field setting laid significantly more eggs on single milkweed plants that were accessible from top to bottom, without visual obstruction, compared to single plants surrounded by, but not touching, ornamental grasses of equal height (**Figure 5**). Milkweeds screened by the grasses received almost no eggs over the 2 week trial.

# DISCUSSION

Numerous programs1−<sup>7</sup> encourage individual landowners, citizen scientists, and organizations in residential areas to establish gardens with milkweed and nectar plants to help offset habitat loss across the monarch's breeding range, and to increase connectivity among habitat patches in other land types. Optimizing the conservation value of such gardens is important because of the substantial effort and resources being directed toward them, and because restoring monarchs to a population goal specified in the North American Monarch Conservation Plan will likely require contributions from all land use sectors (Pleasants, 2017; Thogmartin et al., 2017). Indeed, geospatial extrapolations indicate that if all metropolitan areas across the US eastern range were engaged, they could provide nearly a third of the projected milkweed needed to sustain the eastern monarch population (Johnston et al., 2019).

To contribute to monarch conservation, gardens must first attract females to lay eggs. Monarchs find and oviposit on milkweeds in small urban gardens, often with higher egg-loading per plant than in natural habitats (Cutting and Tallamy, 2015;

Stenoien et al., 2015; Baker and Potter, 2018; Geest et al., 2019). The present study indicates that the layout of such gardens strongly influences the extent to which the milkweeds therein are found and used. Results from each of its components; i.e., numbers of eggs and larvae in existing Monarch Waystations, colonization of replicated gardens with different configurations, and oviposition on milkweeds with or without surrounding non-host vegetation, support the hypothesis that at least within small gardens, milkweeds are more susceptible to discovery and oviposition when they are spatially separated from nectar and non-host plants as opposed to being closely intermixed with them.

Host-finding by most butterfly species involves a sequence of behaviors including habitat location, orientation, landing, and plant surface evaluation (Renwick and Chew, 1994). Monarch adults are highly vagile and move extensively between habitat patches with milkweeds and nectar plants, but the relative distances over which they use visual or olfactory cues to locate resources are poorly understood (Zalucki et al., 2016). Caged lab-reared monarchs learned to associate the color and shape of artificial flowers with a nectar reward in the laboratory (Cepero et al., 2015), suggesting they also use such visual cues when orienting to hosts in the field. Upon landing, females engage contact chemoreceptors on their antennae and tarsi to assess plant suitability for oviposition, with flavonol glycosides in asclepiad hosts serving as oviposition stimulants

(Baur et al., 1998). Monarchs encountering natural stands of milkweed tend to lay more eggs on taller plants than on shorter ones, and more eggs per plant on isolated plants, and on plants at the edge of a patch compared to ones in a patch center (Zalucki and Kitching, 1982a,b; Zalucki et al., 2016).

In our study the gardens were standardized by area and botanical composition. All gardens contained the same number of milkweeds, but the interplant distances between milkweeds differed and were systematically greater in the "perimeter milkweed" layout than in the other garden designs. Because monarchs are known to preferentially oviposit on

isolated milkweeds, this may have influenced the results. Our purpose, however, was to find ways to optimize monarch use at the whole-garden scale by comparing same-sized gardens planted in different configurations. Consistent with Pitman et al. (2018), who found higher egg densities in small (<16 m<sup>2</sup> ), low-density (0.1–2 milkweed per m<sup>2</sup> ) milkweed patches in agricultural areas than in larger, higher-density milkweed patches, our small experimental gardens and surveyed Monarch Waystations were readily colonized and used by monarchs.

Visual and chemical stimuli from host and non-host plants can affect specialist herbivores' ability to find and colonize habitat patches, and their behavior in those patches (Tahvanainen and Root, 1972; Root, 1973; Risch, 1981; Finch and Collier, 2000; Bruce et al., 2005). The strength of attractive stimuli for a particular herbivore determines what Root (1973) called "resource concentration" which is affected in turn by density and spatial arrangement of host and non-host plants, and potential interference from non-hosts. (Root, 1973)Resource Concentration Hypothesis predicts that a specialist herbivore approaching a habitat will have greater difficulty locating a host plant when the relative resource concentration is lower. Nonhost vegetation may impair specialists' host-finding by physical obstruction, visual camouflage, making it more difficult for the herbivore to identify correct blends of volatiles produced by host plants against a complex background of volatiles from non-hosts, shading, or otherwise causing host plants to become less attractive or suitable (Tahvanainen and Root, 1972; Root, 1973; Risch, 1981; Bruce et al., 2005). Moreover, "inappropriate" landings on non-hosts may cause specialists to emigrate more quickly from mixed-plant habitat patches of low resource concentration (Root, 1973; Risch, 1981; Finch and Collier, 2000). There is evidence that monarchs are more likely to find and oviposit on milkweeds growing in monoculture agricultural fields than on milkweeds embedded in more botanically diverse habitats such as roadsides, nature preserves, and prairies (Pleasants and Oberhauser, 2013).

Some other diurnal specialist butterflies (e.g., the pipevine swallowtail Battus philenor) that use visual cues, e.g., leaf shape, when approaching host plants for oviposition have more difficulty locating hosts growing amid non-host vegetation than when such vegetation is removed (Rausher, 1981). A similar phenomenon, involving both visual camouflage and physical obstruction, may explain the results from this study. Results of our trial comparing oviposition on individual milkweed plants surrounded or not surrounded by non-host grasses also support the visual camouflage/physical obstruction hypothesis.

Resource concentration and accessibility may also help to explain why female monarchs moving amongst natural patches of milkweed tend to lay more eggs on relatively taller, single, isolated, or edge plants (see above). Indeed, Zalucki and Kitching (1982b) predicted that once a female finds a habitat patch, her movements will be determined by local environmental stimuli; e.g., host plant spacing, flowering plants, and edges, as well as her physiological condition. Those movements determine patch use, and how quickly a patch is "lost" by the butterfly wandering out of it.

An alternative hypothesis for why we found fewer monarch eggs and larvae in gardens having the milkweeds closely intermixed with nectar and non-host plants is that predatory invertebrates might be more abundant in such gardens, or might more readily move from non-host plants to prey on monarchs on adjacent milkweeds. However, our pitfall traps and direct inspections of milkweed plants found no evidence that garden design affected abundance of any predator group. We did not measure parasitism, or losses to birds, vespid wasps, or other flying predators, but there is no reason to expect those mortality agents would be any more or less prevalent in gardens having different layouts of the same plants. Indeed, visually-searching predators would seemingly have less difficulty finding monarch larvae on milkweeds not intermixed with other plants which, if affected by garden configuration, would have contributed to per-garden populations opposite of what we found.

Of those landscape features we analyzed, unimpeded northsouth access to gardens was the strongest predictor of monarch egg and larval abundance in citizen-planted Monarch Waystations. Although monarchs foraging locally may approach and leave milkweed patches from all directions (Zalucki and Kitching, 1982b), unimpeded north/south access to gardens may be particularly important for them to be encountered and used when adults are flying predominantly southward during their fall migration or northward during spring migration. North-south access may also be important because availability of nectar sources, particularly during autumn migration, may be critical to monarchs' migration success (Saunders et al., 2019). Interestingly, neither overall percentage of hardscape within a 100 m radius of the gardens, nor the percentage of total (360◦ ) access blocked by buildings, was a significant determinant of monarch use. Several of the gardens with relatively high numbers of monarchs were located close to the east or west side of buildings, which may account for the positive correlation between those factors in the multivariate analysis. Orientation of a garden in relation to structures, not the proximity per se, may affect monarch use. Nevertheless, the two least productive Waystations we surveyed were the only ones located in courtyards where access to them was blocked by structures. Further research on monarch foraging in relation to hardscape and other features of urban landscapes is warranted.

Despite the public's high level of enthusiasm and capacity for monarch-friendly gardening and projections that the urban sector can make important contributions to monarch recovery (Thogmartin et al., 2017; Johnston et al., 2019), the conservation value of such gardens remains uncertain. That urban milkweed gardens have the potential to recruit monarchs, often with higher egg-loading per plant than occurs in natural milkweed stands, is established (Cutting and Tallamy, 2015; Stenoien et al., 2015; Baker and Potter, 2018; Geest et al., 2019). Such gardens, however, could serve as ecological traps if they expose monarch larvae to increased risk of predation, disease, or pesticides (Majewska et al., 2018; Geest et al., 2019). We did not measure egg or larval survival, but earlier studies found no difference in overall survival (Cutting and Tallamy, 2015), or in mortality from parasitic tachinid flies or the protozoan Ophryocystis electroscirrha (Geest et al., 2019) between urban gardens versus more natural sites in meadows or conservation reserves, respectively. We have documented high rates of European paper wasp, Polistes dominula, predation on monarch larvae in some urban gardens (Baker and Potter, unpublished). Given the propensity of this wasp to nest in building eaves, cavities, and other sheltered places associated with human structures (Liebert et al., 2006), it could potentially pose a greater hazard to monarchs in urban settings than in more natural ones.

Regardless of their value in helping to restore the eastern migratory monarch population, Monarch Waystations and similar gardens provide opportunities to engage large numbers of people in reconciliation ecology. While the magnitude of the current extinction crisis is widely recognized by scientists (IPBES, 2019), we are witnessing an "extinction of experience"

#### REFERENCES


(Pyle, 1993; Miller, 2005; Goddard et al., 2010) whereby the US general public, 80% of which now lives in metropolitan areas, is increasingly estranged from the natural world. Gardening for monarchs, whether by individual landowners, school children, or organizations, can help foster personal engagement with nature, providing social and educational connections that enrich urban residents' quality of life, and engendering public support for protecting native species (Miller, 2005; Goddard et al., 2010). Our findings suggest guidelines for designing small gardens that can help make the urban sector's contributions to monarch habitat restoration more rewarding for participants, and of greater value to monarch recovery.

# DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.

# AUTHOR CONTRIBUTIONS

AB and DP conceived the project and contributed to the study design, analyses, and writing the manuscript. AB conducted most of the fieldwork with help from undergraduate field assistants.

### FUNDING

Funding was provided by USDA-NIFA-SCRI grant 2016- 51181-235399 administered through IR4 Grant 2015-34383- 23710, BASF Living Acres Program, US Golf Association, the Horticultural Research Institute, Applewood Seed Co., University of Kentucky Nursery Research Endowment Fund, and USDA-NIFA Hatch Project no. 2351587000.

#### ACKNOWLEDGMENTS

We thank R. Brockman, M. Geis, B. Mach, T. D. McNamara, K. O'Hearn, C. T. Redmond, and L. Wallis for field and lab assistance, G. Munshaw for granting access to the experimental garden site, members of the Lexington Chapter of Wild Ones and others for permission to monitor their Monarch Waystations, R. Bessin for statistical advice, N. Barnes, S. Bhatt, and V. Ligenza for help with Waystation monitoring, and S. Malcolm, J. Lensing, and reviewers for insightful comments and suggestions.


must be added to increase the monarch population. Insect Conserv. Divers. 10, 42–53. doi: 10.1111/icad.12198


**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 Baker 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.

# Is the Timing, Pace, and Success of the Monarch Migration Associated With Sun Angle?

Orley R. Taylor Jr. <sup>1</sup> \*, James P. Lovett <sup>2</sup> , David L. Gibo<sup>3</sup> , Emily L. Weiser <sup>4</sup> , Wayne E. Thogmartin<sup>4</sup> , Darius J. Semmens <sup>5</sup> , James E. Diffendorfer <sup>5</sup> , John M. Pleasants <sup>6</sup> , Samuel D. Pecoraro<sup>7</sup> and Ralph Grundel <sup>7</sup>

*<sup>1</sup> Department of Ecology and Evolutionary Biology, University of Kansas, Lawrence, KS, United States, <sup>2</sup> Monarch Watch, Kansas Biological Survey, University of Kansas, Lawrence, KS, United States, <sup>3</sup> Department of Biology, University of Toronto, Mississauga, ON, Canada, <sup>4</sup> U.S. Geological Survey, Upper Midwest Environmental Sciences Center, La Crosse, WI, United States, <sup>5</sup> U.S. Geological Survey, Geosciences and Environmental Change Science Center, Denver, CO, United States, <sup>6</sup> Department of Ecology, Evolution and Organismal Biology, Iowa State University, Ames, IA, United States, <sup>7</sup> U.S. Geological Survey, Great Lakes Science Center, Chesterton, IN, United States*

#### Edited by:

*Jason Chapman, University of Exeter, United Kingdom*

#### Reviewed by:

*Jacobus De Roode, Emory University, United States Stephen Baillie Malcolm, Western Michigan University, United States*

> \*Correspondence: *Orley R. Taylor Jr. chip@ku.edu*

#### Specialty section:

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

Received: *11 April 2019* Accepted: *30 October 2019* Published: *10 December 2019*

#### Citation:

*Taylor OR Jr, Lovett JP, Gibo DL, Weiser EL, Thogmartin WE, Semmens DJ, Diffendorfer JE, Pleasants JM, Pecoraro SD and Grundel R (2019) Is the Timing, Pace, and Success of the Monarch Migration Associated With Sun Angle? Front. Ecol. Evol. 7:442. doi: 10.3389/fevo.2019.00442* A basic question concerning the monarch butterflies' fall migration is which monarchs succeed in reaching overwintering sites in Mexico, which fail—and why. We document the timing and pace of the fall migration, ask whether the sun's position in the sky is associated with the pace of the migration, and ask whether timing affects success in completing the migration. Using data from the Monarch Watch tagging program, we explore whether the fall monarch migration is associated with the daily maximum vertical angle of the sun above the horizon (Sun Angle at Solar Noon, SASN) or whether other processes are more likely to explain the pace of the migration. From 1998 to 2015, more than 1.38 million monarchs were tagged and 13,824 (1%) were recovered in Mexico. The pace of migration was relatively slow early in the migration but increased in late September and declined again later in October as the migrating monarchs approached lower latitudes. This slow-fast-slow pacing in the fall migration is consistent with monarchs reaching latitudes with the same SASN, day after day, as they move south to their overwintering sites. The observed pacing pattern and overall movement rates are also consistent with monarchs migrating at a pace determined by interactions among SASN, temperature, and daylength. The results suggest monarchs successfully reaching the Monarch Butterfly Biosphere Reserve (MBBR) migrate within a "migration window" with an SASN of about 57◦ at the leading edge of the migration and 46◦ at the trailing edge. Ninety percent of the tags recovered in Mexico were from monarchs tagged within this window. Migrants reaching locations along the migration route with SASN outside this migration window may be considered early or late migrants. We noted several years with low overwintering abundance of monarchs, 2004 and 2011–2014, with high percentages of late migrants. This observation suggests a possible effect of migration timing on population size. The migration window defined by SASN can serve as a framework against which to establish the influence of environmental factors on the size, geographic distribution, and timing of past and future fall migrations.

Keywords: Danaus plexippus, migration, monarch butterfly, Monarch Watch, phenology, sun angle

# INTRODUCTION

The eastern North American migratory population of monarch butterflies (Danaus plexippus) can migrate more than 4,000 km in late summer and early fall from breeding areas encompassing hundreds of millions of hectares across the eastern U.S. and Canada to reach Mexican overwintering sites comprising fewer than 20 hectares (Brower, 1997). The migration begins in early August at northerly latitudes and ends in Mexico as the last monarchs reach overwintering sites by early December. The migration has a leading and trailing edge with new recruits joining the migration as it progresses southward. Although we have learned much about how monarchs navigate during the migration (Reppert and de Roode, 2018), we do not understand what factors determine the pace of the migration and how migration timing may affect monarch survival during the migration. Pace here refers to the distance advanced per day and timing refers to the date at which a monarch was recorded within the migration.

To address these questions, we use data from the long-term monarch tagging program created by Monarch Watch (MW) to explore whether the fall migration from the eastern U.S. and Canada to Mexico is an orderly and predictable process, possibly associated with the daily maximum vertical angle of the sun above the horizon (Sun Angle at Solar Noon, SASN). Although there are many possible cues for monarch migration initiation and pacing, solar cues, weather effects, and daylength may be related to changes in monarch physiology and behavior that initiate and affect the pace of the migration south to Mexico (Barker and Herman, 1976; Reppert and de Roode, 2018). Here we compare the pacing of the migration determined from MW tagging records to the pace that might be associated with monarchs following spatial and temporal variation in temperature, daylength, and SASN across latitudes.

The arrival of the monarch migration at each latitude is characterized by directional flight with a distinct heading or bearing, not of just one individual, but of many monarchs (Perez et al., 1997). These arrivals overlap with the last reproductive monarchs of the previous generation and the emergence (eclosion) of new monarchs. Monarchs can be abundant before the migration begins, especially at the more northerly latitudes. Prolonged emergence of new monarchs that are the product of late-season reproduction can result in the presence of new monarchs late in the migration or even after the migration has progressed beyond a specific location. As a result of these dynamics, monarchs are typically present for 60–80 days at each latitude during the fall as shown by tagging records (Monarch Watch unpubl. data). The migration itself is more limited. It is typically about 28 days from the arrival of the leading edge of the migration to the last detection of directional flight for a specific location but can be shorter at more northerly latitudes or when migrations are delayed significantly by weather. For example, while monarchs may be abundant from 1 August through 10 October at 40◦ N, both dates can shift if temperatures are either below or above long-term averages during the early stages of the migration. There is little movement when temperatures are below 10◦C or above 30◦C.

Changes associated with the apparent path of the sun across the sky affect daily and seasonal patterns of behavior for many animal groups, including insects, crustaceans, amphibians, reptiles, mammals, and birds (Duangphakdee et al., 2009; Dingle, 2014; Lebhardt and Ronacher, 2014; Vogt et al., 2014; Mason, 2017; Warren et al., 2019). Cues from the sun, photoperiod and light polarization, for example, are processed along with other environmental characteristics, such as magnetic fields (Dreyer et al., 2018), weather, and biological time-keeping mechanisms to determine migration phenology by a variety of species (Helm et al., 2013; Åkesson and Bianco, 2017; Muheim et al., 2018). In monarchs, an internal circadian timekeeper is combined with a sense of the sun's horizontal (azimuthal) position into a time-compensated sun compass that helps maintain a consistent bearing during the migration (Perez et al., 1997; Mouritsen and Frost, 2002; Reppert and de Roode, 2018). This compass may work by sensing the angle of polarization of sunlight (evector) and may be supplemented by other modalities such as a magnetic compass or geomagnetic map sense that could enable navigation to the overwintering locations in central Mexico (Reppert and de Roode, 2018). The e-vector that the monarch perceives is related to the sun's vertical angle above the horizon, illustrating that the monarch's integration of information from solar signals can incorporate the sun's vertical position above the horizon as well as the sun's horizontal position (Heinze and Reppert, 2011). Temperature, photoperiod, and milkweed and nectar plant quality interact to change monarch reproductive status at the end of the breeding season, a change that typically precedes initiation of the southward migration to Mexico (Barker and Herman, 1976; Goehring and Oberhauser, 2002; Pocius, 2014). The solar signals may influence the physiological processes affecting migration by, for example, modulating release of juvenile hormone (Zhan et al., 2011).

Prior to the start of the tagging program initiated by Monarch Watch in 1992 and the Journey North monarch observation program (Howard and Davis, 2015), we lacked detailed information on the spatial and temporal dynamics of the monarch migration. It was generally known that most of the fall migrants were seen in September in the north and in October in the south, with initial arrival at overwintering locations in Mexico roughly coinciding with the Day of the Dead (2 November) (Brower, 1995). At that time, many assumed that migration pacing was mainly driven by weather.

Given the predominant north–south direction of the monarch migration and the documented ability of the monarch to sense a variety of solar cues and use them for navigation (Mouritsen and Frost, 2002), it is reasonable to consider whether solar characteristics related to latitude might play a role in the initiation and pacing of the monarch migration. Sun Angle at Solar Noon (SASN) is the sun's maximum daily vertical angle above the horizon occurring daily halfway between sunrise and sunset. Because SASN changes in a set manner through the fall season, being affected by date and latitude, it is a candidate for being a solar cue that might affect timing and pace of the fall monarch migration. Indeed, there are precedents for insects sensing and using sun angle to regulate daily activity patterns—for example, sun angle helps maintain daily activity patterns in honey bees and absconding behavior in Apis florea (Duangphakdee et al., 2009).

Taylor and Gibo formulated an hypothesis that monarch arrival at different latitudes during migration was related to sun angle. This hypothesis was formalized in a life cycle model (Feddema et al., 2004) and predicted timing of monarch arrival at different latitudes during both the spring and fall migrations as a function of sun angle. Taylor and Gibo observed that the leading edge of the early southward migration was associated with an SASN value of about 57◦ . For example, MW observers in Winnipeg, one of the most northerly MW tagging locations, often first reported southward flying monarchs in early August, shortly after SASN reached 57◦ . Monarchs typically arrived at the Monarch Butterfly Biosphere Reserve (MBBR) in the last days of October, again as the SASN value reached 57◦ . These observations suggest that SASN might be a cue for initiation of significant migration events throughout the yearly life cycle of the monarch butterfly and may affect the pace of the fall migration.

The sun angle hypothesis is based on MW observations that have not yet been systematically analyzed. So here we quantify MW data to assess whether tagging observations are consistent with a role for SASN in determining the initiation and pace of the southward migration to the MBBR.

We examine whether the pacing of the monarch migration, as derived from MW tagging data, is consistent with monarchs following a specific SASN or range of SASN values and, if such a relationship occurs, whether there are alternative pacing mechanisms that might be more strongly related. We specifically examine whether the migration pace might alternatively be set by monarchs following certain minimum temperatures on their way south to Mexico or whether the pace might be set by monarchs moving at a constant velocity during daylight hours. Thus, we compare rates of monarch movement derived from MW data to rates of movement we might expect if monarchs moved south to maintain a constant SASN as they reach lower latitudes, to maintain a minimum daily temperature, or to maintain a constant flight velocity during daylight hours. We refer to these three alternatives as the SASN, Temperature, and Velocity Scenarios.

To assess whether the pace of the monarch fall migration is related to the SASN, Temperature, or Velocity Scenarios, we use the location and date of tagging of >1.3 million monarchs tagged in the eastern U.S. and Canada to define the migration pace. We also examine dates and SASN values associated with tagging locations of >13,000 monarchs that were tagged in the eastern U.S. and Canada and recovered in Mexico. We use these recovery data to describe how the SASN at the time and place of tagging is related to the likelihood of a monarch's tag being recovered in Mexico to describe how migration timing might relate to overwintering population size.

#### METHODS

# The Monarch Watch Program and Tagging Protocol

The Monarch Watch Tagging Program<sup>1</sup> began in 1992 with recruitment of volunteers to tag monarch butterflies during the fall migration season in the Midwestern U.S. The program quickly grew to include taggers covering the entire range of the monarch population east of the Rocky Mountains in the U.S. and Canada. Monarch Watch provides participants with handling and tagging instructions and guidelines for expected passage of the migration at each latitude. Volunteers tag monarchs from early August through mid-November. The 9 mm circular tags are applied to the discal cell on the underside of a hindwing, a location close to the center of gravity of the butterfly. Tags weigh 9–10 mg or about 2% the mass of a 500-mg monarch. Low recovery rates of tagged monarchs in Mexico in the early years of this program led to the development of uniquely-coded, weather-resistant tags in 1997. This newer tagging system, and a program to pay residents from the Monarch Butterfly Biosphere Reserve (MBBR) and nearby locations in Mexico for recovered tags, led to a higher tag recovery rate.

# Tag Recovery

Monarchs usually begin arriving at the MBBR in the last days of October. Conspicuous colonies form by mid-November. Tags from some of the monarchs tagged in the U.S. and Canada are recovered in the MBBR. Most of the recovered tags are from dead monarchs found beneath the colonies by guides and ejido members. Because the ratio of untagged to tagged monarchs is likely >20,000:1 (Taylor, pers. obs.), search time to find a tag can be many hours. It is likely that most recovered tags are "discovered" rather than the result of active searches. To reward recovery efforts, representatives of Monarch Watch buy recovered tags from guides and residents in late winter each year. Because residents with tags often do not connect with Monarch Watch representatives each year, it can take 3–4 years to acquire most of the recovered tags from one tagging season. Monarch Watch acquires tags from all overwintering sanctuaries, especially El Rosario, the site that typically has the largest colony. Below, "recovered" or "recoveries" refers to tagged monarchs that successfully arrived at the MBBR in Mexico and whose tags were found.

The number of tags recovered in the MBBR differs among sites and years. The number tagged in the U.S. and Canada, overall size of the population, size of specific colonies and survival during the migration and through winter likely contribute to tag recovery rates. Other site characteristics that might affect recovery include understory density, accessibility of colonies, movement of the colonies during the winter season, increases in overwintering mortality, turnover in guides and economic conditions that motivate searching for tags.

<sup>1</sup>https://monarchwatch.org/tagging

#### Data Preparation

Taggers recorded the location, date, and sex of each butterfly tagged. Starting in 2004, participants were asked to record whether the butterflies were wild-caught or reared then tagged and released. Prior to 2004 the rearing status was less consistently reported. Some analyses in this paper are limited to data from wild-caught monarchs from 2004 to 2015 due to the possibility that recovery rates for these monarchs differ from those for reared monarchs (Steffy, 2015). Because some taggers each year failed to return their data, both the number of monarchs recorded as tagged and recovered each year is an underestimate.

Tagging locations were recorded as site names, usually a city or municipality, or occasionally a local landmark, such as a park. To affix consistent latitudes and longitudes to these location names, we geocoded all records. For each tagging, we associated sanctioned names from the Geographic Names Information System (GNIS) database of place names<sup>2</sup> plus ZIP codes (U.S.) and postal codes (Canada) provided by taggers. We geocoded both by place name and by ZIP code/postal code whenever possible and compared results. Discrepancies of >50 km between place name-determined and ZIP code-/postal code-determined latitudes and longitudes were reexamined individually to determine a reasonable geocoded latitude and longitude. This geocoding process should identify tagging locations to within 50 km of the actual tagging location and usually much closer. The data set analyzed here includes 8,389 tagging locations with unique latitude and longitude.

As noted, the Sun Angle at Solar Noon (SASN) (Woolf, 1968) is the sun's vertical angle above the horizon, calculated at solar noon, and is the time of day when the sun is at its highest point in the sky (**Figure 1**). SASN varies by date and latitude. We calculated the SASN associated with the date and location where each monarch was tagged using the maptools and insol packages in R (Corripio, 2014; Bivand and Lewin-Koh, 2018; R Core Team, 2018) and formulae<sup>3</sup> . These tagging locations may not be at or near the monarch's natal origin since tagged monarchs may have migrated into an area prior to tagging. At a given latitude, SASN declines during the fall migration, as illustrated in **Figure 1**. This decline is related to seasonal changes in the tilt, or declination, of the earth toward the sun, with the rate of change in the tilt speeding up then slowing down during the timeframe of the fall migration. On a given day-of-the-year, SASN increases with decreasing latitude.

The tagging region for the eastern monarch population was the U.S. and Canada east of the Rocky Mountains (Brower, 1995) 4 . Taggers were instructed to tag monarchs from early

sin α = sin δ sin ∅ + cos δ cos ω cos ∅

August to mid-November, which we considered to be the migration period. Based on long-term observations of monarch migrations and the probabilities of reaching Mexico if tagged at a specific time and latitude, we set early and late latitude-specific dates for tagging that eliminated ∼2.3% of tagged monarchs from the analysis. The early line is set from 1 August at 50◦ N latitude to 1 September at 25◦ N and the late line from 31 October at 50◦ N to 15 November at 25◦ N. Finally, we did not eliminate south Florida from the analyses, although the southernmost Florida population is considered to be largely non-migratory (Brower, 1995).

#### Scenarios for Predicting Monarch Movement Rates

We measured rates of movement (◦ latitude day−<sup>1</sup> ) from the breeding grounds to the MBBR by examining how latitude of tagging changed with date. For this calculation of observed movement rates, we selected wild-caught monarchs tagged between 2004 and 2015 whose tags were eventually recovered in the MBBR. We also examined three possible scenarios that might help explain the observed movement rates and thereby the pace of the migration—monarchs following a constant SASN, following maximum temperatures, and daily movement distances proportional to daylength. Temperature affects monarch migration speed (Knight et al., 2019). Daylength affects patterns of migration across many types of insects (Denlinger et al., 2017). Although other environmental cues might be sensed and reacted to during migration, temperature and daylength are good candidates for comparing effects on migration pace with SASN. For each of the three scenarios, we predicted daily movement rates, as degrees of latitude moved per day. We compared these predicted rates to day-to-day changes in latitude of tagging.

Specifically, we estimated how daily monarch movement distances would differ if monarchs moved: (1) daily distances that would result in maintaining a constant SASN each day during the fall migration (SASN Scenario), (2) at a daily rate that would allow monarchs to experience daily maximum temperatures of 20, 25, or 30◦C, based on historical temperatures (Temperature Scenario), or (3) at a constant hourly velocity during 70% of the daylight hours (Velocity Scenario). To compare the outcomes of these different movement patterns, we examined potential movement along a selected migratory pathway. The MBBR is located at ∼19.5◦ N and 100.3◦ W. To maintain a constant pathway, we estimated daily movement rates for a hypothetical monarch as it moved from southern Canada (ca. 49.34◦ N, 100.3◦ W) due south along longitude 100.3◦ W to the MBBR, a 3,300 km migration path. However, because tagging was not conducted in Mexico, comparisons to the three explanatory hypotheses stopped near the U.S.–Mexico border, ca. 26–29◦ N latitude.

We compared the predicted migration rates from the three scenarios to observed movement rates along the selected migratory pathway. The observed rate of movement was calculated for wild-caught monarchs tagged between 2004 and 2015 within 10◦ of longitude of the 100.3◦ W longitude line. For each day of the year (DOY), within each year, we calculated a

<sup>2</sup>https://geonames.usgs.gov/

<sup>3</sup>https://www.esrl.noaa.gov/gmd/grad/solcalc/NOAA\_Solar\_Calculations\_dayxls. A formula for calculating sun angle (from https://www.itacanet.org/the-sun-as-asource-of-energy/part-3-calculating-solar-angles/) is:

Where α is the sun angle, δ is the sun's declination angle which mainly varies by day of year, ∅ is the latitude, ω is the hour angle which equals zero at solar noon. <sup>4</sup>Determined from U.S. Geological Survey shapefile available at: https://water.usgs. gov/GIS/metadata/usgswrd/XML/physio.xml but with modifications to follow the Rio Grande River in the south.

mean daily latitude of tagging, if at least five tagging observations were available. We then calculated a mean tagging latitude for that DOY, by averaging the yearly means across years, if at least 5 yearly means were available to be averaged. Finally, we calculated the observed rate of movement, in ◦ latitude day−<sup>1</sup> , by subtracting the mean latitude on a given DOY from the mean latitude on the previous DOY.

For the constant SASN scenario, we calculated two variants, one maintaining a constant 57.01◦ SASN starting on 7 August 2019 and the second route maintaining a 48.49◦ SASN starting on 2 September 2019. These two SASN values were selected to represent high (early in the migration) and average SASN values, respectively, observed in the tagging data. During the fall migration season, at a given latitude, earlier calendar dates have higher SASN. We calculated the latitude a monarch would have to reach every 5 days to maintain a constant daily SASN during this hypothetical migration. The constant SASN scenario might be thought of as a monarch reaching a latitude daily at which the sun is at a specified SASN, such as 57.01◦ .

For movement rate as a function of temperature (Temperature Scenario), we divided the 3,300-km route into 10 equal sections and moved the hypothetical monarch along the route to place it at the start of each section on a date during which the historical mean daily high temperature was 20, 25, or 30◦C. We selected this target range for daily temperatures to represent temperatures that would be high enough during the day to be compatible with flight and with maintenance of reproductive diapause (oligopause) by females (Kammer, 1970; Pocius, 2014). Although monarchs can fly on sunny days with temperatures as low as 13◦C (Masters et al., 1988), we set the target temperature for this hypothetical journey on the high end of the possible temperature range for flight under the assumption that higher temperatures are energetically advantageous. However, since monarchs frequently use thermals to gain altitude while conserving energy, temperatures at ground level can be misleading (Gibo, 1981). For each of the sectionstarting locations, we examined historical temperature data<sup>5</sup> to find the date during the migration period for which the historical mean daily maximum temperature first hit 20, 25, or 30◦C. Beyond latitude 26.13◦ N (Salinas Victoria, Nuevo Leon, Mexico), the beginning of the eighth section, daily maximum temperatures are similar across the year (e.g., August–December daily maximum range ∼21–26◦C at San Luis Potosi, Mexico, at the start of the ninth section). Because of this lack of variation, we extended this examination of effects of daily temperature on migration pace only from 49.34◦ N to 26.13◦ N, instead of all the way to 19.5◦ N at the MBBR.

For the velocity and daylength test (Velocity Scenario), our hypothetical monarch began migrating on 7 August 2019, flying at velocities of 3–15 km h−<sup>1</sup> , comparable to flight speeds documented in published and unpublished studies (Howard and Davis, 2015). As the monarch moved down the route, we calculated the minutes of daylight (sunset minus sunrise times) available at the monarch's new location and date. We assumed 70% of daylight hours were available for flying, with the rest devoted to activities such as resting and feeding. The product of the available daylight hours multiplied by flight speed then predicted the endpoint of each day's progress down the migration path.

Using geographic origins of monarchs based on an analysis of carbon and hydrogen isotopes, Wassenaar and Hobson (1998) concluded that origins of monarchs were similar among all colonies. To determine whether recovered tags represented a similar pattern, we asked how recoveries from monarchs tagged at 1-day events at one location (Lawrence, Kansas), those tagged at one location over the season for many years (Cannon Falls, Minnesota), all taggings from Iowa, and all other taggings, were distributed across the three major colony sites (El Rosario, Sierra Chincua, and Cerro Pelon) from which recoveries were obtained.

Statistical analyses were performed in R (R Core Team, 2018), including analyses of variance (ANOVA), contingency table analysis, and linear regression (Tabachnick and Fidell, 2001; Fox and Bouchet-Valat, 2019). To describe the relationship between day of year and mean latitude of tagging, we used multivariate adaptive regression spline analysis (Milborrow, 2019). This flexible regression technique models non-linear relationships by breaking a curvilinear relationship into multiple line segments, each with its own slope. We modeled other nonlinear relationships, approximated by Gaussian and exponential decay curves, using SigmaPlot software (Systat Software Inc, 2008). We used root mean square error (RMSE) to quantify the difference between observed rates of movement and rates predicted by the different movement scenarios on a given day (Cort and Kenji, 2005). We compared how much variation in observed movement rates might be accounted for by the three movement scenarios (constant SASN, Temperature, Velocity) by regressing daily movement rates predicted by the scenarios on observed movement rates. We assessed the strength of relationships between the scenarios, and their interactions, and the observed rates in several ways. First, we calculated standardized regression coefficients (β) to compare relative effects of the three scenario movement rates on the observed movement rates (Tabachnick and Fidell, 2001). Second, we calculated squared semi-partial correlations (η 2 ) that indicate how much overall fit (R 2 ) is reduced if an independent variable, or an interaction between independent variables, is deleted from the regression (Tabachnick and Fidell, 2001).

We examine possible effects of migration timing on overwintering population size in the MBBR. The World Wildlife Fund in Mexico<sup>6</sup> measures the area (ha) of the MBBR with substantial presence of overwintering monarchs as an index of overwintering monarch abundance. We use that area coverage as a measure of monarch abundance in the MBBR.

# RESULTS

### Tagging Effort/Success and Relationship of Numbers Tagged to Monarch Abundance

Between 1998 and 2015, Monarch Watch volunteers tagged and reported on 1,385,518 adult monarchs across the U.S. and Canada east of the Rocky Mountains between 1 August and 15 November within the bounds of early and late date and latitude lines described in Methods. Of this number, 13,824 (1.00%) tags were retrieved from the MBBR and nearby locations and are termed recovered tags or simply "recoveries" or "recovered." The remaining 99% are termed "not-recovered" or "non-recovered." The proportion of tagged butterflies that reached the overwintering colonies but were not recovered is not known, so non-recovered tags represent both monarchs that failed to complete migration and those that successfully migrated but were not recovered at the MBBR.

From 1998 to 2015, monarchs were tagged at 8,389 unique locations (**Figure 2**). The abundance of monarchs, weather during the migration, the number and distribution of taggers, and tagging efforts likely affect the number of monarchs tagged. The yearly number of wild-caught monarchs tagged between 2004 and 2015 was significantly correlated (r = 0.85, n = 12, p = 0.0004) with the annual measure of hectares of trees covered with monarch clusters in the MBBR<sup>7</sup> . This result suggests that monarch abundance was a determinant of tagging numbers and that the late summer monarch population was correlated with the size of the overwintering population.

Recoveries of butterflies tagged at different origin locales (Cannon Falls, Minnesota; Lawrence, Kansas; Iowa; and all other U.S. and Canada locations) were distributed similarly among the three major overwintering colony sites in Mexico— El Rosario, Sierra Chincua, and Cerro Pelon (**Table 1**). Although the distribution of recoveries differed among these four origin

<sup>5</sup>https://usclimatedata.com for U.S. locations, https://weatherspark.com for Mexico locations.

<sup>6</sup>https://www.wwf.org.mx

<sup>7</sup>https://monarchwatch.org/a/monpop2019.png

Watch program from 1998 to 2015. Each dot represents a unique location determined from geocoding place names of tagging effort provided by Monarch Watch volunteers.

locales (X<sup>2</sup> = 27.7, df = 9, p = 0.001), about 80% of the recoveries from each of the four tagging origins were recovered in El Rosario while 7–12% were recovered from Sierra Chincua and Cerro Pelon.

# Comparison of Migration Pacing Predicted by Three Scenarios to Observed Migration Pacing

#### The Observed Pace of the Monarch Migration

To assess the pace of the fall migration, we calculated how mean daily latitude of tagging changed between consecutive days-ofyear (DOY) (**Figure 3**). Because tagging data were available only in the U.S. and Canada, the pace is illustrated only as far as the U.S.–Mexico border region. The multivariate adaptive regression spline describing the relationship between mean latitude and day of the year indicated that the migration pace across latitudes was slow (0.155◦ day−<sup>1</sup> ) from DOY 229 to 261 (17 August−18 September), increased to 0.220◦ day−<sup>1</sup> for DOY 262–269 (9 September−19 September), increased further to 0.426◦ day−<sup>1</sup> for DOY 270–285 (27 September−12 October), then slowed to 0.152◦ day−<sup>1</sup> for DOY 286–296 (13 October−27 October). We term these rates the observed rates of daily movement. The final rate estimate (0.152◦ day−<sup>1</sup> ) may be artefactually lowered by the southern limits of tagging (U.S.–Mexico border, ca. 26◦–29◦ N latitude) since monarchs present below this latitude on a given day will not be available for tagging, pushing the mean latitude estimates higher. These rates correspond to ∼17.2, 24.4, 47.3, and 16.9 km moved day−<sup>1</sup> , for the four DOY intervals assuming 111 km per degree of latitude, which represents the typical perpendicular distance between degrees of latitude.

#### Migration Pacing to Maintain Constant SASN, Temperature, and Velocity

Estimates of migration pacing were determined for three pacing scenarios along the 100.3◦ W longitude line from 49.34◦ N latitude to the MBBR (19.56◦ N latitude). **Figure 4A** shows the curve describing daily movement distances to maintain constant SASN vs. DOY. Two example migration situations are illustrated, a departure from 49.34◦ N latitude on 7 August to maintain a constant 57.01◦ SASN and a departure on 2 September to maintain a constant 48.49◦ . The daily pace needed to maintain TABLE 1 | Examples of distribution of recovered tags among locations in the Monarch Butterfly Biosphere Reserve (MBBR) representing season-long records for Cannon Falls, Minnesota, 1-day events in Lawrence, Kansas, the entire record for Iowa and for all recoveries between 1998 and 2017.


*Cannon Falls and Lawrence represent long time series of tagging activity from a specific locale and Iowa represents the area with the most tagging activity.*

constant SASN increases from early August until the beginning of October, peaking at about 0.39◦ latitude day−<sup>1</sup> (43 km day−<sup>1</sup> ) on day 274 (1 October), then decreases.

For the two constant SASN examples, departure on 7 August following an SASN of 57.01◦ requires 80 days to reach the MBBR, while a 2 September departure following 48.49◦ SASN takes 83 days. In each example, the pace initially increases then declines. As departure becomes later, the proportion of the trip with declining pace increases. These calculations indicate that in a constant SASN scenario, the date of reaching any latitude is predictable based on the latitude and date of tagging.

Longitude of origin will affect the distance moved per day to maintain a constant SASN. The due-south course along longitude 100.3◦ W represents the shortest traverse of latitudes to the

MBBR. A monarch originating from longitudes east or west of 100.3◦ W would need to fly further per day to reach lower latitudes associated the rate of change in SASN (**Table 2**). The

every other line numbered). On each graph, dotted curve represents the daily movement rates derived from day-over-day changes in mean Monarch Watch tagging latitudes between days-of-year 229–296 as shown in Figure 3.

TABLE 2 | Average daily movement distance (km day−<sup>1</sup> ) needed to maintain a constant Sun Angle at Solar Noon (SASN) when leaving from different longitudes but from a constant latitude (40◦ ) on the journey to the Monarch Butterfly Biosphere Reserve (MBBR) in Mexico.


*Constant SASN of 57.01*◦ *requires departure on 4 September. Constant SASN of 48.49*◦ *requires a departure on 26 September.*

TABLE 3 | Non-linear regression equations predicting distance monarchs would move per day, between DOY 229 and 296 (17 August–23 October), for three pacing scenarios.


origins to the east or west of 100.3◦ W would not alter the general shape of **Figure 4A** but would increase the daily distances (i.e., magnify the Y-axis) needed to maintain the constant SASN pace.

The constant Temperature Scenario tracked daily maximum temperatures of 20, 25, or 30◦C and predicted daily movement distances that initially increased, then decreased (**Figure 4B**). Root mean square error (RMSE) between the daily movement rate predicted by the Temperature Scenario and by the MW data was lowest, among the three Temperature Scenarios, for the 20◦C scenario. For all three temperatures, there was an initial increase in daily rate of movement, a decline in movement rate, and then a period of slow daily movement.

Finally, we described migration pacing determined by daylength and flight speed. In this constant Velocity Scenario, daily movement distances gradually declined with DOY, regardless of flight speed (**Figure 4C**). The 3-km h−<sup>1</sup> hourly velocity produced the lowest RMSE between the observed movement rate and the constant velocity among velocities from 3 to 15 km h−<sup>1</sup> .

RSME values of other scenario options that exceeded those of the closest fit options are not shown. **Table 3** shows the regression equations predicting daily movement rates (◦ latitude day−<sup>1</sup> ), between DOY 229 and 296, for the three scenarios.

#### Comparison of Observed Migration Pacing to Pacing Predicted by Constant SASN, Temperature, and Velocity Scenarios

The observed migration pacing to the U.S.–Mexico border between DOY 229 and 296 (17 August−23 October) increased until approximately DOY 285 then decreased (**Figure 3**). The constant SASN and constant 20◦C Temperature Scenarios both predicted an initially increasing migration pace, followed by a decline (**Figures 4A,B**). The fastest pace predicted in the scenarios overlapped the DOY with the fastest observed pace. The pattern of the constant Velocity Scenario, steady decline, did not match the observed pacing pattern of increasing then decreasing pace (**Figure 4C**).

For each DOY between 229 and 296, mean location (◦ latitude) was calculated for each of the three scenarios, based on the equations in **Table 3**, and compared to observed mean tagging latitude on that DOY (**Figure 5**). Cumulatively, the SASN scenario predicted arrival to the lower latitudes earlier than the constant Temperature or Velocity Scenarios and earlier than the observed data predict. For example, the SASN model predicted monarchs arriving at ∼30◦ N at a time that the MW observations suggested the mean latitude of the taggings to be between 36 and 38◦ N (**Figure 5**). Thus, the cumulative pacing predicted by following constant SASN was faster than for constant Temperature or Velocity and faster than observed. RMSE for the difference in daily movement distances predicted by the three scenarios compared to rates calculated from the MW observations was lowest for the constant Velocity and similar between constant SASN and constant Temperature Scenarios (**Table 4**). This suggests best overall fit between daily movement distances predicted by the Velocity Scenario and observed from the MW data. Nonetheless, RMSE was similar for the three scenarios, suggesting similar fits for all scenarios.

Considered together in a linear regression analysis, the three scenarios accounted for 77% of the variation in the movement rates observed from the MW data (**Table 4**). Based on the absolute value of the standardized regression coefficients, the movement distances predicted by constant SASN and Velocity had the greatest effect on the observed movement distances and the interaction between SASN and Velocity was the interaction with the greatest effect. Based on squared semi-partial correlations, the overall fit between the scenario predictions and the observed rates of movement was most affected, in order, by the interaction between the Velocity and Temperature movement rates, the interaction between Velocity and SASN movement rates, by SASN movement rates, and by Velocity movement rates. The Temperature × Velocity interaction and the SASN × Velocity interaction accounted for 0.26 and 0.24, respectively, of the total model R <sup>2</sup> of 0.77. SASN and Velocity individually accounted for 0.13 and 0.12 of the total model R 2 .

## Sun Angle Differences Between Monarchs Recovered and Not Recovered at the MBBR

Recovered tags represent a subset of all monarchs tagged within the two-dimensional space defined by latitude and DOY

results below the solid line indicate the predicted latitude was greater (i.e., further north) than observed. Predicted mean latitude was based on a 2,947-km migration from 49.34◦ N to 22.82◦ N latitude along longitude 100.3◦ W.

(**Figure 6**). SASN values at latitude and DOY of tagging for monarchs that are recovered in Mexico were a subset of SASN values of monarchs whose tags were not recovered (**Figure 7**). Tags that were not recovered included monarchs that arrived at the MBBR and individuals that did not make it to the MBBR.

No recoveries occurred among monarchs with associated SASN on the day of tagging between 26.64 and 35.56◦ (n = 7,728, 0.56% of total tagged between 1998 and 2015) or between 64.89 and 72.25◦ SASN (n = 1,807, 0.13% of total). No individuals with SASN lower than 26.64◦ or higher than 72.25◦ occurred in the data set. While tagged butterflies may have been present but not detected at the MBBR, the failure to recover any tags from monarchs tagged with high (>64.89◦ ; early migrants) and low (<35.56◦ ; late migrants) SASN values suggests that monarchs tagged at those SASN values were unlikely to successfully complete the migration.

For the tagged wild-caught monarchs between 2004 and 2015, 90% of the SASN distribution (5th to 95th quantiles) was within the interval 40.9◦–58.5◦ overall and within 46.0◦–56.8◦ for recoveries. This result suggests that a SASN window of ∼46◦– 57◦ at the DOY and location of tagging may be associated with successful arrival of migrating monarchs at the MBBR.

For wild-caught monarchs tagged between 2004 and 2015, a mean of 0.88% ± 0.70 (standard deviation, s.d.) (range 0.12– 2.48%) of non-recoveries had SASN values greater than the maximum SASN value of recoveries and 4.81% ± 4.99 (s.d.) (range 0.18–18.78%) of non-recoveries had SASN values less than the minimum SASN value of recoveries within a year. Thus, TABLE 4 | Regression of daily movement distances (◦ latitude) predicted by maintaining constant SASN, constant daily maximum temperature (20◦C), and constant velocity (3 km h−<sup>1</sup> ) during 70% of daylight hours on daily movement distances predicted from Monarch Watch tagging data 2004–2015.


*<sup>a</sup>Standardized regression coefficient.*

*<sup>b</sup>RMSE, Root mean square error for the difference in daily movement distances predicted by the SASN, Temperature, and Velocity scenarios compared to rates calculated from the Monarch Watch observations (*Figure 3*).*

*r: Pearson correlation between rates predicted from Monarch Watch observations and variable.*

η *2 : squared semi-partial regression coefficient; indicates how much overall R<sup>2</sup> is reduced if variable is deleted from regression equation (Tabachnick and Fidell, 2001).*

\**p* < *0.05;* \*\*\**p* < *0.001. Variables SASN, Temperature, and Velocity were centered by subtracting their means prior to regression analysis. This decreases possible effects of collinearity (Tabachnick and Fidell, 2001).*

about 0.88% of individuals that were not recovered were tagged at high SASN, or relatively early in the migration season, while about 4.81% were tagged at low SASN, or relatively late in the migration. This result suggests that more late migrants than early migrants were associated with SASN values outside of the range associated with monarchs that completed the migration to the MBBR.

### Possible Effects of Migration Timing on Monarch Abundance in the MBBR

We examined whether overwintering abundance of monarchs in the MBBR, measured as area (ha) of the MBBR with substantial presence of monarchs, was related to timing of the fall migration. The years 2004 and 2011–2014 exhibited relatively high percentages of late migrants based on tagging data (**Figure 8**). Since late migrants have low recovery rates, these late migrations may have been associated with higher mortality during migration and thus accounted for the relatively low area cover of monarchs measured at the MBBR in those years [2004 (2.19), 2012 (1.19), 2013 (0.67), and 2014 (1.13)]. Those years represent four of the five lowest hectares of overwinter coverage recorded between 1994 and 2018, the years of data available at the time of this analysis.

Mean SASN at tagging did not change significantly over the study period [F(1, 16) = 1.544, p = 0.23] and runs of values above and below the mean value of SASN were randomly distributed [Wald–Wolfowitz runs test (Caeiro and Mateus, 2014), Z = 0.97, p = 0.33], suggesting the lack of a pattern of high and low SASN values across years (**Figure 9**).

200–350, 19 July to 16 December, *n* = 1,411,214). Isoclines illustrate approximate limit of monarchs recovered (blue dots) (57◦–47◦ ) and not-recovered (red dots) (68◦–36◦ ) in the MBBR. Pink and green lines represent authors' estimates of the demarcation dates and locations separating monarchs that are migrating to Mexico from those that are of the previous breeding generation (left of the pink line), and butterflies with little chance of reaching the MBBR (right of the green line). Pink (early) line stretches from 1 August at 50◦N latitude to 1 September at 25◦N. Green (late) line stretches from 31 October at 50◦N to 15 November at 25◦N. Note the handful of recovered butterflies tagged left of the pink line-they may represent recently emerged monarchs tagged while "staging" before the start of the migration. Also note two recoveries to the right of the green line and the 36◦ SASN isocline-both may be the result of tag code or other errors, since each was tagged in November > 30 days after the last recovery for their respective latitudes. Analyses were limited to the area between the pink and green lines (*n* = 1,385,518).

# DISCUSSION

# Can Following a Constant Sun Angle Predict the Pace of Monarch Fall Migration?

From August through October, the overall pace of the migration, as derived from Monarch Watch (MW) tagging data, can be characterized as slow-fast-slow, or increasing in the first part of this period then decreasing as the monarchs progress southward from northern latitudes to the Texas–Mexico border. This pacing is consistent with analyses of the day-to-day progression southward of fall monarch roosts reported to Journey North (JN) (Howard and Davis, 2015). JN roost data were analyzed for four 20-day periods from 10 August to 27 October. Those data indicated a slow migration advance in the first half of the migration period followed by faster roost advances in late September–early October. In the fourth 20-day time interval (8– 27 October), which incorporates migration in Mexico, the mean rate of travel declined, but the value was not significantly different from the previous time interval (Howard and Davis, 2015). More data are needed to document pacing in this final section of the monarch migration, but the concordance between the MW and JN analyses supports a pattern of an initially slow migration pace that speeds up in the approach to south Texas and then slows down.

Taylor and Gibo (Feddema et al., 2004) hypothesized that migrating monarchs use solar cues for timing the initiation of the migration and for maintaining a migration pace. They suggested that the sun angle, the vertical angle of the sun above the horizon at the daily high point might be that solar cue because they observed that the migration pace varies by dayof-year and latitude and sun angle varies by day-of-year and latitude. We tested this sun angle hypothesis by comparing the pace of migration that would be associated with monarchs maintaining a constant daily sun angle at solar noon (SASN) to the pace derived from MW tagging data. Solar noon is the midpoint between sunrise and sunset and is when the sun is at its highest daily angle. We compared the observed pacing to the predicted pacing associated with three scenarios—monarchs maintaining a constant SASN, moving to maintain a daily maximum temperature of at least 20◦C, or flying at a constant 3 km h−<sup>1</sup> for diminishing hours of daylight per day.

Among the three migration pacing scenarios tested, the pattern of pacing increase and decrease and the peak value of daily movement distances from the SASN scenario matched well with the MW observations (**Figure 4A**). While the pacing pattern associated with SASN was similar to that observed from MW data, the pace predicted by SASN was overall faster than observed. The difference between the SASN and observed scenarios was mainly because the slow parts of the slow-fast-slow cycle in the SASN scenario were faster than observed. We calculated regression models that estimated how well all three scenarios accounted for variation in observed pacing. The results suggest pacing predicted from the three scenarios can account for about 77% of the variation in the observed pacing and that movement rates from the SASN

and Velocity Scenarios and their interaction, plus interaction between Velocity and Temperature, accounted for most of that explained variation. Therefore, the pattern of pacing predicted by following a constant SASN, and the variation in observed pacing accounted for by following SASN, are consistent with the SASN playing a role in determining migration pacing. However, if following SASN is, in fact, an important determinant of the migration pacing, it likely interacts with other factors, such as temperature and daylength in determining the observed pacing pattern. No one of the three scenarios provided the best match with the observed pattern and overall rate of pacing but together accounted for much of the variation in pacing.

Movement distances per day for a monarch traveling at a pace to maintain a constant SASN should peak around 1 October and then slow down (**Figure 4A**). Predicting population pacing based on millions of monarchs potentially advancing at a rate associated with the rate of change in SASN will require further analysis, especially mechanistic analyses of monarch response to SASN.

FIGURE 8 | Percentage of not-recovered tagged monarchs for which the Sun Angle at Solar Noon (SASN) was less than the minimum (open circles) or greater than the maximum (filled circles) SASN of recovered tagged monarchs.

FIGURE 9 | Mean Sun Angle at Solar Noon (SASN) (◦ ) (±95% confidence interval) by year for all data, wild-caught and reared tagged between 1998 and 2015 in the Monarch Watch program. Horizontal line represents mean of yearly means.

# Is Timing of the Migration Related to Overwintering Monarch Abundance?

SASN can be used as a metric that describes migration timing. Because SASN is determined by latitude and time of year, it combines these two parameters when considering how late in the migration period an event occurs. For instance, asking whether a butterfly tagged on 1 October is early or late in a typical fall migration period is not simply answered—it may be late if tagged in Canada or early if tagged in Texas. However, characterizing a tagging event or migration observation as occurring on a day and location when SASN equals 57◦ indicates that event occurred relatively early in the migration period, whether it is associated with initiation of migration in Canada or completion of migration upon arrival at the MBBR.

SASN values associated with monarch taggings suggest there is a temporal window across latitudes linked to successful completion of the migration to the MBBR. Ninety percent of the tags recovered in the MBBR were associated with SASN between 46.0 and 56.8◦ at the date and location of tagging, representing a "migration window" of SASN within which successfully completing the migration is more likely. In years with relatively high percentages of monarchs tagged at low SASN, representing late migrants, the overwintering population size in Mexico was generally low. This observation suggests a possible relationship between late migration and failure to complete the migration. A decline in migration success for later migrants has been shown in tagging records reported by Steffy (2015). A temporally and spatially defined migration window likely exists in many migratory insects, although temporally the window is often small (Bauer et al., 2011). Documenting migration windows across the animal kingdom can be critical to recognizing important changes in migration phenology that can affect survivorship during and after migration due to factors such as climate change (Cotton, 2003; Kelly et al., 2016).

The SASN-defined migration window can serve as a template that allows us to compare the relative pace of migrations among years and the pace of the migrations within specific regions. In addition, we can use the distribution of tagged butterflies across the range of SASN values, and mean SASN values, to characterize an entire migration, or regional parts of the migration, as fast or slow, early or late. We can then look for factors such as temperatures during the migration, summer temperatures and late recolonizations of the northern breeding area that may explain these differences. Although there was no significant indication in the current analysis that mean SASN values exhibited an upward or downward trend through time, a shift in the mean values to lower SASN values might occur with increasing summer and fall temperatures.

As noted, the percentage of monarchs migrating outside of the migration window of 56.8◦–46.0◦ , if substantial, could affect monarch overwintering abundance. Monarchs tagged late in the season (low SASN) yield few tag recoveries. Although the percentages of early and late (high and low SASN, respectively) migrants varied by year, late migrants were more common than early. The percentages of late migrants were high in 2004 and 2011–2014, all years with low area coverage of monarchs overwintering in the MBBR<sup>8</sup> . Many factors may have led to these late migrations. For example, the summer of 2004 was the coldest during this period, 2012 was the earliest recorded spring over much of the monarch's eastern breeding range (Ault et al., 2013) and was followed by high summer temperatures and low precipitation. These weather patterns likely reduced monarch breeding success. The 7 month drought in Texas in 2011 may have affected monarchs completing the migration to the MBBR or surviving once there. Since extreme weather conditions in each of these years likely influenced population growth and may have, in part,

Although we present analyses that the migration timing and success are associated with the pace predicted from SASN values, it is unknown whether SASN itself determines the observed relationship between SASN and the migration pacing documented by the MW data. Monarchs could be responding to other celestial cues such as e-vectors, light intensity, or specific azimuths. It appears that the migration starts for individual butterflies when the SASN declines to about 57◦ , but what of butterflies that eclose later, when the SASN is 47◦ ? These butterflies migrate, but to what stimulus are they responding and how might that stimulus be related to SASN? Again, we have much to learn about how monarchs successfully reach the overwintering sites in Mexico.

Although one of our scenarios showed that it might be possible for a monarch to maintain a reasonable pace if tracking declining SASN along longitude 100.3◦ W (the longitude of the MBBR), the longitude from which a butterfly starts will affect the daily pace needed to maintain a constant SASN. The further east, or west, the starting longitude is from 100.3◦ W, the further a monarch will have to advance daily to reach the latitude needed to maintain a constant SASN. This consideration gives rise to another question. Do monarchs originating east of 100.3◦ W, or later in the migration, fly longer each day to "catch up" with the pace of the changing fall conditions? Understanding the role seasonal events have on population growth and timing of both the emergence of the last generation and temperatures favoring the migration during the fall will be required to determine the dynamics that result in migratory success. Future analyses of the Monarch Watch data should help with that understanding.

While the results illustrate that we can learn much from the tagging data, these data have their limitations. Some of these limitations affect particular measures and interpretations relevant to the monarch migration; others are problems common to most citizen science projects (Brown and Williams, 2019). Although tags failing to adhere to monarch wings and effects of handling while tagging are thought to be minimal, both values could affect the results by increasing mortality during the migration and perhaps during the winter period. Lost tags could result in an underestimate of how many monarchs reach the overwintering sites. Tagging done early in the migration are more likely to apply tags to monarchs that are not yet migrating potentially leading to underestimates of recovery percentages for high SASN situations. While number of butterflies tagged can be used as a population index throughout the range, these measures are likely to result in underestimates of population size simply due to the distribution of taggers, the size of the population, weather events that limit tagging and times available to tag during the week and on weekends. For example, some areas that produce many monarchs are quite large. Low tagging rates in these areas, such as the eastern Dakotas and western Minnesota, are likely to produce underestimates of the population and number of monarchs

led to the lateness of these migrations, linking the tagging and migration success data to population growth and physical factors should lead to a richer understanding of monarch population dynamics.

<sup>8</sup>https://monarchwatch.org/a/monpop2019.png

reaching the overwintering sites in Mexico. Despite these limitations, the long time series represented by the tagging data, the geographic scope of the tagging, and the ability to associate these data with both demographic and weather patterns will continue to yield insights relevant to the dynamics of the monarch population.

#### Summary

The pace of monarch migration determined from Monarch Watch tagging data is initially slow at the northernmost latitudes, faster at mid-latitudes, and slows again at more southerly latitudes. That pacing pattern is similar to what would be expected for an individual monarch that maintained a constant Sun Angle at Solar Noon (SASN) throughout its migration. Whether a causal relationship exists between SASN and monarchs is not known and requires further study; however, SASN is associated with migratory success, since 90% of the recovered tags across all latitudes were tagged within a migration window defined by SASN values of 56.8◦–46.0◦ . Years in which high proportions of monarchs were tagged after SASN reached 46.0◦ exhibited low overwintering numbers, suggesting negative population consequences of late migration. Diverse factors, including SASN, temperature, and daylength, likely combine to determine the pace of each monarch migration.

The migration window defined by SASN can be viewed as a template, a means of standardizing the observations among and within years, that will allow us to assess the influence of factors such as temperatures during the summer and fall, the temporal distribution of migrants and the pace of the migration. Our ability to define a migration window and describe those factors that influence the pace of the fall migration will improve our understanding of how survival during the migration affects overwintering monarch abundance.

#### DATA AVAILABILITY STATEMENT

The datasets for this manuscript are not publicly available pending addition of additional years and further initial analyses. Requests to access the datasets should be directed to OT (chip@ku.edu).

# REFERENCES


# AUTHOR CONTRIBUTIONS

OT created the Monarch Watch Program. JL and OT oversaw data acquisition and input. SP, JL, OT, and RG contributed to data preparation and quality control. OT, JL, DG, EW, WT, DS, JD, JP, SP, and RG contributed to the conception and design of the study and preparation of the manuscript. RG wrote the first draft of the manuscript. OT provided a major revision. RG combined co-authors' comments into the final version. All authors contributed to manuscript revision and read and approved the submitted version.

# FUNDING

Donations to Monarch Watch were used to pay for recovered tags in Mexico. Funding from the U.S. Geological Survey Ecosystems Mission Area helped support preparation of the Monarch Watch data from 2017 to 2019, while funding from the U.S. Geological Survey Land Change Science program supported contributions from DS and JD.

# ACKNOWLEDGMENTS

Millions of monarch butterflies do not get tagged and thousands recovered without the effort of thousands of people across eastern North America since 1992, all of whom we acknowledge and thank. We are indebted to Ann Ryan, Diane Pruden, Debbie Jackson, Gail Morris, Trecia Neal, Carol Pasternak, Janis Lentz, Sarah Schmidt, Dana Wilfong, Cathy Walters, Alfonso Alonso, Carole Jordan, Diane Seaborn-Brown, Veronica Prida, David Kust and family, and many others who helped purchase and recover tags at the overwintering sites in Mexico. We are most appreciative of the numerous donors who have contributed to the tag recovery fund over the years. Many staff members and students assisted with data management in Kansas over the last two decades and we have been aided by many data entry volunteers. We are indebted to the guides and ejido residents who diligently recovered and saved each recovered tag. We thank Jeremy Havens and Tammy Patterson for help with figures. Any use of trade, firm, or product names are for descriptive purposes only and do not imply endorsement by the U.S. Government.

Synthesis, eds. E. J. Milner-Gulland, J. M. Fryxell, and A. R. E. Sinclair (Oxford: Oxford University Press), 68–87.


**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 At least a portion of this work is authored by Ralph Grundel, Samuel Pecoraro, Emily Weiser, Jay Diffendorfer, Darius Semmens, and Wayne Thogmartin on behalf of the U.S. Government and, as regards Dr. Grundel, Mr. Pecoraro, Dr. Weiser, Dr. Diffendorfer, Dr. Semmens, Dr. Thogmartin and the U.S. Government, is not subject to copyright protection in the United States. Foreign and other copyrights may apply. 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.

# Monarch Habitat in Conservation Grasslands

Laura Lukens1,2 \*, Kyle Kasten1,2 \*, Carl Stenoien<sup>3</sup> , Alison Cariveau1,2, Wendy Caldwell1,2 and Karen Oberhauser2,4

<sup>1</sup> Monarch Joint Venture, St. Paul, MN, United States, <sup>2</sup> Department of Fisheries, Wildlife and Conservation Biology, University of Minnesota, St. Paul, MN, United States, <sup>3</sup> Department of Entomology, University of Minnesota, St. Paul, MN, United States, <sup>4</sup> Arboretum, University of Wisconsin–Madison, Madison, WI, United States

#### Edited by:

Stéphane Joost, École Polytechnique Fédérale de Lausanne, Switzerland

#### Reviewed by:

Brian J. Wilsey, Iowa State University, United States David Jack Coates, Department of Biodiversity, Conservation and Attractions (DBCA), Australia Chrisophe François Randin, Université de Lausanne, Switzerland

\*Correspondence:

Laura Lukens llukens@monarchjointventure.org; luke0063@umn.edu Kyle Kasten kkasten@monarchjointventure.org

#### Specialty section:

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

Received: 31 December 2018 Accepted: 16 January 2020 Published: 06 February 2020

#### Citation:

Lukens L, Kasten K, Stenoien C, Cariveau A, Caldwell W and Oberhauser K (2020) Monarch Habitat in Conservation Grasslands. Front. Ecol. Evol. 8:13. doi: 10.3389/fevo.2020.00013 There is strong evidence that a major driver of the decline of eastern North American monarch butterflies (Danaus plexippus) is the loss of breeding habitat in the upper midwestern United States. Grasslands, including conservation areas, provide some of the largest remaining tracts of breeding habitat available to monarchs. While grassland conservation has been well-studied, little is known about how monarchs interact with these areas, or how planting and management practices impact the quality of habitat for monarchs. Here, we evaluate monarch habitat and use by monarchs in 61 conservation grasslands (including restoration sites in the U.S. Department of Agriculture Conservation Reserve Program, U.S. Fish and Wildlife Service Partners for Fish and Wildlife Program, and privately funded restored prairies) in Minnesota, Wisconsin, and Iowa. We documented milkweed (Asclepias spp.) density and diversity, blooming plant frequency and richness, and immature monarch density during the monarch breeding seasons of 2016 and 2017, along with seeding and management histories. Milkweed was observed at 60 of 61 study sites with a mean density of 1,390 plants per hectare (median = 783), a greater density than previously estimated in conservation grasslands. Monarchs were observed at 57 of 61 sites. Asclepias syriaca was the most frequently observed species, regardless of whether it was planted. Asclepias tuberosa and Asclepias incarnata may be the most cost-effective milkweeds to seed in our study geography, given that they were both more likely to be present and occurred at higher densities when planted than when not planted. Forb establishment rate varied across species planted and seeding rates. Increased rates of forb establishment were observed at larger sites, sites planted in the fall, and sites with fewer species in the seed mix. We observed a relatively low frequency of early season nectar sources, suggesting that managers should consider including more early blooming species in seed mixes and on existing conservation lands. We present establishment information for consideration in seed mix design and describe how our findings can be used to inform monarch habitat availability models, future studies, and conservation efforts.

Keywords: monarch butterfly, Danaus plexippus, milkweed, Asclepias, nectar plants, habitat monitoring, habitat management, prairie reconstruction

# INTRODUCTION

fevo-08-00013 February 4, 2020 Time: 17:12 # 2

The decline of the eastern migratory North American monarch butterfly (Danaus plexippus) population has been well documented (Brower et al., 2012; Rendon-Salinas et al., 2018). Annual population estimates in the overwintering sites in Mexico have revealed a steep decline in the area occupied by monarchs over the last two decades (Rendon-Salinas et al., 2018). Measurements of egg density in the northern United States suggest a similar trend during the summer breeding season (Stenoien et al., 2015). Projection models suggest that the monarch decline is worrisome enough to predict a monarch quasi-extinction probability of 11–57% over the next 20 years (Semmens et al., 2016).

While eastern monarchs experience many threats throughout their annual migration cycle, research suggests that a main cause of their population decline is the loss of breeding habitat in the Upper Midwest of the United States (Semmens et al., 2016; Pleasants et al., 2017; Thogmartin et al., 2017a). Monarch larvae consume plants in the genus Asclepias (and a few closely related genera), commonly known as milkweeds. Twenty years ago, a significant portion of monarchs originating in the Upper Midwest utilized milkweed found in agricultural habitats (Oberhauser et al., 2001). Since the introduction of genetically modified herbicide-tolerant row crops (corn and soybeans), milkweed within these crop fields has largely disappeared, significantly reducing the availability of monarch host plants in agricultural settings (Hartzler, 2010; Pleasants and Oberhauser, 2012; Stenoien et al., 2016; Thogmartin et al., 2017b).

Nectar resources are also an important component of monarch habitat. More than 99% of native northern tallgrass prairie has been lost since European settlement, and with it, many of the nectar resources that previously existed in these habitats (Samson and Knopf, 1994). Lark et al. (2015) estimate that 5.7 million acres of grassland in the U.S. were converted to cropland from 2008 to 2012, accounting for 77% of the overall cropland conversion during that time. This, coupled with the loss of milkweed in agricultural fields, has made monarch breeding habitat increasingly rare in much of their eastern breeding range (Pleasants, 2017).

Current population viability models estimate that 1.3– 1.6 billion milkweed stems need to be added throughout the eastern migratory range to bring the eastern migratory monarch population back to a sustainable level (Pleasants, 2017; Thogmartin et al., 2017a). Many habitat conservation or restoration initiatives, including programs like the U.S. Department of Agriculture Farm Service Agency's Conservation Reserve Program (CRP), are being engaged to help reach this habitat target. Despite conservation activities already being implemented, there have been few studies to examine the quality of restored or conserved habitats for monarchs, or how monarchs are using those habitats. Since conservation practitioners rely on a variety of pre- and post-planting tools and methodologies for a successful habitat project, understanding how these factors interact is critical to guiding future monarch habitat conservation activities.

Here, we (1) provide metrics of monarch habitat and use by monarchs in midwestern conservation grasslands and (2) investigate factors important for developing and maintaining monarch habitat and monarch use, including seeding and other management practices. We used an observational approach to address the following questions regarding the establishment and availability of milkweeds and blooming nectar plants, the defining features of monarch butterfly breeding habitat. What is the availability of milkweeds and blooming plants in these conservation grasslands and how do these metrics vary within growing seasons? For each milkweed species, is its inclusion in the seed mix or its seeding rate predictive of its establishment and density? How do seed mix characteristics (diversity and seeding rates), site age, planting season, and other management actions influence the frequency and establishment of milkweeds and blooming plants? Finally, are any site characteristics predictive of immature monarch abundance? Addressing these questions will inform monarch population and habitat modeling, land management decisions, as well as continued research on the importance of conservation lands for supporting monarchs and other pollinators.

# MATERIALS AND METHODS

#### Site Selection

We surveyed 61 conservation sites in Minnesota, Wisconsin, and Iowa during the growing seasons of 2016 (n = 23) and 2017 (n = 38) (**Figure 1**). Landowner and manager participants were recruited through outreach from local conservation offices, researchers, and other conservation stakeholders. Thirty nine sites were enrolled in CRP, 10 were part of the U.S. Fish and Wildlife Service Partners for Fish and Wildlife Program, and 12 were privately funded native prairie restorations that were not enrolled in formal conservation programs. Prior to restoration or conservation practices, most sites (n = 46) were used for row-crop agriculture, primarily corn or soybeans. Other previous land-use types included agricultural conservation land (expired WRP and CRP parcels, n = 4), pasture (n = 2), unmanaged grassland (n = 2), and remnant prairie (n = 1). The remaining six sites did not have information on previous use. All sites were seeded except for the remnant prairie and one unmanaged grassland. Time since seeding (site age) ranged from 1 to 32 years (mean = 8.4, median = 6), based on the most recent seeding of the entire site area, and size ranged from 1 to 38 hectares (mean = 11, median = 7).

# Field Survey Methods

Field sampling procedures followed the Integrated Monarch Monitoring Program (2017 version) (Commision for Environmental Cooperation, 2017; Cariveau et al., 2019a). Sites were surveyed 3–5 times during the monarch breeding season (May–September), with crews examining 150 1 m<sup>2</sup> quadrats (0.5 m × 2.0 m) placed 7 m apart along a series of parallel transects during each visit. The number of transects varied depending on site dimensions and size, and placement

and orientation were randomized for each visit using Geographic Information Systems software (ESRI, 2011).

The following data were collected on milkweeds, blooming plants, and monarchs:


at least one flower open. Opportunistic observations of blooming plant species found outside of quadrats (but within the site boundaries) were recorded separately during quadrat sampling and during the meandering walk survey completed afterward. Data collected in the meandering survey were only used to supplement a species list for each site. Such methodology is useful for detecting rare species that would be less likely to occur in the quadrats (Szigeti et al., 2016; Larson et al., 2018).


#### Seeding and Management History

Landowners or managers provided seeding and management data for each site to identify characteristics that could have influenced milkweed and blooming plant establishment. Seed quantities were reported in one of three units: pure live seed (PLS), bulk, or number of plugs. PLS, referring to the amount of viable seed, provides greater reliability in comparing across seed mixes (Englert, 2007; Houck, 2009), and therefore only sites that reported PLS were used in our seed rate analyses (n = 19). Landowners and managers reported using a combination of management practices, including prescribed burning, mowing, and herbicide use. Because all sites had been mowed or treated with herbicide at least once, yet detailed records of management dates and specific treatment areas were lacking from most managers, we excluded these practices from our analyses. Dates of prescribed burns were known for 55 of the 61 sites; 35 sites were burned at least once since planting (mean number of burns = 1.23, median = 1, range = 0–6).

#### Data Analysis

All statistical tests were performed in R version 3.5.1 (R Core Team, 2018). We excluded Asclepias verticillata (whorled milkweed) from all milkweed density analyses because its growth form and biomass are very different from the other species observed (small statured, dense clusters of stems), and because it was not included in models used to generate milkweed density targets (Thogmartin et al., 2017a). Blooming plant frequency was calculated as the proportion of subplots occupied by blooming plants. We calculated this in three ways: (1) the frequency of each forb species independently, (2) the total frequency of all planted forb species at a site, and 3) the frequency of all forb species at a site including those not planted. Blooming plant establishment rate refers to the proportion of forb species planted that were observed blooming at each site. Milkweed establishment rate refers to the proportion of sites in which a planted species of milkweed was observed. Milkweed colonization rate refers to the proportion of sites in which a species was observed but not planted. Immature monarch density is reported as the sum of the eggs and larvae divided by the number of milkweed plants examined.

We used Fisher Exact Tests to compare the establishment and colonization of the four milkweed species observed, and Kruskal–Wallis tests to determine whether planting milkweed seeds led to greater observed densities. Linear models were used to examine the effect of milkweed seeding rates on observed milkweed density.

To assess the effect of the time of sampling on milkweed density and the frequency of planted blooming species, we built two linear mixed-effects models (lme4 package) (Bates et al., 2015). In these models, we used the ordinal day of visit, with site ID as a random effect, as predictor variables. Date of sampling was treated as a second order polynomial variable because we predicted that milkweed density and blooming plant frequency would have curvilinear relationships with time. Sites at which milkweeds were never detected were excluded from the milkweed density analysis.

We used a two-step process to examine the effects of several predictors on four response variables of interest: milkweed density, blooming plant frequency (planted forb species only), blooming plant establishment rate, and immature monarch density. First, we built multivariate linear regression models using the lmer function in R, including the factors we hypothesized to be most important (**Tables 1A,B**) (lme4 package, Bates et al., 2015). We used generalized linear regression models (GLMs) with a binomial distribution and logit link function for blooming plant frequency and establishment rate. Next, we performed backwards model selection by AIC value using the MASS package step function to identify the relationship between each set of predictors and response variables (Venables and Ripley, 2002).

In these models, milkweed density, blooming plant frequency, and immature monarch density were averaged across visits for each site. We log-transformed milkweed density and immature monarch density to normalize the error terms of these otherwise right-skewed response variables. Site visits in which milkweeds were not detected were excluded from the immature monarch density model. Response variables were visually inspected for egregious outliers, and one site was removed from milkweed models because it had a milkweed density three times greater than the next largest value. Because only two sites were seeded during winter and one of them was never seeded with forbs (only grasses), we excluded winter plantings from all models. Two sites (the remnant prairie and existing grassland) were also excluded since they were never seeded, and therefore, age values were null.

Using Pearson's product moment correlation, we examined the correlation between site age, forb seeding rate, and number of planted forb species. We also examined the relationship between planting season and site age, and the relationship between planting season and forb seeding rate with ANOVA and Tukey's HSD. We considered them significantly correlated if p ≤ 0.05. Variables with statistically significant correlations were included as interactions in each model.

Due to limited sample sizes, burning frequency and forb seeding rate were examined univariately with each response variable. Because a significant portion of our study sites were younger than 3 years old and because plantings are not typically burned until the third or fourth year, we only included sites that were older than 3 years in the burning models (n = 30). Burning frequency was calculated as the total number of entire site burns divided by site age. Only sites that included seeding rate information in PLS were included in forb seeding rate analyses (n = 27).

# RESULTS

#### Milkweed

Across sites, we observed four milkweed species: Asclepias syriaca (common milkweed), A. incarnata (swamp milkweed), A. tuberosa (butterfly milkweed), and A. verticillata (whorled milkweed). At least one milkweed plant was observed at 60 of 61 study sites. Total milkweed density (A. syriaca, A. incarnata,


TABLE 1 | (A) Planting and post-planting management variables included in analyses of milkweed density, blooming plant frequency (planted forb species only), and blooming plant establishment rate. (B) Predictor variables included in immature monarch density analysis.

TABLE 2 | Milkweed density by species across sites (plants/ha) (n = 61).


and A. tuberosa combined) ranged from 0-16,880 plants/ha with a mean of 1,390 plants/ha (median = 783). Individual species densities are listed in **Table 2**.

Milkweed density varied significantly across the sampling period (**Supplementary Figure S1**). Based on the coefficients from the mixed-effects model, milkweed density peaked in mid-July (**Supplementary Table S1**).

Levels of establishment (Fisher Exact Test, p < 0.0001) and colonization (Fisher Exact Test, p < 0.0001) varied among milkweed species (**Figure 2** and **Table 3**). Asclepias syriaca was observed at 24 of 25 sites in which it was planted, and at all 33 sites in which it was not planted (n = 58, two sites were not seeded, one site lacking seed mix data). Asclepias incarnata was observed at 12 of 14 sites in which it was planted, and 21 of 44 sites in which it was not planted. Asclepias tuberosa was observed at 14 of 25 sites where planted, and 5 of 33 in which it was not planted. Lastly, Asclepias verticillata was never observed at sites in which it was planted (0 of 8) but was observed at 13 of 50 sites where it was not planted.

To determine whether planting any milkweed (regardless of seeding rate) leads to greater milkweed density, we compared densities based on whether a given species was planted or not (**Table 3**). When planted, Asclepias incarnata and A. tuberosa had significantly higher densities than when not planted, and A. verticillata and A. syriaca densities did not depend on whether they were planted. Seeding rate had a similarly varied effect on milkweed densities. When milkweed species were combined (A. syriaca, A. incarnata, and A. tuberosa), total milkweed density was significantly correlated with seeding rate (F1,<sup>15</sup> = 12.5, p = 0.003) (**Figure 3A**). Species-specific linear models indicated that A. incarnata was significantly and positively related to seeding rate (F1,<sup>5</sup> = 7.892, p = 0.038, **Figure 3B**). Asclepias syriaca and A. tuberosa densities did not significantly relate to seeding rate (F1,<sup>8</sup> = 0.208, p = 0.661, F1,<sup>10</sup> = 0.800, p = 0.392, respectively, **Figures 3C,D**).

After backwards model selection, seeding rate and site age were left as the best predictors of milkweed density, but neither was significant (**Table 4**). We did not detect any effects of

TABLE 3 | Comparison of the presence and densities of milkweed species at sites where milkweeds were planted (establishment) and at sites where they were not (colonization).


Note that milkweed could have established on its own at some of the sites at which it was planted, but we assumed that it had established from planted seeds; (\*) indicates significance at α = 0.05.

prescribed burn frequency on milkweed density (F1,<sup>29</sup> = 0.069, p = 0.794) (**Supplementary Table S3**).

### Blooming Plant Frequency

The average frequency of planted blooming species (those included in the seed mix) across sites was 0.29 (median = 0.20, range = 0-0.97). In other words, we observed planted blooming species in 29% of the 1 m<sup>2</sup> quadrats sampled, on average, during any given site visit. The average frequency of all blooming species (including weedy and volunteer species) was 0.45 (median = 0.44, range = 0–0.99). During peak bloom for each site (the visit for each site with the highest frequency of blooming plants), the average frequency of planted species was 0.54 (median = 0.60, range = 0.01-0.97), and the average frequency of any blooming species was 0.70 (median = 0.74, range = 0.17–0.99).

The frequency of planted blooming species varied significantly across the sampling period (p = < 0.001) (**Supplementary Figure S2**). Based on the coefficients from the mixed-effects model, blooming plant frequency at these sites peaked on August 2 (**Supplementary Table S2**).

Forb seeding rate was negatively correlated with site age (p = 0.01) and positively correlated with the number of forb species planted (p = 0.049). The number of forb species planted varied across planting seasons (p = 0.014). More forb species were planted in fall than in spring (p = 0.011) but summer was not significantly different from spring or fall (p = 0.132, 0.437, respectively). Similarly, site age varied across planting seasons (p = 0.018). Sites that were planted in the spring were typically older than those planted in summer (p = 0.030). Fall plantings did not significantly differ in age from spring or summer plantings (p = 0.071, p = 0.999, respectively), nor did forb seeding rate differ across seasons (p = 0.144). There was a trend for higher numbers of forb species planted on more recently established sites, but the correlation was not statistically significant (p = 0.102).

After backwards model selection by AIC value, site age remained as the best predictor of blooming plant frequency but was not significant (**Table 4**). We did not detect any effect of forb seeding rate or prescribed burn frequency on blooming plant frequency (p = 0.199, p = 0.962, respectively, **Supplementary Table S3**).

FIGURE 3 | Milkweed seeding rate (kg/ha) and observed density (plants/ha) by species and combined. (A) A. syriaca, A. incarnata, and A. tuberosa combined: Linear model, F1,<sup>15</sup> = 12.5, p = 0.003. (B) A. syriaca: Linear model, F1,<sup>8</sup> = 0.208, p = 0.661. (C) A. incarnata: Linear model, F1,<sup>5</sup> = 7.892, p = 0.038. (D) A. tuberosa: Linear model, F1,<sup>10</sup> = 0.800, p = 0.392.



(\*) Indicates significance at α = 0.05. Milkweed model estimates represent the log of the effect.

#### Blooming Plant Establishment

We documented a total of 288 blooming plant species across sites, with a mean of 36 blooming species per site (median = 34, range = 18–80) and 13 planted species (median = 12, range = 1–33). An average of 47% of blooming species planted at study sites were observed during sampling (median = 46%, range = 21–100%). The average number of forb species included in a seed mix was 29 (median = 29, range = 4–64). On average, we observed ten colonizing species per site that were not included in the mix (median = 9, range = 0–52). Erigeron annuus


TABLE 5 | Establishment and colonization patterns of the 10 most commonly planted forbs across study sites (for all forbs, see Supplementary Table S4).

(daisy fleabane), Cirsium arvense (Canada thistle), and Melilotus alba (white sweet clover) were among the most commonly observed colonizing species across sites (observed at 58, 49, and 38 sites, respectively; see also **Supplementary Table S4**).

The most commonly sown species are listed in **Table 5**. Rudbeckia hirta (black-eyed Susan) and Ratibida pinnata (yellow coneflower) were most successful among these species; they were observed growing at all sites in which they were planted (n = 50, n = 39, respectively) and with the highest average frequency across sites. Astragalus canadensis (Canada milkvetch) and Symphyotrichum novae-angliae (New England aster) were least successful when planted; they were only observed at 28 and 34% of sites in which they were sown (n = 11, n = 9, respectively), and at the lowest average frequency across sites (**Table 5**). Two commonly planted species, Ratibida pinnata and Astragalus canadensis, showed a positive frequency response to seeding rate while others did not (**Figure 4**).

After backwards model selection, the number of planted species, planting season, site size, and the interaction of number of planted species and planting season remained as the best predictors of blooming plant establishment rate (**Table 4**). The number of species planted was negatively related to forb establishment rate while site size was positively related to establishment rate (p < 0.001, p = 0.016 respectively, **Table 4**). Sites seeded in the fall (September–November) had higher establishment rates than those seeded in the summer (July– August, p = 0.031) (spring plantings did not significantly differ from either other season). We did not detect any effect of forb seeding rate or prescribed burn frequency on blooming plant establishment rate (p = 0.265, p = 0.173, **Supplementary Table S3**).

#### Monarch Occupancy

Monarchs (eggs, larvae, or adults) were observed at 57 of 61 sites surveyed. At least one adult monarch was observed at 54 of 61 sites. We observed direct use of habitat including nectaring and oviposition at 34 of 61 sites; at the other sites, adult monarchs were simply observed flying over the habitat. We observed 71 adult monarchs nectaring on 31 different blooming plant species across sites. The species on which they were most frequently observed nectaring were Monarda fistulosa (wild bergamot) (n = 10), Asclepias spp. (n = 8), Liatris spp. (blazing star) (n = 7), and Solidago spp. (goldenrod) (n = 5) (**Supplementary Figure S3**).

Milkweed density remained as the best predictor of immature monarch density after backwards model selection but was not significantly related to monarch density (p = 0.348, **Table 4**).

#### DISCUSSION

#### Milkweed

Regardless of whether or not they were planted, milkweeds were present at almost every site we studied. This suggests that Asclepias species in general, especially A. syriaca, are good colonizers, either due to viable seeds remaining in localized natural seedbanks or by dispersal from surrounding areas. However, since the effectiveness of herbicide tolerant technology has dramatically reduced the number of milkweeds within the agricultural landscape during the last 20 years (Pleasants and Oberhauser, 2012; Stenoien et al., 2016), the potential for future recolonization could be limited because of a reduced natural seedbank. On the other hand, studies in roadsides and urban areas have document a high presence of milkweeds (Kasten et al., 2016; Cariveau et al., 2019b; Johnston et al., 2019), which could support recruitment and colonization. Thus, while colonization success could be altered given changing milkweed abundance on the landscape, our study suggests that land managers can expect some degree of milkweed colonization.

Asclepias incarnata and A. tuberosa were more likely to be present and found at higher densities when they had been planted; A. syriaca was equally likely to be present in sites in which it was not planted, and we did not find A. verticillata in any of the sites in which it had been planted (although it colonized approximately one fourth of sites where it was not planted). Asclepias incarnata was the only species that showed a significant positive correlation between seeding rate and density. Based on these results, A. incarnata and A. tuberosa appear to be the most costeffective milkweed species to include in seed mixes within our study geography. Additional considerations for establishing

meandering walk and therefore has frequency scores of zero).

A. verticillata may be needed if it is a desired species for a conservation project.

Milkweed densities in our study sites peaked during mid-July, suggesting that monitoring efforts intending to capture peak milkweed density for similar sites in this bioregion should occur during mid-summer, after seedlings and ramets have emerged, but prior to senescence. Because milkweed density varies throughout the season, the highest densities that could be used to assess a site's contribution to national milkweed stem targets should occur mid-summer. Although we did not observe any effects of planting season, site age, or burning frequency on milkweed density, these factors may be worthy of further study, given that milkweed and monarch oviposition have been shown to respond favorably to landscape disturbance (Evetts and Burnside, 1972; Baum and Mueller, 2015; Haan and Landis, 2019).

Previous studies have estimated an average density of 277 stems per hectare in CRP lands and 8 stems per hectare in protected grasslands (Hartzler and Buhler, 2000; Hartzler, 2010; Thogmartin et al., 2017a). We observed a mean milkweed density of 1,390 plants per hectare (median = 783), approximately five times higher than these previous estimates. However, given that the landowners volunteered to participate in a study of monarch habitat, it is possible that the milkweed densities in our study sites may be higher than what is present on a random sample of CRP lands or conservation grasslands. Most of the participating landowners in this study expressed a desire to conserve pollinators and wildlife in general and managed their sites for these objectives. Nevertheless, the high milkweed densities on these sites set a standard for a quality that can be achieved across conservation grasslands.

Kasten et al. (2016) found that immature monarch density is positively correlated with milkweed density up to 4,942 plants per acre in roadsides. This suggests that managing for even higher densities than we observed could benefit breeding monarchs. However, managing for highly diverse grassland habitats that are rich with timely blooming species as well as milkweed host plants will benefit many additional species.

# Blooming Plant Frequency and Establishment

Previous studies have identified the importance of nectar plants as a component of monarch habitat (Stenoien et al., 2016;

Thogmartin et al., 2017a,b; Kinkead et al., 2019), and have proposed that a loss of nectar plant resources could be a significant contributor to increased mortality during migration (Agrawal and Inamine, 2018). The general lack of data on monarch habitat as a whole (milkweed plus nectar plants) makes studies like ours even more necessary. Because nectar plants are used by many species beyond monarchs, we stress the importance of tracking and gathering data on nectar availability.

We chose frequency sampling over other options (e.g., densities or aerial cover) because it is highly repeatable across observers and field conditions, is robust across seasonal variation, and is highly efficient (Elzinga et al., 1998). This method enabled us to assess relative frequency of species included in the seed mix and to characterize the blooming plant communities at each site. To detect more rare species, we conducted a meandering walk survey following quadrat sampling (following Szigeti et al., 2016; Larson et al., 2018), creating a more complete species richness list for each site. This allowed us to better characterize plant establishment, and the success of the seed mix.

We observed a higher frequency of blooming species in mid- and late-summer, suggesting greater nectar availability during those times. The relatively low frequency of early season nectar resources suggests that managers should consider including more early blooming species in seed mixes for new plantings and enhance existing conservation lands with these species. While monarchs need nectar resources during both the breeding and migration periods within our study geography, more early-blooming species may be especially important for monarchs arriving from an energy-intensive northward migration (Alonso-Mejia et al., 2011).

Our study sites were planted with a wide range of blooming plant species (194 species across all sites). Similar to milkweed, the establishment rates of these plants were variable and speciesspecific. Some species grew in all or most sites in which they were planted (Rudbeckia hirta, Ratibida pinnata) while others grew in very few (Astragalus canadensis, Symphyotrichum novae-angliae). Other species had high colonization success (establishing in sites where they were not planted). The colonization success of certain species (such as Dalea purpurea and Rudbeckia hirta) may give reason to reduce them in future seed mixes with the expectation that they may appear on their own, especially if their seeds are costly. However, though some species may have higher establishment success across projects, we stress the importance of diversity in conservation grasslands; a diverse group of blooming plants provide nectar resources throughout the growing season, supply host plants for a suite of insects, and are more ecologically resilient (Naeem and Li, 1997; Tilman, 1997; Timberlake et al., 2019).

Though forb seeding rate and blooming plant frequency were not significantly correlated in multivariate models, the sample size for our seed rate analyses was small. More than half of our sites reported seeds in bulk or seedlings or did not have seeding information available. For two planted species in our study, a positive correlation between frequency and seeding rate was observed (Ratibida pinnata, Astragalus canadensis), suggesting that planting more seeds of these species may lead to a greater abundance at a site. Our study does not present enough information to identify site or species characteristics that affect overall forb frequency, and therefore more research is needed to determine those effects.

We did not observe an effect of burn frequency or forb seeding rate on blooming plant frequency or establishment rate. However blooming plant establishment rate responded to seed mix forb diversity, site size, and planting season. Overall, the highest establishment rates were observed at larger sites planted in the fall, when fewer species were included in the seed mixes. However, there were a number of interactions between these variables. Almost half of the sites in our study sample (n = 25) were planted 1–3 years prior to sampling, and these sites tended to have more diverse seed mixes. Because prairie species may take many years to establish, younger sites may not have had a chance to establish as fully as those seeded a decade or more ago, and establishment rate might still increase through time on those sites. Furthermore, higher diversity seed mixes may include species that are more difficult to establish, whereas low-diversity seed mixes may be more likely to include those known to have high establishment success. Diverse seed mixes are valuable if they may yield a fuller array of native prairie species; further research into rates of establishment for various species may assist future conservation efforts.

Fall planted sites had higher establishment rates than sites planted in summer. Fall plantings are favored by many managers in this region to ensure that seeds are cold-stratified prior to the first growing season (Kurtz, 2001), and our findings suggest that this is a good strategy. However, because this was not a randomized experimental design, we recommend continued investigation regarding the efficacy of seeding seasons.

### Monarch Use of Sites

Monarch eggs, larvae, or adults were observed at most sites, suggesting that they provided suitable monarch habitat. Because monarch population estimates were well below historical averages during the study period (Rendon-Salinas et al., 2018), observed monarch densities across sites were also very low, making it difficult to detect any relationships between site characteristics and monarch numbers. More data are needed to better understand how relevant site characteristics (milkweed density, blooming plant richness and abundance) might impact monarch use (Leone et al., 2019).

Adult monarchs were observed nectaring from 31 blooming plant species, confirming that many blooming plant species on conservation grasslands provide nutritional resources for adult monarchs. While we noted that Monarda fistulosa was the most commonly utilized nectar plant, it was also commonly encountered on our sites, so we cannot make conclusions about monarch nectar plant preference.

# CONCLUSION

Conservation grasslands represent an important source of existing and potential monarch habitat (Thogmartin et al., 2017a) and our study demonstrates that they provide milkweed in

abundance; milkweed was observed at nearly every site and at densities much higher than previously estimated for similar grasslands (Thogmartin et al., 2017a). In a landscape drastically transformed by agriculture and development, this conserved habitat is critical for supporting monarchs and other wildlife.

The quality of habitat varied across sites, with a diverse suite of species at some, and few at others. Continued research on a larger sample of sites will further our understanding of the relationship between seeding and management practices and habitat responses across conservation grasslands. Other factors such as landscape context, weather during establishment, and soil characteristics could play a role in the establishment, colonization, and abundance of milkweed and nectar plants (Grman et al., 2015; Kaul and Wilsey, 2019), but these analyses were beyond the scope of our study.

Through the course of this study, multiple landowners suggested that many of the sites were managed for pollinators, and thus were likely to have better habitat resources than other similarly categorized sites. Management actions included manual or chemical weed removal and mowing, targeting species such as wild parsnip (Pastinaca sativa), thistles (Cirsium spp.), and buckthorn (Rhamnus spp.). Such actions were conducted with the intent of benefiting the native prairie plants, but due to a lack of spatial and temporal data on the extent of the actions, we were unable to determine their effects on the plant community. More detailed tracking by land managers can help to inform future conservation effectiveness studies to illustrate the benefits of these practices.

Our study required detailed information on seed mixes, seeding methods, and management actions. Ongoing studies will benefit from cooperation with landowners and managers who keep detailed records of management actions and who are open to sharing their practices with researchers. The protocols we used were prototypes for the Integrated Monarch Monitoring Program (IMMP), which monitors monarch habitat and monarch use throughout the North American breeding range (Cariveau et al., 2019a). The IMMP is an effective tool for addressing these questions. We encourage future researchers, landowners, and conservation practitioners to participate in the IMMP in order to build a more robust dataset for addressing questions relating to the effectiveness of their conservation practices. Ultimately, these data will lead to more efficient and effective conservation for monarchs and other pollinators.

#### REFERENCES


# AUTHOR CONTRIBUTIONS

KO, WC, LL, and KK designed the study. LL and KK performed the research. KO, WC, and AC advised the project. LL, CS, KK, and AC led the data analysis. All authors contributed to the development and writing of the manuscript.

# FUNDING

This work was supported by the USDA's Natural Resources Conservation Service under agreement number 68-7482-16-532, the Monarch Joint Venture, and the Environmental Defense Fund.

# ACKNOWLEDGMENTS

We thank several parties for assistance throughout this study; the Wisconsin Department of Natural Resources, US Fish and Wildlife Service, USDA's Natural Resource Conservation Service, Pheasants Forever Farm Bill Biologists, and Prairie Restorations, Incorporated assisted with identifying potential study sites. Melissa Martin helped identify sites, identify research targets, and served as liaison for the NRCS. Wayne Thogmartin and Diane Larson provided analytical support, and Karen Tuerk provided technical and database assistance. We thank Nicholas Haas, Madeline Esterl, Nicole Biagi, Maisong Lee, Lizzy Lincoln, Alexander Jack, Wesley Marchand, Maud Prineas, Rachel Olson, Tyler Zolczynski, and Evan Carlson for their hard work collecting data and for their dedication to the project. We would like to give a final and very special thanks to the landowners who agreed to be part of this study. Their passion and dedication to conservation is critically important in a time of drastic landscapelevel change.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo.2020.00013/ full#supplementary-material



America: identifying the threatening processes. R. Soc. Open Sci. 4:170760. doi: 10.1098/rsos.170760


**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 Lukens, Kasten, Stenoien, Cariveau, Caldwell and Oberhauser. 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.

# Evidence for a Growing Population of Eastern Migratory Monarch Butterflies Is Currently Insufficient

Wayne E. Thogmartin<sup>1</sup> \*, Jennifer A. Szymanski<sup>2</sup> and Emily L. Weiser<sup>1</sup>

<sup>1</sup> U.S. Geological Survey, Upper Midwest Environmental Sciences Center, La Crosse, WI, United States, <sup>2</sup> U.S. Fish and Wildlife Service, Endangered Species Program, La Crosse, WI, United States

The eastern migratory population of monarch butterflies has experienced a multidecadal decline, but a recent increase in abundance (to 6.05 ha in winter 2018) has led some observers to question whether the population has reversed its long-standing decline and embarked on a trajectory of increasing abundance. We examined this possibility through changepoint analyses, assessing whether a change in trajectory existed over a 25-year times series. We found evidence of a change in trajectory in 2014, but insufficient statistical support for a significantly increasing population since that time (β = 0.285, 95% CI = −0.127, 0.697). If the population estimate for winter 2019 is ≥4.0 ha, we will then be able to credibly assert the population has been increasing since 2014. However, given estimated levels of time series variability, presumed habitat capacity and no recent change in status or trend, there was a 13.5% probability of observing a population estimate as large or larger than was reported for winter 2018. Our analyses highlight the incredible difficulty in drawing robust conclusions from annual changes in abundance over a short time series, especially for an insect that commonly exhibits considerable year-to-year variation. Thus, we urge caution when drawing conclusions regarding species status and trends for any species for which limited data are available.

#### Edited by:

Cheryl Schultz, Washington State University Vancouver, United States

#### Reviewed by:

Stephen Baillie Malcolm, Western Michigan University, United States Natalie Kerr, Duke University, United States

> \*Correspondence: Wayne E. Thogmartin wthogmartin@usgs.gov

#### Specialty section:

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

Received: 29 March 2019 Accepted: 11 February 2020 Published: 26 February 2020

#### Citation:

Thogmartin WE, Szymanski JA and Weiser EL (2020) Evidence for a Growing Population of Eastern Migratory Monarch Butterflies Is Currently Insufficient. Front. Ecol. Evol. 8:43. doi: 10.3389/fevo.2020.00043 Keywords: changepoint, Danaus plexippus, extinction risk, small data, population ecology, time series analyses

# INTRODUCTION

"Short-Term Fluctuations May or May Not Contain Messages About Longer-Term Trends" – Art Shapiro

Populations vary over time in their abundance, and this variability can impart uncertainty to the status and trend of a species. As population dynamics approach extinction, dynamics become more variable (Fagan and Holmes, 2006), which means short-term highs might become higher, even while abundance is declining on average. In addition to the stochastic variation in abundance imposed by the environment, uncertainty in species status and trend arises from population sizes most often being estimated rather than counted; trends being inferred from limited duration time series; and latent characteristics of a population, such as its relation to carrying capacity or quasi-extinction thresholds, generally being inferred properties rather than an observable quantity. Thus, given these various sources of uncertainty, it is difficult enough to determine the trajectory for a population, let

alone any change that may occur in that trajectory, especially one that may occur near the terminus of a time series based on limited data.

Estimates of the population size of the eastern North American migratory population of monarch butterflies (Danaus plexippus, hereafter monarchs) in their overwintering locations in high-elevation oyamel fir (Abies religiosa) forests of central Mexico suggest a long-term decline in abundance. Using a model allowing separation of observation-induced error from natural process variability, Semmens et al. (2016) estimated monarchs declined by 84% between the winters beginning in 1996 (18.19 ha) and 2014 (0.67 ha), with an estimated annual population rate of change of 0.94. This rapid decline in monarch abundance led to widespread concern regarding the imperilment of the species (Brower et al., 2011), including a petitioning of the U.S. Fish and Wildlife Service (USFWS) to consider listing the species under the U.S. Endangered Species Act (ESA) of 1973 (Center for Biological Diversity and Brower, 2014).

The estimated rate of decline (λ = 0.94) in monarchs was, however, considerably uncertain, with credible intervals spanning from as low as 0.69 to as high as 1.30. This uncertainty, in turn, led to considerable uncertainty in the estimates of risk faced by the population; for instance, depending on the quasiextinction threshold chosen, the range of uncertainty in the risk was as much as one or two orders of magnitude wide (i.e., 0–34% at a 0.01 ha quasi-extinction threshold and 7–88% at 0.25 ha). The principal reasons for this large uncertainty in the trajectory of monarchs and their subsequent risk of further decline are the environmental and biological variability this insect faces over its annual cycle and our ability to intuit the species response to this variability with the limited data available from monitoring programs. Density-independent mortality, caused by a wide array of annually variable environmental stressors, is offset against density-dependent reproduction (Yakubu et al., 2004; Drury and Dwyer, 2006; Flockhart et al., 2012; Marini and Zalucki, 2017), and this tension between birth and death processes plays out over multiple generations and across the vastness of eastern North America (Flockhart et al., 2015; Oberhauser et al., 2017). In some years, these processes complement one another, leading to booms or busts in the population (Himes Boor et al., 2018). In other years, increases in one are offset by the other, mitigating any sizeable year-to-year change in population size.

In winter 2018, estimates of monarch abundance in their overwintering areas indicated monarchs increased by 144% over their previous year's abundance, to an index of population size of 6.05 ha (Conanp and World Wildlife Fund-Mexico, 2019). This estimate has led some observers to question whether the population has grown in recent years to the point at which it is no longer at risk. This seemingly simple question is manifold in nature. The question suggests that there may have been a change in the trajectory of the species in recent years, from a population in decline to one of increase, that in turn begs whether the evidence of this change in trajectory supports a reduced risk of quasi-extinction. An alternative possibility could be that the underlying status and trajectory of the population had not changed but instead the species demonstrated the extreme variability in year-to-year abundance that is not uncommon for insects.

To address this question, we conducted a time-series analysis examining whether the observed series of population sizes experienced changes in mean or trajectory anywhere over the 25 year period of record. The population as measured in Mexico reached its nadir in abundance in winter 2013 (Rendón-Salinas and Tavera-Olonso, 2014); we hypothesized that any change in status and any reversal of trend should occur at this point in the time series.

# METHODS

The overwinter index of population size (in hectares) we used in our models was that used by the USFWS in its Species Status Assessment for informing considerations of whether listing under the ESA is warranted. These data ranged from 1984– 2018. With these data, we evaluated two models, a step model ( ) evaluating whether there was a demonstrable change in status (i.e., mean abundance) during the time period and a segmented model ( ) examining whether there was a change in the trend; we specifically tested for a reversal of trend from a period of decline to one of growth. We fit these models in R (R Core Team, 2018) with both the changepoint (Killick et al., 2016) and chngpt (Fong and Gilbert, 2017) packages to ensure correct model outputs (see **Supplementary Datasheet S1**). Assumptions of independent, normally distributed data (on a log<sup>e</sup> scale) with constant variance pre- and post-change were evaluated with Shapiro and Kolmogorov–Smirnov tests and inspection of quantile-quantile and autocorrelation plots. We used an information-theoretic approach (with Akaike's Information Criterion) for selecting the best model among step, segmented, linear (no change in slope), and interceptonly formulations.

Pleasants (2017) suggested there was sufficient milkweed in the upper midwestern United States to support a mean population size overwintering in Mexico of 3.2 ha. He also asserted that in some years, the reported abundance is likely to be lower because of the accumulation of poor conditions faced by the population during its annual cycle, whereas in some years favorable conditions will lead to a population increase higher than 3.2 ha. We calculated the probability from a lognormal distribution of observing a 6.05-ha population relative to the 3.2-ha expected population size. We calculated the variance for this log-normal distribution from the variance of the post-2013 period.

Given that a changepoint was identified and the postchangepoint period was non-significantly increasing (95% confidence interval of the slope parameter overlapping 0) (see Results), we asked the question: How many more years of positive increase would be necessary to provide statistically robust evidence that the population was growing? To evaluate this question, we extrapolated the post-changepoint period abundance given the estimated post-changepoint slope and refit the changepoint model with additional years of extrapolated abundance.

#### RESULTS

When examining the time series of overwinter abundance of the eastern migratory population of monarch butterflies for a change in mean abundance (i.e., step change), we identified a single credible changepoint in winter 2009. For the period preceding this year, mean abundance was 6.69 ha (95% CI = 4.43, 8.94). For the period after winter 2009, mean abundance was 1.52 ha (95% CI ≥ 0.001, 4.68). The population variance was 15% higher in this latter period (σ 2 <sup>≤</sup><sup>2009</sup> =1.32 vs σ 2 <sup>&</sup>gt;<sup>2009</sup> =1.52), exhibiting greater variability at lower population sizes. If the underlying milkweed is currently sufficient to support a winter population of 3.2 ha (Pleasants, 2017), then a population as large or larger than 6.05 ha is expected to occur 13.5% of the time.

Fitting a segmented model, rather than a step model, suggested the best-supported year for the changepoint threshold was 2014 (likelihood ratio test of segmented model with and without changepoint, λ = 8.167, p = 0.0221; bootstrapped 95% CI = 2002, 2026), with 2013 close behind. The slope describing the decline of monarchs in the period before winter 2014 was negative (β = −0.103, **Table 1**), whereas after this winter the population exhibited a non-significant increase, though with confidence intervals >5:1 in favor of an increase (β = 0.285, 95% CI = −0.127, 0.697) (**Figure 1**).

Residuals from these step and segmented models before and after their changepoints were independent, normally distributed about their respective mean, and had constant variance. Comparing the segmented model (AIC = 45.3) with the step model (AIC = 49.5) suggested an 88% probability (odds 7.2:1) that the segmented model served as a better description of the data. Both models were appreciably better than an intercept-only model (AIC = 62.8) and a linear model regressing the log<sup>e</sup> (overwinter estimate) against year (AIC = 51.5).

TABLE 1 | Parameter estimates for the best-supported linear segmented changepoint model for 1994–2018 estimates of overwinter abundance of the eastern migratory monarch butterfly population.


If the winter 2019 population continues the mean rate of increase observed since 2014, then with this single additional year of data, we would have sufficient information statistically to conclude the population was growing (β = 0.399, 95% CI = 0.072, 0.727). Further, if the index of abundance was any value ≥4.00 ha, this amount too would be statistically sufficient (p < 0.05) to support a conclusion that the population was growing. Any value <4.00 ha, however, would cast doubt on a growing population.

#### DISCUSSION

At this time, there is insufficient statistical evidence to confidently assert that the eastern migratory monarch population has significantly grown since winter 2014. If the dynamic of population growth for the few years postwinter 2014 holds, then winter 2019–2020's population size estimate should provide evidence as to whether the trend has credibly changed from one of decline to one of increase.

In a noisy time series, stochastic fluctuations may lead to observed increases over relatively long periods, even when populations have an average negative growth rate. Similarly, stochastic fluctuations may cause a population to decrease, even when the long-term average growth rate is positive. Our analysis and the uncertainty it reveals highlights the difficulty in assessing species status and trend with even a 25-year dataset, especially when interannual variation is high. Semmens et al. (2016) reported a mean declining dynamic through 2014, but one with a non-negligible probability of a possible underlying growth rate that was positive. Their findings showed that two-thirds of the credible interval distribution about their estimate of the population growth rate was <1, indicating that the odds were 2:1 in favor of a declining population. Nevertheless, one-third of the distribution suggested a stable or growing population. With the full set of data through winter 2018 but with different methods, we find that the population prior to the estimated changepoint was similarly in decline (**Table 1**). Conversely, based on the interval width we calculated for the post-2014 trajectory, the odds are roughly 5:1 in favor of an increasing population. Unfortunately, the post-2014 period is too short to confidently conclude, at this time, a reversal in trajectory.

In any time series, the sample size is the number of years, and 10–30 years are often necessary to detect a significant trend even for species with average interannual variation (Urquhart, 2012; White, 2018). Despite the challenge of high interannual variation, the monarch butterfly is an iconic and highly visible species that benefits from strong public interest (Diffendorfer et al., 2014) and a corresponding availability of data (Ries and Oberhauser, 2015). For many species considered for listing under the ESA, even less information is available for evaluating the statistical support for any apparent decline. Thus, the challenge of assessing trend becomes even greater as one examines short-term time series or smaller periods of time within long-running time series; what may initially appear to be a short-term trend may have no statistical support in the context of the population's history. While assessing subsets of a time series could be a useful way to evaluate whether a species is moving toward recovery, caution is warranted when making conclusions based on limited data. Aside from the estimate of trend, other metrics can be useful in such cases, such as whether mean abundance falls below the estimated threshold for a secure population. In the case of the monarch butterfly, the recent mean of 1.52 ha falls well below the threshold of 6.0 ha estimated by Semmens et al. (2016) and established by the three nations of Canada, United States and Mexico as the nearterm population goal for the eastern population of migratory monarch butterflies. If we take this 6.0 ha threshold as a recovery criterion and assume a σ 2 <sup>&</sup>gt;<sup>2009</sup> =1.52, then the population is likely to need to reach a mean of 6.85 ha for 3 years to confidently assert the population has crossed this threshold (analysis not shown). Thus, this mean population size warrants continuing concern given the uncertain growth in recent years and the high year-to-year variability exhibited by this insect species.

#### DATA AVAILABILITY STATEMENT

Publicly available datasets were analyzed in this study. This data can be found here: http://d2ouvy59p0dg6k.cloudfront.net/ img/original/grafica\_ocupacion\_de\_colonias\_monarca.png.

### AUTHOR CONTRIBUTIONS

WT and JS conceived the study. WT conducted the analyses. WT, JS, and EW wrote the manuscript.

# FUNDING

This research was conducted in accordance with official duties as employees of the United States federal government.

#### ACKNOWLEDGMENTS

We appreciate M. Post Ven Der Burg and the two reviewers for comments made on an earlier version of this manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo.2020.00043/ full#supplementary-material

# REFERENCES

fevo-08-00043 February 24, 2020 Time: 17:4 # 5


**Disclaimer:** Any use of trade, firm, or product names is for descriptive purposes and does not imply endorsement by the U.S. Government. The findings and conclusions in this article are those of the authors and do not necessarily represent the views of the U.S. Fish and Wildlife Service.

**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 Thogmartin, Szymanski and Weiser. 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.

# Abies religiosa Seedling Limitations for Passive Restoration Practices at the Monarch Butterfly Biosphere Reserve in Mexico

Gerardo Guzmán-Aguilar<sup>1</sup> , Aglaen Carbajal-Navarro<sup>1</sup> , Cuauhtémoc Sáenz-Romero<sup>2</sup> , Yvonne Herrerías-Diego<sup>1</sup> , Leonel López-Toledo<sup>2</sup> and Arnulfo Blanco-García<sup>1</sup> \*

<sup>1</sup> Facultad de Biología, Universidad Michoacana de San Nicolás de Hidalgo, Morelia, Mexico, <sup>2</sup> Instituto de Investigaciones sobre los Recursos Naturales, Universidad Michoacana de San Nicolás de Hidalgo, Morelia, Mexico

#### Edited by:

Ryan G. Drum, United States Fish and Wildlife Service (USFWS), United States

#### Reviewed by:

Jordi Honey-Roses, The University of British Columbia, Canada Jay E. Diffendorfer, United States Geological Survey (USGS), United States

\*Correspondence:

Arnulfo Blanco-García arnulfoblanco@yahoo.com.mx

#### Specialty section:

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

Received: 31 December 2018 Accepted: 09 April 2020 Published: 13 May 2020

#### Citation:

Guzmán-Aguilar G, Carbajal-Navarro A, Sáenz-Romero C, Herrerías-Diego Y, López-Toledo L and Blanco-García A (2020) Abies religiosa Seedling Limitations for Passive Restoration Practices at the Monarch Butterfly Biosphere Reserve in Mexico. Front. Ecol. Evol. 8:115. doi: 10.3389/fevo.2020.00115 To recover the structure and functionality of a deforested ecosystem, two strategies of ecological restoration are considered: active restoration, which eliminates the disturbance agents and implements strategies to accelerate site recovery, and passive restoration, which eliminates disturbance agents, allowing natural regeneration to occur. Prior to choosing passive restoration, a field evaluation of the potential for natural regeneration is important. In this context, seedling and sapling density as well as patterns of recruitment and survival are appropriate indicators of restoration potential. In the present study, we deduced the potential of sacred fir (Abies religiosa) forest of the Monarch Butterfly Biosphere Reserve to recover by natural regeneration through seedling and sapling density and mortality, since A. religiosa is the dominant tree species in wintering sites of monarch butterfly. In 2015, we evaluated seedling density in 53 sites along an elevational gradient (3050–3550 m above sea level; m a.s.l.). There was a higher density of seedlings and saplings established in canopy gaps, compared to sites under dense forest canopy. Seedling recruitment was higher in sites at intermediate elevations (3050 to 3300 m a.s.l.) than in those at higher elevations. In a second survey, we studied A. religiosa seedling mortality over the dry season of 2016 to identify the environmental variables that cause the high seedling mortality and very low recruitment. Recently emerged seedling mortality was 49.2% at the end of the dry season (June 2016). The highest monthly mortality (14.3%) was recorded in April, a dry and warm month with the lowest values of moss thickness and soil moisture. We found no negative effects of moss layer on seedling mortality; indeed, moss appears to slow soil moisture reduction at the critical end of the warm and dry season. Soil and moss moisture values in April seem to be a critical factor for A. religiosa seedling recruitment, and we expect this condition will deteriorate under projected climatic change scenarios. Thus, the potential of MBBR A. religiosa forest to recover by passive restoration is highly constrained and will require management actions to achieve successful restoration outcomes.

Keywords: Abies religiosa, soil moisture, natural regeneration, seedlings, elevational gradient

# INTRODUCTION

fevo-08-00115 May 11, 2020 Time: 19:23 # 2

The sacred fir (Abies religiosa) is a conifer native to Mexico. It distributes in the mountainous areas (2100 to 3600 m a.s.l.) in central Mexico, presenting monospecific forests between 3000 and 3300 m a.s.l. (Rzedowski, 2006). These forests occur in locations that present very specific geographical, climatic and ecological conditions (Pineda-López et al., 2013), particularly on steep, humid and shaded slopes. One of the most emblematic Abies religiosa forests in Mexico is found in the Monarch Butterfly Biosphere Reserve (MBBR), which acts as refuge and habitat for the monarch butterfly (Danaus plexipus L.) that evades the winter conditions of Canada and the United States by annually migrating south to a few mountainsides in central Mexico (Honey-Rosés et al., 2018). The MBBR encompasses 56,259 ha and was designated as a UNESCO World Heritage Site.

Currently, the Reserve is threatened by various political, social and economic issues that lead to environmental degradation associated with logging activities, expansion of the agricultural frontier and overexploitation and inadequate management of natural resources. This is despite the fact that the Monarch Butterfly Reserve receives a considerable amount of financial resources from national and international organizations to carry out reforestation programs (Honey-Rosés et al., 2011). Between 2002 and 2010, the region received United States \$9.2 million for reforestation programs (SEMARNAT, 2011).

To recover the structure and function of a deforested ecosystem, two strategies of ecological restoration are generally considered: active restoration, which eliminates the agents of disturbance and implements strategies to accelerate site recovery (e.g., tree planting and soil conservation practices), and passive restoration, which eliminates agents of disturbance in the area, relying on natural regeneration (Holl and Aide, 2011).

The process of natural regeneration is one of the most important issues in passive restoration, and can be seen as a continuous cycle of ecological processes, such as the development of seeds and their subsequent dispersal and predation, or the germination and establishment of seedlings, among others. The long-term success and dominance of tree species depends on these ecological processes (Pérez-López et al., 2013). Natural regeneration can be an appropriate option for passive restoration of forests (Pensado-Fernández et al., 2014); however, understanding the relationship between the structure and dynamics of canopy vegetation with seedling density, are crucial for predicting the likely effectiveness of passive restoration strategies (Grime and Hillier, 2000).

Natural regeneration rates are highly variable depending on the ecosystem, landscape context, land use history and passive restoration may not always be successful (Lara-González et al., 2009), taking longer to reach the goals established for the restoration of a site than an active restoration. Such delays in regeneration can sometimes be perceived as failures of passive restoration. Lands subject to passive restoration can be seen in developing countries as abandoned or unused land, which may encourage local people to use these areas for livestock or other activities. An advantage is that passive restoration is generally perceived as a low-cost alternative, although in general it has costs that are often not taken into account such as the purchase of material (fences or barriers) to isolate the ground from disturbance agents and payments for site surveillance (Zahawi et al., 2014). It has the potential to achieve similar levels of biodiversity and environmental services as an active restoration; however, it is only feasible in certain places where the disturbance was not so intense, natural communities are resilient and are far from human communities (Holl, 1999; Zahawi and Augspurger, 1999; Muñiz-Castro et al., 2006; Suding and Hobbs, 2009; Aide et al., 2010; Holl and Aide, 2011).

Due to the prevailing shade conditions throughout the understory, the rates of natural regeneration in temperate forests are reduced or even null in some cases. Temperate forests are renewed by the dynamic of gap formation (which can have both natural and artificial causes), where natural regeneration processes are increased considerably. Natural regeneration in situ, compared to traditional forest plantations, is an appropriate option for ecological rehabilitation on degraded land, especially if protected from livestock (Lara-González et al., 2009; Sánchez-Velásquez et al., 2016).

Populations of tree species differ genetically along elevational gradients, as a response to the selection pressure of temperature and precipitation gradients (Rehfeldt, 1991; Ortiz-Bibian et al., 2017). This makes it advisable to delineate elevational zonings to guide seed and seedling movement in reforestation programs. Castellanos-Acuña et al. (2014) reported a significant morphological differentiation among populations of A. religiosa along an elevational gradient: low-altitude populations have shorter needles and longer cones than high-altitude and these might have important consequences for seed production and seedling quality.

Scientific literature on A. religiosa shade tolerance is at times contradictory; according to Rzedowski (1978), the A. religiosa is a shade tolerant species and canopy gaps contribute to the regeneration of A. religiosa in the Cofre de Perote National Park, in Veracruz, Mexico, and seedling density is considerably greater in gaps than in the understory (Lara-González et al., 2009). However, some authors consider that the species can regenerate naturally in both clearings and understory (Narakawa and Yamamoto, 2001; Sugita and Tani, 2001; Mori and Takeda, 2002), while González et al. (1991) state that the species grows in open places in smaller proportions than in the understory.

Honey-Rosés et al. (2018) studied the drivers of forest cover both inside and outside the MBBR using a combination of remote sensing imagery and field-collected data. They found an increase in forest cover of 5,673 ha occurred between 1986 and 2012: 71% of this recovery was attributed to natural regeneration processes, while active restoration efforts only contributed 3.8%, raising questions about the effectiveness of active restoration. The rest (25%) was attributed to a combination of both techniques. The authors conclude that due to the high potential for natural regeneration in the reserve, management efforts should focus on passive restoration activities instead of investing in active restoration (Honey-Rosés et al., 2018).

While many forest managers may be attracted to the idea of supporting natural regrowth via passive restoration,

various biophysical conditions may impede the successful recruitment of young seedlings in areas that are unhospitable to forest regrowth. For example, Manzanilla (1974) suggests a negative interaction of the moss layer thickness with A. religiosa seedling mortality; since thicker layers will generate a physical barrier that is responsible for the absence of natural regeneration. Local forest technicians and landowners of the MBBR support this assertion.

The objective of this study was to study factors that may affect natural regeneration of A. religiosa seedlings at MBBR, an essential consideration for implementing passive ecological restoration. We studied regeneration capacity through seedling density in response to elevation, canopy closure, and other abiotic factors such moss layer thickness and soil moisture. This research is intended to guide decision-making regarding the implementation of adequate restoration and conservation strategies at the Monarch butterfly overwintering sites in the MBBR.

## MATERIALS AND METHODS

#### Study Site

This study was carried out at Ejido de La Mesa, in the municipality of San José Del Rincon, Estado de México (19◦ 340 , 35.7<sup>00</sup> N and 100◦ 140 , 30.2<sup>00</sup> W), in the central-western part of the Mexican Transvolcanic-Belt.

In September and October of 2015, natural regeneration of A. religiosa seedlings was monitored along an elevational gradient (3050 to 3550 m a.s.l) in the MBBR. The transect range was classified into two different elevational bands: intermediate (3050–3300 m a.s.l.) and upper (3301–3550 m a.s.l.), according to the elevational zoning of Castellanos-Acuña et al. (2014). It was not possible to measure seedling density at the lower elevational band (2800–3050 m a.s.l.), since this is an area with a long history of impact by human settlement and agricultural and livestock activities, and the original A. religiosa trees at this low elevation remain only in small forest fragments.

At both elevational bands (intermediate and upper), we selected 53 sites with and without canopy gaps (canopy type): 25 sites in the intermediate band (11 under forest cover and 14 in gaps) and 28 sites in the upper band (9 under forest cover and 19 in gaps). The area for each gap was different (<400m<sup>2</sup> ): the diameter not less than 15 × 15, nor more than 23 × 23 m (resembling the size of an adult tree canopy). In closed canopy sites a 15 m diameter circle was used (**Supplementary Figure S1**).

The selected sites presented slopes of less than 22◦ (on steeper slopes the effect of the gap decreases due to inclination of the crowns of adjacent trees). Abundance, height and diameter of seedlings (0–2 mm root collar diameter) and saplings (<5 cm DBH) were measured throughout each site with canopy gaps and within the 15 m circle for sites without canopy gaps. Additional parameters measured in each site included: canopy cover, slope, elevation and gap diameter. Percent cover of rocks, shrubs, herbaceous plants, mosses and bare soil was recorded in three 2 × 2 m square quadrats per site.

#### Seedling Mortality Measurements

We conducted an additional survey in the same area, evaluating the mortality of naturally regenerated A. religiosa seedlings throughout the 2016 dry season (from February to early June). Thirty quadrats of 4 m<sup>2</sup> (2 × 2 m) were delimited and distributed in the same elevational bands (15 quadrats each in the intermediate and upper bands). The quadrats were always located beneath forest canopy (>60% of tree cover) with a minimum distance of 50 m apart to avoid spatial autocorrelation. In each quadrat, all of the recently emerged seedlings that appeared to be less than 1-year-old (older seedlings show a lignified stem) were individually labeled. Each month, we recorded alive seedlings and carefully collected dead seedlings for dry weight measurement.

In quadrats with presence of moss, the moss layer thickness was measured monthly in three adjacent sites per quadrat (to avoid disturbing seedlings and moss layer inside the quadrat). We collected small samples of moss and soil adjacent from each quadrat, weighed in situ, and packed fresh in zip sealed plastic bags for subsequent drying in the laboratory to estimate the relationship of volumetric moisture content.

Circular plots of 0.1 ha (17.8 m radius) were established to count adult trees above each quadrat and we measured height and diameter at breast height (DBH) of each tree recorded and grouped in three categories (<25 cm, 25–45 cm, >45 cm) according to Pineda-López et al. (2013) and Manzanilla (1974). Canopy cover was estimated from hemispheric photographs taken with Winscanopy (Regent Instruments Inc.), (Guay, 2014).

The samples of moss and soil were dried in an oven at 70◦C for 5 days and weighed. Dead collected seedlings were divided into their root and aerial parts, which were dried in an oven for measurement of dry weight. The biomass allocation estimates were done to asses if dead seedlings fail to reach the soil beneath the moss layer. When dead seedlings were in moss, we recorded if the roots penetrated the moss and made it into the soil below since local forest technicians have claimed this is the main cause of seedling mortality.

#### Data Analysis

Seedling and sapling density was analyzed using a generalized linear model with a Poisson distribution. The independent variables were elevational band (intermediate or upper), canopy (gap or forest cover) and the interactions among these factors. Linear regression or Spearman rank correlation tests were applied to assess the relationship between the various environmental variables and seedling density.

To determine temporal variation of A. religiosa seedling mortality during 2016 dry season, we performed a repeated measures ANOVA, with a post hoc Tukey paired test (the dependent variable was the number of surviving seedlings per month while elevational band was the independent variable). The temporal comparison was conducted with paired Wilcoxon and Kruskal-Wallis tests.

Moisture content of moss and soil was estimated through the formula of gravimetric moisture: - W% = Ma Ms × 100 , where Ma is the weight of water lost following drying and Ms is the fresh weight of soil or moss (Universidad Nacional de Córdoba, 1993),

representing the percentage or weight of water in 1 g of soil or moss.

The canopy photographs taken were analyzed with the Winscanopy (Regent Instruments Inc.) (Guay, 2014) and percentage of canopy cover was estimated.

In addition, linear and quadratic regressions were performed between seedling mortality and moss layer thickness, moss gravimetric moisture content and soil gravimetric moisture content, to identify a threshold that promoted major seedling mortality.

Finally, we applied a Cox proportional hazards model to analyze the influence of environmental and forest structural variables on seedling survival time. The independent variables were soil organic matter content, moss cover, maximum moisture content of moss, tree density, maximum and minimum thickness of moss layer, minimum moisture content of moss and minimum soil moisture content. A few seedlings disappeared from quadrats during the study. These may have been eaten by herbivores instead of dying but we included these individuals in the analyses. All statistical analyses were performed with the packages R 3.1.3 and JMP 8.0 SAS Institute Inc.

#### RESULTS

#### Seedling and Sapling Density

There was a higher density of seedlings at the intermediate elevational band compared to upper band. Most of the individuals recorded in the intermediate band beneath the forest were seedlings (68%) while canopy gaps harbored a higher proportion of saplings than sites without canopy gaps (**Figure 1**). In the upper zone, in addition to the lower overall density, only 3% of individuals were seedlings, with similar proportions in each canopy type. However, there was a higher density of saplings in gaps. The results of the generalized linear model were significant for both parameters: elevational band (x <sup>2</sup> = 352.7, df = 10, p < 0.001) and canopy type (x <sup>2</sup> = 198.4, df = 10, p < 0.001). Saplings also showed significant differences: elevational band (x <sup>2</sup> = 483, df = 10, p < 0.001) and canopy type (x <sup>2</sup> = 243, df = 10, p < 0.001). However, there was no significant effect of interactions between these independent variables.

#### Understory Conditions

The most common companion species at the sites were Acaena elongata (a shrub of 0.3 to 1 m in height), Alchemilla procumbens (a creeping grass of up to 30 cm in height), and Roldana angulifolia (a shrub of 1 to 2.5 m in height).

Despite the small difference in tree coverage between the gap and forest canopy types (**Table 1**), statistically significant differences were found. The results suggest that A. religiosa seedlings experience suitable conditions in the gaps (intermediate levels of light) for initiation of the natural regeneration process.

The gaps presented a higher shrub and herb coverage than sites without gap, while there was higher coverage of moss beneath the forest canopy. In all four strata, significant differences were present between sites with and without gaps (**Table 1**). No differences in rock and bare soil coverage were observed between gaps and without gaps sites.

A non-parametric correlation of Spearman ranks was conducted between tree canopy openness and seedling density, revealing a weak relationship (p = 0.009, r<sup>s</sup> = 0.351, n = 53).

No significant differences were found when correlating moss cover with seedling density using the Spearman rank coefficient. However, moss cover was negatively related to other understory components (rocks, bare soil, and shrubs), and positively related to herbs (**Table 2**).

#### Seedling Mortality Survey

Six-hundred sixty-one A. religiosa seedlings were marked and monitored in 30 quadrats. In the upper elevational band, 15 quadrats were established and 378 seedlings monitored. In the intermediate elevational band, 15 quadrats were established and 283 seedlings monitored. We found that mortality increased during dry season reaching 48% (in either elevational band) when

TABLE 1 | Mean coverage (%) of different forest strata per canopy type (with and without gap), and significance of the difference between canopy types in each case.


Tree values came from hemispheric photos while the other data came from field collection at each site.

TABLE 2 | Correlations between components of ground cover and seedling density.


The significance (n = 53) of the test is shown above the diagonal, while the correlation coefficient is shown below.

the rainy season began. In either elevational band the highest mortality occurred in April and the lowest in June (**Figure 2**).

### Factors Associated With Seedling Mortality

No significant differences were found between the two elevational bands (**Figure 2**). A low proportion of the seedlings disappeared in the month of April and May and these individuals apparently

obvious cause of death and were used in biomass allocation estimate, and "lost" refers to individuals where the entire plant was gone; letters show significant differences as result of repeated measures ANOVA.

had been consumed by herbivores or had decomposed during the interval between the two monitoring periods (**Figure 3**).

#### Biomass Allocation Estimate

The dead seedlings presented greater average aboveground biomass compared to belowground biomass allocation, 65 and 35% in the upper band, and 68 vs. 32% at the intermediate band. The ANOVA shows significant differences between this biomass allocation, but no differences were found in this respect between elevational bands. We also analyzed the correlation between average aerial biomass and canopy cover of each site but there is not a significant relation between these variables.

# Moss Layer Thickness and Moisture Content

The average initial thickness of the moss layer was 3.2 cm, and this decreased to a minimum value in April (a warm and dry month), at an average of 2.1 cm, before recovering quickly as a consequence of the early rains in June. April was the only statistically different month revealed in the repeated measures ANOVA. February and June showed the highest average values of thickness, with 3.1 and 2.9 cm, respectively (**Figure 4**). There were no statistically significant differences in moss layer thickness between elevational bands for any month.

### Gravimetric Moisture Content of Moss and Soil

The moss layer showed a higher water retention capacity, containing up to 2.6 g of water/g of moss, as well as rapid dehydration and rehydration with precipitation. For both substrates (moss and soil), the lowest water content was observed in April, with statistically significant differences observed compared to the other months. The lowest thickness of the moss layer and the highest seedling mortality was also in

April. In contrast, the highest moisture content was observed in June for both substrates because that month had the highest rainfall (**Figure 5**).

The gravimetric moisture content of the moss showed a positive relationship with canopy cover only in the wettest month (June). The relationship of seedling mortality with monthly moisture content (of moss or soil) is statistically significant and shows clearly that lower humidity values are associated with higher monthly mortality rates (**Figure 6**). A general trend is evident: a mortality rate greater than 4% occurs when a critical threshold of 1 g of water/g of moss, or 0.7 g of water/g of soil, is reached during the dry season.

#### Proportional Risk Analysis

The Cox regression or proportional hazard analysis shows that three parameters had an effect on A. religiosa seedling mortality: soil organic matter content increases 29% the risk of seedling mortality, and tree density surrounding the sites (1%) and moss cover has a significant but weak effect in seedling mortality (**Table 3**).

#### Forest Structure

There were many differences in adult trees surrounding 4 m<sup>2</sup> quadrats between the two elevational bands. Sites in the intermediate band showed an average tree density of 904/ha vs. 529 trees/ha in the upper band (f = 11.1, df = 1, p < 0.002). In the upper band, average tree height was 30.2 vs. 18.2 m in the intermediate band (f = 26.0, df = 1, p < 0.01). DBH was significantly higher in the upper band (f = 13.0, df = 1, p < 0.01), while no significant differences were recorded in the canopy cover (79.5% in intermediate vs. 76.8% in the upper band).

The tree diameter distribution showed that the average density of trees with dbh less than 25 cm is clearly higher in the intermediate band compared to the upper band, (f = 17.2, df = 1, p < 0.001). In the following two categories of DBH (25–45 cm and >45.1 cm), the density decreased and was significantly higher in the upper band (f = 9.1, df = 1, p < 0.001) and (f = 12.4, df = 1, p < 0.001) (**Figure 7**).

#### DISCUSSION

In the present study, several variables were considered to have affected the density of seedlings and saplings of A. religiosa, one of which is elevation. Ortiz-Bibian et al. (2019) found that populations of this species in the central part of their elevational distribution (intermediate zone) exhibit a higher number of viable seeds and greater germination capacity. This pattern could explain the larger number of seedlings and juveniles we recorded in the intermediate band.

FIGURE 5 | Gravimetric moisture content (GMC) of soil (right) and moss (left) over time; (different letters indicate statistical differences among the months as revealed by the repeated measures ANOVA).

FIGURE 6 | Quadratic regression of seedling mortality against: (left) moss moisture content, (right) soil moisture content. (Vertical black line indicates what appears to be a critical humidity threshold in relation to seedling mortality).



Likelihood ratio test = 49, df 6, p < 0.001, n = 662, number of events = 233.

The observed higher seedling and sapling density in gaps compared to forest is similar to the results of Lara-González et al. (2009), who found that regeneration of A. religiosa is favored in sites with greater canopy openness. Manzanilla (1974) reports that the regeneration of A. religiosa occurs clumped in sites with high availability of sunlight.

Regarding the size of the seedlings and saplings, diameter showed a similar pattern (most individuals belonging to smaller categories). However, there were notable differences between elevational bands. At the intermediate band most of the individuals recorded were seedlings (68% in forest and 45% in gaps); while for the upper band, only 3% were seedlings and 97% saplings. This suggests recruitment of seedlings to saplings is limited in the intermediate elevation band. On the other hand, seed limitation, either from a lack of seed production, germination, or early post-germination survival might be occurring at higher elevations. Likewise, in gaps at the intermediate band, a vigorous germination process could be underway, which would ensure that suitable plants are established for the regeneration of the forest.

In relation to canopy type, individuals in gaps had greater size and gaps had the highest percentage of shrub and herbaceous plant coverage. This vegetation could therefore play a "nurse plant" role, which might act to favor the establishment and growth of A. religiosa. Sánchez-Velázquez et al. (2011), and also Blanco-García et al. (2011) measured the effect of nurse plants such as Baccharis conferta and Lupinus elegans in an A. religiosa reforestation trial and documented lower mortality and higher growth of Abies when growing under the canopy of these shrubs.

Bautista (2013) and Lara-González et al. (2009) reported that morphological variables (number and length of lateral buds) and natural regeneration (seedling density) of A. religiosa are favored with increased canopy openness, as confirmed by the present study. Even when canopy cover between gaps and forests was slightly different (92 vs. 99%, respectively) these differences could determine the suitability of the light conditions for Abies religiosa.

The Abies religiosa forests of Mexico are relatively dense because of their closed canopies; the light that comes to the ground is low and the understory is scarce, so the existence of gaps is not common even though its contribution to forest regeneration is very important. It has been observed that in open areas the regeneration is more successful under the canopy of some shrubs that act as nurse plants facilitating fir regeneration (Lara-González et al., 2009).

#### Seedling Mortality Survey

The seedling mortality recorded in our survey was lower compared to results reported for Abies pinsapo (Arista, 1993). In the latter species, the possible factors contributing to seedling mortality were high light intensity, low humidity and competition with herbaceous plants. It was noted that seedlings died quickly as soon as spring and summer began, possibly as a result of water stress since they were exposed to full sunlight. The following year, the same author (Arista, 1994) reported contrasting data for another population of the same species, where seedlings less than 1-year-old presented 45% survival in understory and 82% in an open field, while older seedlings presented survival of 75% in the forest and 83% in an open field. That study indicates low humidity, light and extreme temperatures as the main factors that

contribute to high mortality. Moreover, where humidity was not a limiting factor, mortality was attributed to a possible fungal attack or lack of mycorrhizae.

Ángeles-Cervantes and López Mata (2009) investigated mortality in a cohort of A. religiosa seedlings in patches affected and unaffected by fires, and found that an important factor increasing Abies seedling mortality is desiccation. This is attributable to the layer of moss and accumulated litter, which prevents the root from reaching and penetrating the mineral soil beneath. This could be one of the factors by which sites with denser canopies (which have the highest percentage of moss) present less natural regeneration, since even though moss may constitute a suitable microsite for the germination of the A. religiosa seed, thicker layers of moss actually behave as a barrier for the longer-term persistence of the seedling. This concurs with comments made by nursery managers located within the MBBR, as well as forest technicians, who report that the presence of the moss strongly causes mortality of A. religiosa seedlings and that its partial removal might increase A. religiosa seedling survival.

#### Biomass Allocation Estimate

The difference in recorded biomass allocation may have an effect on seedling mortality, since the failure of the root system to supply water to the plant or to regenerate new roots will lead to a vicious circle of water stress and depletion of carbohydrates, which will cause a delay or a reduction of regrowth, or even the death of the plant, since desiccation of the roots is considered to have the most damaging effect on plant vitality (Brønnum, 2005).

#### Effect of Moss Layer

Our results suggest that the moss layer is not a primary limiting factor for A. religiosa seedling survival. Manzanilla (1974) found that in the A. religiosa forests, the thick layer of moss is responsible for the low natural regeneration, since it acts as a mechanical barrier reaching up to 30 cm in thickness in an understory with abundant vegetation that prevents the seedling root from reaching and anchoring to the mineral soil. Our study did not find moss layers as thick as those reported in Manzanilla (1974) and only 3.8% of dead seedlings roots failed to penetrate the moss layer in our quadrats. Manzanilla (1974) suggests the hypothesis that the negative interaction of the moss with the seedlings will generate a physical barrier that is responsible for the absence of natural regeneration. Local forest technicians and land owners also support this assertion.

Similarly, there have been reports of positive and negative effects on germination and recruitment generated by the organic matter layer, since this layer usually reduces soil temperature and water evaporation, increasing moisture in the soil and promoting better conditions for germination. Nevertheless, it can generate an allelopathic inhibition, reduce the incidence of light or form a physical barrier to the penetration of the seedling roots (Dechoum et al., 2015).

In contrast to the potential negative effects of moss on seedling establishment described above, in our study sites, the moss seems to provide a suitable environment for seed germination and seedling establishment, which is very important for the dispersion capability and establishment of woody species (Dechoum et al., 2015). However, with the onset of the dry season of the year, the moss loses its moisture very quickly, causing thinning of the moss layer and a subsequent loss of soil moisture, leaving the seedlings more exposed to other agents that can potentially cause mortality, such as temperature, solar radiation, lack of environmental humidity. This effect of the humidity has been described by Chen et al. (2015) as influencing the richness and abundance of moss species during the transition from dry to wet periods, and its variation due to the differential tolerance of some species to this abiotic factor.

In our study quadrats, soil moisture seems to be the abiotic factor that most affects the mortality of A. religiosa seedlings both directly and indirectly. For the genus Abies, availability of water is very important at the seedling stage, since several species are extremely sensitive to a moisture deficit in the substrate. Indeed, it is considered the most important factor in the mortality of coniferous seedlings within the first 5 years of growth (Van der Salm et al., 2007; Rodríguez-Laguna et al., 2015). A clear example is the high sensitivity to stress due to desiccation reported for Abies prosera in a study conducted under controlled conditions (Brønnum, 2005). In addition, it is essential to consider the possible impact on ecosystems as a consequence of climate change (Ledo et al., 2015), which is modifying the patterns and frequency of the dry period and will likely have severe effects on the recruitment of seedlings in the forests.

Finally, it is possible that critical environmental thresholds (such as mortality greater than 4% per month with a reduction of 1 g of water/g of moss or 0.7 g of water/g of soil) would be lowered given projected climate change scenarios (Sáenz-Romero et al., 2012). Higher temperatures and lower precipitation could prevent the successful establishment of some tree species or may limit their establishment to favorable years only, ultimately changing the structure and functioning of the forest ecosystem (Von Arx et al., 2013).

#### CONCLUSION

Abies religiosa seedlings are more abundant at intermediate sites (3050 to 3300 m a.s.l.) than at upper (3301 to 3550 m a.s.l.) elevations, where poor establishment and recruitment of seedlings over the last 20 years have been observed. Additionally, canopy gaps play a positive and very important role in seedling recruitment, but a high proportion of seedling failure occurs at intermediate elevations, and the consequent lack of recruitment is an important issue that requires further research.

We found no evidence that the moss layer is responsible for seedling mortality; indeed, it constitutes an excellent moist microsite for seed establishment and germination, as well as protecting the bare soil from excessive moisture loss through evapotranspiration.

The most important factor increasing seedling mortality is soil moisture in the critical warm and dry month of April. This condition is likely to worsen under future scenarios of climatic change, affecting the regeneration of the Abies religiosa forest.

The upper elevational range of the MBBR is experiencing serious changes and active restoration might be needed to maintain forest cover and the ecosystem services it provides for inhabitants of the region, including the overwintering Monarch butterfly colonies.

#### REFERENCES

Aide, T. M., Ruiz-Jaén, M. C., and Grau, H. R. (2010). "What is the state of tropical montane cloud forest restoration?," in Tropical Montane Cloud Forests Science for Conservation and Management, eds L. A. Bruijnzeel, F. N. Scatena, and L. S. Hamilton (Cambridge: Cambridge University Press), 101–110.

Passive restoration practice in MBBR is favored by a high seed production and germination but constrained by a low seedling density (especially beneath closed canopies in upper elevation sites) and low recruitment (especially in intermediate elevation sites). Forest management activities might be needed to promote gap formation and improve seedling recruitment to sapling stage.

#### AUTHOR CONTRIBUTIONS

AB-G and CS-R conceived the research project. GG-A, AC-N, and AB-G carried out the field measurements and conducted the statistical analysis. LL-T and YH-D provided helpful comments during the development of the project and proposed and conducted specific statistical analyses. All of the co-authors revised and contributed to the manuscript. AB-G led the writing of the manuscript.

# FUNDING

Financial support was provided to AB-G by the UMSNH Coordinación de la Investigación Científica; the Basic Research CONACYT Fund (Ciencia Básica-2014-242985) and the UNAM Laboratorio Nacional de Análisis y Síntesis Ecológica (LANASE 2018-293701).

### ACKNOWLEDGMENTS

We thank Sr. Francisco Ramírez Cruz and Sra. Doña Petra Cruz-Cruz, for the facilities for site selection and seedling measurement at Ejido La Mesa. Without their help, this study would not have been possible. The MBBR staff helped to select sites for fieldwork and granted the license required to conduct the research. Thanks to Marcos Sandoval-Soto, Beatriz Guerrero-Carmona, Nancy Farías-Rivero, Miriam Linares, Jorge Herrera-Franco, Francisco Loera-Padilla, and other UMSNH students for assistance with taking measurements. We also thank two reviewers and an English proofreader for improving the quality of this manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo.2020.00115/ full#supplementary-material

FIGURE S1 | Two different canopy types: on the left, a gap; on the right a site without gap (beneath the forest).



in situ: el caso de los pinos y la rehabilitación en el parque nacional cofre de perote. Bot. Sci. 92, 617–622.


**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 JD, declared their involvement as co-editors in the Research Topic, and confirm the absence of any other ongoing collaboration.

Copyright © 2020 Guzmán-Aguilar, Carbajal-Navarro, Sáenz-Romero, Herrerías-Diego, López-Toledo and Blanco-García. 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.

# Evaluating the Migration Mortality Hypothesis Using Monarch Tagging Data

Orley R. Taylor Jr. <sup>1</sup> \*, John M. Pleasants <sup>2</sup> , Ralph Grundel <sup>3</sup> , Samuel D. Pecoraro<sup>3</sup> , James P. Lovett <sup>4</sup> and Ann Ryan<sup>4</sup>

*<sup>1</sup> Department of Ecology and Evolutionary Biology, University of Kansas, Lawrence, KS, United States, <sup>2</sup> Department of Ecology, Evolution and Organismal Biology, Iowa State University, Ames, IA, United States, <sup>3</sup> U.S. Geological Survey, Great Lakes Science Center, Chesterton, IN, United States, <sup>4</sup> Monarch Watch, Kansas Biological Survey, University of Kansas, Lawrence, KS, United States*

The decline in the eastern North American population of the monarch butterfly population since the late 1990s has been attributed to the loss of milkweed during the summer breeding season and the consequent reduction in the size of the summer population that migrates to central Mexico to overwinter (milkweed limitation hypothesis). However, in some studies the size of the summer population was not found to decline and was not correlated with the size of the overwintering population. The authors of these studies concluded that milkweed limitation could not explain the overwintering population decline. They hypothesized that increased mortality during fall migration was responsible (migration mortality hypothesis). We used data from the long-term monarch tagging program, managed by Monarch Watch, to examine three predictions of the migration mortality hypothesis: (1) that the summer population size is not correlated with the overwintering population size, (2) that migration success is the main determinant of overwintering population size, and (3) that migration success has declined over the last two decades. As an index of the summer population size, we used the number of wildcaught migrating individuals tagged in the U.S. Midwest from 1998 to 2015. As an index of migration success we used the recovery rate of Midwest tagged individuals in Mexico. With regard to the three predictions: (1) the number of tagged individuals in the Midwest, explained 74% of the variation in the size of the overwintering population. Other measures of summer population size were also correlated with overwintering population size. Thus, there is no disconnection between late summer and winter population sizes. (2) Migration success was not significantly correlated with overwintering population size, and (3) migration success did not decrease during this period. Migration success was correlated with the level of greenness of the area in the southern U.S. used for nectar by migrating butterflies. Thus, the main determinant of yearly variation in overwintering population size is summer population size with migration success being a minor determinant. Consequently, increasing milkweed habitat, which has the potential of increasing the summer monarch population, is the conservation measure that will have the greatest impact.

#### Edited by:

*Constanti Stefanescu, Granollers Museum of Natural Sciences, Spain*

#### Reviewed by:

*Naresh Neupane, Georgetown University, United States Tyson M. Wepprich, Oregon State University, United States*

> \*Correspondence: *Orley R. Taylor Jr. chip@ku.edu*

#### Specialty section:

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

> Received: *01 March 2019* Accepted: *27 July 2020* Published: *07 August 2020*

#### Citation:

*Taylor OR Jr, Pleasants JM, Grundel R, Pecoraro SD, Lovett JP and Ryan A (2020) Evaluating the Migration Mortality Hypothesis Using Monarch Tagging Data. Front. Ecol. Evol. 8:264. doi: 10.3389/fevo.2020.00264*

Keywords: monarch, butterfly, migration, milkweed, tagging, recovery rate, monarch decline

# INTRODUCTION

Since the late 1990s, the monarch butterfly, Danaus plexippus, population has declined significantly based on measurements made at the Mexican overwintering grounds (Brower et al., 2011; Semmens et al., 2016). Identifying the cause or causes of the decline is important in order to focus conservation measures appropriately. Two explanations for the decline in the size of the overwintering population dominate the literature. The first, known as the "milkweed limitation" hypothesis, posits that the decline in the number of milkweed host plants in the major summer breeding area in the Upper Midwest of the U.S. (**Figure 1**) has led to a reduction in the size of the migratory population (Pleasants et al., 2017). The second, known as the "migration mortality" hypothesis, posits that the resources and conditions during the fall migration have declined resulting in an increase in mortality during the migration and a decline in the overwintering population (Agrawal and Inamine, 2018).

The milkweed limitation hypothesis is supported by data showing that in the early 2000s the majority of monarch production came from common milkweed, Asclepias syriaca, in corn and soybean fields in the Midwest (Oberhauser et al., 2001) and that the abundance of those milkweeds declined precipitously due to glyphosate herbicide use in those fields (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Pleasants et al., 2017; Thogmartin et al., 2017a; Saunders et al., 2018). The loss of the milkweeds from corn and soybean fields began in the late 1990s with the adoption of glyphosate-tolerant crops. Milkweeds had been nearly eliminated from these fields by 2006 (Pleasants, 2017). During this period, an estimated 71% of the monarch production potential of milkweeds on the Midwest landscape was eliminated, amounting to 25 million hectares of agricultural habitat that no longer had milkweeds (Pleasants, 2017). The subsequent decrease in the availability of milkweed is thought to have limited the size of the summer breeding population. Support for this hypothesis comes from the pattern of decline in milkweed availability that parallels the decline in the size of the overwintering population (Pleasants et al., 2017). Further support comes from the strong correlation between yearly late summer Midwest monarch egg production and yearly overwintering population size (Pleasants and Oberhauser, 2013; Pleasants et al., 2017).

The migration mortality hypothesis was proposed to explain the results of studies that found a disconnection between monarch numbers measured during the summer and early fall and the size of the overwintering population, and no decline in the summer population in contrast to the decline in the overwintering population (Davis, 2012; Davis and Dyer, 2015; Ries et al., 2015a,b; Inamine et al., 2016; Agrawal and Inamine, 2018). Other studies found no correlation between the size of the migratory population passing through prominent peninsular points and the size of the overwintering population (Badgett and Davis, 2015; but see Crewe and McCracken, 2015). These observations led to the presumption that there had to be another explanation for the monarch overwintering population decline, and these authors proposed an alternative hypothesis.

encompassing the area from 40 to 50◦ latitude and 65 to 80◦ longitude (outlined in blue). What we are calling the Total Area is the Midwest and Northeast combined. The NDVI values (Saunders et al., 2019) come from the region that encompasses the area from 30 to 40◦ latitude and 90 to 105◦ longitude (outlined in green). The dark blue square indicates the location of the overwintering colonies. Butterflies were tagged in other sectors besides the Midwest and Northeast but those data are not included in this study.

The migration mortality hypothesis states that mortality during the fall migration has a significant effect on monarch numbers, accounts for much of the year to year variation in overwintering numbers, and is an important contributor to the long-term decline of the monarch population. The possible causes of migration mortality include degradation of habitat for nectar resources needed during migration, increased parasite load, and road kills (Agrawal and Inamine, 2018). Support for this hypothesis requires data showing that these potential mortality factors have increased during the period of monarch decline. However, those mortality factors that have been examined to date show no increase. Saunders et al. (2019) found that parasite load had not increased over time. They did find that the greenness of Texas and the surrounding region in the fall, thought to be an important nectaring area for migration success, explained some of the variation in overwintering population size but had not declined over time. Despite the lack of supporting evidence, the migration mortality hypothesis has continued to be posited as a possible explanation for the monarch decline (Agrawal, 2019; Popkin, 2020). This assertion has created some uncertainty with regard to the value of the extensive monarch conservation efforts focused on establishing more milkweed habitat (Thogmartin et al., 2017b).

Besides testing the migration mortality hypothesis indirectly by examining the changes in factors that might affect migration success, as done by Saunders et al. (2019), the hypothesis could be tested directly by examining changes in migration success over time. Here we examine migration success using the probability that a butterfly tagged during migration will be recovered in Mexico (recovery rate). Migrating monarch butterflies have been tagged since 1992 by >10,000 citizen scientists through the Monarch Watch<sup>1</sup> (MW) program. Tagging is conducted from August 1 to November 15 throughout the eastern monarch population range (**Supplementary Figure 1**). Tag recoveries are made by guides and residents in Mexico who search for tagged butterflies among those that have died beneath the colonies or along the trails in the oyamel fir forests. MW representatives visit the overwintering sites each season to acquire these recovered tags. From its inception to the present, this program has tagged over 1.8 million monarchs with more than 18,000 tags recovered in Mexico. In this paper, we test the migration mortality hypotheses using a subset of these data, the tagging and recovery data from 1998 to 2015 (about 1.4 million tagged individuals and about 14,000 tag recoveries in Mexico).

To determine whether there is support for the migration mortality hypothesis using the tagging data, we addressed the following questions:

(1) Is there is a disconnection between the size of the summer population and the size of the overwintering population as the migration mortality hypothesis presumes? To examine this question we used the number of tagged individuals as a measure of the summer population size. When comparing number tagged with overwintering population size, we have considered the total number tagged but have also subdivided the total tagged individuals into those tagged in the Midwest and those tagged in the Northeast (**Figure 1**). The Midwest has been identified as the major source of monarchs in the overwintering population (Wassenaar and Hobson, 1998; Flockhart et al., 2017). Butterflies from the Midwest use the Central Flyway during fall migration whereas those from the Northeast use the Atlantic or Eastern Flyway (Howard and Davis, 2009). We have analyzed migration success for the Midwest and Northeast separately because it is possible that migration mortality might have changed in different ways according to which flyway is used.


# METHODS

#### Tagging

Monarch tagging kits are issued by Monarch Watch to volunteers broadly distributed east of the Rocky Mountains in the U.S. and Canada each August-September (**Supplementary Figure 1**). The number of tags distributed to each participant or group is recorded. Tags are applied to migratory monarchs during a 3 months period from early August into November. The circular tags (diameter 9 mm) are applied to the discal cell on the underside of a hindwing, a location close to the center of lift and gravity of the butterfly. The mass of the tags (10 mg) is about a 2% of the mass of an average monarch and tags are unlikely to affect flight performance. Tagging data are sent to MW as a digital or hard copy. The date, location and identity of the person tagging each butterfly is logged into a database. Additional information gathered includes the sex of each butterfly, and starting in 2004, whether the butterfly was wild-caught or reared in captivity and released. Each tag bears an individual code and the codes of recovered monarchs are matched to the person and the data on returned data sheets.

As an index of summer population size, we used the number of individuals that were tagged north of 40◦ latitude north and east of 100◦ longitude west (**Figure 1**). This region includes the primary production area for monarchs (Flockhart et al., 2017). Although some individuals south of 40◦ latitude were tagged, our goal was to estimate the size of the summer population before significant migration mortality had occurred so these data were not included.

To determine whether tagging effort might have changed over the last two decades, we examined the number of tags distributed every year. The number of tags distributed could be

<sup>1</sup>monarchwatch.org

a measure of the level of interest and possible effort volunteers put into tagging.

The butterflies tagged each year include individuals that were caught and tagged while migrating and individuals reared to adulthood from eggs or larvae found on milkweeds and tagged and released. While both the number of wild-caught and reared individuals may reflect the size of the summer population, our analysis is limited to wild-caught monarchs since the data show that reared monarchs have a lower recovery rate than their wild counterparts (Steffy, 2015). Beginning in 2004 volunteers were asked to record whether the butterflies were wild-caught or reared. For butterflies tagged before 2004, we estimated the proportion of those tagged that were reared using the procedure described in **Supplementary Appendix 1** and adjusted the number tagged to approximate only wild-caught individuals. Although the MW tagging program began in 1992, the current tagging system including protocols and a program for purchasing recovered tags did not become well-established until 1998. Hence, the focus on data from 1998 to 2015. Because it may take several years before all of the recovered tags from a particular year are returned to MW, data beyond the year 2015 have not been included.

We have examined migration success of butterflies tagged in the Midwest and the Northeast separately. The region we are calling the Midwest encompasses the area from 40 to 50◦ latitude north and 80 to 100◦ longitude west (**Figure 1**). This region corresponds to the Midwest plus Northwest and most of North Central as defined by Flockhart et al. (2017) and the Midwest as defined by Agrawal and Inamine (2018) and the North Central plus mid-Central regions of Stenoien et al. (2015). The region we are calling the Northeast encompasses the area from 65 to 80◦ longitude north and 40 to 50◦ latitude west (**Figure 1**). This region corresponds to that defined as the Northeast by Flockhart et al. (2017) and Agrawal and Inamine (2018) and the Mideast plus Northeast regions defined by Stenoien et al. (2015). The Total area is the combination of the Midwest and Northeast regions.

Other measures of the size of the summer population exist so we examined the correlation between those measures and the number tagged. The number of tagged butterflies was compared with the Midwest NABA (North American Butterfly Association—www.naba.org) summer butterfly counts and an index of Midwest monarch egg production. NABA counts were obtained from Saunders et al. (2019). These are midsummer counts of adults that will produce the generation that migrates to Mexico. As an index of egg production, we used the average maximum number of eggs per stem for each year obtained from the Monarch Larva Monitoring Project (MLMP<sup>2</sup> ). These eggs will produce the adults that will migrate to Mexico. Because both NABA counts and eggs per stem data did not include sampling in agricultural fields, in comparing these data with number of butterflies tagged we only used data from 2006 to 2015 to avoid the sampling bias in NABA counts and eggs per stem for the period before 2006 (Pleasants, 2017; Pleasants et al., 2017; and see Discussion).

#### Tag Recovery

Monarchs usually begin arriving at the overwintering sites in the last days of October and conspicuous colonies form by mid-November (Monarch Watch, 2019). Tags are recovered from dead monarchs found beneath the colonies by guides and ejido (local community) members throughout the winter months. To reward their search efforts, MW representatives purchase the recovered tags from guides and residents in late winter each year. People with tags save them in the hope that they will be present when MW representatives arrive to buy tags. Some residents hold on to tags for many years. These delays in connecting sellers with buyers means that it may take 3–4 years before there is a nearly complete picture of recoveries for any given year. While MW representatives purchase tags from people living near most of the colonies open to the public, over 80% of the tags are obtained from the site of the largest colony, El Rosario, with the majority of the remainder obtained from Cerro Pelon and Sierra Chincua, all sites within the State of Michoacan, Mexico (**Figure 1**).

The recovery rate for any particular year is calculated as the number of tags recovered divided by the number of butterflies tagged. Note that this is not the same as mark-recapture; the number of untagged butterflies encountered while searching for tagged one is not counted. The recovery rate is the product of three factors, (1) the probability that a tagged butterfly will arrive in Mexico (migration success), (2) the probability that a tagged butterfly will die while in Mexico (overwinter mortality), and (3) the probability that someone will find that tagged dead butterfly (detection probability). Each of the probabilities that make up recovery rate may vary over the years for a variety of reasons. We are most interested in the variation in migration success over the years. To use recovery rate as a measure of migration success, we assumed that the annual variation in overwinter mortality and detection probability was random with respect to migration success and to year. Because the migration mortality hypothesis posits that migration success has declined substantially, such a trend should be apparent in the recovery rate despite random variation in overwinter mortality and detection probability.

The detection probability may vary due to seasonal changes as monarchs seek water or colonies shift in location which can result in off-site mortality where tags are not likely to be found. Other site issues that affect detection include the density of the understory, with denser cover limiting the ability to spot tags. There are also human factors involved in detection, such as the accessibility of the colonies to the searchers, the shifting population of searchers due to the turnover in guides, the overall number of searchers and the economic conditions that motivate the searchers. We have no quantitative or anecdotal information suggesting that any of these factors might have changed over time so we have assumed that variation in detection is random.

Overwinter mortality varies from year to year due to predation by birds and mice, and open canopies that contribute to greater exposure and mortality (Calvert et al., 1979; Glendinning et al., 1988; Brower, 1996; Brower et al., 2004). Mortality due to predation and exposure to average winter conditions are treated

<sup>2</sup>monarchlab.org/mlmp

as normal mortality in our analysis. We do not have direct measures of normal overwinter mortality but shrinkage in the size of the colonies over time, which may be related to mortality, appears to be similar among years (Vidal and Rendón-Salinas, 2014). So, we have assumed the annual variation in normal overwinter mortality to be random.

However, there were 3 years in the data set when the assumption of normal overwinter mortality was not met. Major winter storms resulted in mass mortality in 2002, 2004, and 2016. These storms occurred after the colony sizes had been measured and affected recovery rates for the 2001, 2003, and 2015 populations. Approximately 70% mortality was estimated for the 2001 and 2003 populations and 40% mortality for the 2015 population as a result of the winter storms (Brower et al., 2004, 2017; Taylor, 2004). Mass mortality events create a larger pool of dead butterflies from which recoveries are made. Consequently, the recovery rates were substantially higher for these years and cannot be compared to the recovery rates for years with normal mortality. We have excluded the data for these years from the analysis of yearly changes in recovery rate.

We tested the migration mortality hypothesis three ways using tag recovery rates. (1) We examined the correlation between recovery rate and the size of the overwintering population. This hypothesis posits that migration success rather than the size of the summer population will be correlated with overwintering numbers. (2) We examined the extent to which recovery rate could explain overwintering population size after accounting for the effect of summer population size. We tested the model that (overwintering hectares) = (summer population size) × (recovery rate). Taking the log of these variables results in log (overwintering hectares) = log (summer population size) + log (recovery rate). This approach allows the variables to be used in a multiple regression. (3) In addition, we examined the trend in recovery rate over the period from 1998 to 2014 to determine if there had been a decline in migration success that would account for the decline in the size of the overwintering population.

Since migration success depends on acquiring sufficient lipid stores for the journey to Mexico and overwintering, the areas in Oklahoma and Texas that provide nectar resources to migrating butterflies are important (Brower et al., 2015). Autumn greenness based on satellite imagery can be a proxy for nectar plant availability. Saunders et al. (2019) examined the NDVI (Normalized Difference Vegetation Index) for the region indicated in **Figure 1** for the period September 15–October 15 as a measure of nectar availability. They found a correlation between NDVI and overwintering population size after accounting for the role of the size of the summer population. We examined the correlation between NDVI for this region and recovery rate. NDVI values from 2000 to 2014 were obtained from Saunders et al. (2019).

All variables were log<sup>10</sup> transformed before statistical analyses. Relationships among variables were analyzed using regression procedures in the Data Analysis package of Excel<sup>3</sup> and JMP (SAS)<sup>4</sup> .

#### RESULTS

#### Number Tagged

#### Comparison With Overwintering Population Size

The first test of the migration mortality hypothesis is to examine whether the size of the summer population is correlated with the size of the overwintering population; the hypothesis presumes that it does not. We used the number of monarchs tagged within the summer breeding region as an indicator of late summer population size. The number of monarchs tagged in the Midwest and Northeast portions of the geographic sectors north of 40◦ latitude north and east of 100◦ longitude west (**Figure 1**) are shown in **Table 1** for each year from 1998 to 2015. To determine how well the number of tagged individuals explained the variation in overwintering numbers, we subdivided the analyses by Midwest, Northeast and Total area (Midwest plus Northeast). We compared regressions using the Total tagged and Midwest and Northeast tagged alone (**Supplementary Table 1**). Based on the lowest AIC value, log Total number tagged was the best predictor of overwintering hectares. However, log Midwest tagged was a close second. Because we wished to compare number tagged as a measure of population size to other population size measures that are Midwest-based, we focused on the number tagged in the Midwest. The number tagged in the Midwest alone explained 74% of the variation in overwintering hectares (**Supplementary Table 1** and **Figure 2**). The number tagged in the Northeast was somewhat less strongly correlated with overwintering hectares (**Table 2** and **Figure 2**). The number tagged in the Midwest was correlated with the number tagged in the Northeast (**Table 2**). The number of individuals tagged in the Midwest and Northeast both declined over the period from 1998 to 2015 as did the overwintering hectares (**Table 2**).

Two other measures of summer population size, Midwest monarch egg production and Midwest NABA butterfly counts, are shown in **Table 1**. The number tagged in the Midwest was highly correlated with Midwest late summer egg production and also correlated with midsummer NABA counts for the years from 2006 to 2015 (**Table 2**). NABA counts and eggs per stem were highly correlated with each other (**Table 2**). Egg production and NABA counts were also correlated with the overwintering population size (**Table 2**). Thus, three independent measures of summer population size: number tagged, NABA counts, and eggs per stem, were correlated with each other and correlated with the size of the overwintering population. This result is contrary to the prediction of the migration mortality hypothesis.

#### Recovery Rate

#### Relation to Overwintering Population Size and Change Over Years

Another test of the migration mortality hypothesis is to examine whether migration success, as measured by tag recovery rate, is correlated with overwintering population size and whether it has declined over time. The numbers of tags recovered each year in the Midwest and Northeast and these two regions combined (Total) are shown in **Table 3**. The recovery rate for the Midwest was used in these analyses because this region has been shown to be the core area for monarch production (Flockhart et al.,

<sup>3</sup>Microsft.com/Microsoft/Excel

<sup>4</sup> JMP.com


TABLE 1 | The number of wild-caught tagged monarchs for all geographic sectors north from 40 to 50◦ latitude and east from 100 to 65◦ longitude and the numbers tagged for the Midwest and Northeast portions of that region (see Figure 1).

*Number tagged from 1998 to 2003 have been adjusted to estimate wild-caught only (*Supplementary Figure 2 *and* Supplementary Appendix 1*). An* \* *indicates a year with stormrelated mass mortality. Also included are yearly indices of the size of the overwintering population (Monarch Watch, 2019), the average number of monarch eggs per stem from MLMP data (Pleasants et al., 2017), and the NABA butterfly counts for the Midwest from Saunders et al. (2019).*

2015) and because the number tagged in the Midwest was a good predictor of overwinter population size. For the analysis of recovery rates, the mass mortality years, 2001, 2003, and 2015 were not included for reasons explained in Methods. Midwest recovery rate was not correlated with overwintering population size (**Table 2** and **Supplementary Table 1**).

Although the size of the late summer population explained a large amount of the annual variation in the size of the overwinter population, it is possible that some of the variation in overwinter population size is due to annual variation in migration success. To examine this possibility, we ran a multiple regression using Midwest tagging numbers and Midwest recovery rates (**Supplementary Table 1**). The equation for the regression of log MW Number Tagged and log MW Recovery Rate on log Overwintering Hectares (OW) was: log OW = −6.3127 + 1.6466 log MW Number Tagged + 0.2461 log MW Recovery Rate (F2, 12 = 15.5268, p = 0.0005, R 2 , adjusted = 0.6748). The standardized regression coefficients (src) were 0.8602 for log MW Number Tagged and 0.1772 for log MW Recovery Rate. The square of each standard regression coefficient, indicates the amount of variation explained by each variable. Src<sup>2</sup> was 0.74 (p = 0.0001) for log MW Number Tagged and 0.03 (p = 0.2739) for log MW Recovery Rate, indicating that log Number Tagged accounted for about 74% of the variation in log OW while log Recovery Rate accounted for about 3%.

There was no decline in the tag recovery rate over the period 1998–2014 for either the Midwest or the Northeast (**Table 2** and **Figure 3**). The recovery rates for the Midwest were correlated with the recovery rates for the Northeast (**Table 2**).

#### Correlates of the Recovery Rate

The recovery rates for the Midwest were correlated with yearly values for NDVI, an index of greenness of the region that provides nectar for migrating butterflies (**Table 2**). The year 2000, which had the lowest recovery rate, had the lowest NDVI value, and the year 2008, which had the highest recovery rate, had the highest NDVI value (**Table 3** and **Figure 3**).

We also examined whether NDVI itself, could explain any of the annual variation in overwintering hectares (**Supplementary Table 1**). The standardized regression coefficients (src) were 0.8917 for log MW Number Tagged and 0.1668 for log MW NDVI. Src<sup>2</sup> was 0.80 (p < 0.0001) for log MW Number Tagged and 0.03 (p = 0.16) for log MW NDVI, indicating that log Number Tagged accounted for about 80% of the variation in log OW while log NDVI accounted for about 3%. While NDVI is correlated with migration success, just like migration success itself, there was no decline in NDVI from 2000 to 2015 (**Table 2**).

The yearly recovery rates for butterflies from the Northeast were correlated with those from the Midwest (**Table 2**) but consistently lower (**Table 3** and **Figure 3**). The recovery rate for all the normal mortality years 1998–2014 combined was 0.94% for those tagged in the Midwest and 0.24% for those tagged

in the Northeast (**Supplementary Table 2**). Thus, the average recovery rate for the Midwest was 3.93 times greater than for the Northeast (**Supplementary Table 2**). For the mass mortality years 2001 and 2003, the ratio of Midwest to Northeast recovery rates were similar to that of all normal mortality years (4.18 and 3.58) but, for the mass mortality year 2015, the recovery rate for the Midwest was not that much greater than for the Northeast (1.19) (**Supplementary Table 2**). For normal and mass mortality years combined, the recovery rate for the Midwest was 3.65 times that for the Northeast (**Supplementary Table 2**). Thus, a tagged butterfly from the Midwest was about 4 times more likely to be recovered than a tagged butterfly from the Northeast. Of the butterflies recovered in Mexico from the region above 40◦ latitude, 89% came from the Midwest with 11% coming from the Northeast. These percentages contrast to those for number tagged with 69% from the Midwest and 31% from the Northeast.

# DISCUSSION

#### Number Tagged

#### Comparison of Number Tagged and Overwintering Population Size

The migration mortality hypothesis presumes that the size of the summer population in the U.S. Midwest is not correlated with the size of the overwintering population in Mexico. We found that the number of monarchs tagged each season in the Midwest was highly correlated with the size of the overwintering population as was the number tagged in the Northeast, although not as highly (**Table 2**). It could be argued that the number of individuals tagged is more a reflection of the size of the migratory population rather than the summer population. As such, number tagged may already have incorporated some degree of migratory failure and thus would be more likely to be correlated with overwinter hectares. We chose to look at number tagged above 40◦ latitude and east of 100◦ longitude because this is the prime summer breeding region for monarchs (Flockhart et al., 2017). Individuals tagged in this region are closer to their natal origin than those tagged below 40◦ latitude. Butterflies tagged above 40◦ latitude have traveled southwest for at most two 5 × 5 latitude/longitude sectors (**Figure 1**) before capture and have 4–6 5 × 5 sectors remaining in their journey. Although it is certainly the case that some percentage of individuals leaving from a particular natal origin point will not make it to a tagging point above 40◦ latitude, the majority of migration mortality has yet to occur. So, the number of butterflies available to be tagged above 40◦ latitude is similar to the summer population size.

The underlying premise of the migration mortality hypothesis is that there is a lack of correlation between summer and winter population sizes. Studies showing a lack of correlation used summer population estimates based on data sets including the annual butterfly counts by NABA and butterfly counts made in Ohio and in Illinois (Ries et al., 2015a; Inamine et al., 2016; Saunders et al., 2016). However, there are methodological problems with these summer butterfly counts that make them inaccurate measures of population size (Pleasants et al., 2016, 2017). Sampling is limited in geographic scope and time and focuses on the early rather than the late summer population. More significantly, no counts were made in corn and soybean fields in the late 1990s and early 2000s when monarchs and milkweeds were still present in those fields. Surveys made during that period therefore underestimated the actual size of the monarch population. Survey data from after that period do show a correlation with overwintering population size (Crewe et al., 2019; Saunders et al., 2019) as does egg production for the last two decades that incorporates information on milkweed abundance (Pleasants and Oberhauser, 2013; Pleasants et al., 2017). Because migrating butterflies come from all habitats, including corn and soybean fields, the sampling bias seen in butterfly counts in the late 1990s and early 2000s is not an issue.

In comparing eggs per stem and NABA counts with overwintering population size, we were restricted to the years since 2006 to remove the sampling bias for not surveying in agricultural fields. But in comparing number tagged with overwintering population size we can use data for years before

0.90 log NE tag). 95% confidence intervals shown.

#### TABLE 2 | Summary of correlations and probabilities.


*NUMBER TAGGED and OTHER 1998–2015 (N* = *18). RECOVERY RATE excludes the mass mortality years 2001, 2003, and 2015 (N* =*15). NABA counts and eggs per stem include only the years from 2006 to 2015 for reasons discussed in Methods (N* = *10). NDVI for years 2000–2015. MW, Midwest; NE, Northeast; OW, overwintering population.* \* *Indicates the correlation after accounting for the variation explained by number tagged in the Midwest.*

TABLE 3 | The number of recoveries and recovery rates for wild-caught butterflies for the total monarch range north from 40 to 50◦ latitude and east from 100 to 65◦ longitude and for the Midwest and Northeast (see Figure 1).


*Number recovered from 1998 to 2003 have been adjusted to estimate wild-caught only (*Supplementary Figure 2 *and* Supplementary Appendix 1*). Recovery rate is expressed as a percent. An* \* *indicates a year with storm-related mass mortality. The NDVI index of greenness for the region shown in Figure 1 is from Saunders et al. (2019).*

2006 because butterflies during migration have come from all possible habitats, including agricultural fields. For the years 2006–2015, eggs per stem and NABA counts of summer population size were highly correlated with Midwest number tagged counts (**Table 3**). All three of these measures of summer population size were correlated with overwintering hectares (**Table 2**). These results are counter to the foundational assertion of the migration mortality hypothesis that the size of the summer breeding population does not predict the size of the overwintering population (Ries et al., 2015a; Inamine et al., 2016; Agrawal and Inamine, 2018). The relationship between summer population numbers and overwintering hectares suggests that

FIGURE 3 | Tag recovery rate (percent) for the Midwest and the Northeast for different years (*N* = 15). The years 2001, 2003, and 2015 were excluded because a major winter mortality event in those years greatly increased the recovery rate. There is no significant trend in recovery rate. Log MW rec. rate = −31.84 + 0.016 year (*R* <sup>2</sup> = 0.11, *p* = 0.22), Log NE rec. rate = −56.33 + 0.028 year (*R* <sup>2</sup> = 0.10, *p* = 0.25). Note that the graphs have different scales for recovery rate; Northeast recovery rates are always lower.

factors affecting monarch breeding population development from March through October, rather than migration success, primarily determine overwintering population size. While physical factors, such as weather and spring and summer temperatures are important determinants of population growth each year (Saunders et al., 2016, 2018, 2019; Crewe et al., 2019), there is also strong evidence for the role of milkweed availability (Pleasants and Oberhauser, 2013; Flockhart et al., 2015; Pleasants, 2017; Pleasants et al., 2017).

Tagging data arguably can provide a better picture of the size of the migrating population than counts made at peninsula points where monarchs stop over during migration. Counts made at Cape May, New Jersey and Peninsula Point, Michigan were not correlated with the size of the overwintering population (Davis, 2012; Badgett and Davis, 2015), but Crewe and McCracken (2015) did find a positive correlation between counts and overwintering numbers for Long Point in Ontario. Possible reasons for the lack of correlation between stopover points and overwintering numbers include the issue of double counting when counts are made more than once a day or when monarchs remain at stopover points for multiple days due to weather. Other reasons are discussed in Pleasants et al. (2016).

While we used numbers of butterflies tagged as an indicator of population size, we recognize that it is not a perfect representation of the summer population size. The tagging effort is influenced by weather events that limit tagging and the times when people are available to tag. In addition, there are few people tagging west of 95◦ longitude an area known to produce a substantial number of monarchs most years (**Supplementary Figure 1**). High populations are likely to be underestimated by tagging as well since taggers often run out of tags. It is also the case that many taggers only tag during the beginning of the migration, leaving the tail of the migration underrepresented in the tagging records. There was a 27% decline in the number of tags distributed from 1998 to 2015 (**Table 1**) (R <sup>2</sup> = 0.33, p = 0.013). This decline may represent a general loss of interest in tagging, or reduced expectations because of low population size that discouraged tagging. The proportion of tags that are affixed to butterflies reared in captivity has increased (**Supplementary Figure 2**) possibly reducing the effort to tag wild butterflies. In spite of these limitations, the fact that this index of population size is highly correlated with other measures of summer population size and overwinter population size lends it credibility.

There were 3 years where the number of monarchs tagged in the Midwest was outside the 95% confidence interval in the regression of number tagged vs. overwintering hectares (**Figure 2**). In the year 2000 Midwest tagging overestimated the size of the overwintering population (**Figure 2**). The recovery rate for 2000 was the lowest seen (**Table 3** and **Figure 3**) and corresponded to a year when NDVI values were the lowest (**Table 3**). Thus, in this year, drier than normal conditions in the southern region may have significantly affected migration success and reduced the number of individuals arriving in Mexico relative to what was expected from tagging numbers. The number tagged in the year 2014 also overestimated the size of the overwintering population (**Figure 2**). Recovery rate and NDVI were not unusual for that year (**Table 3**). Migration was later than normal in 2014 (Journey North<sup>5</sup> ). Late migrations are associated with low recovery rates (Taylor et al., 2019). It is also possible that there was an increased effort in tagging following the report in 2013 of the lowest overwintering population size ever recorded. In 1998, the number tagged underestimated the size of the overwintering population (**Figure 2**). The recovery rate was not particularly high that year so higher than normal migration success is probably not the explanation. The number of tags applied was the fifth lowest in the record (**Table 1**),

<sup>5</sup> journeynorth.org

suggesting that tagging simply underrepresented the size of the migratory population.

In the regression of number tagged in the Northeast vs. overwintering hectares, the year 2012 was an outlier (**Figure 2**). The number of butterflies tagged in the Northeast in 2012 greatly overestimated the actual overwintering population size. There was an early spring in 2012 (Ault et al., 2013) and monarchs colonized the Northeast earlier than normal (Journey North<sup>5</sup> ). The resulting population in the Northeast, as represented by the monarchs per hour count at Cape May, New Jersey, was the third highest in the 27 years of that program (Cape May Monarch Monitoring Project<sup>6</sup> ). On the other hand, extremely high temperatures in the Midwest, with drought conditions in much of the area (Taylor et al., unpublished data), reduced the 2012 migration in the Midwest and the number tagged in the Midwest to the second lowest total in the record (**Table 1**).

The number tagged in the Midwest from 1998 to 2015 was correlated with the number tagged in the Northeast with an r value of 0.64 (**Table 2**). The Midwest and Northeast populations are partially but not completely independent. There is some evidence that the Northeast is colonized by monarchs that were born in the first generation in the Midwest (Miller et al., 2011). But there is also evidence that some of the monarchs colonizing the Northeast have come directly from the Texas generation (Journey North, 2019).

#### Recovery Rate Relation to Overwintering Population Size and Change Over Years

Migration success, as measured by recovery rate, was not correlated with overwintering population size (**Table 2**). Nor did recovery rates decline from 1998 to 2014 (**Table 2**, **Figure 3**) as would be expected if migration success was driving the decline in the overwintering population. That said, migration success may play a small role in determining the annual variation overwintering population size. After most of the variation in overwintering population size was accounted for by summer population size, migration success accounted for an additional 3% of the variation. In addition to migration success, recovery rate is a function of overwinter mortality and detection probability. We have no data on the annual variation in those two factors. High levels of variation in those two factors could swamp out changes in migration success. However, the migration mortality hypothesis predicts a substantial decline in migration success which should be detectable despite variation in the other two factors. The fact that there was a significant relationship between recovery rate and NDVI, a factor suspected of affecting migration success, also indicates that variation in these other two factors are insufficient to undermine the use of recovery rate as a measure of migration success..

#### Other Correlates of the Recovery Rate

Although recovery rates did not decline from 1998 to 2014, they did vary over those years (**Figure 3**). This variation may be due to a variety of causes, some associated with the conditions during the migration, such as parasite load (Bartel et al., 2011), road kills (Kantola et al., 2019) and nectar availability. However, Saunders et al. (2019) did not find a significant correlation between Ophryocystis elektroscirrha (OE) parasite load and overwintering population size from 2004 to 2015. They also did not find a decline over time in parasite load, suggesting that this factor was not directly involved in the monarch population decline. They did find that the level of greenness along the southern fall migration corridor was correlated with the size of the overwintering population size, after accounting for the role of summer population size. However, they did not find that greenness had decreased over time.

We found that the greenness index, NDVI, was positively correlated with migration success. The lowest NDVI value and the lowest recovery rate was for the year 2000 (**Table 3** and **Figure 3**) and was associated with drought in Texas (drought.gov<sup>7</sup> ). The highest recovery rate observed was in 2008 which had the highest NDVI value (**Table 3** and **Figure 3**). The year 2011, when there was also a drought in Texas (Brower et al., 2015), had the second lowest NDVI and the third lowest recovery rate (**Table 3** and **Figure 3**). The fact that recovery rate is correlated with NDVI, a proxy for nectar availability, which we know to be important for monarch migration success (Brower et al., 2015), lends credibility to the use of recovery rate as a measure of migration success.

In addition to normal overwinter mortality caused by predators, broken wings, exposure or weather conditions, death may also result from inadequate lipid stores obtained during migration (Brower et al., 2015) or neogregarine parasite load that butterflies bring with them (Bartel et al., 2011; Altizer et al., 2015). We considered death due to lipid shortage as add-on mortality. We have to consider the possibility that years with poor migration success may also have increased add-on mortality if butterflies arrive in Mexico with reduced lipid supplies. The recovery rate is based on the product of migration success and overwinter mortality, including add-on mortality. What happens if low migration success is always coupled with high add-on mortality? In a poor migration year, lower migration success will have the effect of decreasing the recovery rate, but add-on mortality will have the effect of increasing the recovery rate. Because of these opposing effects on recovery rate, the question is whether we can compare recovery rates for different years and interpret those differences as differences in migration success. Lipid analysis by Brower et al. (2015) can provide some insight into this question. In 2011, which was a drought year in Texas, they found lower lipid levels in butterflies in Texas than in non-drought years. Surprisingly however, they found that November lipid levels of butterflies at the overwintering sites in Mexico were not lower than in non-drought years. The difference in lipid levels in Texas and Mexico could be due to the acquisition of nectar and lipids as the Texas monarchs continued to the overwintering sites through Northeastern Mexico or selective mortality that eliminated monarchs with low lipid levels. While Brower et al. (2015) did find individuals with low lipid levels in Mexico, their

<sup>6</sup>https://capemaymonarchs.blogspot.com/

<sup>7</sup>drought.gov/drought/states/texas

data suggest that the frequency of such individuals, and the amount of add-on mortality that might accrue from it, was no different between drought year and normal years. Thus, add-on mortality and overall migration success may not be interrelated. If this is the case, recovery rates are primarily a measure of migration success.

Overall, the recovery rate for monarchs tagged in the Midwest from 1998 to 2014 was about four times greater than for monarchs tagged in the Northeast. Thus, monarchs tagged in the Midwest have a higher probability of making it to Mexico than monarchs tagged in the Northeast. The greater distances from Northeast locations to Mexico may account for lower recovery rates from that region due to longer flight times and greater mortality. In addition, monarchs from the Northeast using the Eastern flyway appear to take two paths, one along the east coast from Maine to at least South Carolina and an interior path along the Appalachians. Monarchs then move west along the Gulf coast from Florida to Brownsville, Texas (Journey North<sup>5</sup> ). Brindza et al. (2008) found that butterflies tagged in more inland areas of the Northeast were more likely to be recovered than those tagged in coastal areas. Some east coast migrants may also end up in Florida (Vander Zanden et al., 2018). The Central Flyway, used by the butterflies from the Midwest, provides a clear southwest track to the Texas border with Mexico. Differences in recovery rate between the Central Flyway and the Eastern Flyway may also be the result of differences in wind patterns and temperature. Geographic differences in recovery rate on a finer scale are being investigated.

The significant correlation between the recovery rates for the Midwest and Northeast (**Table 2**) suggests that, although migration success for the Northeast is generally lower than the Midwest, there is an overarching factor that affects both regions similarly. Both Midwest and Northeast monarchs must traverse similar terrain in southern Texas and northeastern Mexico although they may take different routes to get to that point. The correlation between migration success and NDVI suggests that nectar availability in this region affects migration success of all butterflies, no matter their point of origin. In addition, there may be weather related timing delays in migration from 1 year to the next that can affect migration success for both regions.

The recoveries from the 3 years (2001, 2003, and 2015), with mass mortality due to winter storms, were not included in the analysis because the majority of these deaths were decidedly storm related rather than caused by attrition factors occurring in a normal year. Because recovery rate is based on dead butterflies, to use recovery rate as an index of migration success requires the assumption that normal overwinter mortality is a random variable over time. Mass mortality years violate that assumption and are not be included in the temporal analysis. Yet, recoveries in mass mortality years can tell us; (1) that large numbers of tagged monarchs were still alive at the time of these events and (2) that years with normal mortality and mass mortality both provide a random sample of the population, only differing in sample size, as evidenced by the similarity in the ratio of Midwest to Northeast recovery rates between normal and mass mortality years (**Supplementary Table 2**).

# CONCLUSIONS

Two competing hypotheses have been offered as the major reason for the decline in the number of monarchs overwintering in Mexico. The milkweed limitation hypothesis is based on data showing a massive loss of milkweeds due to herbicide use in agricultural fields and additional losses due to changes in land use from the expansion of agriculture and development (Lark et al., 2015; Pleasants, 2017; World Wildlife Fund Plowprint Report, 2018). While weather related factors may play an important role in the annual variation in population size, the decreased abundance of milkweeds and the greater dispersion of milkweed patches effectively caps the maximum size the population can attain under favorable conditions (Pleasants, 2017).

In contrast, the migration mortality hypothesis assumes that the amount of habitat is sufficient to support large monarch populations and that milkweeds are not limiting (Ries et al., 2015a,b; Inamine et al., 2016; Agrawal and Inamine, 2018). Rather, based on a perceived disconnection between observed summer monarch populations and overwintering numbers, the advocates of this hypothesis posit that increasing monarch mortality during the migration over the last two decades accounts for the decline in monarch numbers. Using data of about 500,000 wild-caught monarchs tagged from 1998 to 2015 along with 6,000 tag recoveries in Mexico during this period we show that (1) there was no disconnection between late summer and overwintering numbers, (2) the recovery rate of tagged monarchs (migration success) was not correlated with overwintering numbers, and (3) the recovery rate (migration success) had not decreased over time. In sum, none of the expectations of the migration mortality hypothesis were supported by the tagging and recovery data. Although there was no pattern indicating an increase in mortality during the migration, it is clear that low NDVI values, which indicate drought years and low nectar availability, are associated with lower than expected numbers of monarchs reaching the overwintering sites and lower recovery rates. Thus, migration success may determine some of the variation in the overwintering population size, but the main determinant is the size of the summer population.

Given historic, recent and continuing milkweed losses (Pleasants, 2017), and the importance of the Midwest to the overwintering monarch population (Flockhart et al., 2015), the challenge ahead is how to sustain the monarch population. The "all hands-on deck" approach (Thogmartin et al., 2017b) indicated that returning monarch numbers to a mean of 6 hectares of overwinter habitat occupied, a level that could assure survival of the migration given known causes of mortality (Semmens et al., 2016), would require the restoration of 1.4 billion milkweed stems primarily on landscapes in the Midwest. But we should not ignore the benefit to the monarch population that could come from restoring quality nectar habitats for both the migratory and breeding phases of the monarch annual cycle.

#### DATA AVAILABILITY STATEMENT

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

#### AUTHOR CONTRIBUTIONS

The data for this study came from monarch tagging efforts coordinated and supported by Monarch Watch, under the direction of OT. The data were vetted by JL and AR, and complied and sorted by RG, SP, and JL. Statistical analysis was done by JP and RG. The study was conceived and designed by OT and JP. JP, OT, and RG wrote the manuscript. All authors contributed to the article and approved the submitted version.

#### FUNDING

Donations to Monarch Watch were used to pay for recovered tags in Mexico. Funding from the U.S. Geological Survey Ecosystems Mission Area helped support preparation of the Monarch Watch data.

#### ACKNOWLEDGMENTS

We were indebted to Diane Pruden, Debbie Jackson, Gail Morris, Tricia Neal, Carol Pasternak, Janis Lentz, Sarah Schmidt, Dana

#### REFERENCES


Wilfong, Cathy Walters, Veronica Prida, David Kust and family, and many others who helped purchase and recover tags at the overwintering sites in Mexico. We were most appreciative of the numerous donors who have contributed to the tag recovery fund over the years. Numerous staff members and students have assisted with data managements over the last two decades, and recently, we have been aided by many data entry volunteers. We were indebted as well to the guides and ejido residents who diligently recovered and saved each recovered tag. We wish to thank the tens of thousands of taggers who have so enthusiastically contributed to this project since 1992. Thanks to Eduardo Rendón-Salinas and his crew for their pains-taking efforts to determine the size of the monarch overwintering population every year. Thanks to Sarah Saunders for sharing NDVI data. Thanks to Fan Dai for help with statistical analysis. Thanks to Barbara Pleasants for editing several versions of the manuscript. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fevo. 2020.00264/full#supplementary-material

S. Oberhauser, K. R. Nail, and S. Altizer (Ithaca, NY: Cornell University Press), 117–129.


variation in the natal origin of monarch butterflies overwintering in Mexico over 38 years. Glob. Change Biol. 23, 2565–2576. doi: 10.1111/gcb.13589


**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 Taylor, Pleasants, Grundel, Pecoraro, Lovett and Ryan. 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.