<%BANNER%>
PRIVATE ITEM
Digitization of this item is currently in progress.
Population distribution models: species distributions are better modeled using biologically relevant data partitions
CITATION PDF VIEWER
Full Citation
STANDARD VIEW MARC VIEW
Permanent Link: http://ufdc.ufl.edu/AA00012284/00001
 Material Information
Title: Population distribution models: species distributions are better modeled using biologically relevant data partitions
Series Title: BMC Ecology
Physical Description: Book
Language: English
Creator: Gonzalez, Sergio C.
Soto-Centeno, J. Angel
Reed, David L.
Publication Date: 2011
 Notes
Abstract: Background: Predicting the geographic distribution of widespread species through modeling is problematic for several reasons including high rates of omission errors. One potential source of error for modeling widespread species is that subspecies and/or races of species are frequently pooled for analyses, which may mask biologically relevant spatial variation within the distribution of a single widespread species. We contrast a presence-only maximum entropy model for the widely distributed oldfield mouse (Peromyscus polionotus) that includes all available presence locations for this species, with two composite maximum entropy models. The composite models either subdivided the total species distribution into four geographic quadrants or by fifteen subspecies to capture spatially relevant variation in P. polionotus distributions. Results: Despite high Area Under the ROC Curve (AUC) values for all models, the composite species distribution model of P. polionotus generated from individual subspecies models represented the known distribution of the species much better than did the models produced by partitioning data into geographic quadrants or modeling the whole species as a single unit. Conclusions: Because the AUC values failed to describe the differences in the predictability of the three modeling strategies, we suggest using omission curves in addition to AUC values to assess model performance. Dividing the data of a widespread species into biologically relevant partitions greatly increased the performance of our distribution model; therefore, this approach may prove to be quite practical and informative for a wide range of modeling applications.
 Record Information
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution.
Resource Identifier: doi - 10.1186/1472-6785-11-20
System ID: AA00012284:00001

Downloads

This item is only available as the following downloads:

( PDF )


Full Text

PAGE 1

RESEARCHARTICLE OpenAccessPopulationdistributionmodels:species distributionsarebettermodeledusing biologicallyrelevantdatapartitionsSergioCGonzalez1,3 ,JAngelSoto-Centeno1,2* andDavidLReed1AbstractBackground: Predictingthegeographicdistributionofwidespreadspeciesthroughmodelingisproblematicfor severalreasonsincludinghighratesofomissionerrors.Onepotentialsourceoferrorformodelingwidespread speciesisthatsubspeciesand/orracesofspeciesarefrequentlypooledforanalyses,whichmaymaskbiologically relevantspatialvariationwithinthedistributionofasinglewidespreadspecies.Wecontrastapresence-only maximumentropymodelforthewidelydistributedoldfieldmouse( Peromyscuspolionotus )thatincludesall availablepresencelocationsforthisspecies,withtwocompositemaximumentropymodels.Thecomposite modelseithersubdividedthetotalspeciesdistributionintofourgeographicquadrantsorbyfifteensubspeciesto capturespatiallyrelevantvariationin P.polionotus distributions. Results: DespitehighAreaUndertheROCCurve(AUC)valuesforallmodels,thecompositespeciesdistribution modelof P.polionotus generatedfromindividualsubspeciesmodelsrepresentedtheknowndistributionofthe speciesmuchbetterthandidthemodelsproducedbypartitioningdataintogeographicquadrantsormodeling thewholespeciesasasingleunit. Conclusions: BecausetheAUCvaluesfailedtodescribethedifferencesinthepredictabilityofthethreemodeling strategies,wesuggestusingomissioncurvesinadditiontoAUCvaluestoassessmodelperformance.Dividingthe dataofawidespreadspeciesintobiologicallyrelevantpartitionsgreatlyincreasedtheperformanceofour distributionmodel;therefore,thisapproachmayprovetobequitepracticalandinformativeforawiderangeof modelingapplications.BackgroundSpeciesdistributionmodeling(SDM)hasbecomea commontoolforunderstandingspatialdistributionpatternsofbiodiversityworldwide[1-4].ThegoalofSDM istobuildamodelpredictingtherelativeprobabilityof occurrenceofaspeciesacrossgeographicspacecommonlyusingenvironmentaldata(i.e.climate,vegetation, soil,etc.)andadatasetofknownpresenceorpresence/ absencelocalities.Theterms ecologicalnichemodel, environmentalnichemodel ,and speciesdistribution model haveallbeenusedtodescribethistypeofmodelingintheliterature;forthesakeofsimplicitywewill usespeciesdistributionmodeling.SDMtechniquescontinuetoevolvewithanincreasinglybroadrangeof applicationsfromconservationplanning[5,6],topredictingspeciescolonizationandabundance[4,7,8],predictingdiseaseoutbreaks[1],andunderstanding phylogeographicpatterns[ 9].Methodsofproducing SDMsvarywiththetypeofdataavailable,purpose,and softwareused. Thereisadirectlinkbetweenclimateandthedistributionofplantspecies[10].Becauseclimateisacausalfactorinthedistributionofplantspecies(andplantspecies assemblages),climaticpatternsatvariousspatialscales directlyaffecthabitattypesandcommunityproductivity. Thus,climateisconsideredaproxyforagivenspecies environmentalniche.Becauseofthevariablesinvolvedin buildingSDMs,itisimportanttokeepinmindthat SDMsarepredictingaspecies fundamentalnicheas *Correspondence:sotocenteno@ufl.edu Contributedequally1FloridaMuseumofNaturalHistory,DivisionofMammals,Universityof Florida,DickinsonHall,Gainesville,FL32611,USA FulllistofauthorinformationisavailableattheendofthearticleGonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 2011Gonzalezetal;licenseeBioMedCentralLtd.ThisisanOpenAccessarticledistributedunderthetermsoftheCreative CommonsAttributionLicense(http://creativecommons.org/licenses/by/2.0),whichpermitsunrestricteduse,distribution,and reproductioninanymedium,providedtheoriginalworkisproperlycited.

PAGE 2

probabilityofoccurrence[11, 12],nottherealizeddistribution,whichisaffectedbymanyextrinsicfactorsthat maynotbeaccountedforinthemodel.Thefundamental nichesofspeciesareconsideredtobeconservedover evolutionarytime[11],whichhasallowedclimate-based SDMstobesuccessfulinpredictingtheoccurrenceof speciesorcloselyrelatedspeciesatpreviouslyunsampled localities[13-16]. Modelingspecieswhosedistributionsspanlargeenvironmentalorhabitatvaria tionmaybeproblematic becausedistributionmodelstendtohavehigherratesof omissionerror(i.e.underprediction)inthepredicted speciesdistributions[2,17,18].Insuchcases,themodels mayindicateregionalspecializationofperipheryorisolatedpopulations.Commissionerrors(i.e.theoverpredictionofdistributions)mayresultfromarestrictionof therealizeddistributionduetobioticinteractionsor geographicbarrierstorangeexpansion.Inanattemptto overcomeomissionerrorsinSDMsforwidespreadspecies,Osborne&Suarez-Seoane[17]modeledspecies distributionsbyspatiallypartitioningtheirdataintogeographicquadrantsandintoconcentricringstomodel eachdatapartitionseparately.Hernandezetal.[18]suggestedthatfutureresearchshouldfocusonmodeling broaddistributionsinsubunitsthatarebasedondistinct geneticlineagesorrecognizedsubspecies. Mostspeciesinthegenus Peromyscus arewidespread, withapositivecorrelationbetweenspeciesrangeand numberofrecognizedsubspecies[19],suggestingthat localspecializationiscommonwithinspeciesofthis genus.With15recognizedsubspecies[20]andgeneticallystructuredpopulations, P.polionotus isanexcellent modelspeciesfordevelopingnewmethodsofdatapartitioningtoovercometheproblemsassociatedwithmodelingthegeographicdistributionsofwidelydistributed species. Theoldfieldmouse( Peromyscuspolionotus ),also knownregionallyastheFloridabeachmouse,iswidespreadthroughoutthesoutheasternUnitedStates(Figure1).Morphologicalandgeneticdifferenceshavebeen documentedbetweensubspecies[21-25].Molecularevidencesuggeststhatlittleornogeneflowoccurs betweenthehighlystructuredpopulationsonislands alongFlorida spanhandle[25]orbetween P.p.rhoadsi Figure1 P.polionotusinsoutheasternUnitedStates .Distributionmapofthe15supspeciesof P.polionotus redrawnfromHall(1981). Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page2of10

PAGE 3

onFlorida scentralridge,and P.p.niveiventris ,onthe Atlanticcoast[21].Highlydifferentiatedpopulationsof Peromyscus aresuggestedtobetheresultofstronglocal adaptation[19,23,26].Coastaldunepopulationsalong theGulfofMexicoarephenotypicallymoresimilarto populationsalongtheAtlanticCoast(especiallyincoat color)thantoneighboringpop ulations,suggestingthey areundersimilarselectivepressuresintheirdisjunct coastalenvironments[23].Throughoutitsrange, P. polionotus isthreatenedbydevelopmentandinvasive speciesexertingcompetitionandpredationpressureson itspopulations,andthesubspecies P.p.decoloratus has beenconsideredextinctsince1950[19,26]. Giventheevidenceforlocaladaptationinthiswidespreadspecies,previousresearchwouldpredictthatan SDMoftheentirespecieswouldunderpredictthegeographicdistributionof P.polionotus .Amoreaccurate predictionofthespecies distributionmayresultfroma biologicallyinformedspatialpartitioningoflocalitydata. Totestthishypothesis,wemodeledthedistributionof P polionotus inthreeways;wemodeledthewholespeciesdistributionatonce,we partitionedlocalitydata intofourgeographicquadrantsfollowingOsborne& Suarez-Seoane[17],andwepartitionedlocalitydataby the15recognizedsubspeciesof P.polionotus .ResultsAllindividualandcompositemodelsproducedAUC valuesabove0.84,whichareconsistentwithAUCvalues reportedintheliteratureforothertaxa[9,17,18,27]. DespitehavingahighAUCvalue(0.899),themodel basedontheentiredatasetfailedtopredicttheoccurrenceof P.polionnotus inplaceswhereitisclearly knowntooccur(Figure2a),mostobviouslyomittingthe distributionsofthesubspecies P.p.colemani and P.p. polionotus.Partitioningthedatabygeographicquadrants(Figure2b)andbysubspecies(Figure2c)producedmodelsthatareprogressivelybetter,bothin termsofpredictingtheknowndistribution,andin termsoftheirAUCscores.TheaverageAUCvalueof theindividualmodelsusedtobuildthegeographic quadrantcompositemodelwas0.927,whereastheaverageAUCvalueoftheindividualmodelsthatwereused tobuildthesubspeciescompositemodelwas0.976. PartitioningthedataintogeographicquadrantsproducedfourmodelswithAUCscoresof0.844,0.968, 0.905,and0.993(clockwisefromnortheast;Figure3). Thequadrantcompositemodel(Figure2b)predicted highprobabilitiesofoccurrenceinareasthatthefull speciesmodel(Figure2a)hadomitted.However,the quadrantcompositemodel(Figure2b)showedpoor resolutioninpartsofnorthernGeorgiaandpeninsular Florida.Whenthedatasetwaspartitionedaccordingto thecurrentlyrecognizedsubspecies(Figures4and5), eachsubspeciesmodelperformedwellbasedonAUC scores.Twelveoutof15modelshadAUCscores between0.97and1.0;theexceptionsbeing P.p.colemani (0.917), P.p.polionotus (0.851),and P.p.trissyllepsis (0.5).Thepoorperformanceofthemodelfor P.p. trissyllepsis wasduetoinsufficientdata(n =2)forthe populationandwasomittedfromthecompositemodel. Figure2 Maxentdistributionsfor P.polionotus usingdifferent datapartitions .Predictedspeciesdistributionsusingpresencedata fortheentirespeciesatonce(A),bydividingthepresencedata intoquadrants(B),andbysubspecies(C). Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page3of10

PAGE 4

Althoughsuperficiallyallmodelsdevelopedfromdata partitionsseemtoperformwellbasedonAUCvalues, ratesofomissionbetweenthesemethodsshowadifferent perspective.Figure6showsthreeomissioncurvesforthe whole-speciesmodel(Figure6a),thequadrantmethod (Figure6b)andthesubspeciesmodel(Figure6c).The curvesshowomissionerror(Y-axis)asafunctionofpredictedprobabilityofoccurrenc e(X-axis).Betterperforming modelsbasedonthelogisticoutputofMaxenthavefewer omissionerrorsevenaspredic tedprobabilityofoccurrence reachesmaximalvalues.Thewhole-speciesmodel(Figure 6a)andthequadrantmodel(Figure6b)haveomission errorratesthatincreaselinearlywithincreasingprobability ofoccurrence.Incontrast,thesubspeciesmodel(Figure 6c)hasrelativelylowomissionerrorratesthatonlybegins toincreasewhenpredictedprobabilityofoccurrence reacheshighervalues,whichispreferable.DiscussionWesuggestthatpartitioningdatainabiologicallymeaningfulway(asopposedtogeographically)canhelpto overcomehighomissionratesindistributionmodelsof widespreadspecies[17,18,28].Despitehavingarelatively highAUCvalue(0.899),themodelbuiltuponthewhole species distributionfailedtopredictknownlocalitiesof P. polionotus ,whichcanbeobservedbycomparingFigure1 toFigure2a.Spatiallypartitioningthedataintoquadrants producedamuchbetterdistributionmodelaftercombiningfourregionalmodelswithAUCvaluesrangingfrom 0.844to0.993,whichcanbeseenbycomparingFigure1 toFigure2b.The14AUCscoresfortheSDMsbasedon subspeciespartitioningrangedfrom0.851to1.0,whichis notsubstantiallydifferentfromAUCvaluesobtainedfor thequadrantmodels.However,theaccuracyandincreased resolutionofthecompositeofthesubspeciesmodelscan beseenincomparingthethreepanelsofFigure2andby comparingFigure1toFigure2c.LimitationofAUCvaluesforassessingpredictive performanceAUCvaluesarecommonlyusedasindicatorsofmodelfit [9,17,18,27],andhighvaluesforallthreemethodsinour Figure3 Maxentdistributionbasedonthequadrantmethod .Predictedspeciesdistributionof P.polionotus estimatedfrompresencedata modeledseparatelyinfourquadrants(A-D). Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page4of10

PAGE 5

studywouldsuggestthateachmethodproducedhighly accuratemodels.Furthermore,themodestincreasein AUCvalueswithevergreaterdatapartitioningwouldsuggestthateachsuccessivepartitioningschemeproduced,at best,onlyslightlybetterfittingmodels.However,thisfindingismisleadingwhenonecomparesthepredicteddistributionstotheknowndistributionforthespecies.The trendofincreasingAUCscoresmayindicatethedirection ofchangeinaccuracy,butitfailstocapturethemagnitude ofimprovementinthepredicteddistributionsofthequadrantcompositemethodandthesubspeciescomposite method.Thisfailureis,inpart,duetothefactthatthe AUCscoresforthegeographicquadrantmethodandsubspeciesmethodarecompositesof4and14combined models(respectively),andtheaccuracy(orinaccuracy)of individualmodelsiscompoundedwhentheyarecombined.Thisfactalone,however,cannotfullyexplainthe observeddiscrepancybetweenthevastlyimprovedmodel predictionandthemodestlybetterAUCscores. AUCvaluescanbemisleadingwhenassessingamodel spredictiveabilityforseveralreasons.TheAUC measuresdiscriminationandnotaccuracy perse ,thus ignoringthegoodnessoffitofamodel[29].TheAUC valuealsotakesintoaccounttheperformanceofthe modelattheextremeleft(aswellastheright)ofthe ROCcurve(see[29]foradetails),aregionthatisnot operationallymeaningfulinourcase.Weareonlyinterestedinthresholdsofpredictedprobabilityofoccurrencegreaterthan0.50becausethatwouldequalthe probabilityofoccurrenceofanullmodel.Thisinclusion oftheareaundertheextremeleftoftheROCcurvecan inflateAUCvalues,whichcanbefurtherinflatedwhen thetotalgeographicextentofthemodelisconsidered. Iftheratiobetweenareasofpresenceandthetotal extentishigh,truepositivesaremorelikelytooccurby chancealone[28].Becausethisratiochangeswitheach oftheindividualmodelsbuiltondifferentdatapartitions,AUCvaluesmaynotbeusefulinaccuratelycomparingrelativemodelperformancebetweenoramong oursubspeciesandregionalmodels. ItisalsopossiblethattheinflationofAUCvalues observedinthemodelswepresentresultsfromthe Figure4 Maxentdistributionof P.polionotus onpeninsularFlorida .Predictedspeciesdistributionof P.polionotus estimatedfrompresence dataofeachmainlandandpeninsularsubspecies(A-G)analyzedseparately. Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page5of10

PAGE 6

interactionbetweengeographicandenvironmental space.Becauseofthenarrowgeographicspaceat whichmostsubspeciesof P.polionotus occur,locality informationisgeographicallyclustrered;therefore,the environmentalspacesampledbythemodelsmayshow spatialautocorrelationinsomeoftheenvironmental variablesused.InSDM,spatialautocorrelationoccurs whenthevaluesofthevariablessampledatnearby locationsarenotindependentfromeachother[30] andasaresult,measuresofaccuracy(e.g.AUC)can beinflated[31,32].Inourcase,thegeographicclusteringofnarrowlydistributedsubspeciesof P.polionotus maycausespatialautocorrelationandthusinflateAUC values(seeFigure4and5).Nevertheless,withinthese narrowextents,weincludedsamplesspanningthe entiregeographicspacerepre sentingasignificantportionoftheenvironmentalspaceoccupiedbyeachsubspecies.Theresultingmodelsareaccuratetothetrue distributionofthesubspeciesandareabletodetect evensubtlelocalenvironmentalconditionslikely affectingeachsubspeciesdifferently,despitetheseeminglygeographicclustering.Thisfurtheremphasizes thepointthatAUCvaluesprovideanunreliableway toaccuratelycomparerelativemodelperformance.The onlyexceptionstotheissueofinflatedAUCvaluesin ourdatasetare P.p.colemani (AUC=0.917)and P.p. polionotus (AUC=0.851),whicharetheonlytwo widelydistributedsubspeciesspanningamoreheterogeneousenvironmentalspacewherelocalityinformationforthesubspeciesisnotgeographicallyclustered (Figure2).Becauseofthelargergeographicspace occupiedbythesesubspecies,theresultingmodels from P.p.colemani and P.p.polionotus areunlikely affectedbyspatialautocorrelationandthereforedonot showinflatedAUCvalues. Finally,theAUCdoesnotprovideinformationasto thespatialdistributionoferrors.Italsoweighsomission andcommissionerrorsequally,bothofwhichvaryin interpretivemeaningandimportancewiththeintended useofthemodel[29].Becausewedonothavetrue absencedata,wecannotquantifyourcommissionerror rate.However,theomissioncurveshowshowwellthe modelperformsatdifferentthresholds(i.e.thedistributionofomissionerrors).Therefore,theomissioncurve canbeasimportantastheAUCvalueintermsofassessingmodelperformance,ifnotmore-so.Amodelwith Figure5 Maxentdistributionof P.polionotus onFloridapanhandle .Predictedspeciesdistributionof P.polionotus estimatedfrompresence dataofeachFloridapanhandlesubspecies(A-G)analyzedseparately. Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page6of10

PAGE 7

relativelyloweromissionerrorsathigherpredicted probabilitiesofoccurrenceispreferred.QuadrantversussubspeciespartitioningBecauseMaxentdrawspseudo-absencedataatrandom tocalculateAUCscores,itispossiblethatinourquadrantanalysisitdrewfalsepseudo-absencesfromareas outsidethequadrantbeingtested,especiallyinthecase ofthenorthwestquadrant.W hetherthisoccurred,and ifitdid,whetheritcontributedtotheobservedunderpredictionisdebatable.First,therewasfarlessunderpredictionintheotherthreequadrants.Forexample, themodelforthesouthwestquadrant,whichincludes peninsularFloridapredictedoccurrencesinregions wherethespeciesdoesnotoccur.Second,thetwo northernquadrantscoverroughlythesamegeographic areaastwoofthesubspecies( P.p.colemani and P.p. polionotus ),yetthetwonorthernquadrantsfailedto predictareasofknownoccurrencethatthetwosubspeciesmodelspredictedaccurately.Itispossiblethatpoor quadrant-basedmodelsresultedbecausepseudoabsenceswerebeinggeneratedbyMaxentinareaswith truepresencesforeitherofthetwosubspeciesjustoutsideofthequadrantbeingmodeled.Thatis,thearbitraryboundarybetweenthequadrantsobfuscates biologicallymeaningfulboundariesbetweenpopulations orsubspecies.Thus,inusingq uadrant-basedpartitioning,nicheinformationwaslostforthetwosubspecies, whichemphasizesourpointthatbiologicallyrelevant datapartitioninginformsspeciesdistributionmodels.MoleculardataandpopulationdistributionmodelsItiseasytoseehowmoleculardata,suchasDNA sequences,canbeusedtodelineatebiologicallymeaningfulgroups(i.e.clades)withinaspecies,andthat thosecladesmightbepartitionedseparatelyforspecies distributionmodeling,especiallyiftheyaregeographicallydiscrete.But,justasmoleculardatacanimprove methodsofgeneratingSDMs, thefindingsassociated withSDMscanalsoinformtheworkdonebymolecular biologistsstudyingpopulationgenetics,phylogenetics,or phylogeography.WhenSDMsarenonoverlappingfor populationswithinaspecies,theymayberevealing crypticpatternsofdivergencethatwouldbeinteresting tostudywithmoleculardata.Conversely,whenmoleculardatauncoverpopulationstructureorlimitstogene flow,SDMscanbeusedtotesthypothesizedmechanismsofdivergencesuchasnichedifferentiation.ExaminingbothmoleculardataandSDMstogetherhasbeen exploredonlyrecently[6,9]. BuildingSDMsfor P.polionotus bypartitioningdata intosubspeciesandbuildingacompositedistribution modelmitigatedtheproblemofhighomissionratesthat usuallyoccurswhenmodel ingthedistributionsof widelydistributedspecies.ThissuggeststhattheSDM basedonbiologicallyrelevant partitions(subspeciesin ourcase)couldaccommodatevariabilityinthenichesof subspecies,whereasmodelin gthewholespeciesdistributiontogethercouldnot.Thisissupportedbythefact thatspatialpartitioningofdataintoquadrantsproduced modelsthathadregionsofbothunder-andover-prediction(Figure3),whereasthem odelsbasedonpartitioningbysubspeciesshowednos ignsofunderprediction andonlymodestoverlapinthedistributionsofadjacent individualsubspeciescausedbyoverprediction(Figure 5).Theevidenceofhighleve lsofpopulationstructure betweenlocallyadaptedpopulations[19,21-23,25,26] mightbedrivingtheimprovementweseeinthecompositemodelbasedonsubspeciesdistributions.Inour case,weshowthatmoleculardataatthepopulation levelimprovedmodelaccu racy.Furthermore,inthe absenceofdetailedmolecularinformationonthepopulationsstudied,researcherscouldgeneraterelevantdata partitionsusingalternativedatasourcessuchas Figure6 Omissioncurvesasameansofassessingmodel performance .Omissioncurve(thingreenline)forthespecieslevel modelof P.polionotus (A),thenorthwestregionofthequadrant method(B),andthesinglesubspecies P.p.phasma ofthe subspeciesmethod(C).MeanAUCvalueforreplicaterunswas 0.899(std.dev.=0.105),0.905(std.dev.=0.119),and1.0(std.dev.= 0.001),respectively. Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page7of10

PAGE 8

subspeciesdelimitations,morophologicaldifferencesor otherphenotypictraits.Phylogeographicimplicationsfor P.polionotusGuisan&Zimmermann[28]encouragecollaboration withevolutionarybiologistsandpopulationgeneticists incaseswherewidespreadspeciesarebeingmodeled. Morerecently,Rdderetal.[9]discussedhowavariety oftechniquesincludingmolecularecologyandenvironmentalnichemodelingcanbecomplimentaryin answeringphylogeographicquestions.Thecaseissuch here,whereourmethodofpartitioningdatawasbased largelyontheliterature,whichincludespopulation geneticstudiesthathavebeenconductedon P.polionotus .Conversely,asmolecularworkhelpedtoinformour models,ourmodelsalsoshedlightonandconfirm resultsofstudiesonthespecies geneticstructure,and possiblyevolutionarytrajectory.Forexample,theSDMs for P.p.rhoadsi and P.p.nivieventris donotoverlap andaregeographicallydiscrete(Figure4and2c),which isconsistentwiththegeneticresultsofDegneretal. [21]andcurrenttaxonomy(Figure1).Thesoutheastern quadrantmodel,however,hasverypoorresolutionof thisfinescaledistinction(Figure3c).Similarly,themodelsfor P.p.polionotus and P.p.colemani capturethe knownextentoftheirrespectiveranges(Figure5),while thetwonorthernquadrantmodelsdonot(Figure3,a and3b),implyingthatthesetwosubspeciesoccupydifferentclimaticniches. Climateappearstoplayanimportantroleindefining inlandandinlandvs.coastalsubspecies(e.g.Figure4AD).However,despitetherebeinggeneticdifferencesin thebeachmousesubspecieslocatedintheFloridapanhandle(Figure5A-F),therewasconsiderableoverlapin theirpredicteddistributi ons,suggestingthatclimate maynotbetheprimaryfactordefiningtherangeof thesesubspecies.Predictednicheoverlapusually occurredbetweenadjacentcoastalsubspecies(Figure 5A-F)andonlyoncebetweencoastalandinlandsubspecies(see P.p.albifrons and P.p.sumneri ,Figure5F-G). Studieshaveshownthatthecoastalbeachmousepopulationsreflectpatternsof localadaptationandstrong selectionfavoringcrypticcoloration[22-25].Therefore, incaseswhereclimatichabitatonadjacentcoastalbeachesmightbesimilar,vicariance(i.e.coastalinlets)and strongselectionforcoatcolorationaremorelikelythan climatetomaintainthedistinctivenessofcoastalbeach mousesubspecies.ConclusionsUsingabiologicallymeaningfulmethodofpartitioning thedatafromwidelydistributedspeciesgenerateda compositeSDMof P.polionotus thatmoreaccurately reflectedtheknowndistributionofthespeciesthanthe processofanalyzingthewholespeciesatonceorpartitioningthedataintogeographicquadrants.Osborne andSuarez-Seoane[17]notethatgeographicbaseddata partitioning(e.g.quadrants)maynothaveworkedwell duetotheabsenceofanybiologicalbasisforpartitioning.Wecontendthatourstudyconfirmsthatstatement. WealsoprovideanexampleofhowSDMscanbeboth informedbyaswellasinformphylogeographicstudies atthepopulationandspecieslevels.Modelingawidespreadspeciesusingbiologicallymeaningfuldatapartitionshasthepotentialtogreatlyincreasethe performanceofdistributionmodelswhileonlyrequiring basicmanipulationinGISsoftware.Thus,thistechnique mayprovetobequitepracticalforawiderangeof modelingapplications.Despitetheincreasinguseand popularityofENMs,acompletelyobjective,accurate, andfullyacceptedmeasureofperformanceofpredictive distributionmodelsisstillelusive[9,27,29].Wesuggest usingboththeAUCandomissioncurveonacontextual basistoassessmodelperformance.MethodsWecreatedspeciesdistributionmodelsfor P.polionotus in MaxentusingtheWorldClimclimatelayers.Maxentuses theprincipleofmaximumentropydensityestimationto generateaprobabilitydistributionbasedonpresence-only data[33,34].Ithasbeenshowntoproducemoreaccurate modelswithlowersamplesiz esthanotherdistribution modelingsoftware[18,35].WeusedtheWorldClimCurrentBioClimclimatelayersat30arc-secondsresolution (about1km2).Theselayersarebasedondatafrom19502000andcomprise19bioclimaticvariablesrepresenting annualtrends,seasonality,andextremesofprecipitation andtemperature[36].Weusedtheentiresetof19climaticvariablesbecausewedidnotmakeany apriori assumptionsofcorrelationamongthesevariables.We clippedtheWorldClimlayersinESRIArcGIS9.3to includetheextentofthespeciesgeographicrangeinour models(N35.00,E-77.0,W-92.0,S25.00). Presencedatawasobtainedfromcollectionlocalities ofmuseumspecimensof P.polionotus identifiedtosubspecies(LouisianaStateMuseumofNaturalScience, MichiganStateUniversityMuseum,NationalMuseum ofVertebrateZoology,AmericanMuseumofNatural History,UniversityofMichiganMuseumofZoology, UniversityofKansasBiodiversityInstitute,SamNoble OklahomaMuseumofNaturalHistory),foundonthe onlinedatabasesMammalNetworkInformationSystem [37]andGlobalBiodiversit yInformationFacility[38]. RecordslackingGPScoordinates,butwithspecificwrittenlocalityinformationweregeoreferencedfollowing MaNISprotocolsusingGoogleEarthandtheU.S. BoardofGeographicNames (BGN)GeographicNames InformationSystem(GNIS)[39].Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page8of10

PAGE 9

Toobtainamodelofdistributionfortheentirespecies,amodelrepresentingt hemeandistributionwas producedinMaxentusingacr oss-validationapproach ofallspecimenlocalities.Thecross-validationfunction splitthedatasetinto n samples.Ineachofthe n replicates,asinglespecimenwastestedsequentiallyagainst allremainingsamples(i.e. n1),whichformedthe trainingsetoflocalities[ 40].Thiseliminatedtheneed topartitionadatasetintolargetrainingandtestingsets. Thisapproachisusefulwhendealingwithespecially smalldatasets,wheresplittingthedatawouldresultina trainingsetofinsufficientsize. SimilartoOsborneandSuarez-Seoane[17],wespatiallypartitionedourdataintogeographicquadrants (northeast,southeast,northwest,southwest)basedon theunweightedcentroidofourdataset.Usingthesame methodsasdescribedabove,weranmodelsforeachof thefourdatapartitions.Wenotethatweonlypartitionedourpresencedata.Therefore,inthiscase,the pseudo-absencesdrawnbyMaxentaredrawnfromour completeworkingextent(notmerelythequadrant beingexaminedinisolation).Thesefourmodelswere thencombinedtoproduceacompositemodelofprobabilityofoccurrencefortheentirespecies.Thiswas doneinESRIArcGISusingtheSpatialAnalysttoolbox tocreateanewrasterbasedonthefourindependently modeledquadrants.Whentwoormorequadrantspredictedoccurrenceatasinglepoint,weusedthehigher probabilityofoccurrencevalueinourcompositespecies distribution. Weproducedasecondcompositemodelbypartitioningourpresencedataintothe15recognizedsubspecies of P.polionotus andmodelingthedistributionsofeach subspeciesseparately.Thesubspecies P.p.trissyllepsis lackedsufficientdatatobuildafunctioningmodel,so thatsubspecieswasomitted.Theremaining14subspeciesmodelswherecombinedinArcGIS,asdescribed above,toproduceacompositemodelofprobabilityof occurrencefortheentirespecies. Thefinallogisticoutputsofeachmodelwereusedto assessourresults.Theareaunderthecurve(AUC)of receiveroperatingcharacteristic(ROC)plotwasusedto evaluatemodelperformance.TheAUCisathreshold independentmeasureofmodelperformance,wherean AUCvalueof1indicatesoptimalperformance,and AUC=0.5indicatesamodelperformingnobetterthan arandomlygeneratedone.Themeanandrangeofthe AUCvaluesofeachgroupofmodelsusedinthecompositeswerecomparedinanattempttogivearelative valueof goodness forthetwocomposites.Acknowledgements WethankR.Fletcherforcommentsonanearlydraftofthismanuscript.JASCthanksRDBarrilitoforsupport.ThisworkwassupportedbygrantstoDLR fromtheUniversityofFloridaResearchOpportunitySEEDFundandthe NationalScienceFoundation(DEB0717165andDEB0845392).Publicationof thisarticlewasfundedinpartbytheUniversityofFloridaOpen-Access PublishingFund. Authordetails1FloridaMuseumofNaturalHistory,DivisionofMammals,Universityof Florida,DickinsonHall,Gainesville,FL32611,USA.2DepartmentofBiology, UniversityofFlorida,Bartram-CarrHall,Gainesville,FL32611,USA.3Fort LauderdaleResearchandEducationCenter,UniversityofFlorida,3205 CollegeAve.,Davie,FL33314,USA. Authors contributions SCG,JASC,andDLRconceivedanddesignedtheexperiment.SCGandJASC carriedoutdataanalysis.SCG,JASC,andDLRdesignedandwrotethe manuscript.Allauthorsreadandapprovedthefinalmanucript. Received:27January2011Accepted:19September2011 Published:19September2011 References1.PetersonAT,Snchez-CorderoV,BeardCB,RamseyJM: Ecologicniche modelingandpotentialreservoirsforChagasdisease,Mexico. Emerging InfectiousDiseases 2002, 8 :662-667. 2.GuisanA,ThuillerW: Predictingspeciesdistribution:offeringmorethan simplehabitatmodels. EcologyLetters 2005, 8 :993-1009. 3.WiensJJ,GrahamCH: Nicheconservatism:Integratingevolution,ecology, andconservationbiology. AnnualReviewofEcology,Evolution,and Systematics 2005, 36 :519-539. 4.VanDerWalJ,ShooLP,JohnsonCN,WilliamsSE: Abundanceandthe environmentalniche:environmentalsuitabilityestimatedfromniche modelspredictstheupperlimitoflocalabundance. TheAmerican Naturalist 2009, 174 :282-291. 5.AllenCR,PearlstineLG,KitchensWM: Modelingviablemammal populationsingapanalyses. BiologicalConservation 2001, 99 :135-144. 6.ScobleJ,LoweAJ: Acaseforincorporatingphylogeographyand landscapegeneticsintospeciesdistributionmodellingapproachesto improveclimateadaptationandconservationplanning. Diversityand Distributions 2010, 16 :343-353. 7.PetersonAT: Predictingthegeographyofspecies invasionsvia ecologicalnichemodeling. TheQuarterlyReviewofBiology 2003, 78 :419-433. 8.PetersonAT,RobinsCR: Usingecological-nichemodelingtopredict barredowlinvasionswithimplicationsforspottedowlconservation. ConservationBiology 2003, 17 :1161-1165. 9.RdderD,WeinsheimerF,LttersS: Moleculesmeetmacroecology: Combiningspeciesdistributionmodelsandphylogeographicstudies. Zootaxa 2010, 60 :54-60. 10.RehfeldtGE,CrookstonNL,WarwellMV,EvansJS: Empiricalanalysesof plant-climaterelationshipsforthewesternunitedstates. International JournalofPlantSciences 2006, 167 :1123-1150. 11.PetersonAT: Conservatismofecologicalnichesinevolutionarytime. Science 1999, 285 :1265-1267. 12.PearsonRG,DawsonTP,LiuC: ModellingspeciesdistributionsinBritain:a hierarchicalintegrationofclimateandland-coverdata. Ecography 2004, 27 :285-298. 13.BentlageB,PetersonAT,CartwrightP: Inferringdistributionsof chirodropidbox-jellyfishes(Cnidaria:Cubozoa)ingeographicand ecologicalspaceusingecologicalnichemodeling. MarineEcologyProgress Series 2009, 384 :121-133. 14.GuisanA,BroennimannO,EnglerR,VustM,YoccozNG,LehmannA, ZimmermannNE: Usingniche-basedmodelstoimprovethesamplingof rarespecies. ConservationBiology 2006, 20 :501-511. 15. RaxworthyCJ,Martinez-MeyerE,HorningN,NussbaumRA,SchneiderGE, Ortega-HuertaMA,TownsendPetersonA: Predictingdistributionsof knownandunknownreptilespeciesinMadagascar. Nature 2003, 426 :837-841. 16.RisslerLJ,ApodacaJJ: Addingmoreecologyintospeciesdelimitation: ecologicalnichemodelsandphylogeographyhelpdefinecryptic speciesintheblacksalamander(Aneidesflavipunctatus). Systematic Biology 2007, 56 :924-942.Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page9of10

PAGE 10

17.OsborneP,Suarez-SeoaneS: Shoulddatabepartitionedspatiallybefore buildinglarge-scaledistributionmodels? EcologicalModelling 2002, 157 :249-259. 18.HernandezPA,GrahamCH,MasterLL,AlbertDL: Theeffectofsamplesize andspeciescharacteristicsonperformanceofdifferentspecies distributionmodelingmethods. Ecography 2006, 29 :773-785. 19.BlairF: Ecologicalfactorsinspeciationofperomyscus. Evolution 1950, 4 :253-275. 20.HallER: TheMammalsofNorthAmerica. 2edition.JohnWileyandSons; 1981,980. 21.DegnerJF,StoutIJ,RothJD,ParkinsonCL: Populationgeneticsand conservationofthethreatenedsoutheasternbeachmouse(Peromyscus polionotusniveiventris):subspeciesandevolutionaryunits. Conservation Genetics 2007, 8 :1441-1452. 22.VignieriSN,LarsonJG,HoekstraHE: Theseleviveadvantageofcripsisin mice. Evolution 2010, 64 :2153-2158. 23.MullenLM,HoekstraHE: Naturalselectionalonganenvironmental gradient:aclassicclineinmousepigmentation. Evolution 2008, 62 :1555-1570. 24.SteinerCC,RmplerH,BoettgerLM,SchnebergT,HoekstraHE: The geneticbasisofphenotypicconvergenceinbeachmice:similarpigment patternsbutdifferentgenes. MolecularBiologyandEvolution 2009, 26 :35-45. 25.MullenLM,VignieriSN,GoreJJA,HoekstraHE: Adaptivebasisof geographicvariation:genetic,phenotypicandenvironmentaldifferences amongbeachmousepopulations. ProceedingsoftheRoyalSocietyB 2009, 276 :3809-3818. 26.HumphreySR,BarbourDB: Statusandhabitatofthreesubspeciesof peromyscuspolionotusinflorida. JounalofMammalogy 1981, 62 :840-844. 27.FieldingAH,BellJF: Areviewofmethodsfortheassessmentof predictionerrorsinconservationpresence/absencemodels. EnvironmentalConservation 1997, 24 :38-49. 28.GuisanA,ZimmermannNE: Predictivehabitatdistributionmodelsin ecology. EcologicalModelling 2000, 135 :147-186. 29.LoboJM,Jimnez-ValverdeA,RealR: AUC:Amisleadingmeasureofthe performanceofpredictivedistributionmodels. GlobalEcologyand Biogeography 2008, 17 :145-151. 30.ElithJ,LeathwickJR: Speciesdistributionmodels:Ecologicalexplanation andpredictionacrossspaceandtime. AnnualReviewofEcology,Evolution, andSystematics 2009, 40 :677-697. 31.VelozSD: Spatiallyautocorrelatedsamplingfalselyinflatesmeasuresof accuracyforpresence-onlynichemodels. JournalofBiogeography 2009, 36 :2290-2299. 32.SeguradoP,AraujoMB,KuninWE: Consequencesofspatial autocorrelationforniche-basedmodels. JournalofAppliedEcology 2006, 43 :433-444. 33.PhillipsS,AndersonR,SchapireR: Maximumentropymodelingofspecies geographicdistributions. EcologicalModelling 2006, 190 :231-259. 34.PhillipsSJ,DudkM: ModelingofspeciesdistributionswithMaxent:New extensionsandacomprehensiveevaluation. Ecography 2008, 31 :161-175. 35.ElithJ,GrahamCH,AndersonRP,DudkM,FerrierS,GuisanA,HijmansRJ, HuettmannF,LeathwickJR,LehmannA,LiJ,LohmannLG,LoiselleBA, ManionG,MoritzC,NakamuraM,NakazawaY,OvertonJMcC,Townsend PetersonA,PhillipsSJ,RichardsonK,Scachetti-PereiraR,SchapireRE, SobernJ,WilliamsS,WiszMS,ZimmermannNE: Novelmethodsimprove predictionofspecies distributionsfromoccurrencedata. Ecography 2006, 29 :129-151. 36.HijmansRJ,CameronSE,ParraJL,JonesPG,JarvisA: Veryhighresolution interpolatedclimatesurfacesforgloballandareas. InternationalJournal ofClimatology 2005, 25 :1965-1978. 37. MammalNetworkInformationSystem. [http://manisnet.org]. 38. GlobalBiodiversityInformationSystem. [http://www.gbif.org]. 39. USBoardonGeographicNames:GeographicNamesInformation System. [http://geonames.usgs.gov/pls/gnispublic]. 40.ArajoMB,PearsonRG,ThuillerW,ErhardM: Validationofspecies-climate impactmodelsunderclimatechange. GlobalChangeBiology 2005, 11 :1504-1513.doi:10.1186/1472-6785-11-20 Citethisarticleas: Gonzalez etal .: Populationdistributionmodels: speciesdistributionsarebettermodeledusingbiologicallyrelevant datapartitions. BMCEcology 2011 11 :20. Submit your next manuscript to BioMed Central and take full advantage of: Convenient online submission Thorough peer review No space constraints or color gure charges Immediate publication on acceptance Inclusion in PubMed, CAS, Scopus and Google Scholar Research which is freely available for redistribution Submit your manuscript at www.biomedcentral.com/submit Gonzalez etal BMCEcology 2011, 11 :20 http://www.biomedcentral.com/1472-6785/11/20 Page10of10


xml version 1.0 encoding utf-8 standalone no
mets ID sort-mets_mets OBJID sword-mets LABEL DSpace SWORD Item PROFILE METS SIP Profile xmlns http:www.loc.govMETS
xmlns:xlink http:www.w3.org1999xlink xmlns:xsi http:www.w3.org2001XMLSchema-instance
xsi:schemaLocation http:www.loc.govstandardsmetsmets.xsd
metsHdr CREATEDATE 2012-07-27T16:06:41
agent ROLE CUSTODIAN TYPE ORGANIZATION
name BioMed Central
dmdSec sword-mets-dmd-1 GROUPID sword-mets-dmd-1_group-1
mdWrap SWAP Metadata MDTYPE OTHER OTHERMDTYPE EPDCX MIMETYPE textxml
xmlData
epdcx:descriptionSet xmlns:epdcx http:purl.orgeprintepdcx2006-11-16 xmlns:MIOJAVI
http:purl.orgeprintepdcxxsd2006-11-16epdcx.xsd
epdcx:description epdcx:resourceId sword-mets-epdcx-1
epdcx:statement epdcx:propertyURI http:purl.orgdcelements1.1type epdcx:valueURI http:purl.orgeprintentityTypeScholarlyWork
http:purl.orgdcelements1.1title
epdcx:valueString Population distribution models: Species distributions are better modeled using biologically relevant data partitions
http:purl.orgdctermsabstract
Abstract
Background
Predicting the geographic distribution of widespread species through modeling is problematic for several reasons including high rates of omission errors. One potential source of error for modeling widespread species is that subspecies and/or races of species are frequently pooled for analyses, which may mask biologically relevant spatial variation within the distribution of a single widespread species. We contrast a presence-only maximum entropy model for the widely distributed oldfield mouse (Peromyscus polionotus) that includes all available presence locations for this species, with two composite maximum entropy models. The composite models either subdivided the total species distribution into four geographic quadrants or by fifteen subspecies to capture spatially relevant variation in P. polionotus distributions.
Results
Despite high Area Under the ROC Curve (AUC) values for all models, the composite species distribution model of P. polionotus generated from individual subspecies models represented the known distribution of the species much better than did the models produced by partitioning data into geographic quadrants or modeling the whole species as a single unit.
Conclusions
Because the AUC values failed to describe the differences in the predictability of the three modeling strategies, we suggest using omission curves in addition to AUC values to assess model performance. Dividing the data of a widespread species into biologically relevant partitions greatly increased the performance of our distribution model; therefore, this approach may prove to be quite practical and informative for a wide range of modeling applications.
http:purl.orgdcelements1.1creator
Gonzalez, Sergio C
Soto-Centeno, J Angel
Reed, David L
http:purl.orgeprinttermsisExpressedAs epdcx:valueRef sword-mets-expr-1
http:purl.orgeprintentityTypeExpression
http:purl.orgdcelements1.1language epdcx:vesURI http:purl.orgdctermsRFC3066
en
http:purl.orgeprinttermsType
http:purl.orgeprinttypeJournalArticle
http:purl.orgdctermsavailable
epdcx:sesURI http:purl.orgdctermsW3CDTF 2011-09-19
http:purl.orgdcelements1.1publisher
BioMed Central Ltd
http:purl.orgeprinttermsstatus http:purl.orgeprinttermsStatus
http:purl.orgeprintstatusPeerReviewed
http:purl.orgeprinttermscopyrightHolder
Sergio C Gonzalez et al.; licensee BioMed Central Ltd.
http:purl.orgdctermslicense
http://creativecommons.org/licenses/by/2.0
http:purl.orgdctermsaccessRights http:purl.orgeprinttermsAccessRights
http:purl.orgeprintaccessRightsOpenAccess
http:purl.orgeprinttermsbibliographicCitation
BMC Ecology. 2011 Sep 19;11(1):20
http:purl.orgdcelements1.1identifier
http:purl.orgdctermsURI http://dx.doi.org/10.1186/1472-6785-11-20
fileSec
fileGrp sword-mets-fgrp-1 USE CONTENT
file sword-mets-fgid-0 sword-mets-file-1
FLocat LOCTYPE URL xlink:href 1472-6785-11-20.xml
sword-mets-fgid-1 sword-mets-file-2 applicationpdf
1472-6785-11-20.pdf
structMap sword-mets-struct-1 structure LOGICAL
div sword-mets-div-1 DMDID Object
sword-mets-div-2 File
fptr FILEID
sword-mets-div-3


!DOCTYPE art SYSTEM 'http:www.biomedcentral.comxmlarticle.dtd'
ui 1472-6785-11-20ji 1472-6785fm
dochead Research article
bibl
title
p Population distribution models: species distributions are better modeled using biologically relevant data partitions
aug
au id A1 ce yes snm Gonzalezmi Cfnm Sergioinsr iid I1 I3 email sergeman@ufl.edu
ca A2 Soto-CentenoJ AngelI2 sotocenteno@ufl.edu
A3 ReedLDaviddlreed@ufl.edu
insg
ins Florida Museum of Natural History, Division of Mammals, University of Florida, Dickinson Hall, Gainesville, FL 32611, USA
Department of Biology, University of Florida, Bartram-Carr Hall, Gainesville, FL 32611, USA
Fort Lauderdale Research and Education Center, University of Florida, 3205 College Ave., Davie, FL 33314, USA
source BMC Ecology
issn 1472-6785
pubdate 2011
volume 11
issue 1
fpage 20
url http://www.biomedcentral.com/1472-6785/11/20
xrefbib pubidlist pubid idtype doi 10.1186/1472-6785-11-20pmpid 21929792
history rec date day 27month 1year 2011acc 1992011pub 1992011
cpyrt 2011collab Gonzalez et al; licensee BioMed Central Ltd.note This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
abs
sec
st
Abstract
Background
Predicting the geographic distribution of widespread species through modeling is problematic for several reasons including high rates of omission errors. One potential source of error for modeling widespread species is that subspecies and/or races of species are frequently pooled for analyses, which may mask biologically relevant spatial variation within the distribution of a single widespread species. We contrast a presence-only maximum entropy model for the widely distributed oldfield mouse (it Peromyscus polionotus) that includes all available presence locations for this species, with two composite maximum entropy models. The composite models either subdivided the total species distribution into four geographic quadrants or by fifteen subspecies to capture spatially relevant variation in P. polionotus distributions.
Results
Despite high Area Under the ROC Curve (AUC) values for all models, the composite species distribution model of P. polionotus generated from individual subspecies models represented the known distribution of the species much better than did the models produced by partitioning data into geographic quadrants or modeling the whole species as a single unit.
Conclusions
Because the AUC values failed to describe the differences in the predictability of the three modeling strategies, we suggest using omission curves in addition to AUC values to assess model performance. Dividing the data of a widespread species into biologically relevant partitions greatly increased the performance of our distribution model; therefore, this approach may prove to be quite practical and informative for a wide range of modeling applications.
bdy
Background
Species distribution modeling (SDM) has become a common tool for understanding spatial distribution patterns of biodiversity worldwide abbrgrp
abbr bid B1 1
B2 2
B3 3
B4 4
. The goal of SDM is to build a model predicting the relative probability of occurrence of a species across geographic space commonly using environmental data (i.e. climate, vegetation, soil, etc.) and a dataset of known presence or presence/absence localities. The terms ecological niche model, environmental niche model, and species distribution model have all been used to describe this type of modeling in the literature; for the sake of simplicity we will use species distribution modeling. SDM techniques continue to evolve with an increasingly broad range of applications from conservation planning
B5 5
B6 6
, to predicting species colonization and abundance
4
B7 7
B8 8
, predicting disease outbreaks
1
, and understanding phylogeographic patterns
B9 9
. Methods of producing SDMs vary with the type of data available, purpose, and software used.
There is a direct link between climate and the distribution of plant species
B10 10
. Because climate is a causal factor in the distribution of plant species (and plant species assemblages), climatic patterns at various spatial scales directly affect habitat types and community productivity. Thus, climate is considered a proxy for a given species' environmental niche. Because of the variables involved in building SDMs, it is important to keep in mind that SDMs are predicting a species' fundamental niche as probability of occurrence
B11 11
B12 12
, not the realized distribution, which is affected by many extrinsic factors that may not be accounted for in the model. The fundamental niches of species are considered to be conserved over evolutionary time
11
, which has allowed climate-based SDMs to be successful in predicting the occurrence of species or closely related species at previously unsampled localities
B13 13
B14 14
B15 15
B16 16
.
Modeling species whose distributions span large environmental or habitat variation may be problematic because distribution models tend to have higher rates of omission error (i.e. underprediction) in the predicted species distributions
2
B17 17
B18 18
. In such cases, the models may indicate regional specialization of periphery or isolated populations. Commission errors (i.e. the overprediction of distributions) may result from a restriction of the realized distribution due to biotic interactions or geographic barriers to range expansion. In an attempt to overcome omission errors in SDMs for widespread species, Osborne & Suarez-Seoane
17
modeled species distributions by spatially partitioning their data into geographic quadrants and into concentric rings to model each data partition separately. Hernandez et al.
18
suggested that future research should focus on modeling broad distributions in subunits that are based on distinct genetic lineages or recognized subspecies.
Most species in the genus Peromyscus are widespread, with a positive correlation between species range and number of recognized subspecies
B19 19
, suggesting that local specialization is common within species of this genus. With 15 recognized subspecies
B20 20
and genetically structured populations, P. polionotus is an excellent model species for developing new methods of data partitioning to overcome the problems associated with modeling the geographic distributions of widely distributed species.
The oldfield mouse (Peromyscus polionotus), also known regionally as the Florida beach mouse, is widespread throughout the southeastern United States (Figure figr fid F1 1). Morphological and genetic differences have been documented between subspecies
B21 21
B22 22
B23 23
B24 24
B25 25
. Molecular evidence suggests that little or no gene flow occurs between the highly structured populations on islands along Florida's panhandle
25
or between P. p. rhoadsi, on Florida's central ridge, and P. p. niveiventris, on the Atlantic coast
21
. Highly differentiated populations of Peromyscus are suggested to be the result of strong local adaptation
19
23
B26 26
. Coastal dune populations along the Gulf of Mexico are phenotypically more similar to populations along the Atlantic Coast (especially in coat color) than to neighboring populations, suggesting they are under similar selective pressures in their disjunct coastal environments
23
. Throughout its range, P. polionotus is threatened by development and invasive species exerting competition and predation pressures on its populations, and the subspecies P. p. decoloratus has been considered extinct since 1950
19
26
.
fig Figure 1caption P. polionotus in southeastern United Statestext
b P. polionotus in southeastern United States. Distribution map of the 15 supspecies of P. polionotus redrawn from Hall (1981).
graphic file 1472-6785-11-20-1 hint_layout single
Given the evidence for local adaptation in this widespread species, previous research would predict that an SDM of the entire species would underpredict the geographic distribution of P. polionotus. A more accurate prediction of the species' distribution may result from a biologically informed spatial partitioning of locality data. To test this hypothesis, we modeled the distribution of P. polionotus in three ways; we modeled the whole species distribution at once, we partitioned locality data into four geographic quadrants following Osborne & Suarez-Seoane
17
, and we partitioned locality data by the 15 recognized subspecies of P. polionotus.
Results
All individual and composite models produced AUC values above 0.84, which are consistent with AUC values reported in the literature for other taxa
9
17
18
B27 27
. Despite having a high AUC value (0.899), the model based on the entire data set failed to predict the occurrence of P. polionnotus in places where it is clearly known to occur (Figure F2 2a), most obviously omitting the distributions of the subspecies P. p. colemani and P. p. polionotus. Partitioning the data by geographic quadrants (Figure 2b) and by subspecies (Figure 2c) produced models that are progressively better, both in terms of predicting the known distribution, and in terms of their AUC scores. The average AUC value of the individual models used to build the geographic quadrant composite model was 0.927, whereas the average AUC value of the individual models that were used to build the subspecies composite model was 0.976.
Figure 2Maxent distributions for P. polionotus using different data partitions
Maxent distributions for P. polionotus using different data partitions. Predicted species distributions using presence data for the entire species at once (A), by dividing the presence data into quadrants (B), and by subspecies (C).
1472-6785-11-20-2 double
Partitioning the data into geographic quadrants produced four models with AUC scores of 0.844, 0.968, 0.905, and 0.993 (clockwise from northeast; Figure F3 3). The quadrant composite model (Figure 2b) predicted high probabilities of occurrence in areas that the full species model (Figure 2a) had omitted. However, the quadrant composite model (Figure 2b) showed poor resolution in parts of northern Georgia and peninsular Florida. When the dataset was partitioned according to the currently recognized subspecies (Figures F4 4 and F5 5), each subspecies model performed well based on AUC scores. Twelve out of 15 models had AUC scores between 0.97 and 1.0; the exceptions being P. p. colemani (0.917), P. p. polionotus (0.851), and P. p. trissyllepsis (0.5). The poor performance of the model for P. p. trissyllepsis was due to insufficient data (n = 2) for the population and was omitted from the composite model.
Figure 3Maxent distribution based on the quadrant method
Maxent distribution based on the quadrant method. Predicted species distribution of P. polionotus estimated from presence data modeled separately in four quadrants (A-D).
1472-6785-11-20-3
Figure 4Maxent distribution of P. polionotus on peninsular Florida
Maxent distribution of P. polionotus on peninsular Florida. Predicted species distribution of P. polionotus estimated from presence data of each mainland and peninsular subspecies (A-G) analyzed separately.
1472-6785-11-20-4
Figure 5Maxent distribution of P. polionotus on Florida panhandle
Maxent distribution of P. polionotus on Florida panhandle. Predicted species distribution of P. polionotus estimated from presence data of each Florida panhandle subspecies (A-G) analyzed separately.
1472-6785-11-20-5
Although superficially all models developed from data partitions seem to perform well based on AUC values, rates of omission between these methods show a different perspective. Figure F6 6 shows three omission curves for the whole-species model (Figure 6a), the quadrant method (Figure 6b) and the subspecies model (Figure 6c). The curves show omission error (Y-axis) as a function of predicted probability of occurrence (X-axis). Better performing models based on the logistic output of Maxent have fewer omission errors even as predicted probability of occurrence reaches maximal values. The whole-species model (Figure 6a) and the quadrant model (Figure 6b) have omission error rates that increase linearly with increasing probability of occurrence. In contrast, the subspecies model (Figure 6c) has relatively low omission error rates that only begins to increase when predicted probability of occurrence reaches higher values, which is preferable.
Figure 6Omission curves as a means of assessing model performance
Omission curves as a means of assessing model performance. Omission curve (thin green line) for the species level model of P. polionotus (A), the northwest region of the quadrant method (B), and the single subspecies P. p. phasma of the subspecies method (C). Mean AUC value for replicate runs was 0.899 (std.dev. = 0.105), 0.905 (std.dev. = 0.119), and 1.0 (std.dev. = 0.001), respectively.
1472-6785-11-20-6
Discussion
We suggest that partitioning data in a biologically meaningful way (as opposed to geographically) can help to overcome high omission rates in distribution models of widespread species
17
18
B28 28
. Despite having a relatively high AUC value (0.899), the model built upon the whole species' distribution failed to predict known localities of P. polionotus, which can be observed by comparing Figure 1 to Figure 2a. Spatially partitioning the data into quadrants produced a much better distribution model after combining four regional models with AUC values ranging from 0.844 to 0.993, which can be seen by comparing Figure 1 to Figure 2b. The 14 AUC scores for the SDMs based on subspecies partitioning ranged from 0.851 to 1.0, which is not substantially different from AUC values obtained for the quadrant models. However, the accuracy and increased resolution of the composite of the subspecies models can be seen in comparing the three panels of Figure 2 and by comparing Figure 1 to Figure 2c.
Limitation of AUC values for assessing predictive performance
AUC values are commonly used as indicators of model fit
9
17
18
27
, and high values for all three methods in our study would suggest that each method produced highly accurate models. Furthermore, the modest increase in AUC values with ever greater data partitioning would suggest that each successive partitioning scheme produced, at best, only slightly better fitting models. However, this finding is misleading when one compares the predicted distributions to the known distribution for the species. The trend of increasing AUC scores may indicate the direction of change in accuracy, but it fails to capture the magnitude of improvement in the predicted distributions of the quadrant composite method and the subspecies composite method. This failure is, in part, due to the fact that the AUC scores for the geographic quadrant method and subspecies method are composites of 4 and 14 combined models (respectively), and the accuracy (or inaccuracy) of individual models is compounded when they are combined. This fact alone, however, cannot fully explain the observed discrepancy between the vastly improved model prediction and the modestly better AUC scores.
AUC values can be misleading when assessing a model's predictive ability for several reasons. The AUC measures discrimination and not accuracy per se, thus ignoring the goodness of fit of a model
B29 29
. The AUC value also takes into account the performance of the model at the extreme left (as well as the right) of the ROC curve (see
29
for a details), a region that is not operationally meaningful in our case. We are only interested in thresholds of predicted probability of occurrence greater than 0.50 because that would equal the probability of occurrence of a null model. This inclusion of the area under the extreme left of the ROC curve can inflate AUC values, which can be further inflated when the total geographic extent of the model is considered. If the ratio between areas of presence and the total extent is high, true positives are more likely to occur by chance alone
28
. Because this ratio changes with each of the individual models built on different data partitions, AUC values may not be useful in accurately comparing relative model performance between or among our subspecies and regional models.
It is also possible that the inflation of AUC values observed in the models we present results from the interaction between geographic and environmental space. Because of the narrow geographic space at which most subspecies of P. polionotus occur, locality information is geographically clustrered; therefore, the environmental space sampled by the models may show spatial autocorrelation in some of the environmental variables used. In SDM, spatial autocorrelation occurs when the values of the variables sampled at nearby locations are not independent from each other
B30 30
and as a result, measures of accuracy (e.g. AUC) can be inflated
B31 31
B32 32
. In our case, the geographic clustering of narrowly distributed subspecies of P. polionotus may cause spatial autocorrelation and thus inflate AUC values (see Figure 4 and 5). Nevertheless, within these narrow extents, we included samples spanning the entire geographic space representing a significant portion of the environmental space occupied by each subspecies. The resulting models are accurate to the true distribution of the subspecies and are able to detect even subtle local environmental conditions likely affecting each subspecies differently, despite the seemingly geographic clustering. This further emphasizes the point that AUC values provide an unreliable way to accurately compare relative model performance. The only exceptions to the issue of inflated AUC values in our dataset are P. p. colemani (AUC = 0.917) and P. p. polionotus (AUC = 0.851), which are the only two widely distributed subspecies spanning a more heterogeneous environmental space where locality information for the subspecies is not geographically clustered (Figure 2). Because of the larger geographic space occupied by these subspecies, the resulting models from P. p. colemani and P. p. polionotus are unlikely affected by spatial autocorrelation and therefore do not show inflated AUC values.
Finally, the AUC does not provide information as to the spatial distribution of errors. It also weighs omission and commission errors equally, both of which vary in interpretive meaning and importance with the intended use of the model
29
. Because we do not have true absence data, we cannot quantify our commission error rate. However, the omission curve shows how well the model performs at different thresholds (i.e. the distribution of omission errors). Therefore, the omission curve can be as important as the AUC value in terms of assessing model performance, if not more-so. A model with relatively lower omission errors at higher predicted probabilities of occurrence is preferred.
Quadrant versus subspecies partitioning
Because Maxent draws pseudo-absence data at random to calculate AUC scores, it is possible that in our quadrant analysis it drew false pseudo-absences from areas outside the quadrant being tested, especially in the case of the northwest quadrant. Whether this occurred, and if it did, whether it contributed to the observed underprediction is debatable. First, there was far less underprediction in the other three quadrants. For example, the model for the southwest quadrant, which includes peninsular Florida predicted occurrences in regions where the species does not occur. Second, the two northern quadrants cover roughly the same geographic area as two of the subspecies (P. p. colemani and P. p. polionotus), yet the two northern quadrants failed to predict areas of known occurrence that the two subspecies models predicted accurately. It is possible that poor quadrant-based models resulted because pseudo-absences were being generated by Maxent in areas with true presences for either of the two subspecies just outside of the quadrant being modeled. That is, the arbitrary boundary between the quadrants obfuscates biologically meaningful boundaries between populations or subspecies. Thus, in using quadrant-based partitioning, niche information was lost for the two subspecies, which emphasizes our point that biologically relevant data partitioning informs species distribution models.
Molecular data and population distribution models
It is easy to see how molecular data, such as DNA sequences, can be used to delineate biologically meaningful groups (i.e. clades) within a species, and that those clades might be partitioned separately for species distribution modeling, especially if they are geographically discrete. But, just as molecular data can improve methods of generating SDMs, the findings associated with SDMs can also inform the work done by molecular biologists studying population genetics, phylogenetics, or phylogeography. When SDMs are nonoverlapping for populations within a species, they may be revealing cryptic patterns of divergence that would be interesting to study with molecular data. Conversely, when molecular data uncover population structure or limits to gene flow, SDMs can be used to test hypothesized mechanisms of divergence such as niche differentiation. Examining both molecular data and SDMs together has been explored only recently
6
9
.
Building SDMs for P. polionotus by partitioning data into subspecies and building a composite distribution model mitigated the problem of high omission rates that usually occurs when modeling the distributions of widely distributed species. This suggests that the SDM based on biologically relevant partitions (subspecies in our case) could accommodate variability in the niches of subspecies, whereas modeling the whole species distribution together could not. This is supported by the fact that spatial partitioning of data into quadrants produced models that had regions of both under- and over-prediction (Figure 3), whereas the models based on partitioning by subspecies showed no signs of underprediction and only modest overlap in the distributions of adjacent individual subspecies caused by overprediction (Figure 5). The evidence of high levels of population structure between locally adapted populations
19
21
22
23
25
26
might be driving the improvement we see in the composite model based on subspecies distributions. In our case, we show that molecular data at the population level improved model accuracy. Furthermore, in the absence of detailed molecular information on the populations studied, researchers could generate relevant data partitions using alternative data sources such as subspecies delimitations, morophological differences or other phenotypic traits.
Phylogeographic implications for P. polionotus
Guisan & Zimmermann
28
encourage collaboration with evolutionary biologists and population geneticists in cases where widespread species are being modeled. More recently, Rödder et al.
9
discussed how a variety of techniques including molecular ecology and environmental niche modeling can be complimentary in answering phylogeographic questions. The case is such here, where our method of partitioning data was based largely on the literature, which includes population genetic studies that have been conducted on P. polionotus. Conversely, as molecular work helped to inform our models, our models also shed light on and confirm results of studies on the species' genetic structure, and possibly evolutionary trajectory. For example, the SDMs for P. p. rhoadsi and P. p. nivieventris do not overlap and are geographically discrete (Figure 4 and 2c), which is consistent with the genetic results of Degner et al.
21
and current taxonomy (Figure 1). The southeastern quadrant model, however, has very poor resolution of this finescale distinction (Figure 3c). Similarly, the models for P. p. polionotus and P. p. colemani capture the known extent of their respective ranges (Figure 5), while the two northern quadrant models do not (Figure 3, a and 3b), implying that these two subspecies occupy different climatic niches.
Climate appears to play an important role in defining inland and inland vs. coastal subspecies (e.g. Figure 4A-D). However, despite there being genetic differences in the beach mouse subspecies located in the Florida panhandle (Figure 5A-F), there was considerable overlap in their predicted distributions, suggesting that climate may not be the primary factor defining the range of these subspecies. Predicted niche overlap usually occurred between adjacent coastal subspecies (Figure 5A-F) and only once between coastal and inland subspecies (see P. p. albifrons and P. p. sumneri, Figure 5F-G). Studies have shown that the coastal beach mouse populations reflect patterns of local adaptation and strong selection favoring cryptic coloration
22
23
24
25
. Therefore, in cases where climatic habitat on adjacent coastal beaches might be similar, vicariance (i.e. coastal inlets) and strong selection for coat coloration are more likely than climate to maintain the distinctiveness of coastal beach mouse subspecies.
Conclusions
Using a biologically meaningful method of partitioning the data from widely distributed species generated a composite SDM of P. polionotus that more accurately reflected the known distribution of the species than the process of analyzing the whole species at once or partitioning the data into geographic quadrants. Osborne and Suarez-Seoane
17
note that geographic based data partitioning (e.g. quadrants) may not have worked well due to the absence of any biological basis for partitioning. We contend that our study confirms that statement. We also provide an example of how SDMs can be both informed by as well as inform phylogeographic studies at the population and species levels. Modeling a widespread species using biologically meaningful data partitions has the potential to greatly increase the performance of distribution models while only requiring basic manipulation in GIS software. Thus, this technique may prove to be quite practical for a wide range of modeling applications. Despite the increasing use and popularity of ENMs, a completely objective, accurate, and fully accepted measure of performance of predictive distribution models is still elusive
9
27
29
. We suggest using both the AUC and omission curve on a contextual basis to assess model performance.
Methods
We created species distribution models for P. polionotus in Maxent using the WorldClim climate layers. Maxent uses the principle of maximum entropy density estimation to generate a probability distribution based on presence-only data
B33 33
B34 34
. It has been shown to produce more accurate models with lower sample sizes than other distribution modeling software
18
B35 35
. We used the WorldClim Current BioClim climate layers at 30 arc-seconds resolution (about 1 kmsup 2). These layers are based on data from 1950-2000 and comprise 19 bioclimatic variables representing annual trends, seasonality, and extremes of precipitation and temperature
B36 36
. We used the entire set of 19 climatic variables because we did not make any a priori assumptions of correlation among these variables. We clipped the WorldClim layers in ESRI ArcGIS 9.3 to include the extent of the species geographic range in our models (N35.00, E-77.0, W-92.0, S25.00).
Presence data was obtained from collection localities of museum specimens of P. polionotus identified to subspecies (Louisiana State Museum of Natural Science, Michigan State University Museum, National Museum of Vertebrate Zoology, American Museum of Natural History, University of Michigan Museum of Zoology, University of Kansas Biodiversity Institute, Sam Noble Oklahoma Museum of Natural History), found on the online data bases Mammal Network Information System
B37 37
and Global Biodiversity Information Facility
B38 38
. Records lacking GPS coordinates, but with specific written locality information were georeferenced following MaNIS protocols using Google Earth and the U.S. Board of Geographic Names' (BGN) Geographic Names Information System (GNIS)
B39 39
.
To obtain a model of distribution for the entire species, a model representing the mean distribution was produced in Maxent using a cross-validation approach of all specimen localities. The cross-validation function split the data set into n samples. In each of the n replicates, a single specimen was tested sequentially against all remaining samples (i.e. n 1), which formed the training set of localities
B40 40
. This eliminated the need to partition a dataset into large training and testing sets. This approach is useful when dealing with especially small datasets, where splitting the data would result in a training set of insufficient size.
Similar to Osborne and Suarez-Seoane
17
, we spatially partitioned our data into geographic quadrants (northeast, southeast, northwest, southwest) based on the unweighted centroid of our dataset. Using the same methods as described above, we ran models for each of the four data partitions. We note that we only partitioned our presence data. Therefore, in this case, the pseudo-absences drawn by Maxent are drawn from our complete working extent (not merely the quadrant being examined in isolation). These four models were then combined to produce a composite model of probability of occurrence for the entire species. This was done in ESRI ArcGIS using the Spatial Analyst toolbox to create a new raster based on the four independently modeled quadrants. When two or more quadrants predicted occurrence at a single point, we used the higher probability of occurrence value in our composite species distribution.
We produced a second composite model by partitioning our presence data into the 15 recognized subspecies of P. polionotus and modeling the distributions of each subspecies separately. The subspecies P. p. trissyllepsis lacked sufficient data to build a functioning model, so that subspecies was omitted. The remaining 14 subspecies models where combined in ArcGIS, as described above, to produce a composite model of probability of occurrence for the entire species.
The final logistic outputs of each model were used to assess our results. The area under the curve (AUC) of receiver operating characteristic (ROC) plot was used to evaluate model performance. The AUC is a threshold independent measure of model performance, where an AUC value of 1 indicates optimal performance, and AUC = 0.5 indicates a model performing no better than a randomly generated one. The mean and range of the AUC values of each group of models used in the composites were compared in an attempt to give a relative value of "goodness" for the two composites.
Authors' contributions
SCG, JASC, and DLR conceived and designed the experiment. SCG and JASC carried out data analysis. SCG, JASC, and DLR designed and wrote the manuscript. All authors read and approved the final manucript.
bm
ack
Acknowledgements
We thank R. Fletcher for comments on an early draft of this manuscript. JAS-C thanks RD Barrilito for support. This work was supported by grants to DLR from the University of Florida Research Opportunity SEED Fund and the National Science Foundation (DEB 0717165 and DEB 0845392). Publication of this article was funded in part by the University of Florida Open-Access Publishing Fund.
refgrp Ecologic niche modeling and potential reservoirs for Chagas disease, MexicoPetersonATSánchez-CorderoVBeardCBRamseyJMEmerging Infectious Diseases20028662lpage 667pmcid 2730326link fulltext 12095431Predicting species distribution: offering more than simple habitat modelsGuisanAThuillerWEcology Letters20058993100910.1111/j.1461-0248.2005.00792.xNiche conservatism: Integrating evolution, ecology, and conservation biologyWiensJJGrahamCHAnnual Review of Ecology, Evolution, and Systematics20053651953910.1146/annurev.ecolsys.36.102803.095431Abundance and the environmental niche: environmental suitability estimated from niche models predicts the upper limit of local abundanceVanDerWalJShooLPJohnsonCNWilliamsSEThe American Naturalist200917428229110.1086/60008719519279Modeling viable mammal populations in gap analysesAllenCRPearlstineLGKitchensWMBiological Conservation20019913514410.1016/S0006-3207(00)00084-7A case for incorporating phylogeography and landscape genetics into species distribution modelling approaches to improve climate adaptation and conservation planningScobleJLoweAJDiversity and Distributions20101634335310.1111/j.1472-4642.2010.00658.xPredicting the geography of species' invasions via ecological niche modelingPetersonATThe Quarterly Review of Biology20037841943310.1086/37892614737826Using ecological-niche modeling to predict barred owl invasions with implications for spotted owl conservationPetersonATRobinsCRConservation Biology2003171161116510.1046/j.1523-1739.2003.02206.xMolecules meet macroecology: Combining species distribution models and phylogeographic studiesRödderDWeinsheimerFLöttersSZootaxa2010605460Empirical analyses of plant-climate relationships for the western united statesRehfeldtGECrookstonNLWarwellMVEvansJSInternational Journal of Plant Sciences20061671123115010.1086/507711Conservatism of ecological niches in evolutionary timePetersonATScience19992851265126710.1126/science.285.5431.126510455053Modelling species distributions in Britain: a hierarchical integration of climate and land-cover dataPearsonRGDawsonTPLiuCEcography20042728529810.1111/j.0906-7590.2004.03740.xInferring distributions of chirodropid box-jellyfishes (Cnidaria: Cubozoa) in geographic and ecological space using ecological niche modelingBentlageBPetersonATCartwrightPMarine Ecology Progress Series2009384121133Using niche-based models to improve the sampling of rare speciesGuisanABroennimannOEnglerRVustMYoccozNGLehmannAZimmermannNEConservation Biology20062050151110.1111/j.1523-1739.2006.00354.x16903111Predicting distributions of known and unknown reptile species in MadagascarRaxworthyCJMartinez-MeyerEHorningNNussbaumRASchneiderGEOrtega-HuertaMATownsend PetersonANature200342683784110.1038/nature0220514685238Adding more ecology into species delimitation: ecological niche models and phylogeography help define cryptic species in the black salamander (Aneides flavipunctatus)RisslerLJApodacaJJSystematic Biology20075692494210.1080/1063515070170306318066928Should data be partitioned spatially before building large-scale distribution models?OsbornePSuarez-SeoaneSEcological Modelling200215724925910.1016/S0304-3800(02)00198-9The effect of sample size and species characteristics on performance of different species distribution modeling methodsHernandezPAGrahamCHMasterLLAlbertDLEcography20062977378510.1111/j.0906-7590.2006.04700.xEcological factors in speciation of peromyscusBlairFEvolution1950425327510.2307/2405335HallERThe Mammals of North Americapublisher John Wiley and Sonsedition 21981980Population genetics and conservation of the threatened southeastern beach mouse (Peromyscus polionotus niveiventris): subspecies and evolutionary unitsDegnerJFStoutIJRothJDParkinsonCLConservation Genetics200781441145210.1007/s10592-007-9295-1The selevive advantage of cripsis in miceVignieriSNLarsonJGHoekstraHEEvolution2010642153215820163447Natural selection along an environmental gradient: a classic cline in mouse pigmentationMullenLMHoekstraHEEvolution2008621555157010.1111/j.1558-5646.2008.00425.x18489719The genetic basis of phenotypic convergence in beach mice: similar pigment patterns but different genesSteinerCCRömplerHBoettgerLMSchönebergTHoekstraHEMolecular Biology and Evolution200926354518832078Adaptive basis of geographic variation: genetic, phenotypic and environmental differences among beach mouse populationsMullenLMVignieriSNGoreJJAHoekstraHEProceedings of the Royal Society B20092763809381810.1098/rspb.2009.1146281729319656790Status and habitat of three subspecies of peromyscus polionotus in floridaHumphreySRBarbourDBJounal of Mammalogy19816284084410.2307/1380610A review of methods for the assessment of prediction errors in conservation presence/absence modelsFieldingAHBellJFEnvironmental Conservation199724384910.1017/S0376892997000088Predictive habitat distribution models in ecologyGuisanAZimmermannNEEcological Modelling200013514718610.1016/S0304-3800(00)00354-9AUC: A misleading measure of the performance of predictive distribution modelsLoboJMJiménez-ValverdeARealRGlobal Ecology and Biogeography20081714515110.1111/j.1466-8238.2007.00358.xSpecies distribution models: Ecological explanation and prediction across space and timeElithJLeathwickJRAnnual Review of Ecology, Evolution, and Systematics20094067769710.1146/annurev.ecolsys.110308.120159Spatially autocorrelated sampling falsely inflates measures of accuracy for presence-only niche modelsVelozSDJournal of Biogeography2009362290229910.1111/j.1365-2699.2009.02174.xConsequences of spatial autocorrelation for niche-based modelsSeguradoPAraujoMBKuninWEJournal of Applied Ecology20064343344410.1111/j.1365-2664.2006.01162.xMaximum entropy modeling of species geographic distributionsPhillipsSAndersonRSchapireREcological Modelling200619023125910.1016/j.ecolmodel.2005.03.026Modeling of species distributions with Maxent: New extensions and a comprehensive evaluationPhillipsSJDudíkMEcography20083116117510.1111/j.0906-7590.2008.5203.xNovel methods improve prediction of species' distributions from occurrence dataElithJGrahamCHAndersonRPDudíkMFerrierSGuisanAHijmansRJHuettmannFLeathwickJRLehmannALiJLohmannLGLoiselleBAManionGMoritzCNakamuraMNakazawaYOvertonJMcCTownsend PetersonAPhillipsSJRichardsonKScachetti-PereiraRSchapireRESoberónJWilliamsSWiszMSZimmermannNEEcography20062912915110.1111/j.2006.0906-7590.04596.xVery high resolution interpolated climate surfaces for global land areasHijmansRJCameronSEParraJLJonesPGJarvisAInternational Journal of Climatology2005251965197810.1002/joc.1276Mammal Network Information Systemhttp://manisnet.orgGlobal Biodiversity Information Systemhttp://www.gbif.orgUS Board on Geographic Names: Geographic Names Information Systemhttp://geonames.usgs.gov/pls/gnispublicValidation of species-climate impact models under climate changeAraújoMBPearsonRGThuillerWErhardMGlobal Change Biology2005111504151310.1111/j.1365-2486.2005.01000.x