# INTEGRATING EMERGING TECHNOLOGIES INTO MARINE MEGAFAUNA CONSERVATION MANAGEMENT

EDITED BY : Peter H. Dutton, Mark Meekan, Lars Bejder and Lisa Marie Komoroske PUBLISHED IN : Frontiers in Marine 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 978-2-88963-287-9 DOI 10.3389/978-2-88963-287-9

#### 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

## INTEGRATING EMERGING TECHNOLOGIES INTO MARINE MEGAFAUNA CONSERVATION MANAGEMENT

Topic Editors:

Peter H. Dutton, Southwest Fisheries Science Center (NOAA), United States Mark Meekan, Australian Institute of Marine Science (AIMS), Australia Lars Bejder, University of Hawaii at Manoa, United States Lisa Marie Komoroske, National Oceanic and Atmospheric Administration (NOAA), United States

Citation: Dutton, P. H., Meekan, M., Bejder, L., Komoroske, L. M., eds. (2019). Integrating Emerging Technologies into Marine Megafauna Conservation Management. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-287-9

# Table of Contents

*05 Editorial: Integrating Emerging Technologies Into Marine Megafauna Conservation Management*

Peter H. Dutton, Lisa Komoroske, Lars Bejder and Mark Meekan

*08 DNA Metabarcoding as a Marine Conservation and Management Tool: A Circumpolar Examination of Fishery Discards in the Diet of Threatened Albatrosses*

Julie C. McInnes, Simon N. Jarman, Mary-Anne Lea, Ben Raymond, Bruce E. Deagle, Richard A. Phillips, Paulo Catry, Andrew Stanworth, Henri Weimerskirch, Alejandro Kusch, Michaël Gras, Yves Cherel, Dale Maschette and Rachael Alderman


Mark Meekan, Christopher M. Austin, Mun H. Tan, Nu-Wei V. Wei, Adam Miller, Simon J. Pierce, David Rowat, Guy Stevens, Tim K. Davies, Alessandro Ponzo and Han Ming Gan


Camrin D. Braun, Gregory B. Skomal and Simon R. Thorrold

*108 TurtleCam: A "Smart" Autonomous Underwater Vehicle for Investigating Behaviors and Habitats of Sea Turtles*

Kara L. Dodge, Amy L. Kukulya, Erin Burke and Mark F. Baumgartner

	- C. Scott Baker, Debbie Steel, Sharon Nieukirk and Holger Klinck

*129 Characterizing the Duration and Severity of Fishing Gear Entanglement on a North Atlantic Right Whale (*Eubalaena glacialis*) Using Stable Isotopes, Steroid and Thyroid Hormones in Baleen*

Nadine S. J. Lysiak, Stephen J. Trumble, Amy R. Knowlton and Michael J. Moore


Stephanie Brodie, Michael G. Jacox, Steven J. Bograd, Heather Welch, Heidi Dewar, Kylie L. Scales, Sara M. Maxwell, Dana M. Briscoe, Christopher A. Edwards, Larry B. Crowder, Rebecca L. Lewison and Elliott L. Hazen

*170 Embracing Complexity and Complexity-Awareness in Marine Megafauna Conservation and Research*

Rebecca L. Lewison, Andrew F. Johnson and Gregory M. Verutes

*181 Underwater Acoustic Ecology Metrics in an Alaska Marine Protected Area Reveal Marine Mammal Communication Masking and Management Alternatives*

Christine M. Gabriele, Dimitri W. Ponirakis, Christopher W. Clark, Jamie N. Womble and Phoebe B. S. Vanselow

*198 Bottlenose Dolphins and Antillean Manatees Respond to Small Multi-Rotor Unmanned Aerial Systems*

Eric A. Ramos, Brigid Maloney, Marcelo O. Magnasco and Diana Reiss

*213 The Utility of Combining Stable Isotope and Hormone Analyses for Marine Megafauna Research*

Alyson H. Fleming, Nicholas M. Kellar, Camryn D. Allen and Carolyn M. Kurle

*228 Assessing Seasonality and Density From Passive Acoustic Monitoring of Signals Presumed to be From Pygmy and Dwarf Sperm Whales in the Gulf of Mexico*

John A. Hildebrand, Kaitlin E. Frasier, Simone Baumann-Pickering, Sean M. Wiggins, Karlina P. Merkens, Lance P. Garrison, Melissa S. Soldevilla and Mark A. McDonald

# Editorial: Integrating Emerging Technologies Into Marine Megafauna Conservation Management

#### Peter H. Dutton<sup>1</sup> \*, Lisa Komoroske<sup>2</sup> , Lars Bejder 3,4,5 and Mark Meekan<sup>6</sup>

*<sup>1</sup> NOAA Fisheries, Southwest Fisheries Science Center, La Jolla, CA, United States, <sup>2</sup> Department of Environmental Conservation, University of Massachusetts, Amherst, MA, United States, <sup>3</sup> Aquatic Megafauna Research Unit, Centre for Sustainable Aquatic Ecosystems, Harry Butler Institute, Murdoch University, Murdoch, WA, Australia, <sup>4</sup> Marine Mammal Research Program, Hawaii Institute of Marine Biology, University of Hawai'i at Manoa, Kaneohe, HI, United States, <sup>5</sup> Zoophysiology, Department of Bioscience, Aarhus University, Aarhus, Denmark, <sup>6</sup> Australian Institute of Marine Science, Crawley, WA, Australia*

Keywords: technology, megafauna, conservation, wildlife, human impacts

#### **Editorial on the Research Topic**

#### **Integrating Emerging Technologies into Marine Megafauna Conservation Management**

Many recent and emerging technological innovations hold great potential to transform the "best-available science" for marine megafauna conservation management, such as remote sensing, telemetry, molecular technologies, unmanned aerial vehicles, bio-acoustics, and animal-borne imaging (O'Brien, 2015; Nowacek et al., 2016; Hays and Hawkes, 2018; Harcourt et al., 2019). This includes both the use of these technologies to address key knowledge gaps in species' biology required for management decisions (e.g., critical habitat use, demographic vital rates, and population connectivity), as well as their application to identify and mitigate human impacts (e.g., distinguishing impact hotspots and forecasting interactions). These technologies are being increasingly employed across a broad diversity of wildlife research contexts; however, there has been highly variable efficacy integrating these new tools into conservation science and translating results into successful management practices and policies (Berger-Tal and Lahoz-Monfort, 2018).

In this special Research Topic, researchers submitted articles addressing how recent and emerging technological innovations are being used to answer the key outstanding biological questions for marine megafauna. The resulting 17 articles illustrate how novel information from different technological applications is informing marine megafauna conservation and discuss challenges, future directions and remaining technological gaps.

### BIO-ACOUSTICS

The collection of passive acoustic data is a rapidly evolving field that is helping to enhance the ability to accurately estimate abundance and determine distribution of marine mammals, particularly for rare or elusive species (Marques et al., 2013). The case study presented by Hildebrand et al. illustrates how passive acoustic monitoring can be applied to detect and analyze signals from two different species of sperm whales (Kogia spp.) to obtain estimates of population density in the Gulf of Mexico (GOM).

### GENETIC AND OTHER MOLECULAR ANALYSES

Collecting non-invasive samples for genetic analysis to identify species, subspecies, or stocks of marine megafauna at sea remains challenging. Baker et al. demonstrate how eDNA methods can be

Edited and reviewed by: *Graeme Clive Hays, Deakin University, Australia*

> \*Correspondence: *Peter H. Dutton peter.dutton@noaa.gov*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *17 September 2019* Accepted: *28 October 2019* Published: *19 November 2019*

#### Citation:

*Dutton PH, Komoroske L, Bejder L and Meekan M (2019) Editorial: Integrating Emerging Technologies Into Marine Megafauna Conservation Management. Front. Mar. Sci. 6:693. doi: 10.3389/fmars.2019.00693*

used to detect specific communities of killer whales, and validate a method that will be useful for collecting DNA from the wake of whales.

Meekan et al. describe an invertebrate DNA (iDNA) approach whereby DNA from whale sharks (Rhincodon typus) was obtained from skin attached to copepods they removed to study population structure in elasmobranchs. McInnes et al. used DNA metabarcoding of black-browed albatross (Thalassarche melanophris) scats as a non-invasive method to identify the variety of fish species found in the diet of seabirds, and discuss uses of this approach to evaluate the extent of scavenging interaction with fishery discards.

The use of stable isotopes has become a valuable technique to understand the biogeography and foraging habits of marine species, while hormone analyses provide a means to assess reproduction, nutrition, stress, and health of individuals and in populations. Fleming et al. explore how combining these two approaches can better inform future marine megafauna conservation and management efforts. They identify four broad areas of research that will require methodological developments.

Lysiak et al. combined stable isotope analysis with analyses of steroid and thyroid hormones on samples from a drowned North Atlantic right whale (Eubalaena glacialis) to illustrate how key physiological indicators traced from baleen can be used to identify the recent anthropogenic impacts on these threatened whale populations.

Meyer et al. evaluated the use of lipid and fatty acid analyses on white sharks, Carcharodon carcharias, for gaining insights into the trophic ecology of marine elasmobranchs.

### AUTONOMOUS UNDERWATER VEHICLES (AUVs)

Technological advances in autonomous underwater vehicles (AUVs) enable direct observation of underwater behaviors and habitats of marine megafauna.

Dodge et al. describe a "smart" AUV that allows direct observation of free-swimming marine animals by concurrently recording video, localization, depth, and environmental data. They use this technology to characterize the diving behavior, foraging ecology, and habitat use of leatherback turtles (Dermochelys coriacea) in a coastal habitat impacted by anthropogenic hazards to inform conservation and management planning.

#### UNMANNED AERIAL VEHICLES (UAVs) OR SYSTEMS (UASs)

Drones, UAVs, or UASs have great potential for overcoming challenges of collecting samples from animals in the wild. Pirotta et al. developed customized UAVs to sample "whale blow," from free-swimming humpback whales (Megaptera novaeangliae), to carry out a population health assessment based on microbiota composition determined to be present in respiratory tracts using genetic sequencing. This study describes a promising new tool for monitoring health in marine megafauna, however the potential disturbance of these systems on marine megafauna are unknown.

Ramos et al. assessed the responses of bottlenose dolphins (Tursiops truncatus) and manatees (Trichechus manatus manatus) to UAS surveillance flight and discuss guidelines that could be developed to minimize animal disturbance.

### SATELLITE TELEMETRY

Dewar et al. Provide new insights on movements, behaviors and habitat use by basking sharks (Cetorhinus maximus) off the coast of California, by analyzing data from pop-off satellite archival transmitting (PSAT) and oceanographic data. They discuss how shifts in vertical and horizontal movement patterns likely reflect changes in prey availability and oceanographic conditions. In order to overcome limitations of using light-based geolocation technology on basking sharks that spend most of their time in deep waters below the photic zone, Braun et al. analyzed depthtemperature profile data recorded by PSAT tags in combination with high resolution models of in situ oceanographic data to determine movement patterns in the NW Atlantic. Both basking shark studies found evidence of long-range movement that illustrates the importance of international cooperation for effective conservation strategies for this threatened species.

Horton et al. provide an extensive analysis of satellite telemetry movement data across vast geographic ranges that shows a fidelity with respect to migratory routes that is associated with gravitational and magnetic coordinates for marine megafauna, including great white sharks (C. carcharias), northern elephant seals (Mirounga angustirostris), and humpback whales (M. novaeangliae).

### OCEAN NOISE MODELING TOOLS

Underwater vessel-generated noise can interfere with the ability of marine mammals to communicate and detect prey. Gabriele et al. examined the ability of harbor seals and humpback whales to communicate with conspecifics under various ambient noise scenarios as a step toward developing tools to assess and mitigate anthropogenic noise.

### SPECIES DISTRIBUTION AND HABITAT MODELING

Determining the spatial and temporal distributions of marine megafauna species is an important challenge in marine megafauna research and conservation. Brodie et al. highlight the utility of including dynamic subsurface variables to improve performance of species distribution models to characterize habitat use of pelagic predators, including blue sharks (Prionace glauca), shortfin mako sharks (Isurus oxyrinchus), common thresher sharks (Alopias vulpinus), and swordfish (Xiphias gladius), in the California Current.

Horton et al. present results of thermographic research that illustrates the potential of infrared videography with various post-processing analytical approaches of thermographic data to create automated platforms for whale detection. They discuss how advances in these technologies provide a non-invasive approach to identify species, collect information for monitoring distributions of cetaceans in remote marine habitats, and reduce ship-strikes in high traffic areas.

Finally, Lewison et al. discuss several emerging technologies that incorporate approaches that account for complexity of ocean systems to address fisheries bycatch and highlight opportunities for advancing marine megafauna research and conservation.

#### REFERENCES


Collectively the studies in this Special Topic illustrate advances in the application of some technologies to marine megafauna conservation. These, and additional emerging technologies will continue to evolve and provide opportunities for further innovation.

#### AUTHOR CONTRIBUTIONS

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

approaches and conservation implications. Anim. Behav. 120, 235–244. doi: 10.1016/j.anbehav.2016.07.019

O'Brien, J. (2015). Perspective: technologies for conserving biodiversity in the Anthropocene. Issues Sci. Technol. 32, 91–97.

**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 Dutton, Komoroske, Bejder and Meekan. 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.

# DNA Metabarcoding as a Marine Conservation and Management Tool: A Circumpolar Examination of Fishery Discards in the Diet of Threatened Albatrosses

Julie C. McInnes 1, 2 \*, Simon N. Jarman3, 4, Mary-Anne Lea<sup>1</sup> , Ben Raymond1, 2 , Bruce E. Deagle<sup>2</sup> , Richard A. Phillips <sup>5</sup> , Paulo Catry <sup>6</sup> , Andrew Stanworth<sup>7</sup> , Henri Weimerskirch<sup>8</sup> , Alejandro Kusch<sup>9</sup> , Michaël Gras <sup>10</sup>, Yves Cherel <sup>8</sup> , Dale Maschette<sup>2</sup> and Rachael Alderman<sup>11</sup>

#### Edited by:

Mark Meekan, Australian Institute of Marine Science, Australia

#### Reviewed by:

Maelle Connan, Nelson Mandela Metropolitan University, South Africa Nuno Queiroz, University of Porto, Portugal

\*Correspondence: Julie C. McInnes julie.mcinnes@utas.edu.au

#### Specialty section:

This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science

Received: 31 May 2017 Accepted: 14 August 2017 Published: 31 August 2017

#### Citation:

McInnes JC, Jarman SN, Lea M-A, Raymond B, Deagle BE, Phillips RA, Catry P, Stanworth A, Weimerskirch H, Kusch A, Gras M, Cherel Y, Maschette D and Alderman R (2017) DNA Metabarcoding as a Marine Conservation and Management Tool: A Circumpolar Examination of Fishery Discards in the Diet of Threatened Albatrosses. Front. Mar. Sci. 4:277. doi: 10.3389/fmars.2017.00277 1 Institute for Marine and Antarctic Studies, University of Tasmania, Hobart, TAS, Australia, <sup>2</sup> Australian Antarctic Division, Kingston, TAS, Australia, <sup>3</sup> Trace and Environmental DNA Laboratory, Department of Environment and Agriculture, Curtin University, Perth, WA, Australia, <sup>4</sup> CSIRO Indian Ocean Marine Research Centre, The University of Western Australia, Perth, WA, Australia, <sup>5</sup> British Antarctic Survey, Natural Environment Research Council, Cambridge, United Kingdom, <sup>6</sup> Marine and Environmental Sciences Centre, ISPA-Instituto Universitário, Lisbon, Portugal, <sup>7</sup> Falklands Conservation, Stanley, Falkland Islands, <sup>8</sup> Centre d'Etudes Biologiques de Chizé, UMR 7372 du Centre National de la Recherche Scientifique-Université de La Rochelle, Villiers-en-Bois, France, <sup>9</sup> Wildlife Conservation Society, Punta Arenas, Chile, <sup>10</sup> Directorate of Natural Resources-Fisheries of the Falkland Islands Government, Stanley, Falkland Islands, <sup>11</sup> Department of Primary Industries, Parks, Water and Environment, Hobart, TAS, Australia

Almost all of the world's fisheries overlap spatially and temporally with foraging seabirds, with impacts that range from food supplementation (through scavenging behind vessels), to resource competition and incidental mortality. The nature and extent of interactions between seabirds and fisheries vary, as does the level and efficacy of management and mitigation. Seabird dietary studies provide information on prey diversity and often identify species that are also caught in fisheries, providing evidence of linkages which can be used to improve ecosystem based management of fisheries. However, species identification of fish can be difficult with conventional dietary techniques. The black-browed albatross (Thalassarche melanophris) has a circumpolar distribution and has suffered major population declines due primarily to incidental mortality in fisheries. We use DNA metabarcoding of black-browed albatross scats to investigate their fish prey during the breeding season at six sites across their range, over two seasons. We identify the spatial and temporal diversity of fish in their diets and overlaps with fisheries operating in adjacent waters. Across all sites, 51 fish species from 33 families were identified, with 23 species contributing >10% of the proportion of samples or sequences at any site. There was extensive geographic variation but little inter-annual variability in fish species consumed. Several fish species that are not easily accessible to albatross, but are commercially harvested or by-caught, were detected in the albatross diet during the breeding season. This was particularly evident at the Falkland Islands and Iles Kerguelen where higher fishery catch amounts (or discard amounts where known) corresponded to higher occurrence of these species in diet samples. This study indicates ongoing

**8**

interactions with fisheries through consumption of fishery discards, increasing the risk of seabird mortality. Breeding success was higher at sites where fisheries discards were detected in the diet, highlighting the need to minimize discarding to reduce impacts on the ecosystem. DNA metabarcoding provides a valuable non-invasive tool for assessing the fish prey of seabirds across broad geographic ranges. This provides an avenue for fishery resource managers to assess compliance of fisheries with discard policies and the level of interaction with scavenging seabirds.

Keywords: scat, trawl fishery, fisheries resource management, Southern Ocean, Thalassarche melanophris, seabird-fishery interaction, fish diversity, seabirds

### INTRODUCTION

Effective ecosystem-based management of commercial fisheries requires information not just on the sustainability of target stocks, but also on the interactions of other marine organisms with fishing operations. Globally, seabirds frequently interact with commercial fisheries through competition for shared resources (Frederiksen et al., 2004; Okes et al., 2009), incidental mortality in fishing gear (Brothers et al., 1999; Sullivan et al., 2006; Watkins et al., 2008; Tuck et al., 2011) and consumption of fishery discards (Garthe et al., 1996; Gonzalez-Zevallos and Yorio, 2006). Seabird survival and breeding success can be reduced by competition with fisheries (Furness and Tasker, 2000; Frederiksen et al., 2004), and incidental mortality in fishing gear can be a major cause of population declines, particularly of albatrosses and large petrels (Weimerskirch and Jouventin, 1987; Barbraud et al., 2008; Phillips et al., 2016). Physical and operational mitigation measures have been developed to reduce seabird mortality (Løkkeborg, 2008; Phillips et al., 2016), including the reduction of fishery discards, which decreases the attractiveness of vessels (Abraham et al., 2009; Pierre et al., 2012). Scavenging birds are attracted to the supplementary food source provided by discards, which may consist of (i) the head, tail and offal of retained catch (commercial species caught at commercial size); (ii) whole fish of commercial species but caught at a non-commercial size; (iii) non-commercial species and (iv) unused baits (in longline fishing). These discards are often fish or other species that may not be naturally accessible. Some populations benefit from the additional food source, with higher breeding success and survival resulting in population growth (Oro et al., 1995; Bertellotti and Yorio, 2000). However, discards can alter food-web structure by providing nutritionally-poor food (Grémillet et al., 2008), or artificially inflating populations of predatory gulls or skuas, which may not be sustainable in the absence of discards or which also prey on smaller seabirds, with potentially major impacts (Phillips et al., 1999b; Foster et al., 2017). The interactions between seabird populations and fisheries are likely to vary over time, space and species; therefore, understanding the nature and extent of these interactions is imperative for effective ecosystem management.

Seabird dietary studies can inform ecosystem risk assessments for fishery management by identifying interactions between fisheries and seabirds for different populations (Phillips et al., 1999a). Understanding the dietary flexibility of seabirds is also fundamental for predicting the responses of individuals and populations to spatial and temporal changes in natural prey abundance, and availability from fisheries, and hence for the effective management of marine resources (Constable et al., 2000). Stomach content and stable isotope analyses are the two main approaches for assessing seabird diet (Duffy and Jackson, 1986; Barrett et al., 2007). The former primarily relies on the use of otoliths and bones to identify fish prey, enabling prey size and meal mass estimates to be obtained. However, discrimination can be poor or impossible if the prey (including larvae or eggs) is small, has no hard parts, or digests quickly; the hardparts are eroded; or those from closely-related species cannot be readily distinguished (Duffy and Jackson, 1986; Barrett et al., 2007). These problems apply in particular to items originating as fisheries offal, as viscera float and are therefore easier to ingest than fish heads with otoliths, particularly those from large species (Thompson and Riddy, 1995). More recent studies have used DNA analysis to identify parts that were not taxonomically diagnostic (Alonso et al., 2014). However, studies using stomach samples are usually restricted to the chick-rearing period, thus focusing on chick rather than adult diet across the annual cycle and usually requires handling of birds.

Stable isotope analysis of blood or feathers does not suffer from the biases associated with differential digestion of prey and can be applied to all stages of the breeding season. This method has been used to determine likely fishery overlaps by comparing the estimated proportions of pelagic and demersal prey, on the assumption that the latter were obtained from fisheries (Granadeiro et al., 2013). However, in most systems stable isotope analyses lack the resolution to identify prey beyond broad trophic groups. DNA metabarcoding of predator scats is a useful alternate or complementary method for assessing seabird diet (Deagle et al., 2007; Bowser et al., 2013). It can provide highlevel taxonomic resolution and does not require prey remains to be physically identifiable (Pompanon et al., 2012). Although the method cannot be used to identify prey size and meal mass, it does give an indication of species occurrence in the diet. Samples can also be collected during all breeding stages (McInnes et al., 2017a) and the technique is non-invasive and requires minimal field time compared to conventional diet sampling, increasing the options for simultaneous sampling across broad spatial scales (Jarman et al., 2013).

The black-browed albatross (BBA, Thalassarche melanophris) has a circumpolar distribution and is the most abundant albatross species in the southern hemisphere (Phillips et al., 2016). Populations have experienced extensive declines which are strongly linked to incidental mortality in longline and trawl fisheries (Phillips et al., 2016). While the population at South Georgia is still declining (Poncet et al., 2017), numbers in the Falkland Islands and on islands off Chile are currently increasing (Wolfaardt, 2013; Robertson et al., 2014, 2017). The increases in Chile have been attributed to a reduction in incidental seabird mortality due to faster sink rates of baited longline hooks associated with a change in fishing practices, and the use of birdscaring (streamer or tori) lines, making hooks less accessible to birds (Robertson et al., 2014). However, longline and trawl fisheries are still thought to cause high mortality of this species elsewhere, especially in the wintering grounds (Yeh et al., 2013; Kuepfer, 2015; Tamini et al., 2015). Fishery resource overlaps with the diet of black-browed albatrosses have been shown at all breeding sites where fish have been characterized, including Iles Kerguelen (Cherel et al., 2000), Diego Ramirez (Arata and Xavier, 2003), South Georgia (Reid et al., 1996; Xavier et al., 2003) and the Falkland Islands (Thompson, 1992). However, the most recent samples used in these studies were collected over 15 years ago (1995, 2002, 2000, and 1991, respectively; Data Sheet 1 in Supplementary Material), over which time fishing operations and regulations, including discarding policies and mitigation requirements, have changed substantially in many regions (Phillips et al., 2016).

We used DNA metabarcoding of BBA fecal DNA to investigate the fish prey consumed at six sites across their breeding range to: (1) determine the fish prey diversity and any spatial and temporal variability; (2) identify any fishery target, bycatch and bait species in the diet of BBA to distinguish regions in which rates and risks of vessel interactions may be greater (and hence efforts to improve discard management and monitoring of compliance with seabird bycatch mitigation may be targeted); and (3) evaluate sources of potential resource competition or food supplementation by fisheries. We use this study to show that DNA metabarcoding can quantify fish diversity and the presence of discards in the diet of seabirds, providing a valuable tool for fishery resource and conservation management.

### METHODS

### Study Sites and Sample Collection

Fresh scat samples were collected from black-browed albatrosses at six breeding colonies over multiple seasons: in austral summers 2013/14 and 2014/15 at New Island and Steeple Jason Island (Falkland Islands), Macquarie Island (Australia) and Bird Island (South Georgia); in 2013/14 and 2015/16 at Canyon des Sourcils Noirs (Iles Kerguelen); and in 2014/15 at Albatross Islet (Chile; **Figure 1**). The majority of samples were collected during the chick-rearing period (December-March) with additional samples collected during incubation in 2014/15 at Steeple Jason Island and New Island, Kerguelen in 2013/14 and during incubation in both years at Macquarie Island (**Table 1**). Sampling years are hereafter termed 2014 for samples collected in 2013/14 and 2015 for 2014/15 samples. This project was approved by the University of Tasmania Animal Ethics Committee (Permit A13745).

A small fragment of the non-uric acid portion (dark part) of each scat was collected using tweezers or a spatula and stored in 80% ethanol. Where possible, fresh scats were obtained (where

FIGURE 1 | Breeding distribution of black-browed albatrosses and sampling sites. Blue dots represent the six colonies where scat samples were collected, and the red dots the remaining colonies not sampled during the study. The inset shows the individual Chilean and Falkland Island colonies. Samples were collected from Albatross Islet, Chile (40–50 breeding pairs, population increasing); New Island (13,343 breeding pairs, population increasing) and Steeple Jason Island (183,135 pairs, population increasing), Falkland Islands; Bird Island, South Georgia (8,264 breeding pairs, population declining); Canyon des Sourcils Noirs, Iles Kerguelen (∼1,200 breeding pairs, population stable); and Macquarie Island (∼200 breeding pairs, population stable; ACAP, 2010; Wolfaardt, 2013; Robertson et al., 2014; Phillips et al., 2016; Poncet et al., 2017)


 primer Actinopterygii (bony fish) Chondrichthyes (sharks skates). but still wet) and the developmental stage of the bird (chick, juvenile or adult) was recorded. Given the low sample sizes remaining when samples were split by site, age, and month, differences between diet of chicks and adults (self-feeding) could not be explored in this study, and therefore samples from different ages were pooled. Further research with greater sample sizes are required to test partitioning of diet by adults for provisioning compared with self-feeding (Davoren and Burger, 1999; Danhardt et al., 2011), and potential dietary differences between breeders, non-breeders and juveniles (Campioni et al., 2016). The foraging ranges of black-browed albatross are greater during incubation than chick-rearing, and the magnitudes of these differences depend on the colony (Wakefield et al., 2011). For example, at South Georgia, mean maximum foraging distances of tracked adults were 980–1,690 km (262–327 h) and 275–505 km (45–77 h) during incubation and chick-rearing, respectively (Phillips et al., 2004). The prey detected in scat samples is likely to reflect the most recent meal consumed by albatross, which is similar to stomach contents analysis. The digestion rates of seabirds are influenced by numerous variables, such as predator species, metabolic rate, meal size, food type, and feeding frequency (Hilton et al., 1998). In sooty albatross (Phoebetria fusca), the mean retention rate of prey ranged from 11 to 15 h, however some prey was still detected up to 50 h after eating (Jackson, 1992). In little penguins (Eudyptula minor) prey could be detected for up to 4 days using DNA metabarcoding (Deagle et al., 2010). The retention time is also likely to vary depending on whether the food is consumed for self-feeding or regurgitated to the chick partially digested. During this study, it is assumed that the prey DNA recovered reflects the most recent foraging trip. For extended foraging trips during incubation, some of the food may not be detected.

#### DNA Metabarcoding

DNA was extracted from albatross scat samples using a Promega "Maxwell 16" instrument and a Maxwell <sup>R</sup> 16 Tissue DNA Purification Kit. PCR inhibitor concentrations were diluted by mixing a small amount (∼30 mg) of the fecal samples in 250µL of STAR buffer (Roche Diagnostics) prior to extraction. Two different DNA markers were amplified. The first used a metazoan primer set that is highly conserved and amplifies a region of the nuclear small subunit ribosomal DNA (rDNA) 18S gene (McInnes et al., 2017a, **Table 2**). For this marker the taxonomic resolution is relatively low; however, it recovers DNA from all animal lineages and provides a broad view of the diet. The second primer pair amplifies a region of the 16S rDNA gene specifically from fish and varies enough to allow species-level identification for most of the targeted fish species (**Table 2**). This primer set was designed based on an alignment of mtDNA 16S sequences from representative Southern Ocean fish that were publicly available on Genbank (a full alignment with sequences in fasta format can be found in Data Sheet 2 of the Supplementary Material). The primer set was designed not to match bird DNA. Primers were tested with fish flesh and scat DNA. All samples were run with the 18S\_SSU primer set first, and those that had fish DNA

the bird was seen defecating or the sample was on the ground

set.

Fish

includes

and

and

presented for those samples from which > 100 food sequences amplified with the 18S\_SSU


Underlined bases in PCR Round 1 are the Miseq tag primer. Bolded bases in PCR Round 2 are an example of the unique tags attached to each sample.

were amplified using the 16S\_Fish primer set (See Image 1 of the Supplementary Material for the DNA metabarcoding workflow).

PCR reactions for each primer set were carried out separately as a two stage process. Stage one PCR reactions (10µL) were performed with 5µL 2 × Phusion HF (NEB), 1µL 100 × Bovine Serum Albumin (NEB), 0.1µL 5 µM of each 18S\_SSU or 16S\_Fish amplification primer (**Table 2**), 0.5µL of Evagreen, 2µL fecal DNA and 1.3µL of water. Thermal cycling conditions were 98◦C, for 2 min; followed by 35 cycles for 18S\_SSU, and 45 cycles for 16S\_Fish, of 98◦C for 5 s, 67◦C for 20 s, 72◦C for 20 s, with an extension of 72◦C for 1 min. Each sample was run in triplicate on a LightCycler 480 (Roche Diagnostics). A negative control containing no template DNA and positive control containing fish DNA were included in each PCR amplification run. If either the negative amplified or the positive failed to amplify, the PCR was re-run. If ≥2 replicates of each sample had a ct score <30 for the 18S\_SSU, or <40 for 16S\_Fish, they were combined to reduce biases produced by amplification from samples with low template concentrations (Murray et al., 2015). Pooled samples were diluted 1:10 for the second stage PCR. In the second stage PCR, a unique tag was attached to each sample (**Table 2**). PCR reactions (10µL) were performed with 5µL 2 × Phusion HF (NEB), 1µL 100 × Bovine Serum Albumin (NEB), 1µL of 1 µM of each tag primer, and 2µL of diluted PCR product from stage one. Thermal cycling conditions were 98◦C, for 2 min; followed by 10 cycles of 98◦C for 5 s, 55◦C for 20 s, 72◦C for 20 s, with an extension of 72◦C for 1 min. Samples were pooled and purified from unincorporated reaction components by washing, utilizing reversible binding to Ampure (Agencourt) magnetic beads following the manufacturer's protocol. Sequencing of PCR products was performed with an Illumina MiSeq high throughput sequencer, using the MiSeq reagent kit V2 (300 cycles).

#### Bioinformatics

Amplicon pools were de-multiplexed based on unique 10 bp Multiplex IDentifiers (MIDs) incorporated in the Illumina twostep MID protocol. Fastq files were processed using USEARCH v8.0.1623 (Edgar, 2010). Reads R1 and R2 from the paired end sequencing were merged using the fastq\_mergepairs function, retaining only merged reads flanked by exact matches to the primers and primer sequences were trimmed. Reads from all samples were pooled and dereplicated, then clustered into broad Operational Taxonomic Units (OTUs) using the cluster\_otus command (-otu\_radius\_pct = 10). Potentially chimeric reads were discarded during this step. Reads for each sample were assigned to these OTUs (usearch\_global -id 0.97) and a summary table generated using a custom R script. Each OTU was identified by BLAST and categorized to closest match using MEGAN 5 (Huson et al., 2007) and the Lowest Common Ancestor (LCA) assignment algorithm. LCA parameters were set at a minimum score of 250 and a top-percent of 5% for the 18S\_SSU and 340 and 5% for the 16S\_Fish. These cut-offs were determined by manually checking a subset of samples against BLAST. Sequences were also manually checked on Genbank to ensure that all species from that genus in the region were represented. Additional flesh samples were obtained at the Falkland Islands (Gras et al., 2016) and through the Australian Antarctic Division and were sequenced and added to Genbank (see Data Availability Section for accession numbers).

OTUs derived from the 18S\_SSU primer set were assigned to class, whereas OTUs derived from the 16S\_Fish primers were classified to genus or species. OTUs were assigned only to genus if there was any uncertainty in the species match, either due to insufficient difference between species in the 16S region amplified, or if species from that genus were not present on Genbank. The geographic distribution of species in each genus was checked in Gon and Heemstra (1990) and Duhamel et al. (2014), and species was assigned if only one occurred within the foraging range of BBA from a particular site. In such cases, the species name in tables and figures is given in parentheses. Samples amplified with the 18S\_SSU primers were included if they contained at least 100 sequences of food DNA, whereas samples amplified for the 16S primers were included if they contained at least 100 sequences of fish DNA (Jarman et al., 2013). Results are presented as the number of samples with a prey item (n), the frequency of occurrence (FOO) and the relative read abundance of sequences (RRA). For FOO calculations, any food item or fish species was deemed present if it comprised >1% of food sequences for 18S\_SSU, or fish sequences for 16S\_Fish. The RRA for 18S was calculated as the total sequences for that prey group divided by the total food sequences for that sample, whereas the RRA for the 16S was the number of sequences for a fish species divided by the total fish sequences for that sample. The RRA was averaged across island or year groups. These multiple measures of diet composition are presented to reduce potential biases in interpretation that might result from consideration of a single metric. The results from the 18S region are presented to show the fish component of the diet and allow calculations of the overall proportion of the population consuming discards. Further details and discussion on the proportions of each prey group for each site can be found in McInnes et al. (2017b).

### Assessing Overlaps between Commercial Fisheries and BBA Prey

Data on fishery catches and target species were provided by the Directorate of Natural Resources of the Falkland Islands Government; the Australian Fisheries Management Authority and the Australian Antarctic Division; the Pecheker database (Martin and Pruvost, 2007) and online Commission for the Conservation of Antarctic Marine Living Resources (CCAMLR) Statistical Bulletins (CCAMLR, 2015). These included fishing effort (total hours for trawl and total hooks for longline); the total catch of target species and the main bycatch species (those comprising >1% of the total catch); the fish bait used in longline fishing operations; and, at Iles Kerguelen, the mass of target and bycatch species that were discarded. No data were available on Illegal, Unreported, and Unregulated (IUU) fishing. During the relevant sampling periods, trawl and longline fisheries were operating during the sampling period within the Falkland Islands Inner and Outer Conservation Zones (FICZ/FOCZ; **Table 2**), with no trawl fishing during January; longline but no trawl fisheries were operating within the Kerguelen Economic Exclusion Zone (EEZ); trawl but no longline fisheries operated close to South Georgia during the sampling period (CCAMLR Division 48.3; excluding March); no fishery was operating in the Macquarie Island EEZ; nor was there a fishery within the Admiralty Sound or Magellan Strait, which are used by foraging birds from Albatross Islet during chick rearing (Arata et al., 2014). Fishery species were defined as any target fish species, or bycatch fish species that made up >1% of the total catch (**Table 3**). Bait species used during fishing operations were also identified. For the main fish species (those contributing >10% of amplified sequences), the depth profile for each species during different age stages were compiled from the literature to determine which were likely to be naturally accessible to albatrosses (Table S1 in Supplementary Material). This study focused on the fishing zones adjacent to the breeding sites, as these are likely to be used more intensively than distant waters by foraging birds during chick-rearing (Phillips et al., 2004; Terauds et al., 2006; Catry et al., 2013; Arata et al., 2014), and secondly, the management of these areas is within the same national jurisdiction as the relevant breeding site. However, we acknowledge that birds may have also interacted with fisheries further from colonies, especially during incubation when BBA are foraging farther from the colony than during chick rearing (Phillips et al., 2004; Wakefield et al., 2011)

### Statistical Analyses

Analyses were carried out using R software (R Core Team). Poisson generalized linear models (GLM) with a log link


The fish species listed are those targeted by the fishery; bycatch species are those that constitute >1% of the total catch.

function were used to test if there were differences in fish species composition between colonies and years, and between years and breeding stages at each site. The model included the count of samples with fish DNA (n) as the dependant variable and predictor variables included fish species (F), year (Y) and breeding stage (S), or colony (C). The base model included the sample size as a function of the main effects (fish species, year, breeding stage, or colony) as well as the year:stage or year:colony interaction. These terms effectively describe the patterns in the data arising from the experimental sampling process (e.g., total number of samples within a given year). The interaction terms, fish:year, fish:stage, or fish:colony were added to the base model to test the effect of year or stage (or colony for the pooled data) on diet composition. The analysis of deviance (with Chi-squared test) and Akaike's information criterion (AIC) were used to compare fitted models and test the significance of predictor terms (Burnham and Anderson, 2002). A linear regression was used to assess the relationship between the proportion of samples with discards and breeding success, based on monitoring of BBA nesting attempts at each colony in the year that the diet samples were collected. Dissimilarity indices were calculated with the Manhattan method using the command "vegdist" in the package "Vegan" (Oksanen et al., 2016). From these indices, a hierarchical clustering was then constructed using the average agglomeration method, and plots created using the package "ggplot2" (Wickham, 2009). The proportion of samples that amplified with the 16S\_Fish primer which contained species that are also caught in fisheries was calculated, and applied to the total number of samples collected that amplified fish with the 18S\_SSU primers.

#### RESULTS

## Diversity, and Spatial and Temporal Variability in Fish Prey of BBA

A total of 1,091 scat samples were collected. DNA was amplified in 793 samples using the 18S\_SSU primers; 372 samples contained at least 100 sequences of food DNA, and 327 contained fish DNA. These samples were then amplified with the 16S\_Fish group specific primers; 295 samples contained at least 100 fish sequences and were included in subsequent analyses (**Table 1**).

Fish were found to be the most common prey group across all sites and years, based on the 18S\_SSU data. In total, 91% of samples contained fish and this made up 72% of sequences (ranging from 73 to 100% of samples, and between 41 and 97% of sequences at different sites; **Table 1**). Chondrichthyes (sharks and skates) were present in 2.7% of samples and comprised 2% of these sequences (**Figure 2**).

The higher resolution provided by the mtDNA 16S marker identified at least 51 fish species, from 33 families in the diet of BBA across the six breeding sites, with 23 species constituting >10% of the amplified sequences for different colonies and years (**Table 4**). The most common fish prey belonged to four families: Nototheniidae (notothens), Channichthyidae (crocodile icefishes), Congiopodidae (horsefishes), and Clupeidae (herrings, sardines and allies; **Tables 5**, **6**). Colonies were clustered into four distinct groups according to fish species composition: (1) Falkland Islands and Albatross Islet, (2) Iles Kerguelen, (3) Macquarie Island, and (4) Bird Island (**Figure 3**). When grouped by family, clusters were similar to fish species, except samples from Steeple Jason in 2015 were more similar to those from Iles Kerguelen due to the high occurrence of Nototheniidae.



<sup>∧</sup> target fishery species; <sup>+</sup>bycatch species, #not naturally accessible. Stars represent the frequency of occurrence (FOO) and relative read abundance (RRA) of amplified fish sequences: \*10–25%, \*\* 25–50%, \*\*\* 50–75% and \*\*\*\* >75%. A colored star is used when the FOO (blue) or RRA (red) were greater than the other measurement. Only species that occurred in more than one sample are included. Species in brackets are those where the genus could be confirmed but the species were either genetically similar [square brackets] or not all in Genbank (round brackets). Where only one species is in the bracket, this is the likely species given spatial distribution. Sites were Albatross Islet, Chile (AI); New Island (NI) and Steeple Jason Island (SJI), Falkland Islands; Bird Island, South Georgia (BI); Canyon des Sourcils Noirs, Iles Kerguelen (KI); and Macquarie Island (MI).

Fish from the family Nototheniidae were common to all groups. Clupeidae was common in group 1, Channichthyidae in groups 2 and 4, and Congiopodidae in group 3. The differences between years were less marked than those among colonies, as the inclusion of colony provided the best model fit (Base model AIC=1,618, F:Y AIC = 1,560, F:C AIC = 1,054, F:C+F:Y AIC = 1,175). Two to eight fish species contributed >10% of the fish prey for each colony-year combination (in either FOO or RRA), and were found in more than one sample (**Table 4**).

#### Albatross Islet

Eight fish species were found in the 49 samples from Albatross Islet; the majority contained Fueguian sprat (Sprattus fuegensis; 88% FOO), and black southern cod or Patagonian rockcod (Patagonotothen tessellata or brevicauda) was the second most common item (29% FOO; **Table 5**, **Figure 4**).

#### Falkland Islands

Eight fish species were identified in the 45 samples from New Island, and contained almost exclusively (>90% of sequences) Fueguian sprat (68% FOO) and rockcod (Patagonotothen sp.; 53% FOO; **Table 5**). There was no difference in the FOO of fish species consumed between years [Base model AIC = 80.27, F:Y AIC = 84.05; χ 2 <sup>7</sup> = 10.22, p = 0.17] or breeding stages [F:S AIC=97.45; χ 2 <sup>14</sup> = 10.82, p = 0.70; **Figure 4**].

Ten fish species were identified in the 51 samples from Steeple Jason Island, of which sprat was the most common species in 2014 (48% FOO), followed by hoki (Macruronus magellanicus; 21% FOO), southern blue whiting (Micromesistius australis australis; 21% FOO), rockcod (17% FOO) and kingclip (Genypterus blacodes; 10% FOO; **Table 5**). In 2015, rockcod was the main item (64% FOO) followed by sprat (18% FOO) and hoki (14% FOO). There was a difference in the fish species consumed between years [Base model AIC = 157.4, F:Y AIC = 152.51; χ 2 <sup>9</sup> = 22.90, p < 0.01] and breeding stages [F:S AIC = 129.3; χ 2 <sup>18</sup> = 64.13, p < 0.001]. When data were adjusted for year, the effect of stage was still significant [F:Y and F:S AIC = 142.5; χ 2 <sup>9</sup> = 46.1, p < 0.001], but not vice versa [F:S and F:Y AIC = 142.5; χ 2 <sup>9</sup> = 4.83, p = 0.85]. This is likely to be an artifact of the timing of sampling, as no samples were collected in incubation in 2014. During incubation in 2015, samples comprised mostly rockcod, whereas in both years,


Number of samples (n), frequency of occurrence (FOO, %) and relative read abundance (RRA, %). Species in brackets are those where the genus could be confirmed but the species were either genetically similar [square brackets] or not all in Genbank (round brackets). Where only one species is in the bracket, this is the likely species given spatial distribution.

samples collected during early chick-rearing were mostly of sprat and in 2014 were of kingclip. During late chick-rearing diet was more diverse, including hoki and rockcod in both years, southern blue whiting in 2014, and sprat in 2015 (**Table 5**, **Figure 4**).

### South Georgia

Sixteen fish species were found in the 68 samples from Bird Island, with two species particularly common in both years: South Georgia icefish (Pseudochaenichthys georgianus; 48 and 42% FOO) and mackerel icefish (Champsocephalus gunnari; 44 and 34% FOO). Marbled rockcod (Notothenia rossii; 26 and 24% FOO), yellow-fin notothen (Patagonotothen guntheri; 11 and 17% FOO) and humped rockcod (Gobionotothen sp.; 11.1 and 17.1% FOO) were also common. In 2014, moray cod (Muraenolepis (microps/orangiensis; 14.8% FOO) was in >10% of samples, whereas in 2015 a large proportion of samples included blackfin icefish (Chaenocephalus aceratus; 29% FOO) and southern driftfish (Icichthys australis; 27% FOO; **Table 6**, **Figure 4**). There was an effect of year [Base model AIC = 200.2, F:Y AIC = 196.3; χ 2 <sup>15</sup> = 33.9, p < 0.01] and breeding stage on fish consumed [F:S AIC = 200.7; χ 2 <sup>15</sup> = 29.5, p = 0.01]. However, although breeding stage was statistically significant, the base model excluding stage still provided a better fit to the data, even when both year and stage were included [F:Y and F:S AIC = 207.3; χ 2 <sup>15</sup> = 18.9, p = 0.21; F:S and F:Y AIC = 207.3; χ 2 <sup>15</sup> = 23.3, p = 0.08].

#### Iles Kerguelen

Eleven fish species were found in the 46 samples from Iles Kerguelen, with the main fish species gray rockcod (Lepidonotothen squamifrons; 53 and 56% FOO) and unicorn icefish (Channichthys rhinoceratus; 33 and 19% FOO) in both years. In 2014, the other common species were skates (Bathyraja sp.; 17% FOO) and moray cod (Muraenolepis marmoratus/orangiensis; 10% FOO), whereas in 2016, Patagonian toothfish (Dissostichus eleginoides) was the second most common item (44% FOO; **Table 6**, **Figure 4**). There were more samples with unicorn icefish during incubation than chick-rearing, whereas all the toothfish was consumed during chick-rearing. There was an effect of year [base model AIC = 113.6, F:Y AIC = 112.8; χ 2 <sup>10</sup> = 20.8, p = 0.03] and breeding stage on the fish species consumed [F:S AIC =113.1; χ 2 <sup>10</sup> = 20.5, p < 0.03].

#### Macquarie Island

Sixteen species were found in the 36 samples from Macquarie Island (**Table 6**). In both years, samples mostly contained Antarctic horsefish (Zanclorhynchus spinifer; 65 and 70% FOO) TABLE 6 | Fish prey of black-browed albatrosses at Bird Island, South Georgia UK (BI); Iles Kerguelen, France (KI) and Macquarie Island, Australia (MI).


Number of samples (n), frequency of occurrence (FOO) and relative read abundance (RRA). Species in brackets were those where the genus could be confirmed but the species were either genetically similar [square brackets] or not all on Genbank (round brackets). Where only one species is in the bracket, this is the likely species given spatial distribution.

and Magellanic rockcod (Paranotothenia magellanica; 31 and 10% FOO). In 2015, one unidentified species, likely from the family Bramidae, made up 20% of samples, although this may reflect the small sample size (n = 10). Fish species composition did not differ between years [Base AIC = 177.4, F:Y AIC = 190.6; χ 2 <sup>15</sup> = 16.8, p = 0.33]; the effect of breeding stage was of borderline statistical significance [F:S AIC = 193.5; χ 2 <sup>30</sup> = 43.9, p = 0.05], but the base model excluding stage still provided a better fit to the data.

## Overlaps between Commercial Fishery Species and BBA Prey

#### Longline Fisheries

Diets of BBA from New Island and Steeple Jason Island did not include any target or bycatch species from longline fisheries operating in the Falkland Islands FICZ/FOCZ. At Iles Kerguelen, diet samples included DNA from the target species, Patagonian toothfish in January of both years (with much higher proportions in 2016; **Tables 6**, **7**) and a bycaught group, skates, in December/January 2014/15 (**Table 7**, **Figure 5**). Bait fish, Scomber scombrus also appeared in samples, but occurred infrequently (<2% of sample sequences). This is a northern hemisphere species used as baits in longline fishing, and is therefore only available to BBA from fisheries. In the Kerguelen EEZ, the amount of toothfish discarded was lowest in November and December 2013 (0.19 and 0.18 t, respectively) and highest in January (1.6 t in 2014 and 2.9 t in 2016). More skates were discarded in December and January 2013/14 (0.3 t in November, 5.3 t in December and 9.4 t in January 2013/14; and 2 t in January 2016), which matched with the relative FOO in the diet in the 2 years.

#### Trawl Fisheries

The trawl fisheries operating in the Falkland FICZ/FOCZ target eight fish species (**Table 3**). No bycatch species made up more than 1% of the reported catch. Fishery target species were found in the diet at both sites in each year (**Table 7**, **Figure 5**). At New Island, the main fishery target species in the diet was rockcod (91% of those samples with a target species); one sample also contained hoki, and one was of kingclip only. At Steeple Jason Island, BBAs consumed five target species in 2014 (rockcod, hake, hoki, southern blue whiting, and kingclip), whereas samples included three target species in 2015 (rockcod, hoki, and hake; **Table 6**, **Figure 6**). The number of samples with fishery target species was lower during early chick rearing (December/January) than either incubation (October-November) or late chick rearing (February-March; **Figure 6A**). This corresponded to the relative catch in the fishery, particularly during January when it was not operating (**Figure 6A**). The main catch species in the fishery was rockcod during incubation, and hoki in late chick rearing (**Figure 6C**). The cluster analysis showed four distinct clusters, with one highlighting the similarity between fishery catch and fish prey of BBA during December 2013 at New Island and October 2014 at both sites, and between fishery catch during March and fish prey of BBA at Steeple Jason (**Figure 6D**).

At South Georgia, the fishery target species (mackerel icefish) and four bycatch species (South Georgian icefish, yellow-fin notothen, gray rockcod, and marbled rockcod) were all recorded in the diets of BBA (**Table 6**), and in a substantial proportion of the samples in both years (**Table 7**, **Figure 5**). The amount of target and bycatch fish species in the diet of BBA did not correspond to the relative catch rates in the fishery. During the sampling period the fishery caught very little mackerel icefish during January 2014 (65 kg), and only 3 tons during February 2014, with 1 ton of South Georgia icefish as bycatch. Over the same period in 2015, the fishery caught 133 tons of mackerel icefish in January, and 144 tons in February, with 70 and 51 tons of yellow-fin notothen bycaught in each month, respectively. Other bycatch included 1 ton of South Georgia icefish, 2 tons of gray rockcod and 4 tons of marbled rockcod, all of which were caught during the 2015 South Georgia groundfish survey.

(October-mid December), early-chick rearing (mid-December-end of January), late-chick-rearing (February onwards). The single sample collected at Macquarie Island during late chick rearing in 2015 was excluded (the DNA sequences were all from the family Bramidae; genus unknown).

#### Breeding Success and Use of Discards

The proportion of sampled birds that had consumed discards was estimated to range from zero at Bird Island to 60% at Steeple Jason. This is based on the conservative assumption that any species which was also available naturally to albatrosses was not considered to have been obtained as a discard (**Table 7**). Breeding success (chicks fledged/eggs laid) during the years that diet samples were collected ranged from zero at Albatross Islet to 84% at New Island. There was a positive correlation between the proportion of samples that contained


TABLE 7 | The proportion of scat samples that contained DNA from target and bycaught species in commercial fisheries operating in adjacent waters during the study.

These are the proportion of samples that amplified with the 16S\_Fish primer (∧), except the final two columns which is the proportion of all samples that contained food DNA amplified with the 18S\_SSU primers (\*). All listed fishery bycatch species constituted >1% of the total catch. Inaccessible fish species—those that occur below the dive depth of albatrosses; includes skates, toothfish, southern blue whiting, rock cod and hoki (Table S1 in Supplementary Material).

discarded fish, and breeding success (r = 0.81, p < 0.001; **Figure 7**).

#### DISCUSSION

This is the first study to use DNA metabarcoding to identify the fish prey diversity of seabird and use this to evaluate the occurrence of fishery discards in the diet across a broad geographic scale. This technique enabled us to identify an extensive diversity of fish in the diet of BBA, including a similar number of species and families as that recorded in all previous published studies for this species combined (31 families and 52 species). We also detected more fish species on average at each sampling site than in the conventional studies based on otoliths and similar numbers to studies using multiple body parts (Data Sheet 1 in Supplementary Material). There was a clear overlap between the species targeted by fisheries operating in adjacent waters, and the diet of BBA at the local colony. This was most evident at the Falkland Islands and Iles Kerguelen where the higher catch rates of target and bycatch species (or the amount discarded, where known) in a particular year corresponded with the relative occurrence in the diet. Our data also highlighted regions, such as South Georgia, where BBA diet overlapped with fishery target species, but the birds likely obtained the fish naturally. In this situation, there is the potential for resource competition but no reason to assume direct interaction with vessels or incidental mortality of foraging adults.

#### Amplification Success

The number of samples that contained food DNA in this study varied between sites and in some cases were quite low. There are numerous factors that can affect the amplification of food DNA including the primers/markers chosen, whether blocking primers were used, sample selection, timing during the breeding season and experience of the field personnel. We chose the combination of the universal metazoan marker (18S) and group specific markers (16S) to get a broad picture of the diet at the population level and specific information on the fish species consumed. Universal metazoan markers are useful dietary markers as they amplify DNA from all eukaryotes, which enables all possible prey groups to be identified. However, they also amplify nonfood DNA such as plant, parasite, and consumer DNA (McInnes et al., 2017a). A consumer blocking primer can increase the detection of food DNA (Vestheim and Jarman, 2008), but was not used in this study as they may inadvertently block similar groups such as other vertebrates like fish (Piñol et al., 2015). This likely reduced the sample size, but provided more reliable results from higher quality samples containing more food DNA. During our study, fresh samples were targeted and the inadvertent collection of dirt and vegetation was minimized where possible. However, with such a large study across a range of environmental conditions this was not always possible. In addition, many samples collected during incubation had little food DNA due to birds fasting. Subsequent to the data collection for this study, optimized scat collection protocols for DNA dietary metabarcoding have been developed that will hopefully improve the amount of DNA detected in future studies allowing diet data to be collected during all breeding stages (McInnes et al., 2017a).

#### Fish Prey Diversity

The fish prey consumed by BBAs varied considerably across their breeding range. Species in the family Nototheniidae were

common in scat samples from the sub-Antarctic sites, as were icefish (Channichthyidae) and horsefish (Congiopodidae). However, aside from the genus Patagonotothen, there were no nototheniids found in the samples from the Falkland Islands and Chile, whereas sprat (Clupeidae) was common. Of those species that contributed >10% of the sequences/samples at any site, 80% of these fish species were likely to be obtained naturally, as they are known to occur at depths accessible to albatrosses (maximum 4.5 m; Prince and Huin, 1994). The remaining species are not known to occur in waters shallower than 4.5 m and are hence likely to be obtained as discards.

This study detected several species that were not identified, or were very uncommon, in the diet of BBA in previous studies, particularly the Fueguian sprat, Antarctic horsefish, and southern driftfish. This was the first study of fish in the diet of BBA at Albatross Islet, and the first published study of BBA diet at Macquarie Island, which may explain some of these discoveries. Sprat was not recorded previously in the diet of BBA at any site (Data Sheet 1 in Supplementary Material), despite being the most common item in our study at the Falkland Islands and Chile. There was an unidentified clupeid in the diet at Diego Ramirez in 2001 and 2002 (Arata and Xavier, 2003), and a small unidentified fish at New Island in 1987 that made up 80% of the fish prey (Thompson, 1992), which may have been sprat. This species has a high biomass across the southern Patagonian shelf as far as the Magellan Strait (Sánchez et al., 1995), Chilean channel waters (Diez et al., 2012) and around the Falkland Islands (Agnew, 2002), and is common in the diet of other seabirds and

FIGURE 6 | Comparison between back-browed albatross fish prey and fishery catch amounts at the Falkland Islands by month from December 2013-March 2014 (excluding February) and October 2014 to March 2015. Solid borders represent New Island; dashed borders represent Steeple Jason. (A) Scats with or without fishery target species (black and gray bars, respectively), compared to the total catch in the fishery (blue line). (B) The proportion of scat samples containing each of the target species. (C) Total catches in the fishery by species. (D) The hierarchical clustering of albatross diet and fishery catch data by month, based on the proportion of sequences (RRA, black text) and proportion of catch (blue text). Clusters were based on dissimilarity indices calculated with the Manhattan method, and hierarchical clustering was constructed using the average agglomeration method (note low sample sizes during January 2014 and 2015). As food DNA may persist in scats for several days (Deagle et al., 2010), there may be some carry-over of prey caught in the previous month if samples were collected early in the month, which was the case in January of both years.

marine mammals in the region (Thompson, 1993; Baylis et al., 2014; Handley et al., 2016). There is also a sprat hotspot to the west of the Falkland Islands, close to both Steeple Jason and New Island, which may explain the prevalence in the diet at these sites (Gras et al., 2017).

Antarctic horsefish was the main fish species consumed at Macquarie Island and is endemic to the Macquarie and Kerguelen plateaus (Duhamel et al., 2014). Antarctic horsefish has been detected previously, but only in low frequency in BBA diets at Iles Kerguelen (Cherel et al., 2000, 2002). Very little is known about the abundance of horsefish or other fish around Macquarie Island. Horsefish have been detected in the diet of gentoo penguins (Pygoscelis papua; 39% FOO) and itinerant New Zealand sea lions at Macquarie Island (Phocarctos hookeri; 63% FOO; Robinson and Hindell, 1996; McMahon et al., 1999); however the majority of fish consumed by other seabirds and seals are myctophids, and to a lesser extent nototheniids (Goldsworthy et al., 2001).

The southern driftfish was detected in a quarter of samples at Bird Island in 2015 and one sample at Iles Kerguelen. It has only been recorded once in the diet of BBA at South Georgia, in 1986 (Reid et al., 1996; Croxall et al., 1997), and rarely in the diet of other seabirds (Croxall et al., 1995; Catard et al., 2000), though has been detected more commonly in seal diets (Guinet et al., 2001; Lea et al., 2008). Southern driftfish have a circumpolar distribution (Gon and Heemstra, 1990), although are rarely caught during trawls in the Scotia Sea (Collins et al., 2012) and none were recorded during a groundfish survey in January 2015, at the time when the scat samples were obtained (Belchier et al., 2015). It is surprising given our results that only one sample was detected in 20 years of conventional diet studies at South Georgia (British Antarctic Survey unpublished data). There are a few possible explanations: most of the previous studies were later in the season (February onwards) and represent chick diet, whereas our samples were from adults; alternatively, driftfish may be consumed as larvae and therefore the hardparts may be undetectable in stomach contents.

The other main fish prey at Bird Island and Iles Kerguelen were similar to the previous studies at each site (Data sheet 1 in Supplementary Material). At South Georgia, mackerel icefish are common prey (Prince, 1980; Reid et al., 1996; Croxall et al., 1997, 1999; Xavier et al., 2003). However, the diversity of fish in our study was much higher than in previous studies at Bird Island using only otoliths, which identified ten fish species overall, and less than five in any year (Data Sheet 1 in Supplementary Material). In comparison, we detected 16 species using DNA metabarcoding, with 13 in each year. Some of this diversity could relate to secondary ingestion; however, all of the species that contributed >10% of the diets (n = 8 and n = 7) were the sole prey item in at least one sample, suggesting they were the primary prey. At Iles Kerguelen, gray rockcod and unicorn icefish are two of the most abundant fish species in the Kerguelen EEZ (Duhamel and Hautecoeur, 2009) and were common in the diet of BBA at Canyon des Sourcils Noirs in a previous study (Cherel et al., 2000). When sample size differences were taken into account, the fish diversity was similar to previous studies at Iles Kerguelen where otoliths, bones and vertebrae were used (Cherel et al., 2000). In our study, there were some fish species identified in just one sample that are not usually found at those sites, such as Trematomus sp. at Iles Kerguelen. These could have originated from scats produced by juvenile or non-breeding birds, or as residual DNA from previous foraging trips far from the islands. For these reasons, we focused on fish species present in at least 10% of samples.

### Overlaps between Commercial Fisheries and Albatross Diet

There were five fish species detected in the scats that were unlikely to be naturally accessible to BBAs during the sampling period due to the known depth profile of fish. These included skates and Patagonian toothfish at Iles Kerguelen, and rockcod, hoki and southern blue whiting at the Falkland Islands. These species were present in fishery catches from the same time period, suggesting vessels were the likely source. At Iles Kerguelen, Patagonian toothfish and skates have no developmental stage where they have been observed at an accessible depth to albatross (Table S1 in Supplementary Material). Skates are demersal and the closest to the surface that toothfish have been recorded is during their larval stage (∼50 m depth), during winter and spring at Iles Kerguelen (Loeb et al., 1993; Mori et al., 2016). Patagonian toothfish were the most common fish in previous BBA dietary studies at Iles Nuageuses in 1994, and second most common at Canyon des Sourcils Noirs in 1994 (Cherel et al., 2000, 2002). These studies and others on wandering albatross (Diomedea exulans) suggest that albatross can consume Patagonian toothfish naturally (Weimerskirch et al., 1997), but how they obtain demersal prey is largely unknown (Cherel et al., 2000). It is possible that albatrosses scavenge prey brought up by deep-diving predators such as seals or whales (Sakamoto et al., 2009). In our study, the occurrence of Patagonian toothfish in the diet of birds from Iles Kerguelen did increase with an increase in discards, however, the amount of discarded toothfish from the fishery was low relative to the large albatross population. This result suggests that albatross may also be consuming Patagonian toothfish as natural prey as well as fishery discards during this study in unknown respective proportions. Although seabird bycatch rates in this fishery were very high in the 1990s and early 2000s (Delord et al., 2005), no albatross mortalities have been reported in recent years (CCAMLR, 2014a). This reflects the adoption of mitigation measures which include night setting, streamer lines, retention of offal during setting and fast hook sink rates (CCAMLR, 2014b). Discarding is still permitted in the Kerguelen toothfish fishery, which is still of concern. Discards increase vessel attractiveness and it is difficult to ensure mitigation is 100% effective for the smaller, more maneuvrable, deeperdiving species, particularly those such as white-chinned petrels (Procellaria aequinoctialis) which, unlike albatrosses, scavenge behind vessels in large numbers during darkness (Phillips et al., 2016). Moreover, individual birds will associate vessels with food, which is problematic if they overlap with fisheries under a different jurisdiction where there is poor compliance with seabird bycatch mitigation. Indeed, wandering albatrosses, which habitually follow vessels, may alter their flight path from 30 km away to approach a fishing vessel (Collet et al., 2015).

At the Falkland Islands, the frequency of rockcod and hoki in the diet corresponded to the relative fishery catches of these species, suggesting they were likely obtained as discards. Although this correlation could also reflect availability of fish stocks, these species are not known to be naturally accessible to albatross. Occurrence varied between sites and breeding stages, and was lowest during early chick rearing, which is consistent with the previous stable isotope study which found that pelagic fish were more common than demersal species (Granadeiro et al., 2013). During early chick-rearing, the fishery catch was zero to low, and therefore there was limited opportunity to exploit discards. However, during incubation and late chick rearing, the frequency of target fishery species in the albatrosses' diet was much higher. The occurrence of fishery species in BBA samples was greater at Steeple Jason than New Island, which is 70 km further south. This may be an artifact of the sampling month, given differences in timing; however, this does not explain all of the variation. The samples collected in the same month (e.g., January) were comparable, but no trawl fishery was operating. The few samples with fishery target species at New Island in November, when catch rates were relatively high, does not seem to match the trend at Steeple Jason for the preceding month. Previous studies at the Falklands also found more offal and discards in the diet of BBA at Steeple Jason than New Island, however, as there were few heads and therefore otoliths, the fish species could not be identified (Thompson, 1992). In the western part of the FICZ, there are two types of fishing grounds: one is to the northwest of Steeple Jason where trawlers target rockcod and one in deeper waters (>200 m) to the west-southwest which targets primarily hoki and southern blue whiting. Previous tracking studies found that Steeple Jason birds were more likely to attend vessels even when the distance to the fishing ground was similar for each colony (Granadeiro et al., 2011). Further research is needed to understand this observation.

The consumption of fishery discards by black-browed albatrosses at both sites in the Falklands puts birds at risk of incidental mortality. An estimated 800 BBAs are killed annually in Falkland Island trawl fisheries (Kuepfer, 2015). Although use of paired streamer lines is compulsory on all vessels, continuous discarding is still permitted (Quintin and Pompert, 2014). At the time of this study, the fishing fleet had limited capacity to retain offal on-board or process this into fishmeal; however, it has been recommended that any new vessels entering the fishery should have capabilities for more effective waste management (Sancho, 2009). Strict discard policies employed by trawl vessels operating in waters within the jurisdiction of CCAMLR have minimized exposure of birds to warp cables by retaining discards on-board until after shooting or hauling of fishing gear; consequently, the occurrence of incidental mortality is close to zero (CCAMLR, 2014b). Implementation of improved discard management measures around the Falkland Islands will be essential to reduce incidental mortality in trawl fisheries in the future (Abraham et al., 2009; Pierre et al., 2012).

### Competition with Fisheries and Reliance on Fishery Discards

South Georgian and mackerel icefishes were the two most common fish consumed by BBA at South Georgia in both years. Although, mackerel icefish is targeted by the fishery, and South Georgia icefish is bycaught, the BBA at Bird Island were likely to have obtained these species naturally. Very few fish were caught by the fishery during the diet sampling period of 2014 and they are known to occur at an accessible depth to albatross. Five other common prey species of BBAs at South Georgia are also caught in the icefish fishery, with bycatch limits set by CCAMLR (South Georgia icefish, marbled rockcod, yellow-fin notothen, humped rockcod, blackfin icefish). Mackerel icefish was the most common fish in the diet of BBA at South Georgia from 1996 to 2000 (Xavier et al., 2003) and in more recent years (British Antarctic Survey unpublished data). Icefish and BBA are both krill predators, and in years of low krill availability, icefish are likely to provide a valuable alternate food source for albatrosses (Reid et al., 1996). The BBA population at South Georgia is declining, and although this appears to be due mainly to incidental mortality during the non-breeding period (Poncet et al., 2017), their breeding success is also lower than conspecifics in the Indian Ocean (Nevoux et al., 2010). During our study, the proportion of krill in the diet was low (**Figure 2**), and over the last 20 years of conventional sampling (in mid-late chick rearing), krill has contributed <20% of the diet in only 4 years, two of which were 2014 and 2015 (18.5 and 5.6%, respectively; British Antarctic Survey unpublished data). Given the decline in krill, and high consumption by BBAs of species that are also targeted or bycaught in the icefish fishery, continued monitoring and evaluation of potential competition for resources is particularly important at this breeding site.

Another area of potential resource competition is off Chile, where there is currently a sprat fishery between 41 and 45◦ S, with annual catch limits of 26,000 tons (Leal et al., 2013). There was a proposal to expand this fishery into Chilean channel waters, where it would be likely to overlap with the foraging areas of BBAs from Chilean colonies. Given the importance of sprat in diets, any expansion of the fishery should consider the resource requirements of other marine species, including BBA, especially at the Albatross Islet colony where the foraging range is restricted (Arata et al., 2014). Globally, a third of fish stocks are fished at unsustainable levels (FAO, 2016), and fisheries are fishing down the food web (Pauly et al., 1998), including smaller fish species like sprat (Leal et al., 2013).

Although, competition with commercial fisheries could have a negative impact on albatrosses by reducing available prey, discards from fisheries can provide a supplementary food source (Bugoni et al., 2010; this study). In our study, breeding success was higher at colonies which had a greater occurrence of fishery discards in the diet samples. At the Falkland Islands where the occurrence of discards was high, the BBA population is increasing (Wolfaardt, 2013). Population increases at Chilean colonies have been attributed to a reduction in bird bycatch in longline fisheries (Robertson et al., 2017). However, high breeding success and a population increase could also reflect greater discard availability. Conversely, at Macquarie Island, where there is no local fishery operating during the breeding season, breeding success of BBAs was lower and the population is stable (ACAP, 2010), and at South Georgia, where the icefish fishery is small and provides few discards, BBAs have the lowest breeding success and the population is declining (Poncet et al., 2017). Many factors can impact breeding success, and a snapshot of diet over 2 years is not definitive. For example, the total failure at Albatross Islet was likely due to predation of eggs and chicks by American mink (Neovison vison; WCS unpublished data). However, availability of discards can influence seabird population trends (Foster et al., 2017), and DNA metabarcoding provides a means of further investigation.

Discards create an unnatural food-web structure, and if they are of low nutritional quality, there may be impacts on growth, breeding success and survival (Rosen and Trites, 2000; Grémillet et al., 2008). For BBA, the increasing population trend and high breeding success at sites where discards were common suggests that these were not nutritionally poor. However, discards could be sustaining an artificially high population and their removal might increase inter and intra-specific competition for available resources. Indeed, the European Union is phasing out the practice of discarding bycatch species and offal from 2015 to 2019, and there are concerns about the negative consequences for scavenging seabirds (Bicknell et al., 2013). Southern blue whiting was the main prey targeted by trawl fisheries around the Falklands Islands up until 2006, at which point the stock collapsed, and rockcod (Patagonotothen ramsayi) increased rapidly (Laptikhovsky et al., 2013). Rockcod is now the main target of the trawl fishery and one of the most common fish in the diet of BBAs at the Falklands during this study. Recent rockcod stock assessments indicate that this species is also beginning to decline (Gras et al., 2017). Monitoring the impact on BBA breeding success and their ability to switch to other resources will be important for assessing the degree to which they have been relying on discards. Similarly, improved discard management in the local trawl fisheries may have implications for the BBA population, particularly in the short-term, and any negative effects might be exacerbated by other threats such as climate change, habitat degradation, introduced pests, or disease, which affect many albatrosses globally (Phillips et al., 2016).

### CONCLUSIONS

This circumpolar study has revealed extensive fish diversity in the diet of BBA using DNA metabarcoding. Many of the fish species in the diet are not known to be naturally available to albatrosses, and were likely obtained by scavenging on discards (non-target fish, processing waste or used longline bait) from fisheries operating adjacent to the colony. Consumption of discards by black-browed albatrosses was detected from the Falkland Islands trawl fishery during incubation and late chick-rearing and from the Iles Kerguelen longline fishery during brood-guard. Our study indicates that improvements in discard management to reduce the attractiveness of vessels and hence incidental mortality of seabirds is likely to have major implications for some albatross populations. DNA metabarcoding of scat samples provides a non-invasive mechanism for quantifying and evaluating the level of interaction between seabirds and fisheries through identification of target and non-target fish, as well as the presence of baits. This provides an avenue for assessing compliance of fisheries with discard policies, and the effects on the level of interaction with scavenging seabirds.

### AUTHOR CONTRIBUTIONS

JM, RA, SJ, ML, and BR conceived and designed the project; RA, RP, AS, PC, HW, AK, contributed samples; JM and SJ designed primers; JM performed laboratory work; JM and BD performed bioinformatics; JM and BR performed statistical analyses; MG, DM, and YC provided fishery data or information on fish biology; JM wrote the first draft of the manuscript, and all authors contributed substantially to revisions.

#### ACKNOWLEDGMENTS

This project was approved by the University of Tasmania Animal Ethics Committee (Permit A13745). Funding was provided by Australian Antarctic Science Grant (4014 and 4122) and the Winifred Violet Scott Charitable Trust; further funding was received from the Falkland Islands Government and from FCT— Portugal through the strategic project UID/MAR/04292/2013 granted to MARE. Falkland Islands fishery catch data were provided by the Directorate of Natural Resources—Fisheries Department of the Falkland Islands Government. Iles Kerguelen fishery data were provided through the Pecheker database with thanks to Guy Duhamel, Nicolas Gasco, Alexis Martin, Patrice Pruvost and Charlotte Chazeau. Macquarie Island data were provided by the Australian Fisheries Management Authority with assistance from Dirk Welsford at the Australian Antarctic Division. South Georgian fishery data was obtained through CCAMLR statistical bulletins. Thanks to the large number of field personnel for scat collections including Javier Arata; fishery observers for obtaining catch data; Mark Belchier and Phillipe Koubbi for advice regarding fish diversity data; the Wildlife Conservation Society for access to Steeple Jason Island and permission to collect samples; and James Marthick and the Menzies Institute (UTAS) for the use of the Miseq Genome Sequencer.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fmars. 2017.00277/full#supplementary-material

#### Data Availability

Fish sequences not currently of Genbank were added including: Sprattus fuegensis, Genypterus blacodes, Iluocoetes fimbriatus, Salilota australis, Icichthys australis, Anotopterus vorax, Halargyreus johnsonii (GenBank accession numbers MF346066-074).


over the Patagonian Shelf. Anim. Conserv. 17, 19–26. doi: 10.1111/acv. 12050


**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 © 2017 McInnes, Jarman, Lea, Raymond, Deagle, Phillips, Catry, Stanworth, Weimerskirch, Kusch, Gras, Cherel, Maschette and Alderman. 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) or licensor 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.

# Assessing the Functional Limitations of Lipids and Fatty Acids for Diet Determination: The Importance of Tissue Type, Quantity, and Quality

Lauren Meyer 1, 2 \*, Heidi Pethybridge<sup>2</sup> , Peter D. Nichols <sup>2</sup> , Crystal Beckmann<sup>3</sup> , Barry D. Bruce<sup>2</sup> , Jonathan M. Werry <sup>4</sup> and Charlie Huveneers <sup>1</sup>

<sup>1</sup> College of Science and Engineering, Flinders University, Bedford Park, SA, Australia, <sup>2</sup> CSIRO Oceans and Atmosphere, Hobart, TAS, Australia, <sup>3</sup> SARDI Aquatic Sciences, South Australian Research and Development Institute, West Beach, SA, Australia, <sup>4</sup> Griffith Centre for Coastal Management, Griffith University, Gold Coast, QLD, Australia

#### Edited by:

Jeremy Kiszka, Florida International University, United States

#### Reviewed by:

Luis Cardona, University of Barcelona, Spain Valentina Franco-Trecu, Facultad de Ciencias, Universidad de la República, Uruguay

> \*Correspondence: Lauren Meyer lauren.meyer@flinders.edu.au

#### Specialty section:

This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science

Received: 18 August 2017 Accepted: 31 October 2017 Published: 24 November 2017

#### Citation:

Meyer L, Pethybridge H, Nichols PD, Beckmann C, Bruce BD, Werry JM and Huveneers C (2017) Assessing the Functional Limitations of Lipids and Fatty Acids for Diet Determination: The Importance of Tissue Type, Quantity, and Quality. Front. Mar. Sci. 4:369. doi: 10.3389/fmars.2017.00369 Lipid and fatty acid (FA) analysis is commonly used to describe the trophic ecology of an increasing number of taxa. However, the applicability of these analyses is contingent upon the collection and storage of sufficient high quality tissue, the limitations of which are previously unexplored in elasmobranchs. Using samples from 110 white sharks, Carcharodon carcharias, collected throughout Australia, we investigated the importance of tissue type, sample quantity, and quality for reliable lipid class and FA analysis. We determined that muscle and sub-dermal tissue contain distinct lipid class and FA profiles, and were not directly comparable. Muscle samples as small as 12 mg dry weight (49 mg wet weight), provided reliable and consistent FA profiles, while sub-dermal tissue samples of 40 mg dry weight (186 mg wet weight) or greater were required to yield consistent profiles. This validates the suitability of minimally invasive sampling methods such as punch biopsies. The integrity of FA profiles in muscle was compromised after 24 h at ambient temperature (∼20◦C), making these degraded samples unreliable for accurate determination of dietary sources, yet sub-dermal tissue retained stable FA profiles under the same conditions, suggesting it may be a more robust tissue for trophic ecology work with potentially degraded samples. However, muscle samples archived for up to 16 years in −20◦C retain their FA profiles, highlighting that tissue from museum or private collections can yield valid insights into the trophic ecology of marine elasmobranchs.

Keywords: biochemical tracer, elasmobranch, biopsy, trophic ecology, white shark, Carcharodon carcharias

### INTRODUCTION

The field of trophic ecology has seen a substantial increase in the number of available techniques and applications across aquatic and terrestrial taxa within the last half century (Layman et al., 2012, 2015; Christiansen et al., 2015; Nielsen et al., 2015; Young et al., 2015; Roslin and Majaneva, 2016). More recently, there has been a growing number of studies moving from traditional stomach-content analysis, which may provide a potentially limited view due to differences in digestibility among prey species (Hyslop, 1980), to time-integrated biochemical methods (reviewed in Traugott et al., 2013; Pethybridge et al., 2018). Lipid and fatty acid (FA) analysis is one such method growing in popularity as it has the capacity to elucidate key biological and ecological aspects, such as an organism's physiology and bioenergetics (Parrish et al., 2007; Pond and Tarling, 2011), and most often, trophic relationships (e.g., Bradshaw et al., 2003; Iverson et al., 2004; Budge et al., 2006). As per the saying "you are what you eat," certain FAs are transferred from prey to predator with minimal modification (Iverson et al., 2004; Budge et al., 2006), allowing certain functional trophic groups to be traced within a food chain. Owing to this broad applicability, more than 29,000 published studies featured FA analysis for marine and aquatic taxa alone between 1990 and 2014 (Rudy et al., 2016).

The applicability of FA analysis is especially pertinent for threatened and iconic species for which lethal sampling, which is often used to obtain stomach contents, is not possible especially for large numbers of specimens. Instead, minimally invasive biopsy techniques are often employed to obtain tissue samples for biochemical studies (e.g., Hooker et al., 2001; Carlisle et al., 2012; Hussey et al., 2012). With the development of specialized biopsy probes (Reeb and Best, 2006; Robbins, 2006; Daly and Smale, 2013), tissue samples can be obtained from free-swimming marine organisms, reducing the stress and detrimental effects of the capture and release process, and enabling the increased use of FA analyses across a number of species, including threatened elasmobranchs (Couturier et al., 2013; Rohner et al., 2013; Every et al., 2016).

The accuracy and reliability of biochemical analyses are dependent on the methods used to collect and store samples. Sampling elasmobranchs in particular poses a series of logistical challenges, due in part to the large proportion of species considered at risk of extinction (Dulvy et al., 2014), leading to samples often being difficult and expensive to obtain. As a result, these samples are often highly valuable and one needs to understand the functional limitations of collecting and storing these tissues to maximize sampling opportunities and reliability of resulting data.

The increasing use of biopsies to collect tissues from elasmobranchs has led to constraints on the type, amount, and quality of tissue collected. Beneath the epidermis, elasmobranchs contain a deep sub-dermal layer of collagen and elastin fibers, which varies in thickness between species (Motta, 1977). The underlying physiological differences between the two tissue types (muscle, a metabolically active and protein-rich tissue vs. sub-dermal tissue, a less bioactive and largely structural tissue composed of elastin and collagen) results in distinct biochemical properties, with the potential to yield different ecological data. This is evidenced by recent isotopic studies on white sharks, Carcharodon carcharias, whereby muscle and subdermal tissue had the same <sup>15</sup>N isotopic signatures, but divergent <sup>13</sup>C signatures, which was attributed to differing tissue-specific incorporation rates (Carlisle et al., 2012; Kim et al., 2012; Jaime-Rivera et al., 2013). How these tissue-specific physiological and biochemical differences manifest in FA profiles remains poorly studied, with most elasmobranch work to date focused on the FA differences between skeletal muscle and the lipid-rich liver (e.g., Schaufler et al., 2005; Pethybridge et al., 2011; Beckmann et al., 2013), myocardial tissue (Davidson et al., 2011, 2014), and blood plasma (Ballantyne et al., 1993; McMeans et al., 2012). However, Every et al. (2016) recently showed differences in FA profiles between muscle tissue and fin clips (a mixed-tissue sample, including cartilage, connective tissue, muscle, vascularization and an outer dermal layer with denticles).

The functional limitations of various biopsy methods also extend to the amount of tissue obtained. With the thick epidermal layer serving as a barrier, collecting sufficient amounts of usable muscle from large elasmobranchs in particular, has proven challenging. The sub-dermal layer of white sharks can be up to 3 cm, hindering the ability to collect the underlying muscle (Jaime-Rivera et al., 2013). Whale sharks, Rhincodon typus, sampled with a biopsy probe penetrating ∼2 cm yielded exclusively sub-dermal tissue (Rohner et al., 2013), whereas the ∼2 cm biopsies of bull sharks, Carcharhinus leucas yielded 5% dermis, 40% sub-dermal and 55% muscle (Daly and Smale, 2013). These differences in the thickness of the sub-dermal layer complicate the collection of elasmobranch muscle samples. Although small amounts of tissue are sufficient for genetic [1 mg dry weight (DW), Kasajima et al., 2004] and stable isotope analysis (∼10 mg DW, Jaime-Rivera et al., 2013), the minimum amount of muscle or sub-dermal tissue necessary for accurate FA analysis remains relatively unknown. Although Every et al. (2016) reported that FA were detectable in fin clips as small as 20 mg and muscle biopsies >10 mg dry weight, minimum sample sizes yielding consistent results were not quantitatively assessed. Such evaluations are vital however, particularly when considering the appropriateness of various biopsy probes, and the applicability of the sampling method across smaller elasmobranch species, from which removing large amounts of tissue is not feasible.

Appropriate sample acquisition, storage and tissue preservation is vital when applying FA analysis techniques, as certain FAs (particularly long-chain(≥C20) polyunsaturated FAs, LC-PUFAs) oxidize when exposed to air, high temperatures, and direct sunlight, leading to tissue degradation and loss of information (Budge et al., 2006). This becomes particularly challenging when there are scarce opportunities for sampling (e.g., for highly mobile, rare, or cryptic species) and when working in remote and hostile field locations (e.g., hot and humid tropics, and offshore sampling sites). Furthermore, despite the growing utilization of non-lethal biopsies, many FA studies use samples taken from deceased elasmobranch carcasses obtained from fisheries bycatch (Pethybridge et al., 2011), beach strandings (Rohner et al., 2013), and shark-control measures (Davidson et al., 2011, 2014; Pethybridge et al., 2014). Given the variable condition of these carcasses, which may have spent multiple days at ambient temperature, there is the high potential for lipid and FA degradation within samples collected via these means. Additionally, FA studies often use tissue samples collected over a long period of time (e.g., 5 years—Davidson et al., 2011, 2014; 2 years—Rohner et al., 2013; 12 years— Pethybridge et al., 2014 and 3 years—Jaime-Rivera et al., 2014), providing another opportunity for unchecked FA degradation throughout these long periods of frozen storage. Several recent studies examining storage procedures have revealed significant species- and tissue-specific lipid and FA degradation over the course of several months held at −20◦C (Refsgaard et al., 1998; Roldán et al., 2005; Phleger et al., 2007; Sahari et al., 2014; Paola and Isabel, 2015; Rudy et al., 2016). To date, the focus of such investigations have remained limited to highly valued commercial teleost (Roldán et al., 2005; Paola and Isabel, 2015; Rudy et al., 2016) and cephalopod species (Gullian-Klanian et al., 2017). Despite this evidence of FA degradation, it remains unassessed for the many archived elasmobranch tissues stored over the period of months to years.

Given the aforementioned lack of information regarding the functional limitations and capabilities of lipid and FA biomarkers for application to highly mobile, rare or cryptic elasmobranchs, this study seeks to assess:


The knowledge gained from addressing these functional limitations will facilitate the more effective use of lipid and FA profiling on biopsied or potentially degraded tissues, allowing them to be employed with greater confidence in a range of ecological studies.

### MATERIALS AND METHODS

#### Sample Collection and Data Compilation

Tissue samples were collected from 110 white sharks, C. carcharias from South Australia (SA), New South Wales (NSW), and Queensland (QLD), Australia between 2000 and 2016 (**Table 1**). Tissues were obtained through punch-biopsies of live, free-swimming white sharks from the Neptune Islands, SA, opportunistically through fisheries bycatch, the NSW Department of Primary Industries Shark Meshing Program and QLD Department of Agriculture and Fisheries Shark Control Program as part of the QLD large shark tagging research program. Samples were frozen and stored from 3 weeks to 16 years at −20◦C, until freeze-drying immediately prior to lipid analysis.

#### Ethics Statement

In South Australia, fieldwork at the Neptune Islands was carried out in accordance with ethics permit #E398, approved by The Flinders University Animal Welfare Committee, and under DEWNR permit # Q26292. In New South Wales, tissue collection under NSW DPI Scientific Collection Permit (P07/0099-3.0 and P07/0099-4) was approved by New South Wales Department of Primary Industries (NSW DPI) Animal Research Authority (ACEC 12/07). Tissue from Queensland was obtained as part of the QLD Shark Meshing Program and QLD Department of Agriculture and Fisheries Shark Control Program as part of the QLD large shark tagging research program under fisheries permit 143005 and QLD Department of Agriculture and Fisheries Shark Animal Ethics Committee approved ethics CA 2010/11/482, CA 2013/11/737, ENV 1709 AEC.

#### Experimental Design

Three sets of comparative lipid and FA analyses were undertaken, each addressing one of the aims; the difference between muscle and sub-dermal tissue, minimum tissue quantity for each tissue, and the effect of tissue degradation on resulting lipid and FA profiles (**Table 1**). To investigate the difference between the muscle and sub-dermal tissue, ∼300 g sections, comprising both muscle and sub-dermal tissue were collected from three deceased white sharks (a, b, and c). Lipid class and FA profiles were assessed across triplicate subsamples from these three sharks (**Table 1**) to incorporate the within-individual variability. Minimum tissue quantity was also assessed in triplicate, across the three sharks, for both muscle and sub-dermal tissue using progressively smaller samples sizes. The tissue degradation analysis was performed in three parts: (i) at ambient temperature, and (ii) short term

TABLE 1 | Sample details across the three study aims, including the number of individual white sharks Carcharodon carcharias and the tissue and lipid parameter analyzed.


DW, Dry weight.

<sup>a</sup>White sharks from the Neptune Islands, SA and NSW.

<sup>b</sup>White sharks from NSW and QLD.

storage at −20◦C (for up to 2 years), and (iii) long-term storage at −20◦C (for up to 16 years). The remaining portions of sharks a and b were then held at room temperature (∼20◦C) for 4 days, and muscle and sub-dermal tissue were sub-sectioned in triplicate, every 24 h. Immediately prior to sub-sectioning, ∼1 cm of the outermost edge was removed and discarded, allowing the sample to be taken from the interior of the tissue section. This was to minimize incidentally measuring the co-occurring effects of oxygen-contact induced FA oxidation on the samples. Only sharks a and b underwent the ambient temperature degradation trial, as there was insufficient remaining tissue from shark c.

The remaining 107 white shark muscle samples were used to assess both short- to mid-term (1 month up to 2 years) and long-term (1 month up to 16 years) FA profile degradation associated with storage at −20◦C (**Table 1**). Forty-five samples from the Neptune Islands, SA and 10 of the 31 samples from NSW were processed within 2 years of being obtained and thus these were assessed together for short- to mid-term degradation (1 month up to 2 years). These results were grouped into 3 months bins for statistical analysis. Sixty-two muscle samples (31 from NSW, 31 from QLD) were assessed together for longterm freezer degradation (1 month up to 16 years). This excluded the 45 Neptune Islands samples included in short-term freezer degradation analysis, limiting the potential confounding factor of collection location within long-term degradation. These longterm freezer degradation results were also grouped into bins for statistical analysis, with group 1 = 0–1 years at −20◦C, 2 = 1.1–2 years, 3 = 3–5 years, 4 = 6–10 years, 5 = 11–16 years.

### Lipid Extraction

Total lipid was extracted using the modified Bligh and Dyer method (Bligh and Dyer, 1959). Briefly, samples were left overnight in a one-phase CH2Cl2:CH3OH:milliQ H2O mixture (10:20:8 mL) before the solution was broken into two phases by the addition of 10 mL CH2Cl<sup>2</sup> and 10 mL of 9 g NaCL L−<sup>1</sup> saline milliQ H2O. The lower phase containing the lipid fraction was drained into a round bottom flask and the solvent removed using a rotary evaporator. The lipid was re-suspended in CH2Cl<sup>2</sup> and transferred to a 2 mL vial and dried under N<sup>2</sup> gas until a constant weight was noted. The total lipid extract (TLE) was then re-suspended in 1.5 mL of CH2Cl2.

#### Lipid Content and Class Analysis

Water content, reported as percent of tissue wet weight, was determined for each sample by taking weights before and after freeze-drying at −82◦C for 72 h and calculating the wet to dry ratio. Similarly, the lipid content was calculated by subtracting tissue dry weight prior to lipid extraction from the weight of the resulting TLE, then multiplied by the wet to dry ratio, and reported as percent of tissue wet weight.

Lipid class composition [triacylglycerols (TAG), phospholipids (PL), sterols (ST), wax esters (WE), and free fatty acids (FFA)] were measured using an Iatroscan Mark V TH10 thin layer chromatrograph coupled with a flame ion detector (TLC-FID). TLE from each sample was analyzed in triplicate. Aliquots of TLE were spotted onto chromarods and developed for 25 min in a polar solvent system [70:10:0.1 v/v/v, C6H14:(C2H5)2O:CH3COOH]. Rods were oven dried at 100◦C for 10 min and analyzed immediately. SIC-480 Scientific Software was used to identify and quantify the areas of the resulting peaks.

### Fatty Acid Analysis

An aliquot of the TLE was transferred into a teflon-lined screw cap glass test tube and trans-methylated with 3 mL of CH3OH: CH2Cl2:HCl (10:1:1 v/v/v) for 2 h at 80◦C. The tube was then cooled in a water bath, and 1 mL MilliQ H2O was added. The resulting fatty acid methyl esters (FAME) were extracted into a 2 mL glass vial using three washes of C6H14: CH2Cl<sup>2</sup> (4:1 v/v), each thoroughly mixed and then the tube centrifuged at 2,000 rpm for 5 min. The resulting FAME were dried under N<sup>2</sup> gas prior to the addition of 1.0 mL of C<sup>19</sup> internal injection standard solution in preparation for gas chromatography (GC) and GC-mass spectrometry (GC-MS) analysis.

Each FAME sample was injected into an Agilent Technologies 7890B GC (Palo Alto, California USA) equipped with an Equity-1 fused silica capillary column (15 m × 0.1 mm internal diameter and 0.1 mm film thickness), a flame ionization detector, a splitless injector, and an Agilent Technologies 7683B Series auto-sampler. At an oven temperature of 120◦C, samples were injected in splitless mode and carried by helium gas. Oven temperature was raised to 270◦C at a rate of 10◦C per min, and then to 310◦C at a rate of 5◦C per min. Peaks were quantified using Agilent Technologies ChemStation software (Palo Alto, California USA). The identities of the peaks were confirmed using a Finnigan Thermoquest DSQ GC-MS system. All FAs were converted from chromatogram peak area to percentage of total area.

#### Statistical Analysis

Of the 50 total FAs detected, 21 (with averages >0.1% of total FAs across either tissue type, in quantities of 100 mg non-degraded muscle and 80 mg of non-degraded sub-dermal tissue) were used for multivariate analysis comparing the differences in profiles across factors. Statistical analysis was undertaken in PRIMER 7 (Plymouth Routines in Multivariate Ecological Research, Clarke et al., 2014) +PERMANOVA. We used Principal Coordinates Analysis (PCO) of Bray-Curtis similarity matrices calculated from the square-root transformed data to determine clustering of individual samples. To test the differences between factors we used permutational analysis of variance (PERMANOVA) with Monte Carlo simulations denoted as p(MC) on the unrestricted raw values to account for the small sample sizes. PERMANOVA analyses used factors nested within shark to incorporate the triplicate samples from each individual shark. Significance was determined by p < 0.05. Following significant ANOSIM tests, similarity percentage (SIMPER) analyses were undertaken to quantify the contribution of each parameter to the separation between the designated groups.

Additionally, the sum of the saturated (SFA), monounsaturated (MUFA), total polyunsaturated fatty acids (PUFA), ω3 PUFA and the ratio of ω3 PUFA:ω6 PUFA and EPA+DHA/16:0 were calculated per replicate. We used nested (factor within shark) PERMANOVA analysis with Monte Carlo simulations to assess the response of individual lipid classes, FA values, and FA metrics (aforementioned sums and ratios). Permutational analysis of multidimensional dispersion PERMDISP denoted at p(perm) was used to determine the relative amount and statistical significance level of the dispersion within factor groups.

#### RESULTS

#### Muscle vs. Sub-dermal Tissue

White shark muscle was high in water content 82.1 ± 1.1% wet weight (WW) and low in lipid content (0.6 ± 0.1% WW), with a wet to dry ratio of 4.1 ± 0.2. Sub-dermal tissue contained even lower amounts of total lipid (0.4 ± 0.2% WW), which was on average 33% less lipid than the muscle tissue.

The lipid class profiles of both tissues were dominated by PL (**Table 2**) followed by ST, which were 13.5% (as % of total lipid) more abundant in sub-dermal tissue than muscle. ST contributed the greatest source of dissimilarity between the tissue types (46%) as determined by SIMPER, and when assessed individually, was the only lipid class significantly different between the tissues [p(MC) = 0.001] (**Table 2**).

Muscle tissue contained primarily PUFA 39.2 ± 8.0%, mostly consisting of 22:6ω3 (docosahexaenoic acid, DHA) and 20:4ω6 (arachidonic acid, ARA) (**Table 2**). SFA contributed 33.9 ± 4.7%, dominated by 16:0 and 18:0. MUFA contributed the remaining 21.8 ± 4.2% of the muscle tissue FA profile, nearly half of which was 18:1ω9. Sub-dermal tissue contained similar relative levels of PUFA (32.8 ± 3.5%) dominated by 20:4ω6 and 22:6ω3, and SFA (33.3 ± 3.2%) mostly 18:0 and 16:0, with MUFA (26.1 ± 2.7%) primarily consisting of 18:1ω9 (**Table 2**).

Muscle and sub-dermal tissue comparisons resulted in distinctly different FA profiles [Nested PERMANOVA: shark p(MC) = 0.439, tissue p(MC) = 0.001, **Figure 1**]. The difference was primarily driven by high levels of 22:6ω3 in the muscle (SIMPER 17% dissimilarity contribution), followed by 18:1ω7 (8.4%), 20:4ω6 (6.2%), and i15:0 (5.6%) (**Table 2**). Sixteen of the 21 individual FAs were found to be significantly different [p(MC) < 0.05] across the two tissue types (**Table 2**), with only 16:0, 18:0, 20:1ω9, 16:3, and 22:4ω6 not significantly different between the two tissues.

Muscle tissue samples showed greater dispersion than the sub-dermal tissue [p(perm) = 0.030; **Figure 1**] across the three individual sharks. However, this difference in tissue-specific dispersion was not seen within the three triplicate samples of sharks a, b, and c [Shark a p(perm) = 0.600, Shark b p(perm) = 0.456, Shark c p(perm) = 0.812].

#### Minimum Sample Size

The progressively smaller muscle tissue increments (100, 50, 25, and 12 mg DW) showed no statistical difference between size groups [p(MC) = 0.28], or difference in dispersion (PERMDISP means of 5.2, 4.1, 5.3, 6.6 for the 100, 50, 25, and 12 mg samples, respectively, p > 0.05). Principal coordinates analysis showed that the clustering is not driven by tissue amount, but by individual shark (Nested PERMANOVA p(MC) = 0.28 nested within shark p(MC) = 0.001), with shark c separating from sharks a and b (**Figure 2A**).

TABLE 2 | Total lipid content, relative proportions of lipid classes and fatty acids (FA) (as percent of total lipid or FA) (mean ± standard deviation) of muscle and sub-dermal tissue (wet weight, WW) from Carcharodon carcharias.


p(MC)-values were determined by Nested PERMANOVA with Monte Carlo simulation, with tissue nested within shark.

SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids. The suffix i denotes branched fatty acids from the iso-series. FALD- fatty aldehyde analyzed as dimethyl acetal.

Data presented are for 21 components, with a cut off of 0.5%.

p(MC) indicated the p value determined by PERMANOVA run with Monte Carlo simulations.

#, Statistical significance determined by p(MC), denoted by: \*p < 0.05, \*\*p < 0.01, \*\*\*p < 0.001, NS, Not significant (p > 0.05).

Sub-dermal tissue increments (85, 40, 20, and 10 mg DW) revealed differing FA profiles with decreasing tissue amounts [p(MC) = 0.042 for tissue size], with the difference between the two larger (85 and 40 mg) and two smaller (20 and 10 mg) amounts driven by 18:1ω9, i15:0, 22:6ω3, and 20:4ω3 (**Table 3**).

The difference in FA profiles is exacerbated by an increase in dispersion with decreasing tissue size (**Figure 2B**), particularly between the two smaller 10 and 20 mg tissue samples and the two larger 85 and 40 mg sample sizes (**Table 3**).

variation attributed to each axis (PCO1 and PCO2).

#### Lipid Class and FA Degradation at Ambient Temperature (20◦C)

The lipid class profiles from the muscle tissue showed no differences across the 4 day period at 20◦C [p(MC) = 0.127]. However, the muscle tissue showed a significant shift in FA profile over the 4 day period at 20◦C [p(MC) = 0.009], with significant (p < 0.05) differences between the fresh samples and days 1, 2, and 3 (**Figure 3A**). This was mostly driven by changes in 18:4ω3, 22:6ω3, and 18:0 (SIMPER analysis). PERMANOVA analysis of individual FA found significant differences in 18:0, total SFA, 18:4ω3, 18:2ω6, 20:5ω3, 22:6ω3, total PUFA, total ω3 PUFA, and the ω3:ω6 ratio, but not 20:5ω3 + 22:6ω3/16:0 (EPA+DHA/16:0).

Additionally, PERMDISP analysis revealed a significant decrease in dispersion when the tissue was left at ambient temperature [p(perm) = 0.03]. The mean dispersion for the fresh tissue (4.9) was significantly larger than the 2.0, 2.1, and 2.0 dispersion means for days 1, 2, and 3, respectively (p < 0.05), but not significantly different than the 3.0 dispersion mean at day 4 [p(perm) = 0.104]. Similar to the muscle lipid class profile, the sub-dermal tissue did not show any significant differences across the 4 day period at 20◦C [p(MC) = 0.183]. There was also no discernible shift in FA profile over the 4 day period [p(MC) = 0.141; **Figure 3B**]. Unlike the muscle tissue, there were no differences in the level of dispersion between the groups [overall p(perm) = 0.631].

## FA Degradation of Frozen Tissue (−20◦C)

The FA profiles showed distinct degradation across the 24 months spent in the −20◦C freezer, regardless of the location where the sharks were caught [location p(MC) = 0.317; time in freezer nested within location p(MC) = 0.008]. Within group comparisons reveal differences primarily between group 2 (3–6 months in the freezer) and all other groups, aside from group 1. Group 1 (0–3 months in the freezer) was only different to group 7 (the 19–21 month period) (**Table 4**). SIMPER analysis reveal that these differences were driven largely by 18:0, 22:6ω3, 18:2ω6, 16:0, and 18:4ω3 across the groups. Similar to the unfrozen, controlled muscle degradation trial, the total FA profile degradation manifests in changes to the level of dispersion, which decreases significantly with the amount of time spent in the freezer [p(perm) = 0.001, **Table 4**].

When assessing freezer-based degradation of archived samples over a long time frame (up to 16 years), there was slight discernible degradation, however, the capture location of the white sharks was more highly significant than period in the freezer [p(MC) = 0.002 vs. 0.045]. For the sharks captured in NSW, none of the group level comparisons showed significant degradation [all p(MC)-values > 0.05], and within the QLD samples, only the difference between group 2 and 4 (1.1–2 years and 5.1–10 years) was significant [p(MC) = 0.041]. Unlike the short-term freezer degradation and the unfrozen muscle degradation trial, there was no decrease in dispersion with the longer storage period [p(perm) = 0.620].

#### DISCUSSION

Lipid class and FA analysis are increasingly used to describe the trophic ecology of a range of species, including elasmobranchs, necessitating greater understanding of the functional limitations of collection and storage methodologies. Here, we determined that muscle and sub-dermal tissue were not directly comparable, as they had tissue-specific lipid class and FA profiles. We also provide the first estimation of the minimum amount of muscle and sub-dermal tissue required to provide reliable FA profiles, which validated the suitability of minimally invasive sampling methods such as punch biopsies. Additionally, we determined that muscle tissue stored at ambient temperature was compromised after as little as 24 h, making muscle samples from beach strandings and fisheries bycatch potentially unreliable for accurate determination of dietary sources. Yet, sub-dermal tissue retained stable FA profiles under the same conditions, suggesting it may offer a more robust tissue for trophic ecology work with potentially compromised samples. However, muscle samples archived for up to 16 years in −20◦C retain their FA profiles, highlighting that muscle tissue from museum or private collections can yield valid insights into the trophic ecology of marine elasmobranchs. Knowledge gained from addressing these functional limitations will facilitate the more effective use of lipid and FA profiling on biopsied or potentially degraded tissues for the white shark, and in addition for other species, allowing them

FIGURE 2 | Principal coordinates analysis (PCO) of the fatty acid profiles from white shark muscle and sub-dermal tissue across differing tissue sizes. Principal coordinates analysis (PCO) of (A) muscle, and (B) sub-dermal tissue from three white shark Carcharodon carcharias individuals (a–c), analyzed in triplicate across differing tissue sizes in mg dry weight (DW). Eigenvalues denote the percent of variation attributed to each axis (PCO1 and PCO2).

FIGURE 3 | Principal coordinates analysis (PCO) of the fatty acid profiles from white shark muscle and sub-dermal tissue across 4 days of degradation at 20◦C. Principal coordinates analysis (PCO) of fatty acid profiles from (A) muscle, and (B) sub-dermal tissue from two white shark Carcharodon carcharias individuals (a and b), analyzed in triplicate across 4 days of degradation (0, indicating fresh tissue, 1, 2, 3, and 4 indicating the number of days left at 20◦C prior to analysis). Eigenvalues denote the percent of variation attributed to each axis (PCO1 and PCO2).

to be employed with greater confidence in a range of ecological studies.

#### Muscle vs. Sub-dermal Tissue

The lipid classes of the muscle tissue, dominated by PL (87%), were consistent with previously reported values for white sharks (92 ± 5%, Pethybridge et al., 2014) whereas the sub-dermal tissue contained higher relative levels of sterols (ST), closely resembling the profile of whale shark sub-dermal tissue (21 ± 4%, Rohner et al., 2013). Regardless of ST contribution, both tissues were dominated by PL, with relatively little contribution from the neutral lipids (triacylglycerols, wax esters, FFA) responsible for metabolic energy storage (Sargent et al., 1999). This affirms the understanding that both muscle and sub-dermal tissue contain little capacity for metabolic energy storage, unlike elasmobranch livers, which are high in lipid content and dominated by triacylglycerols (Beckmann et al., 2013; Pethybridge et al., 2014).

Tissue differences across 16 of the 21 FAs (contributing >76% of total FA) are likely a reflection of divergent functions and underlying physiology. For example, 22:6ω3 and other key essential FAs including 18:2ω6, 20:4ω6 (ARA), 20:5ω3 (EPA), which serve as indicators for a range of trophic pathways differed between the two tissues. As such, the variation in FAs that accounted for the separation between muscle and sub-dermal tissue indicates that interpretation of a species' diet would be greatly affected by the tissue from which the FA profiles is derived,

TABLE 3 | The differences between sub-dermal tissue sizes for white shark Carcharodon carcharias.


p(MC) values were determined by Nested PERMANOVA with Monte Carlo simulation (three replicates nested within three sharks, n = 3) and the primary fatty acids (FA) driving the significantly different groups determined by SIMPER percent contribution. FAs are listed in order of decreasing contribution. Listed PERMDISP p values indicate the significance of the differences in dispersion between the tissue sizes. \*p < 0.05, \*\*p < 0.01, \*\*\*p < 0.001.

and thus the profiles of the different tissues are not directly comparable.

Recent studies have suggested that differences in FA profiles between muscle and sub-dermal tissue of euryhaline elasmobranches are species-specific (Every et al., 2016). However, when we include the FA profiles of manta ray muscle from Couturier et al. (2013) and whale shark sub-dermal tissue, from Rohner et al. (2013) in the PCO with our white shark samples, the manta ray and whale shark FA profiles align with the tissuespecific clusters (**Figure 1**). This suggests that the difference in FA profiles between muscle and sub-dermal tissues are not limited to white sharks, but extends to other species and across trophic levels.

The sub-dermal tissue serves as a key structural component, with a slower metabolic turnover rate than muscle (assessed in relation to divergent isotopic signatures by del Rio et al., 2009). As such, these tissues may therefore present complementary results, reflecting diets incorporated across different time frames (Every et al., 2016). Given the opportunity to collect both tissue types through non-lethal biopsies, further investigations comparing the tissue-specific FA incorporation rates should be undertaken. Results discerning the time-frame of both tissue's FA profiles would provide the opportunity to assess multiple temporal scales of an individual's trophic history, valuable additional information when investigating individual specialization, location specific, seasonal, or ontogenetic dietary shifts.

### Minimum Samples Size

Muscle biopsies of variable forms have previously been developed to collect samples for genetic and isotopic studies, e.g., punch biopsies (Robbins, 2006; Daly and Smale, 2013) or thickgauged needles (Baker et al., 2004). Based on the ability of samples as small as 12 mg DW (= 49 mg WW) to provide consistent FA profiles, our study shows that sufficient tissue samples are collected by standard biopsy darts (e.g., Daly and Smale, 2013; Jaime-Rivera et al., 2013) including the small dart assessed by Robbins (2006) which obtained 6.6–122 mg of total tissue. Although not stated what proportion of these biopsies were muscle, the large quantity of tissue obtained (up to 122 mg WW) suggests that sufficient muscle can be TABLE 4 | The differences between groups of samples combined by time spent frozen at −20◦C for 55 white shark Carcharodon carcharias samples from the Neptune Islands, South Australia and throughout New South Wales.


p(MC) values determined by Nested PERMANOVA (freezer group nested within sampling location) with Monte Carlo simulation between binned freezer groups (1 = 0–3 months) (2 = 4–6 months) (3 = 7–9 months) (4 = 10–12 months) (5 = 13–15 months)(6 = 16–18 months)(7 = 19–21 months) (8 = 22–24 months), the primary fatty acids (FA) driving the significantly different groups determined by SIMPER percent contribution. FA are listed in order of decreasing contribution. Listed PERMDISP p values indicate the significance of the differences in dispersion between the groups. \*p <0.05, \*\*p < 0.01, \*\*\*p < 0.001.

collected. Furthermore, biopsy needles (14-gauge, 4 cm long, double-barreled Tru-Cut needles), designed to collect 60 mg WW of tissue from small teleosts are also sufficient to collect tissue for FA analysis (Baker et al., 2004; Logan and Lutcavage, 2010). This ability to obtain FA profiles from small amounts of muscle validates the suitability of minimally invasive sampling methods, and allows trophic ecologists to apply FA analyses to smaller elasmobranchs than previously thought, without the need for lethal sampling. Additionally, multiple studies investigated the variation in muscle-derived FA profiles across different anatomical sites, and found no significant differences (Davidson et al., 2011; Pethybridge et al., 2014). Thus, these biopsy methods can be reliably used regardless of variation in sampling site, furthering the applicability of signature FA analyses. Furthermore, FA profiles can be obtained from the lipids extracted during standard sample preparation for isotopic analysis (Marcus et al., 2017). Therefore, the minimal tissue quantities already retrieved for SIA provide researchers with the opportunity for distinct and complementary FA analyses from the same non-lethal tissue biopsies, without the need to prioritize one of the two datasets. Considering the small amount of muscle necessary, minimally invasive biopsy methods collect sufficient muscle tissue to undertake FA analysis which can be paired with existing standard sample preparation for isotopic analysis, enhancing the method's suitability for ongoing work in trophic ecology.

#### TABLE 5 | Individual fatty acid degradation, assessed by days at −20◦C, months stored at −20◦C, and years stored at −20◦C.


PERMANOVA p(MC) of the full profiles noted with \*p < 0.05, \*\*p < 0.01, \*\*\*p < 0.001. PERMANOVA p(MC) significance of individual fatty acids set at p < 0.01, and denoted as either non-significant (NS) or significant (S). Month and Year data has been binned for analysis. Months at −20◦C binned as (1 = 0–3 months) (2 = 4–6 months) (3 = 7–9 months) (4 = 10–12 months) (5 = 13–15 months)(6 = 16–18 months)(7 = 19–21 months) (8 = 22–24 months). Years at −20◦C binned as 1 = 0–1 years, 2 = 1.1–2 years, 3 = 3–5 years, 4 = 6–10 years, 5 = 11–16 years.

FAs with no significant degradation across any of the three trials, 14:0, 15:0, 16:0, 17:0, 18:1ω9, 6MUFA, 20:2ω6, 20:4ω6, 22:5ω6.

In contrast to muscle tissue, the FA profiles of sub-dermal tissue smaller than 40 mg DW became highly variable, indicating a minimum reliable tissue quantity of 40 mg DW (= 184 mg WW), which is more than three times the minimal requirement for muscle. This is potentially due to the difference in PL concentration between the two types of tissue of the lipid profile. Combined with the lower lipid content, the lower relative PL contribution in the sub-dermal tissue may explain the comparatively larger minimum sub-dermal tissue quantity, as the ST, which are found in higher abundance in the sub-dermal tissue, do not contribute to the FA pool. This larger minimum tissue quantity required for sub-dermal tissue compared to muscle may limit the applicability of many aforementioned non-lethal biopsy methods. For example, the biopsy method yielding the second highest tissue volume provided only 80– 172 mg WW of sub-dermal tissue (Daly and Smale, 2013), which is not sufficient for reliable FA analysis. Only the Reeb and Best's dart head (Reeb and Best, 2006) which retained an average of 0.35 cm<sup>3</sup> of sub-dermal tissue when trialed by Jaime-Rivera et al. (2013), obtained potentially suitable tissue quantities. Furthermore, biopsies from small elasmobranchs are unlikely to yield sufficient tissue, as the thickness of the sub-dermis is greatly reduced. For example, sub-dermal tissue layers in atlantic sharpnose shark, scalloped hammerhead and dusky smooth-hound sharks ranged 0.02–0.16 cm (Motta, 1977), compared to white sharks averaging 1.1 cm (Jaime-Rivera et al., 2013) and whale sharks exceeding 2 cm (Rohner et al., 2013).

### Degradation

The consistently low levels of FFA in muscle and sub-dermal tissue throughout the degradation trial contrasts with findings across marine taxa, which highlight large increases in FFA from enzymatic hydrolysis of several non-polar lipid classes (Fernández-Reiriz et al., 1992; Kaneniwa et al., 2000; Losada et al., 2005). The difference between our findings and the pervasive trends in previous studies may be attributable to species- and taxa-specific enzymatic processes. Rudy et al. (2016) and Kaneniwa et al. (2000) hypothesized that total lipid content drove the species-specific differences in the level of observed lipid class and FA degradation amongst teleost species, with the "fatty" fish most susceptible. Compared with the six teleosts assessed in Rudy et al. (2016), white sharks were orders of magnitude leaner, with muscle containing 0.6% lipid WW and sub-dermal tissue 0.4% lipid WW (vs. 10.3–2.9% WW in teleosts). The low lipid content may explain the lack of discernable lipid class degradation across both tissues and the comparative stability in FA profiles within the sub-dermal tissue. Given the aim of determining the functional limitations of using elasmobranch specimens not immediately frozen, for example from fisheries bycatch and shark mitigation measures, our results indicate that lipid classes from muscle and subdermal tissues are not convoluted by degradation within a 4 day period.

The lipid-poor sub-dermal tissue also showed no discernible shift in FA profile or level of dispersion through exposure to ambient temperature for 4 days. However, the FA profiles derived from muscle tissue immediately changed, with a decrease in dispersion observed after 24 h, potentially compromising the ability to distinguish between individual samples. This advocates for exploring the use of sub-dermal tissue over muscle in situations when samples have been left at ambient temperature, and should be the subject of controlled feeding trails to assess the capacity for sub-dermal tissue to reflect diet. Our earlier findings, however, highlights that such FA profiles based on sub-dermal layers cannot be directly compared to FA profiles from muscle and that this discrepancy should be accounted for.

Muscle segments stored at −20◦C showed significant FA profile shifts in both assessment periods, highlighting concerns regarding the capacity to accurately use archived samples. Results in this study suggest that although there may be some level of FA degradation, the time frame at which this occurs and processes involved remains unclear. It is also plausible that the difference in the 3–6 months group is not driven by the time spent in the freezer, but by the influence of unassessed biotic factors (e.g., individual's state of maturity, sex, season of capture). The comparison of FA profiles from archived samples stored for 1– 16 years did not provide further clarification and showed no clear differences in FA profiles. Furthermore, neither trial's FA profiles decreased in dispersion, a pattern characteristic of FA degradation in the ambient temperature trial. Regardless of the degradation that might be occurring through long-term storage, differences between locations (NSW vs. QLD) remained, further suggesting that frozen samples may retain viable and indicative FA signatures.

The shift in the relative proportions of individual FAs of the muscle tissue illustrates the complex nature of FA degradation at both 20 and −20◦C. Our study found that SFA, driven primarily by 18:0, can remain constant during some time periods, but also decreased drastically through other periods. The MUFA, unchanged at 20◦C, demonstrated some resistance to degradation, with no shifts in either individual MUFA, or the PMUFA. Unexpectedly, they showed variable patterns of alteration in the early month of storage, suggesting that they are prone to degradation at −20◦C, consistent with findings across other taxa (**Table 5**, e.g., teleosts in Rudy et al., 2016 and octopus in Gullian-Klanian et al., 2017). PUFA are more reactive owing to their numerous double-bonds and are especially prone to degradation (Refsgaard et al., 1998; Paola and Isabel, 2015; Rudy et al., 2016; Gullian-Klanian et al., 2017). However, shifts in relative levels of PUFA of white sharks, including key dietary indicators 22:6ω3 (DHA) and 20:5ω3 (EPA), were only distinguishable in the ambient temperature trial, and not in either the short- or long-term −20◦C analysis (with the exception of 18:2ω6). Additionally, the polyene index (EPA+DHA/16:0), a well-established metric for tissue degradation, thought to be ubiquitous across taxa (Jeong et al., 1990; Paola and Isabel, 2015),showed no decrease across any trials (**Table 5**). The present study shows that white shark muscle PUFA might not show the stark degradation seen in the muscle tissue of other species. Given the relative importance of PUFA, as essential FAs and key dietary markers, these findings suggest that elasmobranch samples may retain these key FAs throughout extensive storage at −20◦C.

#### CONCLUSION

Our findings indicate that muscle and sub-dermal tissue contain distinct FA profiles and differing individual FAs, many of which are key trophic indicators. As such, these tissues are not directly comparable. They may, however, present complementary trophic information reflecting differing time frames, providing the opportunity to garner additional information from non-lethal biopsies. The minimum tissue amount for sub-dermal tissue was 40 mg DW (184 mg WW), whereas muscle samples as small as 12 mg DW (equating to 49 mg WW) retained consistent FA profiles. This makes FA analysis an ideal tool for elucidating trophic ecology of rare or endangered elasmobranchs for which lethal sampling is inappropriate. Degradation of muscle tissue occurred within the first 24 h at ambient temperature, unlike sub-dermal tissue, which revealed no discernible degradation across 4 days. As such, the use of deceased organisms, from shark mitigation strategies, by-catch, or beach strandings should be undertaken with caution, ensuring that preservation occurs within 24 h. Muscle tissue appears to retain viable and indicative FA signatures across long periods of frozen storage (up to 16 years), advocating for the use of archived samples, especially in cases where sampling opportunities are rare or opportunistic. Overall, lipid class and FA analysis can be reliably assessed from small tissue quantities derived from minimally invasive, nonlethal biopsies, deceased elasmobranchs preserved within 24 h and archived samples, proving a robust toolset for elucidating the trophic ecology of rare and endangered wildlife.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

LM, CH, CB, HP, PN, JW, and BB: conceived and designed the experiments; BB, JW, CB, and CH: provided tissue samples; LM: performed the experiments, analyzed the data with the help of CH, CB, PN, and HP; LM wrote the manuscript with the advice of CH, CB, HP, PN, JW, and BB; all authors provided editorial advice.

#### FUNDING

This work was supported by a research grant from the Save Our Seas Foundation (grant number RPF14/553).

#### ACKNOWLEDGMENTS

This work was supported by the Save Our Seas Foundation. We thank the Queensland Department of Agriculture and Fisheries Shark Control Program as part of the QLD large shark tagging research program and the QLD white shark tracking program for providing tissue samples. We also thank the New South Wales Department of Primary Industries Shark Meshing Program and the South Australian Research and Development Institute for collecting and storing tissue material. The white shark tourism operators; Andrew Fox and the Fox Shark Research Foundation, Calypso Star Charters and Adventure Bay Charters staff and crew provided ongoing support and in-kind contributions, facilitating sample collection within South Australia.


into the feeding ecology and ecophysiology of a complex top predator. PLoS ONE 9:7877. doi: 10.1371/journal.pone.0097877


**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 © 2017 Meyer, Pethybridge, Nichols, Beckmann, Bruce, Werry and Huveneers. 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) or licensor 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.

# iDNA at Sea: Recovery of Whale Shark (Rhincodon typus) Mitochondrial DNA Sequences from the Whale Shark Copepod (Pandarus rhincodonicus) Confirms Global Population Structure

Mark Meekan<sup>1</sup> , Christopher M. Austin2, 3, 4, 5, Mun H. Tan<sup>2</sup> , Nu-Wei V. Wei <sup>5</sup> , Adam Miller <sup>2</sup> , Simon J. Pierce<sup>6</sup> , David Rowat <sup>7</sup> , Guy Stevens <sup>8</sup> , Tim K. Davies <sup>9</sup> , Alessandro Ponzo<sup>10</sup> and Han Ming Gan2, 3, 4 \*

*<sup>1</sup> Australian Institute of Marine Science, Indian Ocean Marine Research Centre (MO96), Crawley, WA, Australia, <sup>2</sup> Centre for Integrative Ecology, School of Life and Environmental Sciences, Deakin University, Geelong, VIC, Australia, <sup>3</sup> School of Science, Monash University Malaysia, Petaling Jaya, Malaysia, <sup>4</sup> Genomics Facility, Tropical Medicine and Biology Platform, Monash University Malaysia, Petaling Jaya, Malaysia, <sup>5</sup> Research Institute for the Environment and Livelihoods, Charles Darwin University, Darwin, NT, Australia, <sup>6</sup> Marine Megafauna Foundation, Truckee, CA, United States, <sup>7</sup> Marine Conservation Society, Seychelles, Seychelles, <sup>8</sup> The Manta Trust, Dorchester, United Kingdom, <sup>9</sup> MRAG Ltd., London, United Kingdom, <sup>10</sup> Large Marine Vertebrates Research Institute Philippines, Bohol, Philippines*

The whale shark (*Rhincodon typus*) is an iconic and endangered species with a broad distribution spanning warm-temperate and tropical oceans. Effective conservation management of the species requires an understanding of the degree of genetic connectivity among populations, which is hampered by the need for sampling that involves invasive techniques. Here, the feasibility of minimally-invasive sampling was explored by isolating and sequencing whale shark DNA from a commensal or possibly parasitic copepod, *Pandarus rhincodonicus* that occurs on the skin of the host. We successfully recovered mitochondrial control region DNA sequences (∼1,000 bp) of the host via DNA extraction and polymerase chain reaction from whole copepod specimens. DNA sequences obtained from multiple copepods collected from the same shark exhibited 100% sequence similarity, suggesting a persistent association of copepods with individual hosts. Newly-generated mitochondrial haplotypes of whale shark hosts derived from the copepods were included in an analysis of the genetic structure of the global population of whale sharks (644 sequences; 136 haplotypes). Our results supported those of previous studies and suggested limited genetic structuring across most of the species range, but the presence of a genetically unique and potentially isolated population in the Atlantic Ocean. Furthermore, we recovered the mitogenome and nuclear ribosomal genes of a whale shark using a shotgun sequencing approach on copepod tissue. The recovered mitogenome is the third mitogenome reported for

#### Edited by:

*Rob Harcourt, Macquarie University, Australia*

#### Reviewed by:

*Celine Frere, University of the Sunshine Coast, Australia Gail Schofield, Deakin University, Australia*

\*Correspondence:

*Han Ming Gan han.gan@deakin.edu.au*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *08 October 2017* Accepted: *07 December 2017* Published: *18 December 2017*

#### Citation:

*Meekan M, Austin CM, Tan MH, Wei N-WV, Miller A, Pierce SJ, Rowat D, Stevens G, Davies TK, Ponzo A and Gan HM (2017) iDNA at Sea: Recovery of Whale Shark (Rhincodon typus) Mitochondrial DNA Sequences from the Whale Shark Copepod (Pandarus rhincodonicus) Confirms Global Population Structure. Front. Mar. Sci. 4:420. doi: 10.3389/fmars.2017.00420*

**42**

the species and the first from the Mozambique population. Our invertebrate DNA (iDNA) approach could be used to better understand the population structure of whale sharks, particularly in the Atlantic Ocean, and also for genetic analyses of other elasmobranchs parasitized by pandarid copepods.

Keywords: eDNA, sharks, minimally-invasive sampling, copepod, control region, population genetics, parasite, commensal

#### INTRODUCTION

The term environmental DNA (eDNA) refers to DNA extracted from cells that are not collected directly from a target organism, but are obtained from the environments they inhabit such as oceans, river water, soil, and air (Ficetola et al., 2008; Fonseca et al., 2010; Andersen et al., 2012; Sigsgaard et al., 2016). Invertebrate-derived DNA (iDNA) is an offshoot of this approach that involves the extraction of genetic material of animals via the flesh-eating or haematophagous invertebrates that parasitise them (Schnell et al., 2015; Schubert et al., 2015; Lee et al., 2016). To date, most iDNA studies have focused on terrestrial vertebrates and have extracted host DNA from insects, ticks, or leeches. There have been no analogous studies in marine environments, despite the potential usefulness of the approach for sampling large marine megafauna such as cetaceans, sirenians, pinnipeds, marine reptiles (sea turtles), elasmobranchs, and teleosts. For such taxa, iDNA sampling offers both ethical and practical advantages as it is minimally-invasive and thus preferable to the direct collection of blood or tissue, particularly where target species are rare or endangered.

The whale shark (Rhincodon typus), similar to many large marine vertebrates, poses significant challenges for researchers trying to understand their biology and ecology and for managers attempting to develop conservation strategies (Graham and Roberts, 2007; Rowat et al., 2009). Tagging, genetic and modeling studies suggest that individuals can disperse widely (Sequeira et al., 2013), although there is evidence of isolation between populations in the Indo-Pacific and Atlantic oceans (Vignaud et al., 2014). However, the scale at which population structure can be discerned is dependent on sample sizes, with low numbers reducing analytical power to reject the null hypothesis of a panmictic population (Castro et al., 2007). For whale sharks, low sample size tends to be a result of the rarity of the species and the difficulties in obtaining biopsies (usually of skin tissue) for genetic analyses, which requires appropriate permits and ethical approvals for invasive sampling of a species that is categorized as "Endangered" by the International Union for the Conservation of Nature and the use of trained divers and technicians. In this situation there are many benefits to the use of an iDNA approach, provided that external invertebrate parasites or commensals that harbor the intact DNA of the whale shark host can be successfully identified. Although eDNA techniques have recently been used to obtain whale shark DNA (Sigsgaard et al., 2016), iDNA still offers a major advantage because it enables haplotypes to be linked to individual whale sharks for the analysis of the genetic structures of populations.

Copepods (Crustacea, Maxillopoda, Copepoda) are an excellent candidate species for iDNA studies on marine vertebrates. Free-living forms are a dominant element of marine zooplankton and may even exceed insects in terrestrial environments in terms of sheer abundance of individuals. Less recognized is that approximately half of the nearly 30,000 described species live in parasitic or commensal associations with a diverse range of taxa, including fish and mammals (Boxshall, 2005). The order Siphonostomatoida contains largely parasitic copepods that feed on the blood, epidermal tissue, or mucus of many marine teleost fishes, sharks, and rays. Approximately 550 genera, representing nearly 40 families, are placed in the order and include economically important species such as sea lice (Brachiura) that parasitise farmed fish (Gunn and Pitt, 2012). Eleven siphonostomatoid families have been reported as symbionts of a diverse range of elasmobranchs (Dippenaar, 2009) and one family, the Pandaridae, is composed of 23 genera that include species that are ectoparasites or commensals of large sharks (Cressey and Boyle, 1978; Walter and Boxshall, 2017). Within the Pandaridae, the genus Pandarus currently has 17 recognized species of which P. rhincodonicus (Norman et al., 2000) is noteworthy as it is appears to be associated exclusively with the whale shark, where it is predominately found on the leading and trailing edges of fins and on the lips. Although it is thought to be a commensal that feeds off bacteria and other microorganisms on the skin of the shark (Norman et al., 2000), dietary studies of the species are incomplete.

Here, we demonstrate that P. rhincodonicus sampled from sites across the Indian Ocean contain sufficient DNA from their whale shark host to routinely recover mitochondrial sequences that allow analyses of the genetic structure of host populations, and the recovery of the complete whale shark mitogenome and ribosomal nuclear genes. The implications of these findings are discussed in relation to the use of copepods for iDNA studies of marine teleosts and sharks and the ecological status of Pandarus spp. as commensals.

### MATERIALS AND METHODS

#### Sampling and Sequencing

A total of 45 copepods were collected from 31 individual whale sharks sampled in Baa Atoll in the Maldives, at Ningaloo Reef, Western Australia, off Tofu Beach, Mozambique and off the coast of Mahe in the Seychelles as part of an earlier study using approved procedures under the Western Australian Department of Parks and Wildlife (WADPW) and University of Tasmania (UTAS) ethics permits (WADPW: SF009814 and SF009227; UTAS: 2255 and 2307) (Vignaud et al., 2014) (**Supplementary Table 1**). Copepods were scraped off the edges of the fins or lips using a plastic knife by a snorkeler who swam alongside an unrestrained animal, collected in an aquarium net and brought back to the vessel where they were preserved in 100% ethanol. DNA extraction was performed on whole copepod specimens using the DNeasy Blood and Tissue kit (Qiagen, Halden, Germany). Of the 45 copepods we collected, 44 were selected for the amplification of whale shark the mtDNA control region, using primers WSCR1-F and WSCR2- R, and following the PCR protocol as described by Castro et al. (2007). PCR products were purified using the Viogene Gel/PCR DNA Isolation System Kit (Viogene Biotek Corp, Taitung, Taiwan) and sequenced on an ABI 3130xl Genetic Analyzer (Applied Biosystems, Foster City, CA) located at Charles Darwin University.

The remaining copepod, isolate 366.1, was used for partial genome sequencing. Briefly, 1 µg of genomic DNA (gDNA) was sheared to 300 bp using a Covaris Focused Ultrasonicator (Covaris, Woburn, MA), and subsequently processed with Truseq DNA library prep kit (Illumina, San Diego, CA) according to the manufacturer's instructions. This was followed by nextgeneration sequencing on a fraction of a HiSeq 2000 lane (Illumina, San Diego, CA), with a run setting of 2 × 100 bp, for the initial goal of recovering the copepod mitogenome (Austin et al., 2016) and microsatellite loci (unpublished).

#### Mitochondrial Control Region Analysis

Authenticity of the DNA sequences of the mitochondrial control region of whale sharks were validated by BLAST searches, and sequences were aligned with previously generated sequences (Castro et al., 2007; Vignaud et al., 2014) using MAFFT version 7.310 with the option "–adjustdirection" activated (Yamada et al., 2016). The 5′ and 3′ ends of the alignment were manually trimmed using MEGA 6 (Tamura et al., 2013) so that each aligned control region had the same flanking sequence (**Supplementary Figure 1**). The trimmed alignment was subsequently de-gapped using Seqret (Rice et al., 2000) and clustered with cd-hit-est at a 100% sequence identity and 100% sequence length coverage cut-off as implemented using the –c 1.0 and –a 1.0 setting (Li and Godzik, 2006).

#### Analysis of Population Differentiation

The mitochondrial control regions of whale sharks generated using iDNA were combined with 613 sequences published by previous studies of whale shark populations (Castro et al., 2007; Ramírez-Macías et al., 2007; Schmidt et al., 2010; Vignaud et al., 2014). Identical iDNA sequences from the same whale shark host were removed prior to genetic analysis, to eliminate the chance of biasing estimates of population genetic differentiation (**Supplementary Table 1**). Arlequin suite version 3.5.2.2 (Excoffier and Lischer, 2010) was used to estimate global and pairwise φST. A median-joining haplotype network was constructed using PopArt version 1.7 (Leigh and Bryant, 2015) and to simplify network representation, only the common haplotypes (a haplotype with more than one sequence representation) were used to generate the network graph.

### Recovery of Whale Shark Whole Mitochondrial Genome and Nuclear Ribosomal Genes

Assembly used a baiting and iterative mapping approach as implemented in MITObim version 1.8 (Hahn et al., 2013) from the complete mitogenome of a whale shark sampled off Taiwan (Accession Number: NC\_023455) as the reference sequence (Alam et al., 2014). In-silico circularization and annotation of the assembled mitogenome followed (Iwasaki et al., 2013; Gan et al., 2014). To improve alignment accuracy and specificity, mitogenome coverage was calculated by mapping the raw paired-end reads to the assembled mitogenome sequence using Bowtie2 version 2.3.2 (Langmead and Salzberg, 2012) with the setting "—score-min L,0.2,−0.2." BRIG was used to visualize the mitogenome annotation and read mapping coverage (Alikhan et al., 2011). Similar Bowtie2 mapping setting was used to map the raw reads to a whale shark assembly contig containing the nuclear ribosomal genes (18S rRNA and 28S rRNA) and the read alignment was subsequently visualized using Integrative Genomics Viewer (Thorvaldsdottir et al., 2013).

### RESULTS

### Mitochondrial Control Region Analysis

A ∼1,000 bp DNA sequence fragment of the whale shark mitochondrial control region was successfully generated from the extracted gDNA from 44 copepods representing 31 sharks. Haplotype clustering confirmed a 100% sequence match for DNA sequences amplified from copepods collected from the same whale shark host (**Supplementary Data 1**). An initial alignment of the 31 newly-generated iDNA control region sequences with the 613 sequences provided by previous studies (Castro et al., 2007; Schmidt et al., 2010; Vignaud et al., 2014; Walter and Boxshall, 2017) yielded a 1,910 bp product (**Supplementary Data 2**). Trimming using 5′ and 3′ -ends produced a final alignment length of 750 bp (**Supplementary Data 3**), from which a total of 163 unique haplotypes were evident (**Supplementary Data 4**). The Ningaloo Reef population had the highest number of sampled control region sequences (n = 163) and also had the greatest number of haplotypes (h = 36) (**Figure 1A**). Interestingly, the observed haplotype diversity for the Isla Holbox population was 50% lower (h = 11) than that of Djibouti population (Vignaud et al., 2014), despite similar sample sizes (**Figures 1A,B**).

Of the 31 deduplicated iDNA sequences reported in this study, five were found to represent replicate individuals from the Ningaloo population previously sequenced by Vignaud et al. (2014) using skin samples (**Supplementary Table 1** and **Supplementary Data 4**). Despite the relatively low number of sequences reported by our study that represented sequences from new whale sharks (N = 27), our study contributed a 39 and 22% increase in the number of individuals sampled from the Mozambique and Seychelles populations, respectively (**Figure 1A**). In addition, four novel haplotypes for control region sequences of whale sharks were identified from copepod samples.

### Population Structure of the Whale Shark

Global φST was weak but significantly different from zero, indicating limited gene flow among some sampling locations. This pattern was driven by a single population, with all pairwise φST calculations associated with the Isla Holbox population differing significantly from zero (average φST of 0.2), whereas all estimates of φST did not (**Figure 2**). This implied that the population at Isla Holbox was genetically isolated from an otherwise panmictic population spread across the Indo-Pacific Ocean. Three common haplotypes occurred at similar frequency across the sampling range (**Figure 3**). A few private haplotypes occurred in the Ningaloo Reef, Philippines and Seychelles populations. Consistent with φST estimates, the population at Isla Holbox was the most genetically differentiated, as it lacked one of the three common ancestral haplotypes (H108, 110, 125, 150, and 152), and had a high frequency of others that were rare (H67, H68, and H76).

### Recovery of the Whale Shark Mitogenome and Nuclear Genes from Shotgun Sequencing of Copepod Tissue

The complete mitochondrial genome of the whale shark was successfully assembled from partial genome sequencing of an iDNA sample from a copepod (isolate 366.1; **Figure 4**). Mapping of the 14 million paired-end reads to the reconstructed whale shark mitogenome gave a mapping rate of 0.006% (900 reads) representing approximately 5× mitogenome coverage (**Figure 4**). The mitogenomic composition of the whale shark were extracted was very similar (>99% identity) to those reconstructed from sharks from Taiwanese waters (Accession codes: KC633221 and NC\_023445). A comparison of the 13 genes coding for mitochondrial proteins indicated one or two nucleotide mismatches, mostly associated with non-synonymous mutations found in the atp6, cox1, cytb, nad2, nad4, and nad5 genes. In addition, we also observed reads mapping to the whale shark 18S and 28S rRNA genes (**Supplementary Figure 2** and **Supplementary Data 5**), indicating the presence of whale shark nuclear DNA in the copepod extracted gDNA, a phenomenon that will be useful for future population genetic studies of these sharks based on nuclear markers.

### Data Deposition

Mitochondrial control regions were submitted to NCBI under the accession numbers MF872682-MF872725. Raw reads from the shotgun sequencing of the copepod were submitted to Sequence Read Archive under the run number SRR4111090 and the reconstructed whale shark mitogenome has been assigned the accession number MF872726.

### DISCUSSION

Our study is the first demonstration of the use of iDNA sampling of a marine invertebrate in order to obtain mitochondrial genetic information from an elasmobranch host. We successfully amplified mitochondrial DNA fragments of whale sharks in the size range of ∼1,000 bp from a copepod, one of the largest host DNA fragments to be recovered by an iDNA study (Schnell et al., 2012, 2015; Lee et al., 2016; Pérez-Flores et al., 2016; Rodgers et al., 2017). This suggests the presence of largely intact whale shark DNA in or on copepods, although the exact source could not be identified as copepods were analyzed whole. Targeted efforts to amplify DNA fragments of whale sharks from the intestine and epidermal layers of P. rhincodonicus will be useful to resolve this issue and clarify the role of the copepod as a commensal or parasite. We found iDNA sequences from multiple copepods sampled from the same whale shark host to be identical, suggesting copepods have a persistent, long-term association with their host shark, which contrasts with the more generalist and mobile host associations of other invertebrate ectoparasites in terrestrial and aquatic ecosystems, such as leeches, blow flies, and mosquitoes (Calvignac-Spencer et al., 2013). However, these findings should be treated with some degree of caution and a multi-locus approach using microsatellite

markers (Olson et al., 2012) might offer a more powerful means to validate these findings. As reported by Vignaud et al. (2014), our estimates of genetic structure across the Indian, Pacific, and Atlantic oceans indicated the presence of two distinct populations, one in the Indo-Pacific and the other in the Atlantic Ocean. Isla Holbox was the only location sampled in the Atlantic Ocean and the minimally-invasive iDNA approach could potentially be used to improve the coverage of sampling in this region.

We also demonstrated that complete mitochondrial genome and nuclear ribosomal RNA sequences of whale sharks could be obtained using iDNA sampling. Although not the first for the species (Alam et al., 2014; Chen et al., 2016), the sequences generated by our study represent the first for the Mozambique population of whale sharks. The successful recovery of the mitogenome and nuclear ribosomal RNA from the shotgun sequencing of P. rhincodonicus was probably due to the high copy number of mitochondrial and nuclear ribosomal genes in somatic cells, coupled with high sequencing depth (∼14 million paired-end reads). Efficiency of the recovery of whale shark mitogenomes from copepod tissue samples might be improved by sequencing gDNA extracted from the part of the copepod body that contains higher concentrations of whale shark tissue. In addition, given the intactness of whale shark gDNA (>1,000 bp) that was present in P. rhincodonicus, long range PCR of additional mitogenome regions followed by high throughput amplicon sequencing could also be explored to dramatically increase the coverage of the whale shark mitogenome for population genetic and phylogeographic applications (Deiner et al., 2017; Pavlova et al., 2017). Given that mtDNA only represents a fraction of the evolutionary history of a species due to its strict maternal inheritance and small genome size (16 kb), studies of the population genetics of whale sharks may benefit from the combination of both mtDNA and nuclear data

transcriptional orientation of each protein coding gene is indicated by the red arrows.

(Godinho et al., 2008; Schmidt et al., 2009; Vignaud et al., 2014). The presence of nuclear ribosomal RNA reads in the copepod shotgun sequencing library suggests that it is possible to amplify both nuclear DNA and mtDNA of whale sharks from the copepod and the recent publication of the first draft genome for the whale shark (Read et al., 2017) will greatly facilitate the design of a targeted next-generation sequencing panel to enhance future studies of whale shark population genetics.

The Pandaridae comprises 23 genera and more than 100 species, many of which are ectoparasites or symbionts of large sharks (Walter and Boxshall, 2017), and there are nearly 30 other families of copepods that include species that are parasites of fishes (Boxshall, 2005). These offer an opportunity to extend iDNA methods to other taxa of large marine vertebrates. Conversely, the presence of gDNA of whale sharks in the extracted DNA of P. rhincodonicus will need to be considered as a possible contaminant in future whole genome sequencing or population genomic projects that target this copepod.

#### AUTHOR CONTRIBUTIONS

MM and CA: Original concept, sample collection, analyses, interpretation of results, writing, proof-reading. MT: Bioinformatics analysis. N-WW: Control region sequencing. AM: interpretation of results, writing, proof-reading. SP, DR, GS, TD, and AP: sample collection proof-reading. HG: analyses, writing, interpretation of results, proof-reading.

#### FUNDING

Funding for this study was provided by the SeaWorld Research and Rescue Foundation, the Save our Seas Foundation, the Monash University Malaysia Tropical Medicine and Biology Platform, Monash University Malaysia School of Science, Quadrant Energy, Australian Institute of Marine Science, The Western Australian Department of Environment and Conservation, The Four Seasons Resorts Maldives. Funding for SP came from Aqua-Firma, the Shark Foundation, the Save Our Seas Foundation and two private trusts.

#### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2017.00420/full#supplementary-material

Supplementary Table 1 | Sample collection information.

Supplementary Figure 1 | Visualization of aligned control region sequences (one representative from each study). Arrows indicate regions upstream and downstream of the 5′ and 3′ ends, respectively, that will be trimmed off for subsequent haplotype clustering.

Supplementary Figure 2 | Bowtie2 alignment of Illumina reads generated from copepod gDNA to the whale shark genomic region containing 18S and 28S nuclear rRNA genes. Red and blue arrow bars indicate forward and reverse pair-end reads, respectively.

Supplementary Data 1 | Cd-hit clustering of trimmed whale shark control region sequences that were amplified from copepods.

Supplementary Data 2 | Alignment of whale shark control region sequences.

Supplementary Data 3 | Trimmed alignment of whale shark control region sequences based on the conserved region depicted in Supplementary Figure 1.

Supplementary Data 4 | Cd-hit Clustering of trimmed whale shark control region sequences.

Supplementary Data 5 | Paired-end reads mapped to the whale shark genomic region containing the 18S and 28S ribosomal RNA genes.


ocean populations of the whale shark, Rhincodon typus. PLoS ONE 4:e4988. doi: 10.1371/journal.pone.0004988


**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 reviewer GS declared a shared affiliation, with no collaboration,with several of the authors, CA, MT, AM, and HG, to the handling Editor.

Copyright © 2017 Meekan, Austin, Tan, Wei, Miller, Pierce, Rowat, Stevens, Davies, Ponzo and Gan. 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) or licensor 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.

# Route Fidelity during Marine Megafauna Migration

Travis W. Horton<sup>1</sup> \*, Nan Hauser <sup>2</sup> , Alexandre N. Zerbini 3, 4, 5, Malcolm P. Francis <sup>6</sup> , Michael L. Domeier <sup>7</sup> , Artur Andriolo5, 8, Daniel P. Costa<sup>9</sup> , Patrick W. Robinson<sup>9</sup> , Clinton A. J. Duffy <sup>10</sup>, Nicole Nasby-Lucas <sup>7</sup> , Richard N. Holdaway <sup>11</sup> and Phillip J. Clapham<sup>3</sup>

<sup>1</sup> Department of Geological Sciences, University of Canterbury, Christchurch, New Zealand, <sup>2</sup> Center for Cetacean Research and Conservation, Avarua, Cook Islands, <sup>3</sup> National Marine Mammal Laboratory, Alaska Fisheries Science Center, National Marine Fisheries Service, NOAA, Seattle, WA, United States, <sup>4</sup> Cascadia Research Collective, Olympia, WA, United States, 5 Instituto Aqualie, Juiz de Fora, Brazil, <sup>6</sup> National Institute of Water and Atmospheric Research, Wellington, New Zealand, <sup>7</sup> Marine Conservation Science Institute, Fallbrook, CA, United States, <sup>8</sup> Laboratório de Ecologia Comportamental e Bioacústica, ICB, Universidade Federal de Juiz de Fora, Juiz de Fora, Brazil, <sup>9</sup> Ecology and Evolutionary Biology, University of California, Santa Cruz, Santa Cruz, CA, United States, <sup>10</sup> Department of Conservation, Auckland, New Zealand, <sup>11</sup> Palaecol Research Ltd., Christchurch, New Zealand

#### *Edited by:*

Lars Bejder, Murdoch University, Australia

#### *Reviewed by:*

Phil Bouchet, Centre for Marine Futures, School of Animal Biology, University of Western Australia, Australia Nuno Queiroz, University of Porto, Portugal

*\*Correspondence:*

Travis W. Horton travis.horton@canterbury.ac.nz

#### *Specialty section:*

This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science

*Received:* 25 September 2017 *Accepted:* 11 December 2017 *Published:* 21 December 2017

#### *Citation:*

Horton TW, Hauser N, Zerbini AN, Francis MP, Domeier ML, Andriolo A, Costa DP, Robinson PW, Duffy CAJ, Nasby-Lucas N, Holdaway RN and Clapham PJ (2017) Route Fidelity during Marine Megafauna Migration. Front. Mar. Sci. 4:422. doi: 10.3389/fmars.2017.00422 The conservation and protection of marine megafauna require robust knowledge of where and when animals are located. Yet, our ability to predict animal distributions in space and time remains limited due to difficulties associated with studying elusive animals with large home ranges. The widespread deployment of satellite telemetry technology creates unprecedented opportunities to remotely monitor animal movements and to analyse the spatial and temporal trajectories of these movements from a variety of geophysical perspectives. Reproducible patterns in movement trajectories can help elucidate the potential mechanisms by which marine megafauna navigate across vast expanses of open-ocean. Here, we present an empirical analysis of humpback whale (Megaptera novaeangliae), great white shark (Carcharodon carcharias), and northern elephant seal (Mirounga angustirostris) satellite telemetry-derived route fidelity movements in magnetic and gravitational coordinates. Our analyses demonstrate that: (1) humpback whales, great white sharks and northern elephant seals are capable of performing route fidelity movements across millions of square kilometers of open ocean with a spatial accuracy of better than 150 km despite temporal separations as long as 7 years between individual movements; (2) route fidelity movements include significant (p < 0.05) periodicities that are comparable in duration to the lunar cycles and semi-cycles; (3) latitude and bedrock-dependent gravitational cues are stronger predictors of route fidelity movements than spherical magnetic coordinate cues when analyzed with respect to the temporally dependent moon illumination cycle. We further show that both route fidelity and non-route fidelity movement trajectories, for all three species, describe overlapping in-phase or antiphase sinusoids when individual movements are normalized to the gravitational acceleration present at migratory departure sites. Although these empirical results provide an inductive basis for the development of testable hypotheses and future research questions, they cannot be taken as evidence for causal relations between marine megafauna movement decisions and geophysical cues. Experiments on model organisms with known sensitivities to gravity and magnetism, complemented by further empirical observation of free-ranging animals, are required to fully explore how animals use discrete, or potentially integrated, geophysical cues for orientation and navigation purposes.

Keywords: navigation, gravity, moon, humpback whale, great white shark, elephant seal, tracking, g-space

#### INTRODUCTION

During some of the most spectacular yet least understood events in nature, millions of animals migrate between widely separated habitats without getting lost. In fact, animals from at least three different phyla are able to relocate previously inhabited sites, including chordates, arthropods, and molluscs (Switzer, 1993). With respect to marine megafauna, humpback whales (Megaptera novaeangliae), great white sharks (Carcharodon carcharias), and northern elephant seals (Mirounga angustirostris) maintain interannual site fidelity to specific seasonal habitats (Oliver et al., 1998; Wedekin et al., 2010; Anderson et al., 2011) despite undertaking long-distance migrations that span thousands of kilometers of open ocean (Le Boeuf et al., 2000; Nasby-Lucas and Domeier, 2012; Garrigue et al., 2015). Direct observation of the same whales, sharks, and seals in the same areas, year after year, demonstrates that all three species have well-developed navigational abilities that enable high levels of spatiotemporal movement accuracy and precision. Despite our awareness of these remarkable movements, a mechanistic understanding of how marine megafauna navigate remains elusive.

One of the main reasons why we do not yet understand the mechanics of marine megafauna navigation is the fact that we have not yet identified the coordinate space in which navigation occurs. Several fundamental questions remain unanswered for most migratory species, including: What exogenous cues are used for orientation purposes? What reference frame(s) and reference datum(a) are applied during the transduction and neurological processing of these cues? To what extent are endogenous cues integrated with exogenous cues during navigation?

In this study, we explore these knowledge gaps through empirical analyses of humpback whale, great white shark, and northern elephant seal route fidelity movements from both magnetic and gravitational geophysical perspectives. We focus our analyses on magnetic inclination and vertical gravitational acceleration cues (see Nomenclature for a glossary of terms) as experimental studies have suggested both can serve as exogenous sources of orientation information during animal movement (Wiltschko and Wiltschko, 1972; Larkin and Keeton, 1978; Keeton, 1979; Kanevskyi et al., 1985; Light et al., 1993; Lohmann and Lohmann, 1994; DeVries et al., 2004; Putman et al., 2011; Blaser et al., 2013, 2014). By focussing our analyses on route fidelity movements, our research is directly relevant to positional orientation during migration, one of the greatest unknowns in animal navigation science (Gould, 2004). By considering a diverse suite of species and scientific disciplines, our research represents a direct response to growing calls for more integrative research on animal migration and navigation (Bowlin et al., 2010; Hays et al., 2016).

Humpback whales, great white sharks and northern elephant seals are ideally suited to empirical analysis of animal movement due to the fact that telemetry datasets for all three species include remarkable examples of asynchronous migratory route fidelity. Route fidelity is similar to site fidelity in that both refer to the repeated utilization of migratory destinations at distinctly different times. In contrast to site fidelity, route fidelity refers to the repeated utilization of well-defined migration routes by either the same individual during multiple independent migrations (i.e., intra-individual route fidelity) or different individuals migrating separately (i.e., inter-individual route fidelity). In this study, we present and analyse 22 humpback whale, great white shark, and northern elephant seal route fidelity movements, including multiple examples of intra- and inter-individual route fidelity. Our dataset includes humpback whales in the South Atlantic and South Pacific Oceans, great white sharks in the North and South Pacific Oceans and northern elephant seals in the North Pacific Ocean.

#### MATERIALS AND METHODS

#### Satellite Telemetry

Humpback whales were tracked using published methods (Garrigue et al., 2015). In brief, a carbon-fiber pole (Heide-Jørgensen et al., 2003) or a modified pneumatic line-thrower (Heide-Jørgensen et al., 2001) pressurized to ∼10 bar with compressed air was used to implant transdermal location-only SPOT radio-frequency platform transmitting terminal (PTT) satellite tags (Wildlife Computers, Redmond, WA, U.S.A.) into the upper flank of each whale near the base of the dorsal fin. Transmitters were duty cycled to maximize battery life as described in Zerbini et al. (2006) and Hauser et al. (2010). In all references to PTT tag numbers in the current study, the two digits to the right of the decimal point correspond with the abbreviated Julian calendar year in which the satellite-monitored movement was initiated.

Great white sharks were tracked using similar technology deployed by different methods (Domeier et al., 2012; Francis et al., 2012). In brief, SPOT5 PTT tags were affixed using 3– 4 small plastic or stainless steel bolts in the apex of the dorsal fin of each temporarily restrained individual. Unlike marine mammals, great white sharks do not have to surface to breathe. Thus, position data from this species are much more sporadic than they are for whales and seals. Location estimates for white sharks were received only when animals spent enough time at the surface, with the dorsal fin above the waterline, allowing the ARGOS (Argos, CLS Group) satellite array to receive three or more consecutive transmissions from the tag. Surface swimming behavior varied among individual sharks, resulting in varying resolution migration data for each shark.

Northern elephant seals were also tracked using PTT satellite tags (Robinson et al., 2012) which were attached to the head of each seal during onshore residence in Año Nuevo State Park (ca. 37.11◦N; 122.33◦W), California, U.S.A. Elephant seal locations were estimated using the ARGOS satellite system based on transmissions received when the PTT was out of the water.

Route fidelity movements were identified from larger satellite telemetry datasets by visual inspection of PTT track maps using GIS software. Movement tracks, or portions of tracks, that visually overlapped for extended periods (i.e., coincident animal location symbols or track lines for multiple days in geographic coordinates) when viewed at a scale of 1:10,000,000 were identified as route fidelity movements. At this scale, an 8 point map symbol is approximately 10 km wide. The three South Atlantic humpback whale route fidelity movements were identified from inspection of a total of 12 long-distance migration tracks (i.e., 25%). The four South Pacific humpback whale route fidelity movements were identified from inspection of a total of 13 intra-tropical movement tracks (i.e., 31%). The two South Pacific great white sharks demonstrating both intra- and interindividual route fidelity were identified from inspection of longdistance PTT tracks of three different sharks (i.e., 67%). The intra-individual route fidelity movements of two North Pacific great white sharks were identified from inspection of the PTT tracks of 11 individual sharks (i.e., 18%). The four northern elephant seal tracks were identified from inspection of a total of 74 long-distance migration tracks (i.e., 5%).

Raw animal location estimates downloaded from the ARGOS system were processed using a 20 km/h velocity filter and combined to determine single average daily locations for each individual. Transmissions were not received by the ARGOS system from all PTT tags on all days. Average daily locations were only determined for those calendar dates on which velocityfiltered locations were received. Gaps in the tracking datasets were not filled by interpolation due to the fact that we do not know the coordinate space in which navigation was performed (Horton et al., 2014). All animal tracking research reported here was carried out in accordance with animal ethics consents given to the authors by their home institutions and/or relevant government agency.

#### Astronomical and Geophysical Variables

We determined multiple astronomical, magnetic, and gravitational cues present at the whale, shark and seal locations recorded by the PTT animal tracking devices (see Nomenclature for a glossary of terms). Astronomical cues, including moon illumination, were calculated using published astronomical algorithms (Meeus, 1991). Moon illumination is a unitless time-dependent quantity ranging between 0 (i.e., new moon) and 1 (i.e., full moon) across the average 29.53-day-long synodic (i.e., moon illumination) cycle. We used it as a direct proxy for time to facilitate comparative analyses of multiple asynchronous individual telemetry tracks in a single panel.

With respect to magnetic cues, we determined main field plus rock anomaly field magnetic coordinates, including magnetic inclination, field intensity and declination, for all animal locations using the Enhanced Magnetic Model (Maus, 2010). Of the seven different magnetic variables used to define the position of Earth's magnetic field from a geocentric perspective in both spherical (F, I, D) and Cartesian (X, Y, Z, H) coordinate spaces (for definitions see Nomenclature; Horton et al., 2014), magnetic inclination (I) is the most widely associated with animal orientation and navigation (e.g., Wiltschko and Wiltschko, 1972; Light et al., 1993). Extensive experimental research has demonstrated that the movement behaviors of diverse species change in response to changes in magnetic inclination. These results have been variably interpreted as evidence for utilization of magnetic inclination as: (1) a navigational compass that facilitates directional orientation (Wiltschko and Wiltschko, 1972), or, (2) part of a metaphorical bi-coordinate "magnetic map" that facilitates positional orientation (Putman et al., 2011). Regardless of how magnetic inclination is used during animal navigation and orientation, compelling experimental evidence demonstrates that it is part of the system by which many migratory species find and follow specific movement trajectories.

Thus, a key question emerges: Are route fidelity movements compatible with orientation relative to magnetic inclination? We explored this question by plotting magnetic inclination vs. moon illumination for the route fidelity movements reported above. Considering the fact that these highly directional movements were performed at distinctly different times in distinctly different locations, strong correlations and systematic nonrandom patterns in the data would be suggestive of a potential spatiotemporal orientation behavior informed by cues associated with magnetic inclination.

With respect to gravity, we determined local gravitational accelerations associated with both latitude and bed-rock density using the International Gravity Formula (Götze, 2014) and the International Gravimetric Bureau's 2 × 2 arc-min (i.e., ∼3.7 × ∼3.7 km) World Gravity Map (Balmino et al., 2012), respectively. Latitude (gL) and bed-rock (gB) vertical gravitational accelerations are reported in Gals (i.e., cm sec−<sup>2</sup> ).

Of the multiple factors that determine gravity (i.e., g) at a given place and time, latitude has the largest effect. At 20◦ north or south latitude, the theoretical g<sup>L</sup> is 978.637 Gal. In contrast, g<sup>L</sup> is 981.070 Gal at 50◦ north or south latitude, equating to a 2.433 Gal range in g<sup>L</sup> over this 30◦ range. Importantly, g<sup>L</sup> is a trigonometric function of latitude with the most rapidly changing g<sup>L</sup> values occurring in the middle latitudes and the most gradually changing g<sup>L</sup> values occurring near the geographic equator and poles. For example, an animal that migrates from the equator to 30◦ south or north latitude would experience a change in g<sup>L</sup> of 1.292 Gal, roughly half the range in g<sup>L</sup> experienced by migrating between 20◦ and 50◦ latitude. To facilitate understanding, a 30,000 kg humpback whale requires 900 N (i.e., 202 lbs) less buoyancy force to remain effortlessly afloat in a tropical habitat, where g<sup>L</sup> is ∼979 Gal, relative to a high-latitude habitat, where g<sup>L</sup> is ∼982 Gal.

The shape and density of Earth also affects local g. However, the difference in geologically imparted Bouguer gravity anomalies from one location to the next is on the order of 10 to 100 mGal, roughly 1 to 10% as large as the latitudinal effects on g across the geographic range of the tracks we report. Since both latitudinal and geological effects on g are temporally independent, we additively combined these two gravitational variables into a single spatially dependent coordinate, gL+gB, that is determined solely by an animal's location.

The whales, sharks, and seals we studied experienced a wide range of gravitational conditions during their individual migrations. Latitude dependent gravity values (gL) ranged between 978.287 Gal (i.e., cm/sec<sup>2</sup> ) and 981.737 Gal. Local bedrock-dependent gravity values (gB) ranged between +0.025 Gal and +0.587 Gal. Using these spatially dependent gravitational variables, we constructed bivariate plots of all three species' movements through gravitational coordinates, or g-space. We define g-space as the bicoordinate system that includes the sum of g<sup>L</sup> and g<sup>B</sup> as the spatially dependent variable and moon illumination as the temporally dependent variable.

As an animal moves, its g<sup>L</sup> value will change as its latitudinal position changes, its g<sup>B</sup> value will change as the density of the Earth's crust over which it is swimming changes, and the magnitude of the tidal gravity vector will change in unison with lunar phase. In contrast, for an animal that stops its migratory movement and begins residence in a single area, only moon illumination would continue to change, and the g<sup>L</sup> and g<sup>B</sup> values it experiences would remain constant. Pure east-west movements would experience changing g<sup>B</sup> values and moon illuminations while the latitudinal component (gL) remains constant.

#### Data Analysis Methods

The satellite telemetry, astronomical, and geophysical data generated by our research were analyzed using a variety of statistical tools. Piecewise linear regression breakpoint analysis (Muggeo, 2008) performed on latitude/longitude time-series plots was used to estimate the date on which open ocean migration was initiated at the individual scale. Latitude timeseries plots were used on all individuals with the exception of the North Pacific great white sharks, where longitude timeseries were used due to the predominantly meridional nature of these tracks. Spectral analysis using the Lomb-Scargle algorithm (Scargle, 1982; Press et al., 1992; Hammer et al., 2001) was used to determine significant (α = 0.05) periodicities present in latitude time-series plots at the individual scale. Sinusoidal regression (Press et al., 1992; Hammer et al., 2001) was performed on geophysical (i.e., magnetic and gravitational) cue vs. moon illumination plots in order to determine the proportion of data variance explained (i.e., coefficient of determination; R 2 ) by a sine-function model of the data.

## RESULTS

### Route Fidelity Movements

Inspection of the humpback whale, great white shark, and northern elephant seal tracking datasets revealed multiple examples of migratory route fidelity (**Table 1**). Migration tracks exhibiting route fidelity were followed by: humpback whales tagged off the coasts of Brazil and Rarotonga, Cook Islands; great white sharks tagged off southern New Zealand and Guadalupe Island, Mexico; northern elephant seals tagged in Año Nuevo State Park, California, U.S.A. Our satellite tracking results demonstrate that these diverse marine megafauna demonstrated a remarkable ability to find and follow near-identical paths despite the fact that none of these individuals was ever within 100 km of the same geographic location at the same Julian calendar date and time.

### South Atlantic Humpback Whale Route Fidelity Movements

Three humpback whales tagged off Brazil (PTT identification numbers: 87760.08; 87761.08; 87769.08) swam near-identical south then southeast directed asynchronous migration paths away from Abrolhos Bank off the eastern coast of Brazil (ca. −39◦W; −19.8◦ S; **Figure 1A**; Zerbini et al., 2006) and toward higher-latitude feeding grounds in the South Atlantic Ocean (ca. −50 to −60◦ S; **Figure 1A**; Zerbini et al., 2011). All three whales followed approximately the same ∼1,550 km southerly path away from Abrolhos Bank, during the first 12 (PTT 88760.08), 15 (PTT 87769.08), and 16 (PTT 87761.08) days of their temporally distinct migrations (**Figure 1A**). These three whales migrated through a migratory corridor that was at most ∼100 km wide despite swimming across a vast expanse of open-ocean at different times. Average swimming speeds during these southdirected movements ranged between 4.3 and 5.0 km/h (±1.1 to 1.4 km/h, ±SD).

Following these initial movements, two of the whales (PTT 87760.08 and PTT 87769.08) turned to a southeast heading until both tags stopped transmitting another 17 and 24 days later, respectively (N.B. PTT 87761.08 stopped transmitting ca. −38◦W; −34◦ S). During these southeast-directed movements, 88760.08 and 87769.08 swam an additional 1,991 and 2,581 km, respectively, through a <85 km wide open-ocean corridor despite migrating ∼15 days apart (**Figure 1A**). The average swimming speeds during these southeast-directed movements were 4.8 and 4.4 km/h (±0.9 and ±1.5 km/h, ±SD) for whales 87760.08 and 87769.08, respectively. For comparison, we include two additional migration tracks of humpback whales that migrated through the same corridor, but along distinctly different geographic coordinate trajectories, in 2005 (PTT 10946.05, **Figure 1A**) and 2012 (PTT 111871.12, **Figure 1A**).

### South Pacific Humpback Whale Route Fidelity Movements

Four humpback whales tagged off Rarotonga, Cook Islands (PTT identification numbers: 37282.07; 120946.14; 120947.14; 121195.14) swam similar asynchronous northwest directed migration paths away from Rarotonga, in the southern Cook Islands (ca. −159◦W; −21◦ S; **Figure 1B**), to Tutuila, American Samoa (ca. −171◦W; −14◦ S; **Figure 1B**). All four of these route fidelity movements began in mid-September, albeit 7 years apart (one in 2007, three in 2014). All four whales followed similar ∼1,300 km long northwest directed paths to Tutuila, American Samoa, during the first 12 to 14 days of their intratropical movements despite migrating at distinctly different times TABLE 1 | Platform transmitting terminal (PTT) tag deployment information.


\*Average Velocity was determined during the period of continuous and directed open-ocean movement. Individuals not known to perform route fidelity movements (i.e., non-route fidelity tracks) are indicated by a double asterix (\*\*).

(**Figure 1B**). The Rarotonga-Tutuila migratory corridor used by these whales was at most 150 km wide in the area ∼300– 400 km northwest of Rarotonga and 50–100 km wide elsewhere (**Figure 1B**). The open-ocean swimming speeds for the four intra-tropical movements ranged between 3.9 and 5.0 km/h.

#### South Pacific Great White Shark Route Fidelity Movements

Two great white sharks departed the waters off southwest New Zealand in early to mid-July, 2014 and 2015 (PTT 55614), and early to mid-September, 2014 (PTT 55615). Although the two sharks followed distinctly different paths between New Zealand and the southeast Great Barrier Reef, Australia, shark 55614 followed a near-identical route between New Zealand and Australia in both 2014 and 2015 (**Figure 2A**). Additionally, sharks 55614 and 55615 both followed a similar return migration path from coastal waters off Byron Bay, Australia, to coastal waters off southwest New Zealand (**Figure 2A**) in late November, 2014 (PTT 55615), and early December, 2014 (PTT 55614).

The satellite tracking results acquired for these two sharks include examples of both intra-individual (i.e., PTT 55614, New Zealand to Australia in 2014 and 2015) and inter-individual (i.e., PTT 55614 and PTT 55615, Australia to New Zealand in 2014) route fidelity. Like both the South Atlantic and South Pacific humpback whale datasets, satellite tracking demonstrates that these two sharks can find and follow <150 km wide migratory corridors during temporally separated open-ocean movements that are well in excess of 1,000 km distance during multiple weeks of continuous swimming. For comparison, we include a third great white shark (PTT 55612) that performed a round-trip migration between Stewart Island, New Zealand, and the Loyalty Islands, New Caledonia (ca. 167◦E; −21◦ S; **Figure 2A**), during 2013.

#### North Pacific Great White Shark Route Fidelity Movements

Two great white sharks tagged off Guadalupe Island, Baja, Mexico (PTT identification numbers: 19787; 20720) swam similar to near-identical asynchronous west-southwest directed migration paths away from Guadalupe (ca. −118◦E; 29◦N) to the central northeast Pacific shared offshore foraging area, or SOFA (ca. −134◦W; 23◦N; **Figures 2B,C**; Domeier et al., 2012). These movements were repeated each year from 2009 to 2012, by shark 19787 (**Figure 2B**), and 2009 to 2011, by shark

FIGURE 2 | Route fidelity and non-route fidelity movements of satellite-monitored great white sharks, including: (A) great white sharks in the South Pacific Ocean, (B,C) great white sharks in the North Pacific Ocean. Symbol sizes and color hue correspond with average daily track velocity (km/h) as shown in the legends. Unique platform transmitting terminal (PTT) tag numbers to the left of the PTT number decimal point correspond with individual sharks and are represented by symbol shape; digits to the right of PTT number decimal point correspond with the abbreviated year in which the whale was tagged and are represented by symbol color family (i.e., red, yellow, brown, etc.). Addition (PTT 55612.13) and black/gray diamond (PTT 55615.14) symbols in (A) correspond with long-distance non-route fidelity migrations. 20720 (**Figure 2C**). The tracking data indicate that shark 19787 consistently departed Guadalupe several weeks earlier than shark 20720. Specifically, shark 19787 departed Guadalupe between:


Whereas, shark 20720 departed Guadalupe between:


Piecewise linear regression breakpoint analysis of longitude vs. time profiles for these seven separate migrations suggests that they likely began on, or about, January 15, 2009, January 24, 2010, February 4, 2011 and January 9, 2012 for shark 19787, and March 22, 2009, April 1, 2010 and March 18, 2011 for shark 20720.

Despite the relatively low temporal resolution of the Mexico great white shark dataset, these data reinforce the observation that great white sharks are capable of migrating through welldefined <150 km wide migratory corridors with a high degree of inter-annual route fidelity (**Figures 2B,C**). Four years in a row, shark 19787 followed the same west/southwest-directed path from Guadalupe to the SOFA (**Figure 2B**). Shark 20720 followed a similar but consistently more southerly path than 19787 every year between 2009 and 2011 (**Figure 2C**).

#### North Pacific Northern Elephant Seal Route Fidelity Movements

Four northern elephant seals tagged at Año Nuevo State Park, California, U.S.A. (PTT identification numbers: 39454.04, 13365.04, 39455.05; 62036.11) swam similar to near-identical asynchronous northwest directed migrations between the central California coast and the central North Pacific Ocean/Gulf of Alaska (**Figure 3**). In contrast to previous research documenting intra-individual route fidelity in male northern elephant seal tracks (Le Boeuf et al., 2000), the seal tracks we analyzed demonstrate inter-individual route fidelity achieved by three females and one male. These seals departed the California coast at Año Nuevo State Park between June 9, 2004 and February 21, 2011 (**Table 1**). Despite migrating across a >6 year period, these four seals followed the same ∼150 km wide migration route during their individual 30–38 day, and 2,500–4,000 km long, northwest-directed migrations (**Figure 3**).

Like the humpback whale and great white shark datasets we analyzed, satellite tracking demonstrates that these four seals were able to find and follow a well-defined migratory corridor during asynchronous long-distance open-ocean movements spanning several weeks of continuous swimming. Although we focus our analysis on the initial route fidelity stage of these elephant seal migrations, these same seals dispersed across ∼3 million km<sup>2</sup> of the North Pacific Ocean during subsequent non-route fidelity stages of their individual migrations. For comparison, we include two additional migration tracks of northern elephant seals that followed distinctly different

FIGURE 3 | Route fidelity and non-route fidelity movements of satellite-monitored northern elephant seals. Symbol sizes and color hue correspond with average daily track velocity (km/h) as shown in the legends. Unique platform transmitting terminal (PTT) tag numbers to the left of the PTT number decimal point correspond with individual seals and are represented by symbol shape; digits to the right of PTT number decimal point correspond with the abbreviated year in which the whale was tagged and are represented by symbol color family (i.e., green, yellow, black, etc.). Yellow multiplication (PTT 133774.14) and addition (PTT 120078.14) symbols correspond with long-distance non-route fidelity migrations.

geographic coordinate trajectories in 2014 (PTT 120078.14 and 133774.14, **Figure 3**).

#### Timing of Route Fidelity Movements

Satellite-monitored PTT tracking of humpback whales, great white sharks, and northern elephant seals in the Atlantic and Pacific reveals that all three species have navigational systems capable of reproducing near-identical movements across vast expanses of featureless ocean (**Figures 1**–**3**). At a minimum, these remarkable examples of navigational reproducibility require both precise and accurate spatial orientation. The fact that all three of the sharks that were tracked for more than 1 year (i.e., PTT identification numbers: 55614, 19787, 20720) also showed a migratory fidelity to specific times of the year further suggests there is a temporal component to open-ocean navigation and migration.

Indeed, one of the distinctive features of the route fidelity movements we report is the fact that they occurred at different times. The asynchrony of these movements is important: little navigational information can be gained from analysis of two or more marine migrants that are swimming together as the coordinate space trajectories followed by these animals would be indistinguishable from any geophysical or environmental perspective. Thus, it is the combined effects of the movement asychrony and the temporal dependence of orientation cues available from the environment that make route fidelity movements potentially novel indicators of navigational behavior. We focused our temporal analysis of the route fidelity movements described above on whether or not a systematic temporal pacing was present.

Spectral analysis using the Lomb-Scargle periodogram algorithm (Scargle, 1982) performed on individual route fidelity latitude-time datasets demonstrates that significant (p < 0.05) periodicities are present in all of the route fidelity movements with 15 or more average daily PTT locations. The fact that significant periodicities were not detected in the route fidelity tracks with <15 locations in total does not necessarily mean that these movements lacked temporal pacing. Rather, the relatively small number of PTT locations in these spatially and temporally shorter tracks may simply preclude periodicity detection using the Lomb-Scargle algorithm.

Significant periodicities were detected in three South Atlantic humpback whale, two South Pacific white shark and four North Pacific elephant seal route fidelity movements (**Figure 4**). These nine tracks represent the nine longest distance and duration annual movements in our dataset. None of the shorter duration intra-tropical Rarotonga humpback whale tracks, nor any of the Guadalupe Island white shark tracks, included a detectable significant periodicity. Spectral analysis of the nine longest duration tracks detected significant periodicities with average peak powers of 15 days (SD = 3 days; n = 3; **Figure 4**) and 27 days (SD = 2 days; n = 7; **Figure 4**). These significant periodicities are not unlike the period (range = 29.3–29.8 days) and semiperiod (range = 14.6–14.9 days) of the lunar illumination cycle (i.e., synodic cycle). Periodicities in these ranges are not entirely unexpected given the growing number of studies demonstrating that lunar illumination is strongly correlated with a variety of organismal behaviors including animal movement (e.g., Larkin and Keeton, 1978; Grau et al., 1981; Baird et al., 2003; Tsukamoto et al., 2003; DeVries et al., 2004; Fraser, 2006; Pinet et al., 2011; Erisman et al., 2012; Schmidt et al., 2012; Sudo et al., 2014).

The tracks we studied further demonstrate that synodic periods are also present in the time gaps separating individual movements (Figure S1). For example, the three 2008 South Atlantic whales tracks that individually include significant semisynodic periodicities (**Figure 4**), also demonstrate an intertrack semi-synodic separation that is maintained throughout the southward migrations of these whales (Figure S1A). This ∼15 day separation persists despite a reduction in movement velocity between ∼30◦ S and ∼40◦ S latitude in all three tracks (**Figure 1A**). Two other South Atlantic humpbacks that followed distinctly different geographic coordinate paths (i.e., non-route fidelity movements; **Figure 1A**) across the same southeastdirected migratory corridor in 2005 (PTT 10946.05) and 2012 (PTT 111871.12) performed their migrations 36 and 52 lunar synodic cycles before and after the 2008 whales, respectively (Figure S1A).

Despite the fact that no significant periodicities were detected within any of the individual Rarotonga humpback whale route fidelity movements, these four intra-tropical movements occurred approximately 0.5, 86.0, or 86.5, synodic cycles apart (Figure S1A). Similar time gaps, ranging between 0.5 and 70.5 synodic cycles, separate both the route fidelity and non-route fidelity movements of the white sharks and elephant seals we tracked (Figures S1B–D) suggesting there is a strong temporal component to marine megafauna movement.

Time-series analyses of the humpback whale, great white shark, and northern elephant seal movements we report demonstrate that these movements include significant and systematic temporal periodicities that are similar in duration to the synodic and semi-synodic cycles. However, the synodic cycle is just one of the many quasi-monthly lunar cycles (i.e., sidereal, anomalistic, tropical, etc.) modulated by the relative position of the moon. The empirical results we report demonstrate that humpback whales, great white sharks and northern elephant seals are capable of performing long distance movements that include both spatial and temporal fidelity to well-defined migratory trajectories.

#### Movements in Magnetic Coordinates

Our analyses demonstrate that magnetic inclination is a strong predictor of route fidelity movements (**Table 2**), consistent with the hypothesis that magnetic inclination informs navigation. At the population level, route fidelity movements appear as overlapping in-phase or antiphase sinusoids when magnetic inclination values are plotted against moon illumination (**Figure 5**). These unexpected systematic and symmetrical correlations require temporal pacing: even individuals that followed the same track at the same speed would not be expected to show this pattern unless their movements were initiated at similar, or antithetical, times in the synodic cycle. The systematic nature of the movement trajectories shown in **Figure 5** is reinforced by the fact that the overlapping in-phase/antiphase TABLE 2 | Significant (p < 0.05) sinusoidal regression correlation coefficients for route fidelity movements in geophysical coordinates relative to moon illumination.


(n) Corresponds with the number of average daily locations included in the open-ocean route fidelity movement analyzed. Gray shading highlights the largest correlation coefficient determined for each route fidelity movement. Bold-face and italicized font indicates the second largest correlation coefficient determined for each route fidelity movement.

pattern persists despite temporal separations between individual tracks as long as 7 years. Although these magnetic inclination vs. moon illumination trajectories are highly unexpected, they are consistent with the synodic and semi-synodic periodicities revealed by spectral analysis.

Perhaps even more surprising, however, is the finding that the route fidelity movements of individual species appear to overlap despite extreme geographic separations between the different populations studied. For example, both the South Atlantic and South Pacific humpback whale route fidelity tracks, and both the South Pacific and North Pacific great white shark route fidelity tracks, describe near continuous magnetic inclination vs. moon illumination trajectories (**Figures 5A–D**). Six of the seven humpback whale movements we report departed coastal breeding grounds, in areas with magnetic inclination values ca. −38 to −39◦ , within a few days of full or new moon (**Figure 5A**). Similarly, most of the South Pacific and North Pacific great white sharks arrived at, or departed from, coastal locations in Australia or Mexico, in areas with magnetic inclination values ca. ±53 to ±55◦ , within a few days of full or new moon.

Sinusoidal regression with respect to the moon illumination cycle reveals that magnetic inclination is the strongest magnetic

FIGURE 5 | Magnetic inclination vs. moon illumination plots. (A,B) Humpback whales in the South Atlantic and South Pacific Oceans, (C,D) great white sharks in the South Pacific and North Pacific Oceans, (E,F) northern elephant seals in the North Pacific Ocean. Symbols as in Figures 1–3. In (B,D,F), average daily location symbols have been removed, and the tracks of PTT numbers 87769.08, 120946.14, 55614.15, 55615.15, 19787.10, 19787.11, and 39455.05 are plotted against the upper reverse moon illumination axis. These seven tracks are thus mirrored across the 0.5 (i.e., 50%) moon illumination value. (B,D,F) are included in order to show the symmetrical in-phase or antiphase distribution of the data with respect to the synodic cycle. These 7 mirror-image tracks are plotted as lighter colored lines. All other tracks (n = 15) are plotted against the lower moon illumination axis and are shown as darker colored lines. Magnetic inclination values in (C,D) are plotted as absolute values due to the fact that the two white shark populations reside in opposite magnetic hemispheres. Magnetic inclination axis values span the same range in all panels to facilitate comparisons between the three species.

spherical coordinate predictor of individual and populationlevel route fidelity movements (**Table 2**). However, these same movements also exhibit significant sinusoidal correlations with respect to both magnetic field intensity and magnetic declination in the majority of tracks analyzed (**Table 2**). Thus, it is not possible to conclusively rule out potential roles for any of the spherical magnetic coordinates in whale, shark or seal navigation based on these analyses alone.

However, magnetic inclination explains a higher proportion of the sinusoidal regression model variance at the individual track level (median = 96%; average = 93%; range = 83–99%; SD = 4.2%; n = 19) than magnetic field intensity (median = 92%; average = 84%; range = 42–99%; SD = 16.4%; n = 19) or magnetic declination (median = 79%; average = 80%; range = 52–99%; SD = 14.2%; n = 17). In contrast, when all of the route fidelity tracks from all populations of all species are combined, magnetic field intensity is the strongest magnetic coordinate predictor, explaining 33% of the model variance (**Table 2**), followed by magnetic inclination (28%) and magnetic declination (21%).

#### Movements in Gravitational Coordinates

G-space plots of marine megafauna route-fidelity migrations reveal that these long-distance open-ocean movements describe highly symmetrical trajectories (**Figure 6**) not unlike the magnetic inclination trajectories reported above. Several aspects of these results are notable.

First, all three South Atlantic humpback whales departed the southeast corner of Abrolhos Bank (ca. gL+g<sup>B</sup> = 978.7 to 978.8 Gal) at antithetical positions of the synodic cycle, 2–3 days prior to full or new moon (**Figure 6A**). These whales maintain this symmetry, about a quarter moon (i.e., 0.5 or 50% moon illumination) mirror plane, throughout their migrations toward higher-latitude feeding grounds (**Figure 6B**). For example, 88760.08 and 87769.08 both pass the gravitational half-way point in their migrations (ca. gL+g<sup>B</sup> = 980.42 Gal; −33.1◦W; −37.4◦ S) when the moon is either full or new (**Figure 6A**). This spatiotemporal symmetry is a direct consequence of the fact that all three whales systematically reduced their swimming speeds by ∼1.5–2.0 km/h as they approached the gravitational half-way point of their southward migrations (**Figure 6A**).

Second, similar symmetrical g-space trajectories are present in the four South Pacific humpback whale route fidelity tracks (**Figure 6A**). Despite migrating at distinctly different calendar dates across a 7 year period, all four whales departed Rarotonga, Cook Islands (ca. −159.8◦W; −21.2◦ S) at antithetical positions of the moon illumination cycle within hours of full or new moon. All four whales arrived off the southeast coast of Tutuila, American Samoa (ca. −170.8◦W; −14.4◦ S), near the subsequent full or new moon (**Figure 6A**). As recognized in the South Atlantic dataset, these movements are symmetrical across the gravitational half-way point (ca. 979.01 Gal; −167◦W; −16.3◦ S in geographic coordinates; **Figure 6B**). This symmetry exists despite each whale: (1) following slightly different routes during the first week of their movements to Tutuila (**Figure 1B**); and (2) swimming at different speeds during different stages of their movements (**Figure 1B**).

Third, although the great white shark tracks we analyzed predominantly demonstrate intra-individual route fidelity, these tracks have many of the same characteristics as the humpback whale movements when plotted in gravitational coordinates (**Figure 6C**). For example, during its near-identical 2014 and 2015 northward migrations to Australia, white shark 55614 departed southwest New Zealand within 24 h of the July, 2014, full moon and the August, 2015, new moon. During these geographically coincident northward migrations, 55614 passed the gL+g<sup>B</sup> midpoint value of 980.42 Gal within 24 h of both the July, 2014, new moon and August, 2015, full moon (**Figure 6D**), at a geographic position located ∼350 km southeast of Sydney, Australia. During its intervening 2014 southward migration, 55614 passed the same 980.42 Gal gravitational midpoint within 24 h of the December, 2014, new moon (**Figure 6C**) at a geographic position located ∼500 km east of where it passed the same g-space midpoint during its northward passage (**Figure 2A**).

Despite following distinctly different northward and southward geographic coordinate routes during its seasonal migrations between New Zealand and Australia (**Figure 2A**), the g-space trajectories of white shark 55614's movements are all but indistinguishable from the g-space trajectories followed by three South Atlantic humpback whales (**Figure 6A**) and at least one other south Pacific white shark (PTT 55615; **Figures 6C,D**). Our analyses further show that South Pacific white sharks 55614 and 55615 followed diametrically opposed g-space trajectories during their 2014 return migrations from Australia despite performing these migrations approximately 15 days apart (**Figures 6C,D**, Figure S1C).

Fourth, the seven examples of intra-individual route fidelity performed by the two North Pacific great white sharks tagged off Guadalupe Island provide further empirical evidence of the observed pattern of symmetrically distributed route fidelity movements in gravitational coordinates (**Figure 6C**). For example, every year between 2009 and 2012, white shark 19787 swam a near-identical route from Guadalupe Island to the SOFA (**Figure 2B**), passing the gL+g<sup>B</sup> midpoint (ca. 979.47 Gal) of its migration route near the: (1) January, 2009, new moon; (2) January, 2010, full moon; (3) February, 2011, full moon; (4) January, 2012, new moon (**Figures 6C,D**). Similarly, great white shark 20720 passed the same gL+g<sup>B</sup> midpoint (ca. 979.47 Gal) within 48 h of both the March new moon and the April new moon during its 2009 and 2011 migrations to the SOFA, respectively (**Figures 6C,D**).

Fifth, g-Space plots of the near-identical routes followed by four North Pacific northern elephant seals further reinforce the movement patterns described above. All four seals followed gspace trajectories that were highly symmetrical across the gL+g<sup>B</sup> gravitational mid-point between Año Nuevo and the Aleutian Trench (ca. 981.1 Gal; **Figure 6E**) despite departing the coast at Año Nuevo State Park at distinctly different positions in the moon illumination cycle. These same g-space trajectories are also highly symmetrical across a mirror plane projected through the 50% moon illumination position in the synodic cycle (**Figure 6F**), not unlike the pattern present in both the humpback whale and white shark trajectories. These mirror-image g-space trajectories

FIGURE 6 | Latitude and bedrock dependent gravity vs. moon illumination "g-space" plots. (A,B) Humpback whales in the South Atlantic and South Pacific Oceans, (C,D) great white sharks in the South Pacific and North Pacific Oceans, (E,F) northern elephant seals in the North Pacific Ocean. Symbols as in Figures 1-3. In (B,D,F), average daily location symbols have been removed, and in (B,F) the same tracks are plotted against the upper reverse moon illumination axis as in Figures 5B,F. In (D), PTT numbers 19787.09, 19787.12, 20720.09, 20720.10, and 20720.11 are plotted against the upper reverse moon illumination axis and PTT numbers 19787.10 and 19787.11 are plotted against the lower moon illumination axis. All mirror-image tracks plotted against the upper moon illumination axis are displayed as lighter color lines. gL+gB corresponds with the sum of the latitude (gL ) and bedrock (gB) dependent vertical gravitational accelerations at each individual location (see Nomenclature).

are present due to the fact that all four seals passed the latitude and bed-rock dependent 981.1 Gal position within 6-h of a quarter-moon (**Figures 6E,F**) despite these movements being separated by as much as 7 years in time.

As part of our analysis of the recurrent pattern of symmetrical g-space trajectories observed in all three species, we tested the gravitational datasets for significant correlations using sinusoidal regression. Like we found for the spherical magnetic coordinate movement trajectories, sinusoidal regression of the gravitational trajectories demonstrates that latitude and bedrock dependent gravity is a strong predictor of the individual route fidelity movements (median = 98%; average = 97%; range = 83–99%; SD = 4.1%; n = 20; **Table 2**). The highly symmetrical in-phase or antiphase g-space trajectories we report reinforce the strong inter-annual pacing of route fidelity movements with respect to the moon illumination cycle (**Figure 6**).

Our results demonstrate that the spatially dependent gravitational cue, gL+gB, is a stronger predictor than spherical magnetic coordinates at the individual animal, population, species and inter-species levels (**Table 2**). This unexpected finding is supported by the facts that: (1) gravity is the strongest predictor of 15 of the 22 route fidelity movements we analyzed, whereas magnetic inclination is only the strongest predictor for 4 of the individual movements (**Table 2**); (2) gravity is the strongest route fidelity movement predictor for 3 of the 5 marine megafauna populations we studied (**Table 2**); (3) gravity is the strongest predictor of the route fidelity movements of all three species (**Table 2**); (4) gravity cues explain 69% of the sinusoidal regression model variance when all of the route fidelity data are concatenated, whereas magnetic inclination only explains 28% of the variance (**Table 2**; **Figure 7**).

### DISCUSSION

#### Geophysical Navigation

None of the significant correlations we report demonstrate causality between geophysical orientation cues available from the environment and navigational decisions. Such causal relations can only be established by experimental testing under controlled conditions. However, the strong and systematically patterned correlations we report can be used as an empirical data-based platform from which hypotheses can be proposed.

For example, magnetic coordinate projections of an animal's repeated utilization of a well-defined geographic coordinate migration route will describe distinctly different magnetic coordinate trajectories due to changes in Earth's main magnetic field through time. Yet, the opposite scenario is also true. The geographic coordinate paths followed between migratory destinations might instead systematically shift through time in response to magnetic secular variation in situations where magnetic cues are the primary source of orientation information. Thus, the extent to which an individual migrant uses magnetic cues for navigational purposes can be further explored through multi-annual tracking of long-lived individuals. For large and elusive species that are difficult to study under controlled conditions, long-term repeat tracking represents an important opportunity to better understand their navigation at sea.

The widespread deployment of PTT tags on animal migrants facilitates longitudinal studies, and long-term tracking studies are viable for some species (e.g., Berthold et al., 2004; Alerstam et al., 2006). Based on the empirical results presented in the current study, we hypothesize that the magnetic coordinate trajectories followed by individual migrants over multiple migratory cycles will describe overlapping and symmetrically distributed paths when plotted against the moon illumination cycle despite changes in the raw geocentric magnetic coordinate values caused by secular variation of the magnetic field. If true, this hypothesis would provide evidence in support of temporally modulated navigation with respect to Earth's magnetic field. Given the relatively small changes in magnetic field conditions from 1 year to the next, tracking studies that span a decade or more will produce the strongest results.

With respect to gravity, the possibility that animal navigation is informed by cues derived from Earth's spatially and temporally dynamic gravitational field was suggested at least 40 years ago (Larkin and Keeton, 1978). However, the possibility that animal orientation is informed by gravitational cues remains largely untested. The empirical results we present suggest further experiments, like those performed by Blaser et al. (2013, 2014), Fisahn et al. (2015), and Cresci et al. (2017) will improve our understanding of the role, if any, gravity plays in animal orientation. We particularly encourage experimental tests on model organisms (e.g., zebrafish, Danio rerio; honey bees, Apis mellifera; eels, Anguilla spp.) that are sensitive to both magnetic (Kirschvink, 1981; Tesch et al., 1992; Osipova et al., 2016) and gravitational cues (Korall and Martin, 1987; Moorman and Shorr, 2008; Cresci et al., 2017). Integrated analysis of telemetry datasets will further help identify the ways in which geophysical cues are used for navigational purposes.

Long-distance inter-hemispheric migrants are also attractive targets for future studies. The presence or absence of symmetrical magnetic or gravitational coordinate movement trajectories across either of the spatially distinct magnetic or geographic equators has the potential to provide significant insight into how navigation is performed. The variable separation between the magnetic and gravitational equators at different longitudes presents additional opportunities to determine if emergent patterns in magnetic or gravitational trajectories are reproducible between tracks of related species that separately migrate across the Pacific and Atlantic Oceans.

Tests for non-random and reproducibly patterned movement trajectories, such as the symmetrical geometry of the g-space trajectories we report, can also be more deeply explored for a variety of geophysical and astronomical orientation cues in migratory domains with distinctly different cue distributions. For example, animals that migrate through middle latitude positions from higher or lower latitude habitats are particularly attractive due to the eccentric and anomalous geometries of magnetic, gravitational and astronomical cues through both space and time. Specific targets for future research in this area might include comparisons between populations that migrate across the South Atlantic magnetic anomaly vs. other ocean basins (Figures S3–S8).

Our analyses suggest that gravity might play an important role in long-distance animal navigation, either in concert with magnetic cues or in isolation. Confirming and elucidating this role requires integrated experimental testing and animal tracking. We hypothesize that spatially dependent gravity cues, when plotted against temporally dependent gravity cues, such as tidal gravity, will describe reproducible movement trajectories at the individual level. We further hypothesize that similarly reproducible movement trajectories will be less pronounced, if not absent, when the same individual's movements are analyzed with respect to other geophysical and astronomical cue/coordinate systems.

#### Route Fidelity vs. Non-route Fidelity

Although this study specifically addresses route fidelity movements, it is important to consider whether or not non-route fidelity movements describe geophysical coordinate trajectories that are similar to the route fidelity movement trajectories we report.

Perhaps one of the most unexpected results animal tracking studies have revealed is the extreme diversity in movement trajectories followed by animal migrants when plotted in geographic coordinates (e.g., Block et al., 2011). Few individuals follow the same path. Our research reinforces this observation: of the animals we tracked, we found only 13% (i.e., 15 out of 113 individuals) achieve intra- or inter-individual route fidelity.

However, our dataset may underestimate the prevalence of route fidelity due to the fact that the vast majority of the individuals we tagged were tracked for only one migratory cycle or less. It is also possible that our use of two-dimensional geographic coordinate space projections, when classifying movements as either route fidelity or non-route fidelity, is a flawed approach. Perhaps geographically distinct individual movements become more alike when they are viewed from a different perspective? In an effort to explore this possibility, we compared route-fidelity to non-route fidelity movement projections in geophysical coordinates.

Non-route fidelity humpback whale geophysical coordinate trajectories (Figures S2A,B) have a similar overall sinusoidal shape and pattern as the route fidelity tracks from the same population (**Figures 6A,B**, **7A,B**). These similarities include multiple segments of both magnetic and gravitational coordinate non-route fidelity movement trajectories that overlap the route fidelity movements for extended periods (Figures S2A,B). However, there are also periods, spanning several days of animal movement, during which the non-route fidelity geophysical coordinate trajectories do not overlap the route fidelity trajectories (Figures S2A,B).

The non-route fidelity great white shark trajectories show a similarly ambiguous pattern. These tracks describe largely coincident magnetic and gravitational trajectories during the first-half of the northward movements away from New Zealand for both sharks, yet, distinctly different trajectories in the secondhalf of the northward movements (Figures S2C,D). Similar variations are also apparent during the southward movement of 55612.13 (Figures S2C,D).

Similarly equivocal patterns are also present in the nonroute fidelity northern elephant seal trajectories. In this case, neither of the non-route fidelity tracks overlap the magnetic coordinate route fidelity trajectories (Figure S2E). However, in marked contrast to the magnetic trajectories, both of the nonroute fidelity tracks extensively overlap with the gravitational route fidelity trajectories (Figure S2F).

The varying degrees of similarity between non-route fidelity geophysical coordinate trajectories and route fidelity trajectories suggests that the navigational system(s) being utilized are calibrated at the individual level.

To explore this possibility, we normalized both magnetic inclination and gL+g<sup>B</sup> values to the values present in core areas inhabited by each individual immediately prior to the onset of long-distance movement. For the northern elephant seals, magnetic inclination and gL+g<sup>B</sup> values were normalized to the values present at Año Nuevo State Park, California, U.S.A. For the South Pacific great white sharks and the South Atlantic

bivariate spatiotemporal coordinate spaces. Lines in (A) colored as in Figure 5. Lines in (B) colored as in Figure 6.

FIGURE 8 | Site-normalized South Atlantic humpback whale magnetic inclination and gravitational trajectories. In this figure geocentric magnetic inclination and latitude and bedrock dependent gL+gB values have been normalized to the values present at the locations inhabited immediately prior to the onset of long-distance open-ocean migration at the individual level. Site-normalized magnetic inclination (A) and gravitational acceleration (B) values are displayed as percentages and are plotted against the moon illumination cycle as in Figures 5–7. Route fidelity movements are shown as light blue lines and non-route fidelity movements are symbolized as indicated in the figure legends. Non-route fidelity track symbol colors correspond with day of the year and exceed 365 days when the tracked movement spanned Julian calendar years.

FIGURE 9 | Site-normalized South Pacific great white shark magnetic inclination and gravitational trajectories. In this figure geocentric magnetic inclination and latitude and bedrock dependent gL+gB values have been normalized to the values present at the locations inhabited immediately prior to the onset of long-distance open-ocean migration at the individual level. Site-normalized magnetic inclination (A) and gravitational acceleration (B) values are displayed as percentages and are plotted against the moon illumination cycle as in Figures 5–7. Route fidelity movements are shown as dark red lines and non-route fidelity movements are symbolized as indicated in the figure legends. Non-route fidelity track symbol colors correspond with day of the year and exceed 365 days when the tracked movement spanned Julian calendar years.

humpback whales, magnetic inclination and gL+g<sup>B</sup> values were normalized to the values present at each individual's last known continental shelf location prior to the onset of open-ocean migration. The South Pacific humpback whale and North Pacific great white shark tracks were not included in this analysis due to uncertainties regarding departure site locations for these individuals.

When compared to site-normalized route fidelity migrations, the site-normalized non-route fidelity geophysical coordinate movement trajectories exhibit several distinctive features (**Figures 8**–**10**). First, site-normalized non-route fidelity magnetic inclination trajectories generally do not overlap with route fidelity magnetic inclination trajectories. Second, site-normalized magnetic inclination trajectories exhibit limited symmetry across the 50% moon illumination mirror plane (**Figures 8A**, **9A**, **10A**). Third, both site-normalized route fidelity and site-normalized non-route fidelity gravitational coordinate trajectories predominantly plot as either overlapping in-phase or anitphase sinusoids for all three species (**Figures 8B**, **9B**, **10B**). Fourth, in comparison to the magnetic inclination trajectories, site-normalized gravitational trajectories exhibit a more pronounced symmetry across both the gL+g<sup>B</sup> midpoint values in the migratory domain utilized by each species and the 50% moon illumination mirror plane (**Figures 8B**, **9B**, **10B**).

The systematic nature of the highly symmetrical site-normalized trajectories shown in **Figures 8**–**10** suggest there is a temporally modulated triggering and/or pacing to long-distance migratory movement behavior. Considering the high specific gravity and extreme crystallographic symmetry of magnetic biominerals, we hypothesize that exogenous magnetic and gravitational cues are integrated components of a spatiotemporal orientation system that is calibrated at the individual level.

#### Future Directions

Knowledge gaps in our understanding of how animals navigate limit our ability to assess and anticipate the sensitivity of migrating animals to perturbations resulting from environmental change, anthropogenic activities, and predator-prey distribution (Hays et al., 2016). Empirical approaches, such as animal telemetry, can be used to develop and assess data-based models of animal movement, with the strongest models creating opportunities to inform conservation and management decision making with respect to both space and time. At a fundamental level, telemetry studies remain uniquely powerful ways to inform marine conservation through databased demonstrations of when and where marine megafauna are located across vast expanses of open-ocean (Shillinger et al., 2008; Gredzens et al., 2014; Maxwell et al., 2016; Dawson et al., 2017). Technological advances in animal telemetry, combined with advances in remote sensing (e.g., Figures S3–S8), have created unprecedented opportunities

FIGURE 10 | Site-normalized North Pacific northern elephant seal magnetic inclination and gravitational trajectories. In this figure geocentric magnetic inclination and latitude and bedrock dependent gL+gB values have been normalized to the values present at the locations inhabited immediately prior to the onset of long-distance open-ocean migration at the individual level. Site-normalized magnetic inclination (A) and gravitational acceleration (B) values are displayed as percentages and are plotted against the moon illumination cycle as in Figures 5–7. Route fidelity movements are shown as dark green lines and non-route fidelity movements are symbolized as indicated in the figure legends. Non-route fidelity track symbol colors correspond with day of the year and exceed 365 days when the tracked movement spanned Julian calendar years.

to retrospectively extract and analyse the geophysical and oceanographic conditions experienced by individual migrants during long-distance migration. Integrated analysis of the data produced by satellite telemetry and remote sensing tools strengthens our understanding of movement behavior at the individual scale.

The data and analyses we report provide a platform for future research that specifically targets both the spatial and temporal aspects of long-distance animal migration. The recently proposed "chord and clock" model of animal navigation (Horton et al., 2014) explicitly includes exogenous temporal cues, consistent with the highly correlated movements we report here. The integrated spatiotemporal perspective that defines the "chord and clock" model provides a novel parallel to the widely accepted, yet purely spatial, positional and directional orientation frameworks proposed by Griffin (i.e., Type III "true" navigation; Griffin, 1952) and Kramer (i.e., "map and compass" navigation; Kramer, 1961).

Future research on the biogeophysics of animal navigation will facilitate the development of a mechanistic understanding of how animals find their way. Given the speed at which oceanic environments are currently changing, technologicallydriven data-based tests of the observation that geophysical cues are strong predictors of the open-ocean movements of diverse marine megafauna are urgently required.

#### ETHICS STATEMENT

All animal tracking research reported here was carried out in accordance with animal ethics consents given to the authors by their home institutions and/or relevant government agency. Northern elephant seal tracking was performed in accordance with approvals granted by the University of California at Santa Cruz Institutional Animal Care and Use Committee and under National Marine Fisheries Service permits #786-1463 and #87-143. Humpback whale tracking was performed in accordance with approvals granted by the Brazilian

#### REFERENCES


Environmental Agency (IBAMA), permit #009/02/CMA/IBAMA and process #02001.000085/02-27 and by the Cook Islands Government, Office of the Prime Minister, to Cook Islands Whale Research. Great white shark tracking was performed in accordance with approvals granted by Secretaria de Medio Ambiente y Recursos Naturales (SEMARNAT), Comision Nacional de Areas Naturales Protegidas (CONANP), and the New Zealand Department of Conservation.

#### AUTHOR CONTRIBUTIONS

TH and RH conceived of the study. TH designed the current study, performed the magnetic, gravitational and astronomical analyses, prepared all of the figures and wrote the manuscript. NH, AZ, MF, MD, AA, DC, PR, CD, NN-L, and PC provided the humpback whale, great white shark and northern elephant seal PTT tracking data used in the current study. All authors contributed to the revision of the initial manuscript.

#### ACKNOWLEDGMENTS

Shell Brasil funded humpback whale tagging in Brazil. Great white shark tracking in New Zealand was funded by: New Zealand Ministry of Business, Innovation and Employment; New Zealand National Institute of Water and Atmospheric Research; New Zealand Department of Conservation. Northern elephant seal tracking was funded in part by U.S. Office of Naval Research grant N00014-08-1-1195 to DC and conducted under NMFS Permit 14636.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2017.00422/full#supplementary-material

with pigeon homing–a GPS tracking study. J. Exp. Biol. 217, 4057–4067. doi: 10.1242/jeb.108670


whales Megaptera novaeangliae in the Southwest Atlantic Ocean. Mar. Ecol. Prog. Ser. 313, 295–304. doi: 10.3354/meps313295

Zerbini, A. N., Andriolo, A. R., Heide-Jørgensen, M. P., Moreira, S. C., Pizzorno, J. L., Maia, Y. G., et al. (2011). Migration and summer destinations of humpback whales (Megaptera novaeangliae) in the western South Atlantic Ocean. J. Cetacean Res. Manag. 3, 113–118.

**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 © 2017 Horton, Hauser, Zerbini, Francis, Domeier, Andriolo, Costa, Robinson, Duffy, Nasby-Lucas, Holdaway and Clapham. 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) or licensor 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.

### NOMENCLATURE

**Anomalistic month—**the period required for the moon to return to perigee following the preceding perigee (average = 29.6 days) **Apogee (lunar)—**the point in its orbit at which the moon is farthest from Earth

**Bouguer gravity anomaly (spherical)—**location dependent difference in gravitational acceleration between normal gravity and observed gravity caused by variations in the shape and density of Earth (for a more explicit definition, see Balmino et al., 2012)

**D (magnetic declination)—**location dependent angle in the horizontal plane between Earth's magnetic field and geographic north expressed positive to the east

**Declination (lunar)—**angle between the moon and the celestial equator

**F (magnetic field)—**location dependent magnetic field flux density (informally: intensity) of Earth's magnetic field expressed, in Standard International Units, as nanotesla (nT); radial distance coordinate {ρ} in spherical coordinate space {ρ, θ, ϕ} descriptions of Earth's magnetic field

**Gal—**unit of gravitational acceleration (1 Gal is equivalent to 1 cm s−<sup>2</sup> )

**gB—**spherical Bouguer gravity anomaly

**gL—**latitude-dependent gravitational acceleration

**H (magnetic)—**horizontal component of Earth's magnetic field (F)

**I (magnetic inclination)—**location dependent angle in the vertical plane between Earth's magnetic field and the horizontal (expressed positive downwards)

**Perigee—**the point in its orbit at which the moon is closest to Earth

**PTT—**platform transmitting terminal

**Sidereal month—**the period required for the moon to complete one full orbit relative to a fixed star's position (average = 27.3 days)

**SPOT—**smart position or temperature transmitting tag

**Synodic—**the period required for the moon to complete one full illumination/phase cycle (average = 29.5 days)

**Tropical month—**the period required for the moon to complete one full orbit relative to the ecliptic (average = 27.3 days)

**X (magnetic)—**geographic north-south component of Earth's magnetic field (F) in the horizontal plane

**Y (magnetic)—**geographic east-west component of Earth's magnetic field (F) in the horizontal plane

**Z (magnetic)—**vertical component of Earth's magnetic field (F)

# Thermal Imaging and Biometrical Thermography of Humpback Whales

Travis W. Horton<sup>1</sup> \*, Alice Oline<sup>2</sup> , Nan Hauser <sup>3</sup> , Tasnuva Ming Khan<sup>4</sup> , Amelie Laute<sup>3</sup> , Alyssa Stoller <sup>3</sup> , Katherine Tison<sup>3</sup> and Peyman Zawar-Reza<sup>5</sup>

*<sup>1</sup> Department of Geological Sciences, University of Canterbury, Christchurch, New Zealand, <sup>2</sup> Environment Program, Colorado College, Colorado Springs, CO, United States, <sup>3</sup> Center for Cetacean Research and Conservation, Rarotonga, Cook Islands, <sup>4</sup> Cornell University, Ithaca, NY, United States, <sup>5</sup> Department of Geography, University of Canterbury, Christchurch, New Zealand*

#### Edited by:

*Lars Bejder, Murdoch University, Australia*

#### Reviewed by:

*Nuno Queiroz, University of Porto, Portugal Phil Bouchet, Centre for Marine Futures, School of Animal Biology, University of Western Australia, Australia*

> \*Correspondence: *Travis W. Horton travis.horton@canterbury.ac.nz*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *28 September 2017* Accepted: *12 December 2017* Published: *21 December 2017*

#### Citation:

*Horton TW, Oline A, Hauser N, Khan TM, Laute A, Stoller A, Tison K and Zawar-Reza P (2017) Thermal Imaging and Biometrical Thermography of Humpback Whales. Front. Mar. Sci. 4:424. doi: 10.3389/fmars.2017.00424* Determining species' distributions through time and space remains a primary challenge in cetacean science and conservation. For example, many whales migrate thousands of kilometers every year between remote seasonal habitats along migratory corridors that cross major shipping lanes and intensively harvested fisheries, creating a dynamic spatial and temporal context that conservation decisions must take into account. Technological advances enabling automated whale detection have the potential to dramatically improve our knowledge of when and where whales are located, presenting opportunities to help minimize adverse human-whale interactions. Using thermographic data we show that near-horizontal (i.e., high zenith angle) infrared images of humpback whale (*Megaptera novaeangliae*) blows, dorsal fins, flukes and rostrums record similar magnitude brightness temperature anomalies relative to the adjacent ocean surface. Our results demonstrate that these anomalies are similar in both low latitude and high latitude environments despite a ∼16◦C difference in ocean surface temperature between study areas. We show that these similarities occur in both environments due to emissivity effects associated with oblique target imaging, rather than differences in cetacean thermoregulation. The consistent and reproducible brightness temperature anomalies we report provide important quantitative constraints that will help facilitate the development of transient temperature anomaly detection algorithms in diverse marine environments. Thermographic videography coupled with laser range finding further enables calculation of whale blow velocity, demonstrating that biometrical measurements are possible for near-horizontal datasets that otherwise suffer from emissivity effects. The thermographic research we present creates a platform for the delivery of three important contributions to cetacean conservation: (1) non-invasive species-level identifications based on whale blow shapes and velocities recorded by infrared videography; (2) reduced ship-strike rates through automated thermographic cetacean detection systems deployed in high traffic areas; (3) monitoring the spatial and temporal distributions of endangered animals in remote habitats.

Keywords: humpback whale, infrared, biometrical thermography, emissivity, blow velocity

## INTRODUCTION

Since the International Whaling Commission (IWC) banned commercial whaling in 1986, many baleen whale species have shown signs of recovery (Thomas et al., 2016). However, all species listed as Least Concern under the International Union for Conservation of Nature and Natural Resources (IUCN) Red List of Threatened Species framework also include threatened subpopulations classified as Vulnerable, Endangered, or Critically Endangered (Thomas et al., 2016). The conservation status, recovery and health of whale populations is very much site and context specific: modern human threats to whales, including ship strikes and entanglement in fishing gear, are not evenly distributed with respect to both space and time (Thomas et al., 2016).

Despite the spatially and temporally dynamic challenges associated with cetacean conservation and protection in the postwhaling era, progress has been made. For example, revisions to shipping lane positions, vessel traffic management plans and mandatory maximum vessel speeds along the eastern coast of North America correlate with significant reductions in North Atlantic right whale (Eubalaena glacialis) deaths due to ship strikes (Laist et al., 2014; Thomas et al., 2016). However, the risks associated with ship strikes remain high elsewhere. Necropsies performed on stranded whales demonstrate that at least one humpback whale (Megaptera novaeangliae), one fin whale (Balaenoptera physalus), and two blue whales (Balaenoptera musculus) are killed by ship strikes off the California coast every year (Redfern et al., 2013). Similar analyses suggest that, on average, one Bryde's whale (Balaenoptera edeni) is killed every year by ship strikes in New Zealand's Hauraki Gulf (Constantine et al., 2015). True cetacean mortality rates due to human activities at sea are almost certainly higher (Kraus, 1990), however, and the annual loss of even a single individual can be significant for smaller populations of long-lived species with low recruitment rates (Laist et al., 2001).

In an effort to reduce the risks ships pose to large whales, the IWC has developed a 3-year (2017–2020) Strategic Plan that seeks to increase the development and use of whale avoidance technologies (Cates et al., 2017). Acoustic and infrared automated cetacean detection systems are attractive and emerging tools for enhanced cetacean conservation (Zitterbart et al., 2013; Nowacek et al., 2016). The ability to detect whale blows, as far away as 5 km using around-the-clock 360◦ infrared scanners outfitted with rigorous detection algorithms (Zitterbart et al., 2013), will benefit many, including marine mammal observers onboard large vessels and land-based scientists studying whale movement behavior (e.g., Perryman et al., 1999) and human-whale interactions along rapidly changing coastlines (e.g., Graber, 2011). Infrared thermography can also facilitate the non-invasive collection and monitoring of fundamental biometrical information, including thermal physiology, injury diagnoses and population surveys (McCafferty, 2007).

Infrared cetacean detection systems also create opportunities for conservation biologists and cetacean ecologists to document the spatial and temporal distribution of animals utilizing remote or inaccessible environments. For example, the Oceania subpopulation of humpback whales, the only migratory humpback whales in danger of going extinct (Childerhouse et al., 2008), seasonally inhabit ∼10 million km<sup>2</sup> of the tropical South Pacific Ocean. Yet, only a handful of scientists, spread across an area of ocean the size of China, actively study these whales. Automated detection systems have the potential to dramatically improve our knowledge of when and where these endangered whales are utilizing highly understudied breeding/calving ground habitats.

However, thermal imaging also has several important limitations. Infrared imaging systems are not inexpensive, particularly so for current high sensitivity models with cryogenically cooled detectors or large focal lengths capable of long-range applications. Infrared detectors also require a direct line of site to the target, yet they can also lose functionality through interaction with sea-spray. The data streams generated by infrared imaging systems are large, creating challenges with data handling, analysis and signal processing. Thermal cameras are also highly inaccurate when imaging scenes from nearhorizontal positions due to emissivity effects (Masuda et al., 1988; Cuyler et al., 1992; see Nomenclature).

The effects of emissivity on the brightness temperatures recorded by a thermal imaging device are extremely relevant to cetacean thermography. For example, as a whale exhales, its breath pushes sea water present in the near-surface water column, or nasal depression, or both, into the overlying atmosphere. From observation points at or near sea-level, this spouting of water droplets immediately and drastically changes the angle at which the whale's blow is being measured by the thermal camera. For example, a 2 m high whale blow will be measured perpendicularly (i.e., measured at a 0◦ zenith angle) from an observation point located 100 m distant and 2 m above sea level. In contrast, the adjacent flat ocean's surface will be measured sub-horizontally at an 89◦ zenith angle. Similar to blows, emergent dorsal fins, flukes or rostrums will also be measured at a relatively low zenith angles in relation to the adjacent ocean's surface. These rapid changes in the angle at which the object is being imaged will have large effects on the surface brightness temperatures estimated by the thermal imaging device due to the effect zenith angle of observed radiation has on sea water emissivity (Masuda et al., 1988).

The research we present was driven by three primary objectives, all aligned to the IWC's strategic goal of developing large whale avoidance technologies. We sought to: (1) quantify infrared image brightness temperature and brightness temperature anomaly (BTA) values for humpback whale blows, dorsal fins, flukes, and rostrums in both tropical breeding/calving ground and sub-polar feeding ground habitats; (2) calculate humpback whale blow height and blow velocity through coupling of infrared videography with laser-range finding; (3) evaluate the effects of emissivity on thermal imaging data collected from high zenith angle (i.e., oblique to target) positions. Achievement of these objectives creates a platform from which a variety of cetacean conservation tools can be further developed and delivered.

#### METHODS

Thermal images of humpback whale surfacing features were collected using a Forward Looking Infrared camera (FLIR A615, FLIR Systems, Inc.,) and analyzed using FLIR Tools+ software (FLIR Systems, Inc.,). The FLIR A615 we used had a focal length of 24.6 mm, 25◦ × 19◦ field of view, F-number of 1.0, infrared resolution of 480 × 640 pixels and a detector pixel pitch of 0.017 mm pixel−<sup>1</sup> . The camera's detector comprised an uncooled Vanadium Oxide (VoX) long–wavelength (i.e., 7.5– 14µm) microbolometer (see Nomenclature) with a thermal sensitivity of <0.05◦C. Infrared images were captured every 0.04 s (i.e., 25 Hz) but frame rates as high as 200 Hz can be achieved with the A615's high-speed windowing option. The A615 was powered by a small 12-volt battery externally strapped to the camera's casing. The A615 was also connected to a FZ-G1 ToughPad tablet computer (Panasonic Corporation) via a high-speed USB cable. A GoPro Hero4 camera (GoPro, Inc.,) was affixed to the top of the A615 for contemporaneous visible wavelength image collection. This study was carried out in accordance with the recommendations of the Cook Islands Government. The protocol was approved by the Office of the Prime Minister, Cook Islands Government.

In Rarotonga, infrared and visible wavelength images were collected either ∼2 m above the ocean surface while onboard a Cook Islands Whale Research vessel, or from shore-based positions ∼5–10 m above sea level on the island's northwest coast (**Figure 1**). In Sitka Sound, all images were recorded ∼4 m above the ocean surface while onboard a commercial whale watching cruise arranged by the Sitka Sound Science Center as part of the annual Sitka Whale Fest (e.g., **Figures 1D,E**). Despite these variable imaging heights, our entire dataset was collected at >85◦ zenith angles (i.e., <5 ◦ off horizontal) due to the range in distances at which whales were imaged. A Nikon Forestry Pro laser rangefinder was used to determine whale distances whenever possible. All measurements were made during Beaufort wind force scale numbers 2–4 and similarly ranked World Meteorological Organization (WMO) Sea State codes.

Brightness temperatures were extracted from individual thermal images using the line measurement tool available in FLIR Tools+. Two lines for temperature data extraction were drawn across each image: the first line was drawn vertically through the background scene immediately adjacent to the targeted whale feature (i.e., blow, dorsal fin, fluke, rostrum, **Figure 1**), and the second line was drawn vertically such that it passed through the maximum brightness temperature included within the targeted whale feature. Thermal benchmarks included within each image, such as the steep thermal gradient across the ocean–atmosphere boundary, were used to align the pixels included in each line's thermal profile (**Figure 2**). Once aligned, the brightness temperatures recorded by each line were subtracted from each other in order to calculate BTA-values at the individual pixel scale for each whale feature analyzed (**Figure 2**).

Because the A615's pixel pitch and focal length were known, independent measurement of whale distances by laser rangefinding allowed us to estimate blow height from thermal image pixel measurements by combining the optical lens equation,

$$
\left(\frac{1}{\frac{blow}{distance}\{m\}}\right) + \left(\frac{1}{\frac{image}{distance}\{m\}}\right) = \frac{1}{\frac{focal}{length}}\tag{1}
$$

with the magnification equation,

$$
\left(\frac{image}{height} \text{(m)}\right) = \left(\frac{image}{distance} \text{(m)}\right)
$$

$$
\left(\frac{blue}{distance} \text{(m)}\right)
$$

and the camera's pixel pitch,

$$pixel\,pitch\left(\frac{m}{pixel}\right) = \frac{image}{image} \, (m) \tag{3}$$

Equation (1) can be rearranged to,

$$\frac{1}{\frac{image}{distance} \text{(m)}} = \left(\frac{1}{\frac{focal}{length} \text{(m)}}\right) - \left(\frac{1}{\frac{blue}{distance} \text{(m)}}\right) \tag{4}$$

Equation (2) can be arranged to,

$$
\begin{split}
\langle \text{llow}\_{\text{left}} \rangle\_{\text{(m)}} &= \left( \frac{1}{\left( \begin{matrix} 1 \\ \text{diameter} \end{matrix} \right)} \right) \times \left[ \begin{pmatrix} \text{image}\_{\text{(m)}} \\ \text{height} \end{pmatrix} \right] \\
&\times \left( \begin{matrix} \text{blow} \\ \text{distance} \end{matrix} \right) \end{split}
\tag{5}
$$

and equation (3) can be rearranged to,

$$\underset{height}{image}(m) = \text{pixel pitch}\left(\frac{m}{pixel}\right) \times \underset{height}{image}(pixel). \tag{6}$$

Substituting equations (4) and (6) into equation (5) gives,

$$\begin{aligned} \mathop{\text{bigow}}\limits\_{height} \{ m \} &= \left[ \left( \frac{1}{\underset{length}{\text{(}m\text{)}}} \right) - \left( \frac{1}{\underset{distance}{\text{(}m\text{)}}} \right) \right] \\ &\times \left[ \left( \underset{pitch}{\text{(}pixel} \left( \frac{m}{pixel} \right)} \times \underset{height}{\text{(pixel)}} \right) \right] \\ &\times \left( \underset{distance}{\text{(}ldistance} \left( m \text{)} \right) \right] \end{aligned} \tag{7}$$

parentheses (i.e., 0◦–10◦C) correspond with sub-polar thermal image brightness temperatures shown in (E).

(G) images of a nostril and adjacent rostrum at ∼10 m distance in tropical waters; visible (H) and infrared (I) images of a blow, rostrum and dorsal fin at 40 m distance in tropical waters; visible (J) and infrared (K) images of a footprint at ∼30 m distance and 50 s following fluke in tropical waters. Temperature scale numbers in

which simplifies to,

$${}^{block}\_{
left}\left(m\right) = \left[\frac{\begin{pmatrix} \text{pixel}\left(\frac{m}{\text{pixel}}\right) \times \text{image}\left(\text{pixel}\right) \times \text{distance}\left(m\right) \\ \text{left}}{\text{length}}\right)\right]$$

$$-\left(\frac{\text{pixel}\left(\frac{m}{\text{pixel}}\right) \times \text{image}\left(\text{pixel}\right)}{\text{height}}\right).\tag{8}$$

Blow heights were estimated using Equation (8) every 0.04 s following blow initiation. Image pixel heights were measured using FLIR Tools+ and blow distances were measured by laser range-finding as described above.

#### RESULTS

In total, we determined BTA profiles for 174 humpback whale blows, 20 dorsal fins, 9 flukes, and 20 rostrums. An equivalent number of whale features were analyzed from each of the two study areas, with the exception of flukes, for which 6 were imaged in Alaska and only 3 were imaged in Rarotonga. Of the 87 blows analyzed in each study area, 32 Rarotonga blows and 16 Alaska blows were imaged at distances <150 m. Of these, only 10 blows from each study area were recorded in the 100–150 m range.

Average BTA profiles demonstrate that humpback whale blows, dorsal fins, flukes and rostrums appear as thermal anomalies of similar magnitude relative to adjacent ocean water (**Figure 3**). For example, 100–150 m distant blows in Rarotonga and Alaska appear as 20–30 pixel-wide thermal anomalies that are ∼3 ◦C warmer than the adjacent ocean (**Figure 3A**). Similarly, dorsal fins and flukes in both areas exhibited maximum BTA values ca. 3–4◦C (**Figures 3B,C**), whereas rostrums from both populations were ∼2–3◦C warmer than the adjacent ocean (**Figure 3D**). Ocean water temperatures were measured by perpendicular thermography and satellite observations in both study areas. These measurements indicate surface ocean water temperature was ∼24◦C in Rarotonga, and ∼8 ◦C in Sitka Sound, Alaska, at the time thermal images were recorded.

The shapes of the average dorsal fin, fluke and rostrum BTA profiles differ because these features were recorded across a large range of distances in each study area. Because the Rarotonga whales were generally imaged at closer ranges, the dorsal fin, fluke and rostrum BTA profiles are spread across a larger number of image pixels than the Sitka BTA profiles (**Figures 3B,D**). In other words, the Rarotonga whale features fill a larger portion of the 640 × 480 pixel thermal images because these images were recorded at closer distances. Despite these distance-related differences in BTA profile shape between the study areas, the maximum BTA values for humpback whale blows, dorsal fins, flukes and rostrums (indicated by arrows in **Figure 3D**) we recorded are not significantly different (p >> 0.05, two-tailed t-test, **Figure 3**).

Laser range-finding enabled quantification of the relationship between the pixel-length of individual blows and blow distance for the FLIR A615. As expected, blow pixel-lengths are larger for images recorded at closer range, and blow pixel-length decreases

RB63) recorded at a distance of 87 m using a FLIR A615 infrared camera (A), including two brightness temperature extraction control lines (Li1 and Li2 in A) drawn adjacent to the targeted blow for brightness temperature extraction along the third line (Li3 in A). (B) displays the raw brightness temperature profiles recorded by the A615 camera for all three lines shown in (A), and (C) displays the brightness temperature anomaly of the blow (Li3) relative to background brightness temperatures (Li1). Inset panel in (A) shows a portion of the same scene as recorded by a visible wavelength GoPro Hero4 camera attached to the top of the A615 thermal camera. Prominent features of both the infrared and visible wavelength images are labeled for reference.

with blow distance according to an inverse power relationship (**Figure 4**). Although blows imaged at <200 m range were easily recognizable with the A615 (**Figures 4A,B,D**), a blow imaged at ∼400 m range appeared as an 8 pixel tall ∼0.4◦ C BTA (**Figure 4E**). Higher sensitivity cooled detector thermal imaging devices and/or devices with longer focal lengths would no doubt extend the range at which whale blows might be detectable (e.g., Zitterbart et al., 2013). However, these larger systems are

Individual profiles were aligned such that the image pixel with the largest brightness temperature difference relative to seawater was assigned pixel number zero. Negative pixel numbers correspond with pixels that are skyward of the maximum brightness temperature difference pixel. Positive pixel numbers correspond with pixels that are seaward of the maximum brightness temperature difference pixel. The parabolic shape of each average profile reflects the fact that the individual datasets used to determine the average profiles shown were imaged at different distances with correspondingly different image pixel widths/lengths. For example, because the Rarotonga rostrums were imaged at closer range than the Sitka rostrums, the Rarotonga rostrums span a much larger number of pixels and include positive thermal anomalies across the nostrils that were not captured in any of the Sitka images (D).

currently much more expensive and less maneuverable than the FLIR A615 we used here.

Regardless of the device used or its imaging range, whale blow heights will also vary in response to a number of uncontrollable factors, including: wind shear, the volume of sea water in the nasal depression at exhalation, and the whale's position relative to the ocean surface at which exhalation is initiated. In an effort to partially overcome these complicating factors, we calculated blow heights 0.4 s after blow initiation, the minimum observed period for a blow to achieve its maximum height, for 32 humpback whale blows across an 18–140 m range in distances (mean = 71 ± 38 m, ± SD, **Figure 4**). The pixel height (range = 24–230 pixels, mean = 63 pixels ± 46 pixels, ± SD, **Figure 4**) of each imaged and laser ranged blow was measured using FLIR Tools+. Estimated blow heights at 0.4 s ranged between 1.0 and 3.3 m (mean = 2.2 ± 0.5 m, ± SD, n = 32). In addition to wind, water volume, and whale position, blow heights are also likely to vary with the volume of air being expelled in a specific exhalation. Although untested, focal follows incorporating thermal imaging techniques have the potential to reveal the breathing behaviors of individual whales of different size, maturity, sex and physiological condition.

Utilization of the 25 frames per second videography option enabled us to also estimate humpback whale blow velocity (**Figure 5**). All blows analyzed reached maximum blow height in <1.2 s and the maximum blow height measured was 4.7 m at 0.8 s following blow initiation equating to a 21 km h−<sup>1</sup> velocity for this blow (**Figures 5G,J,K**). Notably, some blows were unambiguously initiated while the nostrils/blowholes were still submerged. Blows of this type exhibited a relatively slow initial acceleration (e.g., **Figures 5G–I**) as the exhaled air pushed into the overlying water column. Individual blows exhibited maximum blow velocities that ranged between 40 and 55 km h−<sup>1</sup> (mean = 13–23 km h−<sup>1</sup> ± 12–18 km h−<sup>1</sup> , ±SD). Maximum blow heights ranged between 2.7 and 4.7 m and occurred 0.76–1.16 s following blow initiation. At 0.4 s following exhalation initiation, the humpback whale blows we recorded were 1.4–3.3 m tall. It is important to acknowledge that these estimates are derived

from the thermal anomalies associated with water droplets that are blasted out of the ocean's surface or nasal depression by exhaled air. Thus, the velocities we calculated must be considered minimum estimates of the true gaseous exhalation velocities achieved by humpback whales.

Our results demonstrate that humpback whale blows, dorsal fins, flukes and rostrums present as similar magnitude brightness temperature anomalies (BTA) in both tropical (Rarotonga, Cook Islands) and sub-polar (Sitka Sound, Alaska, U.S.A.) environments despite an ∼16◦C difference in ocean surface temperature between the two study areas. This occurs due to emissivity effects associated with the oblique nearhorizontal imaging angles used in the current study. Thus, absolute temperatures determined from oblique (i.e., subparallel to target) measurement angles do not represent accurate quantifications of whale blow or skin temperatures. Our results also demonstrate how to calculate blow heights and blow velocities by combining target BTA pixel size with target distance as measured by a laser range finder. Although blow acceleration varied both within and between individual blows, our results indicate that humpback whale blows have average instantaneous velocities of ∼4.6 m s−<sup>1</sup> .

FIGURE 5 | Humpback whale blow evolution through time. Visible (A) and infrared (B) images of whale blow RB34 0.16 s after blow initiation at a distance of 130 m. Visible (C)and infrared (D) images of whale blow RB48 0.36 s after blow initiation at a distance of 95 m. Visible (E) and infrared (F) images of whale blow RB65 0.48 s after blow initiation at a distance of 55 m. Visible (H) and infrared (I) images of whale blow RB66 0.64 s after blow initiation at a distance of 40 m. Visible (J) and infrared (K) images of whale blow RB67 0.80 s after blow initiation at a distance of 36 m. (G) displays blow height vs. time (i.e., velocity profiles) for the blows indicated in the legend.

#### DISCUSSION

The infrared radiation emitted by a surface is a function of both the surface's temperature and its spectral emissivity (see Nomenclature). Thermal imaging systems estimate surface temperatures by assigning emissivity values to the imaged scene. However, sea surface emissivity, the ratio of the energy radiated from the ocean's surface relative to a blackbody, further depends on the ocean's surface roughness, refractive index, and the zenith angle from which the surface is being observed (Masuda et al., 1988). Thus, quantitative analyses and accurate interpretations of thermographic datasets collected at sea depend on a large number of variables.

Of these variables, the angle from which the surface is being observed has the largest effect on emissivity and, as a consequence, thermographic temperature estimates (Masuda et al., 1988). For example, the emissivity of perfectly planar sea water at a 0◦ zenith angle (i.e., perpendicular to the sea surface) is ∼0.98 (Masuda et al., 1988). At a 60◦ zenith angle this same surface will have an emissivity of ∼0.92 and at 85◦ (i.e., 5◦ above the horizontal) the emissivity drops to ∼0.36 (Masuda et al., 1988). Using human targets included in our thermographic image dataset, we found that a decrease in surface emissivity of 0.98–0.36 resulted in a 12.2◦C increase in the human skin surface temperature reported by the camera. Similar tests on 25◦C ocean water revealed that a similar magnitude change in emissivity resulted in a 3.5◦C change in sea surface temperature at ∼100 m distance. As suggested by Cuyler et al. (1992), our findings confirm it is inappropriate to assume relatively high emissivity values (i.e., >0.95) in thermographic cetacean research when imaging is performed at high zenith angles.

Thus, the data we report suffers from extreme emissivity effects due to the fact that our thermal images were collected at sub-horizontal observation angles (i.e., zenith angles of ∼85◦–89◦ ). However, the A615 infrared camera we used includes a high sensitivity microbolometer (<0.05◦C); thus, the brightness temperature measurements we report can be considered precise but not accurate. Although the loss of thermographic accuracy due to emissivity effects associated with oblique-angle imaging is problematic for biometrical estimates of cetacean thermoregulation, it is a benefit to cetacean detection.

Brightness temperature anomalies of ∼2–4◦C, like those we report for humpback whale blows, dorsal fins, flukes and rostrums (**Figure 3**), are the consequence of rapid changes in emissivity as the whale feature emerges from the ocean's surface and immediately changes the observation point zenith angle. However, the higher BTA values we report for humpback whale nostrils (ca. 4.5◦C, **Figure 3D**) likely reflect a more accurate approximation of humpback whale skin temperatures due to the closer range at which nostrils were imaged (i.e., at lower zenith angle) and the observed ∼2 ◦C difference between nostril/blowhole temperatures and adjacent (wet) rostrums (**Figure 2D**). The potential utility of thermographic imaging of cetacean nostrils/blowholes for biometrical research purposes should be more deeply explored using aerial drones mounted with high resolution and high frame-rate thermal imaging systems.

One of the primary challenges in cetacean ecology and conservation is determining when and where whales are located. Although our results do not include accurate determinations of whale surface temperatures, they conclusively demonstrate that whale blows and emergent body parts appear as similar magnitude thermal anomalies, ca. 2–4◦C, relative to surface waters in both tropical and sub-polar environments at distance ranging between 100 and 150 m. These thermal anomalies are largely due to emissivity effects associated with thermographic imaging from sub-horizontal positions. Thus, our findings represent an important quantification of the magnitude of the thermal signal from which thermographic cetacean detection algorithms can be developed and refined.

Quantitative constraints on the magnitude, size and duration of whale-derived thermal anomalies can also be used to help restrict the number of false positives and false negatives produced by automated cetacean detection systems that use transient thermal contrast algorithms based on average brightness temperatures (e.g., Zitterbart et al., 2013). Improving automated detection systems in this way should assist applications in windy conditions or large swells, when ocean surface roughness has the potential to produce thermal anomalies of similar magnitude as whale blows due to emissivity effects (e.g., **Figure 4E**). Differentiating cetacean induced anomalies from non-cetacean induced anomalies will also benefit from quantifications of thermal anomaly shapes and their evolution through time. For example, our results demonstrate that the water spouts produced by humpback whale exhalations move at ∼4.6 m s−<sup>1</sup> and accelerate at ∼100–300 m s−<sup>2</sup> . Such biometrical measurements not only provide additional quantifications for the development of automated cetacean detection systems, but also create a platform for species-level identifications using measurable differences in blow geometry and velocity.

#### AUTHOR CONTRIBUTIONS

TH and PZ-R conceived of the study. All authors contributed to the fieldwork and infrared imaging in Rarotonga, Cook Islands. TH performed the fieldwork and infrared imaging in Sitka Sound, Alaska, U.S.A. TH and AO performed all of the thermographic image processing and analysis. TH and AO wrote the initial manuscript and all authors contributed to the revision of the manuscript.

#### ACKNOWLEDGMENTS

This research was supported by the Center for Cetacean Research and Conservation, Cook Islands Whale Research and the Frontiers Abroad programme through generous provision of research vessel access, and travel and accommodation costs. All authors thank the local fishermen, especially the great people at Akura Fishing Charters, for their kind assistance in helping us locate humpback whales offshore of Rarotonga. TH thanks the Sitka Sound Science Center for its kind invitation to present at the 20th Sitka Whale Fest: Whales Through Time and providing an opportunity to record thermal images of the beautiful whales residing in Sitka Sound, Alaska.

### REFERENCES


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

Copyright © 2017 Horton, Oline, Hauser, Khan, Laute, Stoller, Tison and Zawar-Reza. 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) or licensor 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.

### NOMENCLATURE

**Brightness temperature**–temperature measured by the thermal imaging device.

**Brightness temperature anomaly (BTA)**–difference between the brightness temperature of the targeted object and the brightness temperature of the background scene.

**Emissivity (spectral)** – the ratio of the energy radiated from a surface to the energy radiated from a blackbody at the same temperature, wavelength and environmental conditions.

**Microbolometer** – the detector in a thermal imaging device (for further details see: Ostrower, 2006).

# An Economical Custom-Built Drone for Assessing Whale Health

Vanessa Pirotta<sup>1</sup> \*, Alastair Smith<sup>2</sup> , Martin Ostrowski <sup>3</sup> , Dylan Russell <sup>3</sup> , Ian D. Jonsen<sup>1</sup> , Alana Grech<sup>4</sup> and Robert Harcourt <sup>1</sup>

*<sup>1</sup> Marine Predator Research Group, Department of Biological Sciences, Macquarie University, Sydney, NSW, Australia, <sup>2</sup> Heliguy Pty. Ltd., Sydney, NSW, Australia, <sup>3</sup> Macquarie Marine Research Centre, Chemistry and Biomolecular Sciences, Macquarie University, Sydney, NSW, Australia, <sup>4</sup> ARC Centre of Excellence for Coral Reef Studies, James Cook University, Townsville, QLD, Australia*

Drones or Unmanned Aerial Vehicles (UAVs) have huge potential to improve the safety and efficiency of sample collection from wild animals under logistically challenging circumstances. Here we present a method for surveying population health that uses UAVs to sample respiratory vapor, 'whale blow,' exhaled by free-swimming humpback whales (*Megaptera novaeangliae*), and coupled this with amplification and sequencing of respiratory tract microbiota. We developed a low-cost multirotor UAV incorporating a sterile petri dish with a remotely operated 'blow' to sample whale blow with minimal disturbance to the whales. This design addressed several sampling challenges: accessibility; safety; cost, and critically, minimized the collection of atmospheric and seawater microbiota and other potential sources of sample contamination. We collected 59 samples of blow from northward migrating humpback whales off Sydney, Australia and used high throughput sequencing of bacterial ribosomal gene markers to identify putative respiratory tract microbiota. Model-based comparisons with seawater and dronecaptured air demonstrated that our system minimized external sources of contamination and successfully captured sufficient material to identify whale blow-specific microbial taxa. Whale-specific taxa included species and genera previously associated with the respiratory tracts or oral cavities of mammals (e.g., *Pseudomonas*, *Clostridia*, *Cardiobacterium*), as well as species previously isolated from dolphin or killer whale blowholes (*Corynebacteria*, others). Many examples of exogenous marine species were identified, including *Tenacibaculum* and *Psychrobacter* spp. that have been associated with the skin microbiota of marine mammals and fish and may include pathogens. This information provides a baseline of respiratory tract microbiota profiles of contemporary whale health. Customized UAVs are a promising new tool for marine megafauna research and may have broad application in cost-effective monitoring and management of whale populations worldwide.

Keywords: UAV, UAS, drone, blow, humpback whale, microbiota, technology, conservation

#### INTRODUCTION

Conservation biology is entering a new era of innovation, with unprecedented growth across a range of techniques, from genetics and genomics to telemetry and remote sensing (Allendorf et al., 2010; Hussey et al., 2015). Rapid advances in the technology underpinning Unmanned Aerial Vehicles (UAVs also known as Unmanned Aircraft Systems or drones), are driving new and

#### Edited by:

*Peter H. Dutton, National Oceanic and Atmospheric Administration (NOAA), United States*

#### Reviewed by:

*Nuno Queiroz, University of Porto, Portugal Gail Schofield, Deakin University, Australia*

\*Correspondence: *Vanessa Pirotta vanessa.pirotta@hdr.mq.edu.au*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *01 October 2017* Accepted: *12 December 2017* Published: *21 December 2017*

#### Citation:

*Pirotta V, Smith A, Ostrowski M, Russell D, Jonsen ID, Grech A and Harcourt R (2017) An Economical Custom-Built Drone for Assessing Whale Health. Front. Mar. Sci. 4:425. doi: 10.3389/fmars.2017.00425*

**82**

innovative environmental applications (Koh and Wich, 2012; Anderson and Gaston, 2013; Christie et al., 2016; Smith et al., 2016; Duffy et al., 2017). The application of UAVs in conservation science makes it possible to collect information from dangerous and inaccessible environments and answer research questions that were previously limited to the hypothetical (Harvey et al., 2016). UAVs also provide an alternative, safer, quieter and often cost-effective option for monitoring fauna and flora, from individuals and populations to entire ecosystems, and in so doing are replacing expensive manned systems such as helicopters and fixed-wing aircraft (Christiansen et al., 2016; Christie et al., 2016). UAV applications in wildlife research now encompass almost all environments, from arid deserts, through rainforests, oceans to polar regions (Linchant et al., 2013, 2015; Durban et al., 2015; Goebel et al., 2015; Duffy et al., 2017).

UAVs are transforming the way scientists monitor and conserve wildlife (Gonzalez et al., 2016). In the terrestrial world, UAVs have been used for a wide variety of conservation applications (van Gemert et al., 2014; Gonzalez et al., 2016). Some examples include, counting elephants (Loxodonta africana) (Linchant et al., 2013; Vermeulen et al., 2013), UAV surveillance (anti-poaching tools) for elephants and rhinoceros (Diceros bicornis and Ceratotherium simum) (Marks, 2014; Mulero-Pázmány et al., 2014; Hahn et al., 2017), locating chimpanzee nests (Pan troglodytes) (van Andel et al., 2015) and mapping Sumatran orangutan (Pongo abelii) habitat, distribution and density (Wich et al., 2015; Szantoi et al., 2017). UAV applications now extend to the polar regions where they have been used to monitor and estimate abundance of penguin populations (gentoo, Pygoscelis papua, and chinstrap, Pygoscelis antarctica) and estimate size and condition of leopard seals (Hydrurga leptonyx) (Goebel et al., 2015; Ratcliffe et al., 2015). In the marine environment, UAVs are revolutionizing the way marine species can be studied due to their small size, apparent minimal disturbance of wildlife and improved safety for both operators and animals (Nowacek et al., 2016; Fiori et al., 2017). UAVs have been utilized for a wide variety of applications including aerial surveys, monitoring, habitat use, abundance estimates, photogrammetry and biological sampling e.g., whale "blow" (Hogg et al., 2009; Acevedo-Whitehouse et al., 2010; Hodgson et al., 2013; Durban et al., 2015; Pomeroy et al., 2016; Schofield et al., 2017).

There are widespread concerns about the health of marine mammal populations in the face of global anthropogenic stressors (Gulland and Hall, 2007). Yet health assessments typically involves collecting samples from stranded animals, which are often biased as these animals are most likely to be health-compromised (Geraci and Lounsbury, 2005). Sampling exhaled breath or 'blow' from wild whales may therefore provide a more representative assessment of the health status of individuals because samples can be randomly taken from the population. From a single sample of whale blow, scientists may be able to collect respiratory bacteria, lipids, proteins, DNA and hormones (Hogg et al., 2005, 2009; Schroeder et al., 2009; Acevedo-Whitehouse et al., 2010; Hunt et al., 2013, 2014; Thompson et al., 2014; Burgess et al., 2016; De Mello and De Oliveira, 2016; Raverty et al., 2017). This information is important for whale conservation, as it can be collected over time to help monitor the recovery of whale populations postwhaling. Early approaches to sampling whale blow involved passing a cotton gauze or nylon stocking on the end of a carbon fiber pole through the blow when the animal surfaced (Hogg et al., 2009; Hunt et al., 2014). Recent advancements on this method have seen the use of a pole with a number of petri dishes with lids to sample wild killer whales (Raverty et al., 2017). However, this method requires extremely close vessel approaches to whales (Hogg et al., 2009). Given the large size, mass and power of whales, this approach involves high risk to both researchers and to the whale itself. Even under ideal circumstances this method is likely to disturb the animal, potentially compromising the validity of some of the measures such as stress hormones which elevate rapidly (Harcourt et al., 2010). Accordingly, alternative approaches have long been sought. Acevedo-Whitehouse et al. (2010) deployed a single-rotor UAV (a remote-controlled helicopter) to sample whale blow. Their study demonstrated the feasibility of the approach but loss of samples from the UAV as it careers through the sea air proved a potential issue as did contamination from airborne particulate not expired by the whale.

Here we describe a purpose-built UAV designed to sample whale blow in the field with minimal contamination. Our goal was to provide a snapshot of whale health. We specifically targeted northward migrating humpback whales (Megaptera novaeangliae) off the East coast of Sydney, Australia for the collection of baseline microbiota information. The UAV used in our study has a unique combination of features that represent a significant advance over existing UAVs. It is fast, highly maneuverable, durable, waterproof, low-cost (< \$USD 1000) and provides flexible payload mounting options. The UAV is scaled to the sampling gear (in this case a 100 mm petri dish), which is held in a mechanism that allows the dish to be opened/closed during flight–minimizing sample contamination or loss.

## MATERIALS AND METHODS

#### Study Site and Species

All flights were conducted offshore Sydney, Australia (**Figure 1**). Each year from May to November, migratory Group V (Stock E1) humpback whales migrate past Sydney, as they swim from high latitude feeding areas in Antarctica to low latitude breeding waters off Queensland (Chittleborough, 1965). All sampling took place in coastal waters <3 nautical miles from Sydney between 30 May 2017 and 27 June 2017.

#### UAV Design

The UAV is a 4-motor electric multirotor (quadcopter) 500 mm across (motor to motor, diagonally) (**Figure 2A**). It has a relatively high power to weight ratio making it fast, maneuverable, resistant to strong wind gusts and relatively quiet while hovering. It carries the bare minimum of hardware and is operated in 'manual mode' (no GPS or autolevelling assistance) with a heavy reliance of the onboard video feed for control, navigation and sampling operations. The airframe structure of the UAV is a 'sandwich' style construction cut from

carbon fiber plate, with a top shell molded from impact-resistant polycarbonate. This seals against the airframe to create a waterproof compartment which houses the power distribution, flight control, motor control, radio control transceiver, and video transmitter components. The float booms/legs were cut from expanded polypropylene (EPP)—a closed-cell foam, chosen for high strength, resistance to bending loads and excellent water resistance. A clear acrylic tube at the front of the aircraft houses a forward facing, tilting camera that provides a realtime position reference to the pilot (First Person View). The resulting composite structure is light, stiff, strong and waterproof. Buoyancy is provided by the two watertight compartments and EPP foam floats under the arms. In the event of a crash or forced landing over water, the UAV floats in an upright position so it can be recovered or take off again. Two reinforced mounting areas on the top shell accept payloads of around 100 g. For this configuration, the blow-sampling apparatus was mounted at the front. This is a hinged frame which opens to 180 degrees and holds a 100 mm diameter petri dish with suction cups. A servo motor opens and closes the dish remotely, during flight. Airflow testing using smoke indicated the best position for the sampling dish relative to the propellers. A forward-looking waterproof video camera (GoPro <sup>R</sup> Hero SessionTM) is positioned at the rear and logs video to an internal memory card. The dish is in the frame of the recorded video, so the footage can be used to confirm the source of the sampled material.

#### Sampling Method

This study was approved by the Macquarie University Animal Ethics Committee, and carried out in accordance with the Animal Research Authority (2016/010). This research was permitted by New South Wales National Parks and Wildlife Services (NSWNPWS) to fly UAVs over whales in New South Wales coastal waters (permit number SL101743). To adhere to Australian legislative requirements, the UAVs (including backup UAV) were registered with the Civil Aviation Safety Authority (CASA) and operated by a CASA certified operator (Heliguy Pty. Ltd.). All flights were conducted in good weather (no rain, Beaufort < 3), from small research vessels, where the UAV was launched and landed on a launch pad at the bow or stern of the boat. A closed, sterile petri dish with nutrient agar covering the base of the petri dish was secured using eight suction cups affixed on the UAV before each flight.

Members of the team scanned the area for humpback whales. Once an individual was selected, the vessel was driven maintaining a constant speed and distance from the whale (>200 m). Once the respiratory rhythm of an individual was determined (downtime length in minutes), the UAV was launched to coincide with the individual surfacing. The UAV pilot was directed by spotters on the vessel and positioned the UAV with the aid of the live feed from the forward-facing camera. To minimize sample contamination, the petri dish remained closed until just before the whale surfaced, when the dish remotely opened as the UAV accelerated toward and through the densest part of the whale blow, collecting the maximum amount of sample in the dish and lid (**Figures 2B,C** and Supplementary Video 1). The petri dish was immediately closed and the UAV was returned to the vessel. The petri dish containing the sample was removed from the UAV and Parafilm <sup>R</sup> was wrapped around the closed petri dish to secure the sample. All samples were

temporarily stored in a cooler box on ice until further processing in the laboratory at the end of each day.

Attempts were made to sample a different whale each flight. Individuals within a pod were chosen based upon unique markings (e.g., white flanks/patterns/scarring/barnacle arrangements). To ensure the same individual was not sampled twice, a live video feed was used to target individuals. Cross contamination among whales was avoided by not triggering the opening of the flip lid until only the targeted whale respired. Footage collected from the GoPro <sup>R</sup> throughout each flight was used to validate sample collection and eliminate repeated sampling of the same individuals by post-hoc identification. The behavioral response of whales was recorded for each pass using by scoring system of one to three (one: 'no response,' two: 'minor response' minor surface activity such as logging, spy hopping and three: 'severe/elevated Response' e.g., breaching, peduncle throw or chin slap).

#### Air and Seawater Samples

To enable direct comparison of UAV-captured air and whale blow samples with bacteria inhabiting the adjacent seawater, the data were combined with 16S sequence libraries prepared from 26 surface seawater samples. This represents a complete annual cycle, collected from the National Time Series Station known as Port Hacking 100 (PH100). All UAV-captured samples were collected within 20 km of PH100.

### Laboratory Processing of Samples

Initial processing of samples occurred in two stages. First, in an Ultra Violet-sanitized class II biosafety hood, the top of the petri dish lid (non-agar) side was swabbed using a dry sterile cotton tip and then placed in a sterile 1.5 ml tube and stored in the freezer at −30◦C. Secondly, the petri dish (both the lid and nutrient base) was placed in an incubator at 37◦C after the lid was swabbed, simulating average mammalian body temperature 36–37◦C (Whittow, 1987; Cuyler et al., 1992). Plates were observed daily for colony growth. If growth occurred, colonies were counted and a representative number of colonies were picked from each plate, resuspended in 100 µl of sterile water, vortexed for 10 s and immediately frozen at −30◦C until further processing. Plates were then stored in the fridge for future reference if needed.

#### Bacterial DNA Extraction

DNA extractions were conducted using the Quick-DNATM Fungal/Bacterial Miniprep kit (Zymo Research, Irvine, California, USA) with minor modifications to the manufacturer's instructions. Each swab was transferred to a tube containing 1.2 g of ZR BashingBeadsTM (equivalent to ∼half of the portion supplied for each extraction). The original storage tube was rinsed with lysis solution (750 µl) to ensure the complete transfer of material into the extraction tube. The swab was then beadbeaten on a Vortex-Genie <sup>R</sup> 2 (Mo Bio Laboratories/QIAGEN, California, USA) for 20 min at room temperature. All other steps were followed according to the manufacturer's instructions, with the exception that two successive final elutions were carried out, each with 20 µl of sterile DNA elution buffer.

#### Amplification and Sequencing

Amplicons targeting the bacterial 16S rRNA gene (27F−519R; Lane et al., 1985; Lane, 1991) were generated and sequenced for each sample at the Ramaciotti Centre for Genomics (UNSW Sydney, Australia) using 250 bp paired end illumina sequencing according to established protocols (http://www. bioplatforms.com/wp-content/uploads/base\_illumina\_16s\_ amplicon\_methods.pdf).

Amplicons generated from drone-captured air and whale blow were combined with 27F−519R sequences generated from 26 surface (2 m and 10 m depth) seawater samples collected over a complete annual cycle from the nearby National Reference Station (PH100) time series (Dec 2014–Mar 2016). Monthly microbial sampling has been conducted at the Port Hacking100 reference station since 2009 (Seymour et al., 2012). All UAVcaptured whale and air samples were collected within 20 km upstream of this reference station, within 1–3 km from shore. We reasoned that this dataset, which was sampled and sequenced using standardized protocols at the same sequencing center, would provide a comprehensive and unbiased assessment of bacterial species characteristic of seawater in this region, which could be excluded as potential contaminants from the whale blow samples. Whale, air and seawater samples analyzed in this study are detailed in Supplementary Tables 1, 2.

Sequence Operational Taxonomic Units (OTUs) tables were prepared after (Bissett et al., 2016). Briefly, paired-end reads were filtered using Trimmomatic (ILLUMINACLIP: NexteraPE-PE.fa:2:30:10 SLIDINGWINDOW:4:15 MINLEN:76) (Bolger et al., 2014) then merged using PEAR (Zhang et al., 2014). The combined amplicon data were clustered into OTUs at 97% sequence similarity using an open reference OTU picking pipeline in USEARCH 64 bit v8.1.1756 (Edgar, 2010), which included de novo chimera detection. Clusters with < 4 sequences were removed, and reads were mapped to representative OTU sequences using USEARCH (97% ID) to calculate read abundances. From an initial pool of 10.5 million paired-end reads, a total of 7.62 million filtered, merged sequences, with chimeras removed, were added to the OTU table. OTU tables were sub-sampled to a constant sampling depth of 10,000 sequences using rarefy in vegan (Oksanen, 2017). All subsequent analyses were conducted on sub-sampled OTU tables. Sequences generated over the course of this project are deposited in the European Nucleotide Archive under project PRJEB23634. All seawater sequence data are deposited in the NCBI Sequence Read Archive PRJNA385736.

#### Data Analyses

Hierarchal clusters of OTU abundance profiles generated from seawater, drone-captured air and whale blow were compared using the simprof test following square-root transformation and conversion to a Bray-Curtis dissimilatory matrix in the r package clustsig (Whitaker and Christman, 2014). Data from samples that were near misses, which would reflect a mixture of air and whale blow microbiota, were set aside from the subsequent statistical analyses. The community structure dissimilarity between samples was observed with non-metric multidimensional scaling. Significant differences in communities sampled in seawater, UAV-captured air or whale blow samples were defined using generalized linear models within mvabund (Wang et al., 2012). Briefly, a negative binomial model was fit to the OTU abundance data and the sample grouping was analyzed using Analysis of Variance (ANOVA). OTUs that were significantly overrepresented in seawater, drone-captured air or specific for whale blow samples were defined using ANOVA with the 'p.uni="adjusted"' option. OTUs were classified against the Silva 123 release database (Quast et al., 2013) using mothur "classify.seqs" with default parameters (v1.36.1, Schloss et al., 2009).

#### Identifying Bacteria Isolated from Agar Plates

Bacterial 16S rRNA genes were directly amplified from cell suspensions obtained from colony picks using conserved primers 27F and 519R (Lane et al., 1985; Lane, 1991). PCR amplifications consisted of 1.0 µl of template and cycle specific for 16S consisted of 95◦C for 10 min, 94◦C for 30 s, 55◦C for 10 s, 72◦C for 45 s and 72◦C for 10 min, and Taq DNA Polymerase (Qiagen). Amplified DNA was prepared for Sanger sequencing using Agencourt <sup>R</sup> AMPure <sup>R</sup> XP beads (Beckman Coulter). Sequences were trimmed to q20, and classified against the Silva Database (version 123).

### RESULTS

A total of 74 flights were conducted over 4 days of sampling. Each pod was considered independent as all whales were on their annual northern migration (Pirotta et al., 2016). Overall, 59 successful samples were collected from at least 48 different whales (11 whales were sampled but not identified via video due to occasional failure of the GoPro <sup>R</sup> camera e.g., low battery or maximum storage capacity reached). Sample volume varied between 50 and 150 µl of exhaled breath. The average opening time of the flip lid was 4 s (min 2 s, max 6 s). The UAV had a maximum flight time (battery time) of 15 min and sampling attempts on average were 4 min 28 s long (range: 27 s to 7 min). The majority of flight time was used to search for the whale's next surfacing position. The time that the UAV was in close proximity to a whale (UAV approximately within 5 m horizontal distance) varied but was on average 53 s (range: 2 s to 2.36 min or 141 s). The most number of samples collected in 1 day was 38. In all cases, there was no behavioral response to the drone (level 1, n = 48). Twice there were strong social interactions that occurred prior to the drone approaching the whales (one tail slap, one breach) but sampling was continued on the group in each case and samples successfully collected.

#### Next Generation Sequencing Results

A total of 7.62 million filtered bacterial 16S ribosomal gene sequences were produced from 59 UAV-captured whale blow and six air samples. These were combined with 0.91 m sequences generated from 26 seawater samples to generate bacterial OTU abundance profiles. Distance-based clustering of blow, air or seawater bacterial community profiles defined at least three significant clusters (simprof, P < 0.05), encompassing one group exclusively composed of seawater, one group exclusively composed of whale blow samples and a third group which clustered the six air samples along with 11 whale-blow samples (**Figure 3A**). Whale blow samples in this group may correspond to UAV sorties that missed, or narrowly missed, capturing whale blow material and were highly correlated with low capture scores based on a visual score of the amount of whale material recovered (Supplementary Table 1).

Bacterial OTUs correlated with seawater, whale blow or air samples were identified using Analysis of Variance (ANOVA) based on generalized linear models fit to the data (Wang et al., 2012). OTU diversity and abundance profiles for air and whale blow were significantly different (p < 0.05) from each other and bear little similarity with communities characteristic of the adjacent seawater. At the Class level whale blow bacteria were dominated by Gammaprotobacteria, Flavobacteriia, Clostridia and Fusobacteria, in contrast to seawater communities, where species composition reflected values typical for sub-tropical waters of the Tasman Sea, i.e., ∼60% Alphaproteobacteria, 15% Cyanobacteria and smaller proportions of Gammaproteobacteria and Flavobacteriia (**Figure 3B**; Seymour et al., 2012).

Overall, whale blow samples displayed the greatest OTU diversity, followed by seawater and air (Supplementary Figure 1). Model-based multivariate analyses identified 198 OTUs that were seawater-specific and 35 OTUs that were significantly correlated with air samples (ANOVA, P < 0.1; Supplementary Tables 3, 4). Successfully collected whale blow samples contained a small proportion seawater and air-specific OTUs, contributing on average 15.7(±10.8)% and 11.5(±4.4)%, respectively, of total sequences. The proportion of air-specific and seawater OTUs in near-miss samples was significantly higher (41.0% and 24.1%, respectively). Subtraction of seawater and air specific OTUs from the total enabled us to define 129 OTUs that were highly specific to whale samples (ANOVA, P < 0.05, **Figure 4**, Supplementary Table 5). Abundant bacterial species identified as whale-blow-specific include multiple OTUs belonging to the genera Cardiobacteriaceae and species Tenacibaculum, followed by OTUs related to Pseudomonas sp. Strain wp33, Leptotrichia sp. and Corynebacteria spp. While these analyses identified which OTUs were highly specific for whale, air and seawater, an addition set of whale-related OTUs could be identified in the remaining non-significant OTUs. We used the following criteria: present in greater than five whales and >100 sequences, to add an additional 145 OTUs that were highly specific to whales but found only in a small proportion of the sampled whale population (5–17 individuals, out of a total of 57) (Supplementary Table 6). Many of the OTUs in this group are closely related to whale-specific OTUs at the genus and species levels, e.g., Cardiobacteriaceae, Tenacibaculum, and Fusibacter strains. However, potential respiratory pathogens were also detected, such as Balneatrix (Gammaproteobacteria), and a range of Gram positive Clostridia and Bacilli, such as Staphylococcus and Streptococcus. In the context of monitoring whale respiratory health, potential pathogens may be present in a subset of the population only. OTUs in this whale-associated group were present in low abundance, and on average constituted 13(±5.7)% of the total sequences detected in each whale sample.

#### Comparison with Culture-Dependent Identification of Whale Blow Microbiota

Bacterial growth was observed on 48 UAV-mounted agar plates exposed to whale blow. Unexposed control plates displayed no bacterial growth. Sequencing of rRNA genes amplified from single colonies identified 18 different bacteria taxa isolated from 19 different whales (Supplementary Table 7). Overall, the most common bacteria identified at the phylum level included Proteobacteria (n = 7), Firmicutes (n = 7) and Actinobacteria (n = 4). Two samples were identified to the family level, Brucellaceae (n = 1) and Microbacteriaceae (n = 1). At the genus level, Micrococcus (n = 3), Acidovorax (n = 3), Bacillus (n = 3), Enterobacteriaceae (n = 2), Paenibacillus (n = 2), Streptococcus (n = 2), and Staphylococcus (n = 2) were most common. Seven whales had more than one bacterium identified. Staphylococcus was identified in both an individual sampled via our UAV.

### DISCUSSION

UAVs are rapidly transforming the way scientists collect information on their study species (Christie et al., 2016; Lowman and Voirin, 2016; Nowacek et al., 2016; Duffy et al., 2017). In whale research, UAVs have enabled sampling methods to be refined and have eliminated the need for close vessel approaches. To our knowledge, this study is the first to successfully demonstrate the use of a purpose-built UAV designed to sample humpback whale blow in Southern Hemisphere waters. The minimal behavioral disturbance observed suggests this method is an excellent, low-impact alternative to pole sampling methods for large, migrating whales. Humpback whales may have been aware of the UAV and did not react or, mostly likely, were not even aware of the UAV's presence. Underwater noise generated from the UAV was likely to be very low level at the heights flown (<10 m), as it is smaller, lighter and has a lower disc loading than comparable off-the-shelf UAVs shown to transmit minimal noise transmission underwater (e.g., SwellPro Splashdrone and the DJI Inspire 1 Pro) (Christiansen et al., 2016). The combination of the waterproof design and the remotely operated flip lid petri dish designed to minimize airborne contamination, is a significant improvement over existing UAV types.

Our results demonstrate that whale blow can be effectively sampled while minimizing species associated with likely sources of contamination, i.e., air and seawater, to define microbes specifically associated with whales. Amplification of DNA extracted from UAV-captured air highlights the sensitivity of PCR-based approaches for detecting microbiota, even from low amounts of extracted DNA, while also demonstrating the sensitivity of this approach to contamination from external sources. The development of a flip-lid sampling system using sterile petri-dishes enabled us to effectively reduce contamination

from typical seawater bacteria, which may exist in aerosols above the sea surface. While the presence of abundant seawater species (Alphaproteobacteria SAR11 and cyanobacteria) in air and whale blow samples is not surprising, the source of some major species detected in air samples is less clear. Some of the most abundant species detected in air samples, Propionobacteria, Arthrobacter, and Staphylococcus, are common commensal organisms of mammalian (human) skin and nasal cavities (Human Microbiome Project Consortium, 2012; Prussin and Marr, 2015). A potential source of some non-marine material may have been contamination during the DNA extraction or amplification procedure, especially when the amount of captured material was low (i.e., for air or near-miss samples). In the context of developing indicators of whale health the presence or absence of species that are common in humans should be interpreted cautiously. Nevertheless, in the UAV-sampled blow where a sufficient amount of material was collected, our analyses indicate that ∼70% of the total sequences were specific to whales, a group of whale associated sequences accounted for a further ∼12% and the remainder could be confidently identified as seawater- or air-specific.

To our knowledge this is the first study to utilize a long-term seawater dataset to identify and subtract seawater bacteria from community profiles of field-captured mammalian samples. The seawater data provided a comprehensive, temporal assessment of the composition of microbial communities present in sea water off Sydney. Critically, a much larger quantity of seawater was collected (2 L) and analyzed in comparison to the whale samples. This method minimized the impact of external sources of contamination and allowed for the greater coverage of the seawater community diversity. We used this resource to filter out all sequences characteristic of seawater to produce a whale blow dataset that could be used as a diagnostic for whale health. The distinct differences observed between statistically-defined bacteria in whale, sea water and air samples indicates that this method was effective for collecting whale microbiota with minimal contamination.

The successful collection of bacterial DNA in this study provides baseline information of microbiota found in migrating humpback whale blow. Due to the infancy of sampling whale breath as an assessment of whale health (Acevedo-Whitehouse et al., 2010; Hunt et al., 2013), it is not clear as to the type of microflora/bacteria species that are considered 'normal' for northward migrating humpback whales off Sydney. Despite this, there are similarities in our collection of bacterial genera from the few studies that have collected blow for the assessment of microbiota (Acevedo-Whitehouse et al., 2010; Denisenko et al., 2012; Hunt et al., 2013). For example, Streptococcus and Staphylococcus genera were detected in our samples and have been detected in the blow of blue whales (Balaenoptera musculus), gray whales (Eschrichtius robustus) and Southern resident killer whales (Acevedo-Whitehouse et al., 2010; Denisenko et al., 2012; Hunt et al., 2013; Raverty et al., 2017). Bacteria from the Streptococcus genus is common in mucous membranes of animals (and humans) and is known to be found in the upper respiratory tract (Krzy´sciak et al., 2013). Streptococcus bacteria has previously been responsible for pneumonia causing death in cetaceans (Acevedo-Whitehouse et al., 2010). Bacillus sp. was also identified via blow collection from western North Pacific gray whales and Southern resident killer whales (Denisenko et al., 2012; Hunt et al., 2013; Raverty et al., 2017).

Next generation sequencing identified Cardiobacteriaceae (family) and Tenacibaculum (genus) to be the most abundant bacterial rRNA genes in whale blow. Cardiobacteriaceae has previously been isolated as a dominant taxa in the respiratory system of "healthy" captive bottlenose dolphins (Tursiops aduncus and, T. truncates) and free-ranging species (T. truncates)

FIGURE 4 | Relative abundance of bacterial taxa identified in seawater, UAV captured air and whale blow. OTUs with abundance <9 across the entire dataset were omitted for clarity. Relative abundances are presented for each group (i.e., seawater, air plus "near-miss" samples and whales, as well as for each sample. Taxa names correspond to the highest taxonomic level identification, full taxonomies are present in Supplementary Tables 3–6) only the top taxa by abundance are shown in the legend.

(Johnson et al., 2009; Lima et al., 2012). These findings may indicate that these genes are part of the normal microflora of dolphins, whilst presence in whales until now was unknown. Cardiobacteriaceae are abundant on humpback whale skin (Gammaproteobacteria genus), as is Tenacibaculum (Apprill et al., 2011, 2014). It may be possible that bacteria found on whale skin also occur within the respiratory tract or epithelial cells. Tenacibaculum has been associated with the microbiome of other marine species such as southern bluefin tuna (Thunnus maccoyii castelnau) (Valdenegro-Vega et al., 2013), while Psychrobacter is part of the thresher shark and rainbow trout skin microbiome (Lowrey et al., 2015; Doane et al., 2017).

The collection of bacterial microbiota is as an indicator of cetacean health is growing (Hogg et al., 2009; Schroeder et al., 2009; Acevedo-Whitehouse et al., 2010; Lima et al., 2012; Hunt et al., 2013; Nelson et al., 2015; Raverty et al., 2017). We were able to sample a number of individuals from a single population over a very short time frame. The use of the waterproof GoPro <sup>R</sup> camera made identification of different individuals reliable and therefore reduced repeated sampling. Our remotely operated "flip dish" design proved effective at reducing possible contamination from the pilot/research team (e.g., breath, touch, clothing) and vessel vapor/fumes. The placement of Parafilm <sup>R</sup> around the dish after sampling ensured that the sample remained unexposed until back in the laboratory for processing. Recently published work by Burgess et al. (2016) found polystyrene dishes (petri dish) to be the most effective surface for sampling whale blow in comparison to other sampling materials like veil nylon and nitex nylon mesh. In addition, the use of ice chilling of our samples for temporary storage was also consistent with Burgess et al. (2016), which found storage in cooler box with ice packs was appropriate for preserving samples (at least for hormones) for daylong fieldwork at sea (<6 h). Our samples only contained a fine mist [we estimated between 50 and 150 µL per sample, similar to amounts collected by Hogg et al. (2009)], and so we were unable to directly pipette samples but we found that swabbing the non-agar lid of the petri dishes to be effective. Variability in blow sample volumes appear to be a common issue (Hogg et al., 2009; Acevedo-Whitehouse et al., 2010) and therefore the need for repeated sampling is recommended. Sample success increased with effort/experience and we recommend effort be made early in any study to improve pilot skill, sample collection, quality and quantity.

While overall highly successful, UAVs still require a high level of skill and effort. Predicting when the whale is about to surface, positioning the UAV and opening the petri dish in time remains challenging. This may be complicated when a whale comes to the surface to breath but does not respire forcefully. When this happens, the plate is exposed to the air and so the UAV must return to the boat so the petri dish can be exchanged, our miss/near-miss rate was 11/59 = 20%. Second, not using an off-the-shelf product requires a high level of UAV competence both to fly and to fix problems as they arise. Third, the flight time for this UAV is 15 min, restricting the number of opportunities for sampling before the UAV must return to the vessel in order to replace the battery. Flight time will increase as battery technology progresses (Nowacek et al., 2016).

Our dataset details the diversity and abundance of the microbiota found in a migrating whale population which provides the baseline to identify pathogenic species. Ultimately, the isolation of pathogens from healthy or diseased animals will be an important step toward understanding the causes of disease and the factors that contribute to virulence. Culturedependent techniques remain a viable option for the surveillance of pathogens in populations. In this study, nutrient agar was an effective way of culturing a subset of whale blow microbiota, including species commonly associated with respiratory disease in mammals. The use of both sides of the petri dish effectively doubled the chance of obtaining bacterial samples. While next generation sequencing has the capacity to probe the diversity of whale blow microbiota, at present, the isolation and identification bacteria from agar plates can be achieved within 3–5 days, compared to a practical timeframe of weeks for illumina sequencing. Selective media could be used to target potential pathogens in conjunction with opportunistic sampling of diseased or distressed animals.

### CONCLUSIONS

Our purpose-built UAV proved highly successful in sampling whale blow for microbial community analysis. It is cost-effective, has low risk of contamination and greatly reduces disturbance of whales. Future applications include other free-ranging whale species (e.g., southern right whales, Eubalaena australis), as well as sampling smaller cetaceans (e.g., dolphins). Our UAV is useful addition to the conservation scientist's tool box, enabling collection of health information and therefore the ability to monitor changes in individual health as populations recover and to provide an early warning system for potential future changes.

### AUTHOR CONTRIBUTIONS

Paper conception: VP, AS, RH, MO, IJ, and AG. Experiment design: VP, AS, RH, and IJ. Field work: VP, AS, and RH. Laboratory work: VP, DR, and MO. Analysis and interpretation of data: VP, MO, RH, IJ, AG, and DR. Wrote paper: VP, MO, RH, IJ, AG, DR, and AS.

### FUNDING

This project was supported by Macquarie University.

#### ACKNOWLEDGMENTS

This research was conducted under the Macquarie University Animal Ethics Committee Animal Research Authority 2016/010 and the NSW Office of Environment and Heritage Scientific Research permit SL101743. VPirotta was supported by an Australian Postgraduate Award Research scholarship. Molecular work was supported by the Australian Research Council grant DP15102326. Thank you to Guy Alexander from Heliguy for providing the legal requirements for the commercial UAV operations and workshop resources for building and testing the UAVs. We would like to thank Roads and Maritime Services, NSW and Dean Cropp from Barefoot Charters for their support in this research project. We appreciate their time and use of vessels for sampling. We thank Oliver Masens, Jemma Geoghegan, Adam Wilkins, Gemma Carroll and Ben Pitcher for their assistance in the field. We thank the Cape Solander Whale Migration Study, in particularly Wayne Reynolds, Sue Rennie Wright and Mark McGeachie. Thank you Duan March from Dolphin Marine Magic and Shona Lorigan from ORRCA for their collection and transport of blow sample from the stranded humpback whale in Sawtell beach, New South Wales. We thank Patrick da Roza for assistance in the laboratory. Thank you to Sally Browning and Airpig Productions for their donation of UAV batteries and Finn Lewis for providing UAV components.

#### REFERENCES


Monthly sampling at the Port Hacking Reference Station was supported by the Integrated Marine Observing System (IMOS). We would like to acknowledge the contribution of the Marine Microbes Project consortium (https://data.bioplatforms.com/ organization/pages/bpa-marine-microbes/consortium) in the generation of data used in this publication. The Marine Microbes Project is supported by funding from Bioplatforms Australia. Bioplatforms and IMOS are supported through the Australian Government National Collaborative Research Infrastructure Strategy (NCRIS).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2017.00425/full#supplementary-material

of estimating the health status of a population," in 26th European Cetacean Society Conference (Galway).


liquid chromatography–mass spectrometry. J. Chromatogr. B 814, 339–346. doi: 10.1016/j.jchromb.2004.10.058


**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 © 2017 Pirotta, Smith, Ostrowski, Russell, Jonsen, Grech and Harcourt. 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) or licensor 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.

# Integrating Archival Tag Data and a High-Resolution Oceanographic Model to Estimate Basking Shark (Cetorhinus maximus) Movements in the Western Atlantic

#### Camrin D. Braun1,2 \*, Gregory B. Skomal <sup>3</sup> and Simon R. Thorrold<sup>2</sup>

*<sup>1</sup> Biological Oceanography, Massachusetts Institute of Technology-Woods Hole Oceanographic Institution Joint Program in Oceanography/Applied Ocean Science and Engineering, Cambridge, MA, United States, <sup>2</sup> Biology Department, Woods Hole Oceanographic Institution, Woods Hole, MA, United States, <sup>3</sup> Massachusetts Division of Marine Fisheries, New Bedford, MA, United States*

Basking shark (*Cetorhinus maximus*) populations are considered "vulnerable" globally and "endangered" in the northeast Atlantic by the International Union for the Conservation of Nature (IUCN). Much of our knowledge of this species comes from surface observations in coastal waters, yet recent evidence suggests the majority of their lives may be spent in the deep ocean. Depth preferences of basking sharks have significantly limited movement studies that used pop-up satellite archival transmitting (PSAT) tags as conventional light-based geolocation is impossible for tagged animals that spend significant time below the photic zone. We tagged 57 basking sharks with PSAT tags in the NW Atlantic from 2004 to 2011. Many individuals spent several months at meso- and bathy-pelagic depths where accurate light-level geolocation was impossible during fall, winter and spring. We applied a newly-developed geolocation approach for the PSAT data by comparing three-dimensional depth-temperature profile data recorded by the tags to modeled *in situ* oceanographic data from the high-resolution HYbrid Coordinate Ocean Model (HYCOM). Observation-based likelihoods were leveraged within a state-space hidden Markov model (HMM). The combined tracks revealed that basking sharks moved from waters around Cape Cod, MA to as far as the SE coast of Brazil (20◦S), a total distance of over 17,000 km. Moreover, 59% of tagged individuals with sufficient deployment durations (>250 days) demonstrated seasonal fidelity to Cape Cod and the Gulf of Maine, with one individual returning to within 60 km of its tagging location 1 year later. Tagged sharks spent most of their time at epipelagic depths during summer months around Cape Cod and in the Gulf of Maine. During winter months, sharks spent extended periods at depths of at least 600 m while moving south to the Sargasso Sea, the Caribbean Sea, or the western tropical Atlantic. Our work demonstrates the utility of applying advances in oceanographic modeling to understanding habitat use of highly

#### Edited by:

*Mark Meekan, Australian Institute of Marine Science, Australia*

#### Reviewed by:

*Ian David Jonsen, Macquarie University, Australia Uffe Høgsbro Thygesen, Technical University of Denmark, Denmark Heidi Dewar, Southwest Fisheries Science Center (NOAA), United States*

> \*Correspondence: *Camrin D. Braun cbraun@whoi.edu*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *28 September 2017* Accepted: *18 January 2018* Published: *08 February 2018*

#### Citation:

*Braun CD, Skomal GB and Thorrold SR (2018) Integrating Archival Tag Data and a High-Resolution Oceanographic Model to Estimate Basking Shark (Cetorhinus maximus) Movements in the Western Atlantic. Front. Mar. Sci. 5:25. doi: 10.3389/fmars.2018.00025* migratory, often meso- and bathy-pelagic, ocean megafauna. The large-scale movement patterns of tagged sharks highlight the need for international cooperation when designing and implementing conservation strategies to ensure that the species recovers from the historical effects of over-fishing throughout the North Atlantic Ocean.

Keywords: movement ecology, satellite archival telemetry, migration, mesopelagic, oceanographic modeling, site fidelity

### INTRODUCTION

The basking shark, Cetorhinus maximus (Gunnerus 1765), is the second largest fish species, attaining weights of up to 4 tons and lengths up to 12 m (Sims, 2008). It is known to inhabit boreal to tropical (Skomal et al., 2004) waters circumglobally and is most often observed on continental shelves (Sims et al., 2006). Despite its size and widespread distribution, major gaps in our understanding of basking shark ecology remain. Population size and structure are currently unresolved and information about fisheries interactions is limited (Sims, 2008). Although there is evidence to suggest population recovery in some areas following exploitation (Witt et al., 2012), the lack of information about key life history traits, population size, movements, and habitat use is problematic as global anthropogenic pressures on elasmobranchs continue to rise (Dulvy et al., 2008; Ferretti et al., 2010).

Basking sharks exhibit life history characteristics that make them particularly vulnerable to exploitation, including low fecundity, slow growth and maturity, and long gestation times (Compagno, 1984; Sims, 2008). There is, therefore, concern over the status of basking shark populations worldwide, and the species is listed on Appendix II of the Convention on the International Trade in Endangered Species (CITES) and Appendices I and II of the Convention for the Conservation of Migratory Species of Wild Animals (CMS). It is also considered "vulnerable" globally and "endangered" in the northeast Atlantic by the International Union for the Conservation of Nature (IUCN).

Historically, information on the ecology of large pelagic animals has been constrained to scarce observations that are limited geographically (Templeman, 1963; Squire J. L. Jr., 1990; Francis and Duffy, 2002). Almost all of our knowledge of basking shark ecology, for instance, comes from surface observations in coastal waters (Sims et al., 2006; Sims, 2008). Yet recent evidence from electronic archival tags suggests that perhaps the majority of their lives are spent offshore at depths below the euphotic zone (Skomal et al., 2004). Indeed, the rapid development of electronic tag technologies has provided a powerful means of gaining detailed information about the behavior of marine species (Block et al., 2011). Pop-up satellite archival transmitting (PSAT) tags have been particularly helpful in ocean environments as data are relayed back to the researcher via satellite upon tag release from the individual (e.g., Block et al., 2011). These tags have provided a wealth of information on sharks (Berumen et al., 2014; Werry et al., 2014), rays (Braun et al., 2014; Thorrold et al., 2014), and large teleost fishes (Braun et al., 2015a) by eliminating the need to physically recover the tag at the end of the deployment.

While electronic tags have revolutionized the study of movement ecology in the ocean, a significant hurdle remains when attempting to track marine fishes compared with terrestrial counterparts. Tags using Argos or Global Positioning System (GPS) locations require the tag antenna to break the water surface long enough for communication with satellites to be established (Argos) or a snapshot of the satellite constellation to be received (GPS). Researchers have, therefore, relied mostly on PSAT tags that use light-level geolocation in which a threshold algorithm is used to detect solar altitude above the horizon from which estimates of longitude (local noon) and latitude (sunrise/sunset) can be estimated (Hill and Braun, 2001). While sea surface temperature (SST) and bathymetry can improve these estimates (Galuardi et al., 2010; Lam et al., 2010), light-based geolocation requires occupation of the photic zone to record adequate light data for geolocation, and even estimates with quality light data can be error prone (Braun et al., 2015b). However, a number of marine species rarely, if ever, experience enough downwelling light or spend adequate time at the surface to determine their position with PSAT tags (Skomal et al., 2004; Aarestrup et al., 2009; Peklova et al., 2012). Animals that spend significant time at depths below the photic zone have, therefore, proved extremely difficult to track in ocean ecosystems (e.g., Skomal et al., 2004; Dewar et al., 2011).

The use of PSAT tags to track basking shark movements has proved particularly difficult in the northwestern Atlantic as basking sharks spend months at a time below the euphotic zone where light-based geolocation is impossible (Skomal et al., 2004). We have recently developed a new geolocation approach that combines all the physical data collected from archival tags, including light levels and depth-temperature profiles, in a likelihood framework to more accurately track the movements of tagged fishes in the ocean (Braun et al., 2018). Our method uses a purely diffusive animal movement model (e.g., Brownian motion) with behavior state switching (migratory or resident states based on a priori movement speeds) coupled with observations of the environment (e.g., in situ or modeled oceanography) to estimate the posterior distribution of the state (e.g., animal position and behavior) in a hidden Markov model (HMM) framework. Depth-temperature profiles provide diagnostic oceanographic signatures that, along with other data sources like light, SST, and maximum depth, may be leveraged to help constrain position (Skomal et al., 2004; Aarestrup et al., 2009).

Satellite tags have been deployed on basking sharks in the Atlantic since the pioneering work of Priede (1984). Yet, basking shark movements and ecology remain poorly understood. Here, we present the results of an intensive tagging effort that deployed 57 PSAT tags on adult basking sharks during summer months in waters adjacent to Cape Cod, Massachusetts. Profiles recorded by the tags were integrated with high-resolution oceanographic model outputs or in situ climatological data to construct likelihoods and improve geolocation estimates for basking sharks. The data provide a rare assessment of the largescale movements and migratory behavior of the ocean's second largest fish. The information is, in turn, a prerequisite for any attempts to estimate abundance and population structure of basking sharks in the Atlantic Ocean.

### METHODS

#### Study Area and Tagging

We opportunistically deployed a variety of PSAT tags on basking sharks near Cape Cod, Massachusetts (USA) in the Northwest Atlantic (NWA) between 2004 and 2011 (**Table 1**). Total length of each individual was estimated relative to the tagging vessel and, where possible, the pelvic region was visually inspected to determine sex. Tags were applied by a professional harpoon fisherman into the dorsal musculature near the base of the first dorsal fin (Chaprales et al., 1998). This research was performed in accordance with the Woods Hole Oceanographic Institution's Animal Care and Use Committee (IACUC) protocol #16518.

#### Tag Types

Three types of PSAT tags were deployed on basking sharks (**Table 1**). These tags (Models Mk10-PAT, Mk10-AF, miniPAT; Wildlife Computers, Inc., WA, USA) logged depth, temperature, and light level data every 10 s (Mk10-AF) or 15 s (Mk10- PAT, miniPAT) to onboard memory. All tags recorded light data for geolocation purposes, and the Mk10-AF tag housed a Fastloc GPS receiver for acquiring high-resolution location information. Software in the tags summarized the highresolution archived data into depth-temperature profiles at 8 depths (between minimum and maximum depth occupied for the summary period) for a 6, 12, or 24-h period depending on tag programming. These data were compiled into a single daily summary profile for data analysis. Tags also transmitted a summary of an individual's time of occupation within designated depth or temperature bins at 6, 12, or 24-h resolution that was also compiled into daily summaries. Depth and temperature bin number, resolution, and extent differed slightly among tag type and year of tag deployment, but all were compiled to encompass the same depth (<10, 10–25, 25–50, 50–200, 200– 400, 400–1,000, >1,000 m) and temperature bins (<7, 7–9, 9–11, 11–13, 13–15, 15–17, 17–19, 19–21, 21–23, 23–25, >25◦C) for subsequent analysis. Results from the compilation of this timeat-depth and time-at-temperature data represented percent time of each deployment day that an individual occupied each of the common depth or temperature bins (shown above). Seasons were delimited in the analyses by the respective solstice and equinox dates for a given year.

At pre-programmed dates during tag deployment (range of programmed deployment duration 129–361 days), tags were released from the animal using a corrosive burn wire. After the tags released and floated to the surface, summarized data were transmitted to Argos satellites until battery failure. Transmitted data were decoded with manufacturer software (WC-DAP 3.0, Wildlife Computers, Inc., Redmond, WA), and light-based geolocation estimates were calculated and evaluated using tag manufacturer software (WC-GPE2). All subsequent analyses were conducted in the R Statistical Environment (R Core Team, 2016).

#### Geolocation Methods

We estimated most probable tracks for PSAT-tagged basking sharks using the HMMoce package (Braun et al., 2018) for R (R Core Team, 2016). This approach leverages light-levels, SST, depth-temperature profiles, and maximum depth data recorded by PSAT tags, with empirical oceanographic data and model outputs, to construct likelihoods of the tagged individual's movements. Likelihoods are convolved in a spatiallygridded HMM that computes posterior probability distributions to estimate the most likely state (position and behavior) of the animal at each time point, which was daily in this study. Parameter estimation is performed on a 1◦ grid (for improved computation speed), and full model runs use a 0.25◦ grid. In double-tagging experiments, HMMoce was shown to recreate movement trajectories with mean pointwise error of 141 km (range 93–183 km, n = 4) based on light and SST data that represented only 25 and 50% of the deployment days, respectively (Braun et al., 2018), although the geolocation error will likely vary with oceanographic regime and animal behavior.

Briefly, HMMoce estimates location and behavior from electronic archival tags. This involves: (1) calculating spatiallygridded observation likelihoods at each time point based on tag and environmental data; (2) forming the state-space model and estimating model parameters; and (3) model selection and interpretation. At each daily time step, we calculate a likelihood of the animal's position L(x<sup>t</sup> ) on the grid:

$$L(\mathbf{x}\_t) = \begin{bmatrix} L\_1(\mathbf{x}\_t) \ \cdot \ L\_2(\mathbf{x}\_t) \ \dots \ L\_n(\mathbf{x}\_t) \end{bmatrix}$$

where 1:n indicates individual, observation-based likelihoods formed for each type of input data at each time point [e.g., LSST(x<sup>t</sup> )]

Observation-based likelihoods were derived from in situ SST, light-based longitude, and depth-temperature profile data collected by the tags, using five separate likelihood calculations as follows and filtered using a bathymetric mask. (1) An SST likelihood was generated for tag-based SST-values integrated according to an error term (±1%) and compared to remotelysensed SST from daily optimally-interpolated SST (OI-SST, 0.25◦ resolution) fields (Reynolds et al., 2007; Banzon et al., 2016). (2) Light-based longitude likelihood was derived using estimates of longitude from GPE2 software (Wildlife Computers, Inc.), which facilitated visual checking of light curves. Depth-temperature profiles recorded by the tag were compared to (3) daily reanalysis model depth-temperature products from the HYbrid Coordinate Ocean Model (HYCOM, 0.08◦ resolution; Bleck, 2002; Chassignet et al., 2007), and (4) monthly climatological mean depth-temperature data from the World Ocean Atlas 2013 (0.25◦ resolution; Locarnini et al., 2013) at standard depth levels TABLE 1 | Summary information from satellite tag deployments on *Cetorhinus maximus* in the NW Atlantic.


*Identification number of each individual is shown along with the tag model. All tags were manufactured by Wildlife Computers, Inc. (Redmond, WA, USA). Est. Length, the total length (m) of the individual tagged as estimated from the tagging vessel; Sex, male (M) or female (F) where determination was possible by visual observation of presence or absence of claspers between the pelvic fins, no entry indicates that sex could not be confidently determined; Pop Lat/Lon, coordinates of tag detachment location; Deploy Duration, number of days between tag deployment and detachment; Max Depth, the deepest depth (m) reported by the tag during the deployment; Track Distance, cumulative distance of most probable track; Light, SST and depth-temperature profile (PDT) columns indicate percent of deployment days with light-based location estimates, sea surface temperature data and depth-temperature profiles. Observation likelihoods are those observations used in HMMoce to construct the most probable track for each tagged animal: L, light-based longitude; S, sea surface temperature; H, HYCOM depth-temperature profiles; W, World Ocean Atlas depth-temperature profiles; O, integrated Ocean Heat Content; F, Fastloc GPS; DD, data deficient.*

*<sup>a</sup>Tracks published in Skomal et al. (2004).*

*<sup>b</sup>Depth data published in Curtis et al. (2014).*

*<sup>c</sup>Tag was physically recovered.*

*<sup>d</sup>Maximum depth capability of this tag model.*

*<sup>e</sup>No track was constructed. This is a straight-line (displacement) distance from tagging location to pop-up.*

available in these products. Individual likelihood surfaces for each depth level were then multiplied together for an overall profile likelihood at that time point. (5) Ocean Heat Content (OHC) was obtained by integrating the heat content of the water column above the minimum daily temperature to the most shallow depth recorded by the tag for both the tag profiles and HYCOM fields (Luo et al., 2015).

All observation-based likelihoods were formed using integrated likelihood calculations (Le Bris et al., 2013). For example, daily SST likelihoods were constructed as:

$$L\_{\mathrm{SST}}(\mathbf{x}\_t) = \int\_{\mathrm{SST}\_{\mathrm{min}}}^{\mathrm{SST}\_{\mathrm{max}}} N\left(t; \,\mu\_z, \sigma\_z\right) dz$$

where N is a normal probability distribution function, µ<sup>z</sup> the remotely-sensed SST grid cell value, and σ<sup>z</sup> the grid cell standard deviation. The same integration approach was performed on the other observation likelihoods. For 3D likelihoods, this approach was performed at each relevant standard depth level in the environmental dataset and integrated limits were tag-based minimum and maximum temperatures recorded (or predicted by linear regression) at that depth level. Standard deviation for all likelihood calculations was calculated with a "moving window" mean using the focal() function in the raster package (Hijmans, 2016) for R to incorporate ∼0.25◦ of environmental data around each grid cell. Start and end locations and available GPS data (from the MK10-AF tag) were seeded as known positions in all model runs.

The resulting observation likelihoods (in various combinations; **Table 1**) were used in a two-step Bayesian state-space approach to estimate the posterior distribution of the state (in this case, a joint probability distribution of location and behavior at each time point). We considered "resident" and "migratory" behavior states that corresponded to fixed speeds of 0.4 m s−<sup>1</sup> (34.5 km d−<sup>1</sup> ) for residency (following Curtis et al., 2014) and an order of magnitude higher (4 m s−<sup>1</sup> , 345 km d −1 ) for migratory movements. These speeds represent maximum diffusion allowed per day (1,200 and 120,000 km<sup>2</sup> d −1 for resident and migratory daily diffusion, respectively) and were represented by Gaussian kernels (see documentation for HMMoce::gausskern for more information) that were convolved with observation likelihoods at each time point. Probability distributions were first calculated forward in time using alternating time and data updates of the current state estimate using a HMM filter on the derived likelihood grid. Parameter estimation was performed using an iterative Expectation-Maximization framework (Woillez et al., 2016). The HMM smoother recursion was the final step that worked backwards in time using filtered state estimates and all available observation data to determine smoothed state estimates. This step provided the time marginal of the probability distributions based on observations (posterior distributions). Distributions are summed for each behavior state and time step to determine the most likely behavior state for each time step. HMMoce calculates the mean or mode of the posterior distribution grid, at each time step, to estimate the animal's most probable track. Model selection was performed using Akaike Information Criterion (AIC). Resulting most probable track estimates represented daily location and most likely behavior state at that time point. Cumulative track distances were calculated using great-circle distance calculations between estimated daily locations using the rdist.earth function in the fields (Nychka et al., 2015) package for R.

The posteriors were summed across behavior states for additional inference on seasonal habitat use, which were conceptually similar to a residency (see Equation 5, Pedersen et al., 2011) or utilization distribution (Royle and Dorazio, 2008). This approach was used to incorporate uncertainty around most probable track estimates that is included in the posteriors, as opposed to traditional utilization distribution calculations based on, for example, kernel density (e.g., Berumen et al., 2014).

### RESULTS

We tagged 57 basking sharks spanning sub-adult (∼500–600 cm) and adult (>600–700 cm) life stages (range 549–762 cm males, 549–823 cm females) and both sexes (10 females, 3 males, 31 unknown). Forty-five (79%) of the 57 PSAT tags deployed between 2004 and 2011 reported. Eight tags released prematurely, and one of the tags had no useable data. Data from 37 of the remaining 44 tags contained sufficient information for further analysis (**Table 1**). These deployments averaged 234 days (SD 85 days, range 79–424 days). There was no evidence of tagging-induced mortality. Of the 35 tags that transmitted data (excluding two that were physically recovered), we received data representing 7% (median, range 1–44%), 26% (median, range 4–61%), and 52% (median, range 7–91%) of deployment days with light-based position estimates, SST, and depth-temperature profile data, respectively. The remaining two tags were physically recovered: one tag washed ashore in The Bahamas after 133 days at liberty and one was located on a beach in Rhode Island still attached to the deceased shark after a 78 day deployment. The full archival record was analyzed for these two deployments and contained light-based position estimates and SST data for 5– 51 and 66–89% of deployment days, respectively, during which the animal occupied the surface (SST) or euphotic zone (light). Transmitted and archival profile data were available for more deployment days than either light-based position estimates or SST data in all but one of the reporting tags. One individual (B28) was tagged with a Fastloc GPS tag which reported 4 GPS snapshots over 3 days during winter (Dec. 22, 23, 26). These locations were considered known in the model runs for this individual, and no other usable GPS positions were acquired.

For a given tag, varying amounts of each data type were obtained due to behavioral variability and individual differences in data transmission. Model selection favored HYCOM-based profile likelihoods (**Figure 1**) in 34 of 37 track calculations. Of the remaining three individual geolocation analyses, one favored OHC-based profile likelihoods, one WOA-based profile likelihoods, and one model selection used only light and SST observations. Available light and SST data were not used in the selected model for four and six individual tag datasets, respectively (**Table 1**). Nearly all model outputs indicated the "migratory" behavior state was more likely once the tagged

boxes show the oceanographic regions discussed in the text, Figure 7, and Table 2 and correspond to New England (red), Sargasso Sea (green), Antillean Arc

individual left the New England shelf (76% of off-shelf position estimates), and this behavior remained dominant throughout the Sargasso Sea region (77% of off-shelf position estimates). Shelf habitats near New England and from the Antillean Arc to the Amazon Delta were characterized by a higher likelihood (∼50% of on-shelf position estimates) of "resident" behavior (e.g., slower, more tortuous movements).

(purple) and South America (blue).

While all tags were deployed off the northeastern coast of the U.S., most probable tracks showed a wide range of individual movements (**Figure 2**). For individuals with sufficient data to perform the geolocation analysis (n = 34), track distances ranged from 4,009 to 17,387 km (mean 10,136 ± 3,988 SD) spanning 79– 424 days (mean 207 ± 107 SD). Several of the sharks showed relatively directed, long-range movements south from the tagging location in New England to the Puerto Rico Trench (n = 4), Antillean Arc (n = 3), and Amazon Delta (n = 3) up to 17,387 km (6,200 km displacement) from the tagging location (**Figure 2**). Three individuals made transequatorial movements.

Movements of tracked sharks demonstrated strong seasonality (**Figures 2**, **3**) with individuals occupying coastal waters in high latitudes during the summer before moving south in fall (**Figures 2**, **3**), and all but one individual (B26) departed New England by January. This individual remained along the shelf edge between New England and the Grand Banks for the winter and returned to the New England canyons by late February (B26 in **Figure 4**). All other tagged sharks overwintered in habitats as close as the Sargasso Sea and as far as the northeastern coast of Brazil before beginning to return to New England waters in late spring and early summer (**Figures 2**, **3**). Seven tags were deployed for >300 days, including one for 423 days, and five of them transmitted sufficient data for track estimation. Six of these seven tags popped up in New England waters ∼1 year after tagging (e.g. B20, B22 in **Figure 5A**), while the remaining tag reported near the Amazon Delta and represented the furthest southerly movements observed in this study (**Figures 2**, **5**). Eighteen tags exhibited deployment durations >250 days, ten of which (59%) exhibited return migrations to the NWA, including one pop-up location 60 km from the tagging location 1 year later (B21). There was no significant difference in mean track distance between males and females (t-test, p = 0.4633), although male sample size was low (n = 3), and a linear regression analysis found no significant relation between shark size and extent of movement (p = 0.27, R <sup>2</sup>= 0.05) or minimum latitude occupied (p = 0.48, R <sup>2</sup>= 0.02).

Long-distance migrations often co-occurred with large vertical excursions and led to occupation of several distinct water masses throughout the year. Binned vertical histogram data (**Figure 3**) were used to quantify where in the water column sharks tended to frequent. Overall, extensive vertical excursions characterized basking shark dive behavior when an individual left the continental shelf region of the eastern US (**Figures 3**, **4**, **6**). Twenty-one individuals spent time below 1,000 m, and it was likely that only limitations in earlier tag technology (maximum

depth capability of 980 m) prevented those individuals' tags from recording similar behavior. The maximum depth recorded by a tag (shark B42) was 1,504 m and recorded temperatures at depth in this study ranged from 4.2 to 29.9◦C. Recorded SST-values from all individuals ranged from 7.4 to 29.9◦C (median 18.3◦C). Overall, 63% of basking shark depth-temperature data was 8– 18◦C, 87% was between 6 and 20◦C, and all individuals made occasional forays into temperatures well-outside those bounds (**Figure 7**). In fact, one individual (B26) remained at northern latitudes (from Cape Cod to the Grand Banks) during winter and experienced <12◦C ambient water temperatures for >3 months (B26 in **Figure 4**; range 4.8–12◦C from Nov 1 to Feb 15).

Vertical habitat envelopes described the distinct water masses across the study area (from coastal New England to open ocean off Brazil), their depth-temperature characteristics, and the vertical behavior observed in each water mass (**Figure 7**, **Table 2**). Generally, individuals spent much of their time in the epipelagic zone (<200 m) during summer months at northern temperate latitudes where temperatures were typically <20◦C (**Figures 4**, **6**, **7**). However, during the fall, the majority of tagged individuals transitioned from the epipelagic orientation of the summer months to residency in the mesopelagic zone during the winter in which they cumulatively spent >60% of time between 400 and 1,000 m (**Figures 3**, **4**, **6**). Based on depth-temperature profile data, sharks remained below the euphotic zone for 27% (median; range 0–90%) of fall, winter and spring deployment days for which data existed, and this behavior exhibited no relationship with individual size or sex, although male sample size was low (n = 3). Temperature profiles from these periods of mesopelagic occupation indicated this behavior occurred largely in the Sargasso Sea where warm (14–20◦C) water penetrates deep in to the water column (profile C in **Figures 1**, **7**) resulting in relatively warm water at depth (e.g., B20 and B22 in **Figures 6**, **7**). However, some sharks overwintered further south in the Guyana Basin and off the Brazilian shelf as indicated by warmer surface temperatures and a stronger temperature gradient with depth (e.g., profile B in **Figure 1**, B36 in **Figures 6**, **7**). Sharks generally inhabited warmer waters throughout winter at low latitudes, despite prolonged deep-water occupation, than the surface waters that they inhabited during summer months (**Figure 7**, **Table 2**).

FIGURE 6 | Daily depth-temperature profiles (row 1) and time-at-depth profiles (row 2) for three representative basking sharks (tracks plotted in Figure 5B). Note differing time scales (x-axis) among individuals.

Shark B22 provided a good example of the distinct water masses traversed during a 1-year deployment, with a complete round trip migration starting and ending in the tagging region (**Figures 4**, **5**). This individual occupied a well-mixed, cool surface layer in the Gulf of Maine during October before moving through the Gulf Stream and into the northern Sargasso Sea in November. This individual occupied the northern Sargasso from December to March before moving back into a more uniformly cool layer in April and May near Cape Hatteras. By June, both the estimated track and water characteristics indicate this individual had returned to the shelf-edge waters near New England and onto the shelf near Cape Cod by late September (**Figure 4**).

#### DISCUSSION

It is increasingly clear that pelagic fishes throughout the global ocean conduct long-range migratory movements (e.g., Block et al., 2011; Skomal et al., 2017) and connect the surface and deep ocean through meso- and bathy-pelagic dive behavior (Braun et al., 2014; Thorrold et al., 2014). The basking sharks tagged in the present study were no exception, making some of the longest horizontal movements of any ocean species tagged to date (Block et al., 2005; Bonfil, 2005; Hays et al., 2006; Skomal et al., 2017). Tagged individuals moved through several distinct water masses of the western Atlantic, and spent significant time in the mesopelagic, demonstrating the ability of basking sharks to

TABLE 2 | Summary statistics for vertical habitat envelopes in Figure 7 by region and season.


*Reported values are formatted as median (minimum–maximum) for sea surface temperature (SST), minimum daily depth (Min Z), maximum daily depth (Max Z), and minimum daily temperature (Min T). Temperatures are* ◦*C and depths are in meters. Sample sizes (N) indicate total number of data points (not individual profiles) and are shown for each region-season combination. Blank combinations in the table indicate no data were collected for that combination. Note these data were restricted to the spatial areas of interest as shown in* Figure 1 *and may not exactly match reported statistics in the text which included all data.*

traverse a wide range of environments from the surface to deep ocean across a 25◦C temperature range.

Movements through distinct water masses often coincided with varying periods of deep water occupation. Nearly all tagged individuals demonstrated a shift from residency in surface waters to deep water occupation in the meso- and bathypelagic during colder months that may explain the apparent disappearance of basking sharks during winter (Parker and Boeseman, 1954). While our results corroborate previous studies that suggest seasonally variable dive behavior (Sims et al., 2003) and southward migration during winter (Doherty et al., 2017), sharks in this study made much more extensive movements throughout the open ocean than those observed in similar studies elsewhere (Doherty et al., 2017) and spent up to several months at mesopelagic depths. Sharks tagged in the northeast Atlantic (NEA) did make dives to similar maximum depths (∼50% of tagged individuals dove below 1,000 m; Doherty et al., 2017) but averaged >80% of time above 200 m and <10% deeper than 500 m (Sims et al., 2003; Doherty et al., 2017). The mesopelagic occupation observed in this study suggests this behavior is much more ubiquitous among NWA basking sharks as they move throughout the open ocean than their NEA conspecifics that remain oriented to the shelf. This apparent difference and may be a product of the oceanography experienced (e.g., warm, homogenous depth-temperature profiles in the Sargasso Sea) by these individuals in the open ocean of the NWA.

The other main difference in behavior among these regions is the winter migration strategy. NEA basking sharks moved south from Ireland and the UK to the Bay of Biscay, but despite tagging 70 basking sharks with satellite tags, only one individual traversed >20◦ of latitude after summer occupation of the far northern latitudes (Doherty et al., 2017). In contrast, winter movements at and beyond this scale were more commonly observed in the NWA (Skomal et al., 2004 and this study). These observed movements demonstrate that tropical environments do not pose a barrier to basking shark movements and refute the suggestion that this species is largely restricted to temperate latitudes (Sims, 1999; Sims et al., 2003; Gore et al., 2008; Doherty et al., 2017).

The long-distance movements by basking sharks in this study are likely driven, at least in part, by the dynamic oceanographic environment of the western Atlantic Ocean. The NWA, in particular, is punctuated by strong seasonal fluxes in pelagic primary productivity (Miller and Wheeler, 2012) and temperature (Talley, 2011). The warm water and high productivity attract many species to the temperate NWA during summer (e.g., basking sharks, Curtis et al., 2014; white sharks, Skomal et al., 2017). While it is clear basking sharks are able to tolerate sub-12◦C water for months at a time (B26 in **Figure 4**; Sims, 2008), individuals in this study spent much of their time overwintering in warm, mesopelagic waters. In fact, as a whole, sharks spent more time in warmer water during deep occupation periods in winter as they moved south than they did during summer. While the function of this deep occupation is unknown, the Sargasso Sea is a relatively stable, warm water mass during winter months and may host prey opportunities for basking sharks in the mesopelagic, including a substantial deep scattering layer that overlaps with basking shark depth use (400–600 m; Irigoien et al., 2014) and potentially co-occurring anguillid eel spawning aggregations (Wysujack et al., 2015). These migrations away from the northern winter may also be associated with hotspots of relatively high production at lower latitudes (e.g., Brazilian shelf; Mourato et al., 2014). Movements in this study demonstrated orientation to shelf edge habitats, particularly along the northern coast of Brazil during winter, that likely host persistent fronts (Le Fèvre, 1987; Sims, 2008) and thus relatively high primary production even at low latitude. While basking sharks have been shown to orient to persistent seasonal fronts (Miller et al., 2015), most individual tracks in this study demonstrated intense occupation of near-shelf regions that was punctuated by lengthy offshore excursions. Thus, perhaps the combination of favorable growth energetics associated with warm overwintering habitat (relative to overwintering at temperate latitudes) and food availability drive southerly movements away from temperate latitudes for winter and the mesopelagic occupation in (sub)tropical waters observed

here. However, further work is needed to test the role of energetics and food resources as drivers of basking shark migrations.

Movement patterns of tagged basking sharks may also be associated with reproduction (Skomal et al., 2004). Basking sharks are commonly observed along the northeastern US during summer, presumably to forage; however, mating may also occur during this period while sharks are aggregated and potential courtship behavior has been observed (Wilson, 2004). Subsequent movements into the tropical Atlantic and occupation of mesopelagic depths may be a predator avoidance or parturition strategy as these environments are characterized by mild, stable conditions. This may further explain the lack of observations of pregnant females despite prolonged coastal fisheries in the NEA (Sims, 2008). Thus, while we did not observe significant differences in movement between sexes, the females that undertake long-range southerly migrations may be exploiting stable environmental conditions for gestation and parturition, and the stable habitat and relative lack of predators may provide suitable nursery habitat for neonates. The presence of <2.5 m TL basking sharks in the Gulf of Mexico during spring (Hoffmayer et al., 2011) lends some support for this hypothesis as it suggests that parturition is occurring during winter months in tropical or subtropical waters. The wide variation in movement patterns (>50◦ range in latitude) suggests these migrations were not driven by a localized mating event somewhere in the Atlantic. Unfortunately, we were unable to sex a significant portion of tagged individuals in this study due to tag application methods, and the limited sample size of sexed individuals indicates no difference in movements between sexes that may further clarify reproductive hypotheses.

Highly variable dive behavior, including extended forays away from the photic zone, exhibited by basking sharks made traditional light-based geolocation difficult in our study. Thus, we employed a recent advance (based on extensive work by Pedersen et al., 2008, 2011) in geolocation analysis methods to supplement missing light data with other forms of data recorded on the tag (Braun et al., 2018). Depth-temperature profiles, in particular, provided substantially more information to be used for geolocation than light and SST data used in traditional geolocation approaches. These profiles provided observations that were used for geolocation when tagged individuals were away from the surface and the tags were unable to collect light and SST metrics. In addition, the profile data yielded diagnostic depth-temperature profiles that were compared to modeled or in situ oceanographic data to reduce geolocation error (Braun et al., 2018). By using the high-resolution (0.08◦ ) HYCOM reanalysis product, we were able to leverage the synoptic daily coverage of an oceanographic model that incorporates available in situ data to improve geolocation estimates. While previous tracking studies have highlighted the potential for error when using HYCOM outputs to represent the extremely dynamic Gulf Stream eddy field (Braun et al., 2018), the majority of basking sharks in this study moved latitudinally and spent relatively little time in the most dynamic regions of the NWA.

Model outputs also indicated a higher likelihood of "residentlike" movements in productive shelf habitats around New England and off the Antilles and South America. It is likely these restricted movements are indicative of foraging in these relatively productive shelf habitats (Mourato et al., 2014). In contrast, migratory movements (4 m s−<sup>1</sup> ) were more likely in pelagic waters, including during overwintering in the Sargasso Sea. Because of model formulation, the higher speeds that we classified as "migratory" may also be more likely, overall, due to the scale at which the observation likelihoods are formulated. For instance, if tag-based SST corresponds to remotely sensed SST over a broad area (e.g., Sargasso Sea), we may expect migratory behavior to be more likely than the resident behavior that would result from more constrained likelihoods (e.g., tag-based SST matching more closely to a confined region). While this approach is significantly more computationally-intensive than traditional light-based geolocation approaches (see Table S2 in Braun et al., 2018), comparing tag data directly to in situ and/or modeled oceanographic profiles from the same time frame results in a more realistic representation of shark movements and the oceanographic environment they inhabit.

The basking shark tracks documented here represent the largest scale movements reported for basking sharks, including one individual's estimated track distance covering >17,000 km, and the deepest dive recorded by a basking shark (1,504 m). The observed tracks further expand the known range of basking sharks reported by Skomal et al. (2004). We recorded three individuals making transequatorial migrations yet no tagged individuals made significant longitudinal movements toward the NEA. North-south movements were, therefore, much more common in the portion of the NWA population sampled here than east-west movements that may, in turn, limit the exchange of genetic material between the NWA and NEA. In contrast, Gore et al. (2008) found that one of two satellite-tagged basking sharks moved from the Isle of Man to the eastern coast of Newfoundland in <3 months. In addition, there is little evidence for genetic structuring of basking sharks in the Atlantic (Hoelzel et al., 2006), suggesting sufficient connectivity to at least maintain panmixia between NEA and NWA populations.

### CONCLUSION

The current reliance on light levels for geolocation of many marine fishes renders geolocation impossible when tagged individuals spend significant time below the euphotic zone. Tagged sharks in this study spent significant time at mesopelagic depths, particularly during winter, at which light levels were too low for geolocation. We supplemented light-based geolocation with position estimates generated by matching depth-temperature profiles collected by the sharks' tags to in situ or modeled oceanographic profiles. Our approach provided considerably more information on movement patterns than are typically available from PSAT data with limited light-level information, providing a valuable method for studying marine species that do not frequent the euphotic zone. The resulting basking shark tracks demonstrated large-scale movements up to over 17,000 km from Cape Cod to southern Brazil, winter residency in New England waters, and a range of behaviors in between. Most individuals exhibited seasonal movements into the Sargasso Sea during winter and multiple deployments of sufficient duration captured the return migration to Cape Cod the subsequent summer. Basking sharks in this study traversed multiple distinct water masses through the western Atlantic and exhibited basin-scale movements that warrant international cooperation for adequate management of this species. Winter habitat use was characterized by occupation of mesopelagic waters at low latitudes during which individuals often left the surface for months at a time. This cryptic deep-water overwintering provides impetus for further study of this poorly understood species.

#### AUTHOR CONTRIBUTIONS

GS and ST: designed the study and conducted the tagging; CB: performed the analysis and wrote the manuscript. All authors contributed ideas throughout the process and read, commented on, and approved the final manuscript.

### REFERENCES


#### ACKNOWLEDGMENTS

We thank B. Galuardi and C. H. Lam for contributing analysis code, and H. Dewar, U. Thygesen and I. Jonsen for valuable feedback on the manuscript. We gratefully acknowledge funding from the US National Science Foundation (OCE 0825148), the National Aeronautics and Space Administration (NNS06AA96G), the Massachusetts Environmental Trust, and the Federal Aid in Sport Fish Restoration Program. Computational support was provided by the AWS Cloud Credits for Research program. CB was funded by the Martin Family Society of Fellows for Sustainability Fellowship at the Massachusetts Institute of Technology, the Grassle Fellowship and Ocean Venture Fund at the Woods Hole Oceanographic Institution, and the NASA Earth and Space Science Fellowship. Funding for the development of HYCOM has been provided by the National Ocean Partnership Program and the Office of Naval Research. Data assimilative products using HYCOM are funded by the U.S. Navy. Computer time for HYCOM was made available by the DoD High Performance Computing Modernization Program.


**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 Braun, Skomal and Thorrold. 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 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.

## TurtleCam: A "Smart" Autonomous Underwater Vehicle for Investigating Behaviors and Habitats of Sea Turtles

Kara L. Dodge<sup>1</sup> \*, Amy L. Kukulya<sup>2</sup> , Erin Burke<sup>3</sup> and Mark F. Baumgartner <sup>1</sup>

*<sup>1</sup> Biology Department, Woods Hole Oceanographic Institution, Woods Hole, MA, United States, <sup>2</sup> Oceanographic Systems Laboratory, Applied Ocean Physics and Engineering Department, Woods Hole Oceanographic Institution, Woods Hole, MA, United States, <sup>3</sup> Massachusetts Division of Marine Fisheries, New Bedford, MA, United States*

Sea turtles inhabiting coastal environments routinely encounter anthropogenic hazards, including fisheries, vessel traffic, pollution, dredging, and drilling. To support mitigation of potential threats, it is important to understand fine-scale sea turtle behaviors in a variety of habitats. Recent advancements in autonomous underwater vehicles (AUVs) now make it possible to directly observe and study the subsurface behaviors and habitats of marine megafauna, including sea turtles. Here, we describe a "smart" AUV capability developed to study free-swimming marine animals, and demonstrate the utility of this technology in a pilot study investigating the behaviors and habitat of leatherback turtles (*Dermochelys coriacea*). We used a Remote Environmental Monitoring UnitS (REMUS-100) AUV, designated "TurtleCam," that was modified to locate, follow and film tagged turtles for up to 8 h while simultaneously collecting environmental data. The TurtleCam system consists of a 100-m depth rated vehicle outfitted with a circular Ultra-Short BaseLine receiver array for omni-directional tracking of a tagged animal via a custom transponder tag that we attached to the turtle with two suction cups. The AUV collects video with six high-definition cameras (five mounted in the vehicle nose and one mounted aft) and we added a camera to the animal-borne transponder tag to record behavior from the turtle's perspective. Since behavior is likely a response to habitat factors, we collected concurrent *in situ* oceanographic data (bathymetry, temperature, salinity, chlorophyll-*a*, turbidity, currents) along the turtle's track. We tested the TurtleCam system during 2016 and 2017 in a densely populated coastal region off Cape Cod, Massachusetts, USA, where foraging leatherbacks overlap with fixed fishing gear and concentrated commercial and recreational vessel traffic. Here we present example data from one leatherback turtle to demonstrate the utility of TurtleCam. The concurrent video, localization, depth and environmental data allowed us to characterize leatherback diving behavior, foraging ecology, and habitat use, and to assess how turtle behavior mediates risk to impacts from anthropogenic activities. Our study demonstrates that an AUV can successfully track and image leatherback turtles feeding in a coastal environment, resulting in novel observations of three-dimensional subsurface behaviors and habitat use, with implications for sea turtle management and conservation.

Keywords: autonomous underwater vehicle AUV, CTD, entanglement, habitat, foraging behavior, jellyfish, leatherback sea turtle, video camera

#### Edited by:

*Lisa Marie Komoroske, University of Massachusetts Amherst, United States*

#### Reviewed by:

*Larisa Avens, Southeast Fisheries Science Center (NOAA), United States Sabrina Fossette, National Oceanic and Atmospheric Administration (NOAA), United States*

> \*Correspondence: *Kara L. Dodge kdodge@whoi.edu*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *03 January 2018* Accepted: *05 March 2018* Published: *20 March 2018*

#### Citation:

*Dodge KL, Kukulya AL, Burke E and Baumgartner MF (2018) TurtleCam: A "Smart" Autonomous Underwater Vehicle for Investigating Behaviors and Habitats of Sea Turtles. Front. Mar. Sci. 5:90. doi: 10.3389/fmars.2018.00090*

## INTRODUCTION

Sea turtles inhabiting coastal environments routinely encounter anthropogenic hazards, including fisheries, vessel traffic, pollution, dredging, and drilling (Lutcavage et al., 1996). These urbanized habitats often require successful navigation of multiple threats to avoid injury or mortality. While there is evidence that sea turtles can modify their behavior based on variable environmental conditions (Hays et al., 2006; Schofield et al., 2009), there is currently little information on behavioral adaptations relative to transient human activities. Sea turtle mortalities in coastal habitats frequently show evidence of human interaction that suggests they are unable to avoid some anthropogenic threats, including vessel strike (Hazel and Gyuris, 2006; Tomás et al., 2008), fishery interactions (Peckham et al., 2007; Casale et al., 2010; Hamelin et al., 2017), and debris ingestion (Bjorndal et al., 1994; Mrosovsky et al., 2009). Furthermore, some turtle behaviors (e.g., shallow water diving, affinity for nearshore habitat) may actually exacerbate risk (Hazel et al., 2009). To develop effective mitigation strategies, we need to understand how turtle behavior and habitat choice mediate risk in coastal feeding grounds.

To investigate fine-scale movements and behaviors of sea turtles, researchers have employed a variety of tools and techniques, including direct subsurface observations (Schofield et al., 2006), radio tags (Avens et al., 2003; Brooks et al., 2009), acoustic tags (van Dam and Diez, 1998; Seminoff et al., 2002), archival tags (Southwood et al., 1999; Fossette et al., 2010), animal-borne camera tag packages (Heithaus et al., 2002; Reina et al., 2005), remotely operated vehicles (ROVs) (Patel et al., 2016), and various combinations of these technologies. While these approaches have resulted in a richer understanding of turtle movements and behaviors in foraging and breeding habitats, few studies have concurrently investigated fine-scale turtle behavior and measured in situ habitat characteristics. Since animal behavior is responsive to environmental conditions, accurate interpretation of turtle behavior requires an understanding of their bio-physical habitat.

Recent advancements in autonomous underwater vehicles (AUVs) now make it possible to directly observe subsurface behaviors and concurrently sample the habitats of marine megafauna (Skomal et al., 2015; Kukulya et al., 2016). Here, we describe a "smart" AUV developed to follow and film free-swimming marine animals, and demonstrate the utility of this technology in a pilot study investigating the subsurface behaviors and habitat of leatherback sea turtles (Dermochelys coriacea) in a high-risk coastal environment. We adapted the SharkCam REMUS-100 AUV developed at the Oceanographic Systems Laboratory (OSL) at the Woods Hole Oceanographic Institution (WHOI) (Packard et al., 2013), and coined it "TurtleCam." TurtleCam's animal-following algorithms were modified to continuously locate, follow and film a tagged turtle while simultaneously collecting environmental data along the turtle's track. We conducted our pilot study off Cape Cod, Massachusetts, USA, where leatherbacks are resident during boreal summer and fall months (Lazell, 1980; Dodge et al., 2014). In Massachusetts' coastal waters, the leatherbacks' spatially restricted movements, particularly in bays, sounds, and shoal/ledge habitat, coincide with high densities of fixed fishing gear (e.g., weighted gear set on the sea floor), concentrated commercial and recreational vessel traffic, subtidal aquaculture, and renewable energy operations and development (2015 Massachusetts Ocean Management Plan, http://www.mass.gov/ eea/docs/eea/oceans/ocean-plan/2015-ocean-plan-v1-complete. pdf). Anthropogenic impacts to leatherback turtles in this region are well-documented by the Greater Atlantic Region Sea Turtle Stranding and Disentanglement Networks, with the primary sources of leatherback mortality attributed to vessel strikes and entanglement in fixed fishing gear (Dwyer et al., 2002; Sampson, 2012). We used the TurtleCam system to understand how leatherback behavior and habitat choice mediated risk to anthropogenic threats, with the ultimate goal of incorporating this information into regional conservation and management plans.

### REMUS TURTLECAM SYSTEM

The REMUS TurtleCam system used in this study is custom built at WHOI and available to the research community though collaboration with the WHOI OSL. The system consists of a 25-kHz cylindrical transponder tag (7.6 × 38 cm, 1.7 kg in air, ∼58 g buoyancy in seawater), a 100-m depth rated AUV (203 × 19 cm, 48 kg in air), and a shipboard tracking system for tracking the animal independent of the AUV if desired. The 25-kHz frequency of the TurtleCam transponder tag is well above the known underwater hearing range and sensitivity of leatherback turtle hatchlings (Dow-Piniak et al., 2012). While hearing range and sensitivity is unknown for adult leatherbacks, studies involving adult sea turtles of other species also found that they detected sounds in the low frequency sound range (e.g., Bartol et al., 1999), and we do not expect leatherbacks to perceive the higher frequency sound of the transponder. The transponder is attached to the turtle via a custom-built mechanism with two suction cups, and has an embedded acoustic release system that can be activated via an acoustic signal from the tracking boat or by a preprogrammed depth. Once attached to the turtle, the tag remains in a listening-only standby mode, awaiting coded acoustic signals (pings) from either the vehicle or shipboard tracking system. Depending on the ping duty cycle, the tag can remain in a low power state lasting for up to 3 weeks.

TurtleCam is launched (typically within minutes) after a turtle is tagged, and immediately dives and swims to the tagging location programmed by the vehicle operator between tagging and deployment. While transiting to the last known position, the vehicle interrogates the transponder tag every 3 s using its onboard acoustic system, including a 360◦ Ultra-Short BaseLine (USBL) receiver array, and follows the tag using navigational algorithms for tracking a randomly moving animal. The transponder tag "listens" for a coded signal (interrogation ping) and replies back with two coded signals. The round-trip travel time of the response is calculated by the vehicle and used to determine the range to the turtle. This response is then beam-formed to determine the turtle's bearing relative to the

vehicle, and the vehicle's compass (Model Sparton AHRS-8) is used to transform this into an absolute bearing. From these range and bearing estimates, the turtle's location in earth coordinates (latitude/longitude) can be determined. A second response from the tag is time delayed proportional to depth. The time difference between the two responses is used to determine the depth of the turtle. This combination allows precise location of a tagged turtle in three-dimensional space (**Figure 1**) (Kukulya et al., 2015).

In addition to tracking with TurtleCam, the tagged turtle can be tracked from the boat using the shipboard tracking system (STS). The STS is a "vehicle in a box" that uses a shipboard-mounted USBL array, transducer, ship's global positioning system (GPS) receiver, and compass to independently track the tagged turtle. The vehicle pilot also uses the STS to communicate in real time with the TurtleCam vehicle using bidirectional acoustic modems (Model 1.3, acomms.whoi.edu). The STS enables the real-time positions of the turtle, vehicle and boat to be plotted on the operator's laptop, giving the research team immediate visual feedback on how the system is working and what the turtle is doing. Moreover, the turtles surfaced frequently in our study, and direct visual observations could be made from the boat to corroborate the estimated turtle positions derived from the TurtleCam and STS.

The architecture of the animal-following algorithm allows an operator to change the vehicle's position relative to an animal in real time. The algorithm allows multiple parameters that are preprogrammed and then modified during the mission, including speed near the animal (within 10 m), speed far from animal (beyond 10 m), the distance to the animal above, below, forward, behind, port and starboard (in meters). Since individual foraging and swimming behaviors can be highly variable, having the ability to manipulate the position of the vehicle relative to an animal in real time is critical. Real-time visualization via the STS allowed the operator to make mission parameter changes as needed. The duration of TurtleCam missions are limited by the batteries in the cameras (230 min mean battery life). If desired, the research team can recover the vehicle to change out the batteries and SD cards on the cameras, and then relaunch the vehicle for a second mission. While the vehicle is out of the water, the tagged turtle can be tracked with the STS. Upon mission completion, the tag is acoustically commanded to release from the animal using the STS; the suction cups on the turtle are flooded upon receipt of this command, and the slightly buoyant tag floats to the surface where it can be located via the STS or a VHF radio receiver and recovered.

While tracking the tagged turtle, the vehicle collects video with six high definition cameras (Model HERO3+, GoPro, Inc.; www.gopro.com). Five cameras are mounted on the nose and one rear-facing camera is mounted on the top. We also added a camera (Model HERO Session, GoPro, Inc.; www. gopro.com) to the animal-borne transponder to record behavior from the turtle's perspective. The vehicle can be outfitted with different environmental sensors to collect bio-physical habitat data along the animal's track, depending on the mission objectives. Sensors for our study included a 1,200-kHz up-down looking acoustic Doppler current profiler (ADCP) (Teledyn RDI;

FIGURE 1 | Graphic representation of the TurtleCam system. Continuous, two-way communication between the leatherback turtle's transponder and the vehicle enables the vehicle to locate, follow and film the tagged turtle while simultaneously collecting habitat data along the turtle's path. A camera on the turtle-borne transponder collects video from the turtle's perspective. Turtles were tracked in areas with fixed fishing gear.

www.rdinstruments.com) for altimetry, water current data, and speed over ground measurements, as well as a conductivitytemperature (CT) probe (YSI; www.ysi.com), magnetic heading sensor, pressure sensor, and an environmental characterization optics sensor (ECO puck, SeaBird Scientific; www.seabird.com/ eco-puck) that measured chlorophyll fluorescence and turbidity.

### PILOT STUDY: TRACKING LEATHERBACK SEA TURTLES WITH TURTLECAM

Between September 2016 and September 2017, we tagged, tracked and filmed nine leatherback turtles with the TurtleCam system off Cape Cod, Massachusetts, USA. The full results of this pilot study will be reported elsewhere. Here, we present representative data from a single turtle to demonstrate how the TurtleCam system can be successfully used to simultaneously investigate turtle behavior and habitat characteristics in a gear-dense, highly trafficked, coastal feeding ground.

On 7 October, 2016, we worked with a spotter pilot to locate a subadult leatherback turtle in Nantucket Sound. The turtle was found in an area with fixed fishing gear (traps/pots) intersected by shipping/ferry lanes, with a minimum of eight other leatherbacks. We approached the turtle by boat and tagged it at the surface with the transponder using a pole applicator (**Figure 2**). The transponder was attached to the turtle's carapace with two suction cups, eliminating the need to capture or handle the turtle and potentially impact its behavior (Heaslip et al., 2012). Turtles in our study dove immediately post-tagging, but they appear to quickly resume feeding dives (usually within minutes based on video footage), suggesting that surface application of suction cup tags has a minimal impact on the turtles' natural behavior (Heaslip et al., 2012; Wallace et al., 2015). The total weight of the transponder, camera and attachment materials was less than 1% of the turtle's minimum estimated body weight of 200 kg. We estimated the turtle's carapace length relative to the known length of the tagging vessel, and used published values of mass vs. curved

FIGURE 2 | A researcher prepares to tag a leatherback turtle at the surface with a pole-deployed transponder off Cape Cod, Massachusetts, USA. Photo: Sean P. Whelan, used with permission.

carapace length (James et al., 2005) to estimate a conservative body weight for the turtle.

We tagged the turtle with the transponder at 14:33 GMT and deployed the TurtleCam AUV at 14:39 GMT. We tracked the turtle until 20:29 GMT, recovering the AUV briefly (from 18:27 to 18:45 GMT) to swap out batteries and memory cards in the vehicle cameras. For this mission, the TurtleCam AUV was equipped with an ADCP, CT probe, pressure sensor, and magnetic heading sensor. For a subset of the total mission (16:32 to 20:29), we also deployed a second REMUS-100 AUV "Edgar" that was equipped with a CTD (Model, SBD 49 FastCAT sensor; www.seabird.com), combo flurometer-turbidity (Model, ECO-Triplet; www.seabird.com) ECO puck sensor, up/down 1,200 kHz RDI ADCP, and 900/1,800-kHz dual frequency sidescan sonar (www.marinesonic.com) to sample throughout the entire water column in the vicinity of the tagged turtle. Our mission generated over 5 h of turtle localization data and video, and 2 h of turtle-borne camera footage. The TurtleCam AUV collected over 5 h of biophysical oceanographic data along the turtle's track, while Edgar collected 4 h of environmental data from the surface to the sea floor near the turtle's path (**Figure 3**). At the end of the mission, the transponder tag successfully released from the turtle on acoustic command from the STS and was recovered.

#### Sensor Data

Turtle localization, depth and habitat data were extracted from the AUV using algorithms developed at the WHOI Oceanographic Systems Lab, and sensor data was analyzed in R (R Core Team, 2016). During the 5.5-h tracking period, the turtle's estimated horizontal movements covered approximately 16 km (**Figure 3**). In this shallow, coastal environment, the turtle dove continuously from the surface to the seafloor, occupying a depth range of 0–20 m (mean ± SD = 8.1 ± 4.7 m) (**Figure 4**).

FIGURE 3 | Leatherback turtle track (red line) from beginning (green triangle) to end (red triangle) in Nantucket Sound, Massachusetts, USA in October, 2016. Turtle track reconstructed from the TurtleCam AUV localization data. Track of second AUV "Edgar" (blue line) from beginning (green circle) to end (red circle) near the tracked turtle's path.

The seafloor along the turtle's path was a mix of shoal and channel habitat, with bathymetry ranging from 7.2 to 30.7 m (mean ± SD = 15.5 ± 2.9 m) (**Figure 4**). The water column was wellmixed, with relatively uniform temperature (mean ± SD = 18.4 ± 0.1◦C) and salinity (mean ± SD = 32.07 ± 0.03 psu), and a weak or absent mixed layer (**Figure 5**). We also measured high values of chlorophyll a (mean ± SD = 488.7 ± 58.9 µg/L) and turbidity (mean ± SD = 242.0 ± 24.6 NTU) near the path of the tagged turtle with the second AUV.

#### Video Footage

The video footage from the turtle-borne transponder and AUV cameras was downloaded and backed up on duplicate drives after each mission. The AUV footage from six different cameras was stitched together into a single video mosaic for review (**Figure 6**). High turbidity in our study site resulted in poor visibility (frequently < 1 m), limiting footage of the turtle from the AUV cameras (<1% total time). For this reason, we focused our analysis on the turtle-borne camera footage (**Figure 7**). We built an ethogram and coded the turtle-borne camera footage using the open-source event-logging software BORIS (Friard and Gamba, 2016). This analysis resulted in a detailed time-activity budget and included state events (frequency and duration) and point events (frequency only). We included modifiers for some parameters to capture additional detail from our observations. Parameters in our ethogram can be found in the Time Diagram (**Figure 8**).

During the ∼2 h recording period of the turtle-borne camera, the turtle spent most of its time diving (68%), making 71 dives (**Figure 8**). Consistent with our turtle depth and bathymetry measurements from the AUV, we observed the turtle using the entire water column, spending almost 16% of its time swimming just above the sea floor (**Figure 7A**) and occasionally feeding on jellyfish at or near the bottom. We recorded high feeding rates (over 30 jellyfish per hour), with prey captures consisting mostly (95%) of Atlantic sea nettle (Chrysaora quinquecirrha) (**Figures 7B,C**, **8**). Out of the 78 prey detections recorded, 63 jellyfish were successfully captured and the turtle spent 29% of its time handling jellyfish (**Figure 7C**). The turtle silhouetted its prey 36% of the time by diving to the bottom or just above the bottom, and then looking up toward the surface light to locate prey (**Figure 7D**). The turtle spent less of its time at the surface (15%) and swimming just under the surface (17%) (**Figure 8**). It broke the surface 103 times during the tracking period, and surface time was always associated with respirations (n = 180) (**Figures 7E**, **8**).

We also coded the video data for the turtle's reaction, if any, to the AUV, and presence of non-prey species (**Figure 8**). We recorded 15 apparent reactions to the AUV, which included brief cessation of feeding and movement away (toward the surface if it occurred at depth and toward the bottom if it occurred at the surface), as well as defensive postures (turning its carapace toward the AUV). The first observed reaction was the longest at 3.5 min, while the subsequent reactions were brief (mean = 0.6 min) with the turtle recovering quickly and resuming its diving and feeding behavior (**Figure 8**). This suggests the turtle may have acclimated to the presence of the AUV over time. We recorded the presence of fish, identified mostly as false albacore Euthynnus alletteratus (**Figure 7F**) and jellyfish-commensal larval butterfish Peprilus triacanthus, but this was infrequent (< 4% of the recording time) (**Figure 8**). Although we tracked and filmed the turtle within meters of pot gear, the turbidity and limited visibility precluded collection of underwater footage of the gear.

#### AUV APPLICATIONS

To mitigate threats to animals in the marine environment, it is critical to identify behaviors that exacerbate risk. Characterization of sea turtle behavior is often based on indirect

FIGURE 5 | Temperature (red line) and salinity (blue line) during a TurtleCam AUV mission in Nantucket Sound in October, 2016. Environmental data was sampled along the tagged leatherback turtle's path by the AUV with a conductivity-temperature (CT) probe. The AUV was recovered part-way through the mission (yellow column) to replace camera batteries and memory cards.

mounted on the vehicle top.

measurements and inference, resulting in an incomplete and potentially inaccurate picture of behavior (Seminoff et al., 2006). Direct visual observations are critical to improve and validate interpretation of indirect behavior measurements (Schofield et al., 2006; Seminoff et al., 2006). To correctly interpret habitat-driven behaviors, we also need to concurrently observe and sample habitat during behavior studies. Autonomous underwater vehicles can efficiently meet all of these objectives, resulting in a more holistic picture of marine animal behavior (Packard et al., 2013; Kukulya et al., 2015, 2016; Skomal et al., 2015). The pilot study described here demonstrates proof of concept for using an AUV to study leatherback turtle behavior and habitat in a densely populated, high-risk coastal environment, and it can be easily adapted for other species and habitats with similar conservation concerns.

The "smart" AUV SharkCam has already demonstrated the utility of using versatile autonomous vehicles to study the behavior of large pelagic animals such as great white sharks

(Carcharodon carcharius) (Skomal et al., 2015). Data collected from previous SharkCam studies showed that great white sharks spend the majority of their time swimming in a straight line at a constant speed (Kukulya et al., 2015, 2016; Skomal et al., 2015). Our leatherback turtle study demonstrated that REMUS AUV technology is also capable of making observations of an obligate air breather that dives frequently, is less predictable in its swimming trajectory, and frequently surfaces and hovers. "Smart" AUVs like SharkCam and TurtleCam, which can track a randomly moving target, film it and collect a variety of oceanographic data, offer a revolutionary tool to scientists investigating the subsurface behaviors and habitat of marine megafauna. Future applications for "smart" AUVs include behavior and habitat studies of whales, seals, rays, skates, tuna, and a variety of sea turtles and sharks. As AUV technology improves and tags become miniaturized, the demand and applications for animal-following AUVs will continue to grow and evolve, transforming the ways that scientists can study cryptic marine species.

### SUMMARY AND CONCLUSIONS

We successfully used the TurtleCam system to simultaneously measure and observe leatherback turtle habitat and behavior. Our findings have direct implications for conservation and management of leatherback turtles off Massachusetts and in regions with similar bio-physical oceanographic conditions. The combination of sensor and video data demonstrated that the tagged turtle fed in productive, turbid, and well-mixed habitat associated with shallow (<35 m) shoal and channel bathymetry. In these conditions, the turtle used the entire water column and fed on jellyfish from the seafloor to just under the surface. Jellyfish are known to accumulate around physical gradients (Graham et al., 2001), and the absence of a pycnocline in our well-mixed study site is consistent with jellyfish distribution throughout the water column. In foraging areas with a well-defined pycnocline (e.g., Atlantic Canada), leatherbacks appear to limit their diving to the upper mixed layer, and depth-specific fishing gear modifications may reduce sea turtle entanglement in buoy lines (Hamelin et al., 2014). In the shallow well-mixed habitat off Cape Cod, leatherback turtles are likely to feed throughout the entire water column and can encounter fishing gear anywhere from the surface to the sea floor, making depth-specific fishing gear modifications to reduce sea turtle interactions ineffective. The tagged turtle also exhibited prey-silhouetting behavior, which has been documented in great white sharks (Klimley et al., 1996; Strong, 1996; Skomal et al., 2015), and in leatherback turtles off Atlantic Canada (Wallace et al., 2015). While sharks silhouette their evasive pinniped prey as an ambush tactic, visual predators like leatherbacks may silhouette jellyfish with surface light in response to the murky, turbid conditions in our study area. Prey silhouetting could potentially increase the risk of entanglement in buoy lines of fixed fishing gear if the turtle mistakes a surface buoy or submerged float for their jellyfish prey. While the leatherback was able to correctly identify jellyfish the majority of the time, it did make a close approach to seaweed on three occasions (**Figure 8**). Entanglement risk may be exacerbated by poor visibility coupled with the turtles' primary focus on prey capture/handling.

The tagged turtle only spent 15% of its time at the surface, but the frequency of surfacing (>100 times in ∼2 h) may increase its probability of boat strike. Surface and subsurface swimming (within the top 2 m) accounted for about one third of the turtle's observed behaviors, putting the turtle within easy striking distance of the hull and/or propeller of a range of watercraft (Hazel and Gyuris, 2006). Vessels operating in our study area range from shallow draft (≤1.5 m) recreational vessels to medium draft (3–5 m) passenger ferries, fishing vessels, and yachts (reviewed in https://energy.gov/sites/prod/files/DOE-EIS-0470-Cape\_Wind\_FEIS\_2012.pdf). This is consistent with annual documentation of boat-struck leatherbacks in our study region (Dwyer et al., 2002), though forensic analysis is needed to

identify the primary source(s) of vessel-related mortalities (e.g., Rommel et al., 2007). Interestingly, the tagged turtle did not appear to show an overt behavioral response to the presence (visual or acoustic) of the tracking boat or other vessels in the area, though this assessment was subjective. To quantify behavioral response to sound, future deployments should include an acoustic tag (e.g., Tyson et al., 2017) and incorporate behavioral response studies. More research is needed on the behavioral response of sea turtles to anthropogenic underwater sounds, especially field studies that measure behavioral response experimentally as has been done for marine mammals (Southall et al., 2016).

Data from TurtleCam can also be used to improve energy budget calculations for leatherback turtles and to help define fine-scale foraging habitat requirements. Video imagery can be used to quantify flipper beat frequency, or stroke rate (Reina et al., 2005), which can be a proxy of energy expenditure in some marine animals (Williams et al., 2004; Jeanniarddu-Dot et al., 2016). Stroke rate data can be combined with video-derived time-activity budgets to estimate activity-specific energy budgets for leatherbacks. In addition to stroke rate, TurtleCam also monitored feeding behavior, including prey composition, consumption rate, and abundance, key parameters in bioenergetics models (Chipps and Wahl, 2008). TurtleCam continuously collected imagery and sensor information on the turtle's immediate environment, including both biotic (chlorophyll a, turbidity) and abiotic (bathymetry, temperature, salinity) properties, allowing us to identify and characterize leatherback habitat requirements within a coastal feeding ground.

Limitations of the current TurtleCam system include relatively short tracking durations (≤8 h), the size of the transponder, and the potential to affect the focal animal's behavior. Larger AUVs (e.g., the REMUS-600; depth-rated to 600 m, 211 × 32 cm, 235 kg in air) have greater battery power and can carry more instruments and lights for overnight deployments, but the larger vehicle size increases the costs associated with deployment and operation. The REMUS-100 can be deployed by two individuals from the side of a small vessel, making it a versatile and cost-effective choice for field studies with limited budgets. The current transponder size (7.6 × 38 cm, 1.7 kg in air, ∼58 g buoyancy in seawater) limits its application to relatively large marine animals that can handle the added weight and drag of the instrument. The size and associated drag of the transponder may also impact swimming and diving behavior (Jones et al., 2013). The WHOI Oceanographic Systems Laboratory is in the process of re-engineering the transponder to be more compact and hydrodynamic (∼60% reduction in size and weight), reducing potential impact to natural behaviors and increasing its suitability for other species. Future improvements to the transponder also include incorporating scientific sensors and onboard data storage.

In our study, the attachment of the transponder via suction cups did not appear to have any measurable impact on subsequent behavior, but for species that require handling for tag attachment, there may be a "capture" stress response that can last for several hours (Heaslip et al., 2012; Thomson and Heithaus, 2014). The video footage allowed us to objectively measure the turtle's response to the presence of the AUV. Given the turbidity and limited visibility in our study site, we used a follow distance of 1–2 m between the turtle and the AUV to maximize footage of the turtle. This small distance likely resulted in heightened turtle response to the presence of the AUV, with the turtle appearing to react to the AUV during close follows by modifying its behavior to avoid it by swimming away and/or presenting a defensive posture (turning its carapace toward it). However, the turtle appeared to recover quickly from these events and resume normal diving and feeding behavior. In less turbid conditions with better visibility, a larger follow distance can be maintained and would likely result in fewer reactions from the turtle.

The TurtleCam AUV system is a unique platform that enables researchers to directly observe and study the subsurface behaviors and habitats of marine megafauna. It is a highly versatile platform that can be customized to meet different study objectives through its range of sensors and adaptability for different animal behaviors (e.g., high frequency diving vs. horizontal swimming). In densely populated coastal habitats where endangered and threatened species overlap with multiple anthropogenic threats, understanding animal behavior and habitat is critical to designing and implementing effective mitigation strategies. Our study demonstrates that an AUV can successfully track and image leatherback sea turtles feeding in a coastal environment, resulting in new observations of threedimensional subsurface behaviors and habitat use, with direct implications for sea turtle management and conservation.

#### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the National Marine Fisheries Service Endangered Species Act Section 10 Permit no. 15672-02. The protocol was approved by the Woods Hole Oceanographic Institution Institutional Animal Care and Use Committee (IACUC) no. 21668.

#### REFERENCES


### AUTHOR CONTRIBUTIONS

KD and AK conceived and designed the study. KD, AK, EB, and MB contributed to obtaining funding, the acquisition, analysis, or interpretation of the data, and to drafts and revisions of the manuscript.

### FUNDING

This research was funded by National Oceanic and Atmospheric Administration Grant #NA16NMF4720074 to the Massachusetts Division of Marine Fisheries under the Species Recovery Grants to States program. Additional funding was provided by Jean Tempel, Hydroid Inc., and over 100 Project WHOI donors.

#### ACKNOWLEDGMENTS

For assistance in the field, we thank George Purmont, Michael Dodge, Sean Whelan, Ken Kostel, and Daniel Cojanu. We thank Roger Stokey and the Oceanographic Systems Laboratory for providing REMUS AUV support. We are grateful to Karen Moore Dourdeville and Mass Audubon Wellfleet Bay Wildlife Sanctuary for providing real-time sea turtle sightings information during our field seasons (www.seaturtlesightings.org). We thank Jack Cook and Natalie Renier (WHOI Graphic Services) for assistance with **Figure 1**. The authors acknowledge the use of SEATURTLE.ORG's Maptool program (www.seaturtle.org/ maptool/). Special thanks to Jean Tempel, Hydroid, Inc., and over 100 Project WHOI donors who provided the seed funding for this work. We thank Larisa Avens and Sabrina Fossette for helpful comments that improved an earlier draft of this manuscript.

VA: U.S. Department of the Interior, Bureau of Ocean Energy Management, Headquarters, OCS Study BOEM 2012-01156, 35.


**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 Dodge, Kukulya, Burke and Baumgartner. 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 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.

# Environmental DNA (eDNA) From the Wake of the Whales: Droplet Digital PCR for Detection and Species Identification

#### C. Scott Baker <sup>1</sup> \*, Debbie Steel <sup>1</sup> , Sharon Nieukirk <sup>1</sup> and Holger Klinck <sup>2</sup>

*<sup>1</sup> Hatfield Marine Science Center, Oregon State University, Newport, OR, United States, <sup>2</sup> Bioacoustics Research Program, Cornell Lab of Ornithology, Cornell University, Ithaca, NY, United States*

Genetic sampling for identification of species, subspecies or stock of whales, dolphins and porpoises at sea remains challenging. Most samples have been collected with some form of a biopsy dart requiring a close approach of a vessel while the individual is at the surface. Here we have adopted droplet digital (dd)PCR technology for detection and species identification of cetaceans using environmental (e)DNA collected from seawater. We conducted a series of eDNA sampling experiments during 25 encounters with killer whales, *Orcinus orca,* in Puget Sound (the Salish Sea). The regular habits of killer whales in these inshore waters allowed us to locate pods and collect seawater, at an initial distance of 200 m and at 15-min intervals, for up to 2 h after the passage of the whales. To optimize detection, we designed a set of oligonucleotide primers and probes to target short fragments of the mitochondrial (mt)DNA control region, with a focus on identification of known killer whale ecotypes. We confirmed the potential to detect eDNA in the wake of the whales for up to 2 h, despite movement of the water mass by several kilometers due to tidal currents. Re-amplification and sequencing of the eDNA barcode confirmed that the ddPCR detection included the "southern resident community" of killer whales, consistent with the calls from hydrophone recordings and visual observations.

Keywords: ddPCR, DNA barcoding, taxonomic, Killer whale, eDNA, mtDNA

## INTRODUCTION

Non-lethal genetic sampling for identification of whales, dolphins, and porpoises (cetaceans) at sea remains challenging. Most samples have been collected with some form of a biopsy dart projected with a crossbow (Lambertsen, 1987) or a modified veterinary capture rifle (Krützen et al., 2002). This requires a close approach of a vessel, usually within 10–20 m, while the individual is at the surface. It is also limiting because of access, distribution, or behavior of cetaceans. Some species are rare, cryptic, or both, e.g., beaked whales (Dalebout et al., 2004). Others species are difficult to approach because of their elusive behavior, e.g., the pygmy and dwarf sperm whale. Finally, some species are considered sensitive to disturbance from the close approach of a vessel or the biopsy sample itself (Noren and Mocklin, 2011).

Advances in analyses of environmental (e)DNA now offer an alternative for detection and identification of rare, cryptic, or vulnerable cetacean species. Here the DNA that is shed or excreted from individuals during normal activity can be collected from the environment, concentrated, and

#### Edited by:

*Peter H. Dutton, National Oceanic and Atmospheric Administration (NOAA), United States*

#### Reviewed by:

*Kirsten Jennifer Harper, Southwest Fisheries Science Center (NOAA), United States Stefano Mariani, University of Salford, United Kingdom*

> \*Correspondence: *C. Scott Baker scott.baker@oregonstate.edu*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *16 December 2017* Accepted: *04 April 2018* Published: *19 April 2018*

#### Citation:

*Baker CS, Steel D, Nieukirk S and Klinck H (2018) Environmental DNA (eDNA) From the Wake of the Whales: Droplet Digital PCR for Detection and Species Identification. Front. Mar. Sci. 5:133. doi: 10.3389/fmars.2018.00133* amplified via the Polymerase Chain Reaction (PCR) using primers targeted for specific taxonomic groups. eDNA has been used widely to detect vertebrate species in freshwater systems (Ficetola et al., 2008; Jerde et al., 2011; Ma et al., 2016; Stewart et al., 2017), and is now finding a growing number of applications in the marine environment (Thomsen et al., 2012), including detection and identification of marine megafauna (Foote et al., 2012; Port et al., 2016; Sigsgaard et al., 2016; Andruszkiewicz et al., 2017; Bakker et al., 2017; Gargan et al., 2017). Whales, dolphins and porpoises represent good candidates for eDNA sampling given their known tendency to release cellular DNA in shed skin (Amos et al., 1992), fecal plumes (Parsons, 1999), and the "spout" or blow (Hunt et al., 2013).

The methodology for eDNA sampling is advancing rapidly as the number and range of application increases. One of these advances is droplet-digital (dd)PCR. ddPCR is a powerful new technology for quantifying low levels of DNA by fractionating a PCR reaction into more than 20,000 droplets using an oil emulsion (Doi et al., 2015). Amplification of the target DNA is quantified by incorporating a fluorescent dye directly into the amplicon reaction or into a molecular probe designed to target a specific sequence bracketed by the PCR primers. The targetpositive and target-negative droplets are individually counted by passing them in a single stream through a fluorescence detector similar to a flow cytometer. The ratio of the target-positive to the target-negative droplets is used to estimate the number of copies of the target DNA in the sample, under the assumptions that the target molecules are distributed among the 20,000 droplets according to a Poisson function. Thus, unlike conventional qPCR, ddPCR allows for direct quantification without the need for standard curves, eliminating the variance associated with creating and running standards with each batch (Cao et al., 2016).

Here, we have investigated the potential for ddPCR to detect eDNA from seawater collected following the passage of killer whales, Orcinus orca, in Puget Sound (the Salish Sea). For this, we took advantage of methods previously developed for species identification of cetacean products sold in Japanese and Korea markets (e.g., Baker and Palumbi, 1994; Baker et al., 2006), including a comprehensive reference database of mitochondrial (mt)DNA sequences from most recognized species of cetaceans (Ross et al., 2003; Dalebout et al., 2004). From these reference sequences, we designed primers for short fragments of the mtDNA, referred to as "mini-barcodes," to target killer whales and improve amplification of degraded DNA. We chose killer whales in Puget Sound for this initial investigation because their well-described habits allowed us to locate and sample individuals or groups efficiently (Hauser et al., 2007). Additionally, the local distribution of killer whales includes multiple "communities" and ecotypes, identifiable by distinct vocalizations (Ford, 1991) and distinct mtDNA haplotypes (Parsons et al., 2013). This overlapping distribution includes the critically endangered "southern resident" community of killer whales (Hauser et al., 2007). Our sampling design was intended to quantify eDNA from the "wake of the whales," i.e., from directed sampling assisted by visual or acoustic localization. However, the result of our serial sampling also provide some insight into the potential for spatial sampling of eDNA for the purposes of describing habitat use or estimating "occupancy" in the marine environment.

### MATERIALS AND METHODS

### Field Surveys

Killer whales were located in the waters around the San Juan Islands during small-boat surveys operating out of the Friday Harbor Laboratory (FHL), during August and September 2015 (**Figure 1**). At each encounter with killer whales, the boat was moved into position 200 m behind the whales (to comply with local whale watching regulations, NOAA, 2011). A Lagrangian drifter (MicrostarTM GPS drifter, Pacific Gyre Inc., Oceanside, CA, USA) was launched to maintain position in the water mass after the whales passed. The drifter tracked currents at 1 m beneath the surface and also recorded the sea surface temperature. Location and temperature data was stored in 30 s intervals on the instrument and downloaded after recovery. A Zoom H4n ProTM (Zoom North America Inc., Hauppauge, NY, USA) handheld recorder and an HTI-96MINTM (High Tech Inc., Long Beach, MS, USA) hydrophone were used to monitor vocalizations and, in some cases, to confirm the dialect of the southern resident pods.

After initial positioning, seawater samples were collected from the surface or sub-surface in a 1L, wide mouth, sterile NalgeneTM bottle. Sub-surface samples were collected from ∼50 to 80 cm below the surface using a PVC bilge pump. Surface collections were made at the air/surface interface. All single and serial samples were collected in pairs, one from the starboard and one from the port side of the boat, for a total of 2 L for each sample. Serial samples within an encounter were conducted at 15-min intervals for up to 2 h following the passage of the whales. Note that other events in the field were recorded as "encounters" and that encounter numbers for collection of eDNA are not necessarily consecutive.

### Seawater Filtration and eDNA Extraction

The seawater samples were stored on ice on the boat and returned each evening to the laboratory. Depending on the flow rate, either 1 or 2 L of seawater were filtered through a 0.4 micron, Whatman Cyclopore polycarbonate membrane (GE Lifesciences, USA) using a portable Nalgene filter unit and low pressure vacuum pump at FHL The filters were stored in 1–2 ml of Longmire's solution (Wegleitner et al., 2015) on ice for transport back to our home laboratory. To avoid contamination in the field, we operated in a wet laboratory that had never been exposed to cetacean samples. We decontaminated sample bottles and filter units by soaking overnight in 10% diluted bleach, followed by rinsing with tap water. To test for crosscontamination during sample processing, we also filtered a "negative control" by replacing the seawater sample with tap water that was subsequently extracted and amplified.

The eDNA was extracted from the filters by conventional phenol/chloroform (PCI) methods (Renshaw et al., 2015). The volume or each extraction was adjusted to represent 1 L of seawater and the extracted DNA was re-suspended in 50 µL

FIGURE 1 | The location of 25 encounters with killer whales in the San Juan Islands during August and September 2015. The red dots show the location of the first eDNA sample, after positioning the vessel at a distance of 200 m behind the passage of the whales. The yellow dots show the location of two "no-whale" controls. Note, encounter numbers for collection of eDNA are sequential but not necessarily consecutive.

of TE. To avoid any contamination in the laboratory, all extractions were conducted in a "clean" room that has never been exposed to post-PCR products. We followed other standard protocols for preventing contamination including use of filtered tips, extraction blanks, and PCR blanks. Initial experiments indicated the presence of PCR inhibitors, a common problem in the extraction of eDNA (Jane et al., 2015). Inhibition was reduced (although probably not eliminated) by diluting the extracted eDNA in a 1:1 volume of laboratory grade water. As an additional control for contamination, we chose the southern resident killer whales for our study because no biological sample of this ecotype has ever been processed in our laboratory.

### Primer and Molecular Probe Design

To optimize the detection and identification of eDNA from killer whales, we designed a set of PCR primers using available reference sequences for the mtDNA control region of the known killer whale ecotypes in the North Pacific (**Figure 3**). Many of these primers were slight modifications of primers used routinely for amplification of the cetacean control region and species identification, e.g., dlp4 and dlp5 (Dalebout et al., 2004; Baker et al., 2006). The objective was to design "mini-barcodes" that would amplify degraded eDNA but provide sufficient sequence information for identification of species and ecotypes. The primers Oordlp5Rleft and dlp8G amplified a fragment of 246 base pairs (bp) in length and the primers Oordlp6.5F and dlp8G amplified a fragment of 139 bp in length. These two primer pairs were used for the ddPCR with the EvaGreenTM fluorescent dye (see below). The sequence of the Oordlp6.5F primer was also used to synthesize a molecular probe (FAM labeled custom TaqMan <sup>R</sup> ) for use with the bracketing primers, Oordlp5Rleft and dlp8G. The sequences of primers are included in Supplementary Material (SupMat Table 1).

## Digital Droplet (dd)PCR

The eDNA of killer whales was detected and quantified using a Bio-Rad QX200TM AutoDGTM Droplet DigitalTM PCR System at the Center for Genome Research and Biocomputing at Oregon State University. The system includes an Automated Droplet Generator that generates thousands of droplets from one ddPCR reaction containing the target DNA and a Droplet Reader for detecting the target fluorescence. Amplification of the target DNA was quantified by incorporating a fluorescent dye into the PCR reaction using QX200 ddPCR EvaGreen SupermixTM, or into a TaqMan molecular probe designed to target a specific sequence bracketed by the PCR primers, using ddPCR Supermix for ProbesTM. The target-positive and target-negative droplets are visualized and analyzed using the manufacturer's software, QuantiSoftTM. Quantification of target DNA, in copies/µL of reaction, is based on an assumption of a Poisson distribution of the target DNA among the more than 20,000 droplets from a typical 20 µL reaction (Miotke et al., 2015).

The optimal primer annealing temperature for ddPCR was determined using a gradient PCR. The gradient was run in 2◦C increments from 50 to 60◦C for the EvaGreen assays and from 54 to 64◦C for the probe assay. For all assays 56◦C was found to be the best annealing temperature, with the highest separation between positive and negative droplets. The final thermocycling profile for all assays was as per manufacturer's recommended protocols with the annealing/extensions step adjusted to 56◦C. PCR mastermixs were made in a final volume of 22 µL under the following conditions; for EvaGreen assays, 1x supermix, 100 nM each primer and 5 µL DNA as described below; and for the Probe assay, 1x supermix, 900 nM each primer, 250 nM TaqMan Probe and 5 µL DNA as described below.

All samples were run in duplicate or triplicate, with negative controls (no-template controls) and positive controls included in each ddPCR run. Two of the replicates were used for quantification with the droplet reader and a third was used for reamplification by conventional PCR and Sanger sequencing (see below). All values are expressed as the average of at least two runs, unless otherwise stated. Each experimental sample included 0.5 µL of the 50 µL re-suspended eDNA, using 5 µL of a 1/10 dilution series to reduce pipetting error. This volume of the extracted eDNA was chosen after initial "quenching experiments" showed evidence of inhibitors in conventional PCR experiments and initial ddPCR reactions showed evidence of droplet "rain" (Witte et al., 2016). The positive control was 1 µL of a 1/1,000 dilution of total cellular DNA extracted from a skin biopsy sample of a Hector's dolphin, Cephalorhynchus hectori hectori (Hamner et al., 2012). Based on Qubit Fluorometric Quantification, the concentration of genomic DNA in the positive control was 60 ng/µL, resulting in a mass of 60 pg/reaction after the 1/1,000 dilution. The Hector's dolphin was chosen because it is endemic to New Zealand and, thus, recognizably distinct by DNA barcoding from the cetacean community of the Salish Sea.

#### Re-Amplification and Sanger Sequencing

To confirm the species or ecotypes, we used conventional PCR to re-amplify the target amplicon from the ddPCR reaction after adding 20 µL of TE and breaking the oil emulsion, following manufacturers guidelines (Bio-Rad\_Laboratories\_Inc., 2014). Of the ∼35 µL recovered from breaking the emulsion, 2 µL was added to subsequent conventional PCR reactions under the conditions detailed below. In addition, potential positive samples indicated from ddPCR runs were directly amplified by one or two rounds of conventional PCR under the following reaction conditions; 1x buffer, 2.5 mM MgCl, 0.4µM each primer, 0.2 mM dNTP, 0.25 Units of Platinum TaqTM (ThermoFisher Scientific, USA) and 5 µL of 1/10 dilution of eDNA in a final volume of 20 µL. Six different primer pair combinations were used in various experiments (**Figure 2**) and all were run with the following thermocycling profile; initial denaturation at 95◦C for 3 min followed by 35 cycles of denaturation at 95◦C for 30 s, annealing at 56◦C for 30 s, extension at 72◦C for 60 s, and a final extension step at 72◦C for 5 min. The amplicons were cleaned with Agencourt AMPure XP (Beckman Coulter, USA) and sequenced in both directions with Big Dye terminator chemistry (ThermoFisher Scientific, USA) following manufacturers protocols. Excess dye-terminators were removed with Agencourt CleanSEQ beads (Beckman Coulter, USA) prior to running on an ABI3730xl. Sequences were aligned to known killer whale haplotypes (Parsons et al., 2013) and visually inspected with the software Sequencher 4.1 (Gene Code).

## RESULTS

### Seawater Sampling and eDNA Extraction

We collected seawater from the vicinity of killer whales on 17 encounters during the August field effort and on 8 encounters during the September (**Figure 1**). Group size ranged from 1 to 18, as judged by visual counts (SupMat Table 2). Southern resident killer whales were acoustically identified during some of the encounters (see SupMat Figure 1). The hand-held bilge pump was used for sub-surface sample collection during the 17 encounters in August. Based on this initial experience and a review of the literature (e.g., Moyer et al., 2014) we then changed to collecting samples from the air/surface interface for the eight encounters in September. From the 25 encounters, we collected, filtered and extracted DNA from 71 paired, 1 L samples of seawater from the passage of the whales. There was considerable variation in the number of samples collected during each encounter due to weather, tidal currents and the activity of other vessels. Of the 25 encounter, 11 were represented by a single point sample, 3 encounters by 2 serial samples, 4 by 3 serial samples, 4 by five serial samples and 2 by 9 serial samples (2 h total). Note that encounters were also recorded for other events in the field and that encounter numbers for collection of eDNA are not necessarily consecutive (see **Figure 5**). The well-recorded movements and predictable habits of the killer whales around the San Juan Islands also allowed us to collect two samples at locations where the whales had not been reported on the previous day ("no-whale" encounters).

#### Primer Design and Assay Sensitivity

Combinations of primers Oordlp5Rleft, Oordlp6.5F, and dlp8G tested positive against a "zoo-blot" of samples representing the family Delphinidae (the dolphins and "blackfish," including the killer whales) using conventional PCR and sequencing. Assay sensitivity was then assessed by multiple runs of the positive control using the ddPCR. These showed good precision in paired replicate samples but an ∼2-fold difference in sensitivity, apparently due to length of amplicon and method of incorporating the fluorescent (**Figure 3**). The most sensitive protocol was the EvaGreen incorporation into the shorter amplicon with the Oordlp6.5F to dlp8G primers (139 bp), providing an estimate of 47.75 copies/µl. The least sensitive assay was the molecular probe, nested within the longer fragment of the Oordlp5Rleft to dlp8G primers (246 bp). The EvaGreen incorporation with the Oordlp5Rleft to dlp8G primers was intermediate in sensitivity. The primer pair Oordlp2 and Oordlp4 was designed specifically for killer whales and to include sites considered diagnostic for the southern resident community (see **Figure 7**). Although we repeated the ddPCR experiments with various combinations of primers and probes, we focus here on the most comparable assays using EvaGreen incorporation with the primers Oordlp6.5F to dlp8G.

#### ddPCR Limits of Detection

To assess the relative sensitivity of the ddPCR, we conducted a serial dilution of the positive control using the EvaGreen

FIGURE 2 | Locations of PCR primers and a molecular probe in the mtDNA control region of the killer whale. The probe assay amplified a 246 bp fragment from Oordlp5Rleft to dlp8G, using Oordlp6.5F as the probe. The EvaGreen assay amplified a 139 bp fragment from Oordlp6.5F to dlp8G. \*Primer pairs used for conventional PCR.

incorporation with the Oordlp6.5F and dlp8G primer pair. The serial dilution started with the extracted genomic DNA of the Hector's dolphin (60 ng/µL), at the initial dilution of 1/1,000 used for the positive control in the experimental runs (i.e., a mass of 60 pg). This initial dilution was the starting point for an 8-fold series of 2x dilutions (e.g., 1/2,000 to 1/256,0000) and a negative control. The series included 6 replicates of each dilution. At the 1/1,000 dilution, the estimated concentration was 44.2 copies/µL (SE = 2.06), consistent with values for the positive control in the experimental runs. The decline in the estimates of the dilution series did not show a linear relationship but, instead, showed a stepwise decline (**Figure 4**). At a dilution of 1/32,000, the estimated concentration was 0.51 copies/µl (SE = 0.288). The lower limit of the detection series was 0.175 copies/µl (SE = 0.057), at a dilution of 1/128,000. At a dilution of 1/256,000, there was no detectable DNA. The negative control, however, included non-zero detections (0.05 copies/µL, SE = 0.024), presumably due to measurement artifacts (Hunter et al., 2017).

Relating the 1/128,000 dilution to the estimated mass of the initial positive control (60 ng), suggested a lower end of detection of 0.047 pg of total cellular DNA in a ddPCR reaction (20 µL). At this lower end of detection, however, the lower 95% confidence limits overlapped with the maximum value of the negative controls in the experimental runs (0.12 copies/µL, see SupMat Figure 1). For this reason, we considered samples with > 0.5 copies/µL to have met a strict threshold for a detection. We considered samples that exceeded the average of the negative controls in the experimental runs, > 0.12 copies/µL to have met a relaxed threshold for a detections.

#### Killer Whale eDNA Quantification by ddPCR

We used ddPCR to quantify the copy number of eDNA from 71 samples of seawater collected during 25 encounters after the passage of killer whales and 2 "no-whale" samples, with a series of experimental controls (positive and negative). The results of the 71 individual samples showed considerable variation in detection of eDNA from the killer whales (SupMat Figure 2). The average concentration of eDNA for all 71 experimental samples was 4.08 copies/µL (SE = 2.31) but the distribution of values was non-parametric (Kolmogorov-Smirnov test, p < 0.01), with zero values and a few very high values. In one encounter (#e46), three samples yielded estimates of eDNA greater than the positive control (> 60 copies/µL). If these three samples were excluded, the mean declined to 0.44 copies/µL (SE = 0.14). Using the strict threshold of > 0.5 copies/µL derived from the dilution series, 21 of these samples were likely detections. Using the maximum for the negative controls as the lower threshold (0.12 copies/µL), another 35 samples were "relaxed detections," for a total of 56 samples with strict or relaxed detection. The values for one of the "no-whale" samples fell below the threshold for relaxed detection (0.035 copies/µL) but the other exceeded the relaxed threshold (0.18 copies/µL).

Given the likely serial dependency of samples within a series (see below), we considered it more informative to judge the probability of detection on the basis of an encounter, rather than individual samples (**Figure 5**). Of the 25 encounters, 17 included one or more samples with a concentration of eDNA exceeding 0.12 copies/µL and 10 of these exceeded 0.5 copies/µL. We noted a positive relationship between the number of samples in a series and one or more positive samples, using either threshold for

FIGURE 5 | A sequential stack plot of the number of samples (for each of the 25 encounters) judged to be positive or negative for detection of eDNA under two different thresholds. Each block in the stack represents a sample: dark green are samples with an average of >0.5 copies/µl (strict detection); light green are greater than the maximum of the negative controls, >0.12 copies/µl, but <0.5 copies/µl (relaxed detection). Samples shown in black are considered "no detection," with <0.12 copies/µl. Samples in gray are missing data. Stars indicate samples that provided mtDNA sequences after amplification of eDNA or re-amplification from ddPCR reactions. The blue bracket indicates the encounters during which samples were collected from the air/surface interface.

detection (Spearman's rank order correlation coefficient, R = 0.61 p = 0.001 for 0.12 copies/µL; R = 0.59, p < 0.002 for 0.5 copies/µL). We also noted a positive relationship between pod size and one or more positive samples in the encounter series, using either threshold for detection (Spearman's rank order correlation coefficient, R = 0.53 p = 0.006 for 0.12 copies/µL; R = 0.62, p < 0.001 for 0.5 copies/µL). Finally, we noted a greater detection probability for the eight encounters with sampling from the air/surface interface (encounters #e27 to #e47) compared to the 17 with sub-surface sampling, using either threshold of detection (Fisher's exact test, p = 0.026 for >0.12 copies/µL; p = 0.028 for >0.5 copies/µL).

#### Serial Detection of the eDNA "Wake"

We found evidence of considerable persistence of detectable eDNA in the wake of the whales (**Figure 5**). Six of the encounters showed a positive detection (relaxed threshold) 60 min after the initial sample, two of these showed a positive detection after 90 min and one after 120 min (encounter #e08; see SupMat Figure 3). Although encounters with multiple samples had a greater probability of at least one detection, compared to single sample encounters (see above), there was no clear relationship between the order of serial samples and the quantification of eDNA (**Figure 5**). Of the 25 total encounters, 14 included 2 or more serial samples, 12 included three or more serial samples and 8 included 4 or more serial samples (**Figure 5**). Using a Wilcoxon signed-rank test (two tailed), we found no significant differences in the copies/µL of the first and second sample (n = 14; p = 0.133), the first and third sample (n = 12; p = 0.659) or the first and fourth sample (n = 9; p > 0.05).

However, we did find considerable stochastic variation in the copies/µL of serial samples, as illustrated by the case history of encounter #e11 (**Figure 6**). During this encounter, a pod of 12–18 killer whales was traveling steadily east after we initially positioned ourselves ∼200 m behind the pod. Over the next 60 min we collected samples at 15-min intervals (samples #s45 to #s49) and drifted ∼4 km, with the tidal current. The detection of eDNA varied from an average of 3.5 copies/µL in the first sample (#s45) to a low of 0.14 copies/µL in the second sample but then increased again in samples #s47 and #s48. One hour after the passage of the whales, there was an average of 0.34 copies/µL of eDNA in the subsurface water sample. All five samples exceeded the relaxed threshold of 0.12 copies/µL and three samples exceeded the strict threshold of 0.5 copies/µL. Three of the serial samples was re-amplified and sequenced to confirm the species (see **Figure 5**).

#### Species and Ecotype Identification by eDNA Sequencing

After breaking the emulsion of the triplicate ddPCR reaction, we were able to re-amplify and sequence a targeted fragment of mtDNA from 11 samples representing six encounters (see **Figure 5**). We note that 5 of these encounters had samples that exceeded the strict detection threshold (i.e., >0.5 copy/µl). Two of these samples (#s47 and #s48) are represented in the results for encounter 11 shown in **Figure 5**. One encounter (#e25) included samples that exceeded the relaxed threshold but not the strict detection threshold.

Although the length of these mini-barcodes was not sufficient to distinguish among the different ecotypes, they were sufficient to confirm the source of the eDNA was killer whale. From encounter #e46, however, there were three serial samples (#s94 to #s96) of sufficient quality and quantity to sequence over 700 bp of the mtDNA control region directly from the extracted eDNA (i.e., without an initial amplification by ddPCR). Using available reference sequences to define diagnostic sites for known ecotypes and communities in the North Pacific (Parsons et al., 2013), this sequence length allowed us to confirm that the encounter was a pod of the southern resident community (**Figure 7**). For a second encounter (#e25) we were able to sequence two fragments to confirm a southern resident haplotype after initial amplification of the ddPCR reaction. As we have never held tissue samples

FIGURE 6 | The location of samples collected during encounter #e11 with killer whales on August 12, 2015 and the result of the ddPCR quantification of eDNA from serial samples. (Left) Location of five serial samples (#s45 to #s49) as determined by the GPS of the Lagrangian drifter deployed initially 200 m after the passage of the whales. (Right) Visualization and analysis of five serial samples using the software QuantiSoft. The estimated copies/µL of eDNA from the killer whales measured by the number of target-positive droplets (shown in blue) above the baseline of target-negative droplets (shown as the dark cloud). The purple line shows the upper threshold of the target-negative droplets calculated from the negative controls. The black bars indicate the pairs of replicate samples, separated by the dashed yellow line. The calculated copies/µL are shown above the black brackets for each of the replicate samples. Two of the serial samples was re-amplified and sequenced to confirm the species (see text).

or DNA for this ecotype, we can exclude the potential for any contamination from other sources in our laboratory.

### DISCUSSION

#### Species and Ecotype Identification

We have demonstrated the potential for ddPCR to detect eDNA in samples of seawater after the passage of killer whales in inland waters. Using a relaxed threshold for false positives, we were able to detect eDNA of killer whales in 17 of the 25 encounters (68%). Using a strict threshold, we detected killer whales in 10 of these encounters (40% of the total). By re-amplification of the ddPCR amplicon and conventional sequencing, we were able to confirm that our eDNA detections were species specific and for, two encounters, that the sequences matched the mtDNA haplotype of the southern resident killer whales. Unlike direct sampling methods (e.g., biopsy sampling), our eDNA sampling imposed no disturbance to the whales and did not require an approach within the 200-yard limitation (182 m) of current vessel-approach regulations for killer whales (NOAA, 2011).

#### Limits of Detection

The choice of a threshold for a positive detection will depend on the tolerance of the research objectives for false positives or false negatives, as well as the precision of the measurements. Here our objective was to investigate the sensitivity of ddPCR for detecting and identifying our target species with a directed sampling design, rather than to estimate presence or absence of some unknown occupancy. For our objective, the strict threshold of >0.5 copies/µL was well supported by the limits of detection analysis and the negative controls of each run. Our choice of a relaxed threshold was more challenging. The value of 0.12 copies/µL was consistent with the lower limits of detection analysis and with the maximum of the negative controls in the experimental run but overlapped with the standard error of these negative controls and with other published estimates for limits of detection in ddPCR (e.g., 0.13 copies/µL, 95% CL 0.08–0.21; (Hunter et al., 2017). Although we were able to confirm some of the relaxed detections by re-amplification and sequencing of the mtDNA, a larger sample size and further experimentation with the ddPCR dynamics will be necessary to establish a more robust detection probability for occupancy modeling. This could include a standard dilution series for each plate in range of the assay expectations, as recommended by (Hunter et al., 2017). Limits of detection could also be improved by further technical development to improve the signal to noise ratio for the negative control and by reducing the inhibitors that tend to accumulate with increasing sample concentration (Williams et al., 2017).

#### Variation in Detection

As could be expected from the dynamics of sampling in the marine environment, there was considerable variation in detectable concentrations of eDNA from the killer whales. Although our sample size was not sufficient for multivariate analyses, we found a significant positive relationship for a detection (relaxed or strict) with the number of samples in an encounter, the pod size of the encounter and the collection of samples at the air/surface interface. Sampling at the air/surface

interface compared to the subsurface sampling was a particularly influential variable. All of the 23 samples collected from the air/surface interface during the eight encounters in September showed a positive detection compared to 69% of the 48 subsurface samples from the 17 encounters in August. Surface sampling during one of these encounters yielded exceptionally high concentrations of eDNA, sufficient for sequencing nearly full-length fragments of the mtDNA control region. Likely contributors to this "surface effect" include advection of sloughed skin or feces, and the retention, by surface tension, of cellular DNA expelled from the spout or blow. Of these, an unobserved fecal plume would seem to be the most likely explanation for an anomalously high sample of eDNA. Although it is often assumed that whale feces are buoyant (Roman and McCarthy, 2011), this is likely to vary considerably by the prey type and the species of the consumer. Deployment of an Unmanned Aerial Vehicle (UAV) prior to eDNA sampling could provide evidence of the presence or absence of a fecal plume and species-specific differences in the persistence of a plume (Wolinsky, 2017).

Although the number of samples in an encounter was related to a positive detection, there was no significant difference in the concentration of eDNA (copies/µL) by order of serial sampling. This apparent absence of a sequential dilution effect is puzzling. Our best interpretation is that the fine-scale spatial distribution and vertical stratification of eDNA result in considerable variation from sample to sample within an encounter. This heterogeneity is greater than the potential homogenization of a dilution effect across the limited time span of the serial samples.

#### Persistence of Detection

The ability to detect eDNA persisted for at least an hour after the passage of the whales in six encounters and, in one of these encounters, for up to 2 h, despite movement of the water mass by more than 4 km due to tidal currents. The positive detection of one "no-whale" control suggests amplifiable DNA might persist longer under some conditions (e.g., Piaggio et al., 2014). This persistence bodes well for targeted sampling of elusive species like beaked whales. These species are very longduration divers (e.g., Baird et al., 2006; Schorr et al., 2014) and difficult to approach for biopsy sampling during their limited time at the surface. However, with the assistance of acoustic localization and visual cues, it should be possible to sample eDNA from the approximate location of a dive, substantially increasing the efficiency of genetic sampling. Including localized spatial sampling with serial sampling could help in describing both the extent and persistence of an eDNA "wake" after the terminal dive. Although basic species identification remains the priority for beaked whales (e.g., Dalebout et al., 2014), species-specific primers could also be designed to identify intra-specific variation for stock identification of widely distributed species like Cuvier's beaked whales (Dalebout et al., 2005).

### CONCLUSION

The development of ddPCR, with the incremental advances in primer design, seawater filtration, eDNA extraction, and sequencing, as evidenced here, provide a powerful new methodology for detection and identification of cetacean species, even those that are not easily identified by morphological or acoustic characteristics. If successful in open-ocean conditions, routine eDNA barcoding could complement the interpretation of acoustic and visual surveys now routinely used to monitor cetacean habitat, especially for rare or cryptic species like beaked whales. In general, genomic technology is advancing rapidly and sequencing costs are dropping rapidly, promising to make ubiquitous eDNA sequencing for surveys of biodiversity more efficient and affordable in the near future.

#### ETHICS STATEMENT

This research did not involve any animal manipulation. Seawater samples were collected at a distance greater than 200 meters after the passage of whales, in compliance with current regulations for vessel approach to killer whales in U.S. waters of Puget Sound (200 yards) and proposed guidelines for approach to southern resident killer whales in Canadian waters (200 m).

### AUTHOR CONTRIBUTIONS

CSB: contributed to the initial experimental design, directed the field collection of eDNA, provided oversight of analysis and was responsible for manuscript preparation; DS: participated in field

#### REFERENCES


collection of eDNA, was responsible for all laboratory analyses and for preparation of figure and tables; SN: participated in field collection of eDNA and acoustic recordings and was responsible for acoustic analysis and data archiving; HK: contributed to the initial experimental design, participated in the field collection of eDNA and acoustic recordings and provided oversight of analysis. All authors have reviewed the manuscript and approved submission.

#### ACKNOWLEDGMENTS

This work was funded by the Office of Naval Research (Contract Number: N00014-15-1-2297). We thank Mike Wiese for his interest and support of this research. We thank the following: Anjanette and Nevé Baker for assistance in the field; Bernadette Holthuis and Kristy Kull of the Friday Harbor Laboratory for assistance with accommodation and vessel support; Brad Hanson, John Durban, Holly Fernbach, and Lance Barret-Lennard for advice on field logistics; Candice Emmons for identification of killer whale calls; Jeanne Hyde and Jason Wood for coordination with whale-watching operations; Margaret Hunter (USGS) for technical advise; Taal Levi and Jennifer Allen for technical discussions; and, Anne-Marie Girard Pohjanpelto of the Center for Genome Research and Biocomputing for technical assistance.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00133/full#supplementary-material

rad.com/webroot/web/pdf/lsr/literature/Bulletin\_6407.pdf: Hercules:Bio-Rad Laboratories Inc.


NOAA (2011). NOAA Issues New Rules to Safeguard Puget Sound's Killer Whales.


**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 reviewer KJH and handling Editor declared their shared affiliation.

Copyright © 2018 Baker, Steel, Nieukirk and Klinck. 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 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.

# Characterizing the Duration and Severity of Fishing Gear Entanglement on a North Atlantic Right Whale (Eubalaena glacialis) Using Stable Isotopes, Steroid and Thyroid Hormones in Baleen

Nadine S. J. Lysiak 1,2,3,4 \*, Stephen J. Trumble<sup>2</sup> , Amy R. Knowlton<sup>3</sup> and Michael J. Moore<sup>4</sup>

*<sup>1</sup> Department of Biology, University of Massachusetts Boston, Boston, MA, United States, <sup>2</sup> Department of Biology, Baylor University, Waco, TX, United States, <sup>3</sup> Anderson Cabot Center for Ocean Life, New England Aquarium, Boston, MA, United States, <sup>4</sup> Department of Biology, Woods Hole Oceanographic Institution, Woods Hole, MA, United States*

#### Edited by:

*Lisa Marie Komoroske, National Oceanic and Atmospheric Administration (NOAA), United States*

#### Reviewed by:

*Rebecca Ruth McIntosh, Phillip Island Nature Parks, Australia Nicholas Marc Kellar, Southwest Fisheries Science Centre, National Oceanic and Atmospheric Administration (NOAA), United States*

> \*Correspondence: *Nadine S. J. Lysiak nadine.lysiak@gmail.com*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *16 December 2017* Accepted: *26 April 2018* Published: *15 May 2018*

#### Citation:

*Lysiak NSJ, Trumble SJ, Knowlton AR and Moore MJ (2018) Characterizing the Duration and Severity of Fishing Gear Entanglement on a North Atlantic Right Whale (Eubalaena glacialis) Using Stable Isotopes, Steroid and Thyroid Hormones in Baleen. Front. Mar. Sci. 5:168. doi: 10.3389/fmars.2018.00168* North Atlantic right whales (*Eubalaena glacialis*) are highly endangered and frequently exposed to a myriad of human activities and stressors in their industrialized habitat. Entanglements in fixed fishing gear represent a particularly pervasive and often drawn-out source of anthropogenic morbidity and mortality to the species. To better understand both the physiological response to entanglement, and to determine fundamental parameters such as acquisition, duration, and severity of entanglement, we measured a suite of biogeochemical markers in the baleen of an adult female that died from a well-documented chronic entanglement in 2005 (whale *Eg2301*). Steroid hormones (cortisol, corticosterone, estradiol, and progesterone), thyroid hormones (triiodothyronine (T3) and thyroxine (T4)), and stable isotopes (δ <sup>13</sup>C and δ <sup>15</sup>N) were all measured in a longitudinally sampled baleen plate. This yielded an 8-year profile of foraging and migration behavior, stress response, and reproduction. Stable isotopes cycled in annual patterns that reflect the animal's north-south migration behavior and seasonally abundant zooplankton diet. A progesterone peak, lasting approximately 23 months, was associated with the single known calving event (in 2002) for this female. Estradiol, cortisol, corticosterone, T3, and T<sup>4</sup> were also elevated, although variably so, during the progesterone peak. This whale was initially sighted with a fishing gear entanglement in September 2004, but the hormone panel suggests that the animal first interacted with the gear as early as June 2004. Elevated δ <sup>15</sup>N, T3, and T<sup>4</sup> indicate that *Eg2301* potentially experienced increased energy expenditure, significant lipid catabolism, and thermal stress approximately 3 months before the initial sighting with fishing gear. All hormones in the panel (except cortisol) were elevated above baseline by September 2004. This novel study illustrates the value of using baleen to reconstruct recent temporal profiles and as a comparative matrix in which key physiological indicators of individual whales can be used to understand the impacts of anthropogenic activity on threatened whale populations.

Keywords: baleen, steroid hormones, thyroid hormones, stable isotopes, entantlement, right whale

#### INTRODUCTION

North Atlantic right whales (Eubalaena glacialis Müller 1776) are highly endangered mysticete cetaceans that range in the industrialized coastal waters of the eastern United States and Canada (Kraus and Rolland, 2007; Thomas et al., 2016). Individuals are identifiable by unique patches of cornified skin on the rostrum and bonnet, which become infested by cyamid whale lice and are highly visible against a whale's black skin (Kraus et al., 1986; **Figure 1A**). Since the early 1980s, the New England Aquarium (Boston, MA, U.S.) has curated a photoidentification and sightings catalog for monitoring habitat use, migratory phenology, health, calving rates, and survivorship of right whales (Hamilton et al., 2007)—making them one of the most well studied wild populations on the planet. Despite significant conservation efforts, there are approximately 450 individual right whales remaining, with a trajectory of population decline (Kraus et al., 2016; Pace et al., 2017; Pettis et al., 2017a).

FIGURE 1 | Field photographs and schematic of *Eg2301* (A) with calf (*Eg3310*) in 2002, photograph: New England Aquarium under NOAA permit 14233; (B) with detail of fishing gear entanglement, illustration: Scott Landry, Center for Coastal Studies; (C) post-mortem in 2005, with portion of left baleen rack visible, photograph: U.S. Coast Guard; (D) left baleen rack, reflected to reveal the lingual baleen surface and entangling lines, photograph: Virginia Aquarium Stranding Program under NOAA permit 932-1905/MA-009526.

In recent years, fixed fishing gear entanglements have increased in frequency and severity to become the primary source of serious injury (i.e., injuries that do not result in an immediate mortality but will likely result in subsequent mortality, Moore et al., 2013) and anthropogenic mortality to right whales (Knowlton et al., 2012, 2016; van der Hoop et al., 2013; Hayes et al., 2017). Fixed fishing gear<sup>1</sup> (herein referred to as "gear") is stationary, anchored at least at one end and can include gillnets, long lines, pots, traps, and vertical lines and buoys. It is estimated that 12–16% of the right whale population becomes entangled in gear each year, with approximately 83% of the population showing evidence of at least one entanglement (Knowlton et al., 2012). Entanglements can become chronic when large whales survive an acute gear entanglement and carry some or all of the gear away with them, often resulting in death within 6– 12 months after first detection in the field (Moore et al., 2006; Cassoff et al., 2011). These entanglements are often difficult to assign to a particular fishery or geographic location, or to track in real-time. Chronic entanglements have a variety of sub-lethal consequences including: serious [likely painful] injury (Knowlton and Kraus, 2001; Moore et al., 2005; Moore and van der Hoop, 2012; Moore, 2014), loss of body condition from increased energetic demands due to the additional drag of attached gear and/or impaired feeding (van der Hoop et al., 2015, 2016, 2017), or compromised health and reduced fecundity (Schick et al., 2013; Robbins et al., 2015; Rolland et al., 2016; Pettis et al., 2017b). Longitudinal studies of individuals impacted by gear entanglements are especially important to ascertain the effect(s) and interactions of these potential consequences.

A retrospective biogeochemical approach can provide critical insight regarding how entanglements affect large whales, such as their impact on foraging success, migration behavior, or stress physiology during an interaction with gear. Here, we measured a panel of six hormones and stable isotopes in the baleen tissue of a 12-year old, reproductively active female right whale (catalog number Eg2301) that died from a well-documented chronic gear entanglement. Eg2301 was first seen with attached gear in September 2004, but the extent of the associated injuries suggested that she had been carrying the gear for weeks/months prior to this field sighting. The aim of this study was to validate a novel endocrine method on a well-studied wild cetacean by (i) describing the hormone and stable isotope profiles of Eg2301 during known stressors that were both anthropogenic (i.e., a chronic gear entanglement) and natural (i.e., a calving event); and (ii) approximating the time and location of Eg2301's entanglement acquisition to provide better information for future mitigation of whale-fishery interactions.

Baleen, a series of metabolically inert keratin plates in the upper jaw that comprise the filter-feeding apparatus of mysticetes, is an ideal tissue for recent retrospective, longitudinal analysis due to its growth—which is assumed to occur continuously, year round (St. Aubin et al., 1984; Werth and Potvin, 2016; **Figures 1C,D**). In balaenid species like right whales, adults can grow baleen plates upwards of 8 feet in length, representing approximately 10 years of tissue for retrospective

<sup>1</sup>https://www.st.nmfs.noaa.gov/st4/documents/FishGlossary.pdf

analysis. Similar to mammalian hair, baleen is a cornified tissue that contains circulating endocrine compounds that have been deposited during its growth (Bryan et al., 2013; Ullmann et al., 2016; Cattet et al., 2017; Hunt et al., 2017b). For this study, we analyzed: two adrenal glucocorticoid steroids—cortisol and corticosterone—as elevations of these two hormones indicate the initiation of the vertebrate stress response via the hypothalamicpituitary-adrenal (HPA) axis (Norris, 2006; Romero and Wingfield, 2015). Thyroid hormones—triiodothyronine (T3) and thyroxine (T4)—were analyzed as biomarkers of foraging success since these hormones are regulators of metabolic rate in mammals, and thyroid hormone status correlates with energy expenditure and body condition (reviewed in Mullur et al., 2014; McAninch and Bianco, 2015). Female gonadal steroids progesterone and estrogen—were analyzed as indicators of pregnancy and potentially, estrous (Rolland et al., 2005; Kellar et al., 2013; Hunt et al., 2016a; Burgess et al., in press).

Additionally, we analyzed carbon (δ <sup>13</sup>C) and nitrogen (δ <sup>15</sup>N) stable isotopes in baleen, as they are well-established trophic markers of seasonal diet and foraging location in large whales (Schell and Saupe, 1993; Best and Schell, 1996; Lee et al., 2005; Hobson, 2007; Lysiak, 2008; Newsome et al., 2010; Matthews and Ferguson, 2015; Busquets-Vass et al., 2017). Animals acquire their stable isotope signatures from their diet, with predictable enrichment of both δ <sup>13</sup>C and δ <sup>15</sup>N at each trophic level (Kelly, 2000). Lysiak (2008) documented annual δ <sup>13</sup>C and δ <sup>15</sup>N oscillations in right whale baleen, which were attributed to a whale's foraging on zooplankton with disparate stable isotope signatures during annual migrations through their seasonal feeding habitats. In recent studies, stable isotopes have been used in conjunction with baleen steroid hormone analysis to establish a timeline of tissue growth—which greatly enhances the resolution at which longitudinal hormone concentrations may be interpreted (e.g., biological validations by Hunt et al., 2014, 2016a, 2017a).

### MATERIALS AND METHODS

#### Study Animal

Field observations and sighting records for whale Eg2301 were obtained from the North Atlantic Right Whale Catalog (http:// rwcatalog.neaq.org/; NARWC, 2006a). Eg2301 was born in 1993, and was photographed annually in at least one of the known right whale seasonal habitats in the Gulf of Maine or southeast US (see Supplementary Material). This female whale was first observed with a calf on December 31, 2002 in the southeast US calving ground (**Figure 1A**), and sightings with that calf continued as she migrated northward through Gulf of Maine feeding habitats in 2003. She was last seen with her calf on September 18, 2003 in the Bay of Fundy (New Brunswick, Canada). Eg2301 was not documented again until September 6, 2004, when she was sighted on Roseway Basin (Nova Scotia, Canada) with an extensive gear entanglement, which involved rope wrapped around the left pectoral flipper and cutting across and through the blowhole, and an extensive entanglement in the mouth and baleen (**Figures 1B–D**). Eg2301 was last seen alive on December 8, 2004, off the North Carolina, U.S. coast. The carcass of Eg2301 was discovered on March 3, 2005 on a barrier island off the Virginia, U.S. coast (**Figure 1C**), approximately 6 months after the entanglement was first detected in the field. Given these observational records, Eg2301 could have been entangled for a minimum of 178 days and maximum of 532 days. A necropsy indicated a serious injury to the left pectoral flipper (a deep v-shaped laceration in the soft tissue with extensive periosteal fibro-osseous proliferation around the humerus bone), partial occlusion of the left blowhole, severe dermal abrasions, and emaciated body condition (NARWC, 2006b; Cassoff et al., 2011). A single, full-length baleen plate (with associated gingiva) was collected from the carcass for this study and stored at −20◦C until analysis. Entangling rope was also collected from the carcass, but it could not be attributed to a particular fishery or location.

#### Sample Preparation

The baleen plate from Eg2301 was scrubbed with a plastic bristle brush and a mild shampoo [Herbal Essences] to remove sand and oils. Gingiva were flensed away to expose the unerupted base of the baleen plate. After drying at room temperature, the plate was wiped three times with 95% ethanol. We placed laboratory tape down the midline of the baleen plate, where 2 cm increments were marked for sampling, with "0" starting at the end of the wide, proximal base of the plate (i.e., the attachment point to upper jaw and newest baleen growth). To obtain higher temporal resolution of hormone data in the final year of the animal's life, we sampled the baleen plate at 1 cm intervals between 0 and 24 cm of baleen length. At each sampling point, baleen was ground into a powder using a Dremel rotary tool fitted with a tungsten carbide bit. All tools, surfaces, gloved hands, and the baleen plate were wiped with ethanol between each sampling bout to prevent cross-contamination. Given the limited amount of tissue at the distal tip of the baleen plate, hormone analysis (which requires significantly more tissue biomass—100 mg for hormone analysis vs. 1 mg for isotope analysis) was collected from 0 to 158 cm of baleen, while stable isotopes measurements were collected from the full length of the specimen (0–214 cm).

#### Stable Isotope Analysis

Baleen powder aliquots (1.0 ± 0.2 mg) were packaged into 4 x 6 mm tin capsules in duplicate for δ <sup>13</sup>C and δ <sup>15</sup>N analysis, using a PDZ Europa ANCA-GSL elemental analyzer interfaced with a PDZ Europa 20-20 isotope ratio mass spectrometer (Sercon Ltd., Cheshire, UK) at the University of California Davis Stable Isotope Facility. Baleen stable isotope values are reported in delta notation (δ, in parts per thousand), as the ratio of an unknown sample to an international standard (Vienna Pee-Dee Belemnite limestone for δ <sup>13</sup>C and atmospheric N<sup>2</sup> for δ <sup>15</sup>N):

$$
\delta^{13}\text{C or }\delta^{15}\text{N(\%o)} = [(R\_{sample}/R\_{standard}) - (1)] \times 1000
$$

where R is a heavy-to-light isotope ratio, <sup>13</sup>C/12C or <sup>15</sup>N/14N. Values were normalized using reference materials with an isotopic composition that spanned that of the sample range (i.e., bovine liver, glutamic acid, and nylon 5; δ <sup>13</sup>C range: −27.72 to 37.626‰, δ <sup>15</sup>N range: −10.31 to 47.6‰) and were calibrated to NIST Standard Reference Materials (IAEA-N1, IAEA-N2, IAEA-N3, USGS-40, and USGS-41). Analytical precision, 0.05‰ for δ <sup>13</sup>C and 0.20‰ for δ <sup>15</sup>N, is based on the standard deviation of a repeated internal laboratory standard (glycine). Reference samples and standards were analyzed after every 12 baleen samples.

### Data Analysis

Autoregressive (AR) models and spectral analysis were used to characterize variation and seasonality in the baleen δ <sup>13</sup>C and δ <sup>15</sup>N profiles. Linear trends were removed from each isotope time series before they were fit to high-order autoregressive (AR(p)) models, with model order p selected based on minimum Akaike information criteria (AIC). The spectral peak frequency was converted to samples per period (1/peak frequency) and then multiplied by sample interval (2 cm) to estimate the period length of each times series (after Matthews and Ferguson, 2015). All statistical analysis were performed using JMP 13.

Similar to other balaenid species, North Atlantic right whales exhibit asymptotic baleen (and body) growth, where annual tissue growth decreases with maturation (Best and Schell, 1996; Lubetkin et al., 2008; Lysiak, 2008; George et al., 2016). Given that Eg2301 was a 12-year-old whale, we expect the baleen plate to contain tissue that grew at variable annual rates across the approximately 8-year profile, such that a single estimate of the period of each isotope time series may not be the best characterization of a potentially dynamic annual growth rate. To characterize inter-annual variation in baleen growth rate, we estimated the period of each annual isotope cycle by counting the number of inclusive data points (i.e., isotope maxima to subsequent maxima) on the detrended data.

We used the stable isotope profiles to establish a timeline that indicates the calendar year of deposition for each baleen sample. The timeline begins with sample 0, which was given a date of February 2005 based on observations at necropsy, suggesting that Eg2301 was dead 1–2 weeks prior to the location of the carcass in early March 2005 (NARWC, 2006). We then worked backwards, using the estimated periods of each isotope cycle obtained from counting the number of data points per oscillation, to assign January of each previous calendar year on the baleen profiles. The field sighting record for Eg2301, which indicated migration behavior and potential times of residency in particular seasonal habitats, also informed this timeline. Protracted feeding in one area may manifest itself as a series of near-identical points in a baleen isotope record, as the whale is theoretically ingesting prey of consistent isotope signature during its residency. We noted instances of repeated sightings of Eg2301 in a particular habitat (in the same season and year) and cross-referenced these to the baleen timeline to determine if field observations matched temporally with sections of samples with similar isotope values. The sighting record also indicates that Eg2301 was first seen in the field with a new calf on December 31, 2002. This event provides an additional opportunity to ground-truth the timeline, as samples from this time period should be characterized by a precipitous decline in baleen progesterone concentration (see Hunt et al., 2016a).

### Hormone Extraction

Following the extraction protocol by Hunt et al. (2014), 100 mg of baleen powder was combined with 4.0 ml of 100% methanol in a borosilicate glass tube, vortexed for 20 h at room temperature, and centrifuged for 15 min at 4,000 g. The resulting supernatant was transferred to a clean glass tube and dried down at 45◦C under nitrogen in a dry-block sample evaporator. Samples were reconstituted in 1.0 ml assay buffer (catalog #X065; Arbor Assays, MI, USA) vortexed well, transferred to a cryovial, and frozen at −20◦C until analysis.

#### Hormone Assays

Commercial enzyme immunoassay kits from Arbor Assays were used to analyze baleen progesterone (catalog #K025), 17β-estradiol (catalog #K030), cortisol (catalog #K003), corticosterone (catalog #K014), T<sup>3</sup> (catalog #K056), and T<sup>4</sup> (catalog #K050). Each of these assay kits has previously been biochemically validated for hormone analysis of North Atlantic right whale baleen (Hunt et al., 2016a, 2017a,b). An extensive laboratory validation study by Hunt et al. (2017b) demonstrated that all six assay antibodies exhibited reliable binding affinity to the desired hormone in right whale baleen (i.e., good parallelism; slopes of the linear portions of the binding curves of serially diluted samples and standards are not significantly different), and verified that each assay was able to distinguish a range of concentrations with acceptable mathematical accuracy (i.e., good accuracy; linear regressions of known standard dose vs. observed dose are within r <sup>2</sup> > 0.95 and slope = 0.7–1.3).

The manufacturer's protocols were followed for analysis of all six hormones. Samples, standards, non-specific binding, and blank wells were assayed in duplicate. Any samples that fell outside 10–90% bound on the standard curve were re-assayed; samples with high hormone concentrations (percent bound < 10%) were diluted 2-fold (1:2, or up to 1:256 in some high progesterone samples) while samples with low hormone concentrations (percent bound > 90%) were concentrated within the assay wells at 2:1. Samples with > 10% coefficient of variance between duplicates were re-assayed. Results were converted to nanograms of immunoreactive hormone per gram of baleen. Baseline hormone concentrations were determined for each dataset using an iterative process excluding all points that deviate from the mean + 2 SD until no points exceed this threshold (after Brown et al., 1988). Peaks in hormone values were defined as points exceeding the overall baseline concentration + 2 SD. The corticosterone (compound B) to cortisol (compound F) ratio (i.e., B/F ratio) was calculated for all hormone samples.

### RESULTS

#### Stable Isotope Timeline

The Eg2301 baleen δ <sup>13</sup>C and δ <sup>15</sup>N profiles contained regular oscillations, which are hypothesized to be annual signals (Schell and Saupe, 1993; Best and Schell, 1996; Lee et al., 2005; Hobson, 2007; Newsome et al., 2010; Matthews and Ferguson, 2015; **Figure 2A**). Similar to another study of North Atlantic right whale baleen isotopes (Lysiak, 2008), we observed δ <sup>13</sup>C maxima in the boreal fall/winter while δ <sup>13</sup>C minima occurred in the spring. δ <sup>15</sup>N profiles were slightly out of phase with δ <sup>13</sup>C, with δ <sup>15</sup>N maxima occurring in summer/fall and minima occurring in winter (**Figure 2A**). Spectra from first-order autoregressive models (AR(1)) fit to the stable isotope data estimated periods of 32.4 cm (δ <sup>13</sup>C) and 29.3 cm (δ <sup>15</sup>N). The periods of each individual isotope oscillation, representing annual baleen growth rates, were estimated by counting the number of data points in each isotope cycle. Across the baleen isotope records, we observed a range of 12–17 data points per cycle (x = 15.28 ± 1.70 points, **Figure 2B**). This is equivalent to a mean annual growth rate of 30.56 cm yr−<sup>1</sup> . Eg2301 shows evidence of decreasing baleen growth rate, especially in the most recent years of the isotope record (i.e., 2002–2004, **Figure 2**) when the baleen growth rate declines to 24 cm yr−<sup>1</sup> , which is a stable growth rate reported for other adult female North Atlantic right whales (Lysiak, 2008; Hunt et al., 2016a).

The two methods used to determine the period of each isotope profile, modeling and counting data points, provided similar results (period = 32.4 cm yr−<sup>1</sup> (δ <sup>13</sup>C, via modeling), 30.56 cm yr−<sup>1</sup> (mean annual growth rate, via counting points), or 29.3 cm yr−<sup>1</sup> (δ <sup>15</sup>N, via modeling). Since we observed a decrease in the period of each isotope cycle across the profile (**Figure 2B**), we used the annual periods estimated by the counting method to build a tissue growth timeline that would best reflect the dynamic nature of the annual baleen growth rate. There were three instances of repeated field sightings of Eg2301 within a single habitat area, in the same season and same year. Eg2301 was seen in the Bay of Fundy (New Brunswick, Canada) in July/August/September of 1999 and 2001, and in August/September of 2003 (NARWC, 2006a). When these field sightings were cross-referenced with the sample timeline, we observed sections of baleen with very similar δ <sup>13</sup>C and δ <sup>15</sup>N values (open circles in **Figure 2A**), which suggest Eg2301's protracted feeding on zooplankton with a consistent isotopic signature during these times. The date of first sighting with a new calf (December 31, 2002) coincides with baleen samples with declining progesterone concentrations, occurring just after a sustained peak associated with gestation (see section Discussion). The sighting record of Eg2301 provides important validation for the tissue growth timeline. Given the good alignment of the timeline with these known life history and migratory events, we suggest that an error of ±1 data point (approximately 21.5–30 days) is a reasonable estimate of uncertainty for the dates of deposition assigned to each sample.

#### Hormone Panel

In the early years of the baleen hormone profiles for Eg2301 (i.e., between 1999 and 2001), nearly all samples fluctuate at or below baseline and corticosterone:cortisol (B/F Ratio) ratios are <1.0 (indicating that immunoreactive cortisol was measured at higher concentrations than immunoreactive corticosterone; **Figure 3**). A protracted progesterone peak, two orders of magnitude above baseline, was observed from early 2001 to late 2002 (x = 199.48 ng g−<sup>1</sup> , baseline = 2.249 ng g−<sup>1</sup> ; **Figure 3A**). In 2001, the increase in progesterone coincided with three discrete estradiol spikes (**Figure 3A**), high cortisol (x = 9.02 ng g−<sup>1</sup> , baseline =

FIGURE 2 | (A) Carbon (δ <sup>13</sup>C, gray circles) and nitrogen (δ <sup>15</sup>N, black circles) stable isotope ratios in *Eg2301* baleen. Open circles represent samples matched to repeated field sightings in the Bay of Fundy (New Brunswick, Canada) in 1999 and 2001 (whale seen in July, August, and September) and 2003 (whale, with calf, seen in August and September). Vertical dotted black lines represent estimates for calendar year of baleen growth. Gray box indicates the duration of the progesterone peak associated with the single known pregnancy for *Eg2301*. Vertical dashed gray lines represent the minimum and maximum bounds of entanglement duration; vertical red dashed line is the revised estimate of when the entanglement was acquired. (B) Annual baleen growth rate (cm yr−<sup>1</sup> ).

1.967 ng g−<sup>1</sup> ) and elevated corticosterone (baseline = 4.663 ng g −1 ; **Figure 3B**), B/F ratios predominantly <1.0 (**Figure 3C**), and elevated but variable T<sup>3</sup> (baseline = 0.719 ng g−<sup>1</sup> ) and T<sup>4</sup> (baseline = 0.376 ng g−<sup>1</sup> ; **Figure 3D**). By contrast in 2002, the progesterone peak coincided with a peak in estradiol (x = 6.91 ng g−<sup>1</sup> , baseline = 2.79 ng g−<sup>1</sup> ; **Figure 3A**), low cortisol and elevated corticosterone (**Figure 3B**), B/F ratios >1.0 (**Figure 3C**), and variable fluctuations in T<sup>4</sup> (early 2002) and T<sup>3</sup> (late 2002) (**Figure 3D**). In 2003, we observed one discrete, concomitant increase in progesterone and estradiol (**Figure 3A**) and several instances of elevated T<sup>3</sup> (**Figure 3D**).

At death, in February 2005, all six hormones measured during this study were elevated above baseline (**Figure 3**). T<sup>3</sup> and T<sup>4</sup> showed variable fluctuations above baseline throughout the baleen record, but persistently increased in concentration beginning in June/July 2004 with the highest recorded concentration of T<sup>3</sup> being the last data point on February 2005 (8.48 ng g−<sup>1</sup> ; **Figure 3D**). By August/September 2004, corticosterone showed persistent elevations, increasing to the highest recorded concentration (13.89 ng g−<sup>1</sup> , 3-fold above baseline) by the end of the baleen record (**Figure 3B**). By September 2004, progesterone and estradiol rose continuously

FIGURE 3 | Observed immunoreactive hormone concentrations across the *Eg2301* baleen plate. Vertical dotted black lines represent estimates for calendar year of baleen growth, as determined from stable isotope profiles. Horizontal dotted lines indicate baseline hormone levels. Gray box indicates the duration of the progesterone peak associated with the single known pregnancy for *Eg2301*. Vertical dashed gray lines represent the minimum and maximum bounds of entanglement duration; vertical red dashed line is the revised estimate of when the entanglement was acquired. (A) Progesterone (pink solid line—note logarithmic scale, baseline = 2.249 ng g−<sup>1</sup> ) and estradiol (purple dashed line, baseline = 2.79 ng g−<sup>1</sup> ); (B) Cortisol (red solid line, baseline = 1.967 ng g−<sup>1</sup> ) and corticosterone (orange dashed line = 4.663 ng g −1 ); (C) Corticosterone:cortisol ratio (B/F Ratio), horizontal dashed line at *y* = 1.0 indicates equal concentrations of both hormones; (D) triiodothyronine (T3–blue solid line—note logarithmic scale, baseline <sup>=</sup> 0.719 ng g−<sup>1</sup> ) and thyroxine (T4–green dashed line—note logarithmic scale, baseline = 0.376 ng g −1 ).

above baseline (**Figure 3A**). We observed three discrete spikes of high cortisol during 2004–2005, the highest concentration (4.72 ng g−<sup>1</sup> and approximately 2-fold above baseline) occurred at the end of the hormone record (**Figure 3B**).

### DISCUSSION

Baleen hormone and stable isotope profiles (**Figures 2**, **3**) showed correspondence with documented life history events for this reproductively mature, chronically entangled female right whale. A period of gestation was characterized by high baleen progesterone concentrations and contained two distinct phases, each lasting approximately 1 year: (1) low estradiol, high corticosterone-cortisol in 2001 and (2) high estradiolcorticosterone, low cortisol in 2002 (**Figure 3**). The end of gestation (i.e., parturition and a return to baseline progesterone values) coincides with the timing of the first field sightings of Eg2301 with a calf, on December 31, 2002. Thyroid hormones (T<sup>3</sup> and T4) are periodically elevated across the baleen record and point to periods of short- and long-term food limitation, associated thermal stress, and increased energy expenditure. Finally, the hormone panel shows evidence of a stress response in 2004, during a period when Eg2301 was documented in the field with a chronic gear entanglement.

#### Gestation and Estrous Gestation Timeline

Progesterone, as detected in several balaenid whale tissue matrices (e.g., feces, Rolland et al., 2005, 2012, 2017; blubber, Kellar et al., 2013; respiratory vapor, Burgess et al., in review; and baleen, Hunt et al., 2016a), is a robust indicator of pregnancy. Longitudinal, physiological-based estimates of mysticete gestation period, as provided by this study, are extremely limited. The protracted progesterone peak observed in the Eg2301 baleen record contains 29 data points that, according to our date of growth timeline, are equivalent to approximately 696 days (23 months). These results could indicate that North Atlantic right whale gestation is significantly longer than previously reported in other balaenids.

Based on field observations, stranding data, and whaling records, Best (1994) estimated a gestation period of 357–396 days (12–13 months) for southern right whales (Eubalaena australis). Reese et al. (2001) modeled bowhead whale (Balaena mysticetus) average gestation length at 13.9 months (predictive distribution = 12.8–15.0 months), based on observations of the reproductive tract of specimens harvested during subsistence whaling. Estimates for both species are associated with a high degree of uncertainty given the difficulty of observing small embryos within the female reproductive tract during the initial, non-linear phase of fetal growth (Best, 1994; Reese et al., 2001). Using an endocrine approach, Hunt et al. (2016a) conservatively defined gestation as "uninterrupted baleen samples with progesterone >100 ng g−<sup>1</sup> ," and detected progesterone peaks lasting 540 and 451 days (18 and 15 months, respectively) in two North Atlantic right whales. Though anecdotal, in an endocrine study of North Atlantic right whales, one adult female was sampled in the summer/fall 2004 (fecal progesterone was low) and again in summer/fall 2005 (fecal progesterone was very high), and the whale calved in the winter 2005/2006 calving season (R.M. Rolland, pers. comm., see Rolland et al., 2005). In this case, fecal progesterone was low (i.e., at or below baseline) at approximately 15–17 months prior to parturition, which puts upper bounds on the length of gestation for this particular animal. Taken together, these previous studies highlight the need for further examination of gestation length in this species, with priority to the detection and characterization of the non-linear fetal growth period in early pregnancy.

The temporal mismatch between our observed high progesterone peak and previous estimates of gestation length could potentially be explained by the timeline calculated using stable isotopes. Baleen growth rate varies with age in bowhead (Lubetkin et al., 2008) and right whales (Lysiak, 2008), where calves, juveniles, and sub-adults exhibit faster growth rates than mature adults. Changes in baleen growth rate per year, or inter-annual growth rates, were quantified in this study as the number of data points per isotope oscillation, and if significant confounding changes occurred in baleen growth rate during pregnancy, we should also see the period of annual δ <sup>13</sup>C and δ <sup>15</sup>N signals change coincidentally (i.e., the period of δ <sup>13</sup>C and δ <sup>15</sup>N oscillations should increase during a slower growth rate scenario, or decrease with a faster growth rate). We did observed inter-annual variation in Eg2301 baleen growth, with longer periods seen in older sections of baleen (**Figure 2**). However, during the proposed pregnancy, the 2002 cycle contained 16 data points (baleen growth rate = 32 cm yr−<sup>1</sup> ) and the 2001 cycle contained 17 data points (baleen growth rate = 34 cm yr−<sup>1</sup> ) (**Figure 2B**), suggesting that baleen growth occurred at a relatively consistent rate for the duration of the progesterone peak.

In this study, as in previous work (Schell and Saupe, 1993; Best and Schell, 1996; Lee et al., 2005; Hobson, 2007; Lysiak, 2008; Newsome et al., 2010; Aguilar et al., 2014; Matthews and Ferguson, 2015; Busquets-Vass et al., 2017), we assume that baleen grows continuously throughout the year (i.e., no intraannual or seasonal variability in growth). Given the consistent wear patterns in baleen, such as abrasion by the tongue, intraand extra-oral water flow, and food or sediment particles (Werth et al., 2016), there is strong selective pressure for right whales to consistently maintain their baleen given its integral role in foraging. It is possible that baleen growth rate could change during pregnancy, stress, or food limitation, as variations in the gestational growth rates of keratin-based tissues have been observed in human hair (LeBeau et al., 2011) and cow hooves (Hahn et al., 1986; Mülling et al., 1999). This may add uncertainty and error to our established timeline of baleen tissue growth. While difficult to measure directly in free-ranging cetaceans, future studies should prioritize the investigation of seasonal variability in baleen growth and wear.

Rather than being indicative of one continuous pregnancy, an alternative explanation is that the observed progesterone peak can be divided into two distinct phases: estrous and gestation. In this case, we hypothesize that gestation comprises the second half of the baleen progesterone peak. There is observational evidence for the endpoint of gestation; it ends at parturition in late 2002, when progesterone returns to baseline levels (at baleen length 58) and Eg2301 was first seen in the field with her new calf on December 31 2002; **Figures 1A**, **3A**). However, defining the beginning phase of gestation is more tentative. Cortisol concentrations may be an adequate endocrine biomarker for the transition between estrous and gestation since a very dramatic shift was observed in that hormone between our two proposed phases of the reproductive event (**Figures 3B,C**). This is similar to observations of captive female Asian elephants, where a protracted cortisol peak preceded a progesterone peak in longitudinal serum samples, and indicated a transition from the follicular phase to the luteal phase (Fanson et al., 2014). Under this scenario, gestation begins in late 2001 when cortisol levels return to baseline and B/F ratios increase to >1.0 (at baleen length 88; **Figures 2**, **3**). This provisional gestation period represents 16 baleen samples, and is equivalent to approximately 384 days (12.8 months), using our timeline of baleen growth. This corresponds well to the Best (1994) estimate of a 12–13 month gestation in southern right whales—a closely related species.

#### Estrous Cycling

We hypothesize that a period of estrous, a state of sexual receptivity during which a female is capable of conceiving, comprises the first half of the baleen progesterone peak (i.e., early to late 2001, from baleen length 114–90 cm; **Figure 3A**). During this time, corticosterone, T3, and T<sup>4</sup> are variable but primarily elevated above baseline, estradiol is predominantly low, and cortisol is consistently high−9-fold above baseline and at the highest concentrations observed across the entire baleen record (**Figure 3**). Cortisol (along with other glucocorticoids, GCs) is an index of relative stress (i.e., it indicates activation of the hypothalamic-pituitary-adrenal axis), but also has a role in responding to natural states of increased energetic needs such as migration or reproduction in free-ranging cetaceans (Brann and Mahesh, 1991; Andersen, 2002; Tetsuka, 2007; Rolland et al., 2012; Trumble et al., 2013; Kellar et al., 2015; Hunt et al., 2017a). Under normal physiological conditions, shortterm increases in GCs promote sexual receptivity, stimulate gonadotropins (i.e., luteinizing hormone and follicle stimulating hormone), facilitate ovulation, and ameliorate damage from inflammation (reviewed in Fanson et al., 2014). However, reproductive dysfunction or failure can occur under chronic stress and elevated GCs (Tilbrook et al., 2002). Sighting records and the right whale photographic database do not indicate any significant anthropogenic stressors (i.e., visible entangling gear, new entanglement scars, or evidence of a non-lethal vessel strike) or an overall decline in health for Eg2301 during 2001 (NARWC, 2006a; A. Knowlton, pers. comm, after Schick et al., 2013). These observations, and that Eg2301 successfully completed a full-term pregnancy in 2001–2002 suggest that the observed high cortisol values could be a component of the animal's natural reproductive cycle. This result highlights the importance of combining biomarkers of reproduction with those of stress physiology—as it could be tempting to interpret these high GCs as an indicator of a major disturbance or anthropogenic stressor in the absence of collocated progesterone and estrogen concentrations.

Estrous is poorly defined in mysticetes (Boness, 2009). However, odontocete reproductive studies show estrous cycles as spikes in urinary estrogen conjugates closely followed by luteinizing hormone surges, and cycles lasting between 30 and 41 days in Pacific white-sided dolphins (Lagenorhynchus obliquidens), bottlenose dolphins (Tursiops truncatus), and killer whales (Orcinus orca) (Robeck et al., 2004, 2005, 2009). Additionally, Robeck et al. (2004) observed sequential, natural estrous cycles within a season for one individual L. obliquidens, which entailed continuously elevated urinary progesterone and detection of a corpus luteum—with no ultrasound evidence of pregnancy—for 86 days (and 103 days in the following year). In the context of these findings, we recommend expanding the baleen hormone panel to include a suite of progesterone and estrogen conjugates or metabolites. The interactions among these hormones could provide a more detailed assessment regarding the reproductive physiology of right whales. Future studies should consider non-target analysis using high performance lipid chromatography (HPLC) to accommodate a wider range of prospective analytes.

According to the timeline of baleen growth and stable isotopes, Eg2301 was 8 years old during the proposed period of estrous, became pregnant just before age 9, and was 10 years old just after parturition. These observations are in agreement with population-wide estimates for mean age of sexual maturity (9 years) and mean age at first calving (10.1 years; Kraus et al., 2001, 2007). In addition to a period of potential estrous, a failed pregnancy, pseudopregnancy, or delayed implantation may have occurred during this time. Delayed implantation, a temporary diapause of the embryonic blastocyst, is widespread among the mammalian orders Rodentia and Carnivora, but only one known species of Cetartiodactyla (the roe deer, Capreolus capreolus) exhibits this life history strategy (Renfree and Shaw, 2000; Ptak et al., 2012). Previous studies commonly observed low estrogen concentrations during diapause and then implantation was indicated by a surge in circulating estrogens and progesterone in northern fur seals (Callorhinus ursinus, Daniel, 1981), roe deer (Aitken, 1981) and giant pandas (Ailuropoda melanoleuca, Zhang et al., 2009). If right whales undergo the delayed implantation observed in other marine mammals, our hormone panel suggests that active gestation could have begun as early June/July 2001 (when progesterone peaks) or September 2001 (when estradiol surges above baseline after protracted low levels) meaning that the estimate of gestation period would increase to approximately 16–19 months. Future studies could investigate this further using a broader hormone panel or additional tissue matrices.

#### Food Limitation

Right whales feed via continuous ram filtration, which is accomplished with a complex mouth anatomy that filters preyladen water via hydrodynamic pressure differentials across the baleen plates (Werth, 2004). The entanglement of Eg2301 included rope obstructing the mouth, with extensive, knotted wraps through the right and left baleen plates (**Figures 1B–D**). This gear configuration likely inhibited feeding ability and/or significantly decreased filtration efficiency, potentially leading to a prolonged fasting. As migratory capital breeders with seasonally abundant prey, right whales are well adapted to periods of food limitation, and will catabolize stored energy reserves (e.g., subdermal adipose tissue or blubber) during fasting (Lockyer, 1981). Blubber thickness measurements correspond to a whale's onboard energy balance, and fluctuate with foraging success and reproductive state (Miller et al., 2011; Irvine et al., 2017). At necropsy, the carcass of Eg2301 had extremely thin dorsal blubber (8.5 cm observed, in contrast to 13.4 ± 1.8 cm in non-entangled adult whales; NARWC, 2006b; Miller et al., 2011; van der Hoop et al., 2016). The baleen stable isotope and hormone profiles support these observations, suggesting that Eg2301 experienced severe, prolonged fasting conditions prior to death.

δ <sup>15</sup>N is an indicator of trophic position in traditional food web studies, since the heavy isotope of nitrogen (15N) is preferentially incorporated into consumer tissues from their diet, which results in a systematic enrichment in nitrogen isotope ratio (15N/14N) with each trophic step (Kelly, 2000). Thus, elevations in δ <sup>15</sup>N may correlate with dietary shifts to <sup>15</sup>N-enriched prey, at higher trophic levels. Fasting conditions can mimic trophic enrichment, since an animal is essentially metabolizing muscle tissue during periods of fasting, thereby causing elevated tissue δ <sup>15</sup>N (Castellini and Rea, 1992; Hobson et al., 1993). Given their significant adaptations to seasonal fasting and ability to increase lipid stores, baleen whales usually do not display trophic enrichment (Lysiak, 2008; Aguilar et al., 2014; Matthews and Ferguson, 2015) with periodic food limitation and migration. Eg2301 baleen δ <sup>15</sup>N values exhibit an annual oscillating pattern similar to those observed in other North Atlantic right whales, with the exception of the last few months of the record where δ <sup>15</sup>N increases to 12.78‰, the highest value recorded in any North Atlantic right whale (Lysiak, 2008). This shift from the regular oscillations could indicate the severe depletion of blubber lipids and a heavier reliance on protein (muscle) catabolism as a means to meet ongoing energetic demands (Fuller et al., 2005; Aguilar et al., 2014).

The thyroid hormones triiodothyronine (T3) and thyroxine (T4) are important regulators of metabolic rate in mammals (Norris, 2006; Mullur et al., 2014; McAninch and Bianco, 2015). During fasting, the hypothalamic-pituitary-thyroid (HPT) axis is depressed, which decreases circulating thyroid hormones, basal metabolic rate, and energy expenditure—which could serve as a survival mechanism (Mullur et al., 2014). Contrary to this paradigm, we observed elevated concentrations of immunoreactive T<sup>3</sup> and T<sup>4</sup> in the baleen of Eg2301 prior to her death, when we know that her blubber layer became extremely thin and lipid depleted. Lipolysis, fat oxidation, or catabolism of the blubber lipids in fasting marine mammals serves two major roles; mobilizing stored energy (Pond, 1978; Lockyer, 1981) and generating metabolic water (Ortiz et al., 1978; Ortiz, 2001) in the absence of active feeding. Thyroid-promoted lipolysis could be a mechanism to meet osmotic and energetic demands during periods of prolonged nutritional stress. For example, captive West Indian manatees that experienced reduced food intake (i.e., a diet switch from lettuce to sea grass) exhibited increased serum T<sup>4</sup> levels and decreased body mass (up to 17%, primarily due to loss of fat; Ortiz et al., 2000).

Blubber is a tissue that also reduces thermal conductance and is a critical adaptation to maintain thermal homeostasis (Dunkin et al., 2005; Samuel and Worthy, 2005)—a constant battle for endothermic homeotherms living in seawater. Loss of blubber thickness can correspond to a concomitant loss of thermal insulation. When under thermal stress, thyroid hormones are secreted to promote adaptive or facultative heat production in brown adipose tissue (BAT) (Silva, 2003; Norris, 2006; Mullur et al., 2014; Santini et al., 2014). In odontocete cetaceans, BAT is located at the innermost blubber layer and enveloping the entire body, exclusive of the thermal windows (Hashimoto et al., 2015). Periods of fasting could represent a negative feedback loop, where thyroid-promoted lipolysis degrades the integument for the sake of water balance and energetic homeostasis, which then necessitates elevated thyroid hormones to promote BATmediated thermogenesis.

The thyroid panel of Eg2301 during the entanglement period is also surprising because stress is generally thought to inhibit thyroid secretions (Eales, 1988; Norris, 2006). However, different patterns in thyroid output have been reported for some stressors that increase energetic output (e.g., exercise stress), contrary to what is seen during simple fasting (Uribe et al., 2014; Hunt et al., 2016b). For example, entangled distressed leatherback sea turtles had higher serum T<sup>4</sup> than healthy wild individuals, presumably from the added energetic cost of carrying gear (Hunt et al., 2016b). Eg2301 would have likely expended additional energy for locomotion to compensate for the drag associated with the entangling gear (van der Hoop et al., 2013, 2015, 2016), which could be an example of exercise stress promoting thyroid function.

Acute or periodic fasting may be indicated in right whale baleen by isolated and discrete rises in thyroid hormones T<sup>3</sup> and T<sup>4</sup> (occurring sporadically across the baleen record) and longer-term fasting may be indicated by prolonged T<sup>3</sup> and/or T<sup>4</sup> elevations (occurring on three occasions: 2001, late 2002-early 2003, and mid 2004–2005; **Figure 3D**). In 2001, elevated T<sup>3</sup> and T<sup>4</sup> concentrations in the baleen coincide with the first half of the 2001–2002 progesterone peak (**Figure 3**). While right whales can travel significant distances, undetected, sighting records for this animal do not indicate a migration to the southeast U.S. in 2001. However, in late 2002, sighting records do indicate Eg2301 migrated approximately 1,500 km from the Gulf of Maine to the southeast US calving ground to give birth and nurse a new calf, with a subsequent return trip several months later (NARWC, 2006a). Interestingly, T<sup>3</sup> is elevated during this period, with peaks circa November 2002 and February 2003, when Eg2301 presumably would be traveling during southbound and northbound migrations, respectively (**Figure 3D**). Finally, T<sup>3</sup> elevated above baseline beginning in June 2004, 3 months prior to the detection of the entanglement in the field (**Figure 3D**). This suggests that Eg2301 experienced significant, chronic fasting leading to loss of blubber, and associated thermal stress over a period of approximately 9 months.

### Entanglement

#### Entanglement Duration and Location

Eg2301 was initially observed with a gear entanglement on September 6, 2004 (minimum entanglement duration = 178 days or 6 months in **Figure 3**), however the baleen hormone panel suggests that the animal first interacted with gear several months earlier (mid-2004; see red line in **Figure 3**). Field observations of an extensive mouth/baleen entanglement plus low blubber thickness documented at necropsy support the hypothesis that Eg2301 experienced an extensive period of compromised feeding. In the hormone panel, we observed persistent T<sup>3</sup> elevations in baleen beginning in June 2004 (NARWC, 2006b; **Figures 1B,C**, **3D**). By September 2004, all hormones except for cortisol were elevated above baseline (**Figure 3**). At the September 6, 2004 field sighting (when the entanglement was first discovered), researchers noted that the skin on the left pectoral flipper appeared white underwater—suggesting that the entanglement must have already persisted long enough for the flipper tissue to become necrotic from being tightly wrapped in line (NARWC, 2006b; **Figure 1B**). By the end of the baleen hormone record (February 2005), cortisol also elevated above baseline and δ <sup>15</sup>N increased to a level indicating [fasting-induced] trophic enrichment (**Figure 2A**). Given these observations and our estimation of uncertainty of the baleen timeline, we propose that Eg2301 carried the chronic gear entanglement for a minimum of 9 (±1) months, potentially first encountering the entangling gear as early as June 2004. At that time, there are no field sighting records available for Eg2301 in late spring 2004, so we can only speculate as to her location when becoming entangled. Eg2301's baleen δ <sup>13</sup>C reaches a minima at this time, which is consistent with depleted δ <sup>13</sup>C values of zooplankton collected in the Great South Channel (southern Gulf of Maine) in May/June, as well as trends in population-wide migration behavior and baleen isotopes for other individual right whales (Hamilton et al., 2007; Lysiak, 2008). Though Eg2301 was not observed in this area in 2004, she was seen in the Great South Channel habitat in previous years (i.e., April/May/June 1999–2002 and July 2003; NARWC, 2006a). Additional isotopic markers, such as sulfur, oxygen, or deuterium (deHart and Picco, 2015; Matthews and Ferguson, 2015) could provide more spatial resolution in the baleen isotope profile to better assess the location of gear acquisition for this animal.

#### Stress Physiology

After the carcass of Eg2301 was recovered, researchers confirmed that entangling rope had sliced down to the left humerus bone as well as cut into the skin of the head and blowhole (NARWC, 2006b). These wounds likely caused pain and significant stress (Moore and van der Hoop, 2012). Despite these persistent injuries during the entanglement, we observed paradoxically low cortisol (June 2004—February 2005). This is unexpected since chronic stress, and thus higher cortisol concentrations, should have resulted from this entanglement, as seen in previous studies. Rolland et al. (2017) documented high cortisol in the feces of North Atlantic right whales due to chronic entanglement as well as exposure to noise from vessel traffic (Rolland et al., 2012). We suspect that either baleen is not truly reflective of circulating cortisol levels, or this could indicate insufficient adrenal gland function (i.e., hypothalamic-pituitaryadrenal axis dysfunction or "adrenal fatigue")—the mechanisms of which are poorly understood in mammals (Edwards et al., 2011). The highest cortisol concentrations in Eg2301 baleen were observed during the proposed estrous cycle, rather than during the entanglement, suggesting that baleen cortisol may not be the best biomarker to use for studying large whale response to chronic anthropogenic stressors (**Figure 3B**). In contrast, corticosterone was elevated during proposed estrous, gestation, and entanglement (**Figure 3B**), with the highest values observed in the last 4 months of the baleen record indicating that this hormone might be more informative when studying stress response in large whales. Furthermore, B/F ratios are consistently >1.0 beginning in 2002 (coinciding with the proposed beginning of gestation), meaning that greater concentrations of immunoreactive corticosterone were detected in these samples than immunoreactive cortisol (**Figures 3B,C**). Hunt et al. (2017a) noted 4-fold higher corticosterone vs. cortisol concentrations in North Atlantic right whale baleen and suggested that this species might conform to a dual GC signaling model (Koren et al., 2012), where cortisol is a better index of acute stress and corticosterone better reflects chronic stress. Though untested in this study, future hormone panels might also include the adrenal hormone aldosterone as a stress biomarker. Aldosterone was elevated in the feces of pregnant North Atlantic right whales (Burgess et al., 2017) and is detectable in right whale baleen (Hunt et al., 2017b).

In summary, we used elevations of female gonadal steroids progesterone and estradiol to identify a reproductive event in Eg2301's baleen hormone record. Beginning in September 2004, when she was first seen entangled, both of these hormones were again elevated above baseline and continued to rise until the death of Eg2301. This could be physiological evidence of a second estrous cycle or pregnancy. As capital breeders, right whales need to secure sufficient onboard energy stores in their blubber to support gestation and lactation (Irvine et al., 2017). Miller et al. (2011) observed that right whales had the thickest blubber just prior to becoming pregnant. Body condition for Eg2301 was declining during her entanglement, and would likely not meet the energetic threshold required to become pregnant or carry a fetus to term. Due to the remote location of the carcass, an internal examination was not undertaken; therefore the presence or absence of a fetus was not determined. In tandem with the visual decline in body condition during entanglement, baleen T<sup>3</sup> and T<sup>4</sup> concentrations showed sustained increases potentially indicating significant lipolysis or increased energy expenditure, providing another indicator to rule out a pregnancy event. Elevated progesterone in late 2004 could be attributed to the adrenal glands, which release progesterone (the parent hormone to GCs) in response to stress (reviewed in Herrera et al., 2016). This may contribute to a greater bioavailability of cortisol during the stress response, providing the body with glucose (via gluconeogenesis in the liver) and restoring homeostasis (Kudielka and Kirschbaum, 2005). This physiological response could be considered part of an unsustainable "emergency lifehistory stage" brought on by entanglement (Wingfield et al., 1998; van der Hoop et al., 2016) in a last ditch attempt to prolong survival. Anecdotally, in 2010–2011, increasing progesterone (R. Rolland, pers. comm.) and very high cortisol (Rolland et al., 2017) were observed in the longitudinal fecal samples of an entangled, juvenile [non-pregnant] female right whale (catalog number Eg3911).

### CONCLUSIONS

The panel of biogeochemical markers from Eg2301 baleen allowed us to investigate the longitudinal physiological response of a large whale to a chronic gear entanglement in its industrialized ocean habitat. With a panel of adrenal and gonadal steroid hormones, thyroid hormones, and stable isotopes, we were able to establish a timeline of baleen tissue growth and examine the fluctuations of hormones in response to: a calving event, an approximately 3,000 km seasonal migration, prolonged periods of food limitation, and stress associated with entanglement-induced serious injury. These observations support an updated estimate of minimum entanglement duration for Eg2301 from 6 to 9 months, which enhances our understanding of the timeline of this event and provides insight into where it may have occurred. With the biological validation provided in this study, we can apply this method to future forensic studies where the cause of death of a large whale is uncertain or undetermined, or the timeline of events is not known. This novel study illustrates the value of using baleen to reconstruct recent temporal profiles and as a comparative matrix in which key physiological indicators of individual whales can be used to understand the impacts of anthropogenic activity on threatened whale populations.

### AUTHOR CONTRIBUTIONS

NL and ST conceived of the project and secured funding; ST also provided significant laboratory resources. NL is the principal author and responsible for study design, data collection, analysis and interpretation. AK contributed to the North Atlantic right whale field sightings and necropsy databases. MM conducted the necropsy, collected the baleen plate, and provided support and feedback in study design and interpretation. All authors revised this work critically and provide final approval of the version to be published.

### FUNDING

The Woods Hole Oceanographic Institution's Ocean Life Institute and Marine Mammal Center funded this study and NL was supported by a Postdoctoral Fellowship from Baylor University.

## ACKNOWLEDGMENTS

Right whale baleen was collected under NOAA permit 932- 1905-MA-009526. We are indebted to the US Coast Guard, Sue Barco, and the Virginia Aquarium Stranding Program for facilitating sample collection at a challenging, remote necropsy site. This study was conducted through a data-sharing agreement with the North Atlantic Right Whale Consortium; we acknowledge Phillip Hamilton, the New England Aquarium, and the contributions of many researchers and institutions to this invaluable database. Joy Matthews and David Harris of the UC Davis Stable Isotope Facility ensured the meticulous analysis of our samples. Elizabeth Burgess provided critical advice in the study design and carefully reviewed early versions of this manuscript; Rebel Sanders assisted with preliminary hormone analysis. We thank two reviewers for their contributions to this manuscript.

#### REFERENCES


#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00168/full#supplementary-material

dolphin blubber (Tursiops truncatus). J. Exp. Biol. 208, 1469–1480. doi: 10.1242/jeb.01559


income breeding sperm whales using historical whaling records. R. Soc. Open Sci. 4:160290. doi: 10.1098/rsos.160290


australisrelated with reproduction, life history status and prey abundance. Mar. Ecol. Prog. Ser. 438, 267–283. doi: 10.3354/meps09174


**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 reviewer NMK and handling Editor declared their shared affiliation.

Copyright © 2018 Lysiak, Trumble, Knowlton and Moore. 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 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.

# Basking Shark (Cetorhinus maximus) Movements in the Eastern North Pacific Determined Using Satellite Telemetry

Heidi Dewar <sup>1</sup> \*, Steven G. Wilson<sup>2</sup> , John R. Hyde<sup>1</sup> , Owyn E. Snodgrass <sup>3</sup> , Andrew Leising<sup>4</sup> , Chi H. Lam<sup>5</sup> , Réka Domokos <sup>6</sup> , James A. Wraith<sup>1</sup> , Steven J. Bograd<sup>4</sup> , Sean R. Van Sommeran<sup>7</sup> and Suzanne Kohin<sup>1</sup>

*<sup>1</sup> Life History Program, NOAA Southwest Fisheries Science Center, La Jolla, CA, United States, <sup>2</sup> Hopkins Marine Station, Stanford University, Pacific Grove, CA, United States, <sup>3</sup> Ocean Associates, Inc., Life History Program, NOAA Southwest Fisheries Science Center, La Jolla, CA, United States, <sup>4</sup> Environmental Research Division, NOAA Southwest Fisheries Science Center, Monterey, CA, United States, <sup>5</sup> Large Pelagics Research Center, School for the Environment, University of Massachusetts Boston, Boston, MA, United States, <sup>6</sup> Ecosystems and Oceanography Division, NOAA Pacific Islands Fisheries Science Center, Honolulu, HI, United States, <sup>7</sup> Pelagic Shark Research Foundation, Capitola, CA, United States*

#### Edited by:

*Mark Meekan, Australian Institute of Marine Science, Australia*

#### Reviewed by:

*Nuno Queiroz, University of Porto, Portugal Yannis Peter Papastamatiou, Florida International University, United States*

> \*Correspondence: *Heidi Dewar heidi.dewar@noaa.gov*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *16 January 2018* Accepted: *23 April 2018* Published: *16 May 2018*

#### Citation:

*Dewar H, Wilson SG, Hyde JR, Snodgrass OE, Leising A, Lam CH, Domokos R, Wraith JA, Bograd SJ, Van Sommeran SR and Kohin S (2018) Basking Shark (Cetorhinus maximus) Movements in the Eastern North Pacific Determined Using Satellite Telemetry. Front. Mar. Sci. 5:163. doi: 10.3389/fmars.2018.00163* To fill data gaps on movements, behaviors and habitat use, both near- and offshore, two programs were initiated to deploy satellite tags on basking sharks off the coast of California. Basking sharks are large filter-feeding sharks that are second in size only to whale sharks. Similar to many megafauna populations, available data suggest that populations are below historic levels. In the eastern North Pacific (ENP) Ocean, the limited information on basking sharks comes from nearshore habitats where they forage. From 2010 to 2011, four sharks were tagged with pop-off satellite archival tags with deployments ranging from 9 to 240 days. The tags provided both transmitted and archived data on habitat use and geographic movement patterns. Nearshore, sharks tended to move north in the summer and prefer shelf and slope habitat around San Diego, Point Conception and Monterey Bay. The two sharks with 180 and 240 days deployments left the coast in the summer and fall. Offshore their paths diverged and by January one shark had moved to near the tip of the Baja Peninsula, Mexico and the other to the waters near Hawaii, USA. Vertical habitat use was variable both within and among individuals and changed as sharks moved offshore. Nearshore, most time was spent in the mixed layer but sharks did spend hours in cold waters below the mixed layer. Offshore vertical movements depended on location. The shark that went to Hawaii had a distinct diel pattern, with days spent at ∼450–470 m and nights at ∼250–300 m and almost no time in surface waters, corresponding with the diel migration of a specific portion of the deep scattering layer. The shark that moved south along the Baja Peninsula spent progressively more time in deep water but came to the surface daily. Movement patterns and shifts in vertical habitat and use are likely linked to shifts in prey availability and oceanography. Data collected indicate the potential for large-scale movements and the need for international dialogue in any recovery efforts.

Keywords: basking shark, habitat, diel vertical migration, satellite telemetry, Cetorhinus maximus, foraging ecology

## INTRODUCTION

A long history of human interaction has resulted in the decline of many species of marine megafauna including turtles, tunas, cetaceans, rays and sharks (Springer et al., 2003; Lewison et al., 2004; Marshall et al., 2006; Bradshaw et al., 2008; Dulvy et al., 2008; Croll et al., 2016; ISC, 2016). This list includes the second largest shark, the basking shark (Cetorhinus maximus, Gunnerus, 1765), that can reach 12 m in length and is named for its habit of swimming slowly at the surface (Compagno, 1984; Priede, 1984; Sims, 2008; McFarlane et al., 2009). Similar to many marine mammals, targeted fisheries for basking sharks in the Pacific Ocean ended decades ago (McFarlane et al., 2009). However, while a number of marine mammal populations have rebounded (IWC, 1998; Summarized in Carretta et al., 2009), there is no obvious increase in basking shark populations in the Pacific Ocean (McFarlane et al., 2009). Although there is an increasing body of research on basking sharks in the Atlantic Ocean, very little is known about this species in the Pacific Ocean, hampering efforts to develop a recovery plan and identify potential sources of mortality.

While basking sharks are circum-global in distribution, they are most commonly reported in the temperate, coastal waters of the northern hemisphere in the Atlantic and Pacific Oceans (Compagno, 1984; Ebert, 2003; Sims, 2008; McFarlane et al., 2009; Curtis et al., 2014). In the eastern North Pacific (ENP), aerial surveys, sightings and catch data indicate their range is from Southeast Alaska to Baja California, Mexico. Historically, there are two regions where basking sharks were most commonly observed: the southern coast of British Columbia, Canada, and near Monterey Bay, CA, U.S.A. (Squire, 1967, 1990; reviewed in McFarlane et al., 2009). Sharks from these areas are thought to belong to the same stock based on their proximity and seasonal shifts in abundance (Squire, 1990; Darling and Keogh, 1994; Ebert, 2003; McFarlane et al., 2009). Historical data show that basking sharks were more prevalent in Canada from March through October (Darling and Keogh, 1994; McFarlane et al., 2009), while off CA, peak abundance was from October through March. It should be noted, however, that basking sharks were reported off California (CA) throughout the year (Squire, 1990; Baduini, 1995). While these two regions are considered linked, the full geographic range of this stock is unknown. In the Atlantic Ocean both electronic tagging data (Gore et al., 2008; Skomal et al., 2009; Braun et al., 2018) and genetic analysis (Hoelzel et al., 2006) suggest the potential for large-scale migrations. No electronic tagging or genetic studies have been conducted in the ENP and additional information on migrations and population structure is needed.

While there are currently no targeted fisheries, there is a long history of fishery interactions with basking sharks in the ENP. Off central CA, fisheries took an estimated 700–800 sharks in two periods from 1924 to 1938 and 1946 to 1952 (Phillips, 1948; summarized in McFarlane et al., 2009). They were taken for their liver oil, human consumption, fertilizer, and use in animal feed. In the 1940's the Canadian government initiated an eradication program to prevent sharks from interfering with salmon fisheries. Between entanglement in salmon nets, sport kills, and the eradication program it is estimated that 1,000–2,600 sharks were killed by 1970 when the program ended (McFarlane et al., 2009). Basking sharks have also been taken incidentally in a range of gear types in U.S., Mexican and Canadian waters as well as in high-seas driftnet fisheries (Bonfil, 1994; Darling and Keogh, 1994; McKinnell and Seki, 1998; Larese and Coan, 2008; Sandoval-Castillo and Ramirez-Gonzalez, 2008; McFarlane et al., 2009). In the ENP, basking sharks are now rare in areas where hundreds to thousands of individuals were previously reported and aerial surveys, sightings, and catch data indicate a decline in the population (Squire, 1967, 1990; Darling and Keogh, 1994; Baduini, 1995; COSEWIC, 2007; McFarlane et al., 2009). While a decline in abundance is apparent for both Californian and Canadian waters, there is a high degree of variability in observations across years (McFarlane et al., 2009). This holds true even historically, Jordan (1887) reported that basking sharks would not be seen for 20 years at a time. The cause of this variability has not been determined in part due to the lack of basic information on migratory patterns, geographic distributions, essential habitat, and species rarity. A better understanding of the mechanisms underlying this variability is needed to help determine the cause of short and long-term trends in abundance. It is also critical to determine where sharks go when they are not observed in coastal waters and to more completely identify potential sources of mortality.

Due to concerns about the populations of basking sharks in the ENP and the lack of basic biological data, the National Oceanic and Atmospheric Administration (NOAA) listed basking shark as a Species of Concern in 2010 (NOAA, 2004, 2010). Basking sharks are listed as endangered in the Pacific, Canadian waters (COSEWIC, 2007), and also have a Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES) Appendix II listing (CITES, 2002) which requires that trade be documented. They are also listed by a number of other international organizations (Fowler, 2005; IUCN, 2007). To obtain additional data on basking shark movements and habitat use, two satellite tagging programs funded by NOAA were initiated off CA. These studies provide the first data on the large-scale movements and behaviors of basking sharks in the ENP.

#### MATERIALS AND METHODS

#### Tagging and Data Processing

Given the rare occurrence of basking sharks off the coast of CA, we relied on public sightings or reports to local fishing forums (e.g., bdoutdoors.com) to locate sharks for tagging. When basking sharks were reported, we launched a small vessel or a pair of inflatable skiffs and searched the area where the shark was last seen. On all occasions when sharks were observed, tags were successfully deployed. For tagging, free-swimming basking sharks were approached and the tag anchor was inserted just below the dorsal fin with a long tagging pole (2.7–4 m). All basking shark tagging was done in accordance with protocols approved by the Southwest Fisheries Science Center, Institutional Animal Care and Use Committee.

Tags used were MK10-AF transmitting fast-GPS tags (MK10 version 10.1) and MK-10 pop-up satellite archival tags (PSATs version MK10.2) from Wildlife Computers (Redmond, WA, U.S.A.). Both tag types release after a preprogrammed period and transmit light data for estimating geolocation along with temperature and pressure (depth) data summarized as profiles (PDTs) and histograms. If tags are recovered, the entire archival dataset can be downloaded. Tags were programmed to summarize data over 6- or 24-h sampling intervals. The GPS tags also log time-series data (temperature and pressure) set at userspecified intervals. In addition, when the GPS tag is at the surface, it captures data from the GPS satellite system that can be postprocessed to obtain more accurate locations. If there is sufficient surface time, the data are transmitted prior to the tag's release. The release dates were set for 180 (n = 1) and 240 (n = 3) days following deployment. A final location is estimated by the Argos satellite system after the tag releases from the shark.

The tags were leadered and anchored using different methods (**Table 1**). The three dart types used included (1) a nylon head augmented with spear gun flopper-blades (PM dart; Prince and Goodyear, 2006), (2) a titanium sled dart (Block et al., 1998) and (3) an eight cm JBL slip-tip, spear-point. Tags were leadered with 300 lb. test monofilament that was covered with heat-shrink tubing. Leader lengths ranged from 30 to 45 cm.

The archival and transmitted data were analyzed to characterize habitat preferences. Data from the first 24 h were not analyzed to reduce the possible effects of tagging on behavior (Hoolihan et al., 2011). Sea surface temperatures (SST) were calculated from temperatures recorded in the top 2 m of the water column. For analyses of diel patterns, we used data from sharks A, B, and C for which bin intervals were 6 h. The two bins that encompassed sunrise and sunset were not included. Shark C transited two time zones and the estimated location of the shark was used to shift the time of sunrise and sunset to local time for diel analyses.

For analyses of onshore and offshore habitat use, the data for sharks C and D were clustered and analyzed according to location (see below). For both sharks, location data were missing around the times the sharks moved offshore, data for those days were not included in the analyses. A Kolmogorov-Smirnov (KS) test was used to test for significant differences between temperature and depth histograms. Mean values are reported ± standard deviation unless otherwise indicated.

### Geolocation Estimates

Deployment and pop-up locations, transmitted light data, and intermittent GPS locations were used to estimate latitude and longitude using the state-space Kalman filter model TrackIt (Nielsen and Sibert, 2007). Correction with SST was not possible due to the limited time the sharks spent at the surface and their proximity to the coast during periods of prevalent cloud cover in the summers of 2010 and 2011. For Shark C, PDT records provided enough temperature data at 200 m to enable correction of position estimates using a variant of TrackIt (Lam et al., 2010). Matching was performed at 200 m between tag measurements and World Ocean Atlas 2009 monthly 1◦ -grid climatology (Locarnini et al., 2010). Lastly, bathymetric correction was applied (Galuardi et al., 2010). Sharks C and D were considered to have moved offshore when they made a directed movement to the west and continued on an offshore trajectory (**Figure 1**). Given the geolocation errors it was not possible to use a specific location or distance from the coast as the transition point from near- to offshore. Regional SST and chlorophyll a (chl a) around this transition were obtained through CoastWatch (http:// coastwatch.pfeg.noaa.gov/) and plotted using ArcGIS software (Environmental Systems Research Institute, Redlands, CA). SST data from CoastWatch were also used to characterize SST in the waters northeast of Hawaii.

### Acoustic Backscatter

To infer foraging behavior in the Central Pacific for shark C, we compared vertical movements to data from an acoustic survey conducted near the track of the shark from ∼23◦N, 157◦W to 26◦N, 158◦W on March 27, 2009. Like shark C, the survey was conducted in the subtropical gyre in an area influenced by the westward flowing North Pacific current. Both these factors result in longitudinal homogeneity. The acoustic surveys were conducted on board the NOAA R/V Oscar Elton Sette, using a hull mounted Kongsberg Maritime AS Simrad (Horten, Norway) EK60 split-beam system with 7◦ beam width operating at the 38, 70, and 120 kHz frequencies from the surface to 1,200 m depth. The system was calibrated prior to each survey using a 38.1-mmdiameter tungsten carbide sphere according to standard methods (Demer et al., 2015). Acoustics signals were processed using Echoview software (Hobart, Tasmania) to remove cavitation noise and bubble dropout, ensuring high signal-to-noise ratios. Data, in the form of volume backscattering coefficients (Sv in dB re 1 m−<sup>1</sup> ), were used to examine the vertical characteristics of the scattering layers. Differences in Sv between frequencies were used to assess the relative composition of the shallow scattering layer and deep scattering layer (Mac Lennan et al., 2002).

## RESULTS

### Tag Deployments

Satellite tags were deployed on three sharks off San Diego and one shark in Monterey Bay, CA (**Table 1**). In all cases fork length was estimated at between 5 and 6.1 m but the sharks swam away quickly and it was not possible to determine sex. Data were obtained from all tags. The first two tags released early: shark A off Morro Bay, CA after 51 days (**Figure 1**) and shark B after 9 days when it was recovered from a beach near the tagging location (San Diego, CA). The final two tags released on their programmed dates: shark C after 240 days ∼400 km northeast of Hawaii, U.S.A. and shark D after 180 days ∼1,130 km west of the tip of Baja Peninsula, Mexico. For the three GPS tags (sharks A, B, and C), no transmissions were received prior to release and only four valid GPS-based location estimates were obtained, one for shark A and three for shark C and all from nearshore locations.

#### Tracks

Tracks were estimated for sharks A, C, and D. The average estimated errors for the light-based latitude and longitude were <0.4 degrees. No locations were estimated for shark B due to its short deployment and the close proximity of the deployment and tag recovery locations.

The estimated track for shark A indicates that after tagging on June 6, the shark moved northwest of San Diego toward



*Leader length (cm) and dart type (PM, Prince Musyl; JBL ST, JBL slip-tip, spear-point; MD, metal dart), fork length (FL), deployment date and location and pop-up date and location for all basking sharks deployments.*

the acoustic survey.

the Channel Islands (**Figure 1**). On July 5, the GPS positioned the shark near the continental slope off Point Conception, CA. Between July 5 and 29, when the tag released off Morro Bay, the shark was in the region around Point Conception. Similarly, shark C moved northwest from San Diego after tagging and remained in the area around Point Conception from mid-June until early August (**Figure 1**). Both sharks appear to have stayed over the continental shelf, in the region around Point Conception, including the Channel Islands and Santa Barbara Channel. In early August (between August 3 and 11), shark C left the coast and made relatively directed movements southwest (2,760 km in 59 days, 47 km/day) until early October when it reached ∼150◦W, northeast of Hawaii. Movements then slowed and the shark remained northeast of Hawaii until early February when the tag released.

Shark D, tagged off Monterey in August, remained around Monterey Bay until early November when it moved offshore (between November 7 and 11) and toward the south, remaining well offshore of Point Conception and the Southern California Bight (**Figure 1**). Movements south were relatively directed (1,560 km in 82 days, 19 km/day) until the shark reached the area west of the tip of Baja Peninsula, Mexico where SST was >20◦C. There the shark stopped around January 18 and returned north prior to the tag releasing on January 29.

Regional SST and chl a around the time that the two sharks left the coastal area were examined (**Figure 2**). Shark C left in early August when chl a concentration was high around Point Conception and it remained high after the shark left. SST in the days around the time of departure dropped from ∼17 to ∼15◦C. Shark D left central CA in early November just as a decline in both chl a concentration and SST (from ∼15 to ∼13◦C) became apparent. Neither shark followed obvious surface fronts in SST or chl a in their offshore migrations (**Figure 2**).

#### Temperature and Depth

Temperature and depth data were obtained from all four sharks including an 8-day archival record from shark B. A summary of

temperature and depth experienced across all sharks is provided in **Table 2**. Overall, temperatures ranged from 5 to 24.6◦C and depths were from the surface to 784 m (**Figure 3**). Except for shark C, all sharks came to the surface daily. SST spanned more than 14◦C ranging from 10.4 to 24.6◦C although the average across fish was much narrower (13.6–16.4◦C; **Table 2**). Given

TABLE 2 | Summary of temperature (◦C) and depth (m) for all sharks.


*SST average (*±*SD) and range, overall minimum temperature, and maximum depth nearshore and offshore, and average minimum temperature (*±*SD), and average maximum depth (*±*SD) nearshore and offshore. When the shark remained nearshore no offshore values are given.*

FIGURE 3 | Temperature and depth profiles obtained from the PDT data for each of the 4 sharks. (A) Shark A, (B) shark B, (C) shark C, and (D) shark D. Inset is a histogram of the percent time (X axis) spent in different depth bins (y axis, in m) over the nearshore record separated into day (gray) and night (black) periods for all but shark D. For shark A the dotted lines indicate the approximate points in time at which the behavior is considered to have shifted between deeper and shallower modes. The arrows indicate the approximate time the shark moved offshore.

the variability in vertical habitat use, portions of the tracks were separated into periods when shifts were apparent (**Figure 3**). For shark A, which remained nearshore, periods of shallow vs. deep vertical habitat use were separated. For sharks C and D, nearshore and offshore periods were separated as described above.

#### Nearshore

In nearshore regions, all but shark A showed similar preference for shallower depths, with 75% of the time or more spent in the top 50 m and 95% in the top 100 m with only periodic dives into deeper waters (**Figure 3**). For sharks B, C, and D the average daily maximum depth was between 49 and 113 m. For shark A, the overall depth distribution was deeper (55% spent in the top 50 m and 85% above 100 m) as was the maximum average daily depth (182 m ±132). The increased depth for shark A was also reflected in lower temperatures with >88% of the time between 8 and 14◦C whereas the remaining sharks spent 83–99% of the time between 10 and 18◦C. The overall deeper depths and cooler temperatures for shark A resulted from two periods where the average SST was warmer (SSTs 15.3 ± 1.7◦C vs. 12 ± 1.2◦C) and the water column more thermally stratified (**Figure 3**, **Supplemental Figure 1**).

Differences in nearshore vertical movements within and between individuals were also apparent in the time-series data. The greatest detail is in the two-min archival data (**Figure 4**). Shark B showed a range of dive patterns, making frequent vertical excursions at various depths in relation to the thermocline or spending protracted periods near the surface or at depth with one dive lasting 6.5 h in waters of 11.5◦C. The 10-min, timeseries data also show behaviors similar to those seen for shark B (not shown) with additional patterns including more extensive vertical excursions (**Figure 4**).

#### Offshore

When sharks C and D moved offshore, their temperature and depth ranges expanded, maximum depths increased, minimum temperature decreased and SST increased (**Table 2**, **Figure 3**). Time at temperature was more broadly distributed with ∼95% of the time spent over a 14◦C range (6–20◦C) in comparison to an 8◦C range (10–18◦C) in nearshore habitat (**Figure 5**).

While the temperature and depth ranges expanded for both sharks offshore, habitat use differed. As shark D moved south after leaving Central CA, time spent at depths deeper than 100 m increased and correspondingly, the time spent in the top 5 m decreased to a minimum of 20% (**Supplemental Figure 2**). However, this shark came to the surface each day. Near the end of the record when SST was >20◦C (maximum 22.4◦C), the maximum dive depth decreased, increasing again when SST declined (**Figure 3**) as the shark moved back north.

While the depths for shark C also increased offshore (**Table 2**, **Figure 3**), shark C rarely came to the surface (4% time spent 0– 5 m). As shark C moved west the time spent in deeper waters

Frontiers in Marine Science | www.frontiersin.org

PDT for that day, dotted black line = relative light levels as indicator of day and night in A and B.

increased and over the last 2 months almost 100% of the time was spent in waters deeper than 200 m (**Supplemental Figure 2**). An exception was 10 recorded excursions into shallow waters (<50 m) which lasted 3–16 min (average 7 ± 4). Eight of ten shallow events happened during the bins including sunrise or sunset. There was no apparent link to lunar phase.

#### Diel Patterns

For sharks A, B, and C it was possible to examine diel patterns. Although this was not possible for shark D, given the 24-h bins, the bimodal depth distribution is consistent with a diel pattern (**Figure 5**, **Supplemental Figure 2**). The nearshore records for A and C show a significant increase in day vs. nighttime depths (K-S test, p < 0.05). The sharks spent from 77 to 92% of their time in the top 50 m at night in comparison to 47–65% during the day (**Figure 3**). Also, in the time-series data the average daytime hourly depth increased significantly from 37 m at night (SE ± 2) to 62 m during the day (SE ± 5) for shark A and from 12 m (SE ± 1) to 51 m (SE ± 4) for shark C. For shark B no diel pattern was apparent (**Figure 3**).

The most striking and consistent diel difference was observed for shark C while offshore. While this shark did not come to the surface, the nighttime depths (20:00–6:00 = 250 ± 29 m) were significantly shallower (t-test, p < 0.05) than those during the day (8:00–14:00 = 447 ± 28 m) (**Figure 6**). Given the thermal stratification, even at depth, the shark experienced as much as a 10◦C change in temperature over the course of a day (night 11.4–17.7◦C, day 7.3–13.2◦C). Interestingly, while the morning descent occurred around sunrise, the evening ascent occurred 2– 3 h before sunset. The SST in this region was between ∼23 and ∼25◦C over this period.

#### SST, Temperature Change, and Depth

While there was a high degree of variability in the vertical movements, one pattern that held across locations was the relationship between the SST and the temperature range (dT) experienced on a given day (dT = SST-min temperature). For all sharks, regression analyses showed a significant increase in dT with SST (**Figure 7**) including when near- and offshore regions were separated. An exception to this occurred at the end of the record for shark D when SST was >20◦C and the minimum temperature increased. In comparison, regression analyses of maximum depth and SST showed mixed results and R <sup>2</sup> were lower than for SST and dT (0.04–0.34 vs. 0.64–0.87). Nearshore there was only a significant increase in the maximum depth with SST for shark A (**Figure 3**). Offshore max depth increased with SST for shark C and decreased with SST for shark D.

#### Movements Relative to Sound-Scattering Layers

During the final 2 months of its deployment, shark C traveled into an area where research surveys had mapped the soundscattering layer associated with vertically migrating mesopelagic organisms (**Figure 8**). The shark's nighttime depths correspond with a thin layer of organisms at the bottom of the shallow scattering layer at ∼250–300 m. During the day the depths correspond again to this same thin layer which migrated to its

daytime depth at the top of the deep scattering layer at ∼450– 470 m depth. However, the shark ascended to nighttime depths in advance of the upward migration of the sound-scattering layer. The acoustic signature of this area is consistent with those of smaller crustaceans and gelatinous zooplankton, as well as small fish.

#### DISCUSSION

This study reports on the first electronic tags deployed on basking sharks in the ENP. Results complement existing sightings and fisheries databases with the advantage of providing information on offshore movements where data are sparse. Overall, results reveal that sharks occupy convergence zones in nearshore habitat in the summer and fall and then disperse offshore. Offshore, vertical habitat expands into deeper waters and movements are linked to a specific portion of the sound-scattering layer. Both near and offshore, vertical habitat was highly variably and likely linked to both vertical and geographic patterns in prey availability as well as regional oceanography.

### Geographic Movements and Essential Habitat

Movements and residency patterns over short and medium time frames (days to months) provide insight into essential habitat, which has become a cornerstone of fisheries management (Rosenberg et al., 2000). Of particular interest is where animals choose to spend protracted periods of time as these are presumed to be associated with key ecological needs, typically foraging. A considerable amount of effort has been put into characterizing these areas of residency from animal tracks using a range of approaches including state-space models (Jonsen et al., 2007), first passage time (McKenzie et al., 2009), and fractal analyses (Tremblay et al., 2007). While no modeling was conducted in the present study given the small sample size, examination of the tracks revealed that basking sharks spent up to 12 weeks in specific areas over the continental shelf and slope including San Diego, Point Conception, and the Monterey Bay region. Previous studies in the ENP showed similar residency periods. Baduini (1995) and Darling and Keogh (1994) reported that individual basking sharks spent up to 30 or 42 days in Monterey Bay, CA, or Clayoquot Sound, Vancouver Island, Canada, respectively. Similar short-term residence times have been reported for basking sharks in coastal regions in the Atlantic Ocean (Sims et al., 2003; Gore et al., 2008; Sims, 2008; Skomal et al., 2009; Curtis et al., 2014; Doherty et al., 2017; Braun et al., 2018). Across locations, the continental shelf and slope are important habitat for basking sharks, especially during the summer months.

As mentioned, residency patterns along the coast are likely linked to the availability of forage. While filter feeding cannot be documented using satellite tags, some foraging data for basking sharks are available for the ENP. Both off CA and Canada basking shark occurrence was linked to higher zooplankton densities (Darling and Keogh, 1994; Baduini, 1995). Similar to in other areas, the preferred prey of basking sharks off CA is thought to be calanoid copepods, specifically Calanus pacificus, (Baduini, 1995). C. pacificus developmental stages (C) IV and V contain a large lipid droplet and are energetically dense. The CIV and CV stage of C. finmarchicus are targeted by basking sharks in the western North Atlantic (Baduini, 1995; Siders et al., 2013; Curtis et al., 2014). Off CA the CIV and CV stages occur year around but are most abundant in surface waters from April through October (Johnson and Checkley, 2004), although high concentrations of CV can also be found at depth during periods of diapause (see below).

As with other filter feeders, ideal foraging habitat in coastal waters require mechanisms that concentrate prey (Sims and Quayle, 1998; Sims et al., 2003; Croll et al., 2005; Dewar et al., 2008; Hazen et al., 2013; Scales et al., 2014; Miller et al., 2015). As a result, foraging habitat will depend on physical factors that may include tidal cycles, internal waves, variability in ocean currents, mesoscale features such as fronts and eddies, and bathymetry. Overall, the California Current is a critical habitat for a range of predators across trophic levels (Block et al., 2011). For basking sharks in particular, there are a number of regions in the California Current that have specific forcing mechanisms to concentrate prey. Mesoscale eddies just south of Point Conception have been associated with hot spots for a number of seabird species (Yen et al., 2006). In the Santa Barbara Channel the deep bathymetry and associated hydrography leads to an incredibly dense aggregation of diapaus copepods (13,000 g wet weight (ww) m−<sup>3</sup> ) from 450 to 500 meters in the spring and summer (Alldredge et al., 1984; Osgood and Checkley, 1997; Ohman et al., 1998). These levels are orders of magnitude higher than the estimated threshold density for foraging ∼0.6 g ww m−<sup>3</sup> in basking sharks (Sims, 1999; Sims et al., 2006). While it is not known if basking sharks take advantage of these dense concentrations, which occur at very low oxygen concentrations (∼0.2 ml L−<sup>1</sup> ), it is clear that they are drawn to this region. Finally, the combination of regional upwelling and the topography of the Monterey Canyon leads to dense concentrations of forage in Monterey Bay (Croll et al., 2005). This area is a hotspot for animals that count on forcing mechanisms to aggregate prey including leatherback sea turtles and filter feeding whales (Croll et al., 2005; Block et al., 2011). Interestingly, the region where shark D spent 12 weeks overlapped with satellite-tagged leatherback sea turtles (Benson, pers. comm.) that left the area around the same time as the basking shark. Any effort to define the essential habitat of basking sharks along the West Coast of North America will need to be dynamic in nature and factor in physical forcing.

When away from shore, essential habitat is more difficult to identify, especially based on surface features. Sharks spent considerable periods below the surface and vertical habitat use was variable. If the offshore migrations of females are associated with pupping as speculated for Atlantic basking sharks (Skomal

et al., 2009; Braun et al., 2018), the habitat may be related to the needs of the pups rather than the adult which is a further complication for identifying essential habitat.

In addition to foraging, another key element commonly used to characterize essential habitat across species is SST. For a complete understanding of the preferred SST range, measurements are needed across seasons. Similar to the findings in this study, basking sharks are reported over a broad range of SST. Lien and Fawcett (1986) report the highest catch rates when SST was 8–12◦C, whereas Owen (1984) reported their occurrence at SSTs up to 24◦C. Basking sharks clearly spend time in areas with even higher SST, but primarily occupy deep cooler water in these regions (Skomal et al., 2009; Braun et al., 2018; this study). While their deep-water occurrence and the broad thermal range make identifying preferred SSTs challenging, SST has been shown to be an important predictor of basking shark abundance (Schwartz, 2002; Skomal et al., 2004; Cotton et al., 2005). Cotton et al. (2005) found that basking sharks occurred between SST ∼12 and 15◦C and the number of basking sharks was strongly correlated with SST but only weakly correlated with copepod density. While it may not be possible to extrapolate globally, SST may provide a useful regional guide to patterns in basking shark abundance and distribution.

Movements over periods from weeks to months also provide insight into seasonal migrations. Considering all locations, sharks were off southern CA in the spring and early summer (May– June) and moved north as temperatures warmed. In the summer (July–September) they were either around Point Conception or Central CA. Both sharks with longer tracks left their summer foraging grounds in either the middle of summer or fall. In the fall and winter they were offshore but were 3,260 km apart in late January. North-South seasonal movements are also reported for sharks in both the East and West Atlantic (Sims et al., 2003; Cotton et al., 2005; Skomal et al., 2009; Doherty et al., 2017; Braun et al., 2018). Movements in the West Atlantic Ocean, however, are more extensive than in the East and sharks crossed the equator, moving as far south as Brazil. Shifting between coastal habitat in the summer and fall to offshore waters in the winter and spring is observed across a range of species (Block et al., 2011; Campana et al., 2011; Dewar et al., 2011). More long-term tracks are needed to determine the links between near- and offshore winter grounds, potential sex linked differences in migration, and if and when individuals return to the CA coast. Some fidelity to summer foraging grounds has been documented in other studies (Sims et al., 2000; Hoogenboom et al., 2015; Braun et al., 2018).

A comparison of recent and historic records for the ENP indicates similarities and differences in seasonal and spatial patterns. Similar to previous reports, the regions around Point Conception, Morro Bay, and Monterey Bay appear to still be important habitat (Squire, 1990; Baduini, 1995). Differences are apparent in seasonal patterns. While basking sharks were documented off CA throughout the year, from 1962 to 1985 the peak in abundance was from October through March (Squire, 1990). In the current study all public sightings and tagging events were in the spring and summer (NMFS unpublished data). This is consistent with the more recent data provided by Baduini (1995) who, in the early 1990s, saw some sharks throughout the year but reported peaks in May and August. Based on available data, it appears that in recent decades the coastal waters off CA provide important summer and fall foraging grounds with no reports in winter months.

Another apparent change in the ENP is the drop in observations on the historically important summer foraging grounds off Canada. Since the work of Darling and Keogh (1994) from 1973 to 1992 when 27 individuals were identified, very few animals have been observed off Canada or off the Pacific Northwest (McFarlane et al., 2009; DFO, unpublished data). While a detailed examination of all potential variables is beyond the scope of this paper, there is some evidence of a shift in productivity off Pacific Canada around 1989 (Hare and Mantua, 2000; McFarlane et al., 2000). McFarlane et al. (2000), using a composite index, identified a shift in climate ocean conditions that resulted in a decrease in productivity in a range of fish species. Inter-annual regional shifts in the abundance of basking sharks off the U.K. have been linked to zooplankton abundance (Sims and Quayle, 1998; Sims and Reid, 2002; Doherty et al., 2017). The reduced sightings could also be a function of the natural variability or a decline in the population in the ENP (Squire, 1990; Darling and Keogh, 1994; McFarlane et al., 2009).

### Vertical Movements

Similar to other areas and consistent with their name, basking sharks spent the majority of their time in the upper portions of the water column in nearshore waters. Coastal surface waters, especially in eastern boundary currents, are highly productive and mesoscale features that concentrate prey are common (Barber and Smith, 1981; Pauly and Christensen, 1995; Hazen et al., 2013; Scales et al., 2014). While surface waters were clearly important, there was a high degree variability in vertical habitat. Off the U.K., in regions with little thermal stratification, considerable surface feeding was apparent whereas when waters were thermally stratified, sharks dove deeper and spent less time at the surface (Sims et al., 2003, 2005). Vertical movements for shark A reflect a similar pattern, with higher thermal stratification associated with deeper dives. Filter feeding whale sharks also show a high diversity in vertical activity and habitat use (Gleiss et al., 2013). This likely reflects differences in prey availability with depth although more information on prey distribution is needed.

Vertical habitat use offshore also varied. One explanation for the difference between sharks C and D may be the shoaling of the oxygen minimum zone along the Baja Peninsula, Mexico. Low oxygen has been shown to constrain the vertical movements of a number of pelagic fish species (Carey and Robison, 1981; Brill, 1994; Prince and Goodyear, 2006; Nasby-Lucas et al., 2009). Differences may also be linked to behavioral thermoregulation with shark C staying deep to avoid warm surface waters, which has also been observed in other pelagic fish (Musyl et al., 2004; Weng et al., 2005; Teo et al., 2007). The SST in the area northeast of Hawaii was near the maximum SST of 24◦C reported by Owen (Owen, 1984; see below). As in the nearshore, offshore habitat use is likely influenced by both oceanography and prey availability.

Similar to other diel migrators, shark C is likely targeting organisms associated with the sound-scattering layer (Carey and Robison, 1981; Musyl et al., 2004; Dewar et al., 2011). Based on the acoustic backscatter data, the basking shark appeared to be following the same portion of the sound-scattering layer as it vertically migrated (**Figure 8**). Unfortunately, which taxa they were targeting is not known. While copepods are thought to be the preferred prey (Baduini, 1995; Sims and Reid, 2002; Siders et al., 2013; Curtis et al., 2014), stomach contents from basking sharks foraging in deep waters in the Northwest Pacific Oceans included small crustaceans of up to 5.4 cm (Mutoh and Omori, 1978). One factor that suggests the basking shark was targeting copepods is the early ascent prior to dusk, which is unusual for pelagic fish that target the sound-scattering layer (Carey and Robison, 1981; Carey and Scharold, 1990; Musyl et al., 2004; Dewar et al., 2011). The copepod, C. pacificus, has been observed to begin their ascent one to two h before sunset (Enright and Honegger, 1977) depending on environmental conditions. Another potential reason for the shark's early ascent includes behavioral thermoregulation. Given the up to 10◦C increase in temperature (**Figure 6**), digestion could be increased by twofold using a standard Q<sup>10</sup> temperature coefficient of two. It is not likely linked to hypoxia as oxygen levels off Hawaii are not depleted at these depths. Additional studies including stable isotopes should help resolve offshore foraging habits. Regardless, results suggest a direct link between basking shark vertical movements and a specific portion of the sound-scattering layer.

A pattern apparent both near- and offshore was the consistent relationship between SST and dT. The overall implication is that as SST increases the sharks' vertical niche expands depending on water column characteristics, and opportunities to forage at depth increase. To determine the driving mechanism behind this pattern, a broader comparison examining behaviors, prey availability, bathymetry and water column characteristics across regions is needed (Dewar et al., 2011). While the underlying mechanism is not clear, the ability to predict dT from SST is useful for predicting vertical habitat use and could be used to inform regional vulnerability to fishing gear.

#### Insight Into Tagging Methods

Results also inform electronic tag selection and deployment for basking sharks. While our sample size is too small to be definitive, the spear-point and metal dart provided long-term deployments whereas the PM dart did not. The lack of uplinks and low number of GPS locations indicate that a less expensive PSAT would provide a similar dataset possibly over longer durations (Skomal et al., 2009; Braun et al., 2018).

#### Conservation and Management Implications

As mentioned above, the basking shark population in the ENP is considered to be at a historic low even though targeted removals in the U.S. and Canada ended more than 40 years ago (Phillips, 1948; Darling and Keogh, 1994; COSEWIC, 2007; McFarlane et al., 2009). The decline in the population is likely linked to basking sharks' low intrinsic population growth rates (Compagno, 1984; Squire, 1990; Smith et al., 1998, 2008; McFarlane et al., 2009) that may be compounded by allee effects that act to reduce population growth rates at very low population sizes (Gilpin and Soule, 1986; Dennis, 1989). Another factor to consider is fisheries mortality, both targeted and incidental. However, to accurately determine population status and assess sources of mortality, directed study is required.

There are a number of potential sources of mortality. Given nearshore foraging behaviors, basking sharks are prone to ship strikes and becoming entangled in fishing gear (Darling and Keogh, 1994; COSEWIC, 2007; Larese and Coan, 2008; McFarlane et al., 2009). Off the U.S. West Coast, the majority of bycatch occurred in the large-mesh, drift-gillnet fishery with most takes occurring in the 1980's (Larese and Coan, 2008). Since that time, regulatory mandates including timearea closures and gear modifications to protect sea turtles and marine mammals (Larese and Coan, 2008) likely also protected basking sharks. Off Mexico, while commercial drift gillnet gear is currently prohibited, basking sharks have been taken by artisanal fishers, (Sandoval-Castillo and Ramirez-Gonzalez, 2008) although interactions are rare (Sosa-Nishizaki, pers. comm.). Off Canada, encounters are now also rare and only three sharks have been taken incidentally in the ground-fish trawl fishery since 1996 (COSEWIC, 2007; McFarlane et al., 2009). Observed bycatch in the U.S., Mexico and Canada, at least over the last 20–30 years, has been low.

Of greater concern is the potential for basking sharks to be taken outside the EEZs of the U.S., Mexico and Canada. Given their potential for large-scale movements (this study, Gore et al., 2008; Skomal et al., 2009; Braun et al., 2018), it is likely that the range of basking sharks in the ENP extends into the Central Pacific Ocean and possibly into the Northwest Pacific. A harpoon fishery operated off Japan from the 1700's until 1980 when sightings declined to only a few per year (CITES, 2002). They may also be vulnerable in the Central Pacific, although their propensity to remain deep may provide some protection. Incidental take in the high-seas, large-mesh, driftnet fisheries that operated in the Central North Pacific Ocean from the 1980's until 1994 was estimated at 54 per year (Bonfil, 1994). While the origin is not known, it is clear that undocumented basking shark take continues. The number of marketed fins is more than the number of sharks accounted for in CITES trade documents (Magnussen et al., 2007). One fin can have a value of over 50,000 \$US providing a strong incentive for targeting or retaining basking sharks.

While some populations in the Atlantic may be recovering (Witt et al., 2012), this does not appear to be the case in the North Pacific (Squire, 1990; Darling and Keogh, 1994; COSEWIC, 2007; McFarlane et al., 2009). The listing of basking sharks by numerous national and international bodies indicates a broad concern for the species regionally and globally. Given the high price for basking shark fins, efforts should focus on reducing demand, enforcing existing regulations, better documenting trade, and reducing mortality.

### AUTHOR CONTRIBUTIONS

HD, SW, JH, OS, SK contributed to the conception, design and implementation of the study; HD wrote the first draft of the manuscript; AL, CL, RD, SB wrote sections of the manuscript; JW, RD, and AL performed analyses for different section; SV contributed to the design and implementation of the study. All authors contributed to manuscript revision, read and approved the submitted version.

#### ACKNOWLEDGMENTS

We thank the members of the public and the recreational fishers for their reports that lead to tag deployments including BDoutdoors.com. The NOAA Species of Concern Program provided funding to purchase the tags and for tag deployments. We also thank our U.S., Mexican and Canadian colleagues for their efforts on the ENP Tri-National Basking Shark Working Group.

#### REFERENCES


### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00163/full#supplementary-material

Supplemental Figure 1 | Thermal profiles. Thermal profiles obtained from the PDT data collected during periods when shark A made dives that were shallower (June 21–23, gray diamonds) and deeper (July 9–10, black squares). Note, shark A remained near the coast throughout its deployment.

Supplemental Figure 2 | Depth histograms. Depth histograms from (A) shark C and (B) shark D. Data are aggregated by month and indicated by shading (6 = June, 7 = July, etc...).


question current population structure paradigm. Can. J. Fish. Aquat. Sci. 67, 966–976. doi: 10.1139/F10-033


oceanographic drivers of habitat use for a pelagic seabird. J. R. Soc. Interface 11:20140679. doi: 10.1098/rsif.2014.0679


**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 Dewar, Wilson, Hyde, Snodgrass, Leising, Lam, Domokos, Wraith, Bograd, Van Sommeran and Kohin. 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 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.

# Integrating Dynamic Subsurface Habitat Metrics Into Species Distribution Models

Stephanie Brodie1,2 \*, Michael G. Jacox 1,2,3, Steven J. Bograd1,2, Heather Welch1,2 , Heidi Dewar <sup>4</sup> , Kylie L. Scales <sup>5</sup> , Sara M. Maxwell <sup>6</sup> , Dana M. Briscoe<sup>1</sup> , Christopher A. Edwards <sup>1</sup> , Larry B. Crowder <sup>7</sup> , Rebecca L. Lewison<sup>8</sup> and Elliott L. Hazen1,2

*1 Institute of Marine Science, University of California Santa Cruz, Santa Cruz, CA, United States, <sup>2</sup> Environmental Research Division, NOAA Southwest Fisheries Science Center, Monterey, CA, United States, <sup>3</sup> Physical Sciences Division, Earth System Research Laboratory (NOAA), Boulder, CO, United States, <sup>4</sup> NOAA Southwest Fisheries Science Center, La Jolla, CA, United States, <sup>5</sup> School of Science and Engineering, University of the Sunshine Coast, Maroochydore, QLD, Australia, <sup>6</sup> Department of Biological Sciences, Old Dominion University, Norfolk, VA, United States, <sup>7</sup> Hopkins Marine Station, Stanford University, Pacific Grove, CA, United States, <sup>8</sup> Coastal and Marine Institute, San Diego State University, San Diego, CA, United States*

#### Edited by:

*Lisa Marie Komoroske, National Oceanic and Atmospheric Administration (NOAA), United States*

#### Reviewed by:

*Antonios D. Mazaris, Aristotle University of Thessaloniki, Greece Jonathan D. R. Houghton, Queen's University Belfast, United Kingdom*

> \*Correspondence: *Stephanie Brodie sbrodie@ucsc.edu*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *31 January 2018* Accepted: *05 June 2018* Published: *25 June 2018*

#### Citation:

*Brodie S, Jacox MG, Bograd SJ, Welch H, Dewar H, Scales KL, Maxwell SM, Briscoe DM, Edwards CA, Crowder LB, Lewison RL and Hazen EL (2018) Integrating Dynamic Subsurface Habitat Metrics Into Species Distribution Models. Front. Mar. Sci. 5:219. doi: 10.3389/fmars.2018.00219* Species distribution models (SDMs) have become key tools for describing and predicting species habitats. In the marine domain, environmental data used in modeling species distributions are often remotely sensed, and as such have limited capacity for interpreting the vertical structure of the water column, or are sampled *in situ*, offering minimal spatial and temporal coverage. Advances in ocean models have improved our capacity to explore subsurface ocean features, yet there has been limited integration of such features in SDMs. Using output from a data-assimilative configuration of the Regional Ocean Modeling System, we examine the effect of including dynamic subsurface variables in SDMs to describe the habitats of four pelagic predators in the California Current System (swordfish *Xiphias gladius*, blue sharks *Prionace glauca*, common thresher sharks *Alopias vulpinus*, and shortfin mako sharks *Isurus oxyrinchus*). Species data were obtained from the California Drift Gillnet observer program (1997–2017). We used boosted regression trees to explore the incremental improvement enabled by dynamic subsurface variables that quantify the structure and stability of the water column: isothermal layer depth and bulk buoyancy frequency. The inclusion of these dynamic subsurface variables significantly improved model explanatory power for most species. Model predictive performance also significantly improved, but only for species that had strong affiliations with dynamic variables (swordfish and shortfin mako sharks) rather than static variables (blue sharks and common thresher sharks). Geospatial predictions for all species showed the integration of isothermal layer depth and bulk buoyancy frequency contributed value at the mesoscale level (<100 km) and varied spatially throughout the study domain. These results highlight the utility of including dynamic subsurface variables in SDM development and support the continuing ecological use of biophysical output from ocean circulation models.

Keywords: species distribution modeling, ocean circulation models, remote sensing, spatial ecology, top predator, ROMS, boosted regression trees

## INTRODUCTION

Species distribution models (SDMs) have become a common method to study species spatial ecology, often to support environmental management and conservation (Robinson et al., 2011, 2017). Building SDMs and exploring resultant environmental drivers requires data on species' presence and corresponding environmental information (Elith and Leathwick, 2009; Robinson et al., 2017). In the marine realm, such corresponding environmental information can be obtained from satellite platforms, in situ sources (i.e., data loggers, moorings, under sea vehicles, surveys), and ocean circulation models. Data-assimilative ocean circulation models incorporate available environmental information from satellite and in situ platforms while also adding value in the form of increased spatial and temporal data resolution and elimination of data gaps. Importantly, ocean circulation models can also provide spatiotemporal resolution of the vertical structure of the ocean. These benefits have resulted in the increasing use of ocean circulation models in SDM development (Becker et al., 2016; Scales et al., 2017b).

SDM applications are often used to understand and predict the horizontal and/or vertical spatiotemporal distribution of species (Guisan and Thuiller, 2005; Elith and Leathwick, 2009; Robinson et al., 2017). For marine species, there can often be separation in temporal scales when comparing horizontal and vertical distributions. For example, vertical distribution often reflects behavior on short temporal scales (e.g., diving and foraging occur from minutes to hours), whereas changes in horizontal distributions generally reflect longer temporal scales (e.g., habitat use and migratory behavior occurring over days to months) (Block et al., 2011; Bestley et al., 2015). These vertical and horizontal movements are often linked, and thus integrating both horizontal and vertical dimensions in SDMs could result in model improvement provided that appropriate data are available. The availability of subsurface data from ocean circulation models, compels the need to explore what, if any, benefit comes from integrating vertical biophysical features in SDMs.

Species occurrence data appropriate for use in SDM development can encompass a variety of spatiotemporal scales (Elith et al., 2006). Many sources of marine species occurrence data are not vertically informed, such as fisheries catch data where there is no available information of the depth of catch (Brodie et al., 2015). Despite this data limitation, subsurface metrics that characterize the physical structure of the water column on a horizontal plane can be informative in SDMs. For example, previous studies have used the climatology of the mixed layer depth as a variable (e.g., Dell et al., 2011; Carlisle et al., 2017). Mixed layer depth is an important characteristic of the vertical water column structure (typically 25–200 m depth; Kara et al., 2003) that effects the vertical distribution of nutrients, plankton and corresponding higher trophic levels (Huisman et al., 2006; Behrenfeld and Boss, 2014; Schroeder et al., 2014). The use of climatological subsurface data is common but does not reflect the contemporaneous conditions animals experience and instead indicates the long-term state of the environment (Mannocci et al., 2017). Ocean circulation models, in contrast, can provide subsurface biophysical data at scales contemporaneous to species occurrence data (Scales et al., 2017b).

The goal of this study is to explore the effect of integrating dynamic, model-derived, subsurface variables into SDMs, and the persistence of effect across species. We do this by comparing three SDM simulations: (1) models that only use static variables; (2) models that use a combination of static and dynamic variables but no vertical variables; and (3) models that use a combination of static and dynamic variables, including vertical variables. Here, simulation 1 is not intended as an appropriate option for building a SDM, but rather provides a null model to comparatively assess the value of no dynamic environmental information, while simulation 2 assesses the added value of dynamic surface variables, and finally simulation 3 assesses the added value from dynamic subsurface variables. We use presence-absence catch data for four comparative species that co-exist in the California Current study region: swordfish Xiphias gladius, blue sharks Prionace glauca, common thresher sharks Alopias vulpinus, and shortfin mako sharks Isurus oxyrinchus. The use of four species allows for a broader comparison of the effect of vertical variables in SDMs. Based on existing knowledge of distribution, behavior, diet, and physiology, these species are known to differ in their horizontal and vertical habitat use (**Table 1**). All four study species exhibit some degree of diel vertical migration (**Table 1**), a behavioral phenomenon in pelagic ecosystems where oceanic organisms ascend to the photic zone during night and descend to mesopelagic depths (typically 200–1,000 m) during the day (Robison, 2004). As a result of this daily migration, it is likely that subsurface variables will be important in structuring the horizontal distribution of the study species. We explore the effect of two subsurface variables that quantify the structure and stability of the water column, isothermal layer depth and bulk buoyancy frequency. Given the role subsurface features have on structuring ecosystems, understanding the contribution of subsurface variables on SDM power and performance of pelagic species can further demonstrate the utility of SDMs in support of marine conservation and spatial planning efforts.

### METHODS

### Species and Environmental Data

Species occurrence data were obtained from the NOAA fisheries observer program from the California drift gillnet fishery, which operates at night along the US West Coast (**Figure 1**). This fishery targets swordfish, but also retains other species including common thresher sharks and shortfin mako sharks. Catch of all species, including bycaught species such as blue sharks, are recorded in the observer program which operates at ∼15% coverage across the fishery. The catch data contained the presence or absence of an animal in each set, with species size data not universally available for analysis. The data were temporally limited to 1997 through 2017 to match the availability of environmental data, namely satellite-derived chlorophyll-a (**Table 1**). All analyses described below were performed using R statistical computing (R Core Team, 2017).

A total of 16 environmental variables were available for inclusion in species distribution models, which included three


TABLE 1 | Ecological comparison of swordfish, blue sharks, common thresher sharks, and shortfin mako sharks based on published sources.

*Metrics of comparison include stock status, International Union for Conservation of Nature (IUCN) global conservation status, typical depth occupied, diel vertical behavior, endothermy, diet, and observed temperature range. Where possible, information specific to the California study region was used. IUCN status accesses The IUCN Red List of Threatened Species. Version 2017-3.* <*www.iucnredlist.org*>*. Downloaded on 05 January 2018. <sup>a</sup> ISC (2014b); <sup>b</sup>NOAA (2017); <sup>c</sup>Sepulveda et al. (2010); <sup>d</sup>Sepulveda et al. (2018); <sup>e</sup>Abecassis et al. (2012); <sup>f</sup>Dewar et al. (2011); <sup>g</sup>De Metrio et al. (1997); <sup>h</sup>Markaida and Sosa-Nishizaki (1998); <sup>i</sup>Markaida and Hochberg (2005); <sup>j</sup>Young et al. (2006); k ISC (2014a); <sup>l</sup>Musyl et al. (2011); <sup>m</sup>Preti et al. (2012); <sup>n</sup>Kubodera et al. (2007); <sup>o</sup>Maxwell et al. (in review); <sup>p</sup>Teo et al. (2016); <sup>q</sup>Heberer et al. (2010); <sup>r</sup>Cartamil et al. (2011); <sup>s</sup>Bernal and Sepulveda (2005); <sup>t</sup>Syme and Shadwick (2011); u ISC (2015); <sup>v</sup>Abascal et al. (2011); <sup>w</sup>Bernal et al. (2001).*

Temperature range (◦C) 3–29<sup>e</sup> 9–27l,o 9–21<sup>q</sup> 4–25l,v

teleosts, and myctophidsm,n

static variables, 11 dynamic surface variables, and two dynamic subsurface variables (**Table 2**). Three static variables included bathymetry (z; ETOPO1 obtained from https://www.ngdc. noaa.gov/mgg/global/global.html, interpolated to 0.1◦ ), rugosity (z\_sd; calculated as the standard deviation of z over a 0.3◦ square), and lunar illumination from the "lunar" package in R. Lunar illumination was included in the static grouping as it does not require observation due to its definitive cyclical nature. Lunar illumination was chosen as the fishery gear is deployed at night and illumination is known to affect the vertical distribution of swordfish (Sepulveda et al., 2010; Lerner et al., 2012; Scales et al., 2017b). The majority of dynamic environmental data was sourced from daily fields of an ocean circulation model, namely a data assimilative configuration of the Regional Ocean Modeling System (ROMS) that covers the California Current System from 30 to 48 ◦N and from the coast to 134 ◦W at 0.1◦ (∼10 km) horizontal resolution (http:// oceanmodeling.ucsc.edu/ccsnrt version 2016a; Neveu et al., 2016). The ROMS 0.1◦ spatial resolution is considered sufficient for habitat modeling as this spatial scale combined with a fine temporal scale (daily) acts to minimize bias in similar species distribution models (Scales et al., 2017a). Vertical structure in the ROMS model is resolved by 42 terrain-following vertical levels (Veneziani et al., 2009). Importantly, the ROMS model employed here assimilates available data from satellites and in situ platforms (e.g., ships, moorings, buoys) to provide environmental information that is better than either the model or the observations in isolation. The temporal scale of our study spanned two ROMS iterations, a historical re-analysis (1980– 2010; Neveu et al., 2016) and a near real-time product (2011 present) and as such all ROMS variables were assessed for consistency across the two temporal periods using a time-series analysis.

Eleven dynamic surface ROMS variables chosen included sea surface temperature (SST) and its standard deviation (SST\_sd; calculated over a 0.3◦ square), sea surface height (SSH) and its standard deviation (SSH\_sd; calculated over a 0.3◦ square), surface eastward and northward velocity (su; sv), surface eastward and northward wind stress (sustr; svstr), wind stress curl, and eddy kinetic energy (EKE) (**Table 2**). We also included chlorophyll-a as a dynamic surface variable, but as the assimilative ROMS model is purely physical, we used a combination of satellite-derived products (SeaWiFS and Aqua MODIS, distributed by NASA and obtained from SWFSC Environmental Research Division's ERDDAP; Simons, 2017). Chlorophyll-a and EKE were highly right skewed and were log<sup>e</sup> transformed prior to analysis.

Dynamic subsurface ROMS variables included isothermal layer depth (ILD) and bulk buoyancy frequency (BBV, also known as Brunt-Väisälä frequency). Both variables provide indices of water column structure, respectively the depth of surface mixing and degree of stratification in the upper water column. ILD (m) was calculated as the depth corresponding to a 0.5◦C temperature difference relative to sea surface temperature (Monterey and Levitus, 1997). Daily mean surface temperature was used as a reference point because the typical 10 m reference point is not suitable for regions with strong upwelling, like the California Current (de Boyer Montégut et al., 2004). ILD provides a daily horizontal field (0.1◦ resolution) comparable to dynamic surface variables (Scales et al., 2017b). BBV (s−<sup>1</sup> ) offers a measure of the upper water column stability and was averaged over the upper 200 m of the water column to produce a daily horizontal

field (0.1◦ resolution), where higher BBV values indicate a more stable water column. In areas <200 m deep, BBV was averaged over the entire water column. The two-dimensional structure of ILD and BBV described water column properties best suited SDM development as the catch data used here were not vertically informed (i.e., depth of catch).

#### Species Distribution Models

Three species distribution models were built for each species (swordfish, blue sharks, common thresher sharks, and shortfin mako sharks) using fishery catch data. The probability of species presence was modeled as a function of environmental variables (described above) using a boosted regression tree (BRT) framework from the "dismo" R package (Elith et al., 2008). Three combinations of environmental variables were used to explore the importance of dynamic vertical variables on models. Simulation 1: static only variables (z, z\_sd, lunar); simulation 2: static and dynamic surface variables with no vertical variables; simulation 3: models with static and dynamic surface variables and vertical variables (**Table 2**). Co-linearity between environmental variables was not a prohibitive issue as the BRT framework automatically handles any co-linearity effects (Elith et al., 2008). All BRT models were built using a Bernoulli family appropriate to the response variable of presence (1) and absence (0). The BRTs had a learning rate of 0.01, a tree complexity of 3, and a bag fraction of 0.6 (Elith et al., 2008). The resultant species distribution models describe the probability of species presence, here termed "habitat suitability" due to our use of fisheries catch data as a response variable.

Species distribution models were evaluated using explained deviance, Area Under the receiver operating Curve (AUC), and true skill statistic (TSS). Explained deviance (%) gives an indication of how well the model explains the data, while AUC and TSS assess the predictive performance of models on new data. Explained deviance was calculated as an average of 50 model iterations. This evaluation approach is possible because the bag fraction (0.6) of each model ensures a random selection of data to each tree, resulting in each model iteration being unique. The AUC and TSS were calculated as the average of 50 model iterations, where each iteration was built using 75% of the data and assessed against the remaining 25% of the data. This approach for model evaluation was done for each species (n = 4) and each model simulation (n = 3) and used a learning rate of 0.001 to improve convergence on the reduced dataset (75%). Differences in evaluation metrics between model simulations were assessed for significance using ANOVA and a Tukey Honest Significance Differences (HSD) post-hoc test (Fournier et al., 2017) in the R "stats" package (R Core Team, 2017).

Final SDMs were used to predict and visualize species habitat suitability on two example days, 1st December 2012 and 2015. December was chosen as fishery catch peaks during this month (Urbisci et al., 2016). 2012 represents a neutral year in the California Current ecosystem, with no strong ENSO influences (Bjorkstedt et al., 2012), while 2015 was a year when the California Current was strongly affected by the combination of an El Niño and pre-existing warm anomalies (Jacox et al., 2016). Habitat suitability predictions were made for each species (n = 4), and for model simulations 2 and 3 (dynamic simulations with and without vertical variables). The differences in habitat suitability between simulations 2 and 3 were calculated to visualize the effect of adding vertical variables into models.

### RESULTS

Species catch data along the California coast were not evenly distributed, with the majority of effort concentrated in the southern region (**Figure 1**). A total of 4,719 drift gillnet sets were used in the analysis. There were more swordfish present (n = 2986) than absent in these sets. In contrast, there were more shark species absent than present (blue sharks n = 2163; common thresher sharks n = 1074; shortfin mako sharks n = 2048) in sets.

Species distribution models revealed complex relationships among species presence and environmental variables. The relative importance of each variable varied among species, but the dynamic subsurface variables (BBV and ILD) ranked in the top six variables across all species (**Figure 2**; Table S1). The contribution of BBV and ILD was most prevalent in the swordfish model relative to the shark models (**Figure 2**).

TABLE 2 | Summary of 16 environmental variables included in species distribution models.


*Simulation number indicates which variables are included in each of the three model simulations. ROMS data obtained from http://oceanmodeling.ucsc.edu/ccsnrt version 2016a (Neveu et al., 2016). ETOPO1 obtained from https://www.ngdc.noaa.gov/mgg/global/global.html. Chlorophyll-a data obtained from SeaWiFS and Aqua MODIS, distributed by NASA and obtained from SWFSC Environmental Research Division's ERDDAP (Simons, 2017).*

Variable response curves revealed how the probability of species presence was influenced by each variable (**Figure 3**; Figure S1). For brevity, we describe results for the two vertical variables (ILD, BBV; **Figure 3**), with the remaining variable response curves provided in the Supplementary Material (Figure S1). For swordfish, the probability of presence had a positive correlation with ILD (peaking 40–120 m) and a non-monotonic correlation with BBV (preference between 0.009 and 0.013 s −1 ). For blue sharks, the probability of presence had a positive correlation with ILD (peaking 40–120 m) and a positive correlation with BBV (plateauing at 0.009 s−<sup>1</sup> ). For shortfin mako sharks, the probability of presence had a positive correlation with ILD (less steep slope between 20 and 120 m) and a negative correlation with BBV (plateauing at 0.009 s−<sup>1</sup> ). For common thresher sharks, the probability of presence had a positive correlation with ILD (sharp increase >70 m) and a complex nonlinear correlation with BBV (two troughs at 0.009 and 0.013 s−<sup>1</sup> ).

The addition of dynamic surface and subsurface variables (simulations 2 and 3) to the static model (simulation 1) significantly improved model explanatory power and predictive performance across all species (**Figure 4**; **Table 3**; p < 0.001). Furthermore, the addition of vertical variables (simulation 3) to a non-vertical variable model (simulation 2) typically increased model explanatory power and predictive performance (**Figure 4**; **Table 3**). However, post-hoc analyses revealed the improvement in predictive performance between simulation 2 and 3 was not statistically significant for species with strong responses to bathymetry (blue sharks and common thresher sharks; **Figure 4**). Further, blue sharks were the only species where adding vertical variables (simulation 3) did not significantly improve explained deviance (**Figure 4**; **Table 3**). The addition of vertical variables (simulation 3) to a non-vertical variable model (simulation 2) also resulted in changes to dynamic variable response curves (Figure S2).

Predicted species' habitat suitability for 1st December 2012 and 2015 revealed spatial differences among species, with common thresher shark habitat predicted more suitable inshore of the 1,000 m isobath and the other three species habitat predicted to be more suitable offshore of the 1,000 m isobath and in the Southern California Bight (**Figure 5**; Figure S3). The effect of adding of ILD and BBV to species distribution models was evident in each species spatial prediction (**Figure 5**; Figure S3). Differences in the predicted habitat suitability between simulations 2 and 3 indicated that the vertical variables ILD and BBV contributed to model predictions primarily at a sub-mesoscale level (<100 km). Comparison of species spatial predictions between a neutral year (2012; **Figure 5**) and an El Niño year (2015; Figure S3) revealed that during 2015 all species predicted habitat expanded, and the contribution of ILD and BBV was more prominent in the swordfish and blue shark models.

#### DISCUSSION

Species distribution models (SDMs) are increasingly used to characterize and understand the distributions of marine species. The prevalence of vertical movement behavior in pelagic top predators substantiates the importance of integrating vertical water column structure into SDMs. Using evaluation metrics for SDMs, we showed that integrating dynamic subsurface variables increased explanatory power across all species models, although the degree of improvement to model predictive

simulation 2 (static and dynamic surface variables); and simulation 3 (static and dynamic surface and subsurface variables). Species' model simulations were run 50 times to generate mean ± S.D, with significance denoted by letters (ANOVA and Tukey HSD, *p* < 0.05). Models for each species are shown: swordfish (yellow), blue sharks (blue), common thresher sharks (green), and shortfin mako sharks (red). Letters are only comparable among simulations of the same species.


TABLE 3 | ANOVA and Tukey HSD *post-hoc* results from species distribution model simulation comparison.

*Values are the mean* ± *SD explained deviance (%), AUC, and TSS generated from 50 model iterations for simulation 1 (static variables), simulation 2 (no vertical variables), and simulation 3 (including vertical variables). ANOVA F and p-values for the main effect of simulation are shown, with the post-hoc results denoted by letters.*

performance was species-dependent. Inter-specific variability in results is likely a result of individual species ecology, where species with strong correlations with static variables were less affected by the addition of dynamic subsurface variables (e.g., blue sharks). The positive effect of including dynamic subsurface variables supports their utility in marine SDM development and encourages further exploration of the vertical dimension to species' horizontal distributions and movements. The use of dynamic subsurface variables in SDM development provided further understanding of the processes driving species distributions, which has clear implications for the ongoing management and conservation of marine megafauna.

#### Species Ecological Responses to Variables

The SDMs built here reveal the complex responses four top predators have to static and dynamic environmental variables. The wide range of variables used in model development represent direct (e.g., temperature-dependent physiological effects; Altringham and Block, 1997; Brown, 2004) and indirect (e.g., chlorophyll-a as an indicator of productivity; Armstrong et al., 1995; Polovina et al., 2008) effects on the distribution of these top predators. Here, the dynamic subsurface variables included in SDMs, namely bulk buoyancy frequency (BBV) and isothermal layer depth (ILD), likely have direct and indirect effects on top predator distribution and therefore influence their susceptibility to catch (see below). These variables characterized subsurface water properties on a horizontal plane, and in doing so indicated relative temperature (direct effects) and availability of prey fields (indirect effects).

Bulk buoyancy frequency (BBV) quantifies water column stability, where high BBV values reflect a more stable water column. Increased stability (a more stratified water column) acts as a barrier for upward nutrient flux into the photic zone, affecting the productivity and distribution of prey fields (Haug et al., 1986; Susini-Ribeiro et al., 2013; Behrenfeld and Boss, 2014). Here, top predator response to BBV differs among species, with swordfish, blue sharks, and shortfin mako sharks showing an increased probability of occurrence in areas with intermediate BBV values. Lower BBV values correspond to waters that are highly mixed and less stable, typical of upwelled waters that are cold and oxygen poor (Grantham et al., 2004). While these species are physiologically and biomechanically equipped for foraging in cold and hypoxic waters (**Table 1**; Dickson and Graham, 2004; Wegner et al., 2010; Abecassis et al., 2012), surface waters support recovery from deep vertical dives by allowing thermal regulation of body temperature and a reduction of oxygen debt (Dagorn et al., 2000; Dewar et al., 2011). As a result, intermediate BBV levels may provide a middle ground between prey availability (low BBV values) and suitability of surface waters for recovering from vertical movements (high BBV values). In contrast, common thresher sharks preferred extreme BBV values (low and high), which likely reflects their predominantly coastal distribution (shallow bathymetry preference) as such areas can have the most extreme BBV values (**Figure 5**).

Isothermal layer depth (ILD) was an important variable in the SDMs, and its relative influence was similar among species. Isothermal layer depth is a proxy for mixed layer depth, and indicates the depth where physical water properties (i.e., temperature, salinity, nutrients, oxygen) change dramatically (Robison, 2004). All four species showed a preference for waters with deeper ILDs which indicates a thick homogenous surface layer in the epipelagic zone where temperature and oxygen are higher. This surface layer may provide a thermal and oxygen refuge for pelagic predators (Prince and Goodyear, 2006; Dewar et al., 2011; Carlisle et al., 2017), as many pelagic

FIGURE 5 | Predicted habitat suitability for each species for an example day, 1 December 2012. The first row shows predicted habitats using simulation 2 (static and dynamic surface variables), the second row shows predicted habitats using simulation 3 (static and dynamic surface and subsurface variables), and the third row shows the difference in probabilities between simulations 2 and 3. The fourth row shows four example dynamic variables for the same day (Sea Surface Temperature, Sea Surface Height, Bulk Buoyancy Frequency, and Isothermal Layer Depth). Species are indicated by a black silhouette: swordfish (first column), blue sharks (second column), common thresher sharks (third column), and shortfin mako sharks (fourth column). Contours on the first and second row are at 0.6 and 0.2, contours on the third row are at 0.1 and −0.1, and contours on the fourth row equate to the 25 and 75% quantiles.

predators, including the study species, spend much of their time in surface waters despite foraging in waters below the mixed layer (**Table 1**). The common thresher shark partial response curve showed a unique response to ILD, which appears to be related to a weak negative correlation between ILD and SST (−0.54 Pearson correlation coefficient). As SST has a greater relative influence on common thresher sharks than ILD, the preference for colder SST values better describes occurrence, which results in no strong pattern seen with ILD < 70m. This disconnect between partial effect curves is typical, and while plot interpretation can be challenging when variables are correlated, these plots represent an effective way of visualizing the effects of each variable (Elith et al., 2008). Given the response common thresher sharks showed here, future work could explore the utility of other subsurface variables in describing their habitat suitability, including model-based upwelling indices (e.g., Jacox et al., 2014) that would spatially align with their predominantly coastal distribution.

### Integration of the Vertical Dimension in SDMs

Integrating dynamic subsurface variables into marine SDMs had a positive effect on describing the horizontal habitat suitability of four pelagic species. The mechanism behind this result is likely a combination of the main effect of ILD and BBV within the model framework, as well as the covariation with other variables (Figure S2). This co-variation occurs as a result of including multiple variables in the statistical framework. While certain variables had a higher relative effect than others, no single variable could perform as well as the multi-variable models, or even perform at a standard required for conservation planning (AUC > 0.75; Table S2) (Pearce and Ferrier, 2000). There is a trade-off in the number of variables to include in a SDM, where a simple model is easier to interpret but may come at a cost of decreased predictive performance, while a complex model is challenging to interpret ecologically (and especially as response curves change) but may have increased predictive performance (Friedman et al., 2001). Furthermore, there is potential for overfitting to occur as the number of variables included in SDMs increases, however regularization methods advised for boosted regression trees reduces the risk of overfitting (Elith et al., 2008). Future research could build on our results by including additional species and additional model frameworks (e.g., generalized linear and additive models; machine learning). There is further scope to explore subsurface variables in SDMs (e.g., oxygen) and we acknowledge that the subsurface variables explored here (BBV and ILD) may not be sufficient for all marine species. Typically, environmental variables included in SDMs are best informed from a priori expectations based on species ecology (Fourcade et al., 2018).

Using ocean circulation models for SDM development can maximize the use of catch data and allow model prediction to be done on large spatiotemporal scales. However, not all ocean circulation models are equal and care must be taken to ensure outputs are appropriate for use. The dynamic surface variables obtained from this ocean circulation model are a bestcase scenario of data availability, in that data from satellites and quarterly in situ data surveys are incorporated (data assimilative) but typical issues with satellite-derived information are avoided (i.e., cloud cover, patchiness, resolution mismatch, temporal span of products; Scales et al., 2017a). Ocean circulation models also provide a consistent framework to access data across periods of changing observational assets (i.e., different satellite eras). Output from regional ocean models, and especially data-assimilative models, is unfortunately limited to certain regions and time periods such that its use will be precluded in some SDM development. However, when data are available for the time period and spatial domain of interest, the added benefits of using ocean circulation data can be powerful (Becker et al., 2016; Scales et al., 2017b).

Integrating dynamic subsurface variables improved the explanatory power and predictive performance of SDMs for highly migratory species. The benefits to model explanatory power support the future use and inclusion of such variables, where possible, to get the best ecological understanding of the environmental drivers on species distributions. Improvements to predictive performance, while significant, were not large for a model that already incorporates many dynamic surface variables. For an operational version of such a model (e.g., Hazen et al., 2018) the benefits of including subsurface variables should be weighed against the resources needed to obtain them and evaluate their contribution. However, more generally there is added benefit in using ocean circulation models whether variables are vertical or horizontal, or both—in SDM applications, as ocean circulation models: (i) eliminate data gaps that are prevalent in satellite data and in situ sources; (ii) provide continuity across periods of changing observational assets; (iii) provide all variables at common spatial and temporal resolutions; and (iv) can be configured to predict into the future. Operational models require continuous collection and collation of data products, a process that is greatly streamlined by having a single source for ocean circulation model output rather than multiple remotely sensed data providers. This improved efficacy can support conservation planning, decision-making, and management (Hobday et al., 2018; Stelzenmüller et al., 2018) on near real-time (Maxwell et al., 2015), seasonal (Brodie et al., 2017), and longer timescales (Almpanidou et al., 2016; Ban et al., 2016).

### AUTHOR CONTRIBUTIONS

SB, MJ, SJB, HW, HD, KS, SM, DB, CE, LC, RL, and EH contributed to writing and reviewing of the manuscript. SB, MJ, SJB, HW, KS, and EH contributed to project design. SB, MJ, CE, HD, KS, DB, and SM contributed to data collation. SB conducted analyses and prepared manuscript figures.

### ACKNOWLEDGMENTS

Funding was provided by NASA Earth Science Division/Applied Sciences ROSES Program (NNH12ZDA001N-ECOF); NOAA Modeling, Analysis, Predictions and Projections MAPP Program (NA17OAR4310108); NOAA Coastal and Ocean Climate Application COCA Program (NA17OAR4310268); NOAA's Integrated Ecosystem Assessment program; NOAA Bycatch Reduction Engineering Program Funding Opportunity (NA14NMF4720312); NOAA NMFS Office of Science and Technology; and the California Sea Grant Program (NA140AR4170075). We also thank Lucie Hazen for project management, and Suzy Kohin for assistance and advice on data compilation.

### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00219/full#supplementary-material

#### REFERENCES


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

Copyright © 2018 Brodie, Jacox, Bograd, Welch, Dewar, Scales, Maxwell, Briscoe, Edwards, Crowder, Lewison and Hazen. 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 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.

# Embracing Complexity and Complexity-Awareness in Marine Megafauna Conservation and Research

#### Rebecca L. Lewison<sup>1</sup> \*, Andrew F. Johnson<sup>2</sup> and Gregory M. Verutes <sup>3</sup>

*<sup>1</sup> Department of Biology, San Diego State University, San Diego, CA, United States, <sup>2</sup> Marine Biology Research Division, Scripps Institution of Oceanography, University of California, San Diego, La Jolla, CA, United States, <sup>3</sup> Science Division, National Audubon Society, San Francisco, CA, United States*

#### Edited by:

*Lisa Marie Komoroske, National Oceanic and Atmospheric Administration (NOAA), United States*

#### Reviewed by:

*Graham Pierce, Instituto de Investigaciones Marinas (IIM), Spain Daniel Paul Costa, University of California, Santa Cruz, United States Beatriz Dos Santos Dias, University of Massachusetts Amherst, United States*

#### \*Correspondence:

*Rebecca L. Lewison rlewison@sdsu.edu*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *05 January 2018* Accepted: *28 May 2018* Published: *26 June 2018*

#### Citation:

*Lewison RL, Johnson AF and Verutes GM (2018) Embracing Complexity and Complexity-Awareness in Marine Megafauna Conservation and Research. Front. Mar. Sci. 5:207. doi: 10.3389/fmars.2018.00207* Conservation of marine megafauna is nested within an intricate tapestry of multiple ocean resource uses which are, in turn, embedded in a dynamic and complex ecological ocean system that varies and shifts across a wide range of spatial and temporal scales. Marine megafauna conservation is often further complicated by contemporaneous, and sometimes competing, social, economic, and ecological factors and related management objectives. Advances in emerging technologies and applications, such as remotely-sensed oceanographic data, animal-based telemetry, novel computational analyses, innovations in structured decision making, and stakeholder engagement and policy are supporting complex systems and complexity-aware approaches to megafauna conservation and research. Here we discuss several applications that focus on megafauna fisheries bycatch and exemplify how complex systems and complexity-aware approaches that inherently acknowledge and account for the complexity of ocean systems can advance megafauna conservation and research. Emerging technologies, applications and approaches that embrace, rather than ignore, complexity can drive innovation and success in megafauna conservation and research.

Keywords: complexity, megafauna, bycatch, fisheries management, complexity-awareness

#### INTRODUCTION

"Stop trying to change reality by attempting to eliminate complexity"


Marine megafauna, which we define as large-bodied, ocean dwellers like sea turtles, seabirds, marine mammals, and sharks, have experienced dramatic declines in many ocean regions (Davidson et al., 2012; Paleczny et al., 2015). The conservation of marine megafauna populations worldwide is challenged by a suite of pressures, many stemming directly from human activity, including incidental capture in fisheries or bycatch (Lewison et al., 2014), shipping strikes (Kraus et al., 2005; Panigada et al., 2006), direct harvest (Clapham, 2015; Fisher, 2016; Hofman, 2016), and contaminant exposure and accumulation (Law, 2014). In response to these pressures, research, policy, and education or conservation awareness campaigns regarding marine megafauna conservation issues have increased considerably (Authier et al., 2017). However, while megafauna research, conservation policies, and education efforts have produced important results (Taylor et al., 2000; Boyd et al., 2016; Morin et al., 2016), we suggest that many efforts have yet to adequately embrace the complexity of the systems in which marine megafauna reside. Given the mounting pressure on ocean resources and the growing concerns regarding marine megafauna conservation (National Academy of Sciences, 2017), failing to acknowledge and account for this inherent complexity may hinder much needed advances and success in marine megafauna research and conservation.

### WHAT ARE COMPLEX SYSTEMS ANALYSES?

Complex systems approaches are not a single type of analyses, rather they are a diverse suite of conceptual, analytical, and computational methodologies that can be applied to "wicked" or unstructured problems (Jentoft and Chuenpagdee, 2009; Balint et al., 2011). To understand how complex systems approaches can support megafauna conservation and research, we must first define complex systems and complex system analyses. A complex system is one with a high number and diversity of interacting components or elements (Levin, 1999; Green et al., 2005). Complexity in natural systems arises when the system is influenced by multiple processes operating at disparate spatial and temporal scales—as is the case for ocean systems and many of the processes within them. Originating, in part, from general systems theory (Bertalanffy, 1968; Warren et al., 1998), complex system analysis focuses on capturing the linked and often reciprocal nature of a system's heterogeneous elements (Arthur, 1999; Manson, 2001; Strogatz, 2001; Levin et al., 2012). A complex systems approach contrasts to a reductionist scientific approach which assumes complex, dynamic, emergent phenomena can be described in terms of their individual, constituent parts and their interactions. Complexity can be measured in many forms, including non-linearity, multi-element feedback loops, path-dependence, self-organization, difficulty of prediction, and emergence of qualities not analytically tractable from system components and their attributes alone (Manson, 2001; Bankes, 2002; National Research Council, 2012). Methods and techniques of complex systems science include, but are not limited to, nonlinear dynamic analysis, cellular automata, agentbased modeling, information and network theory, and machine learning (Shalizi, 2006).

Technological and computational advances have increased the tractability of complex system approaches (Levin et al., 1997; Green et al., 2005) and in some ocean research domains, like fisheries science and fisheries policy, complex systems analyses have been adopted to some degree (Wilson et al., 1994; Knowlton, 2004; Anderson et al., 2008; Mahon et al., 2008; Glaser et al., 2013). Although less commonly applied to marine megafauna, studies that have embraced true complex system approaches highlight the utility of embracing complexity. For example, Agent Based Models (ABMs) that simulate the actions and interactions of autonomous agents have been used to successfully evaluate individual decision strategies of boat users in whale watching operations and how these collectively impact local whale populations in the St. Lawrence river, Canada (Anwar et al., 2007) and more recently to investigate optimal strategies for the monitoring of green turtle (Chelonia mydas) populations in Hawaii (Piacenza et al., 2017). Machine learning methods have been used to help devise complex models and algorithms for prediction to classify probable behaviors and to estimate habitat areas of importance in seabird populations (Guilford et al., 2009; Fox et al., 2017) whilst artificial neural networks that identify patterns through unguided simulations have been used to evaluate breeding habitat suitability for New Zealand fur seals (Arctocephalus forsteri) (Bradshaw et al., 2002). Finally, a number of recent advances have been made in understanding marine megafauna behavior by addressing the dynamic state space of behavior and by developing big-data approaches that require no "a priori" assumptions about the behaviors of study animals (Beyer et al., 2013). Other examples include the use of Stochastic Dynamic Programming (SDP) and state-dependent behavioral theory to investigate how disturbance affects pinniped pup recruitment (McHuron et al., 2017), a dynamic state model of blue whale migratory behavior and physiology to explore the effects of perturbations on reproductive success (Balaenoptera musculus) (Pirotta et al., 2018), and a study of tagged southern elephant seals (Mirounga leonina) that identifies intrinsic drivers of movement, to describe the migratory and foraging habitats (Rodríguez et al., 2017). State space models have also been used to characterize dynamic movement of sea turtles (Jonsen et al., 2007; Bailey et al., 2008), seabirds (Dean et al., 2013), other marine mammal species (Moore and Barlow, 2011), and sharks (Block et al., 2011).

One common feature of the examples of traditional complex system analyses is data richness, i.e., traditional complex systems analyses are data intensive. For this reason, a strictly defined complex systems approach may be challenging for many data limited ocean megafauna research and conservation efforts (Pott and Wiedenfeld, 2017). However, we suggest that even when data availability may limit the application of traditional complex system analyses, adopting complexityaware approaches that acknowledge and strive to account for system complexity and adopt the fundamental precepts of complexity will be instrumental in advancing megafauna conservation and research. The term complexity-aware has been used in computing and computer science since the early 1990's (Mukherjee and van der Schaar, 2005). Outside of computer science, the concept of complexity-awareness has more recently been adopted in the context of social change, participatory research and project management by the nongovernmental organization community (Paludan, 2015; US AID, 2016). Complexity-awareness acknowledges the prevalence and importance of non-linear, unpredictable interrelationships, nonlinear causality and emergent properties, in essence, the tenets of a complex systems approach.

While the term complexity-awareness has not yet been widely adopted in marine megafauna conservation or the natural resource community, calls to increase and maintain complexity in ecological research and conservation efforts are growing (Crowder and Norse, 2008; Anand et al., 2010; Stirling, 2010; Parrott and Meyer, 2012; Howarth et al., 2013; Evans et al., 2017; Johnson and Lidstrom, 2018). Much of

this growing body of literature articulates how complexity and complexity-aware frameworks and analyses can be adopted and applied in conservation science, affirming the need to incorporate complexity into the conservation science landscape, particularly in response to the growing threats and stressors on coupled ecological-human systems. **Figure 1** captures this concept and illustrates the complexity landscape as a function of data availability (x axis), conceptual complexity (y axis), and analytical complexity (z axis). While traditional complex systems analyses will be data intensive and typically include a high level of analytical and conceptual complexity, even data-poor applications can adopt a high degree of conceptual complexity and be complexity-aware.

#### HOW CAN COMPLEX SYSTEM ANALYSES AND COMPLEXITY-AWARENESS SUPPORT MEGAFAUNA CONSERVATION IN A FISHERIES BYCATCH CONTEXT?

We illustrate complex systems analyses and complexity-aware approaches in the context of fisheries bycatch, one of the most significant anthropogenic threats to marine megafauna (Lewison et al., 2014). Fisheries bycatch, the incidental capture of unwanted, unused, or unmanaged non-target species (Davies et al., 2009), is symptomatic of one of the central challenges to ocean fisheries—how to balance ecological sustainability with economic and social viability. Megafauna bycatch is a product of susceptibility driven by the distribution, type, and magnitude of fisheries effort, and vulnerability based on ecological characteristics such as life history and species distribution traits of the bycatch species (Lewison et al., 2014). For some megafauna species, such as Pacific leatherback turtle (Dermochelys coriacea), Amsterdam Albatross (Diomedea amsterdamensis), vaquita (Phocoena sinus), Atlantic humpbacked dolphin (Sousa teuszii), and Australian and New Zealand sea lion (Neophoca cinerea and Phocarctos hookeri), fisheries bycatch has been identified as the single largest threat to extant populations (Weimerskirch et al., 1987; Lewison et al., 2004; Chilvers, 2008; Weir et al., 2011; Hamer et al., 2013; Taylor et al., 2016).

In the past decade, research and development of gear and fishing practice modifications have advanced considerably and have made important progress in reducing megafauna bycatch. For some coastal drift and gillnet fisheries, deployment of visual or acoustic deterrents has been shown to substantially reduce seabird bycatch (Melvin et al., 1999; Maree et al., 2014), while acoustic alarms (pingers) have been demonstrated to decrease bycatch for multiple marine mammal species (Dawson et al., 1998; Barlow and Cameron, 2003; Carretta and Barlow, 2011; Mangel et al., 2013; Larsen and Eigaard, 2014), and buoyless nets have been found to reduce sea turtle bycatch (Peckham et al., 2016). The use of turtle exclusion devices (TEDs) can also be highly effective in reducing sea turtle bycatch in trawl fisheries (Crowder et al., 1994; Lewison et al., 2003) as can the simple use of net lights in small scale gillnet fisheries (Ortiz et al., 2016; Virgili et al., 2017). The implementation of circle hooks, alternate baits and bird scaring devices and improved setting practices in longline fisheries has been shown across multiple studies to reduce bycatch of sea turtles and seabirds as well as sharks and other non-target fishes (Gilman et al., 2005, 2007; Watson et al., 2005; Kerstetter and Graves, 2006). A number of these gear modifications have also increased survival rates for animals that are caught and released.

Despite bycatch mitigation innovations and advances, bycatch of megafauna remains a substantial challenge to population viability largely because addressing bycatch is an inherently complex problem. This complexity stems from the need to balance the benefits of bycatch reduction against the costs of altered fishing activity to fishers' livelihoods and culture. In addition, the diversity among the regulatory, logistical, and socio-cultural constraints that create complex context dependencies influencing bycatch reduction efficacy needs careful consideration. Further, complexity associated with data collection, integration, and analysis, and the complex nature of the dynamic ocean itself, with macro-, meso-, and micro-scale temporal and spatial variability in ocean structure, processes and species distributions all must be considered (Hazen et al., 2013). Here, we describe several approaches that address megafauna fisheries bycatch by embracing complex system and complexity-aware analyses or frameworks.

#### DYNAMIC OCEAN MANAGEMENT

Dynamic ocean management (DOM) is an example of a complex systems approach that can support or supplement traditional management strategies to support sustainable fishery targets. DOM is an emerging management paradigm in which management responses change in space and time, at scales relevant for animal movement and human use. What differentiates DOM from static or traditional ocean management approaches is the use of real-time or near real-time data on the shifting physical, biological, socioeconomic, and other characteristics of the ocean and ocean resource users to generate responsive spatial management measures or strategies (Maxwell et al., 2012; Hobday and Hartog, 2014; Lewison et al., 2015). DOM holds promise for bycatch reduction, protected area design (Dunn et al., 2016) and management of populations of highly migratory and protected marine megafauna (Maxwell et al., 2015) because it integrates biological, ecological, environmental, and socioeconomic data collected over multiple spatiotemporal scales to provide information to managers and resource users in near real time (Hobday et al., 2014; Lewison et al., 2015).

While not all DOM approaches adopt traditional complex system approaches, many employ complex systems and complexity-aware ecological informatic or eco-informatics approaches (sensu Hobday et al., 2010; Scales et al., 2017; Brodie et al., 2018; Hazen et al., 2018). These cited examples use innovative digital approaches to the generation, sampling, processing, analysis, visualization, management, and dissemination of ecological, environmental, and socioeconomic data (Michener and Jones, 2012) and account for complexity at a number of levels. Central to these applications is the capacity to acknowledge complex, often non-linear and emergent, relationships between oceanographic and biological data using species distribution models (Elith and Leathwick, 2009; Žydelis et al., 2011; McGowan et al., 2013; Becker et al., 2014, 2016; Hobday et al., 2014), often using complex ocean circulation models. Rapid developments in ocean modeling have supported the integration of species distribution models with Regional Ocean Modeling Systems (ROMS), a family of models that use free-surface, hydrostatic, primitive equations over varying topography (Wang et al., 2016). Complex or complexity-aware DOM applications have been developed to reduce sea turtle bycatch in US Hawaiian fisheries (Howell et al., 2008, 2015), avoid sturgeon-fisheries interaction in the Atlantic (Breece et al., 2017), reduce bluefin tuna bycatch in Eastern Australia (Hobday et al., 2010, 2011), and limit megafauna bycatch in the US West Coast swordfish fishery (Scales et al., 2017; Hazen et al., 2018).

### SPATIALLY-EXPLICIT RISK ASSESSMENT

In coastal fishing zones, resource managers, and planners have the challenging task of balancing ecological, conservation, socioeconomic, and cultural objectives. To protect marine megafauna while supporting fisheries, managers are often asked to map, measure, and monitor the relative and cumulative risks to megafauna, with fisheries bycatch as one of the primary risks. This is particularly challenging in developing countries where managers contend with the need to support local livelihoods, paucity of available data, incongruencies across spatial and temporal scales of available information and often meager monitoring budgets. Accurately characterizing risks to megafauna, and identifying opportunities to reduce bycatch, requires an approach that accounts for the complex relationship between humans and ocean systems. New methods to investigate cumulative impacts of human activities (Halpern et al., 2008; Worm et al., 2009) and drivers of ecosystem risk in marine systems (Patrick et al., 2010; Hobday et al., 2011; Williams et al., 2011) have led to the development of risk-based scenario assessment tools, a complexity-aware approach that uses existing data and knowledge to evaluate the direct effects of human activities, climate change, and other stressors on natural resource conservation and management.

Many applications of spatially-explicit risk assessment are driven and implemented by stakeholders, managers, and policymakers needing a roadmap to understand the complexity of coastal and ocean systems, often in low capacity, data poor settings. Spatial risk assessment aims to synthesize and integrate primary data, literature reviews, expert opinion and other local knowledge in a transparent manner. Common examples of spatiotemporal risk assessment are bivariate analyses that include exposure of a habitat or species to a stressor and some metric for consequence and recovery potential, and the ability of a habitat or species to resist the stressor and recover following exposure. To account for uncertainty or missing data, these assessments can include variable weighting structure and data quality ratings, e.g., weighted averages, to acknowledge data limitations and account for uncertainty. Even in the face of data and capacity gaps, spatially-explicit risk assessments enable users to apply existing information to guide, inform, and identify appropriate survey methods, equipment and strata, establish baselines (Long et al., 2017), focus effort and resources in at-risk areas for the purpose of monitoring fisheries and marine megafauna, and evaluate alternative management scenarios that reduce risk to threatened populations (Henrichs et al., 2010).

Recent applications of tools for spatially-explicit risk assessment demonstrate the importance and utility of this complexity-aware approach in both data rich and data poor contexts (e.g., Guerry et al., 2012). Spatially-explicit risk assessments have been used to support conservation of habitats in coastal Belize (Arkema et al., 2014; Verutes et al., 2017), marine and terrestrial fauna in Washington state, USA (Samhouri and Levin, 2012; Duggan et al., 2017), freshwater lenses (aquifer) in The Bahamas (Holding and Allen, 2014), and dugongs (Dugong dugon) in Sabah, Malaysia (Briscoe et al., 2014). In Belize, the InVEST habitat risk assessment model (naturalcapitalproject.org) is a spatial risk assessment tool that was applied as part of a coastal zoning process where risk to habitats was used to alter inputs to ecological production functions in mechanistic complex, process-based models of spiny lobster catch and revenue, tourism visitation and expenditures, and natural protection provided by coastal habitats during storms (Arkema et al., 2015; Guannel et al., 2016).

Inspired by InVEST, a new spatial tool has been developed to evaluate fisheries bycatch risk and support marine megafauna conservation in developing countries, called ByRA (mmbycatchtoolbox.org). ByRA couples available information about the locations of megafauna with fishing effort categorized by gear type. ByRA outputs are spatially and temporally explicit, utilize existing data sources (e.g., Ponnampalam et al., 2014; Peter et al., 2016), community perspectives, and the human dimension of marine megafauna conservation (Hines et al., 2005; Teh et al., 2015). To date, the tool has been applied for the endangered Irrawaddy dolphins (Orcaella brevirostris) and dugongs (IUCN, 2017a,b) in Malaysia, Vietnam, and Thailand as the implementation of a new trade policy looms for these and other nations that currently export wild-caught seafood to the United States (Williams et al., 2016; Johnson et al., 2017).

### BEYOND BYCATCH ECOLOGY: INTEGRATING ECONOMICS IN SUPPORT OF COMPLEXITY

As the spatially-explicit risk assessment and structured decisionmaking tools illustrate, the complexity of fisheries bycatch extends far beyond biological or ecological factors (Lotze et al., 2017). Bycatch of marine megafauna is also defined by their social, economic, and political contexts (Lewison et al., 2011; Bisack and Magnusson, 2016). Integrating the economic factors and dimensions of bycatch into ecology-focused studies illustrates one key example of supporting complexity-awareness, moving bycatch from a one-dimensional (ecological) to a two-dimensional domain (ecological-economic). While there has been some integration of ecological bycatch research and socioeconomic relevant factors, e.g., calculations of potential biological removal (PBR) are an obvious example of this (Lobo et al., 2010; Jin, 2012; Little et al., 2014; Abbott et al., 2015), socioeconomic considerations of bycatch are often overlooked in the marine megafauna conservation literature (Lent, 2015; Barnes et al., 2016; Alava et al., 2017; Lent and Squires, 2017; Lotze et al., 2017; van Beest et al., 2017).

Economic approaches to megafauna bycatch reductions are limited by data gaps, limited understanding of their effectiveness (Lent and Squires, 2017) and the difficulty of integrating economic valuation functions with ecological production functions (Tschirhart, 2011). One effort to integrate economics and endangered Stellar sea lion (Eumetopias jubatus) bycatch demonstrates the utility of integrated, complexity-aware approaches to bycatch (Finnoff and Tschirhart, 2008). Using economic and ecological dynamic general equilibrium models and applying economic principles such as rational behavior, efficiency, and equilibrium to ecosystem processes, integrated models were used to assess the impact of alternative quotas in a local pollock (Gadus chalcogrammus) fishery on eight bycatch species including otters (Enhydra lutris), killer whales (Orca orcinus), and Stellar sea lions. Related analyses also consider the effects of the pollock fishery on the non-consumptive use of these marine mammals (Finnoff and Tschirhart, 2003b). By embracing the linked economic-ecological complexity in fisheries bycatch, these integrated approaches demonstrate the ability of complexity-aware bycatch analyses to capture the key interactions and trade-offs between target catch and at-risk megafauna populations (Finnoff and Tschirhart, 2003a), and serve as a framework for how to also incorporate linked social or political factors that can strongly influence megafauna bycatch.

#### COMPLEXITY FOR STAKEHOLDER ENGAGEMENT, CONSUMER AWARENESS, AND POLICY

Embracing complexity is equally important outside the scientific community as a part of stakeholder engagement, education and awareness. Because stakeholder groups are the backbone of marine megafauna conservation and support for research (Fulton et al., 2015), stakeholder awareness of the interdependencies and inherent complexities of the megafauna conservation landscape is an essential ingredient to effective conservation (Prell et al., 2009). Although not specific to marine megafauna conservation, Q-methods, participatory mapping and collaborative learning methods are examples of approaches that have been used to help "unpack" complexity surrounding natural resource management and use (Davies et al., 2016; Hagan and Williams, 2016; Niedziałkowski et al., 2018). These methods can clarify and map viewpoints and perspectives of different stakeholder groups without bias, helping to articulate the potentially competing interests of the fishing industry and the marine megafauna conservation community (Prell et al., 2009; Forrester et al., 2015).

Complexity and complexity-awareness is also essential to consumer education and engagement particularly in the context of fisheries, and the co-creation of knowledge among stakeholder groups has been used to support complexity-aware approaches to education and outreach (Steyaert and Jiggins, 2007). Ecolabeling and certification standards, like SeaFood Watch (http:// www.seafoodwatch.org/), RASS (http://www.seafish.org/rass/), or MSC (http://www.msc.org/) can help seafood consumers deal with complex market and supply chains and support effective megafauna conservation as well as reinforce corporate social responsibility commitments (Gutierrez and Thornton, 2014; Caveen et al., 2017; Lent and Squires, 2017). These complexity-aware education approaches allow consumers to move beyond a "not in my backyard" perspective to support meaningful megafauna conservation and sustainable fisheries across intricate, connected global market chains.

Understanding how complexity affects policies that govern bycatch will also be an important aspect to embracing complexity in marine megafauna conservation (Hirsch et al., 2010; Whitty, 2015). Policy approaches that explore policy pathways, such as the Pragmatic Enlightened Model (Edenhofer and Kowarsch, 2015), can help policy-makers understand what varying levels of complexity mean across the tradeoffs and practical consequences of each pathway. A better understanding and integration of complexity into a policy context will also help to strengthen policy development and implementation (Paul Cairney, 2017). Complexity-aware policies are ones that account for the diversity of resource stakeholders, and the reality of multiple resource uses across the seascape (Sayer et al., 2013). The conservation of marine megafauna, in particular, requires policy and governance structures that acknowledge the migratory nature of many megafauna species of conservation concern. Protection of important breeding or feeding grounds within one jurisdiction may prove to be necessary but not sufficient to conservation efforts if sensitive life stages move among unprotected waters, whether intra- or international. Policies that inherently recognize the complex relationships among ecological, social and economic systems and the influence these relationships can have on policy outcomes across jurisdictions can also ensure that wellintentioned megafauna conservation policies do not displace, and in some cases magnify, marine megafauna bycatch, or different risks in other jurisdictions (Lenzen et al., 2012; Lim et al., 2017).

### COMPLEXITY AND UNCERTAINTY: A TRADEOFF

Approaches that embrace complexity often improve the accuracy of how systems are represented and understood. Greater complexity is, however, also commonly associated with increased uncertainty that is borne from the addition of parameters which each have their own uncertainties associated with them (Fulton et al., 2003; Low-Décarie et al., 2014; Winkler, 2016). Conservation practitioners must therefore weigh the benefits of using simplified frameworks alongside more complex approaches. For all conservation applications and frameworks, there will be a fundamental and common challenge of how to balance system complexity while minimizing uncertainty (Collie et al., 2014). In fisheries science, attempts to strike this balance have led to the development of Models of Intermediate Complexity for Ecosystem assessments, or MICE (Plagányi et al., 2014). The MICE approach selects model complexity based on a specific problem statement and the data available, with temporal scales to match the questions being addressed (Essington and Plaganyi, 2013). The MICE approach for strategic complexity integration in fisheries research serves as a useful model for how to increase complexity while limiting sources of uncertainty in megafauna conservation and research efforts. Other fisheries-focused initiatives that attempt to recognize the inherently complex nature of ocean management include the concepts of Ecosystem Based Fisheries Management (EBFM) and Integrated Ecosystem Assessments (IEA), which aim to sustain healthy marine ecosystems and the fisheries they support by accounting for ecosystem complexity and the holistic impact of management decisions on those systems (Pickitch et al., 2004; Levin et al., 2009). Compared to more traditional, single speciesbased approaches to fisheries management, both EBFM and IEA represent how fisheries science has moved toward complexity and complexity-awareness (Marshak et al., 2017). However, the well-described shortcomings with the implementation of EBFM and IEA (Longhurst, 2006; Shelton, 2007; Borgström et al., 2015; Dolan et al., 2016) highlight the challenges that complex approaches in megafauna research and conservation will likely face in operationalizing and implementing complexity in a measurable and meaningful way.

## CONCLUSION

Given the complex ecological, environmental, socioeconomic, and cultural dimensions that govern ocean systems, and thus megafauna conservation, the need for complex systems analyses or complexity-aware approaches will likely not come as a surprise to most readers of this special issue. While research and conservation methods that have approached megafauna conservation from a single element or domain perspective have yielded important insights and accomplishments, as the conservation status of many marine megafauna worsens (Davidson et al., 2012; Paleczny et al., 2015), there is a pressing need to embrace the complexity that governs marine megafauna and the systems in which they reside. Even in the face of limited data and uncertainties, adopting complex systems and complexity-aware approaches to resource management, marine spatial planning and resource and marine policy and education provides a more realistic lens for conservation and research efforts.

The need to view the conservation landscape as a complex system and calls for a "conceptual revolution" in how we approach marine megafauna conservation and research echo similar calls to infuse complexity into other conservationrelated domains (Blaustein and Kiesecker, 2002; Parrott and Meyer, 2012; Lash-Marshall, 2013). By accepting that innovation and effective action in megafauna research and conservation means embracing complexity, and avoiding oversimplification sensu Stirling (2010), the marine megafauna research and conservation community is poised to focus on identifying levels of complexity that are needed to characterize and understand patterns of interest in time and space, and drive real world change. By embracing the inherent complexity of marine systems, conservation scientists and practitioners will be better equipped to provide actionable analyses, data, and information which can be used to protect and conserve megafauna.

## AUTHOR CONTRIBUTIONS

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

## ACKNOWLEDGMENTS

RL was supported in part by NASA Earth Science Division/Applied Sciences Program's ROSES-2012 Ecological Forecasting grant (NNH12ZDA001N-COF). AJ was supported by National Science Foundation grant DEB-1632648 (2017/2018). Focus for this paper was conceived at the California Leatherback Day Celebration, Southwest Fisheries Science Center, October 2016.

### REFERENCES


and drivers of the movements of southern elephant seals. Sci. Rep. 7:112. doi: 10.1038/s41598-017-00165-0


**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 Lewison, Johnson and Verutes. 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 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.

# Underwater Acoustic Ecology Metrics in an Alaska Marine Protected Area Reveal Marine Mammal Communication Masking and Management Alternatives

Christine M. Gabriele<sup>1</sup> \*, Dimitri W. Ponirakis <sup>2</sup> , Christopher W. Clark <sup>2</sup> , Jamie N. Womble<sup>1</sup> and Phoebe B. S. Vanselow<sup>1</sup>

*<sup>1</sup> Glacier Bay National Park and Preserve, Gustavus, AK, United States, <sup>2</sup> Cornell Lab of Ornithology, Bioacoustics Research Program, Ithaca, NY, United States*

#### Edited by:

*Lars Bejder, Hawai'i Institute of Marine Biology, University of Hawaii, United States*

#### Reviewed by:

*Clive Reginald McMahon, Sydney Institute of Marine Science, Australia Gail Schofield, Queen Mary University of London, United Kingdom Luke Rendell, University of St Andrews, United Kingdom*

> \*Correspondence: *Christine M. Gabriele chris\_gabriele@nps.gov*

#### Specialty section:

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

Received: *16 December 2017* Accepted: *17 July 2018* Published: *08 August 2018*

#### Citation:

*Gabriele CM, Ponirakis DW, Clark CW, Womble JN and Vanselow PBS (2018) Underwater Acoustic Ecology Metrics in an Alaska Marine Protected Area Reveal Marine Mammal Communication Masking and Management Alternatives. Front. Mar. Sci. 5:270. doi: 10.3389/fmars.2018.00270* Vessel-generated underwater noise can affect humpback whales, harbor seals, and other marine mammals by decreasing the distance over which they can communicate and detect predators and prey. Emerging analytical methods allow marine protected area managers to use biologically relevant metrics to assess vessel noise in the dominant frequency bands used by each species. Glacier Bay National Park (GBNP) in Alaska controls summer visitation with daily quotas for vessels ranging from cruise ships to yachts and skiffs. Using empirical data (weather, AIS vessel tracks, marine mammal survey data, and published behavioral parameters) we simulated the movements and acoustic environment of whales and seals on 3 days with differing amounts of vessel traffic and natural ambient noise. We modeled communication space (CS) to compare the area over which a vocalizing humpback whale or harbor seal could communicate with conspecifics in the current ambient noise environment (at 10-min intervals) relative to how far it could communicate under naturally quiet conditions, known as the reference ambient noise condition (RA). RA was approximated from the quietest 5th percentile noise statistics based on a year (2011) of continuous audio data from a hydrophone in GBNP, in the frequency bands of whale and seal sounds of interest: humpback "whup" calls (50–700 Hz, 143 dB re 1 µPa source level, SL); humpback song (224–708 Hz, 175 dB SL), and harbor seal roars (4–500 Hz, 144 dB SL). Results indicate that typical summer vessel traffic in GBNP causes substantial CS losses to singing whales (reduced by 13–28%), calling whales (18–51%), and roaring seals (32–61%), especially during daylight hours and even in the absence of cruise ships. Synchronizing the arrival and departure timing of cruise ships did not affect CS for singing whales, but restored 5–12% of lost CS for roaring seals and calling whales, respectively. Metrics and visualizations like these create a common currency to describe and explore methods to assess and mitigate anthropogenic noise. Important next steps toward facilitating effective conservation of the underwater sound environments will involve putting modeling tools in the hands of marine protected area managers for ongoing use.

Keywords: acoustic ecology, Alaska, humpback whale, communication space, harbor seal, National Park, marine protected area, agent-based modeling

## INTRODUCTION

Hearing is a primary sensory modality that marine mammals use to exchange information and detect environmental cues. Effective conservation must include protection of the acoustic environment (Barber et al., 2011) because many marine and terrestrial species are highly social and rely on acoustic communication for vital life functions such as feeding, breeding, and rearing young (Brumm and Slabbekoorn, 2005; Tyack, 2008). Motorized vessels generate underwater noises that overlap in frequency, space and time with marine mammal sounds (Richardson et al., 1995), and these noises presumably hinder effective communication. Several studies have described the increasingly noisy ocean in which these animals live (Payne and Webb, 1971; Malme et al., 1982; Andrew et al., 2002; Erbe, 2002; Hatch et al., 2008; Simard et al., 2008; Bassett et al., 2012; Miksis-Olds et al., 2013; Houghton et al., 2015; Blair et al., 2016; McKenna et al., 2017; Stafford et al., in press). Marine mammals and other taxa are known to be affected by noise and sometimes adjust their behavior to communicate in noisy environments (Terhune et al., 1979; Lengagne and Slater, 2002; Parks et al., 2007, 2009; Di Iorio and Clark, 2009; Holt et al., 2009; Dunlop et al., 2014; Ellison et al., 2016; Shannon et al., 2016; Tennessen and Parks, 2016; Fournet et al., 2018a). Across marine mammal taxa, the specific biological consequences of noise are difficult to pinpoint (Nowacek et al., 2007), but the disruption of social bonds, lost opportunities for mating and feeding, increased energy expenditure in vocalizing louder (Holt et al., 2009; Parks et al., 2010; Fournet et al., 2018a), in addition to a reduced ability to detect predators (Deecke et al., 2002), seem likely to be among the proximate causes that ultimately manifest as population level effects (Bejder et al., 2006).

Protected areas are by no means immune from the effects of noise; fortunately, methods to assess the effects of noise on wildlife in natural areas are advancing (Barber et al., 2011; Hatch et al., 2012). Historically, received sound level has been used as the primary metric for predicting both harmful and behavioral impacts from anthropogenic noise (Southall et al., 2007). However, a growing number of observations indicate that received level does not adequately predict response and that social and other aspects of context are critical factors (Buck and Tyack, 2000; Deecke et al., 2002; Southall et al., 2007; Ellison et al., 2012). In this study, we adapt emerging techniques to quantify lost opportunities for humpback whales (Megaptera novaeangliae) and harbor seals (Phoca vitulina richardii) to communicate in varying noise conditions in a large marine protected area in southeastern Alaska.

Communication sounds must propagate through the acoustic environment from sender to receiver in order for acoustic communication to occur (Wiley and Richards, 1978). The distance over which such communication can occur was first referred to as "active space" in studies of terrestrial species (Marten and Marler, 1977; Brenowitz, 1982; Lohr et al., 2003). The acoustic environment is composed of the aggregate of all sounds at a particular time and place, including both natural and manmade sound sources that might influence the ability of animals to communicate. In the ocean, wind is the dominant natural source of ocean noise (Wenz, 1962), but other natural biotic (animals), natural abiotic (e.g., earthquakes, ice, rain, lightning) and manmade abiotic (e.g., aircraft, boats, energy exploration, construction, sonar) sources (Richardson et al., 1995) are also typical contributors to the marine acoustic environment. The specific frequency bands utilized by a particular species (both passively and actively) for basic life functions can be thought of as the acoustic habitat of that species (Clark et al., 2009; Ellison et al., 2012; Moore et al., 2012; Merchant et al., 2015).

Acoustic masking occurs when sound from one or more sources interferes with a listener's ability to detect, recognize, and/or understand sounds of interest (Marten and Marler, 1977; Richardson et al., 1995). Communication masking occurs when noise reduces a receiving animal's ability to hear sounds from conspecifics, or reduces the likelihood that a vocalizing animal's sounds will be received by conspecifics. For example, Jensen et al. (2009) estimated reductions in the distances over which bottlenose dolphins (Tursiops sp.) and short-finned pilot whales (Globicephala macrorhynchus) could possibly communicate in vessel noise. Clark et al. (2009, 2016) developed a means to quantify the reduction in the area over which a baleen whale's communication sounds are detectable by conspecifics in a present noise condition relative to the area assumed to be available under the historically quiet natural conditions. Expanding on the concept of active space, the Clark et al. (2009, 2016) methods take into account in situ empirical measurements of ambient noise, the source levels of communication sounds and noise sources, and the acoustic habitat of the species of interest, and can be done from the perspective of the vocalizing animal (sender) and/or their intended listeners (receivers). This approach to estimating communication masking has been applied to a variety of cetaceans including fin, humpback, killer, minke, and right whales (Hatch et al., 2012; Williams et al., 2013; Cholewiak et al., 2018).

Arriving at a meaningful estimate of communication masking relative to natural quiet (devoid of manmade noise) relies on choosing an appropriate reference ambient noise condition. Clark et al. (2009, 2016) recommended using the lowest 5th percentile noise level statistic under naturally quiet conditions as a reasonable estimate of this reference condition. This recommendation was based on the hypothesis that for frequencies below 1 kHz, baleen whale auditory thresholds are noise-limited, and there would be a selective advantage for hearing sensitivity to be close to the quietest ambient noise condition (Clark and Ellison, 2004). For any acoustically active marine mammal, the reference condition is intended to capture the historical acoustic environment available without the influence of manmade noise. In estimating mysticete communication masking in Massachusetts Bay, where continuous ship traffic noise makes it impossible to use contemporary noise level statistics to approximate historically quiet conditions, Hatch et al. (2012) and Cholewiak et al. (2018) used a reference noise condition of 10 dB below the empirically measured present day median noise conditions as an approximation of the historical noise condition.

Refinements and new uses of emerging communication masking metrics in marine protected areas create opportunities to understand and mitigate the effects of noise. These studies and others have advanced knowledge on what geographic areas are of highest concern for anthropogenic noise, which species' communication sounds are most affected by vessel-generated noise, and what vessel attributes or operational practices affect noise production (Hatch et al., 2012, 2016; Williams et al., 2013, 2015; Frankel and Gabriele, 2017; Cholewiak et al., 2018).

Humpback whales and harbor seals are two acoustically active marine mammals that are of management concern in Glacier Bay National Park (GBNP or Park), in southeastern Alaska (Womble et al., 2010; Gabriele et al., 2017). The Park functions as one of the largest marine mammal protected areas in the world with regulations to minimize threats to these species and sustain a healthy ecosystem. For example, Park regulations require all vessels to reduce speed in areas of high concentrations of humpback whales and harbor seals (Code of Federal Regulations Title 36, Part 13, Subpart N) and exclude vessels from important seal habitat during pupping and molting periods (Code of Federal Regulations Title 36, Part 13, Subpart N). Both species produce communication sounds in the context of feeding and reproduction during their time in Park waters.

The communication sounds of these two species of interest include calls and mating-related displays produced in a variety of social contexts. We examined three types of vocalizations: (1) the humpback whale "whup" call, heard in Alaska in all seasons, believed to function as a contact call (Wild and Gabriele, 2014; Fournet et al., 2015); (2) humpback whale song, a male display common on winter breeding grounds (Payne and McVay, 1971; Darling et al., 2006) and in southeastern Alaska in the fall (McSweeney et al., 1989; Gabriele and Frankel, 2002); (3) the harbor seal roar, a male breeding display (Van Parijs et al., 2003; Hayes et al., 2004) heard primarily in May through July in Glacier Bay (Matthews et al., 2017a).

Humpback whales are a globally distributed, migratory baleen whale that was profoundly depleted by twentieth-century commercial whaling (Rice, 1978; Ivaschchenko et al., 2013). In 1973, the humpback whale was declared "endangered" under the Endangered Species Act (ESA) and thus became a prime management concern in GBNP (Code of Federal Regulations, Title 36, Part 13, Subpart N). The Hawaii Distinct Population Segment (DPS) is one of nine DPSs worldwide recently removed from the U.S. Endangered Species List, while the Mexico DPS remains listed as "threatened" (NOAA, 2016). Whales from both the Hawaii DPS and the Mexico DPS feed in GBNP waters. The harbor seal is a widely distributed pinniped that occupies various habitats in the Northern Hemisphere (Bigg, 1981). Over the last few decades, harbor seal numbers steeply declined in GBNP (Mathews and Pendleton, 2006; Womble et al., 2010), leading to management concern and regulatory action to minimize disturbance to hauled out seals (Code of Federal Regulations, Title 36, Part 13, Subpart N). In this study, we incorporated empirical data from acoustic monitoring in GBNP and applied the Clark et al. (2009, 2016) approach to quantify the degree to which vessel noise can compromise the ability of vocalizing humpback whales and harbor seals to communicate with conspecifics. Our results characterize the spatial and temporal dynamics of GBNP's underwater acoustic environment over a full year, provide the first comprehensive models of vessel traffic noise in GBNP, and quantitatively estimate the degree to which vessel noise masks humpback whale and harbor seal sounds. An understanding of communication masking over meaningful temporal and geographical scales is highly informative when management decisions are needed to address the effects of anthropogenic noise on species of concern.

### METHODS

#### Study Area

The GBNP encompasses a tidewater glacier fjord with over 2,400 km<sup>2</sup> of marine waters. The Park experiences tourism-related vessel traffic mainly in May through September. The National Park Service has jurisdiction over the marine waters of the Park, and during visitor season controls private and commercial vessel traffic using a permit system. Administrative data indicate the date/time that each vessel enters and exits GBNP. Bounded by land on all sides except its mouth, Glacier Bay is acoustically removed from distant shipping noise. Freight-carrying vessels crossing the Gulf of Alaska bypass Glacier Bay because it is not a thoroughfare and contains no major commercial port.

### Conceptual Approach

Communication masking metrics were calculated using an agentbased model comprised of multiple sound sources (including vocalizing whales and seals, wind-generated ambient noise, and vessel noise) and an underlying grid of potential receivers (see Clark et al., 2009, 2016). We modeled communication masking on 3 days with varying amounts of vessel traffic conditions and wind-generated ambient noise. The model used custom-built Sound Ecology, Detection, and Noise Analysis software (SEDNA, Dugan et al., 2011) to simulate vocalizing marine mammals and noise-producing vessels, known as "animats," that move through three-dimensional space and time according to behavioral rules set in the model (Frankel et al., 2002).

Glacier Bay was partitioned into a grid of 1 km<sup>2</sup> cells with a modeled whale or seal listener (i.e., an animat receiver) in the center of each grid cell (**Figure 1A**). To capture noise conditions over the course of each of the 24-h days, for each grid cell we computed aggregated sound levels from wind and from vessels at 10-min time intervals. These noise levels served as estimated received levels for a modeled animal listening in the center of each grid cell. For each communication sound, we computed CS metrics under three noise conditions: present ambient noise, present ambient and aggregated vessel noise combined, and aggregated vessel noise.

To calculate communication masking indices for each communication sound and noise conditions, we compared their communication space metrics relative to communication space under a naturally quiet ambient noise condition. By this process we derived estimates of communication masking dynamics for humpback whale and harbor seal sounds under different vessel traffic conditions.

FIGURE 1 | Map of Glacier Bay showing the (A) grid of 1 km by 1 km boxes and data layers for the movements of (B) individual vessel, (C) humpback whale or (D) harbor seal animats. After adding wind-generated noise, we estimated received levels for a modeled whale or seal listening in each grid square using a 17 log R propagation model for each 10-min time slice of a 24 h day. To calculate a masking index for each 10-min slice in each grid-square we compared communication space (CS) under Present Ambient noise (PrA) relative to CS under Reference Ambient noise (RA).

#### Ambient Noise Data Collection

We used continuous digital recordings of the acoustic environment from 1 January through 30 December 2011 for this study. These data were collected via a hydrophone in Bartlett Cove near the mouth of Glacier Bay (58.43501 N, 135.92297 W), deployed at a depth of 30.2 m (**Figure 1A**). The system consisted of a calibrated ITC type 8215A broadband omnidirectional hydrophone (nominal sensitivity −178 dB re 1 V/µPa) mounted on an anchoring tripod 1 m above the seafloor. This seafloor is a remnant of a glacial moraine and is fairly flat at a depth of 40–60 m. A submerged 5-mile cable connects the hydrophone to a control unit at Park headquarters, where continuous recordings were made 24 h a day, archived as 5-min sound files (National Instruments 4451 Digital Signal Analyzer, 22.05 kHz sampling rate, 24-bits per sample). The recording system had a flat frequency response from 20 Hz to 20 kHz (±2 dB). The Least Significant Bit Level (LSB) is the lowest noise level that will trigger the first bit in the sensor, and allows determination of the quietest sound that the sensor can measure. The LSB and the Most Significant Bit (MSB, i.e., the loudest sound the sensor can measure) describe the range of dB levels that the sensor can measure, and are essential for determining the amplitudes of ambient noise measured during the study.

The LSB was calculated from the formula:

LSB dB = 20 <sup>∗</sup> log10(Maximum Input Voltage/2(bitdepth−1) – Hydrophone Sensitivity - Hydrophone Preamp Gain - NI4451 Gain. Where:

> MaximumInputVoltage = 10V HydrophoneSensitivity = −178dBre : V/µPa HydrophonePreampGain = 20dB NI4451Gain = 30dB BitDepth = 24bit

With a calculated LSB of 9.5 dB, we derived the MSB (148 dB) and dynamic range of 138.5 dB.

#### Ambient Noise Data Processing

Sound level metrics were computed using the SEDNA, referred to Raven-X (Dugan et al., 2011) 1 in a two-stage process. The AIF audio files were processed on a high performance computing platform. Traditional spectrograms (16,384 FFT, 0% overlap, Hann window) were computed with a 1 µPa calibration reference level, a frequency resolution of 1 Hz, and a temporal resolution of 10 s.

A second-stage Raven-X analyzer then generated broadband and 1/3-octave band metrics that were averaged into 10-min sound level values, referred to as Leq, 10min. The Leq, is the continuous equivalent sound level defined as the single sound pressure level (SPL) that, if constant over the analysis period (e.g., 10 min), would contain the same energy as the actual measured sound level that is fluctuating over that same period.

Sound measurements were computed for each species-specific frequency band (Matthews et al., 2017b; Fournet et al., 2018b) for the communication sound types of interest. The hourly (Leq, 10min) sound levels for species-specific frequency bands were summarized as percentile sound level distributions at hourly and monthly resolutions.

#### Whale, Seal, and Vessel Animat Distributions and Movements

Vessel animats: We used actual vessel records on the number, types, and tracks of vessels in GBNP for 3 days in summer 2011 to represent low, moderate and high vessel traffic conditions. These subjective categories centered on the number of cruise ships, but also included different numbers of medium and small vessels. Our high traffic day (High) had 32 vessels including 2 cruise ships that arrived and departed GBNP 1 hr apart, our moderate traffic day (Moderate) had 26 vessels including one cruise ship, and our low traffic day (Low) had 16 vessels and no cruise ships (**Table 1**). To represent actual vessel schedules and compare the difference in loss of communication space for marine mammals exposed to simultaneous cruise ship events vs. events separated significantly in time (see also Frankel and Gabriele, 2017), we also ran a version of the high traffic day in which the cruise ships arrived 3 h apart. In all scenarios, vessels moved according to AIS tracks gathered from PACTRACS (Marine Exchange of Alaska, unpublished data) or knowledge of the entry time, exit time and maximum speed of each vessel. GBNP uses quotas to manage vessel entries into Glacier Bay during the June—August visitor season. Whale and seal animats: Realistic distributions and movements of animats are critical to realistic estimates of the CS available to a vocalizing animal. Distributions of calling seal and whale animats were based on 2011 visual survey data (**Figures 1C, D**). Whale animats were programmed to move at 3.7 km/h (2 kts). Seal animats started at a haulout, traveled at ∼5.9 km/h (±1.4; range: 0.6–11.6 km/h) (3.2 kts) between the haulout and likely foraging areas, and moved at slower rates of travel [0.6 km/h (0.3 kts)] while foraging. This pattern, based on observations of radio-tagged seals, includes likely searching and foraging behavior characterized by repeated dives in the general same area (Womble et al., 2014). Because harbor seals tend to haul out for several hours daily, at times which may vary due to tide, time of day, or other environmental variables, we incorporated this pattern into the model (Simpkins et al., 2003; Womble et al., 2014). For 50% of the seals in model runs where haul-out behavior was incorporated, sound exposures that occurred between 12:00 and 17:00 were omitted from calculations of the seals' communication masking metrics. Whale and seal animats turned at every 10-min interval with a maximal turn radius of 120◦ . If an animat moved into a water depth less than 10 m, the animat was programmed to turn sharply to return to deeper water (Dolphin, 1987a,b; Dalla Rosa et al., 2008; Witteveen et al., 2008). Seals, whales and vessels were not programmed to approach or avoid each other; thus separation distances were determined solely by vessel course and the movement pattern of an animat.

#### Whale and Seal Communication Sounds

Based on a long-term underwater sound dataset, humpback whale, and harbor seal communication sounds are the most common and conspicuous in the Park (McKenna et al., 2017). We chose and modeled the most relevant frequency band for each of three species-specific sound types based on published reports (**Table 2**; Fournet et al., 2015; Matthews et al., 2017b) and examination of the acoustic data. The low- frequency and high-frequency values bracket the 1/3rd-octave frequency bands for each sound type. We used the best available information on SL and frequency characteristics (Au et al., 2006; Matthews et al., 2017b; Fournet, 2018). We assumed 17 log R sound propagation loss based on measurements in GBNP (Malme et al., 1982). The maximal communication range (CR) for each type of communication sound was based upon published literature (Watkins and Schevill, 1979; Tyack, 2008), and best professional judgment. For harbor seal roars (0.7 km) and humpback whale whup calls (2.3 km), our choices of CR were informed by the median of the empirical distribution of how far away individual animals were acoustically localized in a 2015-2016 study in GBNP. For harbor seal roars, the median and maximal ranges at which harbor seals were localized were 710 and 2026 m, respectively (L. Matthews, unpublished data). Humpback whale

<sup>1</sup> In the initial years of this project we used customized software code referred to as SEDNA (Sound Ecology, Detection, and Noise Analysis), but this evolved into a more comprehensive system now referred to as Raven-X containing SEDNA and DeLMA-HPC (Detection Classification for Machine learning using High Performance Computing).

#### TABLE 1 | Vessel scenarios.


*Detailed vessel specifications are available in* Supplementary Table 1*.*

TABLE 2 | Characteristics of modeled humpback whale and harbor seal communication sounds.


whups were localized at median and maximal ranges of 1,350 m and 2,260, respectively (M. Fournet, unpublished data).

#### Ambient Noise

We calculated ambient noise levels for each 10-min period during a full year (2011) of audio recordings from the Bartlett Cove hydrophone (**Figure 1**). The lowest 5th percentile of band level noise for the year was regressed against wind speed data from the National Data Buoy Center station BLTA2 in Bartlett Cove and the resulting relationship was used to predict the contribution of wind to ambient noise levels for each 10-min time period during each of the 3 modeled days. These levels are referred to as "Present Ambient" **(PrA)** noise levels. To calculate communication masking metrics, a reference ambient noise level (RA) was needed to represent the naturally quiet conditions of historical times. We assumed that Glacier Bay's nighttime ambient noise level could represent this historical reference level because manmade noise is virtually absent at night (and during the winter), and distant ship traffic is only audible when a large vessel transits near the mouth of the bay. Since the difference between median daytime (with shipping) and nighttime ambient noise levels were consistently around 2dB (**Supplementary Figure 1**), we used 2 dB as our reference ambient level (RA). Therefore, in calculating communication masking metrics (see section Communication Space and Masking Index), we subtracted 2 dB from the Present Ambient (PrA) for each 10-min time period to represent historical naturally quiet conditions.

#### Vessel Noise

We modeled aggregate noise field levels from all known vessels in Glacier Bay (**Figure 1B, Supplementary Table 1**). The resultant noise level is referred to as the Present Vessel (**PrV)** noise level. Resultant noise fields were based on vessel SLs, vessel movements, sound propagation, and biologically appropriate frequency bands, at 10-min intervals, for each sound type on each modeled day. Calibrated noise signatures for individual charter, cruise, government, private, and tour vessels were used, as available, or estimated based on published and unpublished sound signatures for similar type vessels (Kipple, 2004, 2010, 2011; Kipple and Gabriele, 2004; Bassett et al., 2012). SL measurements were not available for the day tour catamaran Baranof Wind which is particularly important because it carries passengers daily from Bartlett Cove to the West Arm glaciers and back, during tourist season. Therefore, the Baranof Wind's SL was calculated based on the analysis of opportunistic recordings from the Bartlett Cove hydrophone, paired with AIS tracks as the Baranof Wind traveled at 14 knots past the hydrophone on 3 days in 2011 with a closest point of approach from the hydrophone of 546 m. This yielded an estimated broadband RMS SL of 180 dB re 1 µPa at 1 m for Baranof Wind. This corresponds with a previous estimate by Frankel and Gabriele (2017) of 177.5 dB re 1 µPa at 1 m.

We simulated the tracks for all vessels that were actually in Glacier Bay on each of the three modeled days in 2011 (NPS, unpublished data, **Table 2**) using GPS and AIS tracks for individual charter, cruise, government, and tour vessels. If a vessel's track was not available, we created a proxy track constructed from known destination(s) and speed capabilities of that vessel, or used an AIS track from a similar vessel. Vessel specifications for each modeled day, are provided as **Supplementary Table 1**.

#### Communication Space and Masking Index

We followed the analytical process proposed by Clark et al. (2009, 2016) to calculate an index of communication masking (M) defined as the proportional area that is available to a vocalizing animal under current noise conditions, relative to the potential area that would have been available under the reference noise condition. M is expressed either as a value between 0 and 1 or as a percentage of lost communication space (CS) under present noise conditions. CS, a synonym for active space, is defined as the area within which receivers of the vocalizing animal's communication sounds have the potential to experience a signal excess (SE) of greater than zero. Signal excess (SE) for any potential receiver depends on the signal source level (SL), transmission loss (TL), signal to noise ratio (SNR), detection threshold (DT), directivity index (DI), and a signal processing gain (SG). Like Clark et al. (2009, 2016), we used a recognition differential (RD) term that combines DT, DI, and SG into a single value that encapsulates the animal's ability to detect and recognize a signal in noise. We used RD to weigh SE, where PR = 0.5 at SE = 0 dB, and PR = 1 at SE ≥ 18 dB (**Table 3**). It is important to note that the probability of a receiver recognizing the signal decreases as SE approaches zero, but in some cases a signal of interest can be recognized even with SNR < 1 (Clark et al., 2009, 2016; Cholewiak et al., 2018).

We computed CS estimates at 10-min resolution based on a grid of theoretical receivers for three conditions: (1) present ambient noise (PrA) relative to RA only (CSPrA), (2) present ambient noise (PrA) and aggregate vessel noise (PrS) relative to RA (CSPrA+PrS), and (3) aggregate vessel noise (PrS) relative to RA (CSPrS).

#### Sensitivity Analysis

Modeling, by nature, requires assumptions about parameters within the model. We conducted a sensitivity analysis to investigate how variation in CS estimates can be apportioned to model input parameters, which are subject to some uncertainty. Using the whale, seal, and vessel animats and ambient noise conditions from Day 3, we systematically varied CR, RD, and RA to assess the role of the model parameters (**Table 3**) on the results, using values based on our best professional judgment and lessons learned from previous work (Hatch et al., 2012; Cholewiak et al., 2018). For all three sound types, we chose RA = 2 dB and RD = 6 dB in the main model. For the sensitivity analysis, we varied RD (2, 5, 6, 7, or 10 dB) and RA (1, 2, 3, 5, or 8 dB). Communication range (CR) variations were specific to each sound type because they differ in source level and propagation. For humpback whale


song, we used seven levels of CR between 15 and 54.5 km. For humpback whale calls, we varied CR between 1.5 and 2.8 km. For harbor seal roars, we varied CR between 0.24 and 0.94 km.

## RESULTS

#### Ambient Noise

We converted the year-long acoustic data into diel plots for each of the three species-specific frequency bands (**Figure 2**, **Table 1**). These plots show ambient noise levels at a 10-min resolution for each day of 2011 in the frequency ranges of importance to a singing humpback whale (**Figure 2A**), calling humpback whale (**Figure 2B**), and a roaring harbor seal (**Figure 2C**). Each ambient noise profile is influenced seasonally by wind, rain and biological sounds. Although male harbor seals produce a 3 s roar in the 78–147 Hz band (Matthews et al., 2017b) about once per minute (Matthews, 2017), thus also contributing to ambient noise (Matthews et al., 2017a; McKenna et al., 2017), noise level dynamics are positively correlated with vessel activity levels during the seasonal and daily periods of highest vessel activity near the Bartlett Cove hydrophone.

Seasonally, aggregate noise from vessel traffic is most evident during June, July, and August, tapering off in May and September and much less common from October to April. The 24-h median received sound levels in June, July and August were ∼98 dB re: 1 µPa in the song frequency band, 103 dB in the whup frequency band, and 103 dB re: 1 µPa in the roar frequency band (**Figure 2**; Clark and Ponirakis, in review). For June through August, the daily pattern of traffic entering and leaving the Park resulted in noise levels that were higher during daylight hours (05:00 and 20:00 Alaska Daylight Time [ADT]) than during the night (20:00 to 05:00). This was true in each of the three communication bands: roughly 5 dB louder in the song band, 1.5 dB louder in the whup band and 3 dB louder in the roar band (**Figure 2**; **Supplementary Figure 1**; Clark and Ponirakis, in review). Vessel noise was particularly prominent around 7:30 and 15:30 during the summer months when the loudest ambient noise levels of the day occurred, coincident with vessels passing close to the Bartlett Cove hydrophone (**Figure 1**) during their entrance and departure from the Park.

#### Communication Space and Masking

Communication masking metrics CS and M varied as a function of species-specific sound type and vessel conditions, summarized as follows.

#### Humpback Whale Song

Day 1, Moderate Vessel Traffic: The median communication space for singing humpback whales under the reference ambient noise condition, CSRA, was 1,421 km<sup>2</sup> . When the present ambient and the moderate vessel traffic noise conditions were aggregated, the median communication space, CSPrA+PrV, decreased to 1,200 km<sup>2</sup> (**Figure 3A**). At times, modeled whales under this aggregate noise condition experienced communication space as small as 91 km<sup>2</sup> for parts of the day (**Supplementary Figure 2A**). Similarly, daily median masking levels varied as a function of noise condition: singers lost a total of 35% of their CS under the

FIGURE 2 | Yearlong and daily patterns of ambient background noise in the frequency bands of (A) humpback whale song (224–708 Hz), (B) humpback whale calls (50–700 Hz), and (C) harbor seal roars (40–500 Hz) as recorded from a single hydrophone located in lower Glacier Bay. Aggregate noise from seasonal vessel traffic is evident during June, July and August, while aggregate noise from daily traffic entering and leaving the Park is evident as two vertical bands around 7:30 AM and 15:30 PM Alaska local time (UST-8 or UST-9). Values are in dB (RMS re 1 µPa). Sunrise and sunset times are represented by white and black vertical lines, respectively.

aggregate, PrA + PrV, moderate traffic noise condition, including 28% which was due to PrV in the moderate vessel traffic condition (**Table 4**).

Day 2, High Vessel Traffic: Under the high vessel traffic condition the median CS for singing humpback whales, decreased to 1,390 km<sup>2</sup> under the aggregate noise condition (CSPrA+PrV) from the1,455 km<sup>2</sup> available under the reference ambient noise condition (CSRA) (**Figure 3A**). Modeled whales under the aggregate noise condition experienced communication space as small as 826 km<sup>2</sup> for parts of the day (**Supplementary Figure 2A**). Based on median daily masking indices, singing whales lost a total of 19% of potential CS under the aggregate noise condition, including 13% due to vessel noise (**Table 4, Supplementary Figure 3A**).

Day 3, Low Vessel Traffic: The median CS for singing whales under natural reference ambient conditions (CSRA) was 1,454 km<sup>2</sup> , and decreased to 1,389 km<sup>2</sup> when vessel noise was included (CSPrA+PrV) (**Figure 3A**). Minimum CS under the aggregate noise condition was 713 km<sup>2</sup> for parts of the day (**Supplementary Figure 2A**). Median daily masking indices indicate that singers lost a total of 18% of CS under the aggregate noise, 13% of which was due to vessel noise under the low traffic noise condition (**Table 4**, **Supplementary Figure 3A**).

Overall, median CS for singers in the models decreased by between 18 and 35% due to combined ambient noise and vessel noise with the highest loss due to vessel noise (28%, **Table 4**) occurring on the moderate vessel traffic day (**Figure 3A**).

#### Humpback Whale Whup

Day 1, Moderate Vessel Traffic: The median CS for calling humpback whales under the moderate vessel traffic, reference ambient noise condition (CSRA) was 1.09 km<sup>2</sup> , and decreased to 0.07 km<sup>2</sup> under the aggregate noise condition (CSPrA+PrV) (**Figure 3B**). For long periods of time, modeled whales experienced a total loss of CS (0.00 km<sup>2</sup> ) due to combined natural and vessel generated noise (**Supplementary Figure 2B**). Median daily masking indices indicate that calling whales lost a total of 92% of CS under the aggregate, moderate traffic noise; 51% of which was due to vessels and 41% of which was due to natural ambient noise (**Table 4**, **Supplementary Figure 3B**).

Day 2, High Vessel Traffic: On the high traffic day, (which was also a windy day), the median CS for calling humpback whale under reference ambient noise condition (CSRA) was 0.35 km<sup>2</sup> , and decreased to 0.12 km<sup>2</sup> when vessel traffic was included (CSPrA+PrV) (**Figure 3B**). All of the modeled whales experienced almost a total loss of CS (0.00–0.01 km<sup>2</sup> ) in the morning and afternoon (**Supplementary Figure 2B**). Based on median daily masking indices, calling whales lost a total of 59% of CS under the aggregate, high traffic noise condition including 41% due to natural ambient noise and 18% due to vessels (**Table 4**, **Supplementary Figure 3B**).

Day 3, Low Vessel Traffic: Calling humpback whales under the low vessel traffic, reference ambient noise condition (CSRA) had median CS of 4.68 km2, which declined to 2.79 km<sup>2</sup> when vessel traffic was included (CSPrA+PrV) (**Figure 3B**). Total loss of CS (0.00–0.01 km<sup>2</sup> ) occurred for prolonged time periods for all

modeled whales (**Supplementary Figure 2B**). Calling whales lost a total of 85% under the aggregate, low traffic noise condition, including 40% due to the reference noise condition, an additional 45% due to vessel noise (**Table 4**, **Supplementary Figure 3B**).

Overall, median CS for calling humpback whales in the models decreased by between 59 and 92% of CS due to combined ambient noise and vessel noise with the highest losses due to vessel noise occurring on the moderate and low vessel traffic days (51, 45%, respectively, **Table 4**, **Figure 3B**).

#### Harbor Seal Roar

Day 1, Moderate Vessel Traffic: In some model runs for harbor seal underwater communication sounds, we incorporated haulout behavior that affected their CS as compared with seals that were in the water all day. Median CS for roaring harbor seals that hauled out of the water daily from 12:00 to 17:00 under the reference ambient noise condition (CSRA) was 1.09 km<sup>2</sup> , and decreased to 0.15 km<sup>2</sup> under the aggregate noise condition (CSPrA+PrV) (**Figure 3C**). The median CS for roaring harbor seals that were in the water all day under the low traffic aggregate noise condition was 0.11 km<sup>2</sup> (**Figure 3D**). In both cases, there were times of day when all modeled seals under the aggregate noise condition experienced a near total loss of CS (0.00–0.01 km<sup>2</sup> ) although this affected seals that were in the water all day to a greater degree (**Supplementary Figure 2C**). Based on the median daily masking index, roaring seals who hauled out at mid-day lost slightly less total CS (83%) than seals that stayed in the water all day (87%) with 59 vs. 61%, respectively, due to vessel noise under the aggregate moderate traffic noise condition (**Table 4**, **Supplementary Figure 3C**).

Day 2, High Vessel Traffic: The median communication space for roaring harbor seals that hauled out of the water under CSRA was 0.39 km<sup>2</sup> , which decreased to 0.12 km<sup>2</sup> CSPrA+PrV when vessel traffic noise was included (**Figure 3C**). For seals that were in the water all day, median CS was 0.11 km<sup>2</sup> (**Figure 3D**). In both cases, in the morning and afternoon there were times when 20 of 36 modeled seals under the aggregate noise condition experienced a total loss of CS (0.00 km<sup>2</sup> ) (**Supplementary Figure 2C**). Roaring seals that hauled out of the water from 12:00 to 17:00 lost a total of 49% under the aggregate, high traffic noise condition but only 10% was attributed to vessel traffic noise (**Table 4**, **Supplementary Figure 3C**). In contrast, seals that stayed in the water all day lost 71% of CS under the aggregate, moderate traffic noise condition, with 32% due to vessel noise (**Table 4**).

Day 3, Low Vessel Traffic: The median CS for roaring harbor seals that hauled out of the water under the low vessel traffic under the aggregate vessel noise condition (CSPrA+PrV) was 0.79 km<sup>2</sup> as compared to 0.11 km<sup>2</sup> for seals that were in the water all day (**Figures 3C,D**). In both cases, there were times of day when 6 of 36 modeled seals under the aggregate noise condition experienced a total loss of CS (0.00 km<sup>2</sup> ) (**Supplementary Figure 2C**). Seals that stayed in the water all day lost 39% of CS under the aggregate, moderate traffic noise condition whereas seals that hauled out lost only 11% of CS, based on median daily masking indices (**Table 4**, **Supplementary Figure 3C**).

Overall, for roaring harbor seals, median CS in the models decreased by between 39 and 87% of CS due to combined ambient noise and vessel noise with the highest loss due to vessel noise (61%, **Table 4**) occurring on the moderate vessel traffic day (**Figure 3B**). To varying degrees, seals that hauled out of the water lost up to 32% less CS than seals that stayed in the water all day.

#### Effects of Management Action on Available Communication Space

Cruise ships alone contributed 1–16% of the daily median masking level attributable to vessels, with highest levels under



*Masking index M represents the amount of communication space lost to an animal under a known noise condition relative to the communication space available under the noise condition assumed to occur under naturally quiet conditions and measures the influence of the noise on the vocalizing animal's acoustic habitat.*

moderate vessel traffic (**Table 4**). The single cruise ship on Day 1 had a much larger influence on CS because that day had a lower vessel traffic overall. For singing whales and roaring seals, removing the day tour catamaran decreased the daily median masking level attributable to vessels by 0–13%, and had the least impact on Day 2 with high vessel traffic and the highest natural ambient noise. The substantial remainder of lost CS attributable to vessels (10–51%) was due to noise from other vessel types (**Table 4, Supplementary Table 1**).

The moderate vessel traffic condition, with one cruise ship, showed substantially more lost CS attributable to vessels (35– 87%) than the high vessel traffic (19–71%), which had two cruise ships. On the two-ship day, scheduling cruise ships to arrive and depart within an hour of each other (rather than staggering them 3 h apart) restored the most lost CS to calling whales (12%) and roaring seals that hauled out at mid-day (12%), followed by roaring seals that stayed in the water all day (5%), with no improvement in CS for singing whales (**Table 4**).

#### Sensitivity Analysis

Using the model parameters for Day 3 as a basis, when we varied the Recognition Differential, RD, which encapsulates the animal's ability to detect and recognize a signal in noise, we found the largest mean difference from the default model (where RD was set at 6 for all simulations) at RD = 2, with estimates of masking due to natural ambient noise and vessels ranging from 8 to 12% higher than our main model (**Table 5A**). Our highest value of RD (RD = 10) gave results that were 5 to 12% lower than our main model. However, most differences were in the 1 to 3% range.

When we varied the value of Reference Ambient correction factor (RA), the resulting masking values were within 18% of the findings from the default model (where RA was set at 2 dB**, Supplementary Figure 1**), with the largest masking values when RA was increased to 8 dB (**Table 5B**). Overall, varying RA between 1 and 8 dB yielded masking estimates within 2 to 13 percentage points of the default model. RA exerted the strongest effects on Day 2 (the day with the most vessels and highest natural ambient noise levels) for whale calls and seal roars.

Communication Range (CR) was varied between 15 and 54.5 km for humpback whale song, which yielded masking values within 6% of the default model (**Table 5C**). For humpback whale whups, varying CR between 1.5 and 2.8 km yielded masking estimates that were no different from the default model. For harbor seal roars, varying CR between 0.24 and 0.94 km, yielded masking estimates within 1–5% of the default model, except that when CR was set at 0.24 km the masking estimate was 20% lower than the default model.

### DISCUSSION

In this study, we implemented models to characterize seasonal and diurnal patterns of ambient noise under different vessel traffic conditions in GBNP, one of the largest marine protected areas in the Northern Hemisphere. Our results demonstrate that both natural and manmade noise are seasonally and diurnally prominent in Glacier Bay's underwater sound environment and likely exert substantial impacts on the area over which harbor seals and humpback whales are able to communicate. The communication masking indices we calculated, using the best available empirical data, provided a highly informative first look at the potential effects of aggregate vessel noise in GBNP on communication for these two acoustically active species. We illustrated the usefulness of this information to protected area managers by examining the acoustic contributions of cruise ships relative to the day tour catamaran and aggregate vessel traffic (**Table 4**) and determining that synchronizing ship schedules could slightly improve the availability of acoustic habitat for humpback whale and harbor seal communication. Moreover, we demonstrated that while the masking indices were sensitive to varying some of the model parameters, the resulting differences were usually less than differences attributable to the disparate amounts of natural and vessel-generated noise on each modeled day.

Successful communication has direct consequences to the fitness of individual animals (Brumm and Slabbekoorn, 2005) and plays an indirect but vital role in maintaining the complex animal social networks that affect survival and reproductive success (Snijders and Naguib, 2017). The effects of vessel noise on behavior pertinent to survival and reproductive success have been documented in marine vertebrate and invertebrates (Lusseau et al., 2009; Wale et al., 2013; Voellmy et al., 2014a,b). While few studies have been able to document population level effects (Bejder et al., 2006), it is reasonable to presume that when noise interferes with communication, its effects can be far reaching and ecologically significant (Barber et al., 2009), particularly in highly social aquatic species in which acoustics is the dominant modality. In this study, we demonstrated that vessel noise decreases the area over which humpback whale songs and harbor seal roars can reach their intended listeners, it seems likely that this could reduce mating opportunities to some degree. Similarly, having demonstrated that vessel noise substantially decreases the area over which humpback whale whup calls are available to conspecifics, it is plausible that these lost communication opportunities have implications for foraging success, calf rearing, and social behavior. We know that these pervasive communication behaviors are important to vital life functions but it will be difficult to directly quantify the fitness consequences of communication masking in terms of the reduced caloric intake, survival, or reproductive success of individuals. Moreover, the efficacy of vocal frequency shifts and the energetic costs of compensating for a noisy ambient noise environment by vocalizing louder or more often, and the ramifications of postponing calls until conditions improve are not wellunderstood. Nevertheless, these results represent important first steps toward understanding communication masking in humpback whales and harbor seals in GBNP, and how different combinations of vessels could be managed so as to reduce their aggregate impacts on the Park's acoustic environment and marine mammals.

### Ambient Noise

The aggregate of vessel noise within GBNP strongly influences the underwater acoustic environment with a noticeable seasonal and diurnal pattern (**Figure 2, Supplementary Figure 1**). Vessel noise associated with summer visitation to the Park, primarily in the daylight hours, affected all the dominant frequency bands used by all three sound types, but particularly the harbor seal roar and humpback whale call. GBNP is one of the few places in the world where it is possible replicate historical underwater acoustic environment, due to its remote location and the fact that the vast majority of vessel traffic results from regulated seasonal tourism that predominantly occurs in the daytime. Year round, even in the absence of vessel traffic, Leq broadband noise demonstrated that nighttime levels were about 2 dB lower than during the day (**Supplementary Figure 1**).

#### Communication Space and Masking

Marine mammals and their communication systems evolved in an ocean where wind, rain and other natural sounds at times reduced their ability to communicate, but manmade noise was not a factor. The days we chose to model illustrated typical patterns of natural sound in GBNP, with high wind-generated ambient noise on summer afternoons, and a tendency for natural noise levels to decrease overnight (**Supplementary Figure 1**). Natural and manmade noise both exert a tangible influence on the frequency bands used for communication by GBNP marine mammals (**Figure 3**).

In our choice of reference ambient (RA), our goal was to capture the optimal communication environment available to humpback whales and harbor seals and avoid including any manmade noise. Based on the best available knowledge for GBNP, the 5th percentile ambient noise statistic was the best approximation of natural ambient noise levels devoid of manmade noise events; even the 25th percentile levels showed evidence of vessel noise events (Clark and Ponirakis, in review). Fine-scale sensitivity analysis could help determine what noise statistic between the 5 and 25th percentile would exclude manmade noise and still include some of the louder natural sound conditions in GBNP. Our choice of RA means that


TABLE 5 | Varying model parameters to compare daily median of 10-min communication masking levels resulting from vessels and natural ambient noise.

*(Continued)*

#### TABLE 5 | Continued


*M, Masking; RA, Reference Ambient Correction; RD, Recognition Differential; CR, Communication Range. RD varies in (A), RA varies in (B) and CR varies in (C). Results from default model are shaded, with bold type.*

our analysis emphasizes the extent to which wind, rain and other natural sounds result in communication masking but do not affect our ability to discern masking as a result of vessel traffic. Future researchers applying communication masking metrics to different study areas will need to carefully consider the unique attributes of the underwater sound environment in their area and select a reference ambient noise condition that best represents natural quiet. Above and beyond the effects of natural ambient sound levels, vessel traffic conditions during GBNP's summer visitor season result in substantial communication masking for whales and seals, especially for lower frequency communication sounds (**Figure 3**). In short, vessel noise affects marine mammal communication in GBNP in summer, even on the days with relatively low levels of vessel traffic.

The amount of communication masking was strongly affected by the frequency content of a communication sound. Harbor seal roars, a low frequency, relatively low source level male display had the highest median daily loss of CS (27–44%) to vessel noise even in the absence of cruise ships, especially during the daytime (**Table 4**, **Supplementary Figures 2C, 3C**). Roaring seals lost 4– 28% less CS when we incorporated mid-day haul-out behavior (**Table 4**) so this behavioral trait may reduce the influence of communication masking. However, male harbor seals roar at all hours of the day, thus the predominance of daytime vessel noise still reduces the CS available to displaying males (Matthews et al., 2017a; McKenna et al., 2017) even if some of their conspecifics are out of the water.

Humpback whale whup calls, which are low source level, short duration sounds of relatively low frequency (**Table 2**) experienced median levels of lost CS (18–51%) during all three vessel conditions, even in the absence of cruise ships (15–49%), and especially during daylight hours when most vessels were in the bay (**Table 4, Supplementary Figures 2B, 3B**). Humpback whales produce these putative contact calls at all hours of the day (Wild and Gabriele, 2014), thus reduced or lost communication opportunities likely have a tangible biological cost to the individual whale. Collaborative work in progress that suggests that humpback whales in GBNP adjust their call rate and/or loudness to accommodate vessel noise and are less likely to call in higher vessel noise conditions (Fournet et al., 2018a).

The loud and higher frequency humpback whale song (**Table 2**) experienced less masking overall and cruise ship noise appeared less important to song masking than the other types of vessels (**Table 4**). Daily median values for singing whales indicate that they lost 18–35% of their CS under aggregate vessel noise conditions on the 3 modeled days. Notably, singing humpback whales have also been observed to stop singing abruptly when vessel noise or presence disrupts their behavior (GBNP, unpublished data), representing a complete loss of communication opportunity. The impact of lost communication opportunities on male reproductive success is difficult to assess, especially given that it is unclear how often mating occurs in high-latitude habitats, although song occurs there (Gabriele and Frankel, 2002; Clark and Clapham, 2004). Although scientific understanding of song function is incomplete, it is unwise to discount the biological importance of song in high-latitudes.

### Effects of Management Action on Available Communication Space

Ambient noise levels show two strong bands of noise during the morning passage of vessels into the Park and the evening passage out of the Park (**Figure 2**). Communication masking is clearly higher in noise conditions with vessels relative to those without vessels (**Figure 3**). Independent of cruise ship numbers, daily CS was strongly influenced by the aggregate noise from other vessels. The day tour catamaran alone showed a fairly small influence on lost CS overall, which is not surprising given that it was a single event in the context of the aggregate of numerous other vessels. Vessel traffic schedule management has the potential to mitigate noise impacts: synchronizing the arrival and departure timing of ships restored up to 12% of lost CS and was especially beneficial for calling whales.

Some results did not conform to our expectation that higher numbers of vessels would generate greater communication masking, but the reasons for this discrepancy are informative. The lowest losses of CS occurred under the low vessel traffic condition (17 vessels total and no cruise ship). However, it was surprising that moderate vessel traffic (28 vessels including one 1,423 passenger cruise ship and one U.S. Coast Guard cutter) produced more CS loss than the higher traffic day (34 vessels including two cruise ships equipped to carry a total of 4,690 passengers) (**Figure 3, Supplementary Figures 2, 3**). While the Coast Guard vessels that happened to be present on our modeled days are not a frequent occurrence, their greater sound output (**Supplementary Table 1**) which is equivalent to cruise ships or mega-yachts that often visit (GBNP, unpublished data) skewed the results of the low and medium traffic days that were defined largely by the number of cruise ships. The finding illustrates that the type of vessel traffic is equally important as the number of vessels and, moreover, that the number of large ships can be less important than their noise characteristics (Kipple, 2004, 2010, 2011). Vessel behavior is also an important determinant of the resulting acoustic environment (McKenna et al., 2017). Managers seeking to mitigate underwater noise may be able to select quieter ships, and/or implement vessel speed limits (McKenna et al., 2017) as opposed to simply allowing fewer ships (and visitors) and still minimize communication masking. This is quite important given the National Park Service mission to preserve resources and allow visitor access to parks. The sheer variety of vessels that enter GBNP on a given summer day is highly variable, and has perhaps the strongest effect on the acoustic environment (**Supplementary Table 1**). Future modeling and continued efforts to obtain calibrated sound signatures for a range of vessel types will allow managers to weigh the acoustic influence of various vessels types, enabling access to various user groups with the least impact on the acoustic environment.

The nine scenarios reported here are just a handful of the numerous iterations that could examine a variety of questions relevant to management. By modeling a greater number of days and controlling for ambient noise levels and the exact vessels in each simulation, future efforts will allow us to move toward attributing the differences to a specific factor or factors. For example, reduced ship speeds result in decreased ambient noise levels (McKenna et al., 2017) and sound exposure levels to humpback whales (Frankel and Gabriele, 2017) in Glacier Bay and it stands to reason that communication masking would follow the same pattern.

#### Sensitivity Analysis

The communication masking exercise was sensitive to varying the parameters used, but overall, we found more variation due to differences in natural and vessel-generated ambient noise between days than we found by varying CR, RA, and RD. The greatest change occurred for harbor seal roar with a sudden increase in masking if CR was set at 0.24 rather than the more realistic higher values (**Table 5C**). This suggests underestimating CR can lead to underestimation of communication masking. In contrast, adjusting CR for humpback whale whup had no visible effect on estimates of communication masking (**Table 5C**).Our highest value of RD (RD = 10) gave results that were 5 to 12% lower than our main model. Varying RA yielded the largest communication masking values when RA was increased to 8 dB (**Table 5B**). Given our empirical approach to calculating RA (**Supplementary Figure 1**) we presume that 2 dB is a more appropriate value for GBNP.

### Conclusions and Next Steps

The ecological ramifications of noise interference are prominent in dialogs about biological resource conservation in natural areas (Barber et al., 2011) despite uncertainties about individual fitness consequences. While some vertebrates are known to adapt their communication sounds to compensate for noisy environments (Brumm and Slabbekoorn, 2005; Slabbekoorn and Peet, 2003; Scheifele et al., 2005; Parks et al., 2010; Snijders and Naguib, 2017), this phenomenon and its biological costs are not well-understood (Patricelli and Blickley, 2006). While these unknowns represent the next challenges in the conservation of acoustic habitats, it is known that marine mammals worldwide face environmental changes that are increasingly widespread, complex and difficult to predict. Mitigating manmade noise to maintain high quality acoustic habitats is thus even more important to maximizing marine mammal survival on decadal and evolutionary time scales.

The tools are clearly in place to conduct sophisticated simulations of marine mammal communication masking, and these methods continue to develop. Important next steps toward facilitating effective conservation of the underwater sound environments will involve putting these tools in the hands of marine protected area managers for ongoing use. Future investigators adapting these methods to other areas and/or species may find challenges in areas where the acoustic and oceanographic environments are undocumented, or in species whose communication sound characteristics are not wellunderstood. A key step that protected area managers can take to help bring communication masking modeling tools into routine use is to begin collecting representative vessel sound signatures and baseline ambient noise measurements that can later be used to inform such models.

Underwater acoustics is a complex discipline that many managers without advanced training in acoustics find difficult to understand, potentially creating a formidable deterrent to using the resulting information in management decisions. And yet, noise mitigation is one of management's most powerful tools because, unlike many other types of habitat degradation, noise pollution responds immediately to the reduction or removal of noise. Simple metrics and visualizations (for example, the animation in **Supplementary Material**) like those used here can help managers understand how the relative contributions of vessel classes and operating conditions reduce animal communication opportunities. Quantitative metrics that create a common currency for describing noise impacts help promote conversations among marine protected area managers, and facilitate the testing and sharing of methods to mitigate the negative effects of elevated noise levels. Fostering this understanding more broadly in the management community is essential to improving acoustic habitats for acoustically active species in natural areas worldwide.

#### AUTHOR CONTRIBUTIONS

Conceived and designed the model simulations: CC, CG, and DP. Collected and analyzed the data: CG, DP, CC with contributions to data collection from JW and PV. Advised on the analysis: JW. Wrote the manuscript: CG, with contributions to drafting, critical review, and editorial input from CC, DP, JW, and PV.

#### FUNDING

Major funding for this project was provided by Glacier Bay National Park and Preserve's Marine Management Funds and supplemented by the Cornell University Bioacoustics Research Program. This work was conducted through the National Park Service's Pacific Northwest Cooperative Ecosystem Studies Unit Agreement #9815060532.

#### ACKNOWLEDGMENTS

Thanks to the many vessel operators who provided vessel specifications and/or voluntarily participated in vessel sound signature measurements, including Holland America and Princess Cruise Lines, with funding from the NPS Ocean

#### REFERENCES


Alaska Science Learning Center. Thanks to Blair Kipple, Russ Dukek and Larry Arndt at the Naval Surface Warfare Center, Carderock Division and Mantech, Incorporated, for their longstanding efforts to maintain and improve the data collection and analysis systems that have so greatly advanced the understanding of Glacier Bay's underwater sound environment. Thanks to Christine Erbe and Christopher Bassett for providing vessel signature data, and Peter Dugan, Marian Popescu, Mike Pitzrick, Bobbi Estabrook for vital technical assistance. Many GBNP staff and volunteers helped with the underwater sound monitoring program over the years. This work was funded by GBNP commercial user fees. The manuscript was improved by comments from three reviewers.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00270/full#supplementary-material

Audio 1 | Glacier Bay\_Humpback\_whale\_song\_09\_29\_2009\_15sec.

Audio 2 | Glacier Bay\_Humpback whup\_volley\_04\_22\_2013\_18sec.

Audio 3 | Glacier Bay\_Harbor\_seal roars 06-22\_2001\_15sec.

Video 1 | Communication space model for humpback whale whup Day2.

ambient noise and multiple vessel types on the communication space of baleen whales in the Stellwagen Bank National Marine Sanctuary. Endanger. Species Res. 36, 59–75. doi: 10.3354/esr00875


**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 Gabriele, Ponirakis, Clark, Womble and Vanselow. 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.

# Bottlenose Dolphins and Antillean Manatees Respond to Small Multi-Rotor Unmanned Aerial Systems

#### Eric A. Ramos 1,2 \*, Brigid Maloney 3,4, Marcelo O. Magnasco<sup>4</sup> and Diana Reiss 1,4

<sup>1</sup> Department of Psychology, The Graduate Center, City University of New York, New York, NY, United States, <sup>2</sup> Fundación Internacional para la Naturaleza y la Sostenibilidad, Chetumal, Mexico, <sup>3</sup> Lab of Integrative Neuroscience, The Rockefeller University, New York, NY, United States, <sup>4</sup> Department of Psychology, Hunter College, City University of New York, New York, NY, United States

#### Edited by:

Peter H. Dutton, National Oceanic and Atmospheric Administration (NOAA), United States

#### Reviewed by:

Holly Edwards, Florida Fish and Wildlife Conservation Commission, United States Vanessa Pirotta, Macquarie University, Australia Alana Grech, ARC Centre of Excellence for Coral Reef Studies, Australia

\*Correspondence:

Eric A. Ramos eric.angel.ramos@gmail.com

#### Specialty section:

This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science

Received: 21 December 2017 Accepted: 16 August 2018 Published: 12 September 2018

#### Citation:

Ramos EA, Maloney B, Magnasco MO and Reiss D (2018) Bottlenose Dolphins and Antillean Manatees Respond to Small Multi-Rotor Unmanned Aerial Systems. Front. Mar. Sci. 5:316. doi: 10.3389/fmars.2018.00316 Unmanned aerial systems (UASs) are powerful tools for research and monitoring of wildlife. However, the effects of these systems on most marine mammals are largely unknown, preventing the establishment of guidelines that will minimize animal disturbance. In this study, we evaluated the behavioral responses of coastal bottlenose dolphins (Tursiops truncatus) and Antillean manatees (Trichechus manatus manatus) to small multi-rotor UAS flight. From 2015 to 2017, we piloted 211 flights using DJI quadcopters (Phantom II Vision +, 3 Professional and 4) to approach and follow animals over shallow-water habitats in Belize. The quadcopters were equipped with high-resolution cameras to observe dolphins during 138 of these flights, and manatees during 73 flights. Aerial video observations of animal behavior were coded and paired with flight data to determine whether animal activity and/or the UAS's flight patterns caused behavioral changes in exposed animals. Dolphins responded to UAS flight at altitudes of 11–30 m and responded primarily when they were alone or in small groups. Single dolphins and one pair responded to the UAS by orienting upward and turning toward the aircraft to observe it, before quickly returning to their pre-response activity. A higher number of manatees responded to the UAS, exhibiting strong disturbance in response to the aircraft from 6 to 104 m. Manatees changed their behavior by fleeing the area and sometimes this elicited the same response in nearby animals. If pursued post-response, manatees repeatedly responded to overhead flight by evading the aircraft's path. These findings suggest that the invasiveness of UAS varies across individuals, species, and taxa. We conclude that careful exploratory research is needed to determine the impact of multi-rotor UAS flight on diverse species, and to develop best practices aimed at reducing the disturbance to wildlife that may result from their use.

Keywords: unmanned aerial systems, drone, dolphin, manatee, marine mammal, disturbance, wildlife, monitoring

## INTRODUCTION

Unmanned aerial systems (UASs) are cost-effective and powerful remote-sensing tools used by scientists and wildlife managers to aerially monitor animals and their habitats (Anderson and Gaston, 2013; Linchant et al., 2015). UAS enable similar surveys as manned flight to detect and identify species in remote and inaccessible locations, but at a reduced risk to researchers and improved capacity to collect data for detailed analyses (see Linchant et al., 2015 for review). Aircrafts are often equipped with multiple sensor packages, which enables the simultaneous acquisition of multimodal data such as high-resolution imagery and geospatial data.

UAS have been successfully used to collect biological data on many marine megafauna species (e.g., Kiszka et al., 2016; Johnston et al., 2017). Oftentimes the collection of these data was previously only possible with the use of manned aircraft, close watercraft approaches, and/or invasive sampling methods (e.g., Koski et al., 2009, 2015; Christiansen et al., 2016a). Successful uses of UAS in marine mammal research have included aerial surveys for animal detection, abundance estimation of pinnipeds in breeding colonies, photo-identification of whales, and photogrammetric assessments of body condition and population health (e.g., Koski et al., 2009, 2015; Christiansen et al., 2016a; Adame et al., 2017; Krause et al., 2017). Fixed-winged UAS are most often used to detect and count animals during high-altitude long-distance surveys over large areas for estimates of population abundances and distributions (e.g., Hodgson et al., 2013; Adame et al., 2017). Conversely, low-altitude flights and stable hovering with multi-rotor aircrafts enable close approaches directly to animals, for example, to collect exhaled breath condensate (blow) for health assessments of large whales (e.g., Acevedo-Whitehouse et al., 2010; Apprill et al., 2017; Pirotta et al., 2018). Several studies suggest that UAS result in reduced disturbance to marine mammals when compared with traditional research methods (Acevedo-Whitehouse et al., 2010; Moreland et al., 2015; Arona et al., 2018). Therefore, it is important for researchers to develop effective strategies to safely apply UAS to monitor wildlife species to minimize the risk of negative impacts (Chabot and Bird, 2015; Vas et al., 2015; Smith et al., 2016).

A wide range of species has been documented to exhibit disturbance behaviors to UAS operations in response to UAS (e.g., seabirds, crocodiles, sea turtles, terrestrial and marine mammals; Rümmler et al., 2016; Brisson-Curadeau et al., 2017; Mulero-Pázmány et al., 2017; Bevan et al., 2018). Among marine mammals, pinnipeds exhibited rapid group dispersal following multi-rotor UAS approaches (Pomeroy et al., 2015; Sweeney et al., 2015), but were largely unaffected by fixed-wing UAS flying at high altitudes (Arona et al., 2018), where aircrafts may be relatively undetectable by most wildlife. Multi-rotor UAS operated at altitudes of 9–200 m did not elicit observable behavioral responses in studies of toothed whales (e.g., sperm whales Physeter macrocephalus: Acevedo-Whitehouse et al., 2010; killer whales Orcinus orca: Durban et al., 2015) and baleen whales (e.g., blue whales Balaena mysticetes; gray whales Eschrichtus robustus: Acevedo-Whitehouse et al., 2010; humpback whales Megaptera novaeangliae: Christiansen et al., 2016a). UAS flight had no detectable effects on blue whale respiration or diving behavior during blow collection, but one whale appeared to briefly look up at the UAS (Domínguez-Sánchez et al., 2018). Humpback whales and southern right whales (Eubalaena australis) rarely reacted when approached with multi-rotor UAS at altitudes of ∼4 m during blow collection, but they sometimes exhibited a "bucking" response or a turn of the body toward the aircraft (Kerr et al., 2016). In a different study these two species were not observed to respond to close approaches (<10 m) of a similar aircraft (Christiansen et al., 2016a). Differences in species responsivity can be related to a variety of factors. The type of aircraft in use (e.g., fixed-wing vs. multi-rotor), the flight patterns of the UAS (e.g., hovering or active-search), the proximity of the aircraft to animals, and the directness of its approach may all affect study subjects differently (e.g., Pomeroy et al., 2015; McEvoy et al., 2016; Mulero-Pázmány et al., 2017). This provides incentive to develop species-specific "best practices" for the use of UAS. This includes characterizing the short-term effects of these systems on wildlife to establish criteria for avoiding disturbance (Hodgson and Koh, 2016; Smith et al., 2016).

To date, little work has been done to determine the effects of UAS on delphinid and sirenian species or to evaluate their efficacy for conducting behavioral research on them. Smith et al. (2016) reported an observation of bottlenose dolphins (Tursiops truncatus) chasing the shadow of a multi-rotor UAS flying at ∼20 m altitude. Nowacek et al. (2001) reported similar responses in one bottlenose dolphin in which it briefly avoided (<10 s) the shadow of a helium-filled aerostat balloon as it passed overhead during behavioral follows. Hodgson (2007) reported that dugongs (Dugong dugon) fled from the shadow of a similar aerostat balloon. Studies using fixed-wing UAS flown at high altitudes (100–300 m) did not detect responses in Florida manatees (Trichechus manatus latirostris; Jones et al., 2006) or dugongs(Hodgson et al., 2013). Information on how to best apply these systems for use on dolphins and sirenians can be beneficial to species monitoring, however it is necessary to first evaluate in the field, how animals respond to its different uses.

During ongoing studies in Belize, we investigated the behavioral responses of bottlenose dolphins and Antillean manatees (T. m. manatus) to small multi-rotor UAS flights in shallow coastal habitats. Dolphins and manatees were located using several different flight strategies and were approached and followed at varying altitudes to determine whether these factors influenced animals' responses to the aircraft.

#### METHODOLOGY

#### Study Populations and Data Collection

UAS flights to track and observe bottlenose dolphins and Antillean manatees were conducted in Belize from 2015 to 2017. Data were gathered throughout the year in both the wet and dry seasons. Dolphins were found in several shallow marine ecosystems (mean water depth = 3.6 m, range = 0.5 to 20 m) in both coastal and offshore regions of Belize (**Figures 1A–C**), mostly at Turneffe Atoll Marine Reserve (**Figure 1C**). Turneffe is an offshore marine atoll 36 km east of the mainland coast where a year-round, small population of resident and non-resident dolphins inhabit shallow seagrass lagoons enclosed by mangrove cayes and a fringing coral reef system (Campbell et al., 2002; Dick and Hines, 2011; Ramos et al., 2016, 2018). Flights with manatees were primarily conducted on the leeward side of St. George's Caye (**Figure 1B**), located 9.5 km east of mainland Belize near the Belize Barrier Reef. The clear shallow waters of our study area are sheltered by islands and predominantly seagrass habitats which allowed us visibility to the seafloor (up to 5 m) and to observe animals continuously for the duration of most flights.

Three small (1,240–1,400 g) commercial quadcopter UAS were used: the DJI Phantom 2 Vision + (P2), Phantom 3 Professional (P3), and Phantom 4 (P4). All models were white and had very similar structure and appearance. Each model was equipped with a 4K camera mounted on a gimbal with 3-axis stabilization. Highdefinition video was recorded to an onboard microSD card in MP4 format. To reduce surface glare that restricts visibility into the water, a linear polarizer was placed on the camera lens of each aircraft, and the camera was angled downward at 45–90◦ toward focal animals. The aircraft was piloted by ER with a remotecontrol joystick. Monitoring of the live video feed, as well as relevant flight metrics (e.g., battery levels, distance of the aircraft from the pilot), was performed using the DJI GO application on a tablet (Samsung Galaxy 8 or iPad 9) mounted to the remote control.

The aircrafts were launched from shore or from a small boat with the help of a ground station operator who held the aircraft (P2, P3, or P4) overhead prior to launching it. The pilot remotely controlled the aircraft to distances of <2 km. Flights were between 5 and 22 min in duration and were performed in non-rainy conditions at wind speeds from 0 to 35 kt. Flight movements were categorized as: (i) direct approaches: vertical descents toward animals at speeds of 0.3–1.0 m/s; (ii) horizontal follows: altitude-stable horizontal flight in an effort to closely match animal movement, at flight speeds of 0–5 m/s; or (iii) hovering: stable hovering in place over animals.

#### Bottlenose Dolphin Flights

Dolphin responses were recorded opportunistically during aerial focal and group follows conducted from small boats and from shore (n = 3). Boat surveys to locate dolphins were conducted from several small vessels (6–15 m long) equipped with one or two outboard engines (85–250 HP). When dolphins were sighted, the vessel approached the group's position slowly (to within 30 m) to collect images for photo-identification using a SLR camera with a telephoto lens (75–400 m). Groups were defined as all dolphins <100 m of each other (Shane, 1990). One dolphin or dolphin group was approached and followed continuously for the entire flight. Group size was measured initially from the boat and verified by counting dolphins in aerial video. Dolphins were classified as adults, juveniles/sub-adults, or calves by relative size and based on previous identifications. Sex was determined by viewing the genital region or by repeated observations of an adult with a calf (Mann et al., 2000). After photo-ID was completed, the UAS was launched to an altitude of 20–50 m and navigated over animals. Once over the animals, the aircraft maintained its initial hovering altitude (50, 40, or 30 m) for 30–180 s, then descended vertically (at 5–10 m intervals) in a direct approach until reaching a stable altitude of either 20, 15, or 5 m. Up to six flights were conducted with each group. There was a maximum of 4 vertical descents per flight. To minimize the behavioral effects of the research vessel, during the majority of flights, the boat remained at a distance of 100–1,500 m away from the focal dolphin(s).

### Antillean Manatee Flights

Manatees were sighted during aerial transects using the P3 and P4 models deployed and remotely controlled from shore as a part of ongoing studies and monitoring of local animals (Ramos et al., 2017). Once sighted, animals were randomly selected for UAS exposure. Groups were defined as all manatees <10 m of each other. Multiple different individual manatees and manatee groups were sometimes sighted during a single flight. The aircraft was initially flown to an altitude of 100 m to locate manatees then descended to 60, 50, 40, or 30 m before hovering over manatees for 30–120 s. The UAS descended again and remained at 20, 15, 10, or 5 m during horizontal follows for a maximum of 5 descents per flight. There was evidence from preliminary data collected in 2016 that manatees changed their direction in response to the aircraft's flight path. Overhead flights were conducted to test the hypothesis that UAS flight causes manatees to evade the aircraft. The aircraft was flown in a straight line over the manatees and perpendicular to their swim direction.

Automated UAS imaging flights were conducted on 7 days in 2017 at St. George's Caye using the DJI GS Pro application to gather high-resolution photos that were stitched together to build georectified orthomosaic maps of manatee habitats (**Figure 1B**). The aircraft was flown to its starting point at an altitude of 150 m and flew autonomously in a saw-toothed transect pattern for distances of 3–7 km capturing one 12 MP (4.1 cm/pixel) image at preselected waypoints. Maps were compiled in OpenDroneMap (www.opendronemap.org) and WebODM (www.webodm.org).

#### Data Analysis

Information was extracted from the flight data logs to determine flight effort and examine behavioral responses. The onboard GPS logged a waypoint every 100 ms and stored associated information on altitude (m), speed (m/s), and distances traveled (m). The actual altitude of flight was adjusted according to the height of deployment at 2 m above sea level in boat-based deployments. The website www.airdata.com was used to generate flight reports. We suspected that high wind speeds might raises the chances of animals responding to the UAS, because high windspeeds increase the noise of the rotors (Christiansen et al., 2016b). To test this hypothesis, mean wind speed and maximum gust speed (m/s) per flight were examined to identify if high wind speeds increased the likelihood of animals responding to the UAS.

Flight tracks were examined in Google Earth Pro to groundtruth measurements of manatees' positioning and travel distances (m). Habitats at sighting locations were categorized as follows: seagrass bed (dense or patchy), lagoon, channel, channel edge, cove, reef, or resting holes (holes in the substrate where manatees frequently rest; Bacchus et al., 2009).

Aerial videos of dolphins and manatees were initially reviewed using QuickTime 10.4 Media Player (Apple Inc.,). Animal activity

Map Source: Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AeroGRID, IGN, and the GIS User Community.

data (including location, behavior, and position in the water column) was coded using GriffinVC Behavioral Coding Software (www.github.com/svirs/griffinVC). Response events (RE) were defined as events during which, in the apparent absence of other stimuli, one or more animals appeared to change their behavior following apparent detection of the aircraft. RE began when dolphins visibly changed their behavior in response to the aircraft, and ended when they either returned to preresponse behavior or began a different activity (**Table 1**), and sometimes included repeated UAS-orienting events. Disturbance was defined as RE in which animals exhibited potentially negative responses, such as increased vigilance, flight responses, and changes in short-term movement patterns (Gill, 2007).

Dolphins were identified according to standard photo-ID techniques (Defran et al., 1990) by matching dorsal fin photos to a catalog of known dolphins in Belize (Campbell et al., 2002; Ramos et al., 2018). We determined the minimum number of different dolphins exposed to the UAS throughout the study instead of the exact number because it was sometimes not possible to identify all dolphins when they were in large groups. To assess dolphins' initial response to the UAS, the first flight over each group (n = 48) was considered separately from subsequent flights.

Aerial videos of dolphins were analyzed to assess if bottlenose dolphins changed their behavior in response to the UAS. We identified RE when dolphins exhibited response behaviors such as orienting and turning toward the aircraft in a fashion uncharacteristic to the behavioral state they were in prior to approach. **Table 1** is an ethogram—a catalog of relevant behaviors—lists the non-response and response behaviors exhibited by dolphins. Response behaviors typically involved upward rostrum- and eye-directed movements to the aircraft and repeated changes in position and body orientation. Dolphin observations were first analyzed to identify possible responses to the aircraft using ad libitum sampling of behavior (Altmann, 1974). Behavior states pre-and post-response were identified using continuous sampling. The behavior of individual responding dolphins was coded using all-event sampling to identify each occurrence of seven UAS-oriented behaviors (**Table 1**): "side-roll," "full-roll," "belly-up," "rostrumup," "circular swim," "spin-and-orient," and "breach." The "sideroll," "circular swim" and "breach" were further characterized using modifiers that indicated whether the dolphin performed the activity with an open mouth, while swimming upside down, or while twisting along the body's longitudinal axis, as these behavioral modifications were occasionally observed.

Dolphin swim direction was scored as left or right in cases that the dolphin executed a turn as part of a response to the UAS. Bodily rotation along the longitudinal axis was scored as clockwise or counterclockwise. Visible increases or decreases in dolphins' swim speed following their initial detection were noted. Direct responses to the boat and instances of social behavior were especially likely to be misidentified as UAS-directed behavior because dolphins at this site typically interact with vessels at first approach. We took special care in noting these behaviors and excluded animals' behavior within the first 5 min of sighting a



Modifiers to behavioral events: OM, Open-mouth; IN, Inverted; TW, Twist.

dolphin group, a period when dolphins are frequently observed interacting with the boat (Ramos, personal observation).

High-resolution images of manatees were used to photo-ID individuals with identifying features (primarily scarring) across their bodies and classify age and sex where possible. Images were taken from screenshots of aerial video (taken at altitudes of 6–60 m) in which most of the manatee's body was visible and exposed or just beneath the water's surface (**Figure 2**). Multiple images taken at different angles during a single sighting and over the course of the study were examined for individually-distinctive features located anywhere on the body including: trunk, head, left and right dorsal, and ventral surfaces, and tail (e.g., Flamm et al., 2000; Langtimm et al., 2004). Manatees were classified as adults, juveniles/sub-adults, or calves based on their relative size. Females were identified by the presence of a closely-associated calf (e.g., O'Shea and Langtimm, 1995; Langtimm et al., 2004). Identified individuals were compiled into a photo-ID catalog for the study, which was used to determine the total number of manatees exposed to the UAS and re-identify animals. Manatees that could not be identified because of insufficient markings or poor visibility were excluded from analysis for response behaviors.

Manatee behavior was sampled continuously from the beginning of observations until animals were out of sight. To identify if and how manatees changed their behavior after UAS detection, their activity was coded using the states and events described in **Table 2**. RE were identified when a manatee exhibited rapid shifts in behavior that differed from its pre-exposure activity state, including flight responses. We sampled their behavior continuously for up to 20 min after the initial sighting to accurately document the duration of their responses and detect if they responded multiple times to the aircraft. Changes in behavior and/or swim direction immediately following aircraft movements were considered responses.

Each manatee's swim distance (m) was measured from the location of their initial RE to their location at 1-min post-RE. To perform this measurement, the GPS track for each was overlaid in Google Earth Pro onto the high-resolution map we generated, and the distance between these two location points was measured using the path tool. The exact locations of manatees were groundtruthed by visual review of animal position and static habitat features (e.g., sand patches, the edges of seagrass patches) in videos and the map. We used this method to identify manatee swim direction before aircraft movements, then documented any changes in their swim direction during or directly after these overhead movements relative to the aircraft's direction of flight.

#### Statistical Analyses

Statistics were conducted in GraphPad Prism version 7.00 for Windows (GraphPad Software, La Jolla California USA, www. graphpad.com) and Microsoft Excel 2016. The duration of time that individuals spent responding to the UAS, and the proportion of this response time relative to overall observation time per flight, were calculated for each response event for both species. The proportions of all flights in which dolphins or manatees responded or did not respond to the UAS were compared using a Chi-Square test at p < 0.05. The proportions of individual dolphins responding vs. not responding to the UAS were compared using a Chi-Square test at p < 0.05. Only the first UAS flight for each dolphin group was considered in this comparison). The same test was performed for manatees. The duration of dolphins and manatee responses in their first confirmed response to the UAS (to account for the observation of specific individuals responding on several occasions) were compared between species with a nonparametric Mann–Whitney U-test at p < 0.01. Mean wind speeds and maximum gust speeds were compared across non-response and response flights within species using independent-samples t-tests to determine whether higher wind speeds were associated influence animals' responsiveness to the UAS as p < 0.05.

#### RESULTS

#### Bottlenose Dolphin Responses

We included 56 sightings of dolphins in our analysis. These flights took place on 48 days in 2015 (n flights = 53), 2016

TABLE 2 | Ethogram of the different behaviors of Antillean manatees.


(n = 71), and 2017 (n = 14). Flights were on average 14.7 min in length (SD = 2.6 min), ranging from 3.0 to 17.4 min, for a total of 33.3 h of flight time. Of this total flight time, we directly observed dolphins for a total of 31.4 h. In most sightings, multiple flights were conducted with each group. The groups we observed contained a mean of 5.14 dolphins (SD = 3.8) and ranged from 1 to 17 animals. Flight altitudes across all flights ranged from 5 to 100 m, with a mean of 20.74 m.

Dolphins responded briefly to the UAS in 8 RE. Characteristics of dolphin RE are described in **Tables 3**, **4**. Responses were of short duration and largely consisted of turning and upwarddirected orienting behaviors (shown in **Figures 3A–F** and **Supplementary Video 1**). We observed possible responses to the UAS during 10 other flights; however, these behaviors occurred simultaneously with social interactions, and potential response behaviors were difficult to distinguish from social behavior directed at conspecifics. Dolphins briefly changed their preresponse behavior state then quickly returned to pre-response or a different behavior state. These sometimes involved vessel interactions directly before responses and less so after in which animals actively approached vessel. Dolphins were more likely to respond in the first flight with each dolphin group. Responses were detected in 12.5% of the first flights with each dolphin group, and 85% (n = 6) of all responses occurred during these initial flights.

We identified, at minimum, 68 different dolphins across these sightings. Responses to the UAS were detected in 9 of these dolphins, 13.2% of the animals we identified (**Table 4**). Most responding animals were adults, but two subadults and one calf also showed reactions to the UAS (**Table 3**). The responding groups ranged in size from 1 to 8 dolphins, but most responses only involved a single dolphin or pairs. Dolphins slightly decreased their swim speed during most RE (n = 5), increased in one RE, and maintained their swim speed in two RE. The most frequently observed behaviors were upward-orienting behaviors like the side-roll, the bellyup, and the rostrum-up (**Figure 4**). Across response behaviors that involved rotations (n = 34), dolphins turned left in 71% of behaviors and turned right in 29% of behaviors. Circular swims involved clockwise turns in one RE (11%) and counterclockwise turns in 8 RE (88.9%). Only one dolphin exhibited possibly


Ramos et al. Dolphins and Manatees Respond to UAS

Dolphins responded to UAS flown at a mean altitude of 19.65 m (**Table 4**). The stacked histogram that shows the number of dolphin and manatee responses in **Figure 5** illustrates our finding that most dolphin responses occurred in a narrow range of low altitudes, while manatee responses occurred at a broader range. Most dolphin responses occurred during horizontal follows by the UAS (n = 9), one occurred during vertical descent, and one during ascent. All dolphins began visibly responding when their bodies were fully underwater, and no parts exposed to the surface. The average latency of dolphins' response, from the time of initial UAS exposure to the start of dolphin RE, was 166 s (SD = 2.75 s), and varied across animals from 3 to 455 s. The majority of RE in dolphins occurred in response to the P3; however, this may be due to the fact that the P3 was flown more than the P2 or P4, rather than being due to differences between the aircraft (**Table 3**).

Mean wind speed (mean across all flights = 5.26 m/s, SD = 2.96 m/s) and maximum gust speed (mean = 6.65 m/s, SD = 3.88 m/s) during dolphin response flights were slightly lower than for non-response flights (wind speed: mean = 5.58 m/s, SD = 2.63 m/s; max gust: mean = 6.92 m/s, SD = 3.27 m/s). However, these differences were not significant at p < 0.05.

#### Antillean Manatee Responses

We included 83 sightings of manatees in our analysis. In contrast to dolphin flights, multiple distinct solitary manatees and small groups were followed in most flights. These flights took place on 16 days in 2016 (n flights = 20) and 2017 (n = 53). Flights were a mean of 17.1 min in length (SD = 1.8 min) and ranged from 14.3 to 20.3 min, for a total of 24.3 h of flight time. Of this total flight time, we directly observed manatees for a total of 22.6 h. Flight altitudes across all flights ranged from 5 to 120 m, with a mean of 43.79 m.

A total of 83 different individual and groups of manatees were exposed to the UAS. Thirty-three of these sightings involved single manatees, and 50 involved groups of ≥2 manatees. We sighted 146 manatees, with 84 adults (mean per group = 1.01, SD = 1.3, range = 1–4), 36 juveniles/sub-adults (mean = 0.43, SD = 0.31, range = 0–2), and 26 calves (mean = 0.31, SD = 0.49, range = 0–2). Manatee groups contained a mean of 1.77 manatees (SD = 1.3), ranging from 1 to 6 animals. Photo-ID analysis revealed these 146 manatees consisted of a minimum of 66 and a maximum of 71 distinct individuals accounting for repeated sightings. Some manatees could not be identified because of a lack of scarring or other definable features (n = 5) indicating some repeated flyovers were undetected. We randomly selected individual manatees for exposure to the UAS, but several manatees (n = 17) in addition to these animals were inadvertently exposed to the aircraft because the individual manatees could not be identified at the time of UAS flyover in the lower resolution tablet view. Individual manatees were exposed to the UAS a mean of 2.1 times (SD = 3.2; range = 1–17 times).

Manatee responses were detected in 24% of all exposures (n = 20), and the characteristics of these responses are described in **Table 4**. Manatees responded by quickly changing their


TABLE 4 | Characteristics of bottlenose dolphins and Antillean manatee responses to small unmanned aerial systems.

Time and altitude values shown are mean ± standard deviation, with range in the row below. RE, Response event.

behavior; often a manatee would execute a powerful tailkick (a movement which raised a large plume of silt), and swim quickly away from the area. Many animals continued to respond for several minutes (see **Figures 6A–D** and **Supplementary Video 2**). A total of 29 identified manatees responded to the UAS, in a total of 36 detected manatee responses, that included repeated responses of some animals. Twenty-three of these manatees appeared to respond directly to the UAS. In other cases, the responses of these directlyresponding manatees appeared to cause behavior change in one to four nearby manatees. More adult manatees responded to the UAS (n = 14) than did juveniles/sub-adults (n = 8) or calves (n = 7), but adults were exposed at higher rates. Calves likely responded because of their mothers. Responses between individuals and groups was highly variable; one mother/calf pair was exposed 17 times but only responded in the first UAS approach, while one juvenile/subadult manatee that was exposed 14 times responded with 11 RE.

The UAS caused strong disturbance responses in manatees. In short, the animals that responded fled the aircraft, and they directionally evaded the UAS when pursued. Every manatee that directly responded to the UAS (n = 23) changed their behavior to fleeing. Of these animals, 13 (44.8% of all responding individuals) responded to direct UAS approach upon their first exposure to the aircraft. These animals, which we presume had no previous UAS experience, all changed their behavior following detection of the aircraft from either feeding (n = 11) or milling (n = 2) to fleeing. They included adults (n = 7), juveniles/sub-adults (n = 5), and a single calf. Most of the manatees that appeared to react because of a directly-responding manatee also changed their behavior to fleeing. All manatees responded close to the onset of initial UAS exposure, with an average response latency of 33.4 s (SD = 33.2 s) which ranged from 0 to 120 s. The proportion of manatees (N = 36) that were moving at the onset of the exposure (52%), as opposed to remaining in one place, increased to 69% immediately after exposure. Responding manatees spent an average of 80% of the total flight observation time fleeing the aircraft (SD = 0.18, range = 13–97%). By 1 min after the initial response, manatees that were still responding (n measured = 19) fled a mean distance of 258 m (SD = 163.6 m, range = 2.2–582.0 m). Fleeing manatees typically swam across shallow seagrass flats; 10 manatees fled into the deeper waters of a nearby channel in 4 different RE; and 4 manatees fled into nearby resting holes in 3 different RE.

Animal-directed aircraft movements were more likely to cause responses than either stable hovering or horizontal follows. Manatees responded to direct aircraft approaches (vertical descents) at altitudes of 6–52 m (**Figure 5**). In flights with no responses (n = 63), manatees were tracked for a total of 14.72 h of horizontal follows, with no signs of UAS-induced behavioral changes. Responses to the P4 (n = 12) were more frequent than to the P3 (n = 8), but manatees were exposed primarily to the P4 (n = 63). More responses occurred during direct approaches (70%; n = 14 RE) than during stationary hovering (30%; n = 6 RE). Most responding manatees (95%) were full underwater when they first responded.

UAS flight over manatees after their initial responses consistently elicited changes in swim direction. Changes in manatee swim direction were identified in 77.8% of RE (n = 14). After their first response in 18 RE, manatees responded to 96.2% of overhead flight movements by changing their swim direction by 45–90◦ relative to the trajectory of the aircraft. Manatees typically responded <5 s of the aircraft flying overhead at an altitude range of 6–104 m and continued to respond to each overhead pass with directional evasion. Manatees directionally evaded the aircraft despite aircraft descents of 10 m and ascents of up to 98 m (**Figure 7**).

Mean wind speed (mean across all flights = 6.56 m/s, SD = 3.3 m/s) and maximum gust speed (mean = 9.5 m/s, SD = 5.01 m/s) were higher in manatee response flights compared to nonresponse flights (wind speed: mean = 5.02 m/s, SD = 3.26 m/s; max gust: mean = 7.35 m/s, SD = 4.74 m/s). However, these differences were not significant at p < 0.05.

#### Comparison of Dolphin and Manatee Responses

Manatees were more likely to respond to the UAS than dolphins, and they displayed stronger responses. The difference in frequency of response between the two species was significant, with dolphins responding during 0.05% of UAS flights, vs. manatees which responded in 26% of their flights (p < 0.01, X <sup>2</sup> = 15.7196, df = 1). A greater number of individual manatees responded to the UAS than did individual dolphins (**Table 4**). At minimum, excluding repeated exposures, 19.2% of all manatees observed responded to the UAS, and 10.3% of dolphins responded. However, this difference was not found to be statistically significant (p = 0.1669, X <sup>2</sup> = 1.9109, df = 1).

The duration of responses (both on average, and in total) was longer for manatees than for dolphins, both in individual

RE and across all events (**Table 4**). When controlled for individual identity and repeated responses, manatees responded for significantly longer durations than did dolphins within the first UAS flight per group (Mann–Whitney U-test, U = 1, p = 0.00006).

Dolphin RE occurred at lower altitudes, both on average and in comparison of altitude range, than for manatees (**Figure 5**),

but a Mann–Whitney U-test revealed these differences were not significant (p = 0.1521).

### DISCUSSION

Our findings document the behavioral responses of coastal bottlenose dolphins and Antillean manatees to small multi-rotor UAS. Aircraft activity caused different behavioral responses in dolphins and manatees that depended on both flight- and animalrelated factors. Both species reacted to UAS flights at a broader range of altitudes than previously reported for marine mammals (e.g., Pomeroy et al., 2015). Only a small subset of the dolphins that we tested responded to the aircraft, and that tended to be when they were directly approached and followed. Dolphins who

did react to the UAS changed their behavior briefly, orienting toward the aircraft before returning to their pre-response activity. Only one dolphin responded in a way that appeared potentially negative, indicating a possible disturbance. In contrast, many Antillean manatees exhibited strong disturbance behaviors in response to our aircraft. Those that reacted rapidly fled the area. Manatees responded for significantly longer durations than dolphins, in a higher proportion of flights, and with more severe disturbance responses indicating they were more sensitive to UAS and their effects. Most disturbed manatees continually evaded the pursuing UAS until the end of is flight, changing direction repeatedly as the aircraft flew over them at high altitudes. Flying the aircraft directly over disturbed manatees during the postresponse period consistently provoked the animals to change direction, indicating heightened vigilance and avoidance. These findings suggest that multi-rotor UAS, in Belize and possibly elsewhere, are a more disruptive stimulus for Antillean manatees than for bottlenose dolphins.

### Responses of Bottlenose Dolphins and Antillean Manatees

Dolphins exhibited low overall responsiveness throughout all UAS flights. Animals only reacted in a small proportion of observations. When they did appear to notice and respond to the aircraft, the duration of their responses was short and animals seemed minimally impacted. Dolphins' responses involved investigation of the aircraft (e.g., side-roll, spin-andorient). These behaviors were similar to reports of whales rolling to one side to view UAS (e.g., Kerr et al., 2016; Domínguez-Sánchez et al., 2018), as well as the "alert" and "head-up" behaviors described in seals during aircraft approach (Pomeroy et al., 2015). Dolphins exhibited open-mouth behaviors during side-rolls, which is similar to reports of a sperm whale rolling on its side with mouth agape in response to a fixed-wing manned aircraft (Smultea et al., 2008). This suggests that measuring the incidence of side-turning behaviors may be a useful diagnostic

multiple overhead movements despite ascents to 104 m. The number of responses is listed from the first (initial) to the fifth individual response.

criterion for detecting cetacean responses to UAS. The single dolphin in our study that engaged in excited responses and possible social displays, and potentially agonistic responses may have been trying to evade the aircraft, but it was unclear if this was the case. A lack of responses to the aircraft when it was flown above 30 m suggests that if a small UAS is responsibly piloted, with minimal animal-directed movements at sufficiently high altitudes, dolphins are unlikely to be significantly impacted.

The evasion we observed in responding manatees appeared similar to typical disturbance responses of marine mammals to close vessel approaches (e.g., Williams et al., 2002; Senigaglia et al., 2016). Manatees exhibited a strong sensitivity to multirotor aircraft movements, fleeing from aircraft at altitudes ranging from 6 to 52 m, and repeatedly evading the UAS at altitudes as high as 104 m. The manatee reactions reported here are similar to the bucking behaviors observed in humpback whales and southern right whales during blow collection (e.g., Kerr et al., 2016), both characterized rapid tail movements and apparent evasion following UAS detection. Manatees fled from aircrafts for long durations across seagrass flats, and occasionally into deeper water if it was available. This behavior is similar to the reactions of Florida manatees following boat disturbance (Nowacek et al., 2004), and also resembles observations of seals fleeing haul-out sites into the water following UAS disturbance (Pomeroy et al., 2015). Interestingly, manatees have no regular natural predators but were intensively hunted in Belize from pre-Columbian times (McKillop, 1985) till the late nineteenth century (Bonde and Potter, 1995), pressures that can sometimes cause manatees to shift their activity to avoid human detection (Jiménez, 2002). Manatees in our study area may respond strongly as a result of a combination of historical hunting behavior and daily exposure to boat traffic in the region leading them to evade approaching objects.

Our findings indicate that manatees can be negatively affected by UAS in various ways, including loss of feeding and resting opportunities and possible area avoidance (including avoidance of critical habitats). For example, responding manatees sometimes fled into nearby deep-water channels, where they were at increased risk of encountering boat traffic. These responses might have especially negative results for vulnerable animals; for example, if manatee flight responses cause the separation of mother and calf, there is an increased risk of calf orphaning or calf death (Parente et al., 2004). Repeated evasion by manatees, and persistent and repeated responses by multiple dolphins, suggests that the increases in animal vigilance following disturbance by a UAS can result in shortterm changes in natural behavior patterns. Birds and marine mammals on land have shown similar responses to UAS (e.g., Chabot and Bird, 2015; Pomeroy et al., 2015). Improper use of UAS targeting marine megafauna could be especially harmful to animals with restricted home ranges, as they may be repeatedly driven from core habitats. The ability to continuously follow animals, while greatly beneficial in tracking cryptic species like manatees, could be problematic for animals that cannot evade an aircraft. The strong disturbance responses we observed in manatees suggests regulations should require UAS pilots to exercise extra caution when using these systems near sirenians.

Individual differences in the personality and experiences of animals within each population likely drove differences in their responses to UAS flight. For example, the two repeated responses of a dolphin (18 months apart) and numerous repeated responses of a manatee (repeated in the same week and again 12 months later) suggests some animals may be more susceptible to disturbance than others. Risk may be higher if they are resident to areas with frequent UAS operations (as with these two animals) or in areas where individuals cannot be identified or distinguished. It is unknown whether repeated aircraft exposures caused behavioral habituation or sensitization in dolphins and manatees, and distinguishing these processes from naturally variable levels of tolerance in their populations will require further study (Bejder et al., 2009). For example, a mother/calf pair repeatedly exposed to the aircraft on only visibly responded in the first flight. On the other hand, the manatee previously mentioned responded in its first flight and continued to respond in many flights. Dolphins were most likely to respond to UAS toward the beginning of the first flight to which they were exposed, and tended not to respond again in up to 5 repeated flights in the same sighting. Manatees that responded tended not to be present for repeated flights, as the response flight typically caused them to flee the immediate area. Previous experiences with watercraft may affect the likelihood of animals to respond to UAS, and specific individuals or age/sex classes may vary in their susceptibility to disturbance (e.g., Lusseau, 2006). Numerous manatees were identified using scars acquired during close interactions and collisions with watercraft. Such events could sensitize animals to the noise of nearby motorized engines, such as those of our quadcopters.

### How Did Dolphins and Manatees Detect the UAS?

Both species showed evidence of using multiple sensory modalities during initial detection of the UAS and throughout their responses. Which of these modalities initially alerted animals to the aircraft, however, is still unclear. Dolphin responses involved clear visual orientation toward and investigation of the aircraft, but these orienting behaviors began after having already detected the aircraft above them. It is possible that the dolphins initially heard the sound of the quadcopters' rotors. Because manatees rapidly fled the area, there is insufficient evidence to speculate as to whether they detected the aircraft visually or acoustically. Bottlenose dolphin visual acuity is equally as good in air as it is in water in regular daylight (Herman et al., 1975), while manatees have poor visual acuity both in air and underwater (Bauer et al., 2003). An approaching aircraft presents animals with an increasingly intense and novel stimulus, both acoustically and visually, and each model used here was equipped with four downward-facing lights. These alternated between a red, blinking light, and a green, constant light. Animals may have been able to see these lights (Kerr et al., 2016). The shadow of the aircraft was not visible on the water's surface in most of our videos; furthermore, the position of the UAS relative to the sun made it unlikely that the animals would detect a shadow.

Across all responses, dolphins and manatees reacted primarily when the aircraft was directly or nearly overhead. In this position, the noise of the four active aircraft motors is greatest. The likelihood of detection of this noise, once it penetrates the water's surface, may be increased by reflection and refraction of the rotor noise off the seabed and surface (Erbe et al., 2017). Recent acoustic experiments with UAS (e.g., Christiansen et al., 2016b; Erbe et al., 2017), coupled with the established hearing capabilities of the West Indian manatee (Gerstein et al., 1999; Gaspard et al., 2012) and bottlenose dolphin (Johnson, 1967), indicate that the three aircraft used in this study produce both in-air and underwater sounds that are audible to both species. However, tests of different multi-rotor UAS models suggest that aircraft noise is unlikely to affect most marine mammals when they are underwater, both because the noise is masked by in-air ambient noise and because most of the sound energy fails to penetrate the water's surface (Christiansen et al., 2016b; Erbe et al., 2017). It is still unclear how manatees detected the UAS at altitudes of up to 104 m, or how they detected the aircraft's flight path well enough to directionally evade it. A combination of a change in altitude of the UAS with a subsequent change in its appearance, and a change in the direction of the noise from the UAS may together facilitate the animal's detection of the aircraft. Future research with animals in captivity will be useful for establishing clear behavioral and sensory thresholds for the use of UAS in studying these species.

#### Best Practices for UAS Flight Dolphins and Manatees

Mitigating the negative effects of UAS use requires taxa- or species-specific impact assessments. Flight protocols must be designed according to both data collection needs and local regulations of UAS. Our findings support the need for published best-practices guides to UAS-use (e.g., Hodgson and Koh, 2016; Smith et al., 2016). In addition, here we will propose several guidelines for multi-rotor UAS operations with marine mammals. Each of these guidelines must be further validated in future studies.


visibility (Kerr et al., 2016). The use of low-noise rotors and propellers could also significantly reduce the probability of detection and disturbance. Additionally, sufficient pilot skill and maintenance of equipment are integral to careful and controlled flight, and in preventing crashes, which are potentially dangerous for both animals and operators.

The findings of this study will inform protocols for scientific, management, and recreational use of UAS with marine mammals, both in Belize and across the range of bottlenose dolphins and Antillean manatees. Increasing unregulated recreational UAS use and a lack of resources for effective enforcement may create a problem for these species, especially at coastal destinations where tourism brings high numbers of boaters and swimmers into critical habitat for marine species. To legally fly UAS near wildlife in Belize, permits from several government offices and managing authorities are required. The Wildlife Protection Act of 1982 makes it illegal to harass marine mammals, with harassment including any disturbance that causes changes in behavioral patterns. These restrictions are similar to those imposed in the USA by the Endangered Species Act and the Marine Mammal Protection Act of 1972, which are enforced by the US Fish and Wildlife Service and the NOAA's National Marine Fisheries Service (Smith et al., 2016). Recommended flight protocols can be integrated into international regulations. For example, in the state of Florida a lack of published data on the effects of UAS exposure on manatees has resulted in restrictions on UAS use. Further study in Belize may be used to inform these permitting guidelines in Florida, despite national boundaries. In general, further study to evaluate exposure thresholds will improve the development of protocols for UAS flight with dolphins and manatees.

Findings from this study illustrate the strength of UAS to gather high-resolution observations of animal behavior. Such information can be critical for effective management of marine fauna. We demonstrated that the use of UAS follows can be effective in tracking specific fine-scale behavioral responses, even among individual animals. Orthomosaic maps produced from UAS images enabled us to precisely verify animals' location, to track individuals, and to measure detailed habitat characteristics. In addition, we developed a method for photo-ID of Antillean manatees through UAS-based imagery of their bodies (see **Figure 2**). This strategy was most useful in distinguishing individual manatees in groups and over short time scales (e.g., several weeks, 1 yr), but it is still unclear whether individual animals will be successfully re-sighted using this method over a span of multiple years. There are several limitations to this method of manatee identification that will be examined in detail in an article in preparation (Landeo-Yauri et al., in preparation). First, it was heavily dependent on the clear and calm waters of our study site. This water clarity allowed for reliable detection and tracking of manatees over sustained periods of time, and allowed us to examine individuals' scars and visible marks across their entire body. These conditions may be unavailable in many turbid manatee habitats. Secondly, the use of this method is less effective in identifying manatees with insufficient or impermanent scarring. Thirdly, the need to fly to at low altitude to obtain high-quality imagery may cause manatees to flee. Finally, the behavior of the manatees during flight sometimes made identification challenging or impossible. For example, resting manatees rarely expose more than the tip of their snout above the water, making them more difficult to see. Future studies using UAS for manatee photo-ID should carefully examine detectability of animals across field sites and conditions, prioritizing methods that minimize disturbance to target animals.

Our study was subject to several limitations. Different methodologies were employed to detect and approach dolphins and manatees based on species-specific constraints (e.g., movement patterns and group sizes). This restricted us from using identical approaches to compare the responses of dolphins and manatees to UAS. We were also limited in our acquisition of control data with no UAS present, as it was the UAS itself that enabled us to gather high-resolution overhead video of the animals. These factors may limit the generalizability of our results to other populations. Dolphins were observed primarily during boat-based deployments; these required close vessel approaches for photo-ID, and rapid aircraft launches to maintain sight of dolphin groups. Dolphins regularly interacted with the research vessel during observations, and before and after several RE. These factors may have affected the animals' behavior, changing the likelihood that they would respond to UAS (e.g., Lemon et al., 2006). Our ability to discern response differences between aircraft models was limited by logistics requiring the use of available models. The turning and movement behaviors that we used to identify UAS responses are similar to many dolphin social behaviors, and this may have caused us to overestimate UAS response levels in some cases. Unlike manatees, near-constant dolphin movement made detection of specific flight responses or movements away from the UAS infeasible in most videos. Finally, the measures of behavioral change we employed were restricted to visible behaviors and detectable changes in movement patterns. It is possible that focal dolphins and manatees exhibited changes in acoustic activity or physiological state, which we were unable to detect. Animals sometimes respond to disturbance stimuli with increased levels of stress-related hormones and chronic stress if they are unable to avoid harmful stimuli (e.g., Rolland et al., 2012). For example, American black bears (Ursus americanus) and king penguin (Aptenodytes patagonicus) chicks equipped with cardiac biologgers responded to UAS flights with elevated heart rates, at times with no observable behavioral responses (Ditmer et al., 2015; Weimerskirch et al., 2017). Tests using blow samples collected from whales, or measurement of stress hormones and blood cortisol levels in captive marine mammals, will be valuable in evaluating these "invisible" physiological effects of UAS response. Future studies examining species-specific responses to UAS may prioritize shore-based operations to reduce the bias introduced by a nearby research boat. Such studies may also make use of multi-sensor tags to track study animals, rather than relying on visual evidence alone.

### ETHICS STATEMENT

This study was carried out in accordance with the recommendations of the Belize Fisheries and Forestry Departments [Permit Ref. No. WL/1/1/16(28)] of the Government of Belize. This study was carried out in accordance with the guidelines of the IACUC committee at Hunter College.

### AUTHOR CONTRIBUTIONS

ER, BM, MM, and DR conceived and designed the study. ER and BM collected and analyzed the data. ER, BM, MM, and DR wrote the paper.

### FUNDING

This research was supported in part by grants from the National Science Foundation (grant number PHY-1530544) and The Eric and Wendy Schmidt Fund for Strategic Innovation. ER was partially funded by the Office of Educational Opportunity & Diversity at The Graduate Center, CUNY.

### ACKNOWLEDGMENTS

We thank the countless students, volunteers, and personnel who participated in the field research for this study, including groups with Ecology Project International, National Geographic Student Expeditions, University of Belize, the Environmental Research Institute on Calabash Caye, SEEtheWild, SEE Turtles, and Smithsonian Student Expeditions. Thanks to Sarah Landeo-Yauri for working on photo-ID. Thanks to HelloOcean for support with catamaran research, and The Moorings for their financial and logistical support in donating vessel time to our cause. Special thanks to A. Jeffords for captaining the boat and catching the UAS. Many thanks to L. and J. Searle at ECOMAR for logistical support at St. George's Caye. Thanks to the Oceanic Society for providing financial and logistical support, including housing for ER and other members of our research team. All UAS were owned by The Rockefeller University and ER, and were piloted by ER. This study and all data collection protocols were conducted under research permits granted to ER by several governmental institutions in Belize. Thanks to the Forest Department and Fisheries Department of Belize for permits for marine mammal research, to the Belize Department of Civil Aviation for permission to fly the UAS, and to the Public Utilities Commission for import permits for the crafts. Finally, thank you to Megan S. McGrath and the three reviewers whose comments and recommendations vastly improved this manuscript.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2018.00316/full#supplementary-material

#### REFERENCES


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

Copyright © 2018 Ramos, Maloney, Magnasco and Reiss. 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 Utility of Combining Stable Isotope and Hormone Analyses for Marine Megafauna Research

Alyson H. Fleming1,2 \*, Nicholas M. Kellar<sup>3</sup> , Camryn D. Allen<sup>4</sup> and Carolyn M. Kurle<sup>5</sup>

<sup>1</sup> Department of Paleobiology, National Museum of Natural History, Smithsonian Institution, Washington, DC, United States, <sup>2</sup> Department of Biology and Marine Biology, University of North Carolina at Wilmington, Wilmington, NC, United States, <sup>3</sup> Ocean Associates, Inc., Marine Mammal and Turtle Division, Southwest Fisheries Science Center, National Marine Fisheries Service, National Oceanic and Atmospheric Administration, La Jolla, CA, United States, <sup>4</sup> The Joint Institute for Marine and Atmospheric Research, Protected Species Division, Pacific Islands Fisheries Science Center, National Marine Fisheries Service, National Oceanic and Atmospheric Administration, Honolulu, HI, United States, <sup>5</sup> Division of Biological Sciences, Ecology, Behavior, and Evolution Section, University of California, San Diego, La Jolla, CA, United States

#### Edited by:

Lars Bejder, University of Hawai'i at Manoa, ¯ United States

#### Reviewed by:

Luis Cardona, University of Barcelona, Spain Neil Randell Loneragan, Murdoch University, Australia

\*Correspondence: Alyson H. Fleming FlemingA@SI.edu; alyson.fleming@gmail.com

#### Specialty section:

This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science

Received: 20 December 2017 Accepted: 04 September 2018 Published: 25 September 2018

#### Citation:

Fleming AH, Kellar NM, Allen CD and Kurle CM (2018) The Utility of Combining Stable Isotope and Hormone Analyses for Marine Megafauna Research. Front. Mar. Sci. 5:338. doi: 10.3389/fmars.2018.00338 Stable isotope and hormone analyses offer insight into the health, stress, nutrition, movements, and reproduction of individuals and populations. Such information can provide early warning signs or more in-depth details on the ecological and conservation status of marine megafauna. Stable isotope and hormone analyses have seen rapid development over the last two decades, and we briefly review established protocols and particular questions emphasized in the literature for each type of analysis in isolation. Little has been published utilizing both methods concurrently for marine megafauna yet there has been considerable effort on this front in seabird and terrestrial predator research fields. Using these other taxa as examples, we offer a few of the major research areas and questions we foresee as productive for the intersection of these two methods and discuss how they can inform marine megafauna conservation and management efforts. Three major research areas have utilized a combination of these two methods: (1) nutrition and health, (2) reproduction, and (3) life history. We identify a fourth area of research, examinations of evolutionary versus ecological drivers of behavior, that could also be well served by a combined stable isotope and hormone analyses approach. Each of these broad areas of research will require methodological developments. In particular, research is needed to enable the successful temporal alignment of these two analytical techniques.

Keywords: stable isotope, hormone, mammal, turtle, cetacean, pinnipedia, marine

## INTRODUCTION

Threats to marine megafauna continue to multiply, and management of these top predators is complex, challenging, and costly. Some threats are relatively visible (e.g., ship strikes or decreased sea ice), whereas others are more cryptic (e.g., ocean noise or climate change). Even the most visible instances of natural or anthropogenic impacts are exceedingly difficult to quantify and the population-level consequences for both lethal and non-lethal factors are usually unknown (Taylor et al., 2000; Read, 2008; Robards et al., 2009). Despite substantial research effort to monitor populations through traditional survey methods, major declines in population abundance

are likely to go undetected (Taylor et al., 2007). To date, most management actions are not proactive, but rather occur after a deleterious or catastrophic event (e.g., Deepwater Horizon oil spill or an unusual mass mortality event) when it is often too late for feasible corrective options. Additionally, effects from climate change pose a current and emerging threat with complex and varied consequences, which can be difficult to predict (Cai et al., 2014; Stock et al., 2014). Given the multitude of risks facing marine megafauna and the limitations of current management practices, there is an ever-increasing need to understand responses of these taxa to natural stressors and to anthropogenic activities (National Academies of Sciences, Engineering, and Medicine, 2017).

Stable isotope and hormone analyses are two minimally invasive methods that may allow for finer-scale characterization of the severity, types, and consequences of megafaunal responses to their changing environment. These methods use integrated physiological biomarkers that can provide insight on the health, stress, nutrition, movements, and reproduction of individuals and populations. Such information can provide early warning signs or more in-depth information on the ecological and conservation status of marine megafauna. Consequently, biologists can use these techniques in tandem to better target strategies and timing for intervention, resulting in more anticipatory species management.

#### Stable Isotope Ecogeochemistry

Analyses of stable carbon (δ <sup>13</sup>C) and nitrogen (δ <sup>15</sup>N) isotope ratios of bulk tissues from marine organisms are used to reconstruct habitat use and trophic ecology of animals that are typically cryptic, difficult to monitor, and wide-ranging in their migration and movement patterns (e.g., Kurle and Gudmundson, 2007; Newsome et al., 2010; Authier et al., 2012a; Turner Tomaszewicz et al., 2017). These analyses are informative because the δ <sup>13</sup>C and δ <sup>15</sup>N values from bulk tissues reflect the underlying biogeochemistry driving the stable isotope values in the primary production at the base of the food web (Peterson and Fry, 1987; Trueman et al., 2012; Lorrain et al., 2014). In addition, internal physiological processes in the consumer result in their δ <sup>15</sup>N values increasing predictably relative to those from their prey (Post, 2002; Kurle et al., 2014), and these differences, or trophic discrimination factors, allow for estimations of animal trophic position (DeNiro and Epstein, 1981; McMahon et al., 2015c). The stable isotope values, usually δ <sup>13</sup>C and δ <sup>15</sup>N, from consumer and prey, can be plotted in bivariate space or incorporated into various stable isotope mixing models, to create a picture of population- or species-level isotopic niche space ("isospace"), which allows for a better understanding of an organism's ecological niche (**Figure 1**; Newsome et al., 2007b; Stock et al., 2018).

#### Hormones

Recent developments in hormone analysis provide critical insights into aspects of marine megafauna biology, ecology, and population health that have previously been nearly impossible to obtain. Assessing the impacts of stressors on marine species is often limited to counts of dead or injured animals, with

FIGURE 1 | Schematic illustrating potential factors influencing the stable isotope niche space or isospace of marine megafauna in ocean systems. Stable carbon isotope values provide information regarding animal foraging location as these values most often reflect patterns at the base of the food web (Rau et al., 1982; Fry, 2006; Ben-David and Flaherty, 2012). For example, animals foraging offshore usually have lower δ <sup>13</sup>C values than those foraging nearshore. The spectrum of productivity refers to conditions that influence phytoplankton growth rates as higher nutrient inputs result in faster growth rates leading to higher δ <sup>13</sup>C values for all levels of a food web (Bidigare et al., 1997; Popp et al., 1998). Stable nitrogen isotope values inform animal trophic position and reflect nitrogen processes at the base of the food web that drive stable isotope values for all consumers within that system (see overview in Rau et al., 2003; Somes et al., 2010; Allen et al., 2013; Kurle and McWhorter, 2017; Turner Tomaszewicz et al., 2017). Here, we highlight the natural nitrogen cycle influences on the base of the food web which create differences across ecosystems, but anthropogenic nitrogen loading from point and run-off sources on land can also alter nitrogen signatures of primary producers (Costanzo et al., 2001; Lemons et al., 2011). Fractionation between diet and consumer can vary across species for both carbon and nitrogen and is likely due to diet composition, quality, trophic position, and form of nitrogen excretion (Germain et al., 2013; McMahon et al., 2015b,c; Nielsen et al., 2015; Kurle et al., 2014).

an understanding that data collected from these individuals represent a small unknown fraction of the total population and may be biased toward those experiencing extreme stress. In this way, hormone analysis of samples collected from freeranging individuals may provide earlier awareness of damaging effects and more direct evidence of relationships between stressors, and increases in dead or injured animals. Specifically, reproductive hormones can reveal an animal's sex, maturity, and pregnancy status, enabling interpretations of demographic structure, birth rates, and sex ratios, as well as the potential to assess lost or infrequent pregnancies due to exposure to harmful conditions (e.g., Rolland et al., 2012; Schwacke et al., 2014; Kellar et al., 2017). Corticosteroid hormones ("stress hormones") can elucidate both acute (e.g., ephemeral predator exposure) and chronic conditions (e.g., nutritional deficits) (e.g., Sheriff et al., 2011), whereas thyroid hormones provide

additional information about nutritive state (e.g., Atkinson et al., 2015).

Our paper (1) briefly summarizes the contributions of stable isotope and endocrine analyses to marine megafauna conservation and management to date and the emerging and developing applications of these methods, (2) identifies opportunities for the combination of these methods, which may reveal exciting new insights into the physiology, ecology, and conservation of these species, and (3) outlines gaps and future work required to advance these fields. In this review, we restrict our discussion of marine megafauna to cetaceans, pinnipeds, and sea turtles.

### STABLE ISOTOPE ECOGEOCHEMISTRY IN MARINE MEGAFAUNA RESEARCH

#### Applications of Stable Isotope Analyses

Estimating animal trophic levels and foraging locations has been the classic application of stable isotope data measured in animal tissues for ecological purposes (DeNiro and Epstein, 1978, 1981; Hobson and Welch, 1992). The development of progressively more sophisticated analytical methods, such as stable isotope mixing models (SIMMs) that incorporate multiple parameters, including stable isotopes of elements besides carbon and nitrogen such as sulfur, oxygen, and hydrogen, have allowed for increasingly detailed estimations of animal ecological niche space using stable isotope data (Jackson et al., 2011; Newsome et al., 2012; Hopkins and Kurle, 2016; Rossman et al., 2016; Bowes et al., 2017; Hopkins et al., 2017). There now exist a wide array of modeling frameworks and metrics for categorizing diet, trophic niche, and trophic structure (Bearhop et al., 2004; Layman et al., 2007; Jackson et al., 2011; Newsome et al., 2012; Stock et al., 2018).

Questions of competition and resource partitioning, foraging plasticity, and maternal provisioning have also been investigated with stable isotope methods (Borrell et al., 2006; Kiszka et al., 2010; Fernández et al., 2011; Authier et al., 2012b; Ryan et al., 2013). Ryan et al. (2013) found evidence of resource partitioning amongst sympatric species of rorquals (Balaenopteridae) in the North Atlantic through analysis of baleen isotopes, while Authier et al. (2012b) investigated impacts of maternal feeding strategy on pup weaning mass in southern elephant seals (Mirounga leonine). Stable isotope analyses have also illuminated population structure, examining ecological and trophic differences within species (Witteveen et al., 2009a,b; Barros et al., 2010; Lowther and Goldsworthy, 2011; Giménez et al., 2013). For example, carbon, nitrogen, and sulfur isotopes were used to differentiate putative population groups of bottlenose dolphins (Tursiops truncatus) off Florida (Barros et al., 2010) and carbon and nitrogen distinguished both breeding and feeding groups of humpback whales (Megaptera novaeangliae) in the North Pacific (Witteveen et al., 2009a,b).

Expansion of the ecological applications of stable isotope analyses have allowed for reconstructions of temporal and spatial variations in animal habitat use as species move along migration routes or target specific feeding grounds throughout multiple life stages (Hobson, 1999; Kurle, 2009; Vander Zanden et al., 2010; Authier et al., 2012c; Allen et al., 2013; Carlisle et al., 2014). For example, Turner Tomaszewicz et al. (2017) analyzed the δ <sup>15</sup>N values from individual growth rings in humerus bones collected from dead-stranded North Pacific loggerhead turtles (Caretta caretta) to demonstrate their use of both oceanic and neritic regions during their decades-long stage as juveniles. The analyses of archived tissues allow for expanded temporal reconstructions of animal diets and habitat use on the order of decades or longer (Newsome et al., 2007a; Fleming et al., 2016). Fleming et al. (2016) linked environmental variability in the California Current System to variations in humpback whale diets over 20 years using isotope values from whale skin. However, temporal and spatial investigations of predator diet and trophic level can be complicated by isotopic changes at the base of the food web, which vary by region, season, and year (Kurle et al., 2011). Thus, the degree of baseline variability must be considered, and ideally estimated from lower trophic level sampling, before interpretations of predator ecologies and movements are drawn (Lorrain et al., 2014; Kurle and McWhorter, 2017).

### Compound Specific Stable Isotope Analysis

More recently, advances in compound specific stable isotope analysis of individual amino acids (CSIA-AA) allow for more thorough explorations of trophic level and foraging location than bulk stable isotope analyses. CSIA-AA enables differentiation between isotopic variation due to different biochemical processes at the base of the food web versus changes in a consumer's trophic level (Popp et al., 2007; Chikaraishi et al., 2009; Ruiz-Cooley and Gerrodette, 2012; Lorrain et al., 2014; Ruiz-Cooley et al., 2014; O'Connell, 2017). Thus, even without stable isotope values from temporally or spatially linked lower trophic level organisms, temporal and spatial shifts in predator diet can often be determined. The δ <sup>15</sup>N values from so-called "source" amino acids (essential amino acids for δ <sup>13</sup>C) show little change or isotopic fractionation as they are transferred up the food web, whereas other "trophic" amino acids (non-essential amino acids for δ <sup>13</sup>C) fractionate with increasing trophic level. Comparison of the isotope values from these two categories of amino acids allows for more nuanced interpretation of stable isotope data. This emerging technique has made most use of the δ <sup>15</sup>N values from amino acids (Sherwood et al., 2011; McMahon et al., 2015a), but ecological applications for the δ <sup>13</sup>C values from amino acids are becoming more apparent, especially for delineating amino acid sources in diets of consumers (Larsen et al., 2009, 2013; Nielson and Winder, 2015). While measurements of amino acid isotopes from marine megafauna are increasing (Arthur et al., 2014; Ruiz-Cooley et al., 2014, 2017; Pomerleau et al., 2017; Zupcic-Moore et al., 2017), there is a need for methodological development specific to these taxa as their unique physiologies and isotopic fractionation patterns necessitate different considerations than lower trophic level taxa (e.g., zooplankton, corals) (McMahon et al., 2015b; McMahon and McCarthy, 2016).

The increasing use of CSIA-AA for modern and archived samples will allow for greater understanding of mechanistic links between patterns in oceanographic parameters, stable isotope geochemistry, and marine megafauna responses. Variations in the bulk δ <sup>13</sup>C and δ <sup>15</sup>N values from marine species collected over time and space can be attributed to changes in oceanographic measures that are in turn driven by climatic conditions (Kurle et al., 2011; Ohman et al., 2012; Allen et al., 2013; Kurle and McWhorter, 2017). For example, higher ocean temperatures are related to less nutrient availability, which in turn correlate to slower growth rates and lower δ <sup>13</sup>C values for phytoplankton (Bidigare et al., 1997; Popp et al., 1998; Schell, 2000; **Figure 1**). CSIA-AA of archived marine samples covering longer time periods may therefore allow for reconstruction of productivity and other trends related to long-term climate patterns that may be driving oceanographic properties of interest to bottom-up control of food webs and impacts on top predators (Hückstädt et al., 2017).

### Physiological Considerations for Stable Isotope Applications

The physiology and metabolism of protein utilization are important to consider when using stable isotope analysis to reconstruct foraging patterns in marine vertebrates. First, the time required for full isotopic turnover varies across tissues due to different rates of protein metabolism and can be on the order of a few days (blood plasma, liver), months (muscle, red blood cells), or longer (bones) (Kurle, 2009; Vander Zanden et al., 2015). Accretionary tissues (e.g., vibrissae, ear plugs, and tooth dentine) produce inert layers that preserve their original chemical composition and allow for serial reconstructions over multiple years or even lifetimes. Also to be considered, single tissue types can have variable turnover rates across species and within individuals. For example, sea turtles retain growth layers in cortical bone but mammalian bone remodels and, therefore, integrates multiple years of growth information (Snover et al., 2011; Riofrío-Lazo and Aurioles-Gamboa, 2013). Within the suborder Caniformia, isotopic differences were found between cortical and non-cortical bones within individuals (Clark et al., 2017). Therefore, stable isotope values from various tissues reflect different time periods during which nutrients were ingested and incorporated and should be considered when using these analyses in conjunction with hormone studies (**Figure 2**).

Second, animals undergoing nutritional stress must rely on their own tissue catabolism or other mechanisms to maintain function (see Elia et al., 1999; Aguilar et al., 2014; Borrell et al., 2016), and these adaptations to resource limitation or starvation can vary in their effect on the δ <sup>13</sup>C and δ <sup>15</sup>N values in tissues. Much evidence points to an increase in the δ <sup>15</sup>N values of tissues for animals undergoing protein catabolism when they are starving (Hobson et al., 1993; Polischuk et al., 2001; Cherel et al., 2005; Lohuis et al., 2007; Newsome et al., 2010; Bowes et al., 2014). In contrast, mammals that appear to rely more heavily on fat reserves or other processes that conserve protein during times of fasting appear to demonstrate decreasing or unchanging δ <sup>15</sup>N values when under nutritional stress (Das et al., 2004; Lohuis et al., 2007; Gomez-Campos et al., 2011; Aguilar et al., 2014). One explanation for these inconsistencies may be related to the amount of lipid reserves stored by an organism undergoing nutritional stress, as that appears to influence the degree to which protein versus lipid is catabolized for energy (Elia et al., 1999) and can thus influence an animal's stable isotope values (Aguilar et al., 2014). For mysticetes that are capital breeders, gestation is thought to occur during periods of fasting and requires substantial protein resources. Fetal development may lead to a decrease in δ <sup>15</sup>N values for the mother throughout the pregnancy as the fetus's tissues increase in their δ <sup>15</sup>N values relative to the mother (Borrell et al., 2016). Therefore, it is important to consider the life history of the animal (e.g., capital vs. income breeders) and the mechanisms responsible for the potential inconsistencies in stable isotope markers for nutritional stress.

#### APPLICATIONS OF STRESS AND REPRODUCTIVE HORMONE ANALYSES

Only recently have hormone data been employed regularly in studies of marine vertebrate ecology and conservation. Since their discovery, these biochemicals that signal between cells and organ systems have been primarily measured in clinical settings to help assess health and reproductive conditions of individual animals. Numerous veterinary and human medical studies have created volumes of information regarding their physiological effects, biochemistry, reference ranges, and associated anomalies (Pineda et al., 2003; Melmed et al., 2016). There was significant work done in the second half of the 20th century analyzing various hormones in marine megafauna to understand their unique physiologies (Deroos and Bern, 1961; Malvin et al., 1978; Liggins et al., 1979; St. Aubin and Geraci, 1988, 1989; Hochachka et al., 1995). However, it was not until the 2000s that these analyses became more common for marine wildlife researchers to assess hormones for conservation and physiological ecology studies, and many of these efforts were aimed at establishing baselines (Mansour et al., 2002; Mashburn and Atkinson, 2004; Rolland et al., 2005; Hunt et al., 2006; Kellar et al., 2006; Blanvillain et al., 2011; St. Aubin et al., 2013). Thus, there are relatively few examples of applied studies examining links between animal hormone levels and exposure to potentially harmful human activities or environmental conditions (Rolland et al., 2012; Kellar et al., 2013; Schwacke et al., 2014; Williard et al., 2015).

The specific molecules within steroid (e.g., progesterone and cortisol) and amino-acid derived (e.g., epinephrine and thyroid hormones) hormone classes are structurally identical across most vertebrate species (Horton and Moran, 1996; Pineda et al., 2003; Melmed et al., 2016). Because of this structural similarity among diverse taxa and their stability despite a wide spectrum of harsh field collection conditions, ecology and conservation efforts to date have mostly focused on analyses of these two hormone groups.

While the specific molecules are structurally similar across taxa, their physiological roles can be quite different across both individuals and species. Additionally, interpretation of hormone levels and patterns varies by hormone type and research question. For example, measurement of hormone levels can indicate a physiological state controlled directly by the measured hormones (e.g., hormonal controls on pregnancy and maturation) (Theodorou and Atkinson, 1998; Owens, 1999; Greig et al., 2007; Kellar et al., 2009, 2014; Perez et al., 2011; Vu et al., 2015). Alternatively, hormone levels can be assessed as an indicator of direct response to some external stimuli (e.g., fightor-flight responses) (Gregory et al., 1996; St. Aubin and Dierauf, 2001; Sheriff et al., 2011). In more recent years, there has been another avenue of investigation into changes in hormone levels due to secondary or indirect responses, such as those associated with exposure to high levels of environmental contaminants (Subramanian et al., 1987; Oskam et al., 2003; Trego et al., 2018). The following paragraphs introduce the various types, applications, and interpretations of hormones most commonly used in marine megafauna research.

#### Sample Matrices

Marine megafauna have diverse sample matrices (tissues, body fluids, and physiological end-products) from which hormones are measured. Relative to hormone distribution patterns, the matrices can be subdivided into three types: (1) those that are in dynamic equilibrium (e.g., blood, blubber, most bone tissue, and muscle) with the hormone concentrations generated by the gland of production, (2) those that become relatively static once formed thereby potentially offering a record of previous hormone concentrations (e.g., laminated ear plugs, fur, hair, whiskers, baleen, claws, laminated bone structures, and potentially, teeth, tusks, and epidermal tissue), and (3) those that are biological end products formed then expelled (e.g., blow particulate ("whale snot"), respiratory vapor, feces, urine, saliva, egg shells, and milk). Moreover, as with stable isotopes, each matrix has its own set of dynamics, integrating signals over varying lengths of time depending on each individual hormone and the matrices' chemical characteristics (**Figure 2**). Note that the processing of these matrices for hormone analysis typically has two phases: (1) isolation of the target hormone and (2) analytical measurement. The measurement procedures (e.g., immunoassays, chromatography, and mass spectrophotometerbased analytical analyses) can theoretically be used with any of the matrices; however, isolation procedures vary enormously as they are tailored to each matrix's specific composition and chemistry relative to the target hormone.

#### Progesterone

Progesterone is a particularly informative indicator of cetacean pregnancy status, which researchers have used to estimate pregnancy rates in several populations (Bergfelt et al., 2013; Kellar et al., 2013, 2014; Clark et al., 2016). In situations where known stressors or atypical perturbations are of concern (e.g., exposure of dolphins to chase-and-encirclement fishery activity), relationships between the frequency or magnitude of exposure to these perturbations and pregnancy rates have been assessed (Kellar et al., 2013). Similarly, reproductive success rates (rates of known pregnancies producing viable calves) have been examined with respect to poor prey availability (Wasser et al., 2017) or

exposure to pollutants (Kellar et al., 2017). Progesterone levels in pinnipeds have also been shown to aid in assessing the pregnancy state of individuals (Gardiner et al., 1996, 1999; McKenzie et al., 2005; Greig et al., 2007). However, in these species, there can be overlap in hormone concentration between known nonconceiving females and females predicted to be pregnant based on progesterone concentrations such that the diagnostic power of progesterone is more limited though still informative (McKenzie et al., 2005; Beaulieu-McCoy et al., 2017). In these animals (along with cetaceans and sea turtles), progesterone can aid in maturity state assessments of females and help elucidate estrous activity in conjunction with measurements of other hormones, like estrogens (Licht et al., 1982; Pietraszek and Atkinson, 1994; Gardiner et al., 1996, 1999; Kakizoe et al., 2010; Beaulieu-McCoy et al., 2017). Progesterone concentration has not been widely assessed in marine turtle experimental studies as this hormone is at baseline levels until a reproductively active female commences nesting activities (see overview in Blanvillain et al., 2011).

#### Androgens

Androgen concentrations, especially those of testosterone, have been applied in a variety of ways in marine megafauna ecology and conservation research. In marine mammals, testosterone helps assess maturity states of individual males (e.g., in blubber, Kellar et al., 2009) and, in more static matrices (e.g., laminated ear plugs), it is used to estimate ages of sexual maturation (Trumble et al., 2013). Testosterone measurements can elucidate reproductive seasonality characteristics of populations, information that can help identify time-of-year associated with heightened vulnerability to potential stressors (i.e., periods of breeding; Robeck and Monfort, 2006; Kellar et al., 2009; Vu et al., 2015). More generally, pollutant exposure has been linked to decreased levels of androgens, as well as estrogens, in a number of marine megafauna species creating potential impacts on development and reproduction (Subramanian et al., 1987; Oskam et al., 2003; Trego et al., 2018). Finally, androgen levels are utilized to help determine the sex of individual animals (Allen et al., 2015; Corkeron et al., 2017). This is particularly important for sea turtles as they display no genotypic markers for sex or external secondary sexual characteristics until maturation (males grow longer tails), and, because their sex is temperature dependent, monitoring sex ratios of turtle populations using testosterone can help identify potential negative impacts of changing environmental conditions on their demography (Allen et al., 2015; Braun McNeill et al., 2016; Jensen et al., 2018).

Since hormone concentrations in hatchling sea turtles are minute, the ratio of estradiol to testosterone was used to predict the sex of sacrificed, just-hatched sea turtles (Gross et al., 1995). However, advances in hormone assay technology should allow detection of hormone concentrations in small volumes of blood collected from hatchlings without deleterious effects (e.g., death or low survivorship), providing the opportunity to assess primary sex ratios for all species of sea turtles.

#### Corticosteroids

In marine megafauna, the dominant corticosteroids (cortisol, corticosterone, and aldosterone) serve as biochemical markers of physiological stress response activity; specifically they are most often interpreted as indicators of hypothalamic–pituitary– adrenal (HPA) axis activation in response to perceived threats (Gregory et al., 1996; St. Aubin and Dierauf, 2001; Sheriff et al., 2011). For the HPA axis, glucocorticoids (GC), cortisol and corticosterone, stimulate creation of additional glucose in anticipation of increased energy needs in response to a stressor (Palme et al., 2005; Busch and Hayward, 2009; Blanvillain et al., 2011; Atkinson et al., 2015). GC concentrations have been analyzed in numerous marine mammal and turtle species ranging from individual responses to stressors like restraint or capture of individual animals (Thomson and Geraci, 1986; St. Aubin and Geraci, 1988, 1989; Gregory et al., 1996; Champagne et al., 2012; St. Aubin et al., 2013; Williard et al., 2015; Hunt et al., 2016a), to population-level responses to increased anthropogenic activity (e.g., vessel traffic; Ayres et al., 2012; Rolland et al., 2012). Glucocorticoid concentrations also vary with temperature perturbations (Houser et al., 2011), nutritional deficits (Kellar et al., 2015; Beaulieu-McCoy et al., 2017; Wasser et al., 2017), and pollutant exposure (most notably oiling from the Deepwater Horizon disaster; Schwacke et al., 2014). These may not be stressors in the classical sense and are therefore often referred to as types of "environmental stressors" as there is not necessarily a perception of threat at an individual level. This distinction is important as these conditions do not necessarily stimulate cortisol production, especially as an anticipatory reaction, as seen in a perceived threat-to-self response (also known as a fight-or-flight response).

The other primary corticosteroid, aldosterone, mainly controls the electrolytic composition of blood to regulate blood pressure and blood volume (St. Aubin, 2001). In marine mammals, aldosterone often shows responses to known stressors that are similar to those of GCs (Thomson and Geraci, 1986; St. Aubin and Geraci, 1989; St. Aubin et al., 1996; Houser et al., 2011). A sea turtle study investigated the effects of acute fresh water exposure and found no change in aldosterone or corticosterone production and suggested that compared to marine mammals, sea turtle response to a hypo-osmotic environment might be delayed (Ortiz et al., 2000). Aldosterone often shows greater relative increases compared to cortisol, though at much lower total concentrations (St. Aubin and Geraci, 1989). It is theorized that this may be due in part to the importance of breath-hold diving (and the profound accompanying changes in blood distribution) in these animals as they respond to potential threats (Atkinson et al., 2015). These observations are creating new interest in using aldosterone as another marker of stress, and, though few studies have examined aldosterone levels relative to known stressors outside of experimental settings, this will likely change in the near future.

#### Thyroid Hormones

Along with GC levels, thyroid hormones (thyroxine, T4 and triiodothyronine, T3) can help in the assessment of individual and population-level nutritional conditions (Atkinson et al., 2015). T4 is a prohormone and typically is converted to T3, the active form, to directly bind to receptors and produce biological effects (St. Aubin, 2001; Melmed et al., 2016). These effects

help control entire metabolic rates of individuals by regulating numerous metabolic pathways. Limited food intake generally inhibits the production and activity of thyroid hormones, especially T3 (Pineda et al., 2003; Melmed et al., 2016). This works by (1) blocking the conversion of T4 to T3 and (2) by stimulating the conversion of both to another form, "reverse T3" (rT3). In this form, rT3 still binds to T3 receptors, but produces no biological effects thereby blocking the associated pathways and lowering the overall metabolic rate (Horton and Moran, 1996; Pineda et al., 2003). Though a number of studies have looked at the effects of food limitations on the thyroid concentrations of individuals (Moon et al., 1998, 1999; Rosen and Trites, 2002; Rosen and Kumagai, 2008; du Dot et al., 2009), few have used thyroid hormone measurements to help assess the populationlevel impacts of limited prey availability in non-captive animals (Moon et al., 1998; Ayres et al., 2012; Crocker et al., 2012; Wasser et al., 2017).

#### INTEGRATED APPLICATIONS OF STABLE ISOTOPE AND HORMONE ANALYSES

In addition to the established and emerging applications of both stable isotope and hormone analyses in marine megafauna presented above, the combination and integration of these two methods offer innumerable possibilities for academic and conservation focused research. While little has been published utilizing both methods concurrently for marine megafauna (Hunt et al., 2014, 2016b; Clark et al., 2016), there has been considerable effort on this front in seabird and terrestrial predator research fields. Using these taxa as examples, we offer a few of the major research areas and questions we foresee as productive for the intersection of these two methods and discuss how they can inform marine megafauna conservation and management efforts. Three major research areas have utilized a combination of these two methods: (1) nutrition and health, (2) reproduction, and (3) life history. We identify a fourth area of research, examinations of evolutionary versus ecological drivers of behavior, that could also be well served by a combined stable isotope and hormone analyses approach.

#### Nutrition and Health

Numerous external influences on marine megafauna can impact nutrition and health, including changes in prey availability and/or quality, degraded or inconsistent habitat conditions (due to natural variability or anthropogenic factors), competition, disease, and direct anthropogenic impacts such as entanglement, ocean noise, or ship strikes. To date, the research effort combining stable isotope and hormone analyses has focused largely on the impacts of variable prey and habitat conditions (e.g., Barger and Kitaysky, 2011; Dorresteijn et al., 2012; Bryan et al., 2013, 2014; Lafferty et al., 2015; Boggs et al., 2016). As top predators, marine megafauna depend on abundant prey and/or dense prey patches and they often capitalize on regions of reliably high biological productivity (Hazen et al., 2009; Santora et al., 2011). However, these areas are subject to climatedriven environmental fluctuations that affect marine productivity (e.g., Macklin et al., 2002; Bograd et al., 2009; Sherwood et al., 2011; Stabeno et al., 2012), contributing to the potential for episodic inadequate prey resources for megafauna that can induce stress, increase competition, create nutritional deficits, and lead to potential starvation. Understanding the linkages between climate, habitat, prey availability, and marine mammal diets, and predicting how these variables impact megafauna conservation and management, is increasingly important in the face of ongoing warming (e.g., Howard et al., 2013).

Currently, the influence of climate change on prey availability for top ocean predators is a topic of considerable research effort across various temporal scales (Wolf et al., 2010). To understand the impacts of current and future changes in climate on marine megafauna, analyses of interannual and decadal patterns are often used as proxies to better understand and predict future population responses to long-term change. For example, Dorresteijn et al. (2012) examined impacts of interannual climate variability and timing of ice retreat on food availability for least auklets (Aethia pusilla), a seabird in the Bering Sea. They combined assessment of changes in auklet diet as measured by stable isotope analyses and regurgitated chick meals with changes in the stress hormone, corticosterone, in auklet blood to assess food availability (higher levels of corticosterone are associated with lower food availability). The combination of isotope, regurgitate, and hormone analyses revealed a more in-depth understanding of the species response than any method in isolation, as two different colonies were found to respond differently to the changing climatic and oceanographic conditions. Only one colony showed changes in diet, but both colonies showed increased levels of corticosterone during warm periods, indicating that, while the diet may not have changed for both colonies, the relative foraging effort may have (Dorresteijn et al., 2012). Such insights are particularly helpful in a management context as they provide further metrics of the consequences of environmental changes on top predators.

The combination of stable isotope and hormone analyses has also been utilized to examine the interplay of intra- and interspecies competition in relation to changing prey conditions. Barger and Kitaysky (2011) demonstrated increasing dietary separation (assessed by stable isotopes) among two species of sympatric seabirds (Uria spp.) in response to food limitation (supported by increasing concentrations of stress hormone) and greater niche overlap when food was abundant. Within-species resource competition was investigated by Bryan et al. (2014) in grizzly (Ursus arctos) and black bears (Ursus americanus) in the Pacific northwest through stable isotope Bayesian mixing models and surveys of salmon (Oncorhynchus spp.) abundance. In grizzly bears, cortisol increased in response to lower salmon consumption. However, in black bears, cortisol increased in response to lower salmon availability, which was tightly coupled to increased competition, indicating a stronger link to social competition in black bears. In both bear species, testosterone decreased with increasing salmon availability, which the authors interpreted as evidence of a less competitive environment. Consequences of diet quality can also be investigated with a

combination of hormone and stable isotope analyses. Fairhurst et al. (2015) found that feathers from Leach's storm petrels (Oceanodroma leucorhoa) with higher δ <sup>15</sup>N values, an indication they were foraging at higher trophic levels, were associated with correspondingly lower corticosterone concentrations. Their data suggested a physiological benefit in the form of either a reduced foraging effort or a greater nutrient benefit associated with consumption of higher trophic level prey by the petrels.

Further research directions on this topic could include investigations of health conditions that may be distinct from dietary inadequacies but could be mistaken for responses to nutritional stress without more detailed data. For example, energetic and endocrine responses to diseases resulting from ecotoxicological or immunological factors may also be aided by a combination of hormone and stable isotope analyses. While most diseases in the wild may have a nutritional component, there may be some instances of direct links to a disease or injury stressor that may not be primarily mediated through prey or nutritional stress. In these cases, assessing both hormone levels and stable isotope values can allow biologists to rule out biologists to rule out the potential for interactions between health and diet.

#### Reproduction

Reproductive behavior, biology, and rates are exceptionally challenging to study in marine megafauna and, for many species, the oceanic habitats in which these animals breed remain unknown (Robeck et al., 2001; Blanvillain et al., 2011). For taxa with more visible reproductive behaviors, such as sea turtles and pinnipeds, there has been substantial examination of hormone levels and patterns. However, we are aware of little research combining hormone and stable isotope analyses for better deciphering questions related to reproduction in other marine species. One of the most fundamental research topics that can be developed by the integration of these two methodologies is baseline physiological patterns inherent to particular reproductive or life history stages. For example, does an individual's ontogenetic stage influence diet and stress or diet and reproductive endocrinology? Do the diverse metabolic needs required at different reproductive or life stages lead to different prey preferences or foraging efforts? Addressing these questions will develop critical baselines from which other, more acute, questions can be added.

Some investigations into reproductive questions using both stable isotope and hormone analyses interpreted these analyses separately, whereas others drew integrated conclusions. Both approaches can add value to single discipline studies, however, we encourage future research to consider the interactions and background physiological conditions that may be driving both the hormonal and stable isotope patterns. For example, Hunt et al. (2016b) examined cortisol and corticosterone changes along baleen plates from two female right whales (Eubalaena glacialis) in relation to observed life stage/reproductive events. They used stable isotope values to demarcate time along the baleen plate, as these values change predictably with annual migration cycles, but they did not consider the potential interplay between hormone and stable isotope values. They found that corticosterone was elevated during pregnancy and lactation, whereas cortisol had more variable, brief spikes along the temporal record, suggesting the two glucocorticoids react differently to stressors. Clark et al. (2016) investigated humpback whale hormones and stable isotope values throughout two feeding seasons to examine pregnancy rates and the impact of pregnancy on stable isotope values and found that pregnant females showed different isotope values than males or non-pregnant females. This potentially reflects changes in tissue synthesis, increased use of lipid stores, and reduced excretion of nitrogenous waste, allowing for a proposed model of predictable changes in hormone levels and corresponding stable isotope values over the life of reproductive female humpback whales.

Bird studies have made considerable progress combining hormone and stable isotope analyses, especially to examine the influence of provisioning on reproductive parameters. For example, Barger et al. (2016) found that sympatric species of murres (Uria spp.) changed their foraging behavior by traveling to different areas to forage for alternative prey during the energetically demanding periods of reproduction (incubation and chick-rearing) to possibly reduce competition. Tartu et al. (2014) found that luteinizing hormone may be impacted by mercury load which was in turn influenced by diet and age in snow petrels (Pagodroma nivea). Their findings support previous research that mercury reduce luteinizing hormone, thereby decreasing reproductive fitness, especially in long-lived birds (Tartu et al., 2013; Goutte et al., 2014). Kouwenberg et al. (2013) found that elevated corticosterone promoted foraging during molt in puffins (Fratercula arctica) which led to consumption of higher trophic level diet and increased egg mass during reproduction.

Further research in this area could combine assessments of body condition, stable isotope values, and hormone concentrations in marine megafauna. For example, photogrammetric information on body condition in cetaceans and pinnipeds could be related to reproductive stage, isotopic niche, and concentrations of stress hormones. Availability of food resources for sea turtles could be examined in relation to nesting frequency and reproductive output. Since various reproductive stages are nutritionally demanding, combining stable isotopes with both stress and reproductive hormone analyses may provide greater insight on the extent of fat storage utilization and nutritional condition throughout pregnancy and lactation. These methodological combinations would enable interesting comparisons across income and capital breeders.

#### Life History

Marine megafauna have complex and varied life history patterns accompanied by specific physiological adaptations and behaviors evolved to support these life histories. For example, longdistance annual migrations are common amongst many taxa. Yet, unraveling the specific determinants of these migrations has long been a subject of much research and speculation. Diet, hormones, offspring protection, thermal regulation, and other factors are all potential contributors to the complex process of migration (Brodie, 1975; Corkeron and Connor, 1999; Clapham, 2001), and the combination of stable isotope and hormone analyses has been used to better understand drivers of migration, especially

in birds. For instance, Warne et al. (2015) examined potential determinants of migratory timing in saw-whet owls (Aegolius acadicus). They used stable oxygen (δ <sup>18</sup>O) or δ deuterium (δ <sup>2</sup>H) isotope values to indicate location or arrival of owls on breeding grounds, and they observed that corticosterone was elevated in birds that migrated earlier. Higher corticosterone levels in owl feathers and blood were related to increased foraging and migratory preparedness/body condition which presumably contributed to the early onset of migration in certain owls. Covino et al. (2017) found that in combining analyses of stable hydrogen isotope values, which indicate proximity to breeding grounds, with testosterone concentrations, which correlate with increased breeding preparation, they were able to decipher the timing of breeding preparedness in male songbirds (Mniotilta varia) in relation to their long-distance migrations. During migration, these birds must devote energy to the journey, as well as toward development of breeding characteristics that prepare them for reproduction when they arrive at the breeding ground. The δ <sup>2</sup>H values indicated that male songbird testosterone levels increased as they approached the breeding area, whereas the reproductive schedule for female songbirds did not show such geographically linked timing and requires further exploration.

For many species of migratory marine megafauna, the ability to fast for half the year is routine. Such fasting requires extreme physiological adaptations that are currently poorly understood and that could be greatly informed via combined hormone and stable isotope analyses. Additionally, differences in physiological responses that occur in animals evolved to experience routine (e.g., migratory) fasting versus those forced to endure unexpected and catastrophic fasting (e.g., declining productivity experienced in certain marine systems during climate-induced warming events) could be investigated through these combined methods. For example, elevated δ <sup>15</sup>N values and corresponding high cortisol levels measured in blood from starving animals may indicate more extreme nutritional stress, whereas lower δ <sup>15</sup>N values and higher cortisol levels may indicate normal fasting conditions in a migratory capitol breeder. In the marine environment, stable oxygen (δ <sup>18</sup>O), or δ deuterium (δ <sup>2</sup>H) isotope values can provide additional gradients for tracking movements in the ocean, especially between coastal and offshore habitats, and polar and temperate latitudes (McMahon et al., 2013; MacAvoy et al., 2017). In polar regions, the inclusion of δ <sup>2</sup>H may correlate to sea ice concentration (deHart and Picco, 2015), offering opportunities to examine migration, sea ice conditions, and stress by combining stable isotope and hormone analyses. Finally, it should be mentioned that other technological tools, such as biologging and tagging devices, are natural complements to stable isotope and hormone analyses for the study of migration and life history.

#### Evolutionary and Ecological Drivers

In addition to the above areas of research, the subject of evolutionary versus ecological determinants of population parameters and behavior might also be explored with the combination of stable isotopes and hormone analyses. Yet, to our knowledge, nothing on this subject has been published thus far. We suggest that future studies examine questions that begin to address the topic of plasticity. For example, when resources change, do individuals alter their reproductive or movement behaviors in order to adapt to the new conditions or do they maintain behaviors because they have evolved to do so? When a population is declining or increasing, do they respond differently to changes in their environment? Such questions are often very challenging to address with observational data. Yet the integration of hormone analyses and stable isotope methods, along with other established and emerging population metrics, may enable exploration of reproductive and ecological responses to both external and internal drivers. Improved understanding of individual and population responses to change would be a valuable asset to conservation and management efforts.

#### CONCLUSION AND NEXT STEPS

Combined investigations using stable isotopes and hormones could address questions at a variety of biological levels, progressing from external (e.g., changes in habitat conditions or prey availability) to internal (e.g., physiological) and individual

(e.g., reproduction or migration) to population-level responses (e.g., population abundance) (**Figure 3**). The combination of these analyses in studies of marine megafauna can allow for layering of multiple questions and lines of evidence to inform management decisions or conservation issues. Each of these broad areas of research (**Figure 3**) will require methodological developments as the ecological, evolutionary, and life history investigations evolve.

While there is a suite of methodological developments that would be useful (e.g., further research on storage considerations that enable both types of analyses on the same tissue, etc.), there is one major topic that should be the focus of nearterm efforts to develop this field. More methodological research is needed to enable the successful temporal alignment of these two analytical techniques. For example, while hormones have dedicated metabolic pathways that control their halflife and concentration in each tissue type, stable isotopes are metabolized coincident with the tissue they are in. Thus, these two markers are both subject to, and reflect different physiological processes and time scales which complicate attempts to evaluate them in parallel. Consequently, research is needed on the durations, amplitudes, and incorporation rates of each signal compared to the other, across multiple matrices.

#### REFERENCES


Quantification of the amounts of each incorporated marker, their detectable levels, and the recording rate of each marker will require controlled experiments. Researchers with access to captive animals or large archival collections, such as those in museums and zoological collections, are aptly poised to develop such investigations. These will be pivotal to conservation and management applications of integrated hormone and stable isotope techniques.

#### AUTHOR CONTRIBUTIONS

AF led study design. All authors were involved in manuscript research, writing, figure preparation, and review of final drafts.

#### FUNDING

Funding was provided by the Smithsonian Institution's James Smithson Postdoctoral Fellowship Program, the Smithsonian Institution Women's Committee, the University of California, San Diego, and the Office of Naval Research Grant #N0001417IP00068.


dolphin subpopulation structure around the Iberian Peninsula. Environ. Int. 32, 516–523. doi: 10.1016/j.envint.2005.12.001




diet quality in a marine fish. Limnol. Oceanogr. 60, 1076–1087. doi: 10.1002/ lno.10081




**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 Fleming, Kellar, Allen and Kurle. 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.

# Assessing Seasonality and Density From Passive Acoustic Monitoring of Signals Presumed to be From Pygmy and Dwarf Sperm Whales in the Gulf of Mexico

John A. Hildebrand<sup>1</sup> \*, Kaitlin E. Frasier <sup>1</sup> , Simone Baumann-Pickering<sup>1</sup> , Sean M. Wiggins <sup>1</sup> , Karlina P. Merkens <sup>2</sup> , Lance P. Garrison<sup>3</sup> , Melissa S. Soldevilla<sup>3</sup> and Mark A. McDonald<sup>4</sup>

*<sup>1</sup> Scripps Institution of Oceanography, University of California, San Diego, La Jolla, CA, United States, <sup>2</sup> National Marine Fisheries Service, Pacific Island Fisheries Science Center, Honolulu, HI, United States, <sup>3</sup> National Marine Fisheries Service, Southeast Fisheries Science Center, Miami, FL, United States, <sup>4</sup> WhaleAcoustics, Bellvue, CO, United States*

#### *Edited by:*

*Lars Bejder, University of Hawaii at Manoa, United States*

#### *Reviewed by:*

*Leigh Gabriela Torres, Oregon State University, United States Robert McCauley, Curtin University, Australia*

> *\*Correspondence: John A. Hildebrand jhildebrand@ucsd.edu*

#### *Specialty section:*

*This article was submitted to Marine Megafauna, a section of the journal Frontiers in Marine Science*

*Received: 14 January 2018 Accepted: 06 February 2019 Published: 27 February 2019*

#### *Citation:*

*Hildebrand JA, Frasier KE, Baumann-Pickering S, Wiggins SM, Merkens KP, Garrison LP, Soldevilla MS and McDonald MA (2019) Assessing Seasonality and Density From Passive Acoustic Monitoring of Signals Presumed to be From Pygmy and Dwarf Sperm Whales in the Gulf of Mexico. Front. Mar. Sci. 6:66. doi: 10.3389/fmars.2019.00066* Pygmy sperm whales (*Kogia breviceps*) and dwarf sperm whales (*Kogia sima*) are deep diving cetaceans that commonly strand along the coast of the southeast US, but that are difficult to study visually at sea because of their elusive behavior. Conventional visual surveys are thought to significantly underestimate the presence of *Kogia* and they have proven difficult to approach for tracking and tagging. An approach is presented for density estimation of signals presumed to be from *Kogia* spp. based on passive acoustic monitoring data collected at sites in the Gulf of Mexico (GOM) from the period following the Deepwater Horizon oil spill (2010-2013). Both species of *Kogia* are known to inhabit the GOM, although it is not possible to acoustically separate the two based on available knowledge of their echolocation clicks. An increasing interannual density trend is suggested for animals near the primary zone of impact of the oil spill, and to the southeast of the spill. Densities were estimated based on both counting individual echolocation clicks and counting the presence of groups of animals during one-min time windows. Densities derived from acoustic monitoring at three sites are all substantially higher (4–16 animals/1000 km<sup>2</sup> ) than those that have been derived for *Kogia* from line transect visual surveys in the same region (0.5 animals/1000 km<sup>2</sup> ). The most likely explanation for the observed discrepancy is that the visual surveys are underestimating *Kogia* spp. density, due to the assumption of perfect detectability on the survey trackline. We present an alternative approach for density estimation, one that derives echolocation and behavioral parameters based on comparison of modeled and observed sound received levels at sites of varying depth.

Keywords: passive acoustic monitoring, density estimation, pygmy sprm whale, dwarf sperm whale, Gulf of Mexico

#### INTRODUCTION

Pygmy sperm whales (Kogia breviceps) and dwarf sperm whales (Kogia sima) are deep diving cetaceans that are widely distributed in tropical and temperate waters worldwide (Jefferson et al., 2015). They are typically encountered along the continental slope and in the abyssal plain (Baird et al., 1996; Baird, 2005). Both species are difficult to observe, being entirely pelagic, with faint blows and showing only a low profile while at the water's surface (Jefferson et al., 2015). They are not easily approached using a small boat and have thus far eluded tagging attempts in the field (Baird, 2015). A recent study using passive acoustic monitoring (Hodge et al., 2018) found that Kogia may be more common than suggested by the visual survey record alone.

The pygmy sperm whale is the larger of the two species with a maximum length of 3.5 m and weighing up to 410 kg, while the dwarf sperm whale has a maximum length of 2.7 m and weight of up to 272 kg (Mcalpine, 2018). They live in groups of less than 10 individuals with varying age and sex composition. Group size and their inshore-offshore presence may vary seasonally as documented in the Bahamas (Dunphy-Daly et al., 2008). They have relatively short lives with a maximum known longevity of 23 years (Willis and Baird, 1998; Jefferson et al., 2015). Their primary prey is cephalopods (particularly Histioteuthidae and Cranchiidae), but stomach contents also have shown consumption of fish and crustaceans (West et al., 2009; Mcalpine, 2014). Based on stomach contents and isotope analysis, the two Kogia species may feed at different depths and on slightly different prey (Barros et al., 1998; Willis and Baird, 1998). Kogia may feed both in the water column and at or near the bottom. Their sightings are most frequently reported in water depths between 400 - 1000 m, although they are also seen in deeper waters (Baumgartner et al., 2001).

Much of what we know about Kogia has been inferred from stranding records (Willis and Baird, 1998; Wursig et al., 2000). Strandings of these two species are relatively common in the southeastern United States. They were reported to be the second most common cetacean (after bottlenose dolphins) to strand from North Carolina to Texas between 1978 and 1987 (Odell, 1991), with a total of 189 animals. In the Gulf of Mexico (GOM) there appears to be no seasonal pattern for strandings (Caldwell et al., 1960; Delgado-Estrella and Vasquez, 1998). The relatively high rate of Kogia spp. strandings suggests they may have a higher population than indicated by the relatively few that are observed during visual surveys (Garrison et al., 2010). A combined (K. breviceps and K. sima) abundance of 186 (CV = 1.04) animals is reported within the entire US Gulf of Mexico (GOM) exclusive economic zone (Waring et al., 2013).

On two occasions, pygmy sperm whales in captivity have been shown to produce high-frequency, narrow-band clicks with peak frequencies around 125–130 kHz (Marten, 2000; Ridgway and Carder, 2001; Madsen et al., 2005). Based on field recordings, dwarf sperm whales are known to produce similarly high-frequency clicks (Merkens et al., 2018). The highfrequency echolocation signals of pygmy and dwarf sperm whales are similar to those of phocoenids, cephalorhynchids and two lagenorhynchid species (Au, 1993; Bassett et al., 2009; Kyhn et al., 2009, 2010); however, none of the latter species are known to occur in the GOM (Wursig et al., 2000).

Passive acoustic monitoring in the northern GOM, conducted in response to the Deepwater Horizon oil spill (Merkens, 2013), yielded ample detections of high frequency echolocation clicks that are most likely produced by one or both of the Kogia species; hereafter the term Kogia is meant to imply "most probably" a Kogia species. Here we report on analysis of these data to provide constraints on acoustic signal production and diving behavior. Building on these we derive population density estimates for Kogia spp. at three sites in the GOM, based on the best available information derived from analogy with beaked whales. The acoustic methods we present here provide a new tool for monitoring Kogia spp. and other cryptic populations, a critical aspect of making management recommendations for their conservation.

### METHODS

### Data Collection

The data presented here were collected from five deepwater locations in the northern and eastern GOM (**Figure 1**), an area which is an important habitat for a diverse and abundant group of cetaceans (Davis et al., 2002). The circulation of the northeastern GOM is dominated by the Loop Current, an area of warm water that enters the GOM from the Caribbean and exits through the Florida strait. The oceanographic dynamics of the GOM also include a large freshwater inflow from the Mississippi and other rivers, along with their associated nutrients and sediment loads. The input of nutrients from the Mississippi River creates high phytoplankton productivity and subsequently high zooplankton productivity. Ecosystem dynamics for the deepwater GOM are poorly understood, although it is clear that sperm whales, Kogia and other deepwater cetaceans are important upper trophic level predators.

Acoustic data were collected in the GOM using multiple deployments of High-frequency Acoustic Recording Packages (HARPs) (Wiggins and Hildebrand, 2007) during and following the 2010 Deepwater Horizon oil spill. HARPs are bottommounted acoustic recorders capable of recording continuously at high sample rates (up to 320 kHz) for extended periods. For the GOM deployments, the HARP instrumentation package was located at or near the seafloor with the hydrophone sensor tethered to the instrument and buoyed approximately 10 m above the seafloor. All acoustic data were converted to sound pressure levels based on hydrophone and electronic system calibrations. The hydrophones were composed of two stages, one for low-frequency (<2 kHz) and the other for high-frequency (>2 kHz), although we focus on only the high-frequency band for this paper. The high-frequency stage uses a spherical omnidirectional transducer (ITC-1042, www.itc-transducers.com) which has an approximately flat (±2 dB) sensitivity response of about −200 dBrms re 1V/µPa from 1 Hz to 100 kHz. Each individual hydrophone has a frequency dependent sensitivity supplied by the manufacturer. The signals from the hydrophone transducer are fed into a preamplifier with approximately 50 dB of gain and a 10-pole low-pass filter to reduce highfrequency aliasing effects above 100 kHz and digitized with 16 bits of resolution at 200 kHz sample rate. The response of each preamplifier and filter were measured and these were combined with the hydrophone sensitivity to create a system transfer function unique to each instrument deployment. An alternative configuration capable of higher sampling rates used a spherical transducer (HS-150, www.humbertek.co.uk) with peak response

FIGURE 1 | Long-term recording sites in the Gulf of Mexico with *Kogia* spp. click detections (Mississippi Canyon - MC 28-50.8N 88-27.9W 980 m, Green Canyon - GC 27-33.4N 91-10.0W 1100 m, and Dry Tortugas - DT 25-31.9N 84-38.2W 1,300 m) and those without (Main Pass–MP and DeSoto Canyon–DC). The Deepwater Horizon site is designated with a red star, and *Kogia* spp. visual sightings are given by black asterisks (following Waring et al., 2013).

of about −198 dBrms re 1 V/µPa at 150 kHz, sampled at 320 kHz, with a low-pass filter above 160 kHz.

Long-term deployment data were recorded continuously at 200 kHz for 2-9-month durations during 2010–2013 for the five sites shown in **Figure 1**. Three of the sites (Mississippi Canyon - MC, Green Canyon - GC, and Dry Tortugas - DT) were located in deepwater (at 980, 1,100, and 1,300 m respectively) and had detections for Kogia spp., which are known to be present at deepwater locations throughout the northern GOM (**Figure 1**). Two sites (Main Pass - MP and DeSoto Canyon - DC) located on the continental shelf (at 86, and 268 m respectively) had no detections for Kogia spp, and were not included in subsequent analysis. Details of each HARP deployment are presented in **Supplementary Table 1**. In addition, a deployment sampling at 320 kHz was conducted at the MC site beginning September 20, 2011 for 41 h duration. The latter obtained recordings at 160 kHz bandwidth, sufficient to fully characterize high-frequency echolocation clicks, for comparison with the lower-bandwidth 100 kHz data collected during the remainder of the deployments.

### Signal Description, Detection and Classification

To characterize high-frequency echolocation clicks from the 160 kHz bandwidth data collected in the GOM, signal processing was performed using the MATLAB (Mathworks) based custom software program Triton (Wiggins and Hildebrand, 2007) and other MATLAB custom routines. Long-term spectral averages (LTSAs) were calculated for visual analysis of the recordings, and each instance of energy in the 120–150 kHz band was investigated to find acoustic encounters, periods with continuity of clicking. Individual echolocation signals within these selected encounters were automatically detected using a two-step approach computer algorithm (Roch et al., 2011). The individual click detections were digitally filtered with a 4-pole elliptical band-pass filter with a pass-band between 80 and 140 kHz. Filtering was done on 160 sample points centered on the echolocation signal. Spectra of each detected signal were calculated using Hanningwindowed data centered on the signal. The frequency-related signal parameters peak frequency, center frequency, and −3 dB bandwidth were processed using methods from Au (1993). Click duration was derived from the Teager-Kaiser energy detector output (Roch et al., 2011).

At 100 kHz bandwidth, the HARPs were unable to capture the full frequency range of the Kogia spp. clicks, but the portion of the click energy below 100 kHz was recorded and a small fraction of the energy above 100 kHz was aliased into the passband and thus was recorded as well (as described later in this paper). A multi-step process was used to detect individual Kogia sp. echolocation clicks in these data as well as to identify time windows (of one-min duration) that contained at least one click. Acoustic encounters of Kogia sp. were first identified in the 100 kHz acoustic data using a Teager-Kaiser energy click detector (Roch et al., 2011) and an expert system (based on selecting clicks with peak frequency >70 kHz). All presumed Kogia spp. acoustic encounters were reviewed in a second analysis stage with improved click detection, to remove false detections, and apply a consistent detection threshold. Individual echolocation signals were automatically detected, this time using an energy threshold method during time periods of verified Kogia spp. acoustic encounters. Detections were selected for inclusion when the signal in a 70–99 kHz band exceeded a threshold of 116 dBpp re: 1 µPa. The acoustic encounters were then manually reviewed using comparative panels showing long-term spectral average, received level, and ICI of individual clicks over time, as well as spectral and waveform plots of selected individual signals. Within each encounter, false detections were removed by manual editing, for instance, when the detections were identified as being from sonars, sperm whales or delphinids, identified by having inappropriate spectral amplitude, ICI, or waveform. The entire dataset was examined in this way twice, with the second pass serving to remove false clicks during times when both Kogia sp. and another echolocating species were present. The process was terminated after the second iteration at which time all encounters had been manually verified and the false detection rate for both encounters and one-min time-bins was determined to be less than 1%. We further examined 2,000 randomly selected clicks and found an average false positive detection rate for individual clicks of 9.6% (CV = 0.11). The most common false positive signals identified as Kogia spp. clicks were delphinid and sperm whale clicks with energy above 70 kHz.

The next step was to determine the number of detections per unit effort. A one-min time window was selected for analysis since this is less than the duration of most Kogia spp. encounters. We examined both the number of detected clicks in each onemin time-bin, and the number of one-min time-bins with at least one click. We aggregated detection counts and one-min bin counts into weekly periods. A 1-week period was chosen to provide a sufficient number of one-min time-bins (10,080) for density estimation.

### Diel Cycle

To test for the presence of a diel echolocation pattern, all Kogia spp. click detections were grouped into encounters when individual clicks were less than 10 min apart, defining a start and end time of encounters. These encounters were then assigned to either photoperiod "day" or "night" by extracting sunset and sunrise information from NASA JPL's Horizons Web Service (Giorgini et al., 1996) through the mediator provided in the Tethys metadata service (Roch et al., 2016). The duration of each encounter was calculated in minutes and all durations were summed over each photoperiod. The sum of encounters was normalized by the duration of each photoperiod. A Kruskal-Wallis test was conducted per recording site to test for differences in echolocation behavior based on photoperiod at each site.

### Group Size

Estimates of Kogia spp. group size were derived from acoustic encounters based on overlapping click sequences with consistent ICIs, and compared to visual survey data (Barlow et al., 1997; Baird, 2005; Dunphy-Daly et al., 2008). We selected encounters with high received amplitude (at least one click > 135 dBpp re: 1 µPa), suggesting that the animals were located near to the acoustic sensor. Then we estimated the number of echolocating animals in the group by counting the number of overlaying sequences in the time series, looking for amplitude changes and consistent ICIs (**Supplementary Figure 1**). The basic assumption of this approach is that all animals in a group vocalize and are detected simultaneously at least at some point, so the number of overlapping sequences are an estimate of group size. In addition, it is assumed that each animal within the group, over a short time period, will produce echolocation clicks at a consistent ICI, and that there is a relative consistency of amplitude from one click to the next, given that several clicks are produced per second and the distance and orientation of the animal will not change substantially from one click to the next. This approach will underestimate the number of animals if the spacing between animals in the group is greater than the detection range for their signals, or if the animals do not have periods of simultaneous echolocation. These acoustic group size estimates were compared to visual estimates of group size derived from repeated sighting surveys in the GOM.

#### Detection Probability

Knowledge of the detection probability as a function of horizontal range is needed to estimate the area that is being effectively monitored, which in turn enables density to be estimated. We used a Monte Carlo simulation approach to estimate the detection probability (Küsel et al., 2011), both for single echolocation clicks and for groups of echolocating animals as described in Frasier et al. (2016). This approach is based on modeling the echolocation and orientation behavior of the animals (**Figure 2**). For the models, echolocation is presumed to occur only during a portion of the foraging dive track, during the descent phase and at the dive depth during the foraging phase, based on known behavior of other deep diving cetaceans (Watwood et al., 2006). The descent occurs at a characteristic angle, and at the foraging dive depth the animal may change orientation in elevation angle. Likewise, the animal has a beam pattern that is given by the directivity, side-lobe (90◦ ) source level and back-lobe (180◦ ) level. For a group of animals, larger numbers of animals, and greater differences in their body orientation (elevation and azimuth) increase the detection probability during a given time window. It is assumed that these parameters have a mean and standard deviation that is expressed over many dive cycles. The acoustic receiver was placed at 10 m above the seafloor depth (Wiggins et al., 2012), which varied by ∼300 m between the three GOM sites. The modeling accounted for the volume of water above the receiver by considering both the depth of the animal and the depth of the receiver, and thereby allowed normalization of the density estimate to the maximum area sampled by the receiver. The simulations were conducted at 90-96 kHz since these frequencies best represent the bandwidth of the echolocation clicks collected at 200 kHz sample rate.

Two simulations were used to predict echolocation click detection probability. The first calculated the detection probability for a single echolocation click as a function of range from the acoustic sensor. Transmission loss (TL) was simulated using the ray tracing algorithm Bellhop (Porter, 2011), with inputs including bathymetry, sediment composition, sound speed profile, and mean surface roughness from the Oceanographic and Atmospheric Master Library (OAML). For each site, propagation modeling was conducted in a twodimensional plane (range vs. depth) in four directions (0◦ , 90◦ , 180◦ , 270◦ ). The TL for each simulated click was obtained as an interpolation between these profiles and applied to the peak-peak amplitude. The impact of sound absorption (Francois and Garrison, 1982) on click amplitude was investigated using the approach of Von Benda-Beckmann et al. (2018). It was determined that averaging four frequencies between 90 and 96 kHz for propagation modeling gave the best approximation

FIGURE 2 | Illustration of geometry for model parameters used to simulate the probability of click and group detection with range. The simulation distributes animal locations and orientations using these parameters, and then tests for echolocation click detection at the sensor. Dive track (dotted line) with portion containing echolocation (solid line) during descent phase and at the dive depth during foraging phase. Beam pattern is given by the directivity, side-lobe (90◦ ) source level and back-lobe (180◦ ) level. For a group of animals, the larger number of animals and differences in their body orientation (elevation and azimuth) increase the detection probability during a fixed time window (one-min). A minimum beam amplitude is designated for the group when the sensor is outside the range of orientations where an on-axis click would be received.

for the sound attenuation over the bandwidth of the clicks (assuming an average detection range of ∼700 m). A test of clicks from site MC suggested that their peak-to-peak and RMS signal levels are highly correlated (**Supplementary Figure 2**), justifying application of the calculated TL to estimate peak-to-peak signal level.

A simulation model run involved placement of 100,000 echolocating animals within a 1 km horizontal radius of the acoustic sensor, with depth, orientation and sound production parameters randomly selected for a mean (with uniform distribution over a fixed interval) and a randomly selected standard deviation (uniform over a fixed interval). The simulation was repeated for each receiver, using the sensor depth and bathymetry unique to each site. The click was designated as being detected if its received level equaled or exceeded 132 dBpp re: 1 µPa (this is 16 dB above the detection threshold used in the signal analysis as a means to compensate for the lower bandwidth of the long-term recordings–the rationale for this choice will be discussed in detail later in this paper). No attempt was made to account for changes in noise background since, at these frequencies (80–100 kHz), ambient noise is limited by thermal noise near the sensor, rather than changes in environmental noise (Hildebrand, 2009). A total of 2000 model iterations were run, and the mean probability for click detection was derived from the mean of these, as was the variance.

For estimating the probability of detecting echolocation from a group of animals, the combined orientation of all the animals in the group was allowed to vary during a one-min time-bin. During the descent portion of the dive, a rotation in group elevation angle was allowed. During the foraging portion of the dive, rotation in both elevation angle and azimuth was allowed. The additional freedom of orientation made it more likely that an on-axis click would be received from the group during each one-min time-bin than would be expected for single clicks from individual animals. This accounts for the dynamic search behavior that odotocetes are known to execute during foraging dives, often changing the direction of their echolocation (Teloni et al., 2008). We further assumed that the highest amplitude click was produced at the center location of the group. If the sensor was outside the range of angles subsumed by the group orientation, then a lesser beam amplitude was applied (intermediate between the minimum beam amplitude and the source level). The spread of group members relative to the center location of the group was assumed to be small compared to the maximum range for detection. Similar to the models for individual click detection ranges, 2000 iterations were implemented with 100,000 simulated groups per iteration for each site. The data were averaged in 100 m range bins, and estimates of the group detection probability and variance were calculated.

To obtain the parameters for both the click and the group models, a grid search for model parameters was conducted to minimize the misfit of the model output and data for click received levels at each site. The percentage of clicks detected versus received level was compared for the model and for the recorded data at the three study sites. The goodness-of-fit metric was the sum of the squared misfit for all received level bins (116 dBpp ≤ RL ≥ 140 dBpp) for all sites. This approach assumes that Kogia spp. diving and echolocation parameters are consistent between all three sites. Alternatively, a set of parameters could be independently estimated for each site, although it was judged to be more effective in this case to require consistency across all three sites. A total of 3000 model runs were conducted (1000 each for 3 sites), for both the click and group model, to arrive at the final set of model parameters and detection probabilities (**Supplementary Table 2**). The final parameters were selected to both provide a match to the received level distribution and to give detection probabilities that yielded consistent density values for the click and group methods. A sensitivity analysis was conducted on the final model parameters to determine which had the greatest impact on the final click or group detection probabilities.

#### Vocal Activity

Estimates of vocal activity are needed to estimate density: the click-based method requires an estimate of mean click production rate (r), while the group-based method requires the proportion of one-min bins a group is vocally active (Pv). Ideally, these data would be obtained from the animals being studied; however, due to a lack of auxiliary data for Kogia spp., the click rates used here were derived from a combination of beaked whale acoustic tag data (Johnson and Tyack, 2003), and from the ICIs recorded in this study.

There are four deep diving cetacean species that have sufficient acoustic tag data to estimate the percentage of their dive cycle spent echolocating. They are: Blainville's beaked whale (17%), Cuvier's beaked whale (26%), Risso's dolphin (51%) and sperm whale (67%) (Watwood et al., 2006; Arranz et al., 2016; Warren et al., 2017). It appears that beaked whales are prone to echolocation a smaller percentage of the time than either sperm whales or Risso's dolphin. Likewise, Risso's Dolphins have somewhat larger group sizes (3–10 animals) and are less stealthy than Kogia.

Based on these options for comparison we find Blainville's beaked whale (Mesoplodon densirostris, Md) as the best analog for Kogia given the similarity of their moderate dive depths, small group sizes and stealthy behavior. The proportion of time Kogia spent clicking was estimated from Blainville's beaked whale tag data collected in the Bahamas (Warren et al., 2017) since their respective lengths (4.7 m Md and 3.8 m Kb) and weights (∼1000 kg Md and ∼450 kg Kb) are more similar than for other beaked whale species, and they are presumed to have similar diving behavior. The mean vocal (echolocating) proportion of the Md dive cycle Pcyc was used as a proxy for the proportion of vocally active one-min bins used in groupbased density estimates. The ICI was estimated from the passive acoustic monitoring data in this study, using the mode of the distribution for each site (**Supplementary Figure 3**). The mean proportion of time spent clicking (Pcyc) was divided by the modal ICI to estimate mean click production rate used in click-based density estimates.

To estimate the probability of vocal activity from a group of animals, the synchrony of their echolocation clicking is an important parameter (Hildebrand et al., 2015). The probability of a group being vocally active, Pv, in any given period increases with group size if asynchrony is present as follows:

$$P\nu = P\text{cyc } \left( s - (s - 1) \* o \right) \tag{1}$$

where Pcyc is the proportion of time spent clicking by an individual animal, s is the group size, and o is the pairwise overlap between echolocation bouts of two animals. This expression assumes that each animal added to the group adds both overlapped (simultaneous) and non-overlapped echolocation time to the bout and is appropriate for moderate (∼ <10) group size. For larger group sizes, even a small amount of asynchrony (o > 95%) results in unity for the vocal activity P<sup>v</sup> (the presence of continuous clicking).

Group clicking synchrony (as measured by overlap o) was estimated from the timing of click bouts obtained from acoustic tracking arrays applied to Cuvier's beaked whales (Wiggins et al., 2012; Gassmann et al., 2015). These estimates of synchrony are available for no other cetaceans, including Kogia spp., although the similarity of their group size (∼1–4 animals) and presumed diving behavior (Scott et al., 2001) suggests that the application may be appropriate.

#### Density Estimation

At the finest temporal scale, we determined the presence of Kogia spp. clicks and the number of detected clicks during each onemin time period, and then averaged over a weekly time interval. We estimated animal density using distance sampling-based methods with both click detection (cue-based) and time-bin detection (group-based) approaches (Hildebrand et al., 2015).

A cue-based approach for density estimation requires counting the number of detected clicks, along with knowledge of the click production rate for individual animals and the detectability of individual clicks (Marques et al., 2009). Given nkt detected clicks at site k during week t, in a time period Tkt, density Dkt can be estimated by:

$$
\hat{D}\_{kt} = \frac{n\_{kt} \ (1 - \hat{c}\_k)}{\pi \,\omega^2 \hat{P}\_k \ T\_{kt} \ r} \tag{2}
$$

where P<sup>k</sup> is the probability of detecting a vocal cue that is produced within the radius w from the site, beyond which no detections are assumed to be possible, c<sup>k</sup> is the proportion of false detections, and r is the cue production rate. The variance was obtained using the delta method approximation (Marques et al., 2009).

A group counting approach for density estimation requires detection within a set of short time windows, along with knowledge of group detectability. It further relies on knowledge of both the mean group size s and group vocalization behavior. Using a group counting approach, the estimated density Dkt at site k, during week t is:

$$
\hat{D}\_{kt} = \frac{n\_{kt}}{\pi} \begin{pmatrix} 1 - \hat{\mathfrak{c}}\_k \end{pmatrix} \begin{array}{l} \hat{\mathfrak{s}} \\ T\_{\nu} \end{array} \tag{3}
$$

where nkt represents the number of time intervals (one-min windows) that groups were detected at site k during week t, and Tkt represents the total number of time intervals that were sampled. Detecting a vocalizing group within a horizontal radius of size w has probability P<sup>k</sup> , and the probability of a group being vocally active in a one-min window is Pv, with the proportion of false detections c<sup>k</sup> . At least 3 days of data were required to produce a weekly estimate at the beginning or end of a deployment, otherwise data were associated with the adjacent weekly estimate. Variance is obtained using the delta method approximation, as above.

#### Seasonality and Trend

Kogia spp. density was also calculated on a monthly basis to investigate the potential for seasonal and long-term trends. To avoid confounding between seasonality and trend estimation, a parametric model was used to estimate seasonal and nonseasonal trends in the monthly time series. The raw monthly time series data were first fit with a linear regression based on the Theil-Sen estimator (Sen, 1968) and this trend was subtracted from the original data. The de-trended data were then regressed against a set of monthly indicators. These seasonal indicators were subtracted from the original time series and the final trend was estimated using a least-square linear regression, including estimates for the 95% confidence intervals of the trend.

FIGURE 3 | Click characteristics from *Kogia* spp. encounters recorded with 160 kHz bandwidth at the MC site in the GOM. (A) peak-peak amplitude, (B) peak frequency, (C) −3 dB bandwidth, (D) click duration, (E) RMS amplitude, (F) click center frequency, (G) −10 dB bandwidth, and (H) inter-click interval. Indicated numbers are for all recorded clicks collected during four encounters.

### RESULTS

#### *Kogia* spp. Acoustics

The signals from four Kogia sp. acoustic encounters captured during broadband recording (160 kHz) at site MC (**Figure 3** and **Table 1**) are helpful for understanding how Kogia spp. signals might appear in recordings limited to 100 kHz bandwidth. The broadband recorded signals have energy that extends below 100 kHz (**Figure 4**), and in addition there is a small amount of aliasing of energy from above 100 kHz into the lower frequency band (**Supplementary Figure 4**), based on the instrumental frequency response. Inspection of **Figure 4** suggests that the primary reason for signal detection in the 100 kHz bandwidth recordings is that at least a portion of the signals have bandwidth that extends below 100 kHz. Recording at 100 kHz results in a difference of about 16 dB between the peak-to-peak received signal level as recorded at full (160 kHz) bandwidth and at the reduced (100 kHz) bandwidth (**Figure 5**). Despite the reduced bandwidth and lower received levels, the 100 kHz recordings captured both typical echolocation clicks (used in this study), and the rapid "buzz" clicks that have been associated with foraging attempts (**Supplementary Figure 5**). When modeling the propagation of echolocation clicks we used an average of 90–96 kHz since this best represents the bandwidth of the recorded click, but we compensated for the measured difference of 16 dB between the peak energy TABLE 1 | Comparison of median peak frequency, pulse duration, Inter-click-interval (ICI) and −3 dB bandwidth for *Kogia* spp. from site MC 320 kHz sample rate recording and the same parameters for other reported encounters with *Kogia* including a captive *K. breviceps* and wild *K. sima* in the Bahamas and Guam.


\**Madsen et al. (2005)* #*Merkens et al. (2018).*

that would be available at 117 kHz and what was measured at 100 kHz.

The characteristics of clicks from dwarf sperm whales recorded in the wild near The Bahamas and Guam (Merkens et al., 2018), and a captive pygmy sperm whale that stranded from the Western Atlantic (Madsen et al., 2005), are reported in **Table 1**. The peak frequencies in the MC recordings are slightly lower (117 kHz), and the pulse durations are shorter (62 µs) than in the reported Kogia recordings. The higher peak frequencies from the captive pygmy sperm whale data and the

aliasing of high frequency energy (filtered waveforms are delayed by 125 µs for clarity). The instrumental spectrum noise-floor shown (black line) associated with light blue encounter.

wild but near-surface dwarf sperm whale field recordings could be a function of the close range for recording relative to the MC recordings. Another difference may be that the captive and wild recordings were from animals within a few meters of the surface while the animals recorded at site MC are thought to be at more substantial depth (∼550 m as discussed later). The MC site ICI (81 ms) is similar to both the captive pygmy sperm whale (40–70 ms), and the wild dwarf sperm whale recorded in Guam (90 ms), but it is not possible to assign the MC acoustic encounters to a particular Kogia species, based on the available data.

#### Diel Pattern

The diel occurrence of echolocation click encounters was tested for sites MC, GC and DT individually (**Supplementary Figure 6**). The sums of the duration of encounters per photoperiod for each day were normalized by the duration of each photoperiod resulting in an hourly encounter rate in minutes per photoperiod. If Kogia spp. were detected at any of the three sites during day or night, the median hourly encounter duration was 3 min (0.03 and 10.9 for 10th and 90th percentile). This hourly encounter duration did not show significant differences between photoperiods at all sites (Kruskal-Wallis test). However, when testing the hourly encounter duration for all days of the recording period, including those with no detections, there were significant differences due to a larger number of days than nights with encounters at sites DT and GC, but not at site MC (**Table 2**). The pattern is particularly apparent for site DT where there were almost twice as many days than nights with Kogia spp. encounters.

#### Group Size

A minimum group size distribution was estimated from the acoustic data by examining overlapping sequences of echolocation clicks that occur simultaneously. The acoustic group size estimate (mean = 1.31, median = 1, N = 77) is slightly lower than the visual group size estimate (mean = 1.58, median = 1, N = 52) from surveys conducted in the GOM between 2003 and 2014 (**Figure 6**).

For comparison, previous studies have estimated group size for Kogia spp. based on visual surveys in other locations. In Hawaii Kogia spp. have a mean group size of 2.33 (Baird, 2005). Dwarf sperm whales in the Gulf of California had a mean group size of 2.5 ± 2.3 (Barlow et al., 1997). In The Bahamas the annual median group size was 3.5 (Dunphy-Daly et al., 2008). In addition, the Bahamian dwarf sperm whale group size varied seasonally with smaller groups in the summer (median = 2.5, SD = 1, range = 1–8, N =

TABLE 2 | Comparison of total number of encounters per site (MC, GC, and DT) and number of days or nights with encounters.


*Kruskal-Wallis test results (*χ <sup>2</sup> = *chi-squared, df* = *degrees of freedom, p* = *probability value) for diel patterns per site comparing normalized sum of encounters per photoperiod.*

FIGURE 6 | Acoustic group size for *Kogia* spp. from site MC (A), compared to group size from visual surveys in the GOM (B) based on data collected between 2003 and 2014 as part of SEFSC-NMFS sighting surveys (unpublished data). Mean of acoustic group size = 1.31, median = 1, *N* = 77. Mean of visual group size = 1.58, median = 1, *N* = 52.

34) and larger groups in the winter (median = 4, SD = 2, range = 1–12, N = 20).

The visual group size estimates may be more accurate than the acoustic estimates owing to the clustering of animals at the surface, in contrast, the acoustic detection range is short (<1 km), and the sensors may not simultaneously detect all the animals in a group while they are submerged and echolocating. Both the visual and acoustic group size estimates may miss animals, and therefore should be considered minimum estimates. Although the visual and acoustic group sizes are comparable, we will use the GOM visual group size estimate (mean = 1.58, CV = 0.09) for further calculations given that it is the larger of the two and potentially less prone to missing animals.

#### Detection Probability

A search for Kogia spp. behavioral parameters using the Monte Carlo simulation model fitting resulted in the values given in **Table 3**. The modeled acoustic received level was compared to the observed at each of the three sites, and these distributions are shown in **Figure 7**. The number of clicks at high received levels

TABLE 3 | Monte Carlo simulation parameters used to model probability of detecting individual clicks and groups of clicking animals, along with their modeled sensitivity.


*For each parameter (column 2), the mean (column 3) and standard deviation (column 4) were drawn from a random uniform distribution between the listed ranges associated with that parameter. The sensitivity of the detection probability for clicks (column 5) and groups (column 6) was tested by making small changes in that parameter while holding the other parameters fixed.*

(>135 dBpp re: 1 µPa) was somewhat above the model for MC (**Figure 7A**), but was a good fit to the model at GC (**Figure 7C**) and at DT (**Figure 7E**). A greater diving depth and/or higher source level would improve the fit at MC, but would reduce the goodness of fit at GC and DT. A foraging maximum dive depth of 550 ± 50 m provided the best overall fit, in concert with a source level of 212 ± 5 dBpp re: 1 µPa. Dive descent angle in the click model (67◦ ± 2 ◦ ), and elevation rotation during foraging in the group model (67◦ ± 5 ◦ ) were well constrained. In the click model directivity was constrained to be 23 ± 1 dB, and orientation elevation was allowed to deviate from the horizontal by between 30◦ and 35◦ . Little or no constraint was imposed on the group model parameters elevation rotation while diving (selected to be 40◦ ± 10◦ ) and azimuthal rotation while foraging (selected to be 180◦ ± 10◦ ), and these parameters appear to have little impact on the detection probability. A single set of model parameters was selected to provide the best fit for all three sites, but clearly an improved fit would be obtained from allowing the model to vary for each site.

The Monte Carlo model suggests that the detection probability for Kogia spp. individual clicks (**Figure 8**) is small (∼0.5% within 1,000 m) owing to their directionality, their relatively shallow foraging depth (∼550 m) with respect to the sensor depth (∼1,000 m), and their reduced apparent source level due to the limited bandwidth of the recordings. The probability of click detection is greatest in the region directly above the sensor, and rapidly falls off with range (**Figure 8A**). This is due to both on-axis and off-axis clicks being detected at close range, whereas, only on-axis clicks are detected at greater ranges. The detection probability for a group of animals during a one-min bin is substantially greater (∼40% within 1,000 m). By allowing elevation rotation of ±67◦ for the entire group while foraging, it is much more likely that an on-axis click will be detected. There is a peak in group-detectability at ranges of

300–500 m horizontally from the sensor (**Figure 8B**). At this range expected changes in elevation-orientation during foraging make the detection of on-axis clicks highly likely, whereas when the animal is directly above the sensor, it is less likely to point directly at the sensor and results in fewer on-axis clicks at short horizontal range.

A sensitivity analysis with respect to the model parameters was conducted for detection probability, and the results are presented in Table 3 (two right-hand columns). In general, the probability of click detection is more sensitive to the model parameters than the probability of group detection. This is particularly evident for the source level estimate (212 ± 5 dBpp) which has a 22%/dB change in the click detection probability, but a 9%/dB change in group detection probability. A change in detection probability with source level is expected (necessary) given that higher source levels will make both individual clicks, and clicking from groups, fall above the detection threshold more often. The greater sensitivity of click detection to source level is a product of the overall lower probability of click detection (∼0.5% within 1,000 m) compared to group detection (∼40% within 1,000 m), so that even a slight increase in click detection probability becomes a larger percentage increase for the former than for the latter. Maximum dive depth is another parameter with substantial impact on the detection probability with 17%/50 m change for click counting and 5%/50 m change for group counting. All the other model parameters have only a moderate or little impact on the detection probability, except for the beam directivity which has an 18%/dB change on the probability of click detection.

#### Vocal Activity

The regularity of click timing during echolocation results in a strongly peaked distribution of ICIs for Kogia spp. (**Supplementary Figure 3**) of ∼79 ms (∼13 clicks/s). A secondary peak in the ICI distribution was due to irregular click production, and that some clicks in a sequence would fall below the threshold of the detector and be missed. The ICI's varied somewhat between recording locations, with modal values at site MC being slightly longer (mode = 80.6 ms, N = 35052) than those at site DT (mode = 79.5 ms, N = 4945) and at site GC (mode = 78.1 ms, N = 21077). Owing to variations in the numbers of clicks at each site, we will use a between-site mean ICI of 79.4 ms (CV = 0.02) in the remainder of this analysis. The proportion of time spent clicking during a dive cycle was previously determined for Blainville's beaked whales (as a proxy for Kogia spp.) based on acoustic tag data collected in The Bahamas (Warren et al., 2017). For Blainville's beaked whale, the mean proportion of each dive cycle that contained clicking was 0.165 with CV = 0.075. Estimated click rates (r) were obtained from the product of the mean proportion of the dive cycle spent clicking, and the inverse of the ICI (for Kogia spp.) as follows:

$$r = \frac{0.165}{0.0794} \text{ clicks/sec} = 2.08 \text{ clicks/sec} \qquad \text{CV} = 0.08 \quad \text{(4)}$$

For the group counting method, the vocal synchrony is also needed, and the only species for which this has been estimated is Cuvier's beaked whales (Ziphious cavirostris) (0.67, CV = 0.03) for groups of 2 and 3 animals (Hildebrand et al., 2015). This yields an estimate for the probability of group clicking (Pv) as follows:

Pv = 0.165 ( 1.58 − 0.58 ∗ 0.67) = 0.197 CV = 0.121 (5)

Under this scenario a group of one or two Kogia sp. would be vocally active about 20% of the time across each dive cycle.

#### Density Estimation

An average density for Kogia spp. at each site was estimated for the entire time period using the parameters outlined above. Somewhat higher average densities were observed at the northwestern sites (MC and GC) than for the southeastern site (DT) (**Tables 4**, **5**). Time series of weekly Kogia spp. density estimates, for the period from May 2010 to September 2013, are presented in **Figure 9**. Kogia spp. were present periodically throughout the monitoring period, although the detections fluctuated daily (**Supplementary Figure 7**), presumably as groups of animals moved in and out of the detection range of each instrument.



*Parameters used for density estimation include the average number of clicks per second nkt/Tkt*, *the percentage of false clicks c<sup>k</sup> with associated CV, the expected click rate r with associated CV, the maximum horizontal detection range w, and the probability of click detection P<sup>k</sup> with associated CV.*

TABLE 5 | Average *Kogia* spp. densities derived from group counting by site (MC, GC, DT) given in # of animals per 1000 km<sup>2</sup> .


*Parameters used for density estimation include the average number of one-minute bins with detected groups nkt/Tkt*, *the percentage of one-minute bins with false detections c<sup>k</sup> with associated CV, the expected group size S with associated CV, the probability of group vocal activity P<sup>v</sup> with associated CV, the maximum horizontal detection range w, and the probability of group detection P<sup>k</sup> with associated CV.*

#### Seasonality and Trend

Seasonal trends present in the Kogia spp. density data differ by site (**Figure 10**). Site MC in the northern GOM has higher densities in the spring and summer (May–August) and a deficit in the fall and winter. Whereas, site DT in the southeastern GOM has higher densities in the fall-winter (August–December), although, it has more limited seasonal data available due to gaps in its time series. The magnitude and clarity of the seasonal patterns are greatest at sites MC and DT, while seasonality at site GC is more complex.

The Kogia spp. density time-series were tested for longterm trends, after application of monthly seasonal adjustment (**Table 6**). At all sites, the least-squares annual density change estimate was slightly-to-strongly positive between 2010 and 2013. At site DT, both the click and group density trends had 95% confidence interval limits that did not include zero. Likewise, the site MC click density estimate yielded an annual increase within the 95% confidence interval. All other estimates were consistent with no annual density change based on an annual change of zero being included within the 95% confidence intervals.

#### DISCUSSION

Kogia produce echolocation clicks that are intermediate in peak frequency between those of porpoise such as Harbor porpoise (137 kHz) and beaked whales such as Cuvier's (44 kHz). The fact that Kogia echolocation clicks have some energy below 100 kHz allowed them to be recorded by this study, albeit at reduced amplitude. A primary source of complexity for the current longterm acoustic dataset is the limited bandwidth (100 kHz) of the recordings relative to the peak energy of the GOM Kogia spp. echolocation signals (117 kHz). Analysis of the broadband (160 kHz) dataset suggests a 16 dB loss of signal detectability at 100 kHz relative to broader-band recordings. Future work using higher bandwidth recording (>150 kHz) would eliminate the need to adjust for the limited bandwidth, and may lead to the ability to separate signals between the two Kogia species. It is possible that the results presented here are a conservative estimate of Kogia population density because of the signal loss due to recording at 100 kHz which may not have been completed compensated by the 16 dB adjustment in signal level. Indeed, some of the signals detected by the 160 kHz bandwidth recordings (**Figure 4**) would not have been detected at 100 kHz because of their low signal level below 100 kHz.

The broader bandwidth of our GOM recordings made at depth (**Table 1**) relative to those recorded in the presence of these animals in the wild, presumably closer to the surface, requires more investigation. Pressure effects may be an important factor in the sense that a much smaller volume of air would be available at depth to aid in sound pulse formation and directionality. Perhaps the lower frequency energy of the GOM recordings is related to the Kogia sound production anatomy as it functions at depth, with the lower-frequency component being less directional compared to the narrowband signals observed near the surface. An alternative possibility is that there may be undescribed differences between the two Kogia species, given that the wild recordings are both from K. sima, while the GOM recordings have uncertain species association.

A diel pattern with greater daytime echolocation, and therefore presumably increased daytime foraging, is found for the data at sites GC and DT, but not at site MC. This suggests that Kogia spp. may be opportunistic foragers that target both some species that undergo diel vertical migrations, and some species that do not. It has been previously suggested that Kogia

spp. target a wide diversity of prey and engage in foraging in waters spanning the mesopelagic and bathypelagic zones (West et al., 2009). Since sites GC and DT are deeper than site MC, it is possible that the observed diel pattern reflects deeper foraging during daylight hours, increasing the detectability of their echolocation at the deeper sites during times when their prey may be deeper.

Understanding the probability for acoustic detection of these animals as a function of their range is key to estimating their density. We have used a simulation approach to modeling detection probability for single animals and their clicks, and for groups of animals based on the most detectable click within a

monthly adjustments at sites MC (A), GC (B), and DT (C). Number of months with data (gray dots) for each site.

time window. As a response to the lack of behavioral and sound production information on Kogia spp. we incorporated a grid search model-fitting approach to estimate dive depth, descent angle and other parameters. Having sensors at several depths resulted in different distributions of received level at the different sites, and model parameters were adjusted to be consistent with these distributions. The resulting behavioral parameters appear not to contradict what little is known of Kogia spp. behavior. For instance, a foraging depth of ∼550 m is consistent with stomach contents from stranded animals that suggest oceanic cephalopods are their primary prey (Santos et al., 2006). Likewise, release of a stranded and tagged juvenile pygmy sperm whale (Scott et al., 2001) revealed foraging along the continental slope with moderate dive intervals (∼2–8 min) but little time spent at the surface (9–23%), although longer dive intervals were observed by Barlow et al. (1997). Testing of the full range of behavior parameters derived from the simulation will await collection of more comprehensive tag data at some future date. For robustness, we have chosen to collectively model the behavioral parameters of all three sites. However, it may be that Kogia spp. behavior varies by site and the modeling could be made site specific.

The only reported echolocation source level for Kogia spp. is 175 dB from a captive animal (Madsen et al., 2005) but this value is presumably lower than that of a free-ranging animal. Use of low source level clicks by captive animals may be due


TABLE 6 | Annual trends in *Kogia* spp. density by site and method (column 2 = click method and column 5 = group method).

*Minimum and maximum values are 95% confidence intervals for annual change.*

to the reverberant holding tank and the short ranges to targets within the tank (Au, 1993). Until there are measurements for Kogia echolocation in the wild, we are left with comparing our modeled source level estimate (212 ± 5 dB pp re 1 uPa @ 1 m) with what is known from other odontocetes. The echolocation energy flux density, a product of source level and signal duration, has been shown to scale with body mass for other odontocetes (Jensen et al., 2018). The Kogia energy flux density derived from our modeling and observational data (SL = 203 ± rms re: 1 uPa @ 1 m, duration = 62 µs) compares well to those of odontocetes with similar body mass (shaded area in **Supplementary Figure 8**) suggesting that the output of our modeling is consistent with what would be expected from scaling of known species echolocation.

The proportion of time spent clicking by Kogia was estimated using Blainville's beaked whale tag data obtained in The Bahamas, and group clicking synchrony from Cuvier's beaked whale acoustic tracking data obtained off southern California (Hildebrand et al., 2015). These data were combined with Kogia spp. ICI rates obtained at sites in the GOM to estimate the overall clicking rate. In the future, data on the proportion of time clicking and clicking synchrony should be collected from Kogia spp. to test the validity of the values applied here. If the measured Kogia spp. clicking percentage is found to differ from what was used here, the proportional change can be applied as a linear correction to the density estimates presented.

Stock assessments for the GOM combine the two species of Kogia spp. for abundance estimation owing to the difficulty of differentiating them visually at sea. The most recent abundance estimate for northern GOM Kogia spp. is 186 animals (CV = 1.04) (Waring et al., 2013). This estimate is from the summer of 2009 and covers waters from the 200 m isobaths to the seaward extent of the US Exclusive Economic Zone, an area of approximately 4 × 10<sup>5</sup> km<sup>2</sup> . This gives an average density of 0.46 animals/1000 km<sup>2</sup> for the northern GOM, an order of magnitude lower than the acoustic density estimates presented in **Tables 4**, **5**. One reason for this difference may be that the continental slope sites monitored here acoustically are particularly favorable habitat for Kogia spp. However, sighting distributions (**Figure 1**) do not reveal a preference for slope habitat, indeed the majority of sightings are in non-slope deep waters. The most likely explanation for the mismatch between visual and acoustic densities is the highly likely violation of the g(0) = 1 assumption (perfect detection of animals on the trackline) incorporated into visual survey estimates (Mullin and Fulling, 2004). For such elusive animals that spend a large proportion of their time at depth, the true g(0) must be considerably lower than 1, as suggested by modeling of their dive intervals and detectability changes with sighting conditions (Barlow, 1999, 2015). Together these imply that current stock estimates are potentially underestimated, and that evaluating the assumptions involved in constructing these acoustic density estimates should be a research priority.

Cetacean populations in the GOM may have been impacted by the Deepwater Horizon oil spill, which was underway at the time these recordings were initiated in May 2010, with oil release terminated by late summer 2010. An increasing population trend during 2010–2013 for Kogia spp. at sites DT and MC is suggested by the click density estimation method, and to a lesser extent by the best estimate of the group density estimation method. The annual rates of increase, for instance at site MC, of 23% (click method) or 13% (group method) are both greater than what would be expected to be the maximum rates of increase (∼6%) for a population of odontocetes (Reilly and Barlow, 1986), suggesting that the observed increases are most likely a result of animals moving into proximity of the site, perhaps following larger temporal scale oceanographic cycles and/or habitat recovery in the period after the oil spill. Indeed, site MC densities appear particularly low in the summer and fall of 2010 relative to other years of monitoring when densities peaked in the summer months (**Figure 9**) suggesting the possibility of avoidance of the area during the time of the oil spill, perhaps related to both the presence of oil and to the presence of a large number of vessels in the area as part of the spill response.

#### CONCLUSIONS

Kogia spp. are among the most difficult cetaceans to study in the wild, and repeated attempts to attach tags to them have been frustrated by their avoidance of boats (Baird, 2015). For this reason, they remain poorly studied species, despite the recent technical advances that have allowed detailed study of beaked whales and other deep diving cetaceans. Passive acoustic monitoring provides a window, albeit incomplete, into their behavior and presence.

In the context of an extremely difficult to study species, the population density estimates made here for Kogia are based on the best available proxies for their echolocation and diving behavior. The accuracy of these estimates would be improved with ample acoustic tag data from Kogia, but thus far this has not been possible. Use of passive acoustic data for applying constraints on Kogia echolocation and dive behavior may be the best option we have to date. The approach presented here compares a model of Kogia behavior to the acoustic data through use of the acoustic received level distributions, a step forward over what has been done for previous acoustic density estimation studies (Hildebrand et al., 2015). Using this and further approaches modeling passive acoustic data, it may be possible to study the diving and echolocation behavior of the full range of beaked whale species (Baumann-Pickering et al., 2013), the majority of which have not yet been studied by acoustic tagging.

Passive acoustic data were analyzed to estimate the density of Kogia spp. at three sites in the GOM during and following the Deepwater Horizon oil spill. The population densities obtained from acoustic monitoring at these three sites are all substantially higher than those derived from line transect visual surveys. An increasing interannual density trend is suggested for animals near site MC, within the primary zone of impact of the oil spill, and for site DT, to the southeast of the spill.

Potential bias in the reported density estimates relate to assumptions involved in obtaining detection probability, group size and vocalization rates. A simulation approach was used to obtain the detection probability, although the parameters used in the simulation were constrained by fitting observational data on click received levels. A full optimization for the fit between the parameters used in the model and the observational data is beyond the scope of this study, but it should be a goal for future work. In particular, understanding of how the population density estimate, which is directly related to click or group detection probability, would change with source level, dive depth, beam pattern and other model parameters, should be explored. The model-based approach provides an alternative means for estimating Kogia spp. diving and echolocation behavior and at present may be the only means to obtaining such data given the difficulty of tag attachment on this species. Better species-specific information on group size and vocalization rates is critical for improving the reliability of these density estimates. As better understanding of Kogia spp. echolocation and diving behavior becomes available in the future, it should be possible to revise these density estimates to incorporate improved estimates of the aforementioned parameters.

#### ETHICS STATEMENT

Because the study involved passive acoustic monitoring, no sounds or other disturbances were produced requiring permits or approval from the institutional animal use and care committee.

#### REFERENCES


#### AUTHOR CONTRIBUTIONS

JH and LG funding acquisition. JH, MM, and KM conceptualization. SW instrumentation. SW and KF data curation. JH, SB-P, KF, KM, and MM methodology. JH, KF, MM, and KM data analysis. LG and MS ancillary data. KF, SB-P, SW, and JH software. MM and JH writing–first draft. KF, SB-P, SW, KM, MM, MS, and LG writing–review and editing. JH writing–revisions.

#### FUNDING

Funding for HARP data collection and analysis was provided by the Natural Resource Damage Assessment partners (20105138), the US Marine Mammal Commission (20104755/E4061753), the Southeast Fisheries Science Center under the Cooperative Institute for Marine Ecosystems and Climate (NA10OAR4320156) with support through Interagency Agreement #M11PG00041 between the Bureau of Offshore Energy Management, Environmental Studies Program and the National Marine Fisheries Service, Southeast Fisheries Science Center, and the CIMAGE Consortium of the Gulf of Mexico Research Initiative (SA 12-10/GoMRI-007).

#### ACKNOWLEDGMENTS

We thank S. Murawski and S. Gilbert of USF, and K. Mullin of the SEFSC for project assistance. We thank LUMCON and the crew of the R/V Pelican, as well as I. Kerr and the crew of the R/V Odyssey for assistance with HARP deployments. We thank members of the SIO Whale Acoustic Laboratory including: T. Christianson, C. Garsha, B. Hurley, J. Hurwitz, J. Jones, E. O'Neill, E. Roth, B. Thayre, J. Jordan, and B. Kennedy for assistance with HARP operations and data processing. Tiago Marques, Danielle Harris, Len Thomas and Frants Jensen of St. Andrews University, as well as three reviewers, provided helpful comments on this manuscript. The GOM Kogia data used for this study are archived at https://data.gulfresearchinitiative. org/data/R4.x267.180:0011/ maintained by the Gulf of Mexico Research initiative.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fmars. 2019.00066/full#supplementary-material


information on stomach contents and strandings. Mar. Mamm. Sci. 22, 600–616. doi: 10.1111/j.1748-7692.2006.00038.x


**Conflict of Interest Statement:** MM was employed by WhaleAcoustics LLC.

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 Hildebrand, Frasier, Baumann-Pickering, Wiggins, Merkens, Garrison, Soldevilla and McDonald. 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.