will the california current lose its nesting tufted …influencing puffin population dynamics and...

Submitted 29 November 2017Accepted 28 February 2018Published 22 March 2018

Corresponding authorChristopher J Hart chrish32uwedu

Academic editorAgus Santoso

Additional Information andDeclarations can be found onpage 18

DOI 107717peerj4519

Copyright2018 Hart et al

Distributed underCreative Commons CC-BY 40

OPEN ACCESS

Will the California Current lose itsnesting Tufted PuffinsChristopher J Hart1 Ryan P Kelly1 and Scott F Pearson2

1 School of Marine and Environmental Affairs University of Washington Seattle WAUnited States of America

2Washington Department of Fish and Wildlife Olympia WA United States of America

ABSTRACTTufted Puffin (Fratercula cirrhata) populations have experienced dramatic declinessince the mid-19th century along the southern portion of the species range leadingcitizen groups to petition the United States Fish and Wildlife Service (USFWS) to listthe species as endangered in the contiguous US While there remains no consensuson the mechanisms driving these trends population decreases in the CaliforniaCurrent Large Marine Ecosystem suggest climate-related factors and in particular theindirect influence of sea-surface temperature on puffin prey Here we use three speciesdistribution models (SDMs) to evaluate projected shifts in habitat suitable for TuftedPuffin nesting for the year 2050 under two future Intergovernmental Panel on ClimateChange (IPCC) emission scenarios Ensemble model results indicate warming marineand terrestrial temperatures play a key role in the loss of suitable Tufted Puffin nestingconditions in the California Current under both business-as-usual (RCP 85) andmoderated (RCP 45) carbon emission scenarios and in particular that mean summersea-surface temperatures greater than 15 C are likely to make habitat unsuitable forbreeding Under both emission scenarios ensemble model results suggest that morethan 92 of currently suitable nesting habitat in the California Current is likely tobecome unsuitable Moreover the models suggest a net loss of greater than 21 ofsuitable nesting sites throughout the entire North American range of the Tufted Puffinregardless of emission-reduction strategies These model results highlight continuedTufted Puffin declinesmdashparticularly among southern breeding coloniesmdashand indicatea significant risk of near-term extirpation in the California Current Large MarineEcosystem

Subjects Conservation Biology Ecology Marine Biology Climate Change BiologyEnvironmental ImpactsKeywords Endangered species act Conservation biology Climate change Tufted puffinEndangered species management Habitat loss Species distribution model

INTRODUCTIONWorldwide species are facing increasing challenges associated with rising sea- and airsurface temperatures (Thomas et al 2004) Warming climates have resulted in distributionand community-abundance changes in species ranges across multiple taxa (Parmesan ampYohe 2003) changes to ecological responses including phenological anomalies correlatingwith warming seasonal temperatures (Walther et al 2002) and changes in habitat qualityand distribution (Klausmeyer amp Shaw 2009) Foden et al (2013) found that 83 of birds

How to cite this article Hart et al (2018) Will the California Current lose its nesting Tufted Puffins PeerJ 6e4519 DOI107717peerj4519

66 of amphibians and 70 of corals that were identified as highly vulnerable to theimpacts of climate change are not currently considered threatened with extinction on theIUCN Red List of Threatened Species indicating that speciesrsquo vulnerabilities are likely tobe much greater than conservation status alone would suggest

In recent years in the United States the United States Fish andWildlife Service (USFWS)and the National Marine Fisheries Service (NMFS) have received several petitions to listspecies under the Endangered Species Act due to the impacts of climate change (Siegelamp Cummings 2005 Wolf Cummings amp Siegel 2008) However the link between climatechange and risk to a species can be difficult to assess One approach to examining theselinkages is to model the interaction between climate and suitable habitat for a givenspecies given what is already known about the relationship between the species and itshabitat This approach has become an integral component of conservation planning in aworld of changing environments (Hagen amp Hodges 2006 Richardson amp Whittaker 2010)Ultimately understanding these linkages can help inform conservation assessments andspecies and ecosystem management strategies (Carnaval amp Moritz 2008 Ponce-Reyes et al2017) for example by estimating the likelihood of losing (or gaining) particular suitablehabitats of interest under future climate conditions

Tufted puffins as a pertinent exampleThe Tufted Puffin (Fratercula cirrhata) is an iconic species that is experiencing dramaticpopulation declines across the southern portion of its geographic range (Piatt amp Kitaysky2002) While Tufted Puffin populations in the Alaska Current have remained relativelystable (but see Goyert et al 2017) populations in the California Current large marineecosystem (area of the eastern Pacific Ocean spanning nearly 3000 km from southernBritish Columbia Canada to Baja California Mexico) have declined by approximately90 relative to early 20th century estimates and are currently declining 9 annually(Hanson amp Wiles 2015) The number of occupied breeding-colony sites in WashingtonState has declined by 60 relative to the 1886ndash1977 average and 45 relative to the1978ndash1984 average (Hanson amp Wiles 2015) Range contractions at the southern edge ofthe Tufted Puffinrsquos habitat in both the eastern and western Pacific Ocean have led topreliminary conservation measures the Washington Department of Fish and Wildlife haslisted the Tufted Puffin as endangered at the state level in 2015 with Japanrsquos Ministry of theEnvironment listing the species as endangered in 1993 (Osa amp Watanuki 2002 Hanson ampWiles 2015Washington Fish and Wildlife Commission 2015)

Tufted puffin biology and ecologyTufted Puffins are seabirds belonging to the family Alcidae and nest in colonies locatedon both sides of the North Pacific ranging in North America from the Channel Islands insouthern California (34N) to coastal northern Alaska (68N) (Piatt amp Kitaysky 2002) andin Asia from Hokkaido Japan (43N) through the Kamchatka Peninsula (63N) (Hansonamp Wiles 2015) They are central-place foragers during the nesting season when they digburrows or use crevices for nesting on nearshore rocks islands and sea stacks (Piatt ampKitaysky 2002) During the nesting season puffins exhibit large foraging radii around their

Hart et al (2018) PeerJ DOI 107717peerj4519 226

colonies (up to 40 km eg Menza et al 2016 Fig 12) and are able to carry more thantwenty fish at a time while flying back to the colony to feed their chicks (Piatt amp Kitaysky2002 Hanson amp Wiles 2015) While little is known about the wintering distributionand ecology of Tufted Puffins summer (MayndashSeptember) breeding colonies are welldocumented and provide the most useful biological data for conservation management(Piatt amp Kitaysky 2002) Extensive breeding colony surveys dating back to the early 20thcentury allow us to examine any potential link between climate and species range extent

Tufted Puffins are subject to multiple well-documented ecological stressorsmdashsuch asincreasing eagle predation habitat degradation declining prey availability and fishingnet entanglement (Baird 1991 DeGange amp Day 1991 Ricca Keith Miles amp Anthony2008)mdashbut several mechanisms associated with temperature stress may be importantin driving puffin declines along their southern range boundary Gjerdrum et al (2003)found dramatically reduced growth rates and fledging success (development of fledglingwings and muscles for flight) in years with high sea surface temperature (SST) anomaliesOther researchers cite the nutritional demands of puffin chicks and the prey availabilityand preferences correlating with fledgling success (Hipfner Charette amp Blackburn 2007)suggesting a mechanism for the negative effects of high sea surface temperature on puffinchicks These and other studies point to a link between temperature and demographictrends in the Tufted Puffin and help identify this species as a candidate for distributionmodeling Modeling outputs may help expose proposed interactions between high oceantemperatures prey distribution in the water column and puffin breeding success

As a result of these potential threats and documented population declines the USFWSwas petitioned to list the California Current population of the Tufted Puffin (Fraterculacirrhata) as endangered under the Endangered Species Act (ESA) (Sewell 2014) In orderto respond to this petition the USFWS is currently examining Tufted Puffin status andtrends evaluating threats to its survival the adequacy of existing regulatory mechanisms toconserve the species the loss of its habitat and other relevant factors Given that climatemdashspecifically increasing sea-surface temperaturesmdashmay be a particularly important factorinfluencing puffin population dynamics and ultimately reducing puffin breeding rangeand given the vast geographic extent of puffin nesting sites (34 of latitude and roughly70 of longitude in North America) and historical data on the occupancy of these sites theTufted Puffin is an excellent candidate for species distribution modeling

Species distribution models in conservation planningSpecies distribution models (SDMs) are a powerful way to examine how climate variablesrelate to species geographic distribution and the distribution of suitable habitat (Guisanamp Zimmermann 2000 Guisan amp Thuiller 2005) By associating species occurrence withclimate variables these models can (1) test for associations in space and time betweenputative environmental drivers and changes in species range and (2) project changes insuitable habitat under future climate change scenarios (Bellard et al 2012) SDMs use avariety of underlying statistical models to capture the relationship between habitat andclimate and create detailed outputs highly useful for wildlife management (Carvalho et al2011Guisan et al 2013) Recent Endangered Species Act listing decisions andmanagement


plans have drawn on SDM results to provide critical spatial and temporal conservationinformation For example climate envelope models were used to develop spatially explicitconservation strategies that account for climate change notably in the case of the NorthAmerica Wolverine (Gulo gulo) where the models were the basis of an ESA listing (UnitedStates Fish and Wildlife Service 2016)

Here we use 50 years of nesting-habitat distribution informationmdashranging fromthe Bering Sea to Californiamdashto map Tufted Puffin nesting habitat We use thiscolony occupancy data to model the relationship between nesting habitat and currentenvironmental conditions to project future suitable breeding sites in the same geographicrange We present these results as an example of how this information can be used in bothregulatory (eg Endangered Species Act) and conservation planning contexts

MATERIALS AND METHODSEnvironmental dataEnvironmental data for the current period which we define here as the years 1950ndash2000was downloaded fromWorldClim a set of global climate layers derived from interpolationof monthly climate observations (Hijmans et al 2005 last accessed January 2017) Afterremoving WorldClim bioclimatic variables displaying high collinearity we also consideredfactors relevant to Tufted Puffin breeding phenology including isolating seasons duringwhich puffins breed (spring and summer) and including environmental variables relevantto their prey species such as temperature These processes resulted in our selection ofsix environmental variables for analysis annual temperature range (ATR) mean diurnaltemperature range (MDR) mean temperature of the warmest quarter (MTWQ) annualprecipitation (AP) precipitation of the warmest quarter (PWQ) and distance-to-ocean(DIST) a variable we created to help models discern suitable nesting habitat as occurringonly in rocky coastal habitats within meters of the sea a biological requirement of puffins(Piatt amp Kitaysky 2002) (see Table S1 for measurements and units) Each variable for thecurrent period was scaled to a 5 arcmin grid cell size (ca 10 km times 10 km) After scaling allenvironmental variables within the relevant geographic range were cropped to only includelocations within 200 kilometers of the ocean While this cropping distance includes landnot physically suitable for obligate coastal and island breeders it is important for creatinglarger environmental variable gradients during model construction (Van Horn 2002)

The same six environmental variables above were averaged over the period of 1910-1950to construct a lsquopastrsquo climate regime used to project past Tufted Puffin range Past climatevariableswere selected using gridded climate data obtained frommonthly observations fromthe Climate Research Unit CRU TS v 401 dataset (Harris et al 2014 (crudataueaacuk)last accessed March 2017) Past environmental data was similarly scaled down to the same5 arcmin grid cell size as the current data

Future climateWe selected emissions scenarios Radiative Concentration Pathways (RCP) 45 and 85as defined by the IPCC 5th Assessment Report (IPCC 2014) as future environmentalprojections against which to forecast changes in Tufted Puffin breeding distribution


Downscaled model output for environmental variables for both future RCP scenarios wereaveraged across the following general circulation models Hadley Centrersquos HadGEM2-AO (Collins et al 2011) NOAArsquos GFDL-CM3 (Griffies et al 2011) NASArsquos GISS-E2-R(Nazarenko et al 2015) Institut Pierre-Simon Laplacersquos IPSL-CM5A-LR (Dufresne et al2013) Beijing Climate Centerrsquos BCC-CSM1-1 (Xin Wu amp Zhang 2013) Bjerknes CentrersquosNorESM1-M (Bentsen et al 2013) National Center for Atmospheric Researchrsquos CCSM4(Gent et al 2011) and the Max Planck Institutersquos MPI-ESM-LR (Giorgetta et al 2013)all for the year 2050 (average of 2041ndash2060) (Hijmans et al 2005 last accessed January2017) Using the average of eight prominent climate model outputs helps incorporatevariance in potential future climate projections within our model The 2050 timeframeand these emissions scenarios (roughly speaking a moderate-reduction scenario andbusiness-as-usual scenario with no emission reductions) were selected as the most relevantto the conservation decisions presently surrounding the Tufted Puffin (IPCC 2014)

Species dataSpecies distribution data were obtained courtesy of USFWS Washington Department ofFish and Wildlife and Environment and Climate Change Canada and were derived fromexpansive US and Canada breeding colony surveys conducted by groups including USFWSWashington Department of Fish and Wildlife Alaska Department of Fish and GameEnvironment and Climate Change Canada California Department of Fish and Wildlifeand others (Speich amp Wahl 1989 Hodum et al see Supplemental Information 4 WorldSeabird Union (httpsseabirdsnet last accessed March 2017) British Columbia MarineConservation Analysis (httpbcmcacadatafeatureseco_birds_tuftedpuffincolonies)2017 last accessed May 2017) see Supplemental Information 5) Count data consistedof estimates of numbers of breeding individuals present at known nesting colonies andthe spatial coordinates of those observations Biological data for the lsquocurrentrsquo periodof climate data (see Table S2) represents the most recent survey observation of knownnesting sites from 1950ndash2009 While the climatological data runs until the year 2000biological data from up to 2009 was included to incorporate recent detailed state-widesurveys in both Oregon and Washington information critical to examining trends acrossthe puffinrsquos southern range We converted count data to presenceabsence values giventhe nature of our analysis which asked whether breeding habitat was likely to be suitable(ge1 nesting birds) or not (0) under future conditions Some observations were adjustedgeographically up to one grid cell (ca 10 km) to fall within gridded terrestrial environmentalvariables Observations further than 15 km from terrestrial grids (eg remote islands) wereremoved from the analysis The environmental variables described above were selectedto model potential interactions between climate conditions and puffin range during thebreeding season

Given the low proportion of absence-to-presence observations for Tufted Puffin surveysand potential bias in survey locations we added pseudo-absence (PA) observations(ie generated absence observations existing within the range of the SDM) to all modelsSDMs using both presence and absence have been shown to perform more accuratelythan models relying on presence-only observations (Elith et al 2006 Barbet-Massin et al


2012) PA generation methodology is also important in both model predictive accuracyand avoiding model over-fitting (Barbet-Massin Thuiller amp Jiguet 2010 Barbet-Massin etal 2012) Adapting these recommendations in Barbet-Massin et al (2012) 1000 PAs wererandomly generated twice across the SDM a minimum of 30 km from any presence or trueabsence point

Model parameterizationBecause Tufted Puffins rely heavily on both terrestrial and marine environments forreproduction we initially tested the correlation between sea-surface temperature andair-temperature data across puffin colony observations SST data for this comparisoncomprised an average of mean monthly temperature for June July and August monthsaligning with Tufted Puffin breeding season obtained from the Hadley Centre UK (Rayneret al 2003 (metofficegovukhadobs) last accessed March 2017) and the corresponding airtemperature readings (MTWQ) (Hijmans et al 2005) Both sets of environmental variableswere scaled to a 5 arcmin grid cell size and represented means from the years 1950ndash2000 Ahigh correlation coefficient (r = 096) allowed us to use air temperaturemdashwhich is availablein higher spatial resolutionmdashrather than SST in the final analysis This strong relationshipbetween air- and sea-surface temperature has also been documented across several othermarine and aquatic species distribution studies (Stefan amp Preudrsquohomme 1993 Domisch etal 2013) Additionally within R software (R Core Team 2013) a principal componentanalysis (PCA) (Pearson 1901) was performed to compare variance in environmentalvariables between areas of collapsed colonies (absence) and occupied colonies during thecurrent period This technique can help identify differences in environmental niches ofspecies occurrence data (eg Broennimann et al 2012 Pentildea Goacutemez et al 2014) here weuse it to create an index of environmental variables to identify likely drivers of TuftedPuffin declines after accounting for the covariances among variables

Species distribution modelingModel algorithmsSDMswere constructed with the R package BIOMOD2 (BIOdiversityMODelling) (Thuilleret al 2009 R Core Team 2013) All SDMs were constructed for a spatial range larger thanthe current estimated US Tufted Puffin distribution (180W to 120W longitude and33N to 69N) Using a larger extent both increases the range of environmental gradientsavailable for model construction and introduces novel climates useful for projectingpotential migration (Thuiller et al 2004 Fitzpatrick amp Hargrove 2009 Domisch et al2013) Models were also examined under a subset of all biological and environmental datafrom 126W to 120W and 32N to 485N This portion of the analysis is intended toaccount for the spatial variance of puffin distribution and examine the temperature-habitatrelationship in the California Current large marine ecosystem exclusivelymdashthe portion ofthe range that has experienced the greatest decline and has been petitioned for listing underthe US Endangered Species Act

To help acknowledge and estimate uncertainty three different models using differentstatistical approaches were selected from the BIOMOD framework generalized linearmodels (GLM) (McCullagh amp Nelder 1989) generalized boosting models (GBM also


referred to as boosted regression trees) (Ridgeway 1999) and random forests (RF)(Breiman 2001) The GLM models used a logit link between the response variable meanand combination of explanatory variables (Guisan Edwards amp Hastie 2002) (ie logisticregression) GBMs incorporate regression and machine-learning techniques throughboosting many decision-tree models to increase model performance (Elith Leathwickamp Hastie 2008) Decision models recursively partition sets of explanatory and outcomevariables in a stagewise manner until subsets of data are explained by trees of bifurcatingdecisions (Elith Leathwick amp Hastie 2008) Boosting then sequentially fits decision treesto training data selecting the trees that best fit the data (Elith Leathwick amp Hastie 2008)Finally RF is a machine-learning technique that creates classification trees similar to thosein GBMs but instead uses random bootstrap samples of data and explanatory variablesupon the construction of each tree (Breiman 2001)

The differences in statistical and machine learning approaches across GLM GBM andRF algorithms provides variance across which to test sensitivity between models as well asestimations of model uncertainty (Marmion et al 2009 Rodriacuteguez-Castantildeeda et al 2012)Additionally usingmodels with relativelymore ensemble (GBMandRF) and parsimonious(GLM) approaches to habitat selection as well as utilizing both parametric (GLM) andnon-parametric (RF) techniques provides robust analysis of environmental drivers of rangechange (Marmion et al 2009) and led to the selection of these three model algorithms

Model calibrationHaving generated two variants of the dataset by generating distinct pseudo-absences wethen constructed twenty models for each algorithm (GLM GBM RF) for each datasetvariant for a total of 120 models All models then used past environmental data aswell as future emission scenarios to project both past and future puffin range changesEach model variant performed a random 7030 split of the biological data using 70for model calibration and 30 for model evaluation This technique addresses spatialautocorrelation and is frequently utilized when faced with dependent biological sampling(surveying of species around only areas of known occurrence) (Arauacutejo et al 2005)Model selection and calibration parameters were kept constant between past and currentmodels to maintain consistency and repeatability For all models across all algorithms thedefault model construction options and parameters of the BIOMOD package were used(Thuiller et al 2009)

HindcastingHindcast models were created to examine the past relationship between temperatureand patterns of puffin colony occupancy Hindcasts can increase confidence in futureprojections and help shed light on ecological interactions over time (Raxworthy et al 2003Labay et al 2011) These lsquohindcastrsquo models use current puffin distribution projections andthe past environmental data detailed above to project past puffin colony distribution Alimited amount of historical survey data along the California Current especially in southernOregon and California makes scoring the hindcast projections against past survey datadifficult for this analysis However examining past model projections at specific known


historical locations (such those in Hanson amp Wiles 2015) in a past cooler climate helpsinterpret the influence of a warming climate in the future (Labay et al 2011)

Ensemble modeling and evaluationThe area under the receiver operating characteristic curve (AUC) and the True SkillStatistic (TSS) were the two model metrics used to evaluate model performance AUCmaps sensitivity rate (true positive) against (1-specificity) values (=false-positive rate) andis a popular metric for species distribution model evaluations because it evaluates acrossall thresholds of probability conversion to binary presence or absence (Fielding amp Bell1997 Guo et al 2015) Higher AUC scores represent better model performance with AUCscores between 07ndash08 classified as lsquofairrsquo 08ndash09 as lsquogoodrsquo and 09ndash10 as lsquoexcellentrsquo (Guoet al 2015) TSS scores display (sensitivity + specificity minus1) with sensitivity quantifyingomission errors and specificity quantifying commission errors (Allouche Tsoar amp Kadmon2006 Guo et al 2015 Shabani Kumar amp Ahmadi 2016) TSS scores of zero or less indicatemodel performance nobetter than randomand scores of 10 indicating perfect performanceBoth scores were emphasized in this analysis to provide strong measures of ordinal modelperformance and to predict accuracy in threshold-dependent conservation planning(Allouche Tsoar amp Kadmon 2006 Shabani Kumar amp Ahmadi 2016)

Ensemble models were created using weighted averages of TSS scores both within andacross algorithms while AUC scores were not used in constructing ensemble models butserved as another evaluation metric This technique captures uncertainty stemming fromrandom sampling of the dataset as well as variance across modeling techniques (Gallardo ampAldridge 2013) thereby providing the user with a robust sense of model fit and sensitivityto particular parameters TSS scores below 07 were excluded from the ensemble to removeinfluence from poor predictive models (Arauacutejo et al 2011) A proportional weight decaywas used averaging model weights resulting in weights proportional to TSS evaluationscores Additionally binary conversions which maximized model TSS performance wereused in some range-change analyses Range-change analyses were performed allowingfuture migration to potential suitable future habitat as well as with no potential migrationEnsemble binary thresholds and their impact on projections are noted below

RESULTSModel performanceModels from all three algorithms and especially the ensemble model scored very highin both model performance metrics (Table 1) GLM GBM and RF algorithms displayedmean TSS scores and standard deviations of 0842 plusmn 0020 0898 plusmn 0017 and 0905plusmn 0017 respectively Similarly GLM GBM and RF mean AUC scores were very highindicating good model accuracy (Table 1) Techniques using machine learning methods(RF and GBM) consistently displayed the highest performance by both AUC and TSSscores (Table 1) perhaps due to these machine-learning (GBM and RF) models relyingon boosting and ensemble learning respectively compared to a single regression modelapproach within GLM algorithms Despite the different statistical and learning approaches


Table 1 Evaluationmetrics and range change analysis for ensemble model and bymodel algorithm(N = 40) (A) Model area under the receiver operating characteristic curve (AUC) and true skill statis-tic (TSS) for ensemble model and by algorithm AUC represents sensitivity rate (true positive) against 1-specificity values (false positive) and TSS represents (sensitivity+ specificityminus1) Scores presented aremean plus or minus standard deviation (B) Percent of projected change in range by model algorithmNorth America-wide and US California Current (32Nndash485N) independent analyses Both RCP 45 (45also italicized) and RCP 85 (85 also bolded) represented Scores presented are mean plus or minus stan-dard deviation

Ensemble GLM GBM RF

(A) Model evaluationTSS 920 842plusmn 020 898plusmn 017 905plusmn 017AUC 994 976plusmn 004 985plusmn 004 986plusmn 004

(B) Range change45 Species-wide minus2180 minus2259plusmn 1169 minus1388plusmn 1172 minus1799plusmn 2033

California current minus9268 minus9669plusmn 1461 minus1695plusmn 2012 minus2126plusmn 261085 Species-wide minus2614 minus3137plusmn 1470 minus1409plusmn 1374 minus1923plusmn 2379

California current minus9756 minus9713plusmn 1296 minus2271plusmn 2027 minus2726plusmn 2680

of the selected modeling approaches TSS and AUC scores were high across all techniquesand displayed low variance (Table 1)

Variable contributionResponse plotsAfter initial variable winnowing both model response plots and PCA analysis indicatethat temperature variables ATR and MTWQ are strongly associated with Tufted Puffinbreeding habitat (Figs 1 2) ImportantlyMTWQdisplayed a thermalmaximumof suitablenesting habitat (ie a threshold) around 15 C and the MTWQ variable also displayed themost consensus across model members among all selected variables (Fig 1) This resulthighlights consensus among GLM ensemble members surrounding the 15 C thresholdATR andMTWQ are related to extreme summer temperatures and consensus among GLMensemble members across these variable response plots is consistent with the hypothesisthat summer temperature anomalies influence Tufted Puffin colonies

Conversely MDR and PWQ were not effective in predicting suitable habitat amongall models In fact the probability of Tufted Puffin occurrence remains high across therange of MDR and PWQ values indicating that these variables are not helpful in predictingpuffin occupancy GLMmodels do show a response to increased annual precipitation (AP)values but there remains a lack of consensus among model members around a particularresponse cutoff

Principal component analysisPrincipal component analysis provided further support to the hypothesis identifyingsummer temperature as a primary driver of variance in Tufted Puffin breeding habitat(Fig 2) PCA components 1 (51) and 2 (27) together explained 78 of the variabilityin the data Component 2 with a strong loading of MTWQ of minus0732 and MDR loadingof minus0497 indicates that MTWQ and MDR explain the difference between presenceand absence points as evidenced by the separation of the 95 confidence ellipse along


Figure 1 General LinearizedModel (GLM)model algorithm variable response plots Responsecurves across GLM algorithms for all environmental variables (A) Annual Temperature Range (ATR)response curves (B) Mean Diurnal Temperature Range (MDR) response curves (C) Mean Temperatureof the Warmest Quarter (MTWQ) response curves (D) Annual Precipitation (AP) response curves (E)Precipitation of the Warmest Quarter (PWQ) response curves (F) Distance from Coast (DIST) responsecurves Each colored line represents one GLMmodel run (N = 40) Y -axis displays predicted probabilityof habitat suitability across each variable given other variables are held fixed at their mean value X-axisdisplays environmental variable values (see Table S1 for specific units) Results display distinct cutoffsbetween ATR MTWQ and occurrence probability

Full-size DOI 107717peerj4519fig-1

this component (Fig 2) The other four variables loaded more strongly onto principalcomponent 1 which does not help separate presence and absence points This resultcombined with the MTWQ response curves (Fig 2) indicate the importance of MTWQ inpredicting what habitat is suitable for Tufted Puffins (Figs 1 and 2)

Range forecastsNorth American projectionsmdash2050After binary transformation of the future probabilistic projection maps ensemble modelsagain representative of the weighted mean of all model algorithms and variants project arange loss of approximately 22of currently occupied range under RCP 45 and a range lossof approximately 26 under RCP 85 by 2050 (Table 1) GLMmodels projected the greatestpercent habitat loss across North America under both emission scenarios (Table 1) Therewas uniform agreement across ensemble algorithms in projecting habitat loss even with thepossibility of colonizing new habitat with variability among algorithms as to themagnitude


Figure 2 Principal component analysis loadings 1 and 2 (95 confidence ellipses) for occupied(present) and unoccupied (absent) nesting colonies


of that loss (Table 1) Spatially losses were uniformly projected along the California Currentup to southeastern Alaska (Fig 3) although ensemble projections suggested continuedsuitability of the Aleutian Islands under both emission scenarios (Fig 3) Ensemble modelresults also reflected agreement on the opportunity for northward range expansion (Fig 3)Both the projected southern range contraction and northward range expansion are furtherconsistent with the hypothesized relationship between puffin habitat and temperature

California currentmdash2050Analysis of the California Current region within the overall ensemble models shows nearcomplete loss of suitable habitat between emission scenarios with both RCP 45 and 85(Fig 4) although the individual component models showed variable amounts of habitatloss GLM models projected the most dramatic loss along the California Current with apredicted loss of greater than 96 of suitable habitat under both scenarios (SD = 1461for projections under RCP 45 SD = 1296 for projections under RCP 85) (Table 1)GLMmodels also projected no habitat as likely to become newly habitable in the CaliforniaCurrent under either emission scenario Both GBM and RF models predicted less rangechange with GBM models projecting a mean loss of approximately 23 across algorithmvariants and RF models projecting a mean loss of approximately 27 across algorithm


Figure 3 North-America-wide habitat projection maps Tufted Puffin breeding habitat range projectionmaps Probabilistic maps color bins display percent probability of grid cell representing suitable habitat(A) Current projections (B) 2050 projections under RCP 45 (C) 2050 projections under RCP 85 Mapdata ccopy2017 Google


variants under RCP 85 (SD = 2027 SD = 2680 respectively) Under both RCP 45and 85 ensemble projections display complete loss of likely suitable habitat in Oregonand virtually complete loss in California (Fig 4) GBM and RF models projected smallportions of northwestern Washington would become slightly more likely than not tobecome environmentally habitable by 2050 under both emission scenarios though thoselocations may not exhibit other puffin habitat requirements

HindcastAs stated above limitations on biological survey for the Tufted Puffin make interpretationof hindcast results difficult though past projections further supported the hypothesisof higher temperature limiting suitable nesting habitat See supplementary material forfurther discussion and presentation of hindcast model results

DISCUSSIONEnsemble models uniformly support summer temperature as a predictor of Tufted Puffinbreeding habitat Highmodel evaluationmetrics (Table 1) coupled with strong correlationsbetween temperature variables and Tufted Puffin range change (Figs 1 and 2) provideconfidence that projected warmer summer temperatures are likely to be associated with


Figure 4 California Current habitat projection maps Tufted Puffin breeding habitat range projectionmaps exclusive to the California Current (32Nndash485N) Probabilistic maps color bins display percentprobability of grid cell representing suitable habitat (A) Current projections (B) 2050 projections un-der RCP 45 (C) 2050 projections under RCP 85 Note Probability bin lsquolsquo61ndash78rsquorsquo absent in B and C asprojections do not reflect any habitat within that bin Map tiles ccopyStamen Design underCC BY 30 DataccopyOpenStreetMap underODbL


the loss of greater than 92 of Tufted Puffin breeding habitat in the California Currentunder the examined emission scenarios (Fig 4) North America-wide ensemble modelsproject an overall loss across model algorithms and variants of approximately 22 andapproximately 26 of suitable habitat respectively under moderate emission reductionsand business as usual carbon emissions by 2050 (Table 1) Figure 3 highlights that mostnesting habitat will be lost along the southern portion of current Tufted Puffin range aswell as the opportunity for northward range expansion

Within the California Current ensemble projections show little variance in theprojected range loss under RCP 45 versus RCP 85 However the model algorithmsvaried considerably with respect to the percent of future habitat likely to be suitable inmore northerly latitudes (Table 1) This variance is likely due to the differences in modelingtechniques across algorithms described above and variance in initial estimation of suitablehabitat (Table 1 Fig 4) Both ensemble models project minimal suitable habitat remainingalong the California Current therefore under the hypothesized relationship betweentemperature and suitable habitat there is little left to become unsuitable with increased


warming (Fig 4) Another factor which may contribute to the lack of variance in projectedhabitat loss range-wide under different RCP scenarios in the California Current is therelatively short timeframe of 2050 projections Divergence in the temperature projectionsof RCPs 45 and 85 are amplified after 2050 with less divergence in temperature in theshort term (IPCC 2014)

Important to the interpretation of ensemble projections is the binary transformationof model outputs into suitable and unsuitable categories For range-change analysesprojections of unsuitable habitat represent a weighted average of lt50 probability ofsuitability a cutoff defined by ensemble calibration In some cases we observed a majorityof ensemble members projecting a particular cell as marginally suitable while a minority ofmembers strongly project that cell as unsuitable The subsequent result is unsuitable habitatdespite being marginally suitable in some models This process of binary transformationcan then reflect an aggregate of probabilistic scores instead of the average of a binaryprojection Binary transformations are thus a useful tool to discuss and represent howchanges in climate may affect the likelihood of suitable breeding conditions throughoutTufted Puffin range but are necessarily imprecise in that they mask underlying variabilityAdditionally many sites in the ensemble projections (Fig 4) become unsuitable afterbinary transformations while falling in to the 27ndash44 probability bin Though theseare projected as likely to become environmentally unsuitable they represent marginallysuitable environmental habitat and further research may help determine the suitabilityof other requisite habitat conditions at these sites to provide a broader picture of nestinghabitat suitability at these locations

Examining the variance among model results and the spatial variance in projections isintegral to the interpretation of model results from a conservation perspective (Guisan etal 2013 Porfirio et al 2014) Tufted Puffins are a relatively rare species in the southernportion of their range are hard to survey and occupy small areas of land (Hanson amp Wiles2015) These biological factors contribute to the difficulty of surveying (and thereforemodeling) puffins and can increase variance among model algorithms making ensemblemodels more valuable for interpretation of results Segurado amp Araujo 2004 Hernandezet al 2006 However here we use colony occupancy information rather than countsPreliminary occupancy analysis suggest that colony occupancy can be assessed with a highprobability with a single relatively rapid visit by boat even to a very small colony with fewbirds (Pearson et al see Supplemental File) Thus our colony occupancy approach likelyreflects actual changes in colony occupancy throughout the range In addition trendswere consistent across algorithms in depicting significant losses of suitability for habitatacross the California Current (specifically California and Oregon) British Columbia andeastern Alaska (Fig 3) All algorithms also projected the opportunity for northward rangeexpansion in the face of accelerating northern latitude warming (Fig 3)

If suitable habitat expands northward as projected by our ensemble models biologicaland ecological factors unrelated to climate such as eagle predation requisite nestingsubstrate etc are predicted to continue and likely to influence the probability ofcolonization (Hipfner et al 2012 Hanson amp Wiles 2015) Because colonization isuncertain we depict in Fig 5 the loss of currently suitable habitat in the California


Figure 5 Histogram of habitat loss in the California Current with nomigrationHistogram displayingthe amount of current California Current Extent suitable habitat projected to become unsuitable by 2050under RCP 85 (N = 120) Colors represent model algorithms In this analysis there is an assumption ofno migration or dispersal to potentially suitable new habitat


Current without the possibility of new colonization throughout the extentmdasha worst casescenario but an important component of analysis when examining the threat of climatechange Variance amongmodels as evidenced in Table 1 along the California Current failedto result in more than a handful of consensus areas of suitability (Fig 4) Ensemble modelsdid depict some areas of marginally suitable habitat along central California (Fig 4) Thisresult was likely influenced by a few outlying colonies such as the one in the FarallonIslands California These outlying colonies persist in opposition to the trends seen in othercolonies throughout the southern portion of puffin range Further examination of themechanisms driving puffin declines in their southern range may shed light on either theviability of these outlying suitable habitat projections as potential climate refugia or othermechanisms supporting these outlying colonies

The discussed hard-to-model specific requirements of puffin nesting habitat and otherecological population drivers make fine scale colony-by-colony analysis of extirpationrisk difficult Additionally analysis of regional trends in puffin success can serve to guideresearch examining the specific causal mechanisms driving documented declines whichwould aid in further analysis of colony-by-colony extirpation risk Importantly all modelsand especially ensemble results support the trend of southern range contraction associatedwith warm summer temperatures (Figs 1ndash4) Additionally while limitations on historicalsurvey data make interpretation of hindcasts difficult preliminary hindcasting resulted


in expansion across the southern portion of current puffin habitat (see SupplementalFile) This result is further consistent with the hypothesized relationship between hightemperature and puffin success

Our results are especially salient in light of the ongoing US Fish and Wildlife Servicersquosanalysis of puffin status following Natural Resources Defense Councilrsquos petition to list theCalifornia Current population as endangered When responding to the petition to list thepuffin the Service can list the species throughout its range or can list a distinct populationsegment (DPS) such as the breeding population south of the Canadian border or that in theCalifornia Current While determining which segments comprise a DPS as outlined by theESA requires more analysis our results provide the spatial information to inform the threatthat both of these breeding range segments or lsquolsquopopulationsrsquorsquo will likely face Our resultssuggest all potential distinct populations segments from British Columbia southward facea significant chance of near extirpation or very significant habitat loss under a wide rangeof climate projections by 2050

Conservation planning for species can greatly benefit from defining the portion(s) oftheir range representing habitat critical to their survival (Hagen amp Hodges 2006) Thisdesignation is essential for conservation planning both under the ESA as well as CanadianSpecies at Risk Act in which it is required for listed species as well as for more localizedconservation efforts (Taylor Suckling amp Rachlinski 2005) Figures 3 and 4 highlight areaswhere Tufted Puffins are currently at the highest risk of colony loss (low habitat suitability)Many puffin nesting sites are already managed by the US Fish and Wildlife Service refugesystem and many of these sites are also designated as lsquolsquowildernessrsquorsquo (Speich amp Wahl 1989and United States Fish and Wildlife Service 2017) Habitat projections made for the year2050 permit analysis of critical habitat in terms of species survival as well as proposedconservation efficacy (Suckling amp Taylor 2006 Stein et al 2013) Land acquisition hasproven to be an effective strategy for the management of endangered species and isa strategy that has been utilized for the Tufted Puffin (Lawler White amp Master 2003WDFW 2016) and could be used in the future With limited resources to conserve speciesat the federal level ranking the conservation priorities and temporally analyzing threatscan allow for prudent investment in conservation lands (Lawler White amp Master 2003)Nesting colony sites throughout the Gulf of Alaska are projected to remain suitable andresults indicate the Aleutian Islands are the most likely habitat to both continue to supportlarge populations of Tufted Puffins as well as potentially becoming suitable as new breedingsites (Fig 3) As these results suggest we can use this information to predict areas of futureTufted Puffin habitat to help outline areas for long-term conservation action while alsomapping areas where long-term conservation efforts may prove ineffective Such proactiveconservation steps often result in greater conservation outcomes and are critical for speciesstruggling to adapt to changing climates (Morrison et al 2011)

Mechanisms driving declineUsing the results reflected in Figs 3 and 4 wildlife managers can continue to explorethe causal mechanisms driving the discussed Tufted Puffin population declines andrange contraction Currently numerous pathways are proposed to help determine puffin


breeding success and adult survival such as prey availability SST predation and habitatdegradation (Morrison et al 2011 Hanson amp Wiles 2015) While many prey species donot show significant population trends (MacCall 1996) our results can provide spatialdetails to explore a potential mechanistic explanation vertical prey distribution (Gjerdrumet al 2003) Exact measurements are unknown but based on body size Tufted Puffinsexhibit the deepest maximum forage depths across alcids at approximately 110 m buttypically forage at 60 m or less (Piatt amp Kitaysky 2002) Tufted Puffins also forage muchfurther offshore than most other alcids and in deeper waters along continental shelf breaks(Ostrand et al 1998 Menza et al 2016) Foraging in deeper waters may leave TuftedPuffins susceptible to downward movement of prey species in the water column duringhigh temperatures (Ostrand et al 1998 Gjerdrum et al 2003) Further research aroundthese biological and ecological factors can be combined with our model results to furtherexplore the mechanisms behind the temperature-range relationship for Tufted Puffins(Ostrand et al 1998 Piatt amp Kitaysky 2002)

In addition to uncovering causal mechanisms current conservation efforts are beginningto examine diverging population patterns among related birds Rhinoceros Auklets(Cerorhinca monocerata) Cassinrsquos Auklets (Ptychoramphus aleuticus) as well as TuftedPuffins along the California Current (Greacutemillet amp Boulinier 2009 Morrison et al 2011)While these three alcids fill similar ecological roles recent years have seen dramaticpopulation swings varying among species (eg El-Nintildeo of 1997ndash98) (Morrison et al 2011)Cassinrsquos Auklets have displayed similar ecological sensitivity to changing environmentalconditions and have experienced recent large scale mortality events as recently as 2015(Sydeman et al 2006Wolf et al 2010Hanson amp Wiles 2015) Physiological and ecologicaldifferences between these related seabird species such as forage radius foraging depthand diet composition may provide insights into the mechanisms responsible for thesedifferences in population trends among species (Sydeman et al 2001 Wolf et al 2009Wolf et al 2010 Morrison et al 2011) For example using SDMs to model multiplespecies may provide insights into the relative influence of climate change on populationstrends (Johnson et al 2017)

CONCLUSIONOur analysis shows a strong negative correlation between warm summer temperatures andTufted Puffin nesting range particularly along the California Current Construction ofSDMs utilizing two different emissions scenarios for the year 2050 show southern rangecontraction and suggest a high risk of Tufted Puffin extirpation in the California Currentlarge marine ecosystem Ensemble projections support preliminary analyses suggestingthat temperature is driving the current puffin population declines and colony loss SDMmodel results can provide valuable input for conservation decision processes Specificallyour work provides the foundation for evaluating the threat of climate change and increasedsummer temperatures on Tufted Puffin breeding range


ACKNOWLEDGEMENTSWe would like to thank Shawn Stephenson of the United States Fish and Wildlife ServiceLaurie Wilson of Environment and Climate Change Canada and Robert Kaler of theAlaska Department of Fish and Game for their help in compiling Tufted Puffin surveydata We would also like to thank Eric Ward of the National Oceanic and AtmosphericAdministration for his help in model parameterization

ADDITIONAL INFORMATION AND DECLARATIONS

FundingThe authors received no funding for this work

Competing InterestsThe authors declare there are no competing interests

Author Contributionsbull Christopher J Hart conceived and designed the experiments performed the experimentsanalyzed the data contributed reagentsmaterialsanalysis tools prepared figures andortables authored or reviewed drafts of the paper approved the final draftbull Ryan P Kelly analyzed the data contributed reagentsmaterialsanalysis tools preparedfigures andor tables authored or reviewed drafts of the paper approved the final draftbull Scott F Pearson conceived and designed the experiments contributed reagentsmateri-alsanalysis tools authored or reviewed drafts of the paper approved the final draft

Supplemental InformationSupplemental information for this article can be found online at httpdxdoiorg107717peerj4519supplemental-information

REFERENCESAllouche O Tsoar A Kadmon R 2006 Assessing the accuracy of species distribution

models Prevalence kappa and the true skill statistic (TSS) Journal of Applied Ecology43(6)1223ndash1232 DOI 101111j1365-2664200601214x

ArauacutejoMB Alagador D CabezaM Nogueacutes-Bravo D ThuillerW 2011 Climatechange threatens European conservation areas Ecology Letters 14(5)484ndash492DOI 101111j1461-0248201101610x

ArauacutejoMB Pearson RG ThuillerW ErhardM 2005 Validation of species-climateimpact models under climate change Global Change Biology 11(9)1504ndash1513DOI 101111j1365-2486200501000x

Baird PH 1991 Optimal foraging and intraspecific competition in the Tufted Puffin TheCondor 93(3)503ndash515 DOI 1023071368182

Barbet-MassinM Jiguet F Albert CH ThuillerW 2012 Selecting pseudo-absences forspecies distribution models how where and how manyMethods in Ecology and Evolution 3327ndash338 DOI 101111j2041-210X201100172x


Barbet-MassinM ThuillerW Jiguet F 2010How much do we overestimate futurelocal extinction rates when restricting the range of occurrence data in climatesuitability models Ecography 33(5)878ndash886 DOI 101111j1600-0587201006181x

Bellard C Bertelsmeier C Leadley P ThuillerW Courchamp F 2012 Impactsof climate change on the future of biodiversity Ecology Letters 15(4)365ndash377DOI 101111j1461-0248201101736x

BentsenM Bethke I Debernard JB Iversen T Kirkevaringg A Seland Oslash Drange HRoelandt C Seierstad IA Hoose C Kristjaacutensson JE 2013 The Norwegian EarthSystem Model NorESM1-MndashPart 1 description and basic evaluation of the physicalclimate Geoscientific Model Development 6687ndash720 DOI 105194gmd-6-687-2013

Breiman L 2001 Random forestsMachine Learning 45(1)5ndash32DOI 101023A1010933404324

British Columbia Marine Conservation Analysis 2017Maps Data amp Reports Availableat https bcmcacadatafeatures eco_birds_tuftedpuffincolonies (accessed on May2017)

Broennimann O Fitzpatrick MC Pearman PB Petitpierre B Pellissier LYoccoz NG ThuillerW Fortin M-J Randin C Zimmermann NE GrahamCH Guisan A 2012Measuring ecological niche overlap from occurrence andspatial environmental data Global Ecology and Biogeography 21(4)481ndash497DOI 101111j1466-8238201100698x

Carnaval AC Moritz C 2008Historical climate modelling predicts patterns of currentbiodiversity in the Brazilian Atlantic forest Journal of Biogeography 35(7)1187ndash1201DOI 101111j1365-2699200701870x

Carvalho SB Brito JC Crespo EGWatts ME PossinghamHP 2011 Conservationplanning under climate change toward accounting for uncertainty in predictedspecies distributions to increase confidence in conservation investments in space andtime Biological Conservation 144(7)2020ndash2030 DOI 101016jbiocon201104024

CollinsWJ Bellouin N Doutriaux-Boucher M Gedney N Halloran P Hinton THughes J Jones CD Joshi M Liddicoat S Martin G OrsquoConnor F Rae J SeniorC Sitch S Totterdell I Wiltshire AWoodward S 2011 Development andevaluation of an Earth-System modelndashHADGEM2 Geoscientific Model Development41051ndash1075 DOI 105194gmd-4-1051-2011

DeGange AR Day RH 1991Mortality of seabirds in the japanese land-based gillnetfishery for salmon The Condor 93(2)251ndash258 DOI 1023071368940

Domisch S ArauacutejoMB Bonada N Pauls SU Jaumlhnig SC Haase P 2013Modellingdistribution in European stream macroinvertebrates under future climates GlobalChange Biology 19(3)752ndash762 DOI 101111gcb12107

Dufresne JL Foujols MA Denvil S Caubel A Marti O Aumont O Balkanski Y BekkiS Bellenger H Benshila R Bony S Bopp L Braconnot P Brockmann P Cadule PCheruy F Codron F Cozic A Cugnet D De Noblet N Duvel J-P Eacutetheacute C FairheadL Fichefet T Flavoni S Friedlingstein P Grandpeix J-Y Guez L Guilyardi EacuteHauglustaine D Hourdin F Idelkadi A Ghattas J Joussaume S KageyamaMKrinner G Labetoulle S Lahellec A Lefebvre M-P Lefeacutevre F Leacutevy C Li ZX Lloyd


J Lott F Madec G MancipMMarchandMMasson S Meurdesoif Y MignotJ Musat I Parouty S Polcher J Rio C Schulz M SwingedouwD S Szopa STalandier C Terray P Viovy N Vuichard N 2013 Climate change projectionsusing the IPSL-CM5 earth system model from CMIP3 to CMIP5 Climate Dynamics40(9)2123ndash2165 DOI 101007s00382-012-1636-1

Elith J Graham C Anderson R DudiacutekM Ferrier S Guisan A Hijmans RJ HuettmannFalk Leathwick JR Lehmann A Li J Lohmann LG Loiselle BA Manion G MoritzC NakamuraM Nakazawa Y Overton JMcCM Peterson AT Phillips SJ Richard-son K Scachetti-Pereira R Schapire RE Soberoacuten J Williams S Wisz MS Zimmer-mann N 2006 Novel methods improve prediction of speciesrsquo distributions fromoccurrence data Ecography 29(2)129ndash151 DOI 101111j20060906-759004596x

Elith J Leathwick JR Hastie T 2008 A working guide to boosted regression treesJournal of Animal Ecology 77(4)802ndash813 DOI 101111j1365-2656200801390x

Fielding AH Bell JF 1997 A review of methods for the assessment of prediction errorsin conservation presenceabsence models Environmental Conservation 24(1)38ndash49DOI 101017S0376892997000088

Fitzpatrick MC HargroveWW 2009 The projection of species distribution models andthe problem of non-analog climate Biodiversity and Conservation 18(8)2255ndash2261DOI 101007s10531-009-9584-8

FodenWB Butchart SHM Stuart SN Vieacute J-C AkcakayamHR Angulo A DeVantierLM Gutsche A Turak E Cao L Donner SD Katariya V Bernard R HollandRA Hughes AF OrsquoHanlon SE Garnett ST Sekercioğlu CH Mace G 2013Identifying the worldrsquos most climate change vulnerable species a systematic trait-based assessment of all birds amphibians and corals PLOS ONE 8(6)e65427DOI 101371journalpone0065427

Gallardo B Aldridge DC 2013 Evaluating the combined threat of climate change andbiological invasions on endangered species Biological Conservation 160225ndash233DOI 101016jbiocon201302001

Gent PR Danabasoglu G Donner LJ HollandMM Hunke EC Jayne SR LawrenceDM Neale RB Rasch PJ VertensteinMWorley PH Yang Z-L ZhangM 2011The community climate system model version 4 Journal of Climate 244973ndash4991DOI 1011752011JCLI40831

Giorgetta MA Jungclaus J Reick C Legutke S Bader J Bottinger M Brovkin VCrueger T EschM Fieg K Glushak K Gayler V Haak H Hollweg HD IlyinaT Kinne S Kornblueh L Matei D Mauritsen T Mikolajewicz U MuellerWNotz D Pithan F Raddatz T Rast S Redler R Roeckner E Schmidt H SchnurR Segschneider J Six KD Stockhause M Timmreck CWegner J WidmannHWieners KH ClaussenMMarotzke J Stevens B 2013 Climate and carboncycle changes from 1850 to 2100 in MPI-ESM simulations for the Coupled ModelIntercomparison Project phase 5 Journal of Advances in Modeling Earth Systems5(3)572ndash597 DOI 101002jame20038

Gjerdrum C Vallee AM St Clair CC BertramDF Ryder JL Blackburn GS 2003Tufted Puffin reproduction reveals ocean climate variability Proceedings of the


National Academy of Sciences of the United States of America 100(16)9377ndash9382DOI 101073pnas1133383100

Pentildea Goacutemez FT Guerrero PC Bizama G Duarte M Bustamante RO 2014 Climaticniche conservatism and biogeographical non-equilibrium in Eschscholzia californica(Papaveraceae) an invasive plant in the Chilean Mediterranean Region PLOS ONE9(8)e105025 DOI 101371journalpone0105025

Goyert HF Garton EO BADrummond Renner HM 2017 Density dependenceand changes in the carrying capacity of Alaskan seabird populations BiologicalConservation 209178ndash187 DOI 101016jbiocon201702011

Greacutemillet D Boulinier T 2009 Spatial ecology and conservation of seabirds facingglobal climate change a reviewMarine Ecology Progress Series 391121ndash137DOI 103354meps08212

Griffies SMWintonM Donner LJ Horowitz LW Downes SM Farneti R Gnanade-sikan A HurlinWJ Lee HC Liang Z Palter JB Samuels BLWittenberg ATWyman BL Yin J Zadeh N 2011 The GFDL CM3 coupled climate model char-acteristics of the ocean and sea ice simulations Journal of Climate 24(13)3520ndash3544DOI 1011752011JCLI39641

Guisan A Edwards T Hastie T 2002 Generalized linear and generalized additivemodels in studies of species distributions setting the scene Ecological Modelling157(2)89ndash100 DOI 101016S0304-3800(02)00204-1

Guisan A ThuillerW 2005 Predicting species distribution offering more than simplehabitat models Ecology Letters 8(9)993ndash1009 DOI 101111j1461-0248200500792x

Guisan A Tingley R Baumgartner JB Naujokaitis-Lewis I Sutcliffe PR TullochAI Regan TJ Brotons L McDonald-Madden E Mantyka-Pringle C MartinTG Rhodes JR Maggini R Setterfield SA Elith J Schwartz MWWintle BABroennimann O Austin M Ferrier S KearneyMR PossinghamHP Buckley YM2013 Predicting species distributions for conservation decisions Ecology Letters16(12)1424ndash1435 DOI 101111ele12189

Guisan A Zimmermann NE 2000 Predictive habitat distribution models in ecologyEcological Modeling 135(2)147ndash186 DOI 101016S0304-3800(00)00354-9

Guo C Lek S Ye S Li W Liu J Li Z 2015 Uncertainty in ensemble modelling of large-scale species distribution effects from species characteristics and model techniquesEcological Modelling 30667ndash75 DOI 101016jecolmodel201408002

Hagen AN Hodges KE 2006 Resolving critical habitat designation failures reconcilinglaw policy and biology Conservation Biology 20(2)399ndash407DOI 101111j1523-1739200600320x

Hanson TWiles GJ 2015Washington state status report for the Tufted Puffin OlympiaWashington Department of Fish and Wildlife 66

Harris I Jones PD Osborn TJ Lister DH 2014 Updated high-resolution grids ofmonthly climatic observationsmdashthe CRU TS310 Dataset International Journal ofClimatology 34(3)623ndash642 DOI 101002joc3711


Hernandez PA Graham CHMaster LL Albert DL 2006 The effect of sample size andspecies characteristics on performance of different species distribution modelingmethods Ecography 29(5)773ndash785 DOI 101111j0906-7590200604700x

Hijmans RJ Cameron SE Parra JL Jones PG Jarvis A 2005 Very high resolution in-terpolated climate surfaces for global land areas International Journal of Climatology25(15)1965ndash1978 DOI 101002joc1276

Hipfner JM Blight LK Lowe RWWilhelm SI Robertson GJ Barrett RT Anker-Nilssen T Good TP 2012 Unintended consequences how the recovery of sea eagleHaliaeetus spp populations in the northern hemisphere is affecting seabirdsMarineOrnithology 4039ndash52

Hipfner JM Charette MR Blackburn GS 2007 Subcolony variation in breeding successin the Tufted Puffin (Fratercula Cirrhata) association with foraging ecology andimplications The Auk 124(4)1149ndash1157DOI 1016420004-8038(2007)124[1149SVIBSI]20CO2

Intergovernmental Panel on Climate Change (IPCC) 2014 In Pachauri RK Meyer LAeds Climate change 2014 synthesis report Contribution of working groups I II and IIIto the fifth assessment report of the intergovernmental panel on climate change GenevaIPCC 151 pp

Johnson CN Balmford A Brook BW Buettel JC Galetti M Guangchun LWilmshurstJM 2017 Biodiversity losses and conservation responses in the AnthropoceneScience 356(6335)270ndash275 DOI 101126scienceaam9317

Klausmeyer KR ShawMR 2009 Climate change habitat loss protected areas and theclimate adaptation potential of Mediterranean ecosystems worldwide PLOS ONE4(7)e6392 DOI 101371journalpone0006392

Labay B Cohen AE Sissel B Hendrickson DA Martin FD Sarkar S 2011 Assessinghistorical fish community composition using surveys historical collection data andspecies distribution models PLOS ONE 6(9)e25145DOI 101371journalpone0025145

Lawler JJ White D Master LL 2003 Integrating representation and vulnerabilitytwo approaches for prioritizing areas for conservation Ecological Applications13(6)1762ndash1772 DOI 10189002-5337

MacCall A 1996 Patterns of low-frequency variability in fish populations of theCalifornia current CalCOFI Reports 37100ndash110

MarmionM ParviainenM LuotoM Heikkinen RK ThuillerW 2009 Evaluationof consensus methods in predictive species distribution modelling Diversity andDistributions 15(1)59ndash69 DOI 101111j1472-4642200800491x

McCullagh P Nelder JA 1989Generalized linear models London Chapman and HallMenza C Leirness J White TWinship A Kinlan B Kracker L Zamon J Ballance L

Becker E Forney KA Barlow J Adams J Pereksta D Pearson S Pierce J JeffriesS Calambokidis J Douglas A Hanson B Benson SR Antrim L 2016 Predictivemapping of seabirds pinnipeds and cetaceans off the pacific coast of WashingtonNOAA Technical Memorandum NOS NCCOS 210 NOAA NCCOS Center ofCoastal Monitoring and Assessment Silver Spring 96 pp DOI 107289V5NV9G7Z


Morrison SA Sillett TS Ghalambor CK Fitzpatrick JW Graber DM Bakker VJBowman R Collins CT Collins PW Delaney KS Doak DF KoenigWDLaughrin L Lieberman AA Marzluff JM Reynolds MD Scott JM Stallcup JAVickersW BoyceWM 2011 Proactive conservation management of an island-endemic bird species in the face of global change BioScience 61(12)1013ndash1021DOI 101525bio2011611211

Nazarenko L Schmidt GA Miller RL Tausnev N Kelley M Ruedy R Russell GLAleinov I Bauer M Bauer S Bleck R Canuto V Cheng Y Clune TL Del GenioAD Faluvegi G Hansen JE Healy RJ Kiang NY Koch D Lacis AA LeGrandeAN Lerner J Lo KK Menon S Oinas V Perlwitz J PumaMJ Rind D RomanouA SatoM Shindell DT Sun S Tsigaridis K Unger N Voulgarakis A YaoM-S Zhang J 2015 Future climate change under RCP emission scenarios withGISS ModelE2 Journal of Advances in Modeling Earth Systems 7(1)244ndash267DOI 1010022014MS000403

Osa YWatanuki Y 2002 Status of seabirds breeding in Hokkaido Journal of theYamashina Institute for Ornithology 33107ndash141 DOI 103312jyio195233107

OstrandWD Coyle KO Drew GS Maniscalco JM Irons DB 1998 Selection of forage-fish schools by murrelets and Tufted Puffins in Prince William Sound Alaska TheCondor 100286ndash297 DOI 1023071370269

Parmesan C Yohe G 2003 A globally coherent fingerprint of climate change impactsacross natural systems Nature 42139ndash42 DOI 101038nature01286

Pearson K 1901 On lines and planes of closest fit to systems of points in space Philo-sophical Magazine Series 6 2(11)559ndash572 DOI 10108014786440109462720

Piatt JF Kitaysky AS 2002 Tufted Puffin Fratercula cirrhata Birds of North America7081ndash31

Ponce-Reyes R Plumptre AJ Segan D Ayebare S Fuller RA PossinghamHPWatsonJEM 2017 Forecasting ecosystem responses to climate change across Africarsquos Alber-tine Rift Biological Conservation 209464ndash472 DOI 101016jbiocon201703015

Porfirio LL Harris RMB Lefroy EC Hugh S Gould SF Lee G Bindoff NL MackeyB 2014 Improving the use of species distribution models in conservationplanning and management under climate change PLOS ONE 9(11)e113749DOI 101371journalpone0113749

R Core Team 2013 R a language and environment for statistical computing Vienna RFoundation for Statistical Computing Available at httpwwwR-projectorg

Raxworthy CJ Martinez-Meyer E Horning N Nussbaum RA Schneider GEOrtega-Huerta MA Townsend Peterson A 2003 Predicting distributions ofknown and unknown reptile species in Madagascar Nature 426(6968)837ndash841DOI 101038nature02205

Rayner NA Parker D Horton E Folland C Alexander L Rowell D Kent EC KaplanA 2003 Global analyses of sea surface temperature sea ice and night marine airtemperature since the late nineteenth century Journal of Geophysical Research108(D14)2156ndash2202 DOI 1010292002JD002670


Ricca MA KeithMiles A Anthony RG 2008 Sources of organochlorine contaminantsand mercury in seabirds from the Aleutian archipelago of Alaska inferences fromspatial and trophic variation Science of the Total Environment 406(1ndash2)308ndash323DOI 101016jscitotenv200806030

Richardson DMWhittaker RJ 2010 Conservation biogeographyndashfoundations conceptsand challenges Diversity and Distributions 16313ndash320DOI 101111j1472-4642201000660x

Ridgeway G 1999 The state of boosting Computing Science and Statistics 31172ndash181Rodriacuteguez-Castantildeeda G Hof AR Jansson R Harding LE 2012 Predicting the fate of

biodiversity using speciesrsquo distribution models enhancing model comparability andrepeatability PLOS ONE 7(9)e44402 DOI 101371journalpone0044402

Segurado P AraujoMB 2004 An evaluation of methods for modelling species distribu-tions Journal of Biogeography 31(10)1555ndash1568DOI 101111j1365-2699200401076x

Sewell B 2014 Petition to list the contiguous US distinct population segment of TuftedPuffin (Fratercula cirrhata) under the Endangered Species Act New York NaturalResources Defence Council 51 pp

Shabani F Kumar L Ahmadi M 2016 A comparison of absolute performance ofdifferent correlative and mechanistic species distribution models in an independentarea Ecology and Evolution 6(16)5973ndash5986 DOI 101002ece32332

Siegel K Cummings B 2005 Petition to list the polar bear (Ursus maritimus) as athreatened species under the Endangered Species Act Tucson Center for BiologicalDiversity 154 pp

Speich SMWahl TR 1989 Catalog of Washington seabird colonies US Fish andWildlife Service Biological Report 88(6) United States Department of the InteriorFish Wildlife Service Research and Development Washington DC 510 pp

Stefan HG Preudrsquohomme EB 1993 Stream temperature estimation from air tem-perature Journal of the American Water Resources Association 29(1)27ndash45DOI 101111j1752-16881993tb01502x

Stein BA Staudt A Cross MS Dubois NS Enquist C Griffis R Hansen LJ HellmannJJ Lawler JJ Nelson EJ Pairis A 2013 Preparing for and managing change climateadaptation for biodiversity and ecosystems Frontiers in Ecology and the Environment11502ndash510 DOI 101890120277

Suckling K Taylor M 2006 Critical habitat and recovery In Dale D Goble J MichaelS Frank WD eds The endangered species act at thirty volume 1 renewing theconservation promise Washington Island Press 755ndash789

SydemanWJ Bradley RWarzybok P Abraham Jahncke C Hyrenbach K KouskyV Hipfner JM OhmanM 2006 Planktivorous auklet (Ptychoramphus aleuticus)responses to ocean climate 2005 unusual atmospheric blocking GeophysicalResearch Letters 33Article L22S09 DOI 1010292006GL026736

SydemanWJ Hester MM Thayer JA Gress F Martin P Buffa J 2001 Climate changereproductive performance and diet composition of marine birds in the southern


California Current system 1969-1997 Progress in Oceanography 49309ndash329DOI 101016S0079-6611(01)00028-3

Taylor MFJ Suckling KF Rachlinski JJ 2005 The effectiveness of the endangeredspecies act a quantitative analysis BioScience 55(4)360ndash367DOI 1016410006-3568(2005)055[0360TEOTES]20CO2

Thomas C Cameron A Green R Bakkenes M Beaumont LJ Collingham YC Eras-mus BFN De Siqueira MF Grainger A Hannah L Hughes L Huntley B VanJaarsveld AS Midgley GF Miles L Ortega-Huerta MA Peterson AT PhillipsOLWilliams S 2004 Extinction Risk from climate change Nature 427145ndash148DOI 101038nature02121

ThuillerW Brotons L ArauacutejoMB Lavorel S 2004 Effects of restricting environ-mental range of data to project current and future species distributions Ecography27(2)165ndash172 DOI 101111j0906-7590200403673x

ThuillerW Lafourcade B Engler R ArauacutejoMB 2009 BIOMODmdasha platformfor ensemble forecasting of species distributions Ecography 32(3)369ndash373DOI 101111j1600-0587200805742x

United States Fish andWildlife Service 2016 Endangered and threatened wildlife andplants proposed ruling for the North American Wolverine 81 Federal Register20171670ndash71671

United States Fish andWildlife Service 2017 National wildlife refuge system Availableat httpswwwfwsgov refugeswhmwildernesshtml (accessed on March 2017)

VanHorn B 2002 Approaches to habitat modelling the tensions between patternand process and between specificity and generality In Scott JM Heglund PJMorrison ML Haufler JB Raphael MG Wall WA Samson FB eds Predicting speciesoccurrences issues of accuracy and scale Covelo Island Press 63ndash72

Walther G Post E Convey P Menzel A Parmesan C Beebee TJC Fromentin J-MHoegh-Guldberg O Bairlein F 2002 Ecological responses to recent climate changeNature 416389ndash395 DOI 101038416389a

Washington Department of Fish andWildlife 2016Wildlife area managementplanning framework Web Available at httpwdfwwagovpublications01810wdfw01810pdf (accessed on 21 March 2017)

Washington Fish andWildlife Commission 2015WSR 15-10-022 sect232-12-014Available at http lawfilesextlegwagov lawwsr20151015-10PERMpdf

Wolf S Cummings B Siegel K 2008 Petition to list three seal species under the endangeredspecies act ringed seal (pusa hispida) bearded seal (erignathus barbatus) and spottedseal (phoca largha) Tucson Center for Biological Diversity 139 pp

Wolf S Snyder M SydemanW Doaks D Croll D 2010 Predicting population con-sequences of ocean climate change for an ecosystem sentinel the seabird CassinrsquosAuklet Global Change Biology 161923ndash1935 DOI 101111j1365-2486201002194x

Wolf SG SydemanWJ Hipfner JM Abraham CL Tershy BR Croll DA 2009 Range-wide reproductive consequences of ocean climate variability for the seabird CassinrsquosAuklet Ecology 90742ndash753 DOI 10189007-12671


World Seabird Union 2017 North Pacific Seabird Data Portal Web Available at http axiomseabirdsnetmaps js seabirdsphpapp=north_pacificz=3ampll=5500000-17000000 (accessed on 28 March 2017)

Xin X-GWu T-W Zhang J 2013 Introduction of CMIP5 experiments carried outwith the climate system models of Beijing Climate Center Advances Climate ChangeResearch 4(1)41ndash49 DOI 103724SPJ12482013041


66 of amphibians and 70 of corals that were identified as highly vulnerable to theimpacts of climate change are not currently considered threatened with extinction on theIUCN Red List of Threatened Species indicating that speciesrsquo vulnerabilities are likely tobe much greater than conservation status alone would suggest

In recent years in the United States the United States Fish andWildlife Service (USFWS)and the National Marine Fisheries Service (NMFS) have received several petitions to listspecies under the Endangered Species Act due to the impacts of climate change (Siegelamp Cummings 2005 Wolf Cummings amp Siegel 2008) However the link between climatechange and risk to a species can be difficult to assess One approach to examining theselinkages is to model the interaction between climate and suitable habitat for a givenspecies given what is already known about the relationship between the species and itshabitat This approach has become an integral component of conservation planning in aworld of changing environments (Hagen amp Hodges 2006 Richardson amp Whittaker 2010)Ultimately understanding these linkages can help inform conservation assessments andspecies and ecosystem management strategies (Carnaval amp Moritz 2008 Ponce-Reyes et al2017) for example by estimating the likelihood of losing (or gaining) particular suitablehabitats of interest under future climate conditions

Tufted puffins as a pertinent exampleThe Tufted Puffin (Fratercula cirrhata) is an iconic species that is experiencing dramaticpopulation declines across the southern portion of its geographic range (Piatt amp Kitaysky2002) While Tufted Puffin populations in the Alaska Current have remained relativelystable (but see Goyert et al 2017) populations in the California Current large marineecosystem (area of the eastern Pacific Ocean spanning nearly 3000 km from southernBritish Columbia Canada to Baja California Mexico) have declined by approximately90 relative to early 20th century estimates and are currently declining 9 annually(Hanson amp Wiles 2015) The number of occupied breeding-colony sites in WashingtonState has declined by 60 relative to the 1886ndash1977 average and 45 relative to the1978ndash1984 average (Hanson amp Wiles 2015) Range contractions at the southern edge ofthe Tufted Puffinrsquos habitat in both the eastern and western Pacific Ocean have led topreliminary conservation measures the Washington Department of Fish and Wildlife haslisted the Tufted Puffin as endangered at the state level in 2015 with Japanrsquos Ministry of theEnvironment listing the species as endangered in 1993 (Osa amp Watanuki 2002 Hanson ampWiles 2015Washington Fish and Wildlife Commission 2015)

Tufted puffin biology and ecologyTufted Puffins are seabirds belonging to the family Alcidae and nest in colonies locatedon both sides of the North Pacific ranging in North America from the Channel Islands insouthern California (34N) to coastal northern Alaska (68N) (Piatt amp Kitaysky 2002) andin Asia from Hokkaido Japan (43N) through the Kamchatka Peninsula (63N) (Hansonamp Wiles 2015) They are central-place foragers during the nesting season when they digburrows or use crevices for nesting on nearshore rocks islands and sea stacks (Piatt ampKitaysky 2002) During the nesting season puffins exhibit large foraging radii around their

































Ensemble GLM GBM RF
































































































































































































Ensemble GLM GBM RF

































































































































































will the california current lose its nesting tufted …influencing puffin population dynamics and...

Documents