Habitat-relatedheterogeneityinbreedinginametapopulation oftheIberianlynx
Ne´storFerna´ndez,MiguelDelibesandFrancisco Palomares
N.Ferna´ndez(), M.Delibes andF.Palomares,DeptofApplied Biology, Don˜anaBiological Station,Spanish CouncilforScientificResearch— CSIC,Avda.Mar´ıaLuisas/n,ES-41013 Seville,Spain.(Present addressofN.F.:Dept ofEcology andPlantBiology,Univ.ofAlmer´ıa, Ctra.Sacramentos/n,LaCan˜adadeSanUrbano,ES-04120Almer´ıa,Spain.)
Identifyingattributesassociatedwithgoodbreedinghabitatiscriticalforunderstandinganimalpopulation dynamics.However,theassociationbetweenenvironmentalheterogeneityandbreedingprobabilityhasbeen oftenoverlookedinhabitatanalyses.Weevaluatedhabitatqualityinametapopulationoftheendangered IberianlynxLynxpardinusbyanalyzingspatiotemporalpatternsin breedingrecords.Datasummarizing successful productionoflittersafteremergencefromdensoverfouryearswithin13lynxterritorieswere examined. Wedesignedasetofgeneralizedlinearmixedmodelsrepresentingdifferenthypothesesregardinghow patternsinbreedingrecordsrelatetoenvironmentalheterogeneity.Environmental heterogeneitywasdescribed bytwocharacteristics: 1)alandscapeindexmeasuredinlynxterritoriesindicativeoftime-averaged prey availabilityand2)yearlyvariabilityinpreyabundancenotcapturedwiththisindex.Byincludingtherandom effectofthelynxterritorywealso accountedforotherterritory-specific effectsonreproduction.Wefound significantdifferencesinyearlypreydensitydynamicsamonglynxterritories.However,temporalvariationin preydensitycontributedpoorlytoexplaininglynxbreeding.Themostparsimoniousmodelincludedthe landscapestructureastheonlyeffectexplainingbreedingpatterns.Amultinomial-model-representationof the landscapehypothesisexplainednearly50%ofvariabilityinbreedingrecords.Results pointedtotheexistenceof ahabitatqualitygradientassociatedwithparticularlandscapestructuresinfluencinglynxhabitatselectionand breedingperformance.Underlyingthisgradientwastime-averagedpreyavailability.Probablyasaresultoflong- termfitnessstrategiesinlong-livedterritorialspecies,theshort-termfluctuationsinpreyavailabilityhadaminor influence.Ourresultsillustratehowhabitatinferencescanbeenhancedbyincorporatingthelinkbetween spatiotemporalpatternsinreproductionandenvironmentalheterogeneity.
Habitatselectionisoneofthemostcriticalprocesses influencingindividualfitnessandpopulationdynamics inanimals(Wiens1976,Pulliamand Danielson1991). Recently,ecologistshavereliedonanimaldistribution dataforthe studyofhabitatselectionand habitat- relatedpopulationconstraints(reviewedinBoyceand McDonald1999,GuisanandThuiller 2005).Land- scape-orientedapproaches,inwhichspatialheteroge- neityintheenvironmentisexplicitlyanalyzed,have greatlycontributedtoourunderstandingofthehabitat anddistributionpatternsofmanyanimal species. Further,ithasbeenargued thattherelationship betweenpatternsofspecies occurrenceandlandscape variablesisapromisingapproachfordescribingthe
environmentalrequirementsfor populationandspecies persistencein the frameworkof niche theory and naturalselection(Hirzeletal.2002,Petersonetal.
2002,GuisanandThuiller2005).
Therelationshipbetweenspeciesdistributionand
environmentalheterogeneityisonlyalimiteddescriptor of the necessaryconditions for the persistenceof populations,ashabitat usedoesnot alwaysreflect habitatqualityintermsoffitness(Pulliam2000).For example,individualsmayoccasionally occurinlow- qualityhabitatswhere theydonotbreedorthey reproduceatlowrates,whereasaminorfractionof higherqualityhabitatsmayberesponsibleforthemajor contributiontopopulationproductivity(Sergioetal.
2003,Penterianietal.2004).Thisvariabilityimplies spatiallyheterogeneouspatterns invitalratessuchas reproductionwithinhabitats.Therefore,morerigorous habitatanalysesrequirethestudyofbreedingperfor- manceindifferentlocationswhereindividuals are found.
Recognizingthat ecological patterns can show variabilityatarangeofscales(Levin1992),several studieshaveimprovedourabilitytoidentifyspecies habitatsbyincorporatingahighdegree ofdetail regardingspatialheterogeneityatdifferentscales(Saab
1999,Maneletal.2000,Fleishman etal.2001, Johnson et al. 2004, Ferna´ndez 2005). However, mosthabitatstudiesstillassumeaconstantenviron- mentandrelyonly onsamplesofanimaldistributions takenatonetime-interval,thereforeoverlookingdy-
namicpatterns.Onthecontrary,temporalheterogene- itymayintroduceanadditionalsourceofvariabilityand exert animportantinfluenceonpopulationdynamics
(Southwood 1977,Korpima¨ki1988).Forexample, breedingperformanceinterritorialanimalshavebeen foundtoberegulatedbylandscapestructurewithinthe territorybutalsobyshort-termchangesintheavail- abilityof resourcessuchasfood(Franklinetal.2000). Therefore,the analysisand conservationof animal populationsrequirestheexplicitexaminationofpat- ternsdescribingbothspatialand temporalvariabilityto characterize thespecies habitatsandtheirqualityfor reproduction.
Weinvestigatedspatiotemporalpatternsinrepro- ductioninametapopulationofacriticallyendangered
carnivore,the IberianlynxLynxpardinus. Previous
studieshaveshownthathabitatselectionintheIberian lynxisstronglyinfluencedbydensityof itsstapleprey, theEuropeanrabbitOryctolagus cuniculus.Moreover, the distributionand sizeof lynxterritoriescan be predictedfrom landscapevariablesinfluencingprey density(Ferna´ndezetal.2003,Ferna´ndezetal.2006). However,thefactorsthatinfluencesuccessfulbreeding bylynxareunclear.Inthepresentstudyweaimedto testalternativehypothesesabouttheeffectsofenviron- mental heterogeneityon spatiotemporalpatterns in reproductionintheIberianlynx.Forthis,werecorded theyearlyproductionoflittersafteremergence from dens.Wehypothesizethatlandscapevariablesindicative oftime-averagedpreyavailabilityexplainthefrequency ofthesebreedingrecords.Apositive resultwould indicatethatthelandscapestructureinformsnotonly
thesuitableconditionsforhabitatselectionbutalsothe habitatqualityforbreeding.Thishypothesisassumes constantenvironmenteffectsonreproduction.How- ever, local rabbit abundance within lynx habitats may largelyvaryamongyears(Palomares etal.2001), whichintroduces atemporaldimensioninenviron- mentalvariability.Therefore,wealsohypothesizethat short-term variation in prey abundance influence
spatiotemporalpatternsinbreeding,implyingthatthe qualityof lynxterritories variesfrom one breeding seasontoanother.Preyavailability mayinfluenceyearly breedingprobabilityatdifferentintervals,forexample beforethe matingseason,whenthe predator must defendasuitable territorytoacquireanadequate physicalcondition for reproduction.Similarly,prey availabilityduringand afterthe birth season,when nutritionalrequirementsoffemalesrisingkittensare particularlyhigh,maybeimportant.Toevaluatethe importanceoflocalspatiotemporalpatterns ofprey abundanceonlynxbreeding,weevaluatedthefitof modelsusingrabbitdensityduringtheyear of reproductionandtheyearbeforereproduction.The landscapehypothesis,theshort-termpreyavailability hypothesisandacombinationofbothwereconfronted usingatheoretic-informationapproach.
Methods
Thestudypopulation
Thestudywas performedinDon˜ana,south-western Spain(3780?N, 6830?W), whereapopulationofthe Iberianlynxpersistswithametapopulationstructure. Severalintensiveradiotrackingprogramsbetween1984 and 2001 havealloweddetaileddescriptionsofthe metapopulationstructureand thespatialdistributionof breedingterritories(Palomares etal.2003).Resident individualsweredetectedin nineteenterritoriesdis- tributedamong ninedifferentnuclei,encompassingan areaofca2000km2.Thepopulationsizehasremained stablethroughoutthelast15yr.Inthepresentwork,we restricted the analysesto 13resident female lynx
territoriesdistributedinfive connectedpopulation nucleiwithahighprobabilityofterritoryoccupancy (Revillaetal.2004).Threeofthesenucleiwere insidea highly-protectedareainthe Don˜anaNationalParkand twointhe closesurroundings (Fig.1).Thedistribution and limitsof lynxfemaleterritorieshaveremained highlystableinthesepopulationsduringthedifferent radiotrackingprogramseventhoughdifferentindivi- dualshaveoccupiedthesameterritories.Wedefined territorylimitsusing80%-fixedKernelhomerange estimationsfromradiolocations ofonerepresentative residentlynxineachterritory,foratotaldatasetof
2397locations.Radiotrackingandterritoryestimation proceduresarefullydescribedinPalomaresetal.2000 andFerna´ndezetal.2003.
Datacollection
We carriedout systematictrack surveysto detect breedingwithinterritoriesoftheIberianlynxevery yearfrom2001to2004.Trackcensuses representa
Fig.1.DistributionoffemaleIberianlynxbreedingterri- toriesinfivesubpopulationsinDon˜ana,SWSpain.Thick linesrepresentlynxterritories.Tenterritorieshavebeenused forbreedingwithintheDon˜anaNationalPark(thinline): threeinCotodelRey(CR),fiveinReservaBiolo´gica(RB) andtwoinMarismillas(MA).Threeterritorieswerelocated outsidethePark,twoinAcebuche(AC)andoneinHato Rato´n (HR). Grey areasrepresentshrublandand forests favourableforlynxdispersal.
reliableand cost-effective method for the studyof mammaliancarnivoresinareaslikeDon˜anacontaining theappropriatesandysubstrate.Threetracksurveys wereperformedineveryterritoryeachyearbetween JuneandOctober, duringtheperiodwhenlynxkittens roamintheirmotherterritoriesafteremergingfrom theirnataldensandbeforedispersing(Ferna´ndezetal.
2002).Ineachsurvey,oneobservercoveredallsandy pathsandtrailsinsidethelynxterritory.Lynx tracks reportedthespeciespresence,andkitten tracks,smaller andoftentogetherwithadulttracks,indicatedpresence ofkittensafterthedenningperiod.Inaddition,most breedingrecords(70%)couldbealsoconfirmedfrom cubtaggingintheirnataldens,fromcameratrapping, and fromdirectobservations.Wedid not findany evidenceforbreedingoutsideareasdetectedfrom track surveys.
To examinewhether spatiotemporalpatterns in
breedingwererelatedtointer-annualvariationinprey abundance,weestimatedrelativerabbitabundancefor eachterritoryeveryyearfrom 2000 to 2004. This
estimationwasbasedon rabbit faecalpelletcounts
carriedoutbetweenJulyandSeptemberin246fixed, randomly-distributedplotsof1m2 (Ferna´ndez2005). Ithasbeenshownthatrabbitpelletdensityisstrongly
correlatedwithrabbitdensityandcanbereliablyused formonitoringrabbitsthroughoutlargeareas(Palo- mares2001a,b,Ferna´ndez 2005).Allplotswere sampledeveryyearexceptin2001,whenwecounted pelletsin a random subsampleof 160 plots. The number of sampledplots per territory-yearranged between10 and 53. To evaluatetheconsistencyof densityestimatesinrelationtosamplesizeweestimated the bootstrap distribution of differencesin mean densityusingsubsamplesof53and10plotsfromthe territorywiththelargestsample.Comparisons between simulatedpairsofsamplesrevealed no significant differencesandthedistributionofdifferencesshowed a smalldispersion(SDboot=0.059 for standardized databetween0and 1).Thissupportedthereliabilityof meandensitycomparisonsamongsamplesofdifferent size.
Totest how landscapestructureaffectsbreedingwe usedpredictionsofalandscape-basedhabitatmodelof
rabbitabundance(Appendix 1).This modelwas constructedfor the northern area of the Don˜ana NationalandNaturalparks,wherespatialvariations
inrabbitabundance werecloselylinkedtothelandscape structure(Ferna´ndez2005).Themodelpredictsthat themostfavourable landscapemosaicsforrabbitsare thoseintheproximitytostreams,pondsandlagoons andwithahighcoverofshrubsaswellasecotones betweenshrublandandpastureland.Moreover,ithas beenshownthattheshrubdensityandtheshrubland- pasturelandecotones arealsothemostimportant landscapefeaturesinfluencinghabitat selectionand sizeof territoriesin the Iberian lynx in Don˜ana (Ferna´ndezetal.2003).Therefore,therabbithabitat modelcanprovideastaticmeasureofthelynxhabitat qualitywithclearbiologicalmeaningbasedonavail- abilityofprey.Wemappedthepredictions fromthe rabbithabitatmodelonagrid of1-hahexagonalcells, representingthespatialscaleatwhichthismodelwas built. Then, weestimatedin everylynxterritorya landscapeindexindicativeofrabbitfavourabilityasthe averagevalueofallcellswithintheterritoryboundaries.
Statisticalanalyses
Weanalyzedspatiotemporalvariationsinthe European rabbitabundancewithinsamplingplotsinrelationto theterritoryandtheyear,includingtheirinteraction. Asignificantinteractionwouldindicatedifferencesin thedynamicsofrabbitabundanceamongthedifferent lynxterritories.Thisanalysiswasperformedusinga generalized linearmodel(GLM)withnegative-bino- mialerrordistributionandloglink(McCullaghand Nelder1989).
Then,wetestedwhetherbreedingwithinterritories wasrelatedtospatiotemporal variationsinthedensity
ofrabbits,tothelandscapestructureassociated with time-averaged preydensities,orboth.Wecompared breedinginaparticularyearPr(B)asafunctionof differentcombinations ofthe averagerabbitdensity withintheterritoryduringthatyear(i.e.parturition andkittenrisingperiods),theyearbefore(previousto thematingperiod),andthelandscapeindexdescribed above.Sincethesamelynxterritoriesweremonitored acrossallyearsweusedgeneralized linearmixedmodels (GLMM),anextensionof GLMthatallowsmodelling thecovarianceofrandomeffects(Littelletal.1996)We includedtheterritoryasarandomtermchoosingthe variance-componentsstructure for modellingcovar- iance.Pr(B)wascodedasa binaryresponseusing binomialerrordistributionandlogit link(McCullagh andNelder1989).Wecompared theexplanatory strengthofthedifferenthypothesesconfrontingaset ofcompetingmodels(Table1)whichincludedmost combinationsofvariablesexceptforsomeequations withrabbitdensityduringthebreeding year,because weconsideredthiseffectplausibleonlyincombination withpreydensitybeforethebreedingseason.
Detection of the most parsimonious ecological
hypothesis wasbasedon modelselectionprocedures usingtheAkaike’sinformationcriterionAIC(Burnham andAnderson1998).AICallowscomparingmultiple workinghypothesesandweightingfortheirlevel of supportinthedata.Thegoalistodetectthebesttrade- offbetweenmodelfitandnumberofparameters.We usedasecond-orderderivation,AICc,whichincludesa correctionforsamplesize,and selected the most parsimonioushypothesisaccordingtothemodelwith thelowestAICc.Wealsocalculated theAICc weight (wi) which reports the relativelikelihoodof every hypothesis normalizedacrossthesetofcandidate models.
Weadditionallymodelledtheeffectofthelandscape structureonthenumberofbreeding recordsper territoryusingmultinomialGLM.Inthisanalysiswe orderedtheresponsevariableintofivelevelsfrom0(no breedingduring the study period) to 4 (breeding inallfouryears).Then,wemodelledtheprobability thatbreedingisaboveaparticularlevelasafunction of the territory-averagedlandscapeindex of rabbit
favourability.Modelfitwascomparedwithananalo- gousequationusingtherabbitabundance ineach territoryaveragedoverthefiveyearsofdataasthe predictor.Thesetwoequationsarenon-exclusivemodel representations ofthehypothesisthat the mostim- portanthabitatattributesinfluencing lynxhabitat selectionalsoaffectbreedingfrequency.
AllGLMswereperformedusingtheGENMOD
procedurein the SASSystemfor Windowsv.8.02 (Anon. 1990). GLMMswerefitfirstusingthe GLIMMIXmacroforSAS(Littelletal. 1996).Model
selectionprocedureswerecodedusingtheRv.2.01free
statisticalsoftwareandtheLme4packageformixed models(Anon.2005,BatesandSarkar2006).
Results
ThenumberofIberianlynxbreedingrecordsranged between 7territories during 2002 (53.8%) and
4during2004(30.8%).Seventy-seven percentofall territoriesshowedevidenceofbreedingatleastoneyear,
whereasfouryearsofreproductionoccurredonlyin
2territories(15.4%).Incontrast,presenceofadultlynx wasrecorded,onaverage,within11(84.6%)territories everybreedingseason,confirmingthe highratesof
territoryoccupancy.
Thecorrelationbetweenthelandscapeindexderived
fromrabbithabitatmodels andthetime-averaged rabbit abundance was high (r=0.65; p=0.019; DF=11), confirmingthe valueof this index for
describingprey availabilitywithin lynx territories.
However, wealsorecordedsubstantial differences in preydensityamongyears,droppingfromof43.393.8
SErabbitpelletsm—2 inyear2000to11.691.7 in
year2004.GLMsshowedsignificanteffectsoftheyear, theterritoryandtheirinteraction(Table2),consistent
withhighspatiotemporalvariationsin rabbitabun-
danceandindicatingdifferentdynamicsamongbreed- ingterritories(Fig.2).
Table 1showstherelativefitofthedifferentGLMM
inthelynxbreeding data.Modelsincludingonly variations in rabbit densityas predictorshad low
support in the set, and their weightedlikelihoods
Table1.Setof generalizedlinearmixedmodelsonbreedingoccurrencewithinfemaleterritoriesoftheIberianlynx.Modelswere fitusingyearlybinomialbreedingdatafrom13territoriesinDon˜anaduringtheperiod2001—2004.The bestapproximating model ismarkedinbold.
Model—2LogLik / AICc / Akaikewi0 / Nullmodel(noeffect) / 62.47 / 66.46 / 0.07
1 / Preyabundance— previousyear / 59.70 / 65.70 / 0.09
2 / Preyabundance— presentandpreviousyear / 58.49 / 66.49 / 0.05
3 / Landscape-only / 56.18 / 62.18 / 0.51
4 / Preypreviousyearandlandscape / 55.63 / 63.63 / 0.21
5 / Fullmodel(presentandpreviousyear,landscape) / 55.23 / 65.23 / 0.07
Table2.Negative-binomialgeneralizedlinearmodelforthe Europeanrabbitabundancewithin breeding territoriesofthe Iberianlynx.
Model-effect TypeIIItests
2
Therandomeffectoftheterritorywasnotsignificantin any model(Z51.48;p]0.07).
MultinomialGLMsprovidedabetterrepresentation
oflandscapeeffectsonlynxbreeding(Table3).The soleeffectofthelandscapeindexonthenumberof
DFx
pbreedingrecordswithintheterritorywassignificantand
Year 3 65.90 B0.001
Territory 12 248.59 B0.001
Yearxterritory 24 122.72 B0.001
werenotqualitativelydifferent fromanullmodelof no-effect.Thisindicatedthat,inspiteoftheirlarge magnitude,yearlyvariations in preyavailabilitydid notexplain breedingrecords.TypeIIItestsalso confirmednon-significanteffectsofpreydensityduring the breedingyearorthe yearbefore(GLMMs:allx25
0.29;p]0.14).Incontrast,twohypothesesincluding the effectof the landscapeaccountedfor 70%
probabilityofselectionin the modelset(Tables 1
and3).Thelandscapeindex asasinglepredictorwas superiorwith51%probabilityofselection,representing thebestapproximatingmodelwiththelowestnumber ofparametersandthebestrelativefittothedata.The secondmodelincludedalsotherabbitabundanceinthe territoryduringthe seasonbeforematingandshoweda selectionprobabilityof21%.Therefore,althoughsome effectofshort-termpreyvariationincombinationwith thelandscapestructurecouldnotbe entirelydiscarded, itsrelevancewaslowincomparisonwiththe landscape.
showedanacceptablefittothedata(adjustedgeneral- ized coefficientof determination R2=0.49). The graphicalrepresentationofmodeleffectsillustratethe greatimpactoflandscapestructureon allbreeding response levelsandparticularlyonintermediatelevels, withamaximumprobabilityincrease up to 0.8for producinglittersmorethanoneyear (Fig.3).An alternativemodelshowedthattheeffectoftherabbit abundance averagedover time wasalsosignificant (TypeIII test,x2=4.42; p=0.03) althoughmodel fitwaslowerinthatcase(R2=0.36). Thiswasnot surprisinggiventhecorrelationbetweentime-averaged rabbit abundanceand the landscapeindexreported above.
Discussion
WefoundthatbreedingrecordsintheIberianlynx werecloselyassociated tolandscapevariablesinfluen- cingthemeanabundance ofitsprey.However, breedingdidnotshowaclearcorrelationwithyearly oscillationsinpreyavailabilitywithinlynxterritories. The hypotheses including only prey densities as
Fig.2.TemporalvariationsintheEuropeanrabbitabundancewithinterritoriesoftheIberianlynxbetweenyears2000and
2004.Territoriesaresortedonrabbitabundanceinyear2000.
Table3.Thetwobestgeneralizedlinearmixedmodels(GLMM)fortheyearlybreedingprobabilityandthemultinomialmodelfor breedingfrequencyintheIberianlynxmetapopulationofDon˜ana.ThetwoGLMMsaccountedfor 72%probabilityofselectionin themodelset.TheadjustedgeneralizedcoefficientofdeterminationforthemultinomialmodelwasR2=0.49.
Model-effect / Parameterestimate / SE / TypeIIItestsx2p
DF
BestapproximatingGLMM: Intercept / —5.97 / 2.44
Landscapeindex / 0.08 / 0.04 / 1 / 5.36 / 0.021
SecondbestGLMM
Intercept / —5.85 / 2.39
Meanpreyabundance(previousyear) / 0.22 / 0.40 / 1 / 0.31 / 0.583
Landscapeindex / 0.07 / 0.04 / 1 / 3.73 / 0.061
MultinomialGLM
Intercept1 / —4.72 / 2.06
Intercept2 / —6.12 / 2.28
Intercept3 / —6.94 / 2.43
Intercept4 / —8.09 / 2.61
Landscapeindex / 0.09 / 0.033 / 1 / 6.33 / 0.012
predictorswerenot supportedand onlyonemodel includingboth thelandscape structureand prey abundancebeforereproductioncouldnotbeunequi- vocallydiscarded.Thismodelwaslessparsimonious than thesimplernestedhypothesisconsideringonly landscapeeffects.Inapreviousstudy,Ferna´ndezetal. (2003)identifiedforthesame lynxpopulationthe landscapefactorsinfluencingbreedinghabitatselection andterritorysize, andtheyconcludedthatthe distribution and densityof territoriesrespondedto thespatialstructureofthevegetation.Noticeably,the significantlandscapepredictorsweresimilarin the presentstudy.Shrubcover(particularlyshrubstypical ofmaturevegetation)anddensityofecotonesbetween shrubsandpastureswerethemostsignificantpredictors
Fig.3.Relationship betweenbreedingfrequencywithin femaleterritoriesoftheIberianlynxandthelandscapeindex predictedfromthefittedmultinomialmodel.Theobserved rangeforthelandscapeindexwasscaledfrom0to1.N denotesthe number ofbreedingrecords(levels)foreach probabilitycurve.
ofhabitatselectionandconstitutedthelandscapeindex associatedto breedingrecords.Accordingto these results,habitatselectionandqualityforbreedingin theIberianlynxcanbeobservedastwoaspectsofa singlegradientconnectedto the landscapestructure and,ultimately,tothe averagepreyavailability.Indeed, theprobabilityoffindingalynxbreedingterritoryand thenumberofbreedingrecordsgraduallyincreasewith bettervegetationconditionsforrabbits.Thissupports theexistenceofahabitatqualitygradientintheIberian lynxassociatedto thelandscapestructureregulating territoryselection,densityandreproduction.
However,the amount of variabilityin breeding
recordsexplainedbythelandscapewasonlymoderate
(around50%attendingtothe generalizedcoefficientof determination).Thereprobablyexistacomplexvariety
ofotherfactors not consideredherethatmayaffect breedingsuchastheindividualqualityforreproduc-
tion,femaleexperience, probabilityoffindingmates,
etc.(Palomaresetal.2005).Theseeffectsaredifficultto test inmammaliancarnivoressuchastheIberianlynx,
whereindividualidentification andmonitoringis hamperedby their secretivehabitsand rarity.The
developmentofnon-invasivemethodsforidentification of individuals such as genetic typing from scats
(Taberletetal.1999)mayhelptoaddress these questionsinthefuture.
Differencesinbreedinghabitatqualityobservedin
thisstudyaredirectly relatedtoopportunitiesfor criticalfood resourcesin differentlandscapes.The
Iberian lynx strongly relieson the availabilityof
Europeanrabbitstofeed,apreyspecies thatconsti- tutes80% of its diet (Delibes1980) and limits
thepredatordistribution(Palomares2001a,b).Inthe presentstudy,theassociationbetweenthelandscape
structureandtheavailabilityoftheEuropeanrabbitwas
thebasisforthelandscapeindexthatbestexplained breedingrecords.However,yearlychangesin prey availabilitysuggestedthathabitatqualityforlynxwas not only heterogeneousin spacebut alsoin time. Indeed,weconfirmedsignificantdifferencesamong territoriesandyearsintherabbitabundance,witha generaldecliningtrendovertheyearsbutwithdifferent dynamicsinthedifferentterritories.Breedingoccurred insomelynxterritorieswithB5rabbitpelletsm—2,avery low density which probably falls below the previouslysuggestedthresholdofonerabbitperhectare forlynxreproduction(Palomaresetal.2001).Ithas beensuggestedthatshort-termvariationsinfoodsupply couldmodulatethe suitableconditions for mating,
pregnancyand kitten survivalin other lynxspecies (Breitenmoseretal.1993).Lowpreydensitymayalso leadtoreproductivefailurecausedby high postpartum kittenmortality,asshownfortheCanadalynxand otherfelidssubjecttoperiodsofpreydeclines(Brand etal.1976,Packeretal.1988,MowatandSlough
1998).Whilefluctuationsinrabbitabundancedidnot seemto influencesuccessful litteremergencein the Iberianlynxattheshorttime,wecannotrejectother effectsoffoodlimitationsuchasreducedproductionof kittens,increasedkitten mortality,etc.In addition, imperfectsampling ofpreyabundancecouldhave hinderedthedetection offine-tuningrelationships betweenyearlypreyvariations andlynxbreeding records.
There are severalnon-exclusivereasons whythe hypothesisofshort-termfood-supplyregulationwas notadequateforpredictingthespatiotemporaldistri- butionofbreedingrecordsintheIberianlynx.First, someterritorieshavebeenoccupiedbyresidentfemales thatdidnot breedduringyears ofoptimalprey availability(Palomares et al. 2005). Second,some females havenotreproducedinsometerritories in yearsofrelativelyhighpreyabundance.Forindividuals selecting a habitatto breed,it maybedifficult forecastingwhetherfoodlevelsinthenextseasonwill allowsuccessfulreproduction.Animalsmakinghabitat selectionandbreedingdecisionsoftenrelyonindirect cuesin theirphysical environmentto anticipatethe futurestate ofthehabitat(Cody1981,Oriansand Wittenberger1991).This isconsistent withthe associationoflynxbreedingfrequencywiththemore staticlandscapestructure,whichinformsontheaverage
rabbit abundancebut reportsimperfectlyon yearly availabilitybecausestochasticvariationsinabundance are alsolarge.Third,breedingmayoccurduringyears with lowrabbitabundancewhenresidentlynxattempt tooptimizelong-termterritoryholding.Thealternative ofsearching forotherbreedinghabitatsduringbad yearsisenergeticallycostlyandimpliesahighriskof mortality(Ferrerasetal.1992).Therefore,long-living animalsliketheIberianlynxmaybenefitbystayingin
theirterritoryand attemptingto breedevenduring unfavourableperiodsof preydecline(Breitenmoser etal.1993).
The presentstudyalsoillustrateshowthespatial analysisofavitalparametercanbeusedtounderstand the fitness consequencesof habitat selection.Few studieshaveanalyzedvitalrateswithindifferenthabitats inrelationtospatialandtemporalheterogeneityinthe environment(Sergioetal.2003,Mulleretal.2005), eventhoughtheoretical studiespredictthatspatial variation invitalrateshaveimportantdemographic consequencesforpopulations(Pulliam2000).Resource selectionfunctions,whichareoftenusedin spatial modelsof habitat selection(Boyceand McDonald
1999),canbeeasilyexpandedtoaddressthisrelation- ship,aswehaveshownusingGLMM.Thisinforma-
tion isparticularly relevantforpredictingthe demographic consequencesof habitat selectionin metapopulationswherethe spatialheterogeneityin populationparametersgreatlyinfluencestheirdemo- graphy.Unfortunately,thistypeofassessmentis limited bythelackofdetailedspatialdataonvitalratesfor manyspecies,especiallyforelusiveorrareanimalssuch astheIberian lynx.Yearlybreedingdataisoneofthe mostusefulparametersforevaluating thefitness consequencesofhabitatselection.Higher-qualitydata on breedingcouldalsobeusedto improvehabitat inferences;forexample,ifthenumberandsurvivorship oftheoffspring isalsoheterogeneousinspaceortime. TheintensivemonitoringprogramintheIberian lynx providedexceptionalinformationonthedistributionof littersthatemerged fromdens,butmoreprecise estimatesofbreedingsuccessincludingoffspringsurvi- valweredifficulttoobtain.
Theidentificationofenvironmental variablesrelated tospatiotemporalheterogeneityinreproductionrepre- sentanopportunityto improvedemographicmodel projectionsandtomanage breedinghabitatsinorderto increase the probabilityofpopulationpersistence. Indeed,viabilityanalysessuggestthatincreasingbreed- ingratesinlynxterritoriesis oneofthemosteffective measurestoassistthe speciesconservation(Gaonaetal.
1998).Thestrongdependenceoflynxhabitatselection andbreedingonthetime-averaged rabbitabundance impliesthatimprovingrabbithabitatisparamountto recover high-qualityhabitatsforlynx.Our results indicatethat increasingthe interspersionof shrub- pasturepatcheswillcontributetoincreasemeanprey
densityand lynxreproductionin currentlyavailable habitats.Onepriorityforlynxconservation istesting thesepredictionsexperimentally and quantifyingthe effectoflandscapemanagementonthespeciesrepro- duction.Last,habitatqualitycanbealsoinfluencedby survivalprobabilityin differentenvironments(Pease andMattson1999),anothercritical vitalratefor populationpersistence.Futureresearchshouldaimto
quantifyhabitat-specificreproductionand mortality simultaneouslyinordertoaccountforcriticalpopula- tionparametersindesigning habitatmanagement strategies.
Acknowledgements — ThisstudywasfinancedbytheSpanish Direccio´n General de Investigacio´n, through project BOS2001-2301, and sponsoredby Land Rover Espan˜a. N.Ferna´ndezwas supportedbyapredoctoralgrantofthe SpanishMinistryofScienceandTechnology,andaMarie Curie Fellowship providedbythe EuropeanCommission andhostedattheUFZ-CentreforEnvironmentalResearch Leipzig-Halle (ContractHPMD442 CT-2001-00109). We aregratefultoJ.Roma´n,A.Rodr´ıguez,E.Revillaandtothe staffoftheProjectLife-NaturefortheIberianlynx,Juntade Andaluc´ıa,forhelpingwithsampling.DouglasBruggeman kindlyreviewedthe English.CommentsfromJ.Calzada, andE.Revillaimprovedearlierversionsofthemanuscript.
References
Anon.1990.SAS/STAT user’sguide,ver.6.— SASInst. Anon.2005R:alanguageandenvironmentforstatistical
computing.— RfoundationforStatisticalComputing,
B
Bates,D.andSarkar,D.2006.Lme4:linearmixed-effects modelsusing S4classes.— B
Boyce,M.S.andMcDonald,L.L.1999.Relatingpopula- tions to habitats using resourceselectionfunctions.
— TrendsEcol.Evol.14:268—272.
Brand,C.J.etal.1976.Lynxresponsesto changingsnowshoe haredensitiesincentralAlberta.— J.Wildl.Manage.40:
416—428.
Breitenmoser,U.etal.1993.Predatorsofcyclicprey:isthe Canadalynxvictimorprofiteerofthesnowshoehare cycle? — Oikos66:551—554.
Burnham,K.P.andAnderson,D.R.1998.Modelselection and inference. A practical information-theoretic ap-
proach.— Springer.
Cody,M.L.1981.Habitatselectioninbirds.— Academic
Press.
Delibes,M.1980.FeedingecologyoftheSpanishlynxinthe
CotoDon˜ana.— ActaTheriol.25:309—324.
Ferna´ndez, N.2005.SpatialpatternsinEuropeanrabbit abundanceafterapopulationcollapse.— LandscapeEcol.
20:897—910.
Ferna´ndez,N. etal.2002.Theuseofbreedingdensand kittendevelopmentintheIberianlynx(Lynxpardinus).
— J.Zool.Lond.258:1—5.
Ferna´ndez,N.etal.2003.Identifyingbreedinghabitatforthe
Iberianlynx:inferencesfromafine-scalespatialanalysis.
— Ecol.Appl.13:1310—1324.
Ferna´ndez,N.etal.2006.Landscapeevaluationinconserva- tion:molecularsamplingandhabitatmodelingforthe Iberianlynx.— Ecol.Appl.16:1037—1049.
Ferreras,P.etal.1992.Ratesandcausesofmortalityin afragmentedpopulationofIberianlynxFelispardina
Temminck, 1824.— Biol.Conserv.61:197—202.
Fleishman,E.etal.2001.Modelingandpredictingspecies occurrenceusingbroad-scaleenvironmental variables:an examplewithbutterfliesoftheGreatBasin.— Conserv. Biol.15:1674—1685.
Franklin,A.B.etal.2000.Climate,habitatquality,and
fitnessin northern spottedowlpopulationsin north- westernCalifornia.— Ecol.Monogr.70:539—590.
Gaona, P. et al. 1998. Dynamics and viabilityof a
metapopulationof the endangeredIberianlynx(Lynx pardinus).— Ecol.Monogr.68:349—370.
Guisan, A. and Thuiller, W. 2005. Predicting species
distribution:offeringmorethansimplehabitatmodels.
— Ecol.Lett.8:993—1009.
Hirzel,A.H.etal.2002.Ecological-nichefactoranalysis:how
tocomputehabitat-suitabilitymapswithoutabsencedata?
— Ecology83:2027—2036.
Johnson, C. J. et al. 2004. A quantitativeapproachto
conservationplanning: usingresourceselectionfunctions tomapthedistributionofmountaincaribouatmultiple spatialscales. — J.Appl.Ecol.41:238—251.
Korpima¨ki,E.1988.Effectsofterritoryqualityonoccupancy,
breedingperformanceandbreedingdispersalinTeng- malm’sowl.— J.Anim.Ecol.57:97—108.
Levin,S.A.1992. The problemofpattern and scalein
ecology.— Ecology73:1943—1967.
Littell,R.C. etal.1996.SASsystemformixedmodels.
— SASInst.
Manel,S.etal.2000.Testinglarge-scalehypotheses using
surveys:theeffectsoflanduseonthehabitats,inverte- bratesandbirdsofHimalayanrivers.— J.Appl.Ecol.37:
756—770.
McCullagh,P.andNelder,J.A.1989.Generalizedlinear
models,2nded.— ChapmanandHall.
Mowat,G.andSlough, B.G.1998.Someobservationsonthe
naturalhistoryandbehaviouroftheCanadalynx,Lynx canadiensis.— Can.FieldNat.112:32—36.
Muller,M. et al.2005. Ecologicaland socialeffectson
reproductionand localrecruitmentin the red-backed shrike.— Oecologia143:37—50.
Orians,G. H. andWittenberger,J.F.1991.Spatialand
temporalscalesinhabitatselection.— Am.Nat. 137, suppl.:S29—S49.
Packer,C.etal.1988.Reproductive successinlions.— In:
Clutton-Brock,T.H.(ed.),Reproductivesuccess.Studies ofindividualvariationincontrastingbreedingsystems. Univ.ofChicagoPress,pp.363—383.
Palomares,F.2001a.Comparisonof3methodstoestimate
rabbit abundance in a Mediterranean environment.
— Wildl.Soc.B29:578—585.
Palomares,F.2001b.Vegetationstructureandpreyabun-
dancerequirementsoftheIberianlynx:implicationsfor thedesignofreservesandcorridors. — J.Appl.Ecol.38:
9—18.
Palomares,F. et al.2000. Iberianlynxin a fragmented
landscape:predispersal,dispersal,andpostdispersalhabi- tats.— Conserv.Biol.14:809—818.
Palomares,F.etal.2001.Spatialecologyof Iberianlynxand
abundanceofEuropeanrabbitsinsouthwesternSpain.
— Wildl.Monogr.148:1—36.
Palomares,F.etal.2003.Estructuradelametapoblacio´n de lincesde Don˜ana. — In: In Memoriam al Prof. IsidoroRuizMart´ınez.Univ. de Granada, pp. 505—
526.
Palomares,F.etal.2005.Reproductionandpre-dispersal survivalofIberianlynxinasubpopulationoftheDonana NationalPark.— Biol.Conserv.122:53—59.
Pease,C. M. and Mattson, D. J.1999. Demographyof
theYellowstonegrizzlybears.— Ecology80:957—975. Penteriani,V. etal.2004.Spatialheterogeneityandstructure
ofbirdpopulations:acaseexamplewiththeeagleowl.
— Popul.Ecol.46:185—192.
Peterson,A.T.etal.2002.Distributionalpredictionbasedon ecological nichemodelingofprimaryoccurrencedata.
— In: Scott, J. M. et al. (eds), Predicting species occurrences:issuesofscaleandaccuracy.IslandPress,
pp.617—623.
Pulliam,H.R.2000.Ontherelationshipbetweennicheand distribution.— Ecol.Lett.3:349—361.
Pulliam,H.R.andDanielson,B.J.1991.Sources,sinks,and habitatselection:alandscapeperspectiveonpopulation dynamics.— Am.Nat.(suppl.)137:S50—S66.
Revilla,E. et al. 2004. Effectsof matrix heterogeneity
onanimaldispersal:fromindividualbehaviortometa- population-levelparameters.— Am.Nat. 164:E130— E153.
Saab,V.1999.Importanceofspatialscaletohabitatuse by
breedingbirdsinriparianforests:ahierarchicalanalysis.
— Ecol.Appl.9:135—151.
Sergio,F.etal.2003.Spatio-temporalshiftsingradientsof habitat quality for an opportunistic avian predator.
— Ecography26:243—255.
Southwood, T.R.E.1977.Habitat,thetempletforecological strategies? — J.Anim.Ecol.46:337—365.
Taberlet,P.etal.1999.Noninvasivegeneticsampling:look
beforeyouleap.— TrendsEcol.Evol.14:323—327.Wiens, J.A.1976.Populationresponsestopatchy environ-
ments.— Annu.Rev.Ecol.Syst.7:81—120.
DownloadtheappendixasfileE5005from
B