ArticlePDF Available

Abstract and Figures

Environmental factors shape the spatial distribution and dynamics of populations. Understanding how these factors interact with movement behavior is critical for efficient conservation, in particular for migratory species. Adult female green sea turtles, Chelonia mydas, migrate between foraging and nesting sites that are generally separated by thousands of kilometers. As an emblematic endangered species, green turtles have been intensively studied, with a focus on nesting, migration, and foraging. Nevertheless, few attempts integrated these behaviors and their trade‐offs by considering the spatial configurations of foraging and nesting grounds as well as environmental heterogeneity like oceanic currents and food distribution. We developed an individual‐based model to investigate the impact of local environmental conditions on emerging migratory corridors and reproductive output and to thereby identify conservation priority sites. The model integrates movement, nesting, and foraging behavior. Despite being largely conceptual, the model captured realistic movement patterns which confirm field studies. The spatial distribution of migratory corridors and foraging hot spots was mostly constrained by features of the regional landscape, such as nesting site locations, distribution of feeding patches, and oceanic currents. These constraints also explained the mixing patterns in regional forager communities. By implementing alternative decision strategies of the turtles, we found that foraging site fidelity and nesting investment, two characteristics of green turtles' biology, are favorable strategies under unpredictable environmental conditions affecting their habitats. Based on our results, we propose specific guidelines for the regional conservation of green turtles as well as future research suggestions advancing spatial ecology of sea turtles. Being implemented in an easy to learn open‐source software, our model can coevolve with the collection and analysis of new data on energy budget and movement into a generic tool for sea turtle research and conservation. Our modeling approach could also be useful for supporting the conservation of other migratory marine animals. Our model represents the migration of green sea turtles in the southwest Indian Ocean between nesting and foraging sites. It can cover thousands of km and is contratined by the regional landscape and oceanic currents. Emerging migration corridors were identified, which can be used for identifying priority sites for conservation.
Content may be subject to copyright.
Ecology and Evolution. 2019;00:1–26.    
|
 1
www.ecolevol.org
Received:6April2019 
|
  Revised:2 2July2019 
|
  Accepted:24July2019
DOI: 10.1002/ece 3.5552
ORIGINAL RESEARCH
Modeling the emergence of migratory corridors and foraging
hot spots of the green sea turtle
Mayeul Dalleau1| Stephanie Kramer‐Schadt2,3| Yassine Gangat4| Jérôme Bourjea5|
Gilles Lajoie6| Volker Grimm7,8 ,9
ThisisanopenaccessarticleunderthetermsoftheCreativeCommonsAttributionLicense,whichpermitsuse,distributionandreproductioninanymedium,
providedtheoriginalworkisproperlycited.
©2019TheAuthors.Ecology an d EvolutionpublishedbyJohnWiley&SonsLtd.
1Centred'EtudeetdeDécouvertedes
TortuesMarines(CEDTM),SaintLeu/La
Réunion,France
2DepartmentofEcological
Dynamics,LeibnizInstituteforZooand
WildlifeResearch,Berlin,Germany
3DepartmentofEcology,Technische
UniversitätBerlin,Berlin,Germany
4LIM‐IREMIA,EA2525,UniversityofLa
Réunion,PTU,Sainte‐Clotilde/LaRéunion,
France
5MARBEC,IRD,InstitutFrançaisde
Recherchepourl'ExploitationdelaMer
(Ifremer),CNRS,UniversitédeMontpellier,
SèteCedex,France
6UMREspace‐Dev,Universit yofLaRéunion,
Saint‐Denis,France
7DepartmentofEcological
Modelling,Helmholt zCentrefor
EnvironmentalResearch–UFZ,Leipzig,
Germany
8DepartmentofPlantEcologyandNature
Conser vation,UniversityofPotsdam,
Potsdam‐Golm,Germany
9GermanCentreforIntegrativeBiodiversity
Research(iDiv)Halle‐Jena‐Leipzig,Leipzig,
Germany
Correspondence
VolkerGrimm,DepartmentofEcological
Modelling,Helmholt zCentrefor
EnvironmentalResearch–UFZ,Permoserstr.
15,04318Leipzig,Germany.
Email:volker.grimm@ufz.de
Abstract
Environmental factors shape thespatialdistributionand dynamicsof populations.
Understandinghowthesefactorsinteractwithmovementbehavioriscriticalforeffi‐
cientconservation,inparticularformigratoryspecies.Adultfemalegreenseaturtles,
Chelonia mydas,migratebetweenforagingandnestingsitesthataregenerallysepa
ratedbythousandsofkilometers.Asanemblematicendangeredspecies,greentur‐
tleshavebeenintensivelystudied,withafocusonnesting,migration,andforaging.
Nevertheless,fewattemptsintegratedthesebehaviorsandtheirtrade‐offsbycon‐
sideringthespatialconfigurationsofforagingandnestinggroundsaswellasenviron‐
mentalheterogeneitylikeoceaniccurrentsandfooddistribution.Wedevelopedan
individual‐basedmodeltoinvestigate theimpact of localenvironmental conditions
on emerging migratory corridors andreproductiveoutput and to thereby identify
conservation priority sites. Themodel integrates movement, nesting, andforaging
behavior.Despitebeing largelyconceptual,themodelcapturedrealisticmovement
patterns whichconfirmfieldstudies.Thespatial distributionofmigratorycorridors
andforaginghotspotswasmostlyconstrainedbyfeaturesoftheregionallandscape,
suchasnestingsitelocations,distributionoffeedingpatches,andoceaniccurrents.
These constraintsalso explained themixing patterns inregionalforagercommuni‐
ties. Byimplementing alternative decision strategies ofthe turtles, wefound that
foraging site f idelity and nes ting investment, t wo characteris tics of green turt les'
biology,are favorable strategiesunderunpredictableenvironmental conditionsaf
fecting theirhabitats.Basedonour results, wepropose specific guidelines forthe
regionalconservationofgreenturtlesaswellasfutureresearchsuggestionsadvanc‐
ingspatialecologyofseaturtles.Beingimplementedinaneasytolearnopen‐source
software,our model cancoevolve withthe collectionandanalysisofnew dataon
energybudgetandmovementintoagenerictoolforseaturtleresearchandconser‐
vation.Ourmodelingapproachcouldalsobeusefulforsupportingtheconservation
ofothermigratorymarineanimals.
KEY WORDS
connectivity,corridors,individual‐basedmodel,migration,movement,seaturtle
2 
|
   DALLEAU Et AL.
1 | INTRODUCTION
Manyspeciesmigratetoexploitresourcesheterogeneouslydistrib
utedinspaceandtime(Jorgensen,Dunlop,Opdal,&Fiksen,2008).
Individualsmust allocate theseresourcesinternallytogrowth,sur‐
vival,andreproductioninawaythatmaximizestheirfitness(Martin,
Jager,Preuss,Nisbet,&Grimm,2013;Roff,2002;Siblyetal.,2013;
Varpe, Jørgensen, Tarling,&Fiksen,2008).Animalmigration costs
must therefore be balanced by fitness benefits (Milner‐Gulland,
Fryxell,&Sinclair,2011).Consequently,evensmallchanges,forex‐
ample,inthequalityofbreedingorfeedingpatchescansignificantly
influencelong‐termpopulationsurvival(Fiksen&Jorgensen,2011;
Taylor&Norris,2010).
Thegreensea turtle, Chelonia mydas,isa wide‐rangingspecies
distributed worldwide (Plotkin, 2003; Figure 1) and classified as
endangeredintheIUCN red list(Seminoff,2004). As adults, green
turtlesperformlong‐distancemigrationbetweenfeedingandnest‐
ing sites, w hich are generally s eparated by thousands of kilome‐
ters (Godley etal., 2008). They exhibit strong natal philopatry (or
natal homing)and tend to nestonthesame sitethattheyhatched
(Jensenet al.,2019;Lohmann,Witherington,Lohmann,&Salmon,
1997). The sout hwest Indian Ocea n (SWIO) shelters so me of the
world's majorgreenturtlerookeries(Bourjea,Dalleau,et al.,2015;
Bourjea, Frappier,et al.,2007; Dalleau et al., 2012; Dervilleet al.,
2015; Laure t‐Stepler et al., 20 07; Mortimer, von Bran dis, Liljevik,
Chapman,&Collie, 2011)that aredistributed acrosstheentirere‐
giononoceanicislandsspreadalongtheMozambiqueChanneland
theMascareneplateau (Figure2).Otherminor nestingsitesarelo‐
catedoncontinentalislandsandshoresonthecoastofMadagascar
and East Africa (Bourjea,Ciccione, & Ratsimbazafy, 2006; Garnier
etal.,2012).Seagrassbeds,themaincomponentofadultgreentur
tlediet (Bjorndal,1997),extendalmost continuously over theeast
Africa n coast from Moza mbique to Soma lia and over the wes tern
coastofMadagascar(Figure2;Gullströmetal.,2002),andforaging
greenturtlesareobservedinallcountriesoftheSWIOhostingsea
grass beds(Ballorainetal., 2010; Fulandaetal., 2007;Muir,2005;
Okemwa,Nzuki,&Mueni,2004;Williams,Pierce,Rohner,Fuentes,
& Hamann , 2017). A tracking s tudy (Dallea u, 2013) demonst rated
that (a) the nor thern par t of the Mozambiq ue Channel is a majo r
oceanicmigrationcorridorforpostnestinggreenturtlescapableto
migrate thousands of kilometers,(b)coastal groundsofEast Africa
and West Mad agascar are im portant for aging sites and mig ration
corridors,(c)turtlesfromtheSWIOnestingsitesmakeextensiveuse
ofavailableforaginghabitatsof the whole region, and (d) foraging
grounds areused by turtles originatingfrom different rookeriesof
theSWIO.
ThreatsarehighlyvariableintheSWIOregion,whichisbordered
bycountries andprovinces ofheterogeneouseconomic levels.The
regionhasbeenidentifiedasaspecific“RegionalManagementUnit”
forthegreenturtle,thatis,aspatiallyexplicitpopulationsegments
definedbybiogeographicaldataofthisspecies(Wallace,DiMatteo,
etal.,2010).Long‐termlocalprotectionatnestingsitesisanimport
antcomponentofseaturtleconservation (Chaloupka et al., 2008).
Nevertheless,adultgreenturtlesspendmostoftheirlifetimeonfor
aginggroundswheretheyareexposedtoimportantthreatssuchas
direct exploitationofeggs,meat, and shellsorfisheriesinteraction
(Wallace , Lewison, et al., 2010), es pecially in the SW IO (Bourjea,
2015; Temple et al., 2018; Williams, Pierce, Fuentes, & Hama nn,
2016)where for instance more than between 10,000 and 16,000
green turtleswere estimated tobecaptured by thelocalartisanal
fisherytobesoldinlocalmarketsforconsumptioneachyearonlyin
aportion ofthesouthwestcoastofMadagascar(Humber,Godley,
Ramahery,&Broderick,2011).Thus,conservationplanscanonlybe
effic ient with coordi nated protect ion measures en compassing the
wholespatialscaleofseaturtle'sdistribution.Tofocusconservation
efforts where theyare mostrequired and efficient, it is anurgent
needtounderstandthefactorsthatgovernthespatialdynamicsof
thespecies and the life‐historystrategiesthatleadtoeffectivecy‐
clesofforaging,migration,andnesting.
Several concepts exist to describe how resource patches are
mostefficientlyexploitedbyanimals(Eliassen,Jorgensen,Mangel,&
Giske,2009).AnexampleistheMarginalValueTheorem (Charnov,
1976) that predicts t hat a forager should leave a patch when its
food intake drops below the average food intake on all other
patches.Asanother example, the theoryof IdealFreeDistribution
(Fretwell &Lucas,1970)predictsthattheproportion ofindividuals
FIGURE 1 Thegreenseaturtle,Chelonia mydas(photo:J.
Bourjea/Ifremer)
    
|
 3
DALLEAU E t AL.
exploitingdifferentgivenresourcepatchesshouldbeproportional
to the patches' resource levels. Nevertheless, these two com
plementary concepts (and others) make the somewhat unrealis
tic assumption that foragers can per fectly assess resource levels
and heterogeneity over an entire region and respond accordingly
(Eliassenetal.,2009;Railsback&Harvey,2013).Realityoftendiffers
from thisassumption. For anyspecies, migration towardafeeding
patchrequiresenergeticallycostlymovementsthatmighthavelim‐
itedbenefitsifthetargetfeedingpatchisalreadydepleted.Itseems
reasonabletoassumethatinthecaseofseaturtles,individualshave
little i f any informatio n about the loc ation of the fee ding patches
that are ideal at a giventime. Itthus remainsan openquestion to
whatdegreethedistributionofturtlesonfeedingpatchesisdeter‐
minedbythesites'accessibility,theturtle'sknowledgeoftheirloca‐
tion(foragingsitefidelity),anddistancetothenestingsite.
In addition to feeding patch selection, heterogeneous land‐
scapesarealsolikelytohavestrong effectsonanimal'smovement
patter ns and hence the r esulting conne ctivit y among feedin g and
nesting sites (Graf, Kramer‐Schadt, Fernandez, & Grimm, 2007;
Olden, Schooley, Monroe, & Poff, 2004; Pe'er & Kramer‐Schadt,
2008; Revilla, Wiegand, Palomares, Ferreras, & Delibes, 2004).
Oceanic currents often play a major role in foraging ecology of
marine animals(Bostetal.,2009;Chapman et al., 2011),especially
oceanographic fronts (Scales etal., 2014), andsea turtles'oceanic
movementsaredirectlyaffectedbyoceaniccurrents(Girard,Sudre,
Benhamou,Roos,&Luschi,2006;Luschi,Hays,&Papi,2003).The
early lifestage of marine turtles(thatcan last decades) is oceanic,
andthespatialfateisalsostronglyimpactedbyoceaniccurrentsand
mayhaveconsequences thatprevailand shapethe spatialdynam‐
icsofadult stages (Gaspar&Lalire,2017).Furthermore, terrestrial
areas,with the exceptionofnestinggrounds,constitutebarriersto
sea tur tle's migratio n as well as potentia l navigationa l cues (Hays,
Broder ick, Godley, et al. , 2002). Migrato ry constr aints might th en
differdrasticallyforislandssurroundedbycoastalareaslikeTaiwan
intheChinaSea(Cheng,2000)incomparisonwithoceanicisolated
islandslikeAscensionIslandintheSouthernAtlanticOcean(Luschi
etal.,2003).
Insummary,successfulseaturtle conservationseems to bein‐
trinsicallylinkedtotheforagingandmigrationprocesses,withnatal
homingfornestingbeingoneofthekeyfactorsdrivingseaturtlelife
history.Wethereforedevelopedaspatiallyexplicitindividual‐based
model(Grimm&Railsback,2005;R ailsback&Grimm,2019)toquali
tativelys tudythespatialdynamicsofadultgreenturtleintheSWIO.
Individual‐based modeling, in a large sense, has been used
before to ad dress variou s aspects of s ea turtle eco logy. A first
kindofIBM,thatwascommonlyimplemented,concernsthespa
tialfate of hatchlingsduring their firstyearsinthe openocean.
Lagrangian modeling of passive drift trajectories has allowed
predictingthespatial cycleofjuvenileseaturtles(Blumenthalet
al.,2009;Godleyetal.,2010;Hays,Fossette,Katselidis,Mariani,
&Schofield, 2010; Putman & Naro‐Maciel,2013).Limitsofpas
sive drift are, however, of concern, andmodels including active
FIGURE 2 OverviewoftheSWIO
landscape.Blackpentagonsrepresent
nestingsites:ALD,Aldabra;EUR,Europa;
IRA,Iranja;MAY,Mayotte;MOH,
Mohéli;TRO,Tromelin;VAM,Vamizi.
Sizeofnestingsiteisproportionalto
nestingnumberoffemales.Blackcrosses
representlocationsoffeedingpatches.
Arrowsindicatemajoroceaniccurrents
(Schott,Xie,&McCreary,2009):GW,
GreatWhirl.Redandbluelevelsindicate
meanannualoceaniccurrentintensities;
NEMC,NorthEquatorialMadagascar
current;SC,SomaliaCurrent;SECC,
SouthEquatorialCounterCurrent;SEMC,
SouthEquatorialMadagascarCurrent;
SG,SouthernGyre.Inthelegendofthe
figure,theacronymsdescribetheinput
datasources:UnitedNationsEnvironment
Programme—WorldConservation
MonitoringCentre(UNEP‐WCMC,Green
&Short,2003),AgulhasandSomali
CurrentLargeMarineEcosystemsProject
(ASCLME,www.asclme.org),Geostrophic
andEkmanCurrentObservatory(GECKO,
Sudreetal.,2013)
4 
|
   DALLEAU Et AL.
swimming behavior were developed. Still, movement rules re
mained f airly simple a nd consisted eit her of random m ovement
(Gasparetal.,2012;Putman,Scott,Verley,Marsh,&Hays,2012)
or movement oriented along a gradient of environment al vari
ablesormagneticfields(Putman,Verley,Shay,&Lohmann,2012).
Morerecentlydispersalaffectedbyoceaniccurrentsandhabitat
features was model ed for the western Pa cific leatherba ck tur
tle(Gaspar &Lalire,2017),aswellas the effect ofmultiplecues
on the hom ing behavior of ind ividual gree n sea turtle s (Painter
& Plochocka, 2019). These kinds of IBMs remained focused on
movementanddidnot considerdemographicprocesses suchas
survivalorreproductiveoutput.
Contrastingly,IBMswerealsousedtorepresentpopulationdy
namics of seaturtles(Mazaris,Broder,& Matsinos, 2006; Mazaris,
Fiksen, & Matsinos, 2005; Mazaris & Matsinos, 2006; Piacenza,
Richards,&Heppell,2017), but in these cases, movementwas not
explicit ly implemented . Another typ e of IBMs was used to stu dy
nestingpopulation dynamics suchas consequencesof variable re‐
migration intervals on sea turtles' nesting numbers (Hays, 2000;
Neeman, Spotila, & O'Connor, 2015) or how changes in biologi‐
cal proce sses can influ ence populatio n recovery and as sessments
(Piacenzaetal.,2017).Alsointhesemodels,movementwasalsonot
explicitlyimplemented.
Ourmodelexplicitlyrepresentsmovementofthousandsofsea
turtl es, but we do not incl ude demograp hic processes an d hence
populationdyna mics.Themainpurposeofourmod elistobetterun
derstandhowthefeaturesoftheregionallandscape,suchasnesting
sitelocations,distributionoffeedingpatches,andoceaniccurrents,
constrainthemigratoryandforagingpatternsofgreenturtlesandto
deviseimplicationsfortheconservationofthespeciesintheregion.
Weimplemented alternative foragingandnestingstrategiesacross
the entir e parameter r ange, expres sing qualitat ive strategie s from
being risk prone to risk averse. Themodelthen allowed assessing
theinfluenceandsensitivityofdifferentforagingandnestingstrat‐
egiesinconcertwithfeedingpatchdisturbanceonthereproductive
outputofrookeries.
2 | METHODS
2.1 | Life cycle of green turtles
The green turtle's life begins in the sand of the natal beach. After
emergingfromthe nests, seaturtles'hatchlings join oceanic waters
anddriftwiththecurrents(Carr,1986).Theyremaininoceanicwaters
foryearsinastageknownasoceanicjuvenilestagebeforerecruiting
inneritichabitats(Musick&Limpus,1997).Conditionsofrecruitment
and criteria of site selection remain poorlyunderstood but recruit‐
mentzonesareoftenfairlydistantfromthenatalbeach(Naro‐Maciel,
Becker,Lima, Marcovaldi,& DeSalle, 2007).Atthis stage, known as
the neritic juvenile stage, green turtle's trophic statuspermanently
changesfromomnivorytoherbivory(Musick&Limpus,1997).Itsmain
dietthenceforthconsists most generally ofseagrassesandpossibly
alsoofalgae(Bjorndal,1980).
Atsexualmaturity,seaturtlesexhibitstrongphilopatry,thatis,a
tendencytobreedintheplacetheywereborn(Brothers&Putman,
2013;Miller,1997).Adultsconsequentlymigratebackandforth to
thenatalnestingsiteseveryfewyears(generally2–4years,Troeng
&Chaloupka,2007).Thedurationbetweentworeproductivecycle,
knownasthe“remigrationinterval,”varieswithinandamongpopu‐
lations (Heithaus, 2013) and may dependonpopulation recovering
status , availabilit y of quality food , or distance to fo raging ground
(Troeng & Chaloupka, 2007). Green turtles are capital breeders,
since they d o not feed during re productio n and the reprodu ctive
cycleisbasedonstoredenergeticreserves.Atnestingsite,females
repeatedlyenter the beachshore wherethey lay eggsinthesand.
Postnestingfemalesthenmigratetoresidentneriticforagingareas.
Fordifferentseaturtles'species,asitefidelitytoforagingareas
overmultiplereproductivecycleshasbeenobserved(Limpusetal.,
1992;Marcovaldietal.,2010;Schofieldetal.,2010;Shaver&Rubio,
2008;Tucker,MacDonald,&Seminoff,2014).IntheMediterranean
sea, female green turtleshave been tracked migrating toidentical
foraginglocationsaftersuccessivenestingevents(Broderick,Coyne,
Fuller,Glen,&Godley,2007).InthePacificOcean,ataggingstudy
also demonstrated foraging sitefidelity offemalegreen turtles at
differentspatialandtemporalscale(Readetal.,2014).Nevertheless,
changeinforaging sitehasalsobeen observedsuggesting that for‐
aging site se lection is a p lastic behav ior (Hays, Hobs on, Metcalfe ,
Righton, & Sims, 2006; Marcovaldi et al., 2010; Shaver & Rubio,
2008).
Green turtlespostnestingmigrations consistof oceanic and/or
coastalmovementtopreferredforagingareaswithrelativelydirect
routes(Godleyetal.,2008).Coastalsectionsalongthewaymayaf‐
fordforagingopportunities(Cheng,2000;Godleyetal.,2002) but
coastlinesmayalsobeusedtofacilitatenavigation(Hays,Broderick,
Godley,etal., 2002).Oceaniccurrentsconstrainhomingand post
nestingmovementsbymovingindividuals awayfrom their course
andloweringtheabilitytoorientate(Cerritellietal.,2018;Cheng&
Wang,2009;Girardetal.,2006).
2.2 | Model description
WedescribethemodelfollowingtheODD(Overview,Designcon‐
cepts, and Details) protocol for individual‐based models (Grimm
etal., 2006,2010). The model wasimplemented inNetLogo 4.1.3
and released under NetLogo 5.3.1 (Railsback & Grimm, 2019;
Wilensky,1999).TheNetLogoprogramandalldatafilesrequiredto
runthemodelareavailableunderhttps://www.comses.net/codeb
ases/69863caa‐2f8e‐4412‐a564‐a2826d9d38d3/releases/1.0.0/.
2.2.1| Purpose
Theproximatepurposeofthemodelistounderstandhowthefea‐
turesoftheSWIOregionallandscape,suchasnestingsitelocations,
distributionoffeedingpatches,andoceaniccurrents,constrainthe
migratoryandforagingpatternsofgreenturtles;itsultimatepurpose
istorevealforagingandnestingsitesofhighconservationvalue.The
    
|
 5
DALLEAU E t AL.
modelimplementstheprocessesofforaging,migration,andnest
ingtostudyhowtheyaffectthereproductivepotentialofthemain
regionalrooker ies.Togof urt her,themodelalsoexp loreshowdif fer
entforaging andnestingstrategiesmayaffectreproductiveoutput
andhencepopulationsurvivalinaheterogeneouslandscape.
2.2.2 | Entities, state variables, and scales
The entities of the model are adult female green turtles, square
grid cellsforming a grid that covers theSWIO region (25°E–65°E;
30°S–10°N;Figure2),andperturbations.Theturtles'statevariables
arelocation(gridcell),currentpreferredfeedingpatch,nestingsite,
internal state (“feeding”;“prenesting,” i.e.,on theway to the nest‐
ingsite;“postnesting,”i.e.,onthewayfromthenestingtoafeeding
patch; “nesting”; “foraging‐migration,” i.e., moving between feed‐
ing patches), energy level, foraging strategy,and nesting strategy.
Eachindividualalsohasacoastavoidancedirectionthatdetermines
whetherit willavoidthecoast to theleftortotheright whenitis
encountered.Thatdirectionisreverteddependingonwhetherthe
turtleisinpre‐orpostnestingmigration(Figure3).
Gridcellsarecharacterizedbytheirlocation.Theycanbeoffour
differenttypes:terrestrial,nestingsite,feedingpatch,orjustocean.
Terrestrial cells are barriers tomovement. Nesting sites represent
mainregionalrookeries(Figure2;Table2).Theyaredispersedacross
theregionwithahigherconcentrationinthenorthwestofthemap
(northoftheMozambiqueChannel).Feedingpatches,derivedfrom
telemet ry mapping (Fi gure 2), are charac terized by their re source
levelreflectingtheavailabilityofseagrass,themainforageforgreen
sea turtles.The resource levelof eachfeedingpatch isconstantly
updated (growth or depletion) depending on the numberofturtles
feeding onit. Mostof the feeding patches occurin larger clusters
alongcontinentalshelves.
Under onesimulationscenario,turtle movementisaffected by
oceaniccurrentsderivedfromclimatologymaps:Theturtle'sveloc‐
ityvectorisresultingfromtheturtle'smotorvelocityvectorplusthe
oceaniccurrentvelocityvectoratturtlelocation.Oceancurrentsare
representedvia colorcoding of oceanic gridcells, in theRGB(red,
green,blue)tuple:Theredandbluecomponentswereusedtorep‐
resent,respectively,theeastward andthe northwardcomponents
ofthe seasurfacecurrents. Feeding patches are possibly exposed
toperturbationsthataltertheirproductivity.Perturbationsarerep
resented byalatitudecoordinateandaspatialrangeofaction.The
growth r ates of feeding p atches locate d within the pe rturbat ions'
spatialrangearediminishedwiththeamountofreductiondepend‐
ingonthefeedingpatch'sdistancetotheperturbation'slatitude.
Each simulation lasts for approximately 50 years (36,500 time
steps). Thefirsttwo years (1,500 timessteps)are considered asa
burn‐inperiodwherenomodeloutputisrecorded.Gridcelldimen
sionisapproximately 7 × 7 km; theentire modelworld consists of
567×577gridcells,correspondingto3,969×4,039km.
2.2.3 | Process overview and scheduling
Ateachtimes tep,whichcorrespondstohalfaday,firstallgreentur
tlesandthenallfeedingpatchesareprocessed,bothinrandomized
order and withimmediate updating oftheir state variables. In the
following,thenamesofsubmodels,whicharedescribedindetailin
theODDelement“Submodels,”aregiveninparentheses.
The taskagreenturtle has toperform depends onits internal
state:Iftheinternalstateis“feeding,”itfeeds(win‐energy)andthen
possibl y switches its inte rnal state to “for aging‐migration” (fo rag‐
ing‐migration‐start)whichincludesselectinganotherfeedingpatch
(allocate‐new‐feeding‐patch), or possibly switches to “prenesting”
(prenesting‐migration‐start); if the internal state is “prenesting,”
the tur tle moves toward the n esting site (move‐o ne‐step‐toward)
if it is still out side the detect ion range of the ne sting site, othe r
wisetheinternalstate switchesto“nesting”;iftheinternalstateis
“postnesting,”itmoves toward itscurrentpreferred feeding patch
(move‐one‐step‐toward)if it isstill outside thedetection range of
thefeedingpatch,otherwisethestateswitches to “feeding”;ifthe
internal stateis“nesting,”theturtlenests(nests),whichincludesa
possibleswitchtothes tate“postn esting”;iftheinternalst ateis“for
aging‐migration,” the turtle movesbetweenfeedingpatchesinthe
samewayitmovesonitswaytowardandbackfromitsnestingsite
(move‐one‐step‐toward).
Ateachtimestep,theturtles'energylevelisupdatedbyeither
gainingenergywhilefeedingor losingenergy whilenestingormi
grating.Individualactionsrely on twodecisionstrategies:foraging
FIGURE 3 Schematicrepresentationofcoastavoidancetrajectories.Directionofcoastavoidanceisdeterminedduringfirstprenesting
migration(alternativelyforagingmigration)byprioritizingtheleastturningangle(totheleft,α1,whichissmallerthantotheright,α2,inthis
example).Duringpostnestingmigrationindividualwillavoidthecoastbyturningintheoppositedirectioncomparedtoprenestingmigration
(totherightinthisexample).Anindividualstopsfollowingthecoastwhenitisabletomovewithoutobstacleinthedirectionofthetarget.
Thismaypossiblyleadtodifferenttrajectoriesduringprenestingandpostnestingmigration
6 
|
   DALLEAU Et AL.
strateg yandnestingallocationstrategy.Theforagingstrategyspec
ifieswhetherandwhenaturtleleavesitsfeedingpatchforanother
onedependingontheresourceleveloftheactualfeedingpatch.The
nestingstrategycontrolstheamountofinternalenergyinvestedat
eachnestingevent.Wemodeledtherangeofpossiblestrategiesin
both proce sses, by a singl e index rangin g from 0 to 1. A foragi ng
patchfidelitystrategycloseto1leadstoa“stayerstrategy”whilea
foragingpatchfidelitystrategySFcloseto0leadstoa“moverstrat
egy”(Figure4).Anestingstrategycloseto1leadstoan“investment
strategy”whileanestingstrategycloseto0leadstoa“conservative
strategy.”Weran sets of simulations withvariouscombinations of
foragingandnestingstrategies.
Movement is represented as direct movement toward a se
lected site, which is modified when barriers (islands, mainland)
are encounteredand possibly bypassive drift due to oceanic cur‐
rents. Movement is energetically costly, so that swimming be
tween fo raging patches or fo raging furth er from the nest ing site
hastobebalancedbyagaininforagingconditions.Forthefeeding
patches,growth, depletionbyturtles,andpossiblyperturbationof
the amount of seagrass are considered (seagrass‐stock‐regrowth;
Figure 5). Perturbation represents potential natural or anthropo
genicimpacts(e.g.,climatechange,habitatdestruction,oilspill);its
strengthdependsonlatituderelativetotheperturbation'slocation.
Feedingpatchesthatarenotwithinthespatialrangeofactionofthe
perturbationarenotaffected.
Finally,plotsandfileoutputsareupdated.Outputanalysescom
prisedspatialforagingandmigratingpatternaswellasreproductive
outputatthepopulationscaleinresponsetotheturtle'sstrategies.
Itshould benoted thatthe modeldid not include mortality or the
turtles'lifecycles;calculationofthepopulation'sreproductiveout‐
putcalculationwasbasedonthenumberofnestingeventsandthe
energyindividualsinvestedintoeggswhennesting.
Figure6summarizestheprocessesasimplementedinthemodel.
Figure 7depictsthe categories ofbehavioralstrategies.Modelpa‐
rameter s are specifie d in Table 1. Whe n possible, the mo del was
parameterizedwith field data. Otherwise,parametersweredeter
minedbyinversemodelfittingtothemostrealisticandbiologically
relevantobservations.
FIGURE 4 Foragingpatchfidelitystrategiesandtheirfunctionalrelationships.ThisfigureillustratestheprobabilityPleave,tforaturtleito
leaveapatchpdependingonitsforagingpatchfidelitystrategySF,tandpatchresourcelevelΦp,t. The x‐axisrepresentstheresourcelevelΦp,t
ofthepatchp. The y‐axisisthelevelofprobabilityPleave,tofleavingthepatchattimet.EachcurvedepictstheprobabilityPleave,tofleaving
thepatchdependingonactuallevelofpatchresource.TurtleforagingfidelitypatchstrategySF,tisfixedacrossasinglesimulation.Aforaging
patchfidelitystrategyclosedto0(highercurves)leadstoanoverallhigherprobabilitytoleavethepatch(moverstrategy).Astrategyclosed
to1(lowercurves)leadstoanoverallsmallerprobabilityofleavingthepatch(stayerstrategy)
FIGURE 5 TemporaldevelopmentofpatchresourcelevelΦp,tasafunctionoftimetandnumberofseaturtlefeedingonpatchNp.
The y‐axisrepresentstheresourcelevelΦp,tofthepatchp. The x‐axisrepresentsthetimet.Eachcurvedescribeshowtheresourcelevel
Φp,tevolvesdependingonthenumberofturtlesNp.Duringsimulations,theresourcelevelofapatchisnotlikelytoevolvesmoothlyas
suggestedbythesecurvesasthenumberofturtles'feedingonthepatchmaychangebetweentimesteps
    
|
 7
DALLEAU E t AL.
2.2.4 | Design concepts
Basic principles
Weassumethatturtleshaveaspatialmemoryoftheirpreferredfeed
ingpatchandtheirnestingsite.Abasicenergybudgetofenergygains
duringfeedingandlossesduringmigrationandeggproductiondeter
minesmigrationpatterns,reproductiveoutput,andreturnintervalsto
thenestingsites.Preferredfeedingpatcheswillbeleftinthesearchof
betterpatchesiffeedingefficiencyfallsbelowacertainthreshold;this
can happenbecause too manyturtles are feeding onthis patchorif
regrowthoftheforage,seagrass,isslowduetoperturbations.
Emergence
Foraging (stayer or mover) or nesting (investment or conservative)
strategiesdirectlydeterminerookeryreproductiveoutputviaindivid
ualbehavior.Intuitively,thebestindividualstrategywouldbetofeed
onfeedingpatchesclosetotherookery,thusreducingthecostofmi
gration.However,withconspecificsdepletingtheclosepatches,differ
entstrategiesmightbebeneficial.Therookeries'reproductiveoutputs
consequently emergedf romindividual behavior while searching for
patchesanddecidingonnestingenergyallocation.Furthermore,the
timeintervalbetweeneverybreedingeventemergedfromenergetic
constraints, as well as the distribution of the spatial feeding patch
usagethatwecouldcomparewithtrackingdatafromfieldsurveys.
Sensing
Atany time step, a migrating turtlecould assess the direction ofthe
migrationtarget(itsfeedingpatchoritsnestingsite)andhastheability
toheadtowardit.Inaddition,aturtlecouldsenseandavoidanycoastal
arealocatedwithin 100 km ofits actual location. Turtlesdidnothave
theabilitytosenseoranticipatetheoceaniccurrents.Turtlesperceived
theresourcelevelofthefeedingpatchwhe retheywerefee dingon.Th e
decisiontoleavethefeedingpatchwastakeninresponsetothislevel.
Interaction
There wasno direct interaction between individualsin themodel.
However,indirectinteractionbetweenindividualswasincludedin
directlyviaresourcecompetitionatfeedingpatches.
FIGURE 6 Flowchartofthemodel'sprocesses.(a)Flowchartoftheturtles'nesting‐migration‐foragingcycleshowingthetransitions
betweeninternalstates.(b)Flowchartoftheprocesseddeterminingtheresourceleveloffeedingpatches
8 
|
   DALLEAU Et AL.
Stochasticity
Initialfeedingpatchesareassignedrandomlyaccordingtodecreasing
exponentialprobabilityfunctionofthedistancetothenestingsite.The
initialspatial distributionofthe turtleonfeedingpatchesistherefore
variablebetweensimulations although itisimpacted by the regional
landscape.Duringthecourseofthesimulation,foragingbehavioralso
leadsto temporal and spatialstochasticity.The decisionofleavinga
feedingpatch for another is aprobabilityfunction that relies onthe
foragingstrategyandontheresourcelevelofthefeedingpatch.Thus,
individuals,althoughtheysharethesameforagingstrategyforagiven
simulation,willnotleavethefeedingpatchsimultaneously.Someindi
vidualswillrandomlyleavethepatch earlier,thereforecausing other
individualstoremaininthepatch.Furthermore,thechoiceofthenew
feedingpatchisalsoadecreasingexponentialfunctionofthedistance
tothepatchthatisleft.Turtlesleavingagivenpatchwillnottravelto
thesamefeedingpatchaffectingtheoccupationofthefeedingpatches.
Thestochas ticit yhereisimp lementedtoreflectsourcesofv aria
tionsthatmayactuallyoccurduringforagingphases.Stochasticityin
turtle'sdistributionoverthefeedingpatcheswillaffectspatialusage
of the ocean ic areas as migrat ory corrido rs but also repr oductive
outputofnestingsites.Overnumeroussimulations,wemayidentify
areasthat are ofinterestforfeedingormigration,despite possible
sourcesofrandomvariationsinspatialbehavior.Ontheotherhand,
wemayalsoidentifyrobusttendenciesinreproductiveoutputvari
ationsbetweenrookeries.
Observation
Focusing on model purposes, model outcomes comprised spatial
foragin g and migration pat tern as well as reprod uctive output at
thepopulationscaleinresponsetotheturtle'sstrategies.Tostudy
foragingand migrationpatterns,werespectivelymeasuredfeeding
patch usage and mapped corresponding migration pathways. For
this,wepooledforeachenvironmentalscenariotheresultsfromall
combinationsofthetwobehavioralstrategies.Wefurtherobserved
theremigrationinter valaswellasenergystoragefromwhichwede
ducedareproductiveoutputatrookeries.Thiswasdoneseparately
foreachbehavioralstrategy.
Westudiedspatialpatterns ofthreeforagingstatistics:(a)time
usage,thatis,thesum,overalltimesteps,ofthenumberofturtles
present ona feeding patch at each time step,(b) number of post‐
nesting visits, that is, the number oftimesthat aturtle arrived in
afeeding patchfollowing postnestingmigration, and (c)numberof
foragingvisits,thatis,thenumberoftimesthataturtlearrivedina
feedingpatchfollowingforagingmigration.
Inaddition,wealsostudiedtheforagingpatternsinrelationtothe
preferrednestingsitesoftheforagingturtles.Forthis,wecomputed
two additional metrics: (a) the number of nesting sites from which
nestersoriginatedinagivenfeedingpatchand(b)thediversityindex
ofnestingsitesfromwhichnestersoriginatedinagivenfeedingpatch.
Diversity index calculation HP is derived from Shannon's diversity
indexbasedonthenumberofpostnestingvisits:
withrp,nistherelativeproportionofpostnestingvisitsofpatchp
byturtlesfromnestingsiten,andNNisthenumberofnestingsites
presentinthemodel.
Turtle'sprenestingandpostnestingmigrationswererecordedby
randomlysampling individual's locations approximately every 500
timesteps.Foragingmigrationswerenotrecorded.Migrationpath‐
wayswerethenstudiedusingkernelmethodsfordensityestimation
onsampledlocations(Worton,1995).
Onlyth es ixmainnes tingsites(Europa,Aldabr a,Mayotte,Moli,
Tromelin,Glorieuses;seeTable2forcorrespondingreferences)were
consider ed in the study of th e reproducti ve parameters. Fo r each
nesting site, the three following statistic s were computed: (a) the
meanindividualremigrationinterval defined as themeanduration
betweensuccessivene stingph asesperea chindivid ual(FigureS2);(b)
themeanindividualenergylevelatnestingdefinedasthemeanen
ergylevelofturtlesafterthenestingevent;(c)therookeryoverallre
productiveoutputwhichwascalculatedasafunctionofthenumber
ofnests,theremigrationintervals,andtheenergylevelatnesting.
Tocompute thesestatistics, at each time ka turtle inested at
nestingsiten, we recordedthedate Ti,kandthecorrespondingen
ergylevelafternestingεi,k.Wecomputedtheremigrationintervalas
thetimedifferencesincethepreviousnestingevent,Ti,k,−Ti,k1.We
computedtheoverallreproductiveoutputROnofeachnesting site
nasdirectlyproportionaltotheenergylevelsatnestingεi,kandthe
nestinginvestmentSN:
(1)
H
p=ΣNNrp,n×ln
(
rp,n
)
ln
(
N
N)
(2)
ROn
i
Σ
k
𝜀
i,k
(
T
i,k
T
i,k1)
S
N
FIGURE 7 Categoriesofbehavioralstrategies.Thex‐axis
representsthenestingallocationstrategy.They‐axisrepresents
theforagingpatchfidelitystrategy.Nestingandforagingstrategy
areconstrainedbetween0and1.Anestingallocationstrategy
closedto0leadstowarda“conservative”tendencyandanesting
allocationstrategyclosedto0toan“investment”tendency.A
foragingpatchfidelitystrategyclosedto0tendstowarda“mover
strategy”andaforagingpatchfidelitystrategyclosedto1towarda
“stayerstrategy”
    
|
 9
DALLEAU E t AL.
2.2.5 | Initialization
Thelandscape,inparticularthenumberandlocationofnestingand
feedingpatches,remainedidenticalwithinandbetweensimulations
andwastakenfrominputmaps.Initialresourcelevelofthefeeding
patchwaseithersettoarandompositivevaluesampledfromauni
formdistributionbetweenzeroandmaximumresourcelevelΦmaxor,
ifnodepletionbyturtleswasconsidered,toΦmax.
Mostsimulationswererunwith 7,000 turtles.At the beginning
ofeachsimulation,theturtles'nestingsiteswereallocatedrandomly
TABLE 1 Modelparametersandvariables.Followingvaluesareexpressedindailyunits.Timestepcorrectionwastakenintoaccount
directlyinmodelimplementation
(a) State variables
Turtle NetLogo variable Abbreviation Default value
Location xcor,ycor x,yVariable
Preferrednestingsite gt‐nesting‐site N01–14
Initialfeedingpatch gt‐feeding‐patch F01–47
Currentfeedingpatch gt‐feeding‐patch Fi1–47
Internalstate(prenesting,nesting,postnesting,foraging) gt‐internal‐state
Internalenergylevelattimestept energy‐level εi,tVariable
Coastavoidanceside(leftorright) gt‐avoidance‐side −1(left)or1(right)
Feedingpatch
Location xcor,ycor px,py
Feedingpatchpresourcelevelattimet feeding‐patch‐resource‐level Φp,tVariable
(b) Parameters
World NetLogo variable Abbreviation Default value
Numberofturtles N‐GTURTLES NT7,000
Numberoffeedingpatches N‐FEEDING‐PATCHES NF47
Numberofnestingsites N‐NESTING‐SITES NN14
Perturbationlatitude perturbation‐latitude σy−26° S
Perturbationrange perturbation‐range dσ,max 1,000km
Perturbationintensity perturbation‐intensity σi0.1
Feedingpatchallocationexponent feeding‐patch‐allocation‐exponent λ20
Tur t le
Foragingpatchfidelitystrategy foraging‐fidelity‐strategy SF(0.2,0.4,0.6,0.8)
Nestingallocationstrategy nesting‐allocation‐strategy SN(0.2,0.4,0.6,0.8)
Prenestingthreshold prenesting‐threshold εcycle Variable
Migrationspeed migration‐speed c65km/day
Energylosspermovementstepduringmigration energy‐loss‐migration Δεi,m−1
Energylosspernestingday energy‐loss‐nesting Δεi,n−5
Energygainpertimestepwhilefeeding energy‐gain‐feeding Δεi,p,tVariable
Proportionofintakefromeachpatch intake‐proportion α0.0001
Distancetonestingsite distance‐from‐nesting‐site di,nVariable
Maximumnumberofnestingduration max‐nesting‐duration Tn,max 45days
Feedingpatch
Maximumfeedingpatchresourcelevel maximum‐feeding‐patch‐resource‐level Φmax 1,000
Initialfeedingpatchresourcelevel feeding‐patch‐resource‐level Φ0[01,000]
Regrowthrateoffeedingpatch patch‐regrowth‐rate β2 · α · NT/NF
Slopeofreactiontowardpatchleavingdecision feeding‐patch‐leaving‐slope a100
Thresholdofreactiontowardpatchresourcedepletionlevel
εp
feeding‐patch‐leaving‐threshold b500
Latitudinaldistancetoperturbation distance‐to‐perturbation dp,σVariable
10 
|
   DALLEAU Et AL.
TABLE 2 Estimatednumberofnestingfemalesforeachnestingsitesasavailableintheliteratureandrelativeproportionofnestingfemalesassignedtoeachnestingsiteinthemodel.
Absolutenumberoffemalesactuallyassignedtoeachnestingsiteinthemodelwascalculated,inrespectoftherelativeproportionindicatedhere,ensuringthatthetotalnumberof
individualsinthemodelequalsNT(7,000individuals).Thisdataneedtobetakenwithcautionandneedtobejustifiedbythepaperscited.Comparisonsneedtobedonewithcautionbecause
estimationmethodsvaryforallsites.(*Majornestingsites)
Site Tri gra m
Area of
estimation
Estimated number of fe
males per year Sources
Adjuste d number of fe‐
males per year
Relative proportion of females
assigned in the model
Europa* EUR All 6,00011,000 LeGall,Bosc,Château,and
Taquet(1986)
11,000 10,000
Aldabra* ALD All 3,000–5,000 Mortimer,vonBrandis,etal.
(2011)
5,000 5,000
Mayotte* MAY All 3,000–5,000 Bourjea,Frappier,etal.(20 07) 5,000 5,000
Mohéli* MOH 6beaches 4,410 Bourjea,Dalleau,etal.(2015) 5,000 5,000
Tromelin* TRO All 1,430 Lauret‐Stepleretal.(2007) 2,000 2,000
Glorieuses* GLO 60% 1,480 Lauret‐Stepleretal.(2007) 2,000 2,000
Tanzania TAN All 12 0 –1 5 0 Muir(2005) 150 200
Iranja IRA All 1 0 0 –15 0 Bourjeaetal.(2006) 150 200
JuandeNova JUA All <80 Lauret‐Stepler,Ciccione,and
Bourjea(2010)
70 200
Seychelles(Except
Aldabra)
SEY 13–24 Mortimer,Camille,etal.(2011) 150 200
Mozambique(Vamizi
Island)
VAM 85% 50 Garnieretal.(2012) 60 200
LaRéunion RUN All <5 CiccioneandBourjea(2006) 520 0
Kenya KEN Unknown Okemwaetal.(20 04) Unknown 200
Mauritius(Chagos
Archipelago)
CHA – – 200
    
|
 11
DALLEAU E t AL.
withthe constraintofensuringthatrealistic proportionswere dis‐
tributedover the nestingsites;thatis,thedistributionacrossnest‐
ingsitesfollowstheknownsizeofthenestingpopulation(Table2).
Theinitialfeedingpatchwasalsoassignedrandomlyassumingthat
the probability of afeeding patch to be assigned to aturtle isin‐
versely p roportion al to the distance sepa rating this site fr om the
turtle'snestingsite.Thatis,theinitialdistributionoveracrossfeed‐
ingpatchesfollowsaninverseexponentialdistancefromthenesting
site.This probabilitywascalculated in thesame way as thechoice
ofanewfeedingpatch duringforaging(procedure “allocates‐new‐
feeding‐patch”). The preferred feeding patch may change during
simulationsdependingonitsquality.
Atinitialization, all turtleshave theinternal state“feeding” and
arereleasedatthelocationoftheirfeedingpatch.Theinitialinternal
energylevelε0 is randomlyattributed by sampling from a positive
uniform distributionbetween0andthetotal energyrequiredfora
wholenestingcycle.
2.2.6 | Input data
Maininputsforthemodelarethefunctionalhabitatmap(rookeries
mapsfornestingsites andseagrassesforfeedingpatches)andthe
mapofoceaniccurrents.
Rookeries
Rookery locations are mapped from local knowledge and using
the latest available estimates of the number of annual nesting fe‐
males (respective studies used are cited in Table2). We are using
theupperlimitfieldestimationofnestingfemale'snumbertocom‐
putethe proportion of individualsassociatedwith eachrookeryin
themodel.Theproportionofindividualsassignedtoeachrookeryis
showninTable2.Aminimumof45turtlesisallocatedtothesmall‐
estrookeries.
Feeding patches
Locationsoffeeding patcheswereset up bycombiningmapsfrom
twodistinctsources:theWorldAtlasofSeagrasses(Green&Short,
2003)andtheAgulhasandSomaliCurrentLargeMarineEcosystem
project (ASCLME; ww w.asclme .org). Mapped sea grass beds were
tr a nsfo rmedintofee d ingpatc hes( gridcell s)atl o c ati o nsc orre spond
ingtothelocationofthemainmappedseagrassesbeds.Additional
feedingpatcheswereaddedalongthecoastofSomaliaasthisplace
isknowntohostvastareasofseagrassbedthatarenotmapped in
theciteddatasets(S.Andréfouët,personalcommunication).
Oceanic currents
Tomodeloceaniccurrents,weareusinganannualclimatologymap
that refle cts the mean cur rent velocities in the region. This map
wascomputedbycombiningGEKCOsurfacecurrentdailydatasets
(Sudre, Maes, & Garçon, 2013).Wedid not consider any seasonal
effec t at this stage. To repre sent the 2D curre nts vector ma ps in
themodel,intheRGB(red,green,blue)tuplethatisusedtoencode
colors in N etLogo, the gre en component w as left at zero and t he
values of theredand bluecomponentwereused to represent,re‐
spectively,theeastwardandthenorthwardcomponentsofthesea
surfacecurrents(Figure2).
2.2.7 | Submodels
Winenergy
Whenattimetturtleifeedsonpatchp,itsinternalenergylevelεi,t
isincreased:
withΔεi,p,tbeingthenetgainfrompatchp attimet.Wedonot
explicitlyconsidermetabolic costsfor maintenance as thiswasas‐
sumedaconstantvariableindependentfrominternalstate.Thenet
gainpertimestepΔεi,p,tdepe ndsontheresou rcel evelofthefee ding
patchΦp,t:
withαbeingthedepletioncoefficient.
Foragingmigrationstart
TheprobabilityPleave,i for turtle i to leavetheactualfeedingpatch
foranotheronedependson the resource levelofthe actual patch
Φp,tandonitsownforagingpatchfidelitystrategySF.Thefunctional
relationshipwasmodeledwithalogisticcurve:
where a modulates the steepness of the reaction and b isthe
leaving th reshold. A for aging patch fidel ity strateg y SF close to 1
leadstoa“stayerstrategy.”AforagingpatchfidelitystrategySFclose
to0 leads to a“mover strategy” (Figure 7).Values forparameters
aand b aregiven inTable1. The resulting probability of leaving a
feedingpatchdependingonforagingstrategySFandfeedingpatch
resourcelevelΦp,tis illustratedin Figure 4. This submodel neither
takesintoaccounttravelcostsnorleaving the patchwhenitsfood
intake dro ps below the average f ood intake on all oth er patches,
sincetheenergeticcosttoanotherfeedingpatchthattheturtlehas
nevervisitedshouldbeunknowntoaturtle. Similarly,a turtleona
patchhasnoknowledgeofthepotentialleveloffoodintakeitcould
getfrom otherpatchesasithas to belocatedonapatch to know
thatlevel.
Therefore,thecostofforagingexplorationwillemergefromthe
model.
Allocatenewfeedingpatch
Theselectionofanewfeedingpatchwasdistance‐dependentwith
selectionprobability Pselection determined by an exponential decay
function:
(3)
Ei,t+1=
𝜀
i,t
𝜀
i,p,t
(4)
Δ𝜀i,p,t=𝛼
Φp,t
(5)
Pleave,i
=
(
11
(
1+exp
((
Φ
p,t
+b
)
a
)))
1000
(
1S
F)
(6)
Pselection
=
(
1d
relative)𝜆
12 
|
   DALLEAU Et AL.
where drelative = d−dmin/dmax−dminiscalculatedfromd,thedis‐
tancebetweenanewfeedingpatchandthecurrentfeedingpatch,
and dmin and dmax, th e minimum and maxim um distance bet ween
feedingpatches.λisanarbitraryexponentialdecaycoefficient.This
model assumes thatchoice of a new feeding patchisbased rather
ontheturtles'betterknowledgeof thelocationoffeedingpatches
nearbythanbythosefeedingpatchresourcelevels,whichtheycan
not know. The mi nimum and maxi mum possible d istances ar e not
know ntothetu r tle sbu tusedtoscal ethes pat ialsca leofknowledge .
Moveonesteptowardwith/withoutcurrents
At each time step and for each turtle in migration, spatial loca‐
tionwasupdatedwithafixed speedof2.7km/hr(65km/day)and
a heading toward the se lected patch when not facing t he coast.
Speedvaluewasderivedfrominsitusatellitetrackingmeasurement
(Dalleau,2013). Effectivetraveling speedand directionmay,how‐
ever,beimpactedbyoceaniccurrentsattheturtle'slocation.
During prenesting, pos tnesting or foraging migr ation, at each
timesteptaturtleimovestowardaselectedpatchp,itlosesafixed
amountofenergyΔεi,m(Table1):
Ateachtimestept,aturtle iattemptstomoveonestepinthe
directionofthetarget,whichiseitheritsnestingsiteinthecaseof
prenesting migration oritscurrent preferred feedingpatchin the
caseofpostnestingorforaging‐migration.
Foravoidanceofcoastalgrounds,weimplementedasimplewall‐
followingalgorithm(Figure3).Atagiventimestep,ifmovingaturtle
forwa rd causes this t urtle to enco unter a coasta l grid cell (patch‐
ahead‐is‐coast?),it s swimming direction is modified incrementally
(angle‐step)uptothe minimumangle thatallows to move forward
without encounteringaterrestrial gridcell (see nextparagraph re
gardingthedirectionofrotation).Theturtlethenmovesforward.At
thefollowingtime step,ifpossible, the swimming directionisfirst
modifiedincrementally(angle‐step)toadirectionclosertothedirec
tionofthe target(thefeedingpatchorthenestingsite)thatallows
movingforwardwithoutencounteringagridcell.Ifthedirectionof
thetargetcanbereached,theswimmingdirectionoftheturtleisset
tothetarget'sdirection.Contrarily,iftheswimmingdirectioncannot
bemodifiedandiftheturtlecannotmovesforward,thentheswim
mingdirectionisonceagainmodifiedincrementally(angle‐step)by
theminimumanglethatallowstomoveforwardwithoutencounter‐
ingaterrestrialgridcell.Atthenexttimestep,thesame processis
repeated.Thisalgorithm leadstheturtletofollow the coastuntilit
canfreelymoveinthedirectionofthetargetonceagain.
Regardingtherotationdirection(totheleftortotheright),the
first t ime that a tur tle encounter s a coast, it co rresponds to t he
directionthatleadstotheleastturninganglerequiredtoavoidthe
coast. Therotation directionis thenmemorized(gt‐avoidance‐ro
tation‐direction) and willremain thesame during theduration of
agivenmigration.Nevertheless,therotationdirectionisreverted
wh e n atur t lestar t s apre‐o r apostnestingm i g r ation.T h i srever sion
is implem ented to favor, at least pa rtially, symm etrical migr ation
trajectories between pre‐ and postnesting migration (Figure 3).
With that we ensure thatan equivalent routeis followed onthe
waytoand the waybackfrom thenestingsite. In otherterms,if
theturtlefollowedthecoasttotherightgoingtothenestingsite,
itwillfollowittotheleftonthewayback. Additionally,the rota
tiondirectionis also reseteachtimethataturtlestartsandstops
aforaging migrationsince thesemigrationsareindependentfrom
nestingmigrationsandsincetheymodifythecurrentfeedingpatch
oftheturtle.
Incasetheeffectofoceaniccurrentsonmovementisconsidered,
migrationdirectionismodifiedaccordingtotheoceaniccurrentve
locityat actualturtleposition.Thefinal velocityvectorisresulting
from theturtle's motor velocity vector toward the target plus the
oceaniccurrentvelocity vectoratturtlelocation.Computationally,
this is simp ly implement ed by artifi cially displa cing the targe t site
(feedingpatchornestingsite)ateachtimestep.The“artificial”tar‐
getsite(x′,y′)islocated atthelocationoftheturtle(x,y)towhich
weadded thevectorsum of thevelocity vector in theabsenceof
current(dx,dy)andthecurrentvelocit yvectors(xc,yc).Itwascalcu
latedasfollows:
Thealgorithmstomoveonestepforwardandtoavoidthecoastal
groundsarethensimilarthanintheabsenceofcurrents.
Prenestingmigrationstart
The decision to start prenesting migration depends on the esti‐
matedlevelofenergynecessarytocompletetheentirenestingpro
cess,thatis,theturtlesstopfeedingonlyiftheygainedasufficient
amountofenergytocompletearound‐tripmigrationtothenesting
siteand nestingaction. A turtle therefore startsprenesting migra‐
tion(fromitscurrentfeedingpatchtoitsnestingsite)whenitsen‐
ergylevelεi,treachesapproximatelythetotalenergylevelneededto
completethecycle,εcycle:
where Δεi,mistheenergylostoneachtimeduringmigration,di,n
thedistancefromthecurrentfeedingpatchtothenestingsite,and
cmigrationvelocity.
Nests
Dependingonthe nesting strategyconsidered,an individualcould
either invest a large amount of energy into nesting (“investment
strategy”—thebigspender),therebytradingoffbetweenhighnest‐
ing investment and low nesting frequency (Figure 7). This might
possiblyresultinlargeintervalsbetweennesting,therebyreducing
fitnesswhenconsideredoverlifetimeaverage.Alternatively,anin
dividua l could invest onl y a limited frac tion of energ y for nesting
(“conservativestrategy”—banksaver),thereby reducingthenesting
investmentwithlowernumbersofeggsproducedbutshorteningthe
intervalbetweennestingphases.
(7)
𝜀i,t+1=𝜀i,t−Δ𝜀i,m
(8a)
x=x+dx+xc
(8b)
(9)
𝜀cycle
=
2
𝜀
migration
+𝜀
nesting
=
2
Δ𝜀
i,m
di,n
c
+
SN
Tn, max
Δ𝜀
i,n
    
|
 13
DALLEAU E t AL.
Thenumberoftimestepsspentatnestingsitesdependsonthe
valueoftheparametercharacterizingthenestingstrategySN:
During nesting,ateach timestep t spentata nesting site i, an
individuallosesΔεi,n:
AnestingstrategySNcloseto1leadstoan“investmentstrategy.”
AnestingstrategySNclose to 0 leadstoa“conservative strategy.”
After completing thenestingevent, the turtle goesback to its last
preferredfeedingpatch.
Seagrass stockregrowth
Weconsideredregrowthof seagrassfeedingpatchesbasedona
logisticfunction(Figure5).Uptakeresourcesbyturtleswasden
sity‐dependent (see Bjorndal, Bolten, & Chaloupka, 2000, e.g.,
ofinsitudensity‐dependence);thatisthe individualuptakeper
timestepdecreasedasthenumberofturtlesactuallyforagingon
thepatchincreased.Depending onits foragingstrategy,aturtle
couldtoleratea low patch resourcelevelandavoidcostlyforag
ingmigration(“stayer”tendency)orcouldratherleaveafeeding
patch when itsresource levelis too low(“mover” tendency). At
eachtimestept,theresource level Φp,tofthefeedingpatchpis
updated:
where ΔΦp,tisthenetgrowthofpatchpattimetwhichdepends
ondepletionbyNp,t turtles foraging on this patchattimetandre
growthaccordingtoalogisticgrowthmodel:
where α is thedepletioncoefficient.Thecoefficientβwasad
justed to (a) mai ntain the amo unt of resources r elatively cons tant
acrossthesimulation;(b)makethelong‐termaverageresourcelevel
beingabouthalfofthemaximumresourcelevelcommontoallfeed
ingpatches, this levelwas chosenarbitrarilybutwasshared across
allsimulations;and (c)assumingthat the turtlesareevenly distrib‐
utedoverthefeedingpatches.
Mathematically,thismeansforallpatchesp:
whichgivesthefollowing:
Thedevelopmentof the resource level Φp,t ofa feeding patch
dependingonthenumberofturtlesNp,tforagingonitis illustrated
inFigure5.
Perturbation
Perturbationisdefinedbyalatitudepositionσy,anintensitylevelσi,
anda maximum rangeofactiondσ,max.The impact ofperturbation
onagivenfeedingpatchdependsonits relativelatitudepytoper
turbationlatitudeσy.Perturbationeffectonfeedingpatchresource
levelisinverselyproportionaltothelatitudinaldistancedp,σfromthe
perturbationlatitudepositionσyandisalsod ependsont hereg row th
rateofafeedingpatch.Ateachtimestep,thepatchresourcelevelis
perturbedasfollow:
That is, ifthe feeding patch is within theperturbation range
(dp,σ < dσ,max),thepatchresourcelevelforthenext step(Φp,t+1)is
diminished by a certain delta (ΔΦp,t). Equ ation 15b details how
this delta is calculated: The diminishing delta is proportional to
theperturbationintensity(σi),therelativelatitudetothepertur
bationlocation(dσ,max/dp,σ).Itisalsoafractionoftheactualpatch
resourcelevel(Φp,t).Thecoefficientβiscalculatedtoensureasuf
ficient“global”energylevelinthesystem(seepreviousparagraph
Equation14),with
Equation 15ciscorrecting latitude effectsand showsthe rela‐
tivelatitude of thepatch to thelatitudeofthe perturbation.Note
thatthiswillremainpositiveasperturbationlatitudeissouthofthe
southernsite.
2.3 | Simulation experiments
When possible, the model was parameterized with field data.
Coastlineswere simplifiedfromGeneralBathymetricChartofthe
Ocean (GEBCO) gridded global bathymetry data (www.gebco.net).
Rookery locationsare mapped usingthe latest available estimates
ofannualnestingfemalenumbers (Table2).Oceaniccurrentswere
derived fromclimatology maps(Sudre et al.,2013).Averageswim‐
ming speedduringmigration (65 km/day) was derived fromin situ
satellitetrackingmeasurementonfemalegreenturtlesintheregion
(Dalleau,2013).
Locationsoffeedingpatchesweresetupbycombiningmaps
fromtwo distinctsources: theWorldAtlas of Seagrasses(Green
&Short,2003)andtheAgulhasandSomaliCurrentLargeMarine
Ecosystem project (ASCLME; www.asclme.org). Mapped sea
grass beds were transformed intofeeding patches (grid cells)at
locationscorrespondingto the location of themain mapped sea
grasse s beds. Addition al feeding patche s were added alon g the
(10)
Tn,i
=
SN
Tn, max
(11)
𝜀i,t+1=𝜀i,t𝜀i,n
(12a)
Φp,t+1
p,tΦp,t
(12b)
ΔΦp,t
=𝛽Φ
p,t(
1−Φ
p,t
∕Φ
max)
𝛼N
p,t
Φ
p,t
(13a)
ΔΦp,t0
(13b)
Φ≈Φ
max2
(13c)
Np,t=NTNF
(14)
𝛽
=𝛼N
T
N
F
Φ
max
(
Φ
max
−Φ
max
2
)
=2𝛼N
T
N
F
(15a)
if dp,𝜎
<
d𝜎, max:
Φ
p,t+1
p,t
−Δ
Φp,t
(15b)
ΔΦp,t=
𝜎
i
𝛽
d𝜎, maxdp,𝜎
Φp,t
(15c)
dp,𝜎=py𝜎y
14 
|
   DALLEAU Et AL.
coastofSomaliaasthisplaceisknowntohostvastareasofsea
grassbedthata ren otmapp edint hec ite ddatas ets(S.Andr éfouët ,
pers. communication).
Otherwise,parametersweredeterminedbyinversemodelfitting
tothemostrealistic andbiologicallyrelevantobservations.Forour
simpleenergybudgetmodel,weassumedthatseaturtles'reproduc
tiveactivitiesareconsiderablymoreenergeticallycostlythanswim
mingorforaging(Williard,2013).Here,costofnestingcomparedto
otheractivitieswascalibratedbyaimingforaremigrationintervalin
themodel(timeintervalbetweenindividualnestingseasons)ranging
between2and7yearsacrossallsimulations.Thesevaluesmatchthe
range obs erved worldw ide (Troeng & Chaloup ka, 2007). Se agrass
growthanddensity‐dependentdepletionwereadjustedtomaintain
theamountofresourcesrelativelyconstantacrossasimulation.
Each mod el simulation was ru n for approximately 50 years.
The fir st two years were c onsidered as a bur n‐in period wher e
nomodeloutputwasrecordedinordertoavoidpossibleartifacts
generatedbythearbitrarily choseninitialstateofthemodelen
tities . Wer an simulations un der three envir onmental scen arios:
scenario 1, without oceanic currents; scenario 2, with oceanic
currents;andscenario 3, withoutoceaniccurrents but withlocal
pert urbations (i.e., se lective reduc tion of feeding site s' produc
tivity).Underenvironmentalscenario3,perturbationswereonly
located i n the souther n feeding patch es. We arbitrar ily chose a
singlelocationtosimplifyourunderstandingoftheeffectof the
perturbationinthemodel.Pleasenotethatwedonotconsidera
model wi thout oceani c currents (i .e., scenario 1) as re alistic but
wantedtoassessthe ireffe cts.Exploringunrealisticsce nar iosisan
importantelementofmodelanalysisandhasbeenlistedaspartof
“RobustnessAnalysis”(Grimm&Berger,2016).Foreachscenario,
weranfiverepetitionsforcombinations ofdifferentnestingand
foragingstrategies,respectively,thatis,conservative/investment
tendenciesandmover/stayertendencies(Figure7).Strategyten
dencies w ere fixed and eq ual for all tur tles throug hout a single
simulation.Overall,weranatotalof240simulations(Table3).
2.4 | Observation and analysis of model output
Model outcomes comprised spatial foraging andmigration pattern
aswellasreproductiveoutputinresponsetotheturtle'sstrategies.
Tostudyforagingandmigrationpatterns,werespectivelymeasured
feedingpatchusageandmappedcorrespondingmigrationpathways.
Wepooled, foreach scenario, theresultsfrom all combinationsof
thebehavioralstrategies' tendencies.Therefore,wedidnotassess
the spatial effectsof behavioralstrategies within a givenscenario
butratherbetweenscenarios.However,emergentbiologicalprop
ertiessuchasremigrationinterval,energystorage,andreproductive
output atrookerieswereanalyzedinthelightofbehavioralstrate‐
gieswithineachscenario.
2.4.1 | Feeding patch usage
We studied spatial patterns of three foraging statistics: (a) time
usage,thatis,thesum,overalltimesteps,ofthenumberofturtles
present ona feeding patch at each time step,(b) number of post‐
nesting visits, that is, the number oftimesthat aturtle arrived in
afeeding patchfollowing postnestingmigration, and (c)numberof
foragingvisits,that is, thenumber oftimesthat a turtle arrived in
afeeding patchfollowing foraging migration fromanotherfeeding
patch.
Inaddition,wealsostudiedtheforagingpatternsinrelationtothe
nestingsitesoforigin for theforaging turtles.Forthis, wecomputed
twoadd itionalmetrics:(a)thenumberofnestingsitesf romwhichnest
ersoriginatedinagivenfeedingpatchand(b)adiversityindexofnest
ingsitesfromwhichturtlesoriginatedinagivenfeedingpatch.
2.4.2 | Migration pathways
Turtle's prenesting and postnesting migrations were recorded by
randomlysampling individual's locations approximately every 500
timesteps.Foragingmigrationswerenotrecorded.Migrationpath‐
wayswerethenstudiedusingkernelmethodsfordensityestimation
onsampledlocations(Worton,1995).
2.4.3 | Energy at nesting, remigration interval, and
reproductive output
Only the s ix main and well kn own nesting site s (Europa, Alda bra,
Mayotte, Mohéli, Tromelin, Glorieuses; Figure 2; Table 2) were
considered in the study of the reproductive parameters. Foreach
nestin g site, the three f ollowing stat istics were com puted: (a) the
mean individual remigrationinterval defined as themean duration
betweensuccessivenestingphasespereachindividual;(b)themean
individualenergylevelatnestingdefinedasthemeanenergylevelof
turtlesafterthenestingevent;(c)therookeryoverallreproductive
outputwhichwascalculatedasthesumovereachindividual'snest‐
ingevent,thatis,thesumoftheenergy‐levelratioofeachnesting
turtlebytheremigrationinterval.
TABLE 3 Modelsimulationexperiments.Overall,weranthreescenarios,fourforagingstrategytendencies,fournestingstrategy
tendencies,andfiverepetitionsforeachconfigurationleadingtoatotalof240simulationruns
Scenario Oceanic current Perturbations Foraging strategy Nesting str ategy Repetitions Simulations
Scenario1 No No 4 4 580
Scenario2 Yes No 4 4 580
Scenario3 No Yes 4 4 580
    
|
 15
DALLEAU E t AL.
3 | RESULTS
3.1 | Feeding patch usage
Underenvironmentalscenario1(withoutoceaniccurrentsnorper
turbations),themostfrequented feedingpatcheswerelocatedon
the coasts ofMadagascar, Mozambique,and Tanzania. These re
gionshadhigherlevelsofusageintermsoftimeusage,postnesting
visits,andforagingvisits(Figure8a–c,leftpanels).Thenorthwest
ernpar tofMadagascarappearedasoneofthemostimportantfor
aging regions as thefeedingpatcheslocatedin thisarea showed
the highest levelsof timeusage (Figure 8a,left panel). Least vis
ited areas corresponded to the eastern island sites (Mascarene
andSeychelles)andthe extreme northernsites located alongthe
Somali coast. Regarding thenesting sitesof origin, high levelsof
mixingwereobservedthroughouttheregion(Figure8d,leftpanel).
FeedingpatcheslocatedinthesouthoftheMozambiqueChannel
hadlowvaluesforthediversityindexofnestingsites,thatis,tur
tles'feedinginthesepatchesoriginatedonlyfromfewnestingsites
(Figure8e,leftpanel).FeedingpatchesalongtheSomalicoast,de
spite low usage, showed highdiversity levels in thenesting sites
oforigin.
Underenvironmentalscenario2(includingoceaniccurrents),in
comparisonwithscenario1(withoutoceaniccurrents),maindiffer‐
ences in ti me usage and num ber of visits occ urred in the ea stern
coastof Madagascar (Figure8a–c,centerpanels).Whenconsider
ingthecurrents,thefeedingpatcheslocatedinthisareaexhibited
lowerlevelsoftimeusageandofforagingvisits(Figure8a,c,center
panels)buthighernumbersofpostnesting(Figure8b,centerpanel).
Regardingnestingsitesoforigin,seacurrentsincreasedthevariabil
ityinthediversitypatterns(Figure8d,e,centerpanels).Inthiscase,
patcheslocatedattheedgesoftheregion,suchasthenorthernand
southern sites of the east African coast as well as theisolated is‐
lands,werevisitedbyturtlesfromasmallernumberofnestingsites
(Figure8d,e,centerpanel).
Under environmental scenario3 (with perturbations), southern
feeding patches(ca.15°Sto25°Sof Latitude)wereexposedtoper
turbations.Incomparisonwithscenario1,itinducedlowerlevelsof
postnesting visits in the southern patches (Figure 8b, rightpanel)
and higher levels of foraging migrations (Figure 8c, right panel).
NeverthelessafewsouthernpatchesinthesouthwestofMadagascar
hadexceptionallyhighlevelsoftimeusage.Asanotherconsequence
ofperturbations,pressureonthenorthernpatcheswasincreasedas
theyweremorefrequentlyvisited(Figure8b,rightpanel).
3.2 | Migration patterns
Kernel density analysis (Figure 9a, left panel) showed important
postnestingmigratoryareasaroundtheislandsofthenorthernpart
oftheMozambiqueChannelaswellasaroundtheislandofEuropa,
south of the Channel. Twomigration corridors were observed (a)
a major trident shaped corridor, between the northern coast of
Mozambique,thesoutherncoastofMozambique,andthesouthern
coast of Mad agascar; (b) an d another impo rtant one be tween the
northerncoastofMozambiquetotheComorosArchipelago.Adding
oceaniccurrents(scenario2,withoceaniccurrents,Figure9a,center
panel)mostlymodifiedthe migratory dynamicswithinthenorthern
partoftheMozambiqueChannel.Themigrationcorridorofthisarea
was broad ened to the nor thern coast of Mad agascar and be yond
tothesmallnestingislandofTromelin.Underscenario3(Figure9a,
rightpanel),themaineffectofperturbationinthesouthernpatches
affectedthe tridentshapedcorridor,witha lossofmovementsbe
tweenthesoutherncoastofMozambiqueandthesoutherncoastof
Madagascar.
Whenlookingatanalysesforparticularnestingsites(Figure9b–
d),underscenario2afternesting,individualsfromTromelinmigrated
more frequently with the main current flow, thesouth equatorial
current (SEC), preferably toward the northwest of Madagascar
than the Mascarene islands (Reunion and Mauritius). Individuals
fromAldabraalso migrated preferablyalongtheNorth‐Equatorial
Madagascarcurrent(NEMC)flow(Figure8d,centerpanel).Similar
patternswereobservedfortheothernestingislandsofthenorthern
partoftheMozambiqueChannel:Glorieuses,Mayotte,andMohéli
(results not shown here). Under scenario 3 (perturbed foraging
sites),individualsnestingonEuropa avoidedmigrationtoward per
turbedfeeding patches,inthesouthofthe MozambiqueChannel
(Figure 9b, right panel), and they preferred migration along the
Europa‐NorthMozambiqueaxis, whichreinforcedtheimportance
ofthismajormigrationcorridor.
3.3 | Reproductive output across the region
Underallscenarios,site‐specificresultsshowedhighspatialvariability
inreproductiveoutput(Figure10).EuropaandMohéliIslandshadthe
highestlevelofreproductiveoutput.Glorieuses archipelagohadthe
lowest reproductive output. Impact ofoceanic currents (Figure 10b)
had contr asting influence a cross the region. Fo r a majority of sites
(Mayotte,Mohéli,Aldabra)oceaniccurrentsloweredthereproductive
output,sometimesdrastically(e.g.,inthecase“mover”and“conserva
tive”strategiesinAldabra,Figure10b).Yet,forEuropaIsland,oceanic
cur rent shadpositiveimpact sonreproduct iveoutputregardlessofthe
decisionsstrategies(Figure10b).ForGlorieuses,only“stayer”tenden
ciesledtosuperiorreproductiveoutput.Thepatternsweresimilarbut
combinations of “mover” and “conser vative” tendencies also led to
higherreproductiveoutput.Perturbationsofsouthernfeedingpatches
(Figure10c)hadanegativeimpactonreproductiveoutput,especially
for Europa Is land, the nea rest site from t he pertur bed patches, a nd
particularlyfor“mover”and“conservative”tendencies.Reproductive
outputofallnestingsiteswasaf fectedregardlessofthedecisionstrat
egies(withtheexceptionofGlorieuses).
3.4 | Reproductive output under
behavioral strategies
Detailedresultsregardingenergyatnestingandremigrationinter
vals (durat ion between t wo nesting phas es) are presented in t he
16 
|
   DALLEAU Et AL.
    
|
 17
DALLEAU E t AL.
Supplement. Reproductive output wasmaximal under scenario 1
(Figure 10a). O ceanic current s (scenario 2, Fig ure 10b) or human
perturbations(scenario3,Figure10c)bothhadoverallnegativeim
pactonreproductiveoutput.Nevertheless,insomerarecasesfor
“conservative”nesting tendencies mean reproductive output was
higher when considering ocean currents (Figure 10b, left panel).
Reproductiveoutputunderscenario1waslowerfor“investment”
nestingtendenciesthanfor“conservative”nestingtendencies.Toa
le s s erex tent,i twas a lsos l ightl y decre a singf o r“s t a yer”f o ragin g ten
dencies.Nestingstrategydidnothaveanyinfluenceonthetrendin
reproductiveoutputwhenconsideringoceancurrents(Figure10b,
leftpanel).However,whenintroducingperturbationsthelossinre
productiveoutputwasmorepronouncedin“conservative”tenden
cies than“investment” nesting tendencies(Figure10c, left panel).
On the oth er hand, while fo raging str ategies had lit tle impact o n
reproductiveoutputunderperturbation scenario (Figure9c,right
panel), t hey modified r eproduct ive output in re sponse to ocean ic
currents.Here,weobserveda higher loss in reproductive output
for“movers”than“stayers.”Tosummarize,themodelpredictedthat
“stayer”foragingtendenciesshouldperformbetterwhenmigration
wasstrongly affectedbyoceaniccurrentswhileinperturbedenvi
ronments“investment”nestingtendencieslimitedthelossinrepro
ductiveoutput.
4 | DISCUSSION
4.1 | The importance of landscape configuration on
the spatial ecology of sea turtles
4.1.1| Landscape configuration spatially structures
sea turtles' populations
Migratorycorridorsandforaginghotspotsarecommonlyobserved
for green turtle populationsworldwide (Luschi, Hays, Del Seppia,
Marsh, &Papi, 1998;Readetal.,2014;Stokesetal.,2015;Troëng,
Evans,Harrison, &Lagueux, 2005).By implementingsimple move‐
mentandbehavioraldecision rules,wewereableto reproducethe
mainregionalpatternsobservedthroughgeneticandtrackingstud
ies,aswediscussindetailbelow.Ourresultssuggestthatthespatial
distrib ution of migration c orridors and for aging hot spots is co n‐
strainedbytheintrinsiclandscapeconfiguration,thatis,therelative
locationofnestingsitesandforagingareas,landbarriers,andoceanic
currents.Theinitialchoiceofaforagingsiteforseaturtlesmightin‐
volvemechanismsmorecomplicatedthanthoseimplementedhere,
suchasdriftingpatternandimprintingduringearlylifestages(Scott,
Marsh, & Hays,2014).Nevertheless, themodel demonstratedthat
constraintsoccurringattheadultstagecouldexplainandmaintain
observedspatial patternsinthefield (migration corridors,foraging
area usage, foraging area composition). The adult's environment,
hereintheshapeofcoast albarrie rs,o ceaniccurrents,andperturba
tions,mightmodifymigratoryconnectivitybetweensites.
4.1.2 | Landscape configuration affects
reproductive output
Variations of r eproductive ou tput have been obse rved in various
species of sea turtles through numerous parameters: remigration
interval(i.e.,breedingfrequency),clutchfrequency,clutchsize,size,
andnutritional componentsofeggs, hatching, and emergence suc‐
cess.Someparametersareaffectedbyphysiologicalconstraints;for
example,theclutchsizeisgenerallycorrelatedwiththesizeofthe
female (Broderick,Glen, Godley,& Hays,2003; Hays & Speakman,
1991),orbylocalconditionsatnestingsites;forexample,theemer
gence success highly relies on incubation conditions (Mortimer,
1990).Variations in the parametersremigrationintervaland clutch
frequencyaremainlyattributedtoforagingandmigrationconditions
(Brodericketal.,2003;Hatase,Omuta,&Tsukamoto,2013;Hatase
&Tsukamoto, 2008;Troeng& Chaloupka,2007;VanderZandenet
al.,2014).
In the model, levels of reproductive output were very vari
able between rookeries under identical simulation parameters
(Figure10),suggestingthatlandscapestructureaffectedreproduc‐
tive potential in away that similar behaviorsled to various repro‐
ductiveoutputdepending onrookery location and accessibility to
foraginggrounds.ThereisalackofdatatoallowaSWIOanalysisof
reproductiveparametersthatcouldvalidateourtheoreticalresults
regardingreproductiveoutput.However,thesevariationsareinac‐
cordancewithseaturtlereproductivebiology.Forexample,Troeng
and Chaloupka (20 07) suggested that short remigration intervals
(2–3years) observedinthe rookery of Costa Ricacould be dueto
therelativeproximityoftheforagingsites.
Wealsodemonstratedthatoverallreproductiveoutputheav
ilyreliedonspatialandenvironmentalconditions,andthat,under
these conditions, behavioralstrategiesmightperformdifferently
essentiallybyaffectingtheenergeticlevelatnesting(Figure S1).
Oceaniccurrentsintroducedenvironmentalheterogeneityandun
certaintyalongmigrationpathwaysgenerallyloweringtheoverall
reproductiveoutput.Undersuchconditions,themodelpredicted
that a “stayer” foraging behavior ledto better reproductive out
put at popu lation level. Thi s suggested tha t uncertaint ies along
FIGURE 8 Usageoffeedingpatches.Leftpanelsdescribeusageoffeedingpatchesunderscenario1,centerandrightpanel,respectively,
describefeedingpatchusageunderscenario2andscenario3relativetoscenario1.Rowsofpanelscorrespondtodifferentusagestatistics:
(a)timeusage,thatis,totalnumberoftimestepsspentbyturtlesonthissite;(b)numberofpostnestingvisits,thatis,numberoftimesthat
aturtlereachedthissitefollowingapostnestingmigration,(c)numberofforagingvisits,thatis,numberoftimesthataturtlereachedthis
sitefollowingaforagingmigration,(d)numberofsitesfromwhichforagingturtlesoriginate,(e)diversity(Shannonindex)ofsitesfromwhich
foragingturtlesoriginate;thisindexreflectsthe“proportion”ofnestingsitesfromwhichforagingturtleoriginate.Allstatisticsarecalculated
overtheallsetsofsimulationforeachscenario,thatis,5simulationsforeachofthe4×4combinationsofforagingandnestingstrategies
takenin(0.2,0.4,0.6,0.8;80simulationsperscenario)
18 
|
   DALLEAU Et AL.
thewayduringmigrationmightfavorfidelitytoforaginggrounds,
a commonl y observed b ehavior for sea t urtles (Br oderick et al. ,
2007;Godleyetal.,2008).Althoughataglobalscale,theflowof
currentsispredictable,thisisnotsofromtheindividual'spointof
viewthatcannotbesurewhichcurrentsitwillfaceduringmigra
tion.Inourmodel,thecurrentsintroduceenergeticconstraintsor
FIGURE 9 Kerneldensitiesofmigrationpathwaysunderthethreescenarios.Densitiesforscenario1(leftpanel),scenario2(center
panel),andscenario3(rightpanel)for(a)allindividuals;individualsnestingin(b)Europa,(c)Tromelin,and(d)Aldabra.(b,c,d)Thenesting
islandisrepresentedwiththeblackcross.Kernelswerecalculatedovertheallsetsofsimulationsforeachscenario,thatis,5simulationsfor
eachofthe4x4combinationsofforagingandnestingstrategiestakenin(0.2,0.4,0.6,0.8),byrandomsampling125positionspersimulation
(10,000positionsperscenario)
    
|
 19
DALLEAU E t AL.
FIGURE 10 Reproductiveoutputforthesixmainnestingsites.Reproductiveoutputasafunctionofnestingallocationstrategy(x‐axis)
andforagingpatchfidelitystrategy(y‐axis).Boxedvaluesarepositive.(a)Reproductiveoutputsforscenario1,(b)reproductiveoutputs
forscenario2relativetoScenario1,and(c)reproductiveoutputsforscenario3relativetoscenario1.(ALD,Aldabra;EUR,Europa;GLO,
Glorieuses;MAY,Mayotte;MOH,Mohéli;TRO,Tromelin).Graygradientindicateshighestvalues(darkgray)tolowestvalues(lightgray).A
diagramrepresentationofthisfigureisavailableinFigureS3
20 
|
   DALLEAU Et AL.
savings,buttheimpactontheturtle'senergeticlevel,upscaledto
thepopulationlevel,willemergefromthemodel.Costsofsearch
ingforforagemightbeariskybehaviorleadingtooveralllowerre
productiveoutput,especiallywhenneighboringforagingpatches
areoflowqualityorhavebeenexploitedalreadybyotherturtles.
Likewise, t he model pred icted that mo re “investment ” in re
productionwaspreferablewhenperturbationtookplace.Testing
thishypothesisinthefieldischallenging.However,fortheleath
erback p opulation of Frenc h Guyana, it has b een demonst rated
that a trade‐off exists between thereproduction effort andthe
delaybetweenreproductionevents(Rivalanetal.,20 05),suggest
ingthatlargerreproductiveintervalscouldbecounterbalancedby
higherinvestmentinreproduction.
Interestingly, while f avorable behaviors were identified under
various environmental circumstances, no general “best strategy”
arosefromthemodelresults.Underidenticalparameters,responses
of reproductive output tobehavioral strategies could be opposite
dependingonthelocationofconsideredrookeriesandthereforein
fluencebyotherextrinsicfactors.Landscapeconfigurationcanlead
tonontrivialresponses,whichunderlinestheimportanceofexplic
itlyconsideringspaceandmovement.
4.2 | Migration corridors, foraging hot spots, and
conservation of green turtle in the SWIO
Under multiple scenarios, the model highlighted two connected
provinces,inthenorthandthesouth oftheMozambiqueChannel,
with thei r own structural part icularities in reg ard to green turtle
populat ions. The mode l provided some ex planations to th e origin
of this regio nal pattern a nd also allowed dr awing furt her hypoth‐
eses.Theconclusionsandtheircomparisonwithfieldobservations
are discussed in the following, as well as potential conservation
implications.
4.2.1| Migration and foraging hot spots, relative
contribution of regional rookeries
Anorthernmigrationcorridor emerged inthemodelbetweenthe
northofMadagascarandthenortherncoastofMozambique.From
themodel,wecouldinferthatthepresenceofnumerousnesting
sites in thisarea andtheir central locationrelative tothedistribu‐
tionoftheregionalfeedingpatches(Figure2)islikelytoexplainthe
highdensitiesofmigrating turtlesin thisarea.Consideringoceanic
current s (scenario 2), we fou nd that the west ward Nort h‐Eastern
Madagascar Current (NEMC) tended to widen the northern mi
gration corridor along itseast–west axis. Thismight explain thata
majorit y of tracked nest ing green turt les from Tromelin migrate d
alongtheNEMCcurrent(Dalleau,2013).IntheSouthernpartofthe
MozambiqueChannel,atridentshapedmigrationcorridorwasalso
observed,withahighleveloffrequentationduetothehighnumber
offemalesnestinginEuropaIsland(Bourjea,2015).Theexistenceof
similarmigrationcorridorswasalsooneofthemajorobservationsof
aregionaltrackingstudy(Dalleau,2013).
According to the model, the coastal areas of Africa and
MadagascarborderingthenorthoftheMozambiqueChannelwould
be the most frequented oneby turtlesoriginating from numerous
rookeries. This is consistent with known foraging locations from
field observations (Fulanda et al., 2007; Muir, 2005; Okemwa et
al., 20 04). Further, this is al so in agreement wit h the distribu tion
of foragin g areas of turtl es' satellite‐tra cked from the major ity of
the regional nesting sites (Dalleau, 2013) that identified four re‐
gionalforaginghot spotsofwhichthreeareborderingthenorthof
theMozambiqueChannel in Tanzania,northern Mozambique,and
northern Madagascar.It is worthwhile to mention that the choice
of feeding p atches was not impo sed throughout a simul ation but
emergedfrom turtle'sdecisionsdependingondensityandhabitat
quality.This indicates thatin our modelthemechanismsofhabitat
selectionareworkingwell.
Anotherconclusionofthe model was thatnumerousrookeries
contribute to the nesters compositionof thenorthern part of the
region(5–15°S),while thehighleveloffrequentationofthesouth
ernpart(18–25°S)reliesonasinglerookery,Europa. Thisresult is
inagreementwithregionalgeneticanalysisbasedonmitochondrial
DNA. Indeed, Taquet (2007)showed thatforaging adultgreen tur‐
tles of Tanzania andwesternMadagascar sharehaplotypes mostly
observed in the northern nesting sites (Bourjea, 2015; Bourjea,
Lapègue, et al., 2007), while foraging adult s of South Africa and
strandedadultsofthe southwestofMadagascar sharehaplotypes
mostlyobservedinEuropanestingpopulation(Bourjea,Lapègue,et
al.,2007).
4.2.2 | Implications for conservation
Themodelpromotedareasasmajorregionalmigratoryandforaging
hotspotsforadultfemalegreenturtles.Twoprovinces,inthenorth
and south of theMozambique Channel, with contrasted dynamics
werecharacterized.Theparticularitiesofeachleadtodifferentchal
lengesintermsofconservation.
IntheSWIO,directtakeandcoastalfisheriesbycatchareamajor
threat(Bourjea,2015).Lookingatthemodelresults,wecouldinfer
thathigh levelsofbycatch and direct takereportedalong theeast
Africancoasts(seereviewinBourjea,2015)ofMozambique(Gove,
Pacules,&Gonçalves,2001;Kiszka,2012;Williams,2017;Williams,
Pierce, Hamann, & Fuentes, 2019), Tanzania (Moore et al., 2010;
Muir, 2005), an d Kenya (Mueni & Mwangi , 2001; Okemwa et al.,
2004)mightprobablyaffectallseaturtlenestingpopulationsofthe
region,andmorespecificallynestingpopulationsfromthenorthof
theMozambiqueChannel.
High level of direct take also occurs in the western coast of
Madagascar(Rakotonirina,2011).Amajorityofindividualsaregreen
turtles captured along thesouthwest coast (Humber etal., 2011).
Themodelshowed,inagreementwithgeneticsdata(Taquet,2007),
thatadultindividualsofwesternMadagascarwereessentiallyissued
fromthenestingpopulationofEuropaIsland.Conservationefforts
inthisareawouldthenconsequentlybenefit preferentiallyEuropa
Island'sgreenturtlenestingpopulation.
    
|
 21
DALLEAU E t AL.
Themodel,underaperturbationscenario,alsopointedoutthat
depletingforaginggroundsof the southern MozambiqueChannel
mightraisethepressureonnorthernforaginggroundsandindirectly
haveconsequences on thereproductive output of rookeries atre‐
gional scale.As a consequence,while protection effortsmight tar‐
getspecificareas dependingonthe conservation goals,the model
clearly highlighted that green turtle conser vation should be ap‐
proachedasaregionalmatter.
AsproposedbyWallace,DiMatteo,etal.,2010,ourmodelresults
confirm thatthe “Regional ManagementUnit”identified forgreen
turtlesintheSWIO(Wallace,DiMatteo,etal.,2010)isthebestscale
forthis area,but the model substantiallyhighlighted the existence
ofsubregionsor provinces withdistinct butconnected population
dynamics.IftheRegionalManagementUnitcanbeapprehendedas
awholethevariablecontributionfromnestingtoregionalmigration
andfeedinghotspotsprovesthatlocalizedconservationactionswill
not affect populations in the sameway. Understanding more pre‐
cisely th e regional spatia l dynamics of gre en turtle is pre cious to
conductmonitoringandconservationefforts wherethey are most
neededwithintheRegionalManagementUnit.
4.3 | Outlook on model improvements
Buildingthemodelprovidedvaluableinsightintotheareasforwhich
biologicalinformationisavailable, but also intothe areasforwhich
therearecriticalgapsinspeciesbiologicalandecologicalknowledge.
Seaturtles'spatialecologybenefitstherecentprogressinbiotelem‐
etry andmoreparticularlyinsatellitetracking.There is now a bet‐
terunderstandingonthespatialecologyofseaturtlesinthemajor
oceans(Hays ,2008)andit slongdistan ce smigrationsrelatedtooce
anic environment(Luschi, 2013). Inthe western Indian Ocean, re‐
centresultsusinggenetics(Bourjea,Lapègue,et al., 2007;Bourjea,
Mortimer,etal.,2015),satellitetracking(Dalleau,2013),andspatial
statistics(Dalleauetal.,2012)haveprovidedabetterunderstanding
oftheregionaldynamicsofgreenturtleintheregion.High‐density
trackingdatacanalsobeusedtodevelopcorrelativehabitatmodels
(oftenalsoreferredtoasspeciesdistributionmodels),whichpredict
high‐qualityhabitat;thishasbeendonealreadyfortheloggerhead
seaturtle(Abecassisetal.,2013).
Theconceptof integratingmovement,energetics,andrepro
ductionisnovelforthissystem,andconfirmedimportantareasfor
conservation.Weproposelinkingthephysiologyoftheanimaland
itsphysicalenvironment (foodandcurrents in our case)asaway
forwardinunderstandingmovementdecisionsandemergingpop
ulationpatterns;thisalsooffersnewpredictivetoolstoassessef
fect sofhabit atorclimatechange(Malishev,Bull,&Kearney,2018).
Historically,metabolicphysiologystudieshaveusedrespirometry
toassessmetabolicratesinclosed‐circuitssystemsanddoublyla
beledwater techniquehasalso beenusedtoestimate fieldmeta
bolicrates(Enstippetal.,2016;Wallace&Jones,2008).Thelatest
advancesintechniquessuch as accelerometry mightalsoprovide
better insights (Hayset al., 2016). For various seaturtle species,
there hasbeen increasing knowledge about energetic balance of
specif ic physiologic al states: ne sting (e.g., H ays, Broderi ck, Glen,
& Godley, 2002), migration (Enstipp et al., 2011; Halsey, Jones,
Jones, L iebsch, & Boot h, 2011), or foraging (e. g., Ballorai n et al.,
2013; Ballor ain et al., 2010; Enst ipp et al., 2016). Still it re mains
achallengeformeasuringmetabolic ratesoffree‐ranging turtles,
andnointegrativeeco‐physiologicalmodelexistsyetthatencom
passes and unifies thethree processes together (Williard, 2013).
Additio nally, the physiolog ical fact ors that trig ger nesting m igra
tionattheindividuallevelarestillpoorlyunderstood.Progressin
energeticsofseaturtleswouldbeofkeyvaluetoimproveindivid
ual‐basedmodelingofseaturtles'ecologicalprocessesintheiren
vi ronm entfrom basi cpr inci p les. Ana d diti onalwayofimpr ovin gthe
seaturtles'energybudgetsinthemodel wouldbe usingDynamic
EnergyBudget (DEB)theory(Kooijman,2010),which isageneric
modelthatpredictshowmuchananimalinvestsenergyingrowth,
maintenance,andreproduction,andhowthisdependsontheani
malssizeandmaturity.DEBisincreasinglyusedinindividual‐based
models (G alic, Grimm , & Forbes, 2017; Galic, Su llivan, Grim m, &
Forbes,2018;Martinetal.,2013;Martin,Zimmer,Grimm,&Jager,
2012)andisalsounderdevelopmentforimprovingtheenergybud
get model of a m odel of harbor p orpoise; C. G hallager, perso nal
communication. WhiletheoriginalDEBtheorydoesnotexplicitly
addressmovement,thishasbeenaddedrecentlyforthemovement
oflizards(Malishevetal.,2018).
In the SWI O, population t rends have been i n most cases e sti‐
matedfrom nesting crawls (Bourjea, Dalleau, etal.,2015; Bourjea,
Frappier, et al., 2007;Lauret‐Stepler et al., 2007;Mortimer, 2012;
Mortimer,von Brandis, etal., 2011;Mortimer,Camille, &Boniface,
2011) and individual's reproductive parameters have rarely been
monitored,asnestingsitesarehardlyaccessibleinthisregion.Spatial
comparisonofindividualreproductiveparameterswouldberequired
fora betterassessmentofpopulation'sviabilityas weshowedthat
responsetoenvironmentaluncertaintiessuchasoceaniccurrentor
pert urbations var ied according to ne sting and for aging strateg ies.
Futureimplementationofthe modelshouldthereforealsoinclude
demographic processes. Indeed, whilereproductive potential was
considered,survivalandfecunditywereinfactnotexplicitlyimple‐
mentedinourmodel.Foragingandnestingstrategieswerefixedfor
agivensimulation.Thisisunlikelytobethecaseinrealit ysincevari
ousstrategiesprobablyevolveorcoexist.Ideally,decisionstrategies
should emergefrom themodel. Thiswouldrequireadaptationand
survivaltobealsoimplemented.
Implementingperturbationsenabledustoqualitativelyshowthe
potential ofIBMs topredict spatial, temporal, and survival conse‐
quences of m odifications of t he foraging envir onment. Advance d
modelingcouldprovideaneffectivetooltopredicttheimpactofcli
matechangeonseaturtles'populationsasspatialcomplexityoftheir
life cycle makes prediction hardly accessible (Hawkes, Broderick,
Godfrey,&Godley,2009).Regardingmoredirecthumanperturba
tion,thereisstilllittleliteratureaboutpoachingandartisanalfisher
iesbycatchintheregion(Bourjea,2015;Humberetal.,2011;Temple
et al., 2018). Inc luding human th reats quantit atively in the mo del
wouldmakeitaperfecttoolformanagersanddecisionmakers.
22 
|
   DALLEAU Et AL.
4.4 | Behavioral and spatially explicit modeling for
sea turtles: the quest for the grail
Ourstudyunderlinedthe importanceofspatiallyexplicitmodeling
to spatial e cology and popu lation dynamic s of migratory spe cies.
The model provides new insights on green turtle biology, link‐
ing spatial ecology,and population dynamicsthroughthe useof a
basicphysiological energy budgetandbehavioraldecisionstrategy
model. Weintegratedina spatiallyrealisticcontextthethreemain
processesof theadult biology of seaturtles: reproduction,migra‐
tion,andforaging.Whileitremainedatthisstageaconceptualand
explorativeapproach, the mainbenefitswere to (a)provideanop‐
eration al tool to charac terize the spati al structu re of green tur tle
populationsatregionalscaleandto(b)explicitlyexploretheroleof
landscapeconfiguration (nesting and foraging site distribution,ter‐
restrialbarrierssuchasMadagascar,environmentaldrivers such as
oceaniccurrents)and(c)individual'sdecisionstrategiesonseaturtle
spatialecology.Practical conclusionsprovideimportantconsidera
tion thataddresses large research prioritiesrecently identifiedfor
seaturtles(Reesetal.,2016).
With the improving knowledge onsea turtlebiology at indi
vidual scale, individual‐based approaches should progress and
becomemoreintegrative.Suchknowledgemayallowhighlighting
differentindividualforagingand/ornestingstrategiesthatmaybe
testedinthemodel by implementing adaptiveness of fixed ver
sus plast ic response s to environment al changes (e.g. , Bradshaw,
Hindell,Sumner,&Michael,2004;Railsback&Grimm,2019).The
nextlogicalsteptoimprovethesemodelsrequiresabetterability
toexplicitlyconsider landscapeandmovementinarealisticcon
text.Themodelpresentedhereconstitutesafirststep.Although
explorative,someoftheideasimplementedshouldinspirespatial
ecologist aiming at unifying movement ecology and population
dynamics.
ACKNOWLEDGMENTS
Theauthorsofthisarticlewanttothankthefollowingpersonsfor
their implication in the project and their constructivecomments:
Daniel David (LIM‐IREMIA), Denis Payet (LIM‐IREMIA), Nicolas
Sébastien (LIM‐IREMIA), Joël Sudre (LEGOS, France), and Serge
Andréfouët (IRD, New Caledonia). Research exchange travels for
MDand SKShavebeen funded bytheRUN SeaScienceProgram,
part of the European Union's Seventh Framework Capacities
ProgrammeandcoordinatedbytheFrenchInstituteofResearchfor
Development(IRD).Computationshavebeenperformedonthesu
percomputerfacilitiesoftheUniversityofLaRéunion.MDwantsto
thankpersonally“RégionRéunion”thatfundedhisPhDstudentship.
Wealso would liketothank one anonymousreviewerfor helpful
comments.
CONFLICT OF INTEREST
Nonedeclared.
AUTHOR CONTRIBUTIONS
MDdeveloped, programmed, andanalyzed themodeland led the
writingofthepaper.SK‐SandVGcodevelopedthemodelandcon‐
tributedtowritingthepaper.JB,YG,andGLprovidedinputinterms
ofexpertknowledgeaboutthe species and system addressedand
contributedtowritingthepaper.
DATA AVAILAB ILITY STATE MEN T
Alldatarequiredtorunthemodel,thatis,inputfileswiththeloca‐
tionofforagingandnestingsitesaswellasallinputmaps,areavail‐
able,togetherwiththeNetLogoprogramimplementingthemodel,
theODDmodeldescription,andinstructionsforrunningthemodel,
intheComputationalModelLibraryoftheComSESNetwork:https
://www.comses.net/codebases/69863caa‐2f8e‐4412‐a564‐a2826
d9d38d3/releases/1.0.0/.
ORCID
Mayeul Dalleau https://orcid.org/0000‐0001‐53023298
Volker Grimm https://orcid.org/0000‐0002‐3221‐9512
REFERENCES
Abecassis,M.,Senina,I.,Lehodey,P.,Gaspar,P.,Parker,D.,Balazs,G.,&
Polovina,J.(2013).Amodelofloggerheadseaturtle(Caretta caretta)
habita t and movement i n the Oceani c North Pa cific. PLoS ONE, 8,
e73274.https://doi.org/10.1371/journal.pone.0073274
Ballorain,K., Bourjea, J., Ciccione, S., Kato,A.,Hanuise,N.,Enstipp,M.
R., … Georges, J. Y. (2013).Seasonal diving behaviour and feeding
rhythmsof greenturtlesatMayotte Island.Marine Ecology Progress
Series,483,289–302.https://doi.org/10.3354/meps10301
Ballorain,K.,Ciccione,S.,Bourjea,J.,Grizel,H.,Enstipp,M.,&Georges,
J.‐Y.(2010).Habitatuseofamultispecificseagrassmeadowbygreen
turtlesChelonia mydasatMayotteIsland.Marine Biology,157,2 5 81–
2590.https://doi.org /10.10 07/s00227‐010‐1520‐7
Bjorndal, K. (1980). Nutrition and grazing behavior of the green tur‐
tle Chelonia mydas. Marine Biology, 56, 147–154. https://doi.
org/10.1007/BF00397131
Bjorndal,K. A.(1997).Foraging ecology andnutrition ofsea turtles.In
P.L.Lutz&A.Musick(Eds.),The biology of sea turtles(pp.199–231).
BocaRaton,FL:CRCPress.
Bjorndal, K.A.,Bolten, A.B., & Chaloupka, M. Y.(2000). Green turtle
somaticgrowthmodel:Evidencefordensitydependence.Ecological
Applications,10,269–282.
Blumenthal, J. M., Abreu‐Grobois, F. A., Austin, T. J., Broderick, A.
C., Bruford, M. W., Coyne, M. S., … Godley, B. J. (2009). Turtle
groups or turtle soup: Dispersal patterns of hawksbill turtles in
the Caribbean. Molecular Ecology, 18, 4841–4853. https://doi.
org /10.1111/j.1365‐294X. 20 09.04403.x
Bost, C .‐A., Cotté , C., Bailleul , F., Chere l, Y., Charr assin, J.‐B., Gu inet,
C., … Weime rskirch, H. (20 09). The impor tance of oceano graphic
frontstomarinebirdsandmammalsofthesouthernoceans.Journal
of Marine Systems, 78, 363–376. https://doi.org/10.1016/j.jmars
ys.2008.11.022
Bourjea, J. (2015). Sea turtles: A review of status, distribution and in‐
teractionwithfisheriesintheSouthwestIndianOcean. InR.P.Van
derElst,&B.I.Everett(Ed.),Offshore fisheries of the Southwest Indian
    
|
 23
DALLEAU E t AL.
Ocean: Their status and the impact on vulnerable species. Durban,
SouthAfrica:OceanographicResearchInstitute.
Bourjea, J., Ciccione, S., & Ratsimbazafy, R. (2006). Marine turtlessur‐
veysinNosyIranjaKely,North‐WesternMadagascar.Western Indian
Ocean Journal of Marine Science,5,209–212.
Bourjea, J., Dalleau, M., Derville, S., Beudard, F., Marmoex, C.,
M'Soi li, A., … Frazie r, J. (2015). Se asonality, abu ndance, and fi f
teen‐yeartrendingreentur tlenestingactivityatItsamia,Moheli,
Comoros. Endangered Species Research, 27,265–276. https ://doi.
org/10.3354/esr00672
Bourjea,J.,Frappier,J., Quillard,M.,Ciccione,S., Roos,D.,Hughes,G.,
& Grizel, H . (2007). Mayot te Island: A nother impo rtant gre en tur‐
tle nesting site in the southwest Indian Ocean. Endangered Species
Research,3,273–282.https://doi.org/10.3354/esr00053
Bo u rjea ,J., L apèg u e,S . ,Ga g nevi n ,L., B rode r ick, D., Mortim e r,J. A . ,Ci c cio n e,
S., … Grize l, H. (2007) . Phylogeogra phy of the green t urtle, Chelonia
mydas,intheSouthwestIndianOcean.Molecular Ecology,16,175–186.
Bourjea,J.,Mortimer,J.A.,Garnier,J.,Okemwa,G.,Godley,B.J.,Hughes,
G.,…Muths,D.(2015).Populationstructureenhancesperspectives
on region al management of th e western Indian O cean green tur‐
tle. Conservation Genetics, 16 , 1069–1083. https://doi.org/10.1007/
s10592‐015‐0723‐3
Bradshaw,C.J.A.,Hindell,M.A.,Sumner,M.D.,&Michael,K.J.(2004).
Loyaltypays:Potentiallifehistoryconsequencesoffidelitytomarine
foragi ng regions by sou thern ele phant seals . Animal Behaviour, 68,
1349–1360.https://doi.org/10.1016/j.anbehav.2003.12.013
Broder ick, A. C., C oyne, M. S., Fulle r, W. J., Glen, F., & Godl ey, B. J.
(2007). Fidelit y and over‐wintering of sea turtles. Proceedings of
the Royal Society B: Biological Sciences,2 74, 1533–1538.https://doi.
org /10.1098/rspb. 2007.0211
Broderick, A.C., Glen, F.,Godley, B. J., & Hays,G. C. (2003). Variation
in reproductive output of marine turtles. Journal of Experimental
Marine Biology and Ecology, 288, 95–109. https://doi.org/10.1016/
S0022‐0981(03)00003‐0
Brothers,R.,&Putman,N.F.(2013).Natalhomingandimprintinginsea
turtles.InJ.Wyneken,K.J.Lohmann,&J.Musick(Eds.),The biology
of sea turtles,(Vol.3,pp.59–77).BocaRaton,FL:CRCPress
Carr,A.F.(1986).The sea tur tle: So excellent a fishe.Austin,TX:University
ofTexasPress.
Cerritelli,G.,Bianco,G.,Santini,G.,Broderick,A.C.,Godley,B.J.,Hays,
G.C.,…Åkesson,S.(2018).Assessingrelianceonvectornavigationin
thelong‐distanceoceanicmigrationsofgreenseaturtles.Behavioral
Ecology,30,68–79.https://doi.org/10.1093/beheco/ary166
Chaloupka, M., Bjor