Seabird Trophic Position Across ThreeOcean Regions Tracks EcosystemDifferences

 Tyler Gagne1*, K. D. Hyrenbach2, Molly Hagemann3, Oron Bass4, Stuart Pimm5, Mark

MacDonalrd6, Brian Peck7, Kyle Van Houtan1

 

1Conservation & Research, Monterey Bay Aquarium, United States, 2Hawaii Pacific University, United

States, 3Bernice P. Bishop Museum, United States, 4National Park Service, United States, 5Nicholas

School of the Environment, Duke University, United States, 6Department of Marine and Wildlife

Resources, United States, 7Rose Atoll National Monument, United States Fish and Wildlife Service(USFWS), United States

  Submitted to Journal:

  Frontiers in Marine Science

  Specialty Section:

  Global Change and the Future Ocean

  Article type:

  Brief Research Report Article

  Manuscript ID:

  397546

  Received on:

  15 May 2018

  Revised on:

  15 Aug 2018

  Frontiers website link:  www.frontiersin.org

In review

1

Seabird Trophic Position Across Three Ocean Regions Tracks Ecosystem Differences 1

2

AUTHORS: Tyler O. Gagné,1* K. David Hyrenbach,2 Molly E. Hagemann,3 Oron L. Bass,4 Stuart L. 3Pimm,5 Mark MacDonald,6 Brian Peck,7 Kyle S. Van Houtan1,5 4

AFFILIATIONS: 1 Monterey Bay Aquarium, Monterey, CA 93940 USA; 2 Hawaii Pacific University, 5Kaneohe, HI 96744, USA; 3 Vertebrate Zoology Collections, Bernice Pauahi Bishop Museum, Honolulu, 6HI 96817; 4 South Florida Natural Resources Center, Everglades National Park, Homestead, FL 33030, 7USA; 5 Nicholas School of the Environment, Duke University, Durham, NC 27708 USA; 6 Department of 8Marine and Wildlife Resources, Pago Pago, American Samoa, USA; 7 Rose Atoll Marine National 9Monument, US Fish and Wildlife Service, Pago Pago, American Samoa, USA. 10

*Corresponding author and lead contact: tgagne@mbayaq.org, Tel: +1 (774) 328-1734. 11

MAIN TEXT WORD COUNT: 2068 12

No. FIGURES: 1 13

No. REFERENCES: 26 14

Running Title: seabird trophic positions track ecosystem changes 15

16

17

18

19

20

21

22

23

24

25

26

27

ABSTRACT 28

We analyze recently-collected feather tissues from two species of seabirds, the sooty tern (Onychoprion 29fuscatus) and brown noddy (Anous stolidus), in three ocean regions (North Atlantic, North Pacific, South 30Pacific) with different human impacts. The species are similar morphologically and are similar in the 31trophic levels from which they feed within each location. In contrast, we detect reliable differences in 32trophic position amongst the regions. Trophic position appears to decline as the intensity of commercial 33fishing increases, and is at its lowest in the Caribbean. The spatial gradient in trophic position we 34document in these regions exceeds those detected over specimens from the last 130 years in the Hawaiian 35Islands. Modeling suggests that climate velocity and human impacts on fish populations strongly align 36with these differences. 37

KEYWORDS: trophic ecology; commercial fisheries; ocean memory; global change; machine learning; 38stable isotopes; food webs 39

In review

2

INTRODUCTION 40

The scale and accessibility of marine ecosystems, anthropogenic impact, and industry removal presents 41management and conservation challenges. Consequently, monitoring and evaluation often rely on proxies 42to assess marine ecosystems. Such proxies include manually-surveyed biological indicators, remotely-43sensed environmental data, and state-aggregated fisheries dependent metrics like mean trophic level 44(Pauly et al., 1998;Holt and Miller, 2011;Hunsicker et al., 2016). For these to be effective decision-45making tools, identifying indicators and metrics that accurately and independently quantify impacts is key 46(Link et al., 2009). 47

Food web status has been a widely-used metric for gauging ecosystem state, especially in relation 48to commercial fishery impacts. In particular, fishery-derived mean trophic level is repeatedly used to 49gauge food web status and trends; though its interpretation has been contested (Pimm, 1982;Pauly et al., 501998). Fishery-dependent mean trophic level assessments have inherent biases that may fail to detect 51multiple ecosystem signals such as high trophic-level species loss, prey release, and market targeting 52shifts (Essington et al., 2006;Estes et al., 2011). Ideal biological indicators are those sampled and 53approximated from species representing broad ecosystem patterns (Lyday et al., 2015;Reed et al., 2016). 54The techniques must be robust and reproducible at scale (Gagné et al., 2018). Increasingly, metrics 55independent of fisheries and often involving non-target upper trophic level predators, like marine birds 56and mammals, offer promising approaches for circumventing biases (Holt and Miller, 2011). 57

To assess the status of pelagic food webs, we use compound-specific stable isotope analysis of 58amino acids (CSIA-AA) to compare seabird tissues across three ocean regions. This approach helps 59resolve previous questions of bias from fishery-data based metrics (Pimm, 1982;Pauly et al., 1998) and 60from those relying on bulk stable isotope techniques which have come under scrutiny due to isotopic 61baseline shifts affecting trophic position estimates (Nielsen et al., 2015). 62

We focus on two species of colonial terns that are central place foragers, have large foraging 63ranges (BRNO = 200 km, SOTE = 800 km), and are circumtropically distributed (Hebshi et al., 642008). CSIA-AA determines trophic position robustly by comparing the relative enrichment of the 15N to 6514N ratio (δ15N) in trophic and source amino acids (Nielsen et al., 2015). This approach proposes a 66fishery-independent metric to document food web status amongst locations. It builds upon our previous 67research, that applied this method across a chronology of 130-years (Gagné et al., 2018) within one 68region. To understand potential spatial patterns that the birds may be detecting, we use random forest 69regression models to investigate the influence of various anthropogenic and climate factors on seabird 70trophic position. We reaffirm seabirds are reliable indicators of marine systems, especially through the 71use of CSIA-AA to calculate their trophic position (Lyday et al., 2015), and in recording impacts from 72commercial fisheries extraction. 73

74

METHODS 75

Specimen collection and preparation 76

We sampled brown noddy (Anous stolidus) and sooty tern (Onychoprion fuscatus) feathers from three 77locations in three distinct ocean basins: Rose Island, Rose Atoll National Wildlife Refuge, American 78Samoa (South Pacific); Waimanalo, Oahu, Hawaii (North Pacific); and Bush Key, Dry Tortugas National 79Park, Florida (North Atlantic). From 2013-2015, we collected senesced, fully-emerged flight feathers at 80nesting colonies and additional samples from dead strandings (USFWS permits MB052060-0, 81MB180283-1). 82

Collected feathers were free from debris and other tissues, were stored in heavyweight 83polyethylene bags (ULINETM, 4 mil) with indicating silica gel desiccant (FisherTM grade 48, 4–10 mesh), 84and later debrided with compressed air. We homogenized individual feathers (n = 20) and sent samples to 85the UC Davis Stable Isotope Facility for CSIA-AA. Though our processed specimen data are limited (n = 86

In review

3

20, 6 Hawaii, 4 Florida, 10 American Samoa with an even split of species by location), the sample size 87we collected is characteristic of CSIA-AA studies of marine species (Popp et al., 2007;Chikaraishi et al., 882009;Votier et al., 2010;O’Malley et al., 2012;Ostrom et al., 2017;Peavey et al., 2017) and effective at 89describing possible population patterns. That said, we discuss future research directions that may 90improve upon the sampling and analyses presented here. 91

Trophic position calculation 92

We calculated trophic position, TP, using: 93

𝑇𝑃 = (𝛿'(𝑁*+, − 𝛿'(𝑁.+/ − 𝛽)

𝑇𝐸𝐹+ 1 94

where δ15NTrp is the mean value for six trophic amino acids (alanine, glutamic acid, isoleucine, leucine, 95proline, valine), δ15NSrc is the single source amino acid (phenylalanine), and β (2.42) and TEF (trophic 96enrichment factor = 5.63) are amino acid-specific constants (Nielsen et al., 2015), calculated using 97established methods (Gagné et al., 2018). Variation of the selected constants does not impact the relative 98relationships of TP between species or location, only absolute values, and therefore does not influence our 99aim of detecting differences between regions. For each specimen, we generated 1000 random TP values 100from the δ15N (normal) parameter distributions the analytical lab provided for each amino acid (Fig 1A 101and Table S1). Due to the large (20000 estimates) sample drawn the lab parametrized distributions, tests 102of difference are highly significant (p <0.005) with narrow confidence intervals. Therefore, we report and 103discuss trophic positions effect size and magnitude differences (Sullivan and Feinn, 2012;Greenland et al., 1042016). 105

Model input aggregation 106

We aggregated covariates tied to environmental and human drivers that may describe spatial differences. 107As seabird prey items vary between regions, we represented fishing pressure for each region with catch 108density (catch-per-unit area, or CPUA, measured in tonnes km-1) as per previous studies (Pauly, 1092007;Pauly et al., 2013). This metric is calculated by dividing the cumulative total annual catch from the 110Sea Around Us reconstructed landings by the total area of the EEZ of the location from which the feathers 111were sampled. We also obtained mean trophic level (MTL) by EEZ from the SeaAroundUs database. To 112capture the nature of the potentially relevant climate velocities that may have precipitated the relevant 113seabird prey composition, we calculated the 10-year change during and immediately prior the sampling 114period. We calculated this as the rate of change in remotely-sensed sea surface temperature (Figure S2 A-115B, AVHRR POES 0.1°, 14-day composite) during the period of 2006-2016, consistent with previous 116studies (Hamann et al., 2015;Van Houtan et al., 2015). 117

Model development 118

To explore correlative relationships, we implemented the random forest algorithm to model trophic 119position as a function of the drivers. Random forests models excel in their flexibility to model non-120linearity and complex interactions while also maximizing generalizability, minimized overfitting 121tendency, and easy interpretation (Breiman, 2001). We trained random forest models with conservative 122hyper-parametrization. Hyper-parameters were set at 500 trees, 2 variables tried at each split, a minimum 123node size of 5 observations, and an out-of-bag sample proportion of 0.4 with replacement (Breiman, 1242001). We built centered individual conditional expectation plots to highlight the range of predictor-125response relationships conditional on the range of observations of the predictor values (Goldstein, 2015). 126

Sensitivity analysis 127

We conducted a sensitivity analysis on variable importance rankings using a leave-one-out randomization 128technique for modeling fitting to measure sample stability on model inferences (Fig. 1, Archer and Kimes, 1292008). In this process, we built 500 sets of variable importance values, based on 500 models each run 130from random subsets of the full sample set. As opposed to one set of variable importance rankings, this 131

In review

4

established a distribution of variable rankings under both sample perturbations and model fitting 132stochasticity that is inherent to random forest. This analysis provides a means to account for the influence 133of individual specimens given our relatively limited sample size. 134

135RESULTS 136

The sooty tern and brown noddy share a similar ecomorphology (Gagné et al., 2018). Figure 1A shows 137comparable trophic positions within species that decline amongst locations from American Samoa (mean 138= 4.00) to Hawaii (3.68) to Florida (3.59). Additionally, unlike bulk stable isotope analysis, CSIA-AA 139accounts for nitrogen from varied food web bases. Therefore, the observed shifts in trophic position 140suggest divergence in the trophic positions of available prey, and suggests food web compression or 141simplification. 142

To model potential regional patterns, we use random forest regression models to investigate the 143influence of anthropogenic and climate factors on trophic position. We aggregated covariates tied to 144environmental and anthropogenic drivers that describe spatial differences in fisheries-dependent mean 145trophic level, fisheries landings, and coupled sea surface temperature rate of change. Our models suggest 146that commercial fishing and climate align with the food web differences between regions. Figure 1B-H 147shows the model relationships and the estimated variable importance. 148

The random forest model explained 57.5% of the variance in trophic position, with a root mean 149square error of 0.17. The three continuous variable inputs show well-defined trends. Trophic position 150declines with increasing catch (Fig. 1B), increasing rates of regional sea surface temperature change (Fig. 1511D), and as fishery-dependent mean trophic level declines (Fig. 1C). The categorical variables (Fig. 2E-152G, location, basin, and species) all offer strong and comparable splitting power (Fig. 1H), though little 153insight into the changing conditions effecting TP. In the leave-one-out cross validation (Fig. 1H), root 154mean square error on the out-of-test sets averaged 0.18 with a 95% quantile range of 0.098-0.284. This 155range does highlight some sensitivity to stability of the specimen dataset, which may be improved in 156future studies with additional specimens. However, due to field logistics and laboratory costs, studies 157using CSIA-AA for similar purposes often employ a similar number of replicates as we have here (see 158above). Importantly, however, our sensitivity analysis reveals the modeled signal is distinct from the 159sample variance. 160

161

DISCUSSION 162

Seabirds as an alternative trophic indicator 163

Does higher seabird trophic position describe a less-impacted ecosystem state? Birds at the Western 164Atlantic site had a significantly lower trophic position in both species we measured (Fig. 1A). Our model 165indicates lower trophic positions at this site are consistent with chronically high fishing pressure, sea-166surface warming, and low mean trophic level of fisheries catch. Conversely, our South Pacific site had the 167highest measured trophic position, paired with low fishing pressure, a recent cooling trend, and high mean 168trophic level in fisheries catch. Future studies that compare such questions across more extensive 169geographies and perhaps timelines may be able to decouple model factors, and better reveal the individual 170effects of climate and fisheries on mean trophic level. Such an approach may provide testing beyond the 171hypotheses we generated observationally and refine some of the low-resolution signals we detected. 172Nonetheless, in accord with recent reviews (Estes et al., 2011), we have documented the trophic 173downgrading of two seabird species across a large spatial gradient and described possible contributing 174factors. 175

Trophic position dissimilarity in space appears to reflect declines across time 176

In review

5

The regional differences we describe are consistent with a recent 130-year analysis of trophic position in 177seabirds from the Hawaiian Archipelago. In that study (Gagné et al., 2018), the ensemble trophic position 178for eight seabird species declined an average 0.32 units from 1891-2015. (Sooty tern declined by 0.30 and 179brown noddy by 0.09.) As trophic position may be a reliable metric to track ecosystem changes in one 180ecosystem over time, it may also reflect food web status across spatially-distinct ecosystems. Given this, 181the contemporary trophic position decline we observe when comparing Florida to American Samoa (0.41) 182and Hawaii to American Samoa (0.33) is larger than the 130-year decline observed in Hawaii alone 183(Gagné et al., 2018). Compared more directly, however, the spatial gradient in trophic position between 184Florida and American Samoa equates to 161 years of change in Hawaii described by Gagne et al. (2018) 185in their seabird ensemble (-0.258 change in trophic position per century). When we constrain the 186benchmark to just the sooty tern and brown noddy data from the Hawaii study (-0.157 change per 100 yr) 187the metric jumps to 264 years of change. If we make similar comparisons of our spatial gradient between 188Hawaii and Florida, those numbers are 34 and 56 years, respectively. Confirming previous studies (e.g., 189Fitzpatrick and Keegan, 2007;Jackson, 2008;Huettmann, 2012), this suggests historical overfishing and 190other anthropogenic impacts in Caribbean coastal ecosystems are significantly more advanced than in 191other tropical regions. 192

In sum, we reaffirm that seabirds are reliable indicators of marine systems, especially when we 193use of CSIA-AA to calculate their trophic position. Importantly, our findings align with the widely 194reported consensus that the Gulf of Mexico is a highly impacted ecosystem (Halpern et al., 2008). To 195utilize our approach while also improving upon our constrained sample, we suggest continued sampling 196of seabird feathers from museums and nesting colonies with negligible impacts to wild populations 197(strandings, fisheries interactions). Such a monitoring framework may better inform sampling and 198modeling needs for ecosystem-based management and decision making. Furthermore, the development of 199a larger sample with our approach will better elucidate the nature of covariates and their interactions. 200Active movement tracking and spatially-explicit fishery data, combined with the tissue sampling 201approach here, may further inform these patterns (Cherel et al., 2016). Ultimately, it is becoming clearer 202that seabird trophic position varies cross space and time and that it reliably informs food web status, and 203particularly the far-reaching impacts from commercial fisheries extractions. 204

205

ACKNOWLEDGEMENTS. C. Yarnes performed the CSIA-AA analyses. V. Lam provided fishery data 206from the Sea Around Us database. A. Copenhaver and E. Schick provided logistical support. F. Huettman, 207P. Renaud, and several anonymous reviewers improved earlier versions of this manuscript. 208

DATA AVAILABILITY STATEMENT: Datasets analyzed for this study are included in the data 209repository at: osf.io/4s9ty 210

AUTHORS’ CONTRIBUTIONS. KV and SP designed the study. All authors collected and prepared 211data; TG and KV analyzed data, prepared figures and wrote the manuscript; All authors reviewed the 212manuscript. 213

ETHICS STATEMENT. This research was performed under USFWS permits #MB052060-0, 214MB180283-1 to KH. All research followed the Institutional Animal Care and Use Committee criteria 215under NOAA and HPU. The authors declare no conflicts of interest. 216

FUNDING STATEMENT. This work was supported in part by a Presidential Early Career Award for 217Scientists and Engineers (PECASE) to KV. 218

DECLARATION OF INTERESTS. We have no competing interests 219

220

REFERENCES 221

In review

6

Archer,K.J.,andKimes,R.V.(2008).Empiricalcharacterizationofrandomforestvariableimportancemeasures.222ComputationalStatistics&DataAnalysis52,2249-2260.223

Breiman,L.(2001).Randomforests.MachineLearning45,5-32.224Cherel,Y.,Quillfeldt,P.,Delord,K.,andWeimerskirch,H.(2016).Combinationofat-seaactivity,geolocation225

andfeatherstableisotopesdocumentswhereandwhenseabirdsmolt.Front.Ecol.&Evol..4,3.226Chikaraishi,Y.,Ogawa,N.O.,Kashiyama,Y.,Takano,Y.,Suga,H.,Tomitani,A.,Miyashita,H.,Kitazato,H.,and227

Ohkouchi,N.(2009).Determinationofaquaticfood-webstructurebasedoncompound-specificnitrogen228isotopiccompositionofaminoacids.LimnologyandOceanography:methods7,740-750.229

Essington,T.E.,Beaudreau,A.H.,andWiedenmann,J.(2006).Fishingthroughmarinefoodwebs.Proc.Nat.230Acad.Sci.USA103,3171-3175.231

Estes,J.A.,Terborgh,J.,Brashares,J.S.,Power,M.E.,Berger,J.,Bond,W.J.,Carpenter,S.R.,Essington,T.E.,Holt,232R.D.,andJackson,J.B.(2011).TrophicdowngradingofplanetEarth.Science333,301-306.233

Fitzpatrick,S.M.,andKeegan,W.F.(2007).HumanimpactsandadaptationsintheCaribbeanIslands:an234historicalecologyapproach.EarthEnvironSciTransRSocEdinburgh98,29-45.235

Gagné,T.O.,Hyrenbach,K.D.,Hagemann,M.E.,andVanHoutan,K.S.(2018).Thetrophicsignatureofpelagic236ecosystemsinseabirds.Sci.Adv.4,eaao3946.237

Goldstein,A.,Kapelner,A.,Bleich,J.,Pitkin,E.,(2015).PeekingInsidetheBlackBox:VisualizingStatistical238LearningWithPlotsofIndividualConditionalExpectation.J.Comput.Graph.Stat.24,44-65.239

Greenland,S.,Senn,S.J.,Rothman,K.J.,Carlin,J.B.,Poole,C.,Goodman,S.N.,andAltman,D.G.(2016).240Statisticaltests,Pvalues,confidenceintervals,andpower:aguidetomisinterpretations.Europeanjournal241ofepidemiology31,337-350.242

Halpern,B.S.,Walbridge,S.,Selkoe,K.A.,Kappel,C.V.,Micheli,F.,D'agrosa,C.,Bruno,J.F.,Casey,K.S.,Ebert,C.,243andFox,H.E.(2008).Aglobalmapofhumanimpactonmarineecosystems.Science319,948-952.244

Hamann,A.,Roberts,D.R.,Barber,Q.E.,Carroll,C.,andNielsen,S.E.(2015).Velocityofclimatechange245algorithmsforguidingconservationandmanagement.GlobalChangeBiology21,997-1004.246

Hebshi,A.J.,Duffy,D.C.,andHyrenbach,K.D.(2008).Associationsbetweenseabirdsandsubsurfacepredators247aroundOahu,Hawaii.Aquat.Biol.4,89-98.248

Holt,E.A.,andMiller,S.W.(2011).Bioindicators:usingorganismstomeasureenvironmentalimpacts.Nature249Edu.Knowl.3,8.250

Huettmann,F.(2012).FromEuropetoNorthAmericaintotheworldandatmosphere:ashortreviewofglobal251footprintsandtheirimpactsandpredictions.TheEnvironmentalist32,289-295.252

Hunsicker,M.E.,Kappel,C.V.,Selkoe,K.A.,Halpern,B.S.,Scarborough,C.,Mease,L.,andAmrhein,A.(2016).253Characterizingdriver–responserelationshipsinmarinepelagicecosystemsforimprovedocean254management.Ecol.App.26,651-663.255

Jackson,J.B.(2008).Ecologicalextinctionandevolutioninthebravenewocean.ProceedingsoftheNational256AcademyofSciences105,11458-11465.257

Link,J.S.,Yemane,D.,Shannon,L.J.,Coll,M.,Shin,Y.-J.,Hill,L.,andBorges,M.D.F.(2009).Relatingmarine258ecosystemindicatorstofishingandenvironmentaldrivers:anelucidationofcontrastingresponses.ICESJ.259Mar.Sci.67,787-795.260

Lyday,S.E.,Ballance,L.T.,Field,D.B.,andHyrenbach,K.D.(2015).Shearwatersasecosystemindicators:261Towardsfishery-independentmetricsoffishabundanceintheCaliforniaCurrent.J.Mar.Syst.146,109-262120.263

Nielsen,J.M.,Popp,B.N.,andWinder,M.(2015).Meta-analysisofaminoacidstablenitrogenisotoperatiosfor264estimatingtrophicpositioninmarineorganisms.Oecologia178,631-642.265

O’malley,J.M.,Drazen,J.C.,Popp,B.N.,Gier,E.,andToonen,R.J.(2012).Spatialvariabilityingrowthandprey266availabilityoflobstersinthenorthwesternHawaiianIslands.MarineEcologyProgressSeries449,211-220.267

Ostrom,P.H.,Wiley,A.E.,James,H.F.,Rossman,S.,Walker,W.A.,Zipkin,E.F.,andChikaraishi,Y.(2017).Broad-268scaletrophicshiftinthepelagicNorthPacificrevealedbyanoceanicseabird.Proc.R.Soc.B284,26920162436.270

Pauly,D.(2007).TheSeaAroundUsProject:documentingandcommunicatingglobalfisheriesimpactson271marineecosystems.AMBIO36,290-295.272

Pauly,D.,Christensen,V.,Dalsgaard,J.,Froese,R.,andTorres,F.(1998).Fishingdownmarinefoodwebs.273Science279,860-863.274

In review

7

Pauly,D.,Hilborn,R.,andBranch,T.A.(2013).Fisheries:doescatchreflectabundance?Nature494,303-306.275Peavey,L.E.,Popp,B.N.,Pitman,R.L.,Gaines,S.D.,Arthur,K.E.,Kelez,S.,andSeminoff,J.A.(2017).276

Opportunismonthehighseas:foragingecologyofoliveridleyturtlesintheeasternPacificOcean.=ersin277MarineScience4,348.278

Pimm,S.L.(1982).Foodwebs.NewYork:Chapman&Hall.279Popp,B.N.,Graham,B.S.,Olson,R.J.,Hannides,C.C.,Lott,M.J.,López-Ibarra,G.A.,Galván-Magaña,F.,andFry,280

B.(2007).Insightintothetrophicecologyofyellowfintuna,Thunnusalbacares,fromcompound-specific281nitrogenisotopeanalysisofproteinaceousaminoacids.TerrestrialEcology1,173-190.282

Reed,J.,Shannon,L.,Velez,L.,Akoglu,E.,Bundy,A.,Coll,M.,Fu,C.,Fulton,E.A.,Grüss,A.,andHalouani,G.283(2016).Ecosystemindicators—accountingforvariabilityinspecies’trophiclevels.ICESJ.Mar.Sci.Journal284duConseil,fsw150.285

Sullivan,G.M.,andFeinn,R.(2012).Usingeffectsize—orwhythePvalueisnotenough.Journalofgraduate286medicaleducation4,279-282.287

VanHoutan,K.S.,Halley,J.M.,andMarks,W.(2015).Terrestrialbaskingseaturtlesarerespondingtospatio-288temporalseasurfacetemperaturepatterns.BiologyLetters11.289

Votier,S.C.,Bearhop,S.,Witt,M.J.,Inger,R.,Thompson,D.,andNewton,J.(2010).Individualresponsesof290seabirdstocommercialfisheriesrevealedusingGPStracking,stableisotopesandvesselmonitoring291systems.JournalofAppliedEcology47,487-497.292

293

In review

8

294

295

Fig. 1. Trophic position of two seabirds (A) and conditional relationships with model predictors (B-296G). A, whisker range is 1.5x the interquartile spread. Plot colors correspond to labelled wing silhouettes, 297“SOTE” is sooty tern and “BRNO” is brown noddy. B-G are conditional expectation plots of predictors 298from the model for continuous (B-D) and categorical model inputs (E-G). Relative TP on the y-axis of B-299G refers to the change in trophic position units observed across the range of values of a predictor. Colored 300lines (B-D) and center quantile (E-G) represent the respective mean and median of the partial effect of a 301covariate on the model prediction of TP. Gray bands and quartiles represent the range and distribution of 302predictions, akin to a prediction interval. TP declines with increases in catch density (B) and rates of SST 303increase (D), and declines as Sea Around Us reported catch landings MTL decreases (C). We assessed 304model variable importance with a leave one out sensitivity analysis. Panel (H) shows the proportion of 305model runs where each covariate appeared at each importance rank. Beyond the splitting power of species 306and geography, fisheries related factors rank highly. 307

308

In review

9

SUPPLEMENTARY MATERIAL 309

This supplement presents additional information on methods, figures, and tables that provide added data 310

and clarification. Analysis code and data can be found at the public repository: osf.io/4s9ty 311

DOI 10.17605/OSF.IO/4S9TY 312

Table S1. Quantile summary for main text Figure 1A of CSIA-AA estimated trophic position by 313location and species. Raw data from which distributions were drawn is available in the public repository. 314

315

316

317

Figure S1. SST time series (A) and map regions (B) from which the 10-year rate of change was 318approximated with linear model fits of the annual median. Rates of change was calculated with sea 319surface temperature using the slope coefficient of a linear model on AVHRR SST data (POES 0.1°, 14-320day composite, annual median) during the period 2006-2016. 321

In review