Building relevant ecological indicators with basic data: Species and community specialization indices derived from atlas data

HAL Id: hal-02456171

https://hal.archives-ouvertes.fr/hal-02456171

Submitted on 27 Jan 2020

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of

sci-entific research documents, whether they are

pub-lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diffusion de documents

scientifiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

Building relevant ecological indicators with basic data:

Species and community specialization indices derived

from atlas data

Ruppert Vimal, Vincent Devictor

To cite this version:

Ruppert Vimal, Vincent Devictor. Building relevant ecological indicators with basic data: Species

and community specialization indices derived from atlas data. Ecological Indicators, Elsevier, 2015,

50, pp.1-7. �10.1016/j.ecolind.2014.10.024�. �hal-02456171�

Building

relevant

ecological

indicators

with

basic

data:

Species

and

community

specialization

indices

derived

from

atlas

data

Ruppert

Vimal

,

Vincent

Devictor

*

a

Centred’EcologieFonctionnelleetEvolutive,UMR5175,CNRS,1919RoutedeMende,Montpelliercedex534293,France

b

InstitutdesSciencesdel’EvolutiondeMontpellier,UMR5554,CNRS,UniversitéMontpellier2,PlaceEugèneBataillon,Montpelliercedex534095,France

ARTICLE INFO

Articlehistory: Received12June2014

Receivedinrevisedform24October2014 Accepted28October2014 Keywords: Atlasdata Communityspecialization Co-occurrence Indicators

Intermediatedisturbancehypothesis

ABSTRACT

Speciesandcommunityspecializationhavebecomepopularindicatorstotrackthespatialandtemporal changes of speciesand communitydynamics during current global changes. However, measuring specializationrequiresdetailedandquantitativedescriptionsofhabitatrequirementsorresourceuse, whicharedifficulttoobtainformanyspecies.Here,weproposeandtestanewmethodtoquantifyand maptherelativecompositionofspecialistandgeneralistspeciesinlocalplotscompatiblewithverybasic ecologicaldata,typicallyusedforatlases.Weusedco-occurrencepatternsof1090plantspeciesrecorded intheFrenchMediterraneanregionofLanguedoc-Roussilloninasystematicgridof122555kmatlas cellstoestimatespeciesspecialization.Wethencalculatedtheaveragedspecializationofeachcelland testedseveralexpectedrelationshipsoftheseindices.Inparticular,wetestedtherelationshipbetween speciesrichnessandaveragespecializationandtherelationshipbetweencommunityspecializationand landscapedisturbanceinducedbylanduse.Asexpectedfromstudiesconductedonfine-scaledata,we foundthatspecialistspecieswerethosewithmorerestricteddistributionsandoccurringinricherspecies assemblages.Wealsofoundthatcommunityspecializationwasmaximizedatanintermediatelevelof landscape disturbance. These resultssuggestthat aggregatingspecializationat largespatialscales provides useful species and community level indicators. Estimating specialization level with co-occurrencedataisagoodcomplementaryapproachtotraditionalestimationsofdiversityindicesfor conservationandlandscapeplanning.

ã2014ElsevierLtd.Allrightsreserved.

1.Introduction

Findingthebestindicatorofspeciesandcommunityresponses tolandscapedegradationisanongoingchallengeforecologists. Consequently,ecological indicatorsbased on_“speciesdiversity_” areverypopular;althoughtheirrelevancewasquestionedatan earlystage,whenspeciesdiversitywasconsideredanon-concept (Hurlbert,1971).Newindices(accountingforecological, phyloge-netic,or functionaldifferencesamongspecies)have,thus, been recurrently proposed to complement species diversity metrics (Monnet et al., 2014). Developing more relevant biodiversity indicatorshasbecome,however,ascientific,political,andsocietal issueofgreatimportance(FrederiksenandGudmundsson,2013). Butratherthansearching forthe“best” indicator,authorshave nowrecognizedthatindicatorsarenot“good”or“bad”butthat

theirrelevancedependsonthequestionaskedandonthedata available(Feestetal.,2010).

Toassessthelarge-scaleimpactsoflandscapedegradationon communities, ecological metrics reflecting the dynamics of “losers” versus _{“winners”} within species assemblages were proposedasapromisingapproachinconservationbiogeography (DevictorandRobert,2009).Inparticular,thereplacementrateof habitatspecialist species by generalistswas viewed as a direct signature of a community response to large-scale habitat degradation for animals and plants (Clavel et al., 2010;Abadie etal.,2011).Infact,itisgenerallyexpectedthathabitatspecialists will benefit from stable and undisturbed habitats whereas, generalistsshouldrespondpositivelytohabitatvariability(Colles et al., 2009). These expectations have been widely tested and ecologicalindicatorsbuiltuponthetemporaltrendsofspecialist species have been considered relevant official indicators of sustainable development for use at national and international levels(Gregoryetal.,2005).

Ideally,thespatialortemporal replacementofspecialistsby generalistscanbeestimatedusinglarge-scaleand standardized communitymonitoringprograms(Devictoretal.,2007).Withsuch

* Correspondingauthor.Tel.:+33467144081;fax:+33467143637. E-mailaddresses:[email protected](R.Vimal),

[email protected](V.Devictor).

http://dx.doi.org/10.1016/j.ecolind.2014.10.024

1470-160X/ã2014ElsevierLtd.Allrightsreserved.

ContentslistsavailableatScienceDirect

Ecological

Indicators

data,thespecializationofspeciesandcommunitiescanbederived fromthe statistical relationships reflecting species distribution alonghabitatgradients,bothmonitoredbystandardizedschemes. Inpractice,however,large-scalemonitoringdatafornationalor regionalsurveysarecurrentlybeingcollectedforonlyafewgroups (mostly birds, butterflies, and mammals) and are based on presence–absencedata.Moreover,measuringspecies specializa-tionisoftenimpairedbythelackofhighenoughresolutiondataon habitat requirements or by the difficulty of defining habitat selectionaccurately(PodaniandCsányi,2010).Consequently,two mainapproacheshavebeenusedtoquantifychangeincommunity composition following landscape disturbance: (i) at global or nationalscales,someauthorshaveused crudeclassificationsof species into specialist versus generalist groups. For instance, indicators for the state of the European avifauna rely on the averagetrendofsomespecies,classifiedasbeingspecializedfora givenhabitattype(e.g.,farmlandbirdspecialists)(Gregoryetal., 2005),(ii)incontrast,others haveusedhigh-resolutiondataon detailed species requirements. In this case, continuous and species-specific levels of ecological specialization was derived from standardized protocols, in which habitat or resource preferences could be precisely assessed (Devictor et al., 2007; CorreaandWinemiller,2014).However,methodstoestimatethe specialization level of species and communities using classic ecologicaldata(i.e.,thepresenceorabsenceofspeciesacrosssites) arelacking;although,theycouldhelptotrackthefateofspecies andcommunitiesinmanycontexts.

Interestingly, Fridley et al. (2007) proposed a method to estimate species specialization that only requires presence– absencedata.Itassumesthat speciesco-occurringwithsimilar speciesareusuallythosefoundinsimilarhabitatsandcould,thus, beconsideredspecialists.Conversely,generalistsshouldbewidely distributedacrosshabitatsandthusco-occurwithmanydifferent species.Inotherwords,foragivenspecies,thesimilarityinthe identity of species co-occurring with that species can be considered,according tothis approach, a continuous proxy for specieshabitatspecialization.Fromthisassumption,and provid-ingthatco-occurrencedataareavailable,aspeciesspecialization index(SSI)canbesimplydeducedforeachgivenspeciesusingthe identity of the species co-occurring with that species. This approachcanbeappliedtoanydatasetprovidingthatdifferent species assemblages have been recorded in different locations (Abadieetal.,2011;Boulangeatetal.,2012).Usingthisapproach, specialization was equated to niche breadth to test a specific hypothesis onthe role played bycompetition (Manthey et al., 2011),orspecificfunctionaltraits(Albertetal.,2010)inspecies distribution. Although originally developed for plants, this approachhasalsobeensuccessfullyusedforamphibians(Rannap etal.,2009),andfishes(Munroeetal.,2013).

This approach does not a priori tell whether ecological specializationcanberelevantwhenmeasuredfordatacollected at coarse spatial grain. Indeed, co-occurrence patterns are expectedtoyielddifferenttypesofspecializationwhenestimated at the quadrat, landscape, or regional scale. In this respect, althoughFridley’smethodhasbeenappliedtovariousorganisms indifferentcontexts,itsrelevanceforco-occurrencedataobtained fromspecieslistsrecordedacrosslargespatialscaleshasnotbeen explored(butseeBoulangeatetal.,2012).Furthermore,itremains unclearwhetherspecializationisstillrelevantandsensitivewhen definedatspatialscalesdifferentfromthosemostlikelytocapture habitatselectionandspeciesinteractions.

Oncethespecializationlevelsofspeciesareestimatedandare sensitiveenoughtohabitatdisturbance,thedistributionofspecies and assemblages according totheir specialization level can be investigated.In particular, specialist speciesare expectedtobe more numerous and to concentrate more individuals in less

fragmentedlandscapes(Devictoretal.,2007).Atthecommunity level,acommunityspecializationindex(CSI)ofspecies assemb-lagescanbecalculatedastheaverageofeachspeciesSSIpresentin theassemblage(Devictoretal.,2008).TheCSIisexpectedtobe higher for species assemblages mostly composed of specialist species.Itcanthenbeusedasaninterestingecologicalindicator complementarytomoretraditionalindicatorsbasedondiversity (Filippi-Codaccionietal.,2010;Abadieetal.,2011).Mappingthe CSI can thus provide a picture of spatial variation in the specialization level of communities, which can be related to independentsourcesofdisturbanceorusedasaspatialguideline toidentifysitesofconservationinterest(Devictoretal.,2008).Yet whethersuchacommunityspecializationindexcanbeusedasa relevantecologicalindicatorwithbasicecologicaldatahasnever beenexplored.

Here,weusedalarge-scaleco-occurrencedatasetonplantsto estimateaspeciesspecializationindex(SSI)foreachspeciesanda communityspecializationindex(CSI)foreachgridcell.Wethen specificallytestedseveralhypothesesonSSIandCSIderivedfrom studiesonspecializationconductedwithhigherresolution data and atfiner spatialresolution. Inparticular, weinvestigated (i) whetherandhowthespeciesdistributionwasdependentontheir SSI.Attheassemblagelevel,wetested(ii)therelationshipbetween CSIandspeciesrichness,and(iii)therelationshipsbetweenthese twometricsandlandscapedisturbance.

2.Methods

2.1.Thestudyregion

ThestudywascarriedoutintheLanguedoc-Roussillonregion (27,376km2_{)insouthernFrance,whichencompassesmostofthe}

MediterraneanregionwestoftheRhônevalley(Fig.1).Themain landscape types occurring here are coastal landscapes with

Fig. 1.Thestudyregionandthedistributionofoccurrencedata.Eachdotrepresents aspecieslistrecordedinthedatabase.Thegridcellsof55kmusedforthe aggregationaredelineated.

lagoons,marshes,cliffsanddunes,andlowlandgarrigues.These habitatsareoftenincludedinmosaiclandscapeswithcultivated areas, vast areas of vines, extensive upland limestone plateau areas,andhillyormountainouslandscapesongraniteandschistin the southern tip of the Massif Central and the south-eastern Pyrenees. In the last 50 years, these landscapes have been profoundlymodifiedbyhumanactivitiessothatheavilydisturbed habitatscoexistwithstableones(Thompson,2005;Blondeletal., 2010).First,extensiveandrapidurbanizationspreadaroundtowns andvillagesacrossthelowlandplainsinconjunctionwithmassive proliferationofcoastaltouristresorts.Second,humanpopulation declineinmanyruralareaswasaccompaniedbytheabandonment ofvineyardsand grazing activity, which wasfollowed byrapid naturalreforestationofagriculturalfields(IFEN,2003).

2.2.Speciesdata

Plant species data were compiled by the Conservatoire Botanique National Méditerranéen de Porquerolles (CBNMP), which isin chargeof theregionaldatabaseof allplantspecies presentinthisregion.Originaldatawerecollectedbetweenthe years1611and2009andcorrespondtothecombinationofallthe naturalist inventories, herbariums, bibliographies, reports, and atlasesavailableintheregion.Forthepurposeofthisstudy,only dataonvascularplantspeciescollectedsincetheyear1980were used.Adatabaseof3667specieswith420,659occurrenceswas obtained. All data were then combined and aggregated in a systematic gridof 1227 55km cells to define a standardized spatialscaleresolution(Fig.1).Eachgridcellwasconsideredasite ofspeciesco-occurrence.Therobustnessofourresultstochanges in the scale of aggregationwas also tested with 22km and 1010kmgrids.

In large-scale biodiversity inventories, spatial, and temporal variation in sampling can introduce significant biases into the representation of species distribution. Here, species have been sampled by different observers using several methodologies. Although this variability in the protocol represents a lack of standardization,thespecies selection toestimate specialization (seebelow)furtherensuredthatitisestimated,foreachspecies, withthesamesamplesize.

2.3.Measuringspeciesandcommunityspecializationindices

First,thealgorithmproposedfromFridleyetal.(2007)wasused to measure a species specialization index (SSI) for each plant species.Thisapproachhasrecentlybeenshowntobeasuitable methodfor measuring plantspecialization usinglarge samples acrossveryheterogeneousenvironments(Boulangeatetal.,2012). Foreachgivenspecies,arandomcombinationofsites(50sites)in whichthis speciesoccurs isselected.Asimilarity indexisthen calculated between each pair of sites to re_flect the degree of betweensitesspecies turnoverinthe50 sites.Thisis repeated 100timesforeachspecies.Foreachofthe100repetitions,anew randomsetof50sitesisthusselectedandacorrespondingSSIis calculated.TheoverallSSIofthegivenspeciesiscalculatedasthe averageofthe100SSIobtained(seeFridleyetal.,2007).Notethat speciesoccurringinfewerthan50sitesarenotconsidered,asit wouldbeunreliabletomeasuretheirspecializationlevelfrom co-occurringspecies.Amongthe3667species,1090werepresentin enoughsitestoprovideanSSIsoallanalyseswereconductedon thesespecies.Notealsothat,inthisapproach,theSSIisalways calculatedforeachspeciesfromafixednumberofsites(50).In each combination, the sites are selected randomly across the speciesrangewithinthestudiedarea.Thus,specializationofrare and common species is derived from combinations of species assemblagesofequalsize.ThisapproachprovidesanincreasingSSI

valuefromthemostgeneralist(i.e.,those expectedtoco-occur with more different species and thus generate less similarity betweensites)tothemostspecialistspecies(co-occurringwith moresimilarpoolsofspecies).

Ecologistshaveusedalargenumberofdifferentmeasuresof community dissimilarity(alsocalledbeta-diversityorturnover) withdifferentpropertiesandmeanings(Koleffetal.,2003).Here, we measured similarity using the average of pairwise

b

calculatedamong sites(Baselga,2010).Fortwo sites,

b

indexgivenby

b

ofspeciescommontobothsites,bisthenumberofspeciesthat occurinthe_firstsitebutnotinthesecondandcisthenumberof speciesthatoccurinthesecondsitebutnotinthefirst.Thisindex variesbetween0(allspeciesshared)and1(nospecieshared).We then used SSI=1

b

therefore,alsotheoreticallyboundedbetween0(mostgeneralist) to1(mostspecialist).

Notethat otherdissimilarityindicesincludingspecies abun-dancewhenavailableandindependentofspeciesrichnesscanalso beused(Boulangeatetal.,2012).However,othertraditionalways of measuring similarity between plots (e.g., in partitioning diversity inlocal,regional,and among-sitecomponents(Lande, 1996))shouldbeused withcaution.In fact,a commonthough unwantedpropertyofthesealternativesimilaritymetricsistobe correlated to species richness (Koleff et al., 2003).In Fridley’s algorithm, these other metrics tend to be highly sensitive to species occurringin species-poorhabitats, which havestrongly skewed richnessdistributions (Mantheyand Fridley,2009).The Simpson’spairwiseindex(

b

similarity(Baselga,2010)soweuseditasa goodindexof beta-diversity.

Once an SSI was obtained for each species, a community specializationindex(CSI)wascalculatedforeachgridcellasthe averageoftheSSIvaluesbelongingtothespeciespresentinthis cell(Devictoretal.,2008).Onlycellsinwhichatleast10species were present were considered. The CSI is higher for species assemblageswithmorespecialistspecies(i.e.,withahighSSI)and is,byconstruction,independentofspeciesrichness.

2.4.Measuringlandscapedisturbance

The indicator of spatial disturbanceused was based onthe compilation,foreachsite,ofthreekindsofhumanpressure:road density, urbanization, and agriculture. Note that disturbance is used hereasa generic termwithoutspecific expectationof its negativeorpositiveimpactonplantassemblages:someofthese artificiallandscapemodificationscanbepositiveforsomespecies and negativefor others.For road densityand urbanization,the “road”and“built-up”layersfromtheBDTOPO1/RGEGISdatabase (IGNInstitutGéographiqueNational)wereused.Foragriculture, the_“arableland_”,“mixedagriculture_”, and_“permanentculture_” layersfromthecorinelandcoverdatabase(Bossardetal.,2000) wereused.Foreachsite,theproportionofdisturbanceelements withinthesitewascalculated.Then,adisturbanceindicatorwas calculatedforeachsiteasthemeanvalueofthenormalizedvalue (from0to1)foreachproportion.Thisdisturbanceindicatorwas explicitlytestedinthisregionandwasshowntoprovidearelevant proxy for mapping the spatial distribution of the intensity of human-induced modification of landscape composition (Vimal etal.,2011).

2.5.Dataanalysis

We first focused on the characteristics of more or less specialized species. Wetested whetherthespecies distribution in the area considered was related to their SSI using linear

regression.SSIwas consideredas adependentvariableandthe numberofsitesoccupiedbythespeciestheexplanatoryvariable assuming independence of the observations and a Gaussian distributionoftheerrors.SpecieswithlowerSSI(moregeneralists) wereexpected to occur in many sites in line with the Brown hypothesis(Brown,1984; Gaston, 2003).Indeed, niche breadth shouldreflect the degree to which species requirements meet environmentalconditions.Nichebreadthis,therefore,generally foundtobepositivelyrelatedtospeciesoccupancy(i.e.,generalist are those with broader niches, and thus with wider regional occupancy).

Similarly,howspecialistandgeneralistspeciesweredistributed inrichandpoorspeciesassemblageswastested.Todothis,we usedlinearregressiontotesttherelationshipbetweenSSIandthe average species richness of the assemblage where the species occurs.WeexpectedspecieswithlowerSSItobefoundin species-poor assemblages following lower niche packing in these assemblages.Onthecontrary,weexpectedspecies-rich assemb-lagestofavor specialistspecies, bestable topartition thetotal ecologicalnichespace(Belmakeretal.,2011).

Finally,wetestedwhethertherichnessandCSIofassemblages wererelatedtoenvironmentaldisturbance.Weexpectedspatial dependencyin observations andpotential non-linearityin the relationships.Weusedgeneralizedleastsquare(GLS)modelsto testthisrelationship. GLSare specific weightedregressions in whichadirectmarginalvariance–covariancespatialstructureof theresponsescanbespecified(Zuuretal.,2009).Thisstructure wasfirstinvestigatedusingdifferentformsofsemi-variogramsto accountforspatialdependencyoftheresiduals.Thebestspatial structure (exponential) and corresponding range and nugget werethenaddedtothemodel(usingthefunctionGLSandCorExp inthepackagenlmeinR).R2_{isnotwelldefinedforGLS.Thus,to}

getaroughestimateofthegoodnessoffitforthesemodels,we usedtheR2_{oflinearregressionsaccountingforspatialgradientsin}

whichpolynomialtermsof thecoordinates(x,y, x2_,y2_{,and xy)}

wereadded as covariates (Fortin and Dale, 2005).Community descriptors are often found to be non-linearly related to environmentalvariables.Totest forany potentialhump-shaped relationshipsbetweenspeciesrichness(orCSI)anddisturbance, we used the same model than above with disturbance and disturbance2aspredictors.Allstatisticalanalyseswerecarriedout usingR2.11softwareandthepackage“nlme”forGLSmodels(R DevelopmentCoreTeam,2014).

3.Results

Atthespecieslevel,whencalculatedusinga55kmgrid,the speciesspecializationindex(SSI)rangedfrom0.21to0.43(mean 0.320.03s.e.).Itwasrobusttowardschangeinthespatialscale considered (correlations between SSI calculated at 5km2 _and

2km2_,R2_=0.82,P

<0.001andbetween5km2_and10km2_,R2_=0.83,

P<0.001).Generalistspeciesweremorewidelydistributed(i.e., occurred in more sites) than specialist species (F1,1088=244;

P_<0.0001; R2_{=0.18; Fig. 2a). Interestingly, the shape of the}

relationshipsuggeststhatbothspecialistsandgeneralistscanhave restricteddistributions(i.e.,occurinfew sites)while, compara-tively,onlygeneralistscanhavewidedistributions.Therewasalso astrongpositivelinearrelationshipbetweenspecializationandthe mean species richness of co-occurring species (F1,1088=357;

P<0.0001;R2=0.25;Fig.2b). Inotherwords, specialist species tendtooccurinricherassemblagesthangeneralistsdo.

Atthespeciesassemblagelevel,theCSIincreasedlinearlywith speciesrichnesssuggestingthatrichassemblagesareprincipally composedofspecialists(GLS:F1,1225=26.5;P<0.0001;R2=0.16,

Fig. 3). Note that the variability of CSI values was unevenly distributedalongthespeciesrichnessgradient:poorassemblages

included assemblageswithhigh and low CSIvalues while rich assemblagesmostlyconsistedofspecialists.

The relationships between CSI or species richness and disturbance were poorly described using linear models (not significant for CSI, P=0.23, nor for species richness, P=0.84). However, there were curvilinear relationships betweenspecies richness and disturbance (complete quadratic model: R2=0.16; P<0.0001;quadratictermb= 0.0024,P=0.007).Aspartofthe relationship between CSI and disturbance could potentially be drivenbytherelationshipbetweenCSIandspeciesrichness,the effectofdisturbanceonCSIwastestedwhilecontrollingforspecies richness byaddingit asacovariate. Whenvariation inCSIwas adjusted for variation in species richness, there was still a

Fig.2. Relationshipsbetweenthespeciesspecializationindex(SSI)and(a)the numberofsiteswherethespeciesoccursand(b)theaveragesitespeciesrichness wherethespeciesoccurs.

Fig.3. Relationshipbetweenthecommunityspecializationindex(CSI)ofspecies assemblagesandtheirspeciesrichness.

curvilinear effect of disturbance on CSI (R2_{=16; P}

<0.0001; quadratictermb= 0.25;P=0.003).

Finally,mappingspeciesrichness,landscapedisturbance,and communityspecializationrevealedspecificareasofcongruencies andmismatches(Fig.4).Forinstance,zonezillustratesanareain whichrichnessandCSIarebothofhighvalue.Theseareasareof high conservation interest, yielding many specialist species. Comparatively, in zone ywhere landscape disturbance is high, richness,andCSIarerelativelylow,suggestingthattheseareareas where plant communities have been strongly impacted by disturbance leading to poor assemblages mainly composed of generalists.

4.Discussion

Usingverybasicco-occurrencedata,wewereabletosegregate speciesalongacontinuousgradientofspecialization.Obviously,in thisapproach,“specialization”isnotequivalenttonichebreadthas generallyestimatedusingfinespeciesresponsestoenvironmental conditionsorspecificfunctionaltraits.Inourstudy,specialization rather reflects the tendency of species to occur in different landscapescomposedofdifferentspecies.Insteadofprovidinga traditionalnichebreadthindex,thisapproachgivesanoperational and quantitative metric for measuring the similarity of co-occurringspecies,relevanttolarge-scaleoccurrencedata.

The ecological meaning of this species-specific attribute is scale-dependent.Whileplantspeciesinteractatverylocalscales, theco-occurrencepatternsatlargerscaleswillbemoreinfluenced byregionalanddispersalprocesses.Inotherwords,aspeciescan beconsideredspecialist atalandscapelevelalthough thesame speciescouldbeageneralistforspecifichabitatswithinlandscapes (Devictoretal.,2010;Boulangeatetal.,2012).

Nevertheless,ourspeciesspecializationindex(SSI)estimated withatlasdatawasrobusttowardschangesinthescaleconsidered (22km,55km,or1010km)whichsuggestsweakvariations in landscape composition between such scales. It is therefore, likelythattheSSIcalculatedat55kmalreadycapturesrelevant variabilityinthecompositionofspeciesassemblagesintheregion

considered. This _first result also suggests that large-scale specialization estimated with atlas data reflects an interesting and robust characteristic of species. Moreover, co-occurrence-basedindicesofspecializationwereshowntobecorrelatedwith moredetailedandcommonlyusedmetricsofnichebreadthaswell aswithspecializationmetricsderivedfrommultivariateanalysis includinghabitatvariablesmoreexplicitly(seeBoulangeatetal., 2012forcomparisonofindices).

Other results further suggest that estimating specialization withatlasdataprovidesrelevantresults,inharmonywiththose found with finer estimations of niche breadth. First, when specialization is measured for specific habitats or resources, specialist species are generally foundtobe those withsmaller ranges(Gaston,2003).Here,wefoundthatspeciesco-occurring withmanydifferentspecies(i.e.,withlowSSI)arealsothosewith largerdistributionsin theregionconsidered(Fig.2a),a pattern already documented for the regional _flora of the French Alps (Boulangeatetal.,2012).Atfinescales,greaternichepartitioningis alsoexpectedwhen speciesrichness(andpossiblycompetition) increases(Masonetal.,2008).Here,wealsofoundthatspecialist species tend to occur in richer assemblages (Fig. 2b). The relationshipbetweenCSIand speciesrichness wasalsopositive (Fig.3)showingthatricherassemblageswerethoseconcentrating more specialist species and that, inversely, poorer assemblages werethoseconcentrating moregeneralistspecies.Theseresults support those derived from fine-scale communities in which speciesinteract(Belmakeretal.,2011).

Theserelationshipscould,however,misscomplexcommunity responses to large-scale disturbance. For instance, non-linear (Daveyetal.,2012)andnegative(Filippi-Codaccionietal.,2010) relationships between CSI and species richness have been documented in human disturbed landscapes. Here, we found similarcurvilinear relationshipsbetweenlandscapedisturbance and species richness or community specializationindex,which also correspond to those widely described in the so-called intermediatedisturbancehypothesis(Wilkinson,1999).According to this hypothesis, competitive, and specialist species should dominateandexcludeothersatlowlevelsofdisturbance,andonly

Fig.4.Thespatialdistributionof(a)disturbance(b)speciesrichness(c)communityspecializationindex(thevalueincreasesfrompaletodarkcolors.Themissingvaluesare inwhite).Asexamples,yisazoneofrelativelyhighdisturbancebutwithlowrichnessandCSI.Incontrast,zisanareaoflowdisturbancebuthighCSIandrichness.(For interpretationofthereferencestocolorinthisfigurelegend,thereaderisreferredtothewebversionofthisarticle.)

afewgeneralistspeciescanthriveinhighlydisturbedsites.Species richness should therefore be greater at intermediate levels of disturbance,under which both specialistand generalist species can coexist. Although the mechanisms of such curvilinear relationshipshave been debated, our results can be explained by the Mediterranean mosaic landscapes in which human activitiescancontribute,uptoacertainpoint,toincreasinglocal speciesrichness(Thompson,2005;Blondeletal.,2010).

ThecurvilinearrelationshipbetweenCSIanddisturbancealso suggests that the coexistence of many specialist species is maximizedatanintermediatelevelofanthropogeniclandscape modi_fication.Thispatternisprobablydrivenbythepresence,in this region,of manyspecies co-occurringin human-dominated landscapes.Thisresultalsogeneralizes previousfindings estab-lished with fine-scale data showing that individual specialist species can be associated with habitats disturbed by humans (DavisonandFitzpatrick,2010;Boulangeatetal.,2012).Although many studies have shown that, following habitat disturbance, generalistsshouldreplacespecialists(Devictoretal.,2008),others have documented the relative increase in specialist species in disturbedassemblages(Claveroetal.,2011).Here,wehaveshown thatthedirectionofchangeincommunitycompositioncanalsobe dependenton disturbance intensity. Although a more detailed analysisofthemechanismsleadingtonon-linearrelationshipsis needed,theseresultssuggestthat,asalreadystatedatalocalscale, large-scalespecializationandCSIwillnotalwaysbeasurrogateof ecosystemqualitybutshouldratherbeusedinconjunctionwith other community descriptors (Filippi-Codaccioni et al., 2010; Filippi-Codaccionietal.,2010).

Overall,ourresultssuggestthatmeasuringspeciesand commu-nity specialization (SSI and CSI) generates meaningful results comparedtowhatisexpectedusingmoredetailedandfinerdata onspeciesassociationswithhabitats.Someofourresults,however, could be biased bythe way datawere collected. Inparticular, a greater samplingeffortcouldbeexpectedforrare,charismatic,orendemic species.However, theSSI was calculated only for thosespecies presentinatleast50squares,thus,eliminatingtheeffectofrareand occasional species. The randomization procedure of Fridley’s algorithmalsoremovesmostofthepotential non-uniform collection effortthroughoutthestudied region.Notealso that depending onthe data considered, this technique could induce a link between specializationandrarity.Foragivenspecies,theindexisestimated as the average (after 100 repetitions) of the similarity indices calculatedfrom50sitesrandomlyselectedamongthesiteswhere this speciesoccurs.For rareandlocalized species,the probability that tworepetitionsincludemoreoftenthesamesitesis higher.The variabilityinthesimilarityindicesgeneratedisthuslowerforrare species.Althoughthisdoesnotnecessarilybiasestheindexitself,it mightaffect theconfidence in specializationestimates. A more elaborated algorithm in which the number of repetition is proportionaltospeciesraritycouldbeused.Finally,theobjective ofouranalysiswasnottodefinethelocal variationinSSI orCSIvalues precisely but rather to describe their relative variations across speciesoracrosslargespatialgradients.Thebiasesabove,ifany, wouldhardlyexplainthecurvilinearrelationshipbetweenCSIand disturbanceandthepositiverelationshipbetweenCSIandspecies richness.

We believe that our approach offers an interesting tool to delineate areas of conservation interest based on the spatial variationincommunitycomposition.TheCSIderivedfromatlas datacouldenablemanagerstoimplementdifferentiated conser-vation plans among ecoregions when coupled with other indicators.Asimilaranalysiscanbeconductedonanygroupor data providing that co-occurrence-based specialization can be estimated.Moreover,theestimationofatemporalchangeinthe CSIcalculated in areas where datahave been collectedseveral

times should be a promising route to shed light on biotic homogenization.

Acknowledgements

WethankJamesMolinaandFredericAndrieuatthe Conserva-toireBotaniqueNationalMéditerranéendePorquerollesfortheir advice and access to data. We are grateful to staff at the Conservatoire des Espaces Naturels and to all the naturalists whocollecteddataonwhichthisanalysis wasbased.Wethank CoralieCalvetforherencouragement.Wealsothanktworeviewers for theirconstructivecomments. Thisworkwas funded by the AgenceNationaledelaRecherche(contract05-BDIV-014,ABIME) andtheLanguedoc-RoussillonRegionalCouncil.VincentDevictor wasfundedbytheFondationpourlaRecherchesurlaBiodiversité (FRB,projectsFABIOandPHYBIO).

References

Abadie,J.C.,Machon,N.,Muratet,A.,Porcher,E.,2011.Landscapedisturbancecauses small-scalefunctionalhomogenization,butlimitedtaxonomic homogeniza-tion,inplantcommunities.J.Ecol.99,1134–1142.

Albert,C.H.,Thuiller,W.,Yoccoz,N.G.,Soudant,A.,Boucher,F.,SacconeLavorel,P.S., 2010. Intraspecific functionalvariability: extent,structure and sources of variation. J. Ecol. 604–613. doi:http://dx.doi.org/10.1111/j. 1365–2745.2010.01651.x.

Baselga,A.,2010.Partitioningtheturnoverandnestednesscomponentsofbeta diversity.GlobalEcol.Biogeogr.19(1),134–143.

Belmaker,J.,Sekercioglu,C.,Jetz,W.,2011.Globalpatternsofspecializationand coexistence in bird assemblages. J. Biogeogr doi:http://dx.doi.org/10.1111/j 1365–2699.2011.02591.x.

Blondel, J., Aronson, J., Bodiou, J.-Y., Boeuf, G., 2010. The Mediterranean Basin –Biological Diversity in Space and Time. Oxford University Press, Oxford,UK.

Bossard,M.,Feranec,J.,Otahel,J.,2000.CORINELandCoverTechnicalGuide– Addendum2000.TechnicalReportNo.40.EuropeanEnvironmentAgency, Copenhagen.

Boulangeat,I.,Lavergne,S.,VanEs,J.,Garraud,L.,Thuiller,W.,2012.Nichebreadth, rarityand ecologicalcharacteristicswithin aregional floraspanninglarge environmentalgradients.J.Biogeogr.39(1),204–214.

Brown,J.H.,1984.Ontherelationshipbetweenabundanceand distributionof species.Am.Nat.124(2),255–279.

Clavel, J., Julliard, R., Devictor, V., 2010. Worldwide decline of specialist species:towardaglobalfunctionalhomogenization?Front.Ecol.Environ.9, 222–228.

Clavero, M., Brotons, L., Herrando, S., 2011. Bird community specialization, birdconservationanddisturbance:theroleofwildfires.J.Anim. Ecol.80, 128–136.

Colles,A.,Liow,L.H.,Prinzing,A.,2009.Arespecialistsatriskunderenvironmental change?Neoecological, paleoecologicaland phylogeneticapproaches.Ecol. Lett.12,849–863.

Correa,S.B.,Winemiller,K.O.,2014.Nichepartitioningamongfrugivorousfishesin responsetofluctuatingresourcesintheAmazonianfloodplainforest.Ecology 95,210–224.

Davey,C.M.,Chamberlain,D.E.,Newson,S.E.,Noble,D.G.,Johnston,A.,2012.Riseof the generalists: evidence for climate driven homogenization in avian communities.GlobalEcol.Biogeogr.21,568–578.

Davison,M.A.,Fitzpatrick,J.W.,2010.Roleofhuman-modifiedhabitatinprotecting specialistspecies: a casestudy in the threatened FloridaScrub-Jay. Biol. Conserv.143,2815–2822.

Devictor,V.,Julliard,R.,Jiguet,F.,Couvet,D.,2007.Distributionofspecialistand generalistspeciesalongspatialgradientsofhabitatdisturbanceand fragmen-tation.Oikos117,507–514.

Devictor,V.,Clavel,J.,Julliard,R.,Lavergne,S.,Mouillot,D.,Thuiller,W.,Venail,P., Villeger,S.,Mouquet,N.,2010.Definingandmeasuringecological specializa-tion.J.Appl.Ecol.47,15–25.

Devictor,V.,Julliard,R.,Clavel,J.,Jiguet,F.,Lee,A.,Couvet,D.,2008.Functionalbiotic homogenizationof birdcommunitiesindisturbedlandscapes. GlobalEcol. Biogeogr.17,252–261.

Devictor,V., Robert,A., 2009.Measuringcommunity responses tolarge-scale disturbanceinconservationbiogeography.Divers.Distrib.15,122–130.

Feest,A.,Aldred,T.D.,Jedamzik,K.,2010. Biodiversityquality: aparadigmfor biodiversity.Ecol.Indic.10,1077–1082.

Filippi-Codaccioni,O.,Devictor,V.,Bas,Y.,Julliard,R.,2010.Towardmoreconcern for specialisation and less for species diversity in conserving farmland biodiversity.Biol.Conserv.143,1493–1500.

Fortin,M.,Dale,M.,2005.SpatialAnalysis. AGuideforEcologists.Cambridge UniversityPress.

Frederiksen,P.,Gudmundsson,H.,2013.Policyuseandinfluenceofindicators.Ecol. Indic.35,1–2.

Fridley,J.D.,Vandermast,D.B.,Kuppinger,D.M.,Manthey,M.,Peet,R.K.,2007. Co-occurrencebased assessmentofhabitat generalists andspecialists:a new approachforthemeasurementofnichewidth.J.Ecol.95,707–722.

Gaston,K.J.,2003.StructureandDynamicsofGeographicRanges.OxfordUniversity Press,Oxford,UK.

Gregory,R.D.,vanStrien,A.,Vorisek,P.,GmeligMeyling,A.W.,Noble,D.G.,Foppen, R.P.B.,Gibbons,D.W.,2005.DevelopingindicatorsforEuropeanbirds.Philos. Trans.R.Soc.London.Ser.B,Biol.Sci.360,269–288.

Hurlbert,S.H.,1971.Non-conceptofSpeciesdiversity–critiqueandalternative parameters.Ecology52,577–586.

L’environnementenLanguedocRoussillon,IFEN,Orléans,France.(2003)

Koleff, P., Gaston, K.J., Lennon, J.J., 2003. Measuring beta diversity for presence–absencedata.J.Anim.Ecol.72,367–382.

Lande,R.,1996.Statisticsandpartitioningofspeciesdiversity,andsimilarityamong multiplecommunities.Oikos76,5–13.

Manthey,M.,Fridley,J.D.,2009.Betadiversitymetricsandtheestimationofniche widthviaspeciesco-occurrencedata:replytoZeleny.J.Ecol.97,18–22.

Manthey,M., Fridley, J.D., Peet, R.K.,2011. Niche expansionafter competitor extinction?Acomparativeassessmentofhabitatgeneralistsandspecialistsin thetreeflorasofsouth-easternNorthAmericaand south-easternEurope. J.Biogeogr.38,840–853.

Mason,N.W.H.,Irz,P.,Lanoiselee,C.,Mouillot,D.,Argillier,C.,2008.Evidencethat niche specialization explains species-energy relationships in lake fish communities.J.Anim.Ecol.77,285–296.

Monnet,A.-C.,Jiguet,F.,Meynard,C.N.,Mouillot,D.,Mouquet,N.,Thuiller,W., Devictor, V., 2014. Asynchrony oftaxonomic, functionaland phylogenetic diversityinbirds.GlobalEcol.Biogeogr.:AJ.Macroecol.23,780–788.

Munroe,S.E.M.,Simpfendorfer,C.A.,Heupel,M.R.,2013.Definingsharkecological specialisation:concepts,context,andexamples.Rev.FishBiol.Fish.24,317–331.

Podani,J.,Csányi,B.,2010.Detectingindicatorspecies:someextensionsofthe IndValmeasure.Ecol.Indic.10,1119–1124.

Rannap,R.,Lõhmus,A.,Briggs,L.,2009.Nicheposition,butnotnichebreadth,differs intwocoexistingamphibianshavingcontrastingtrendsinEurope.Divers. Distrib.15,692–700.

Thompson,J.D.,2005.Plant EvolutionintheMediterranean.OxfordUniversity Press,Oxford.

RDevelopmentCoreTeam,2014.R:Alanguageandenvironmentforstatistical computing.R FoundationforStatisticalComputing,Vienna, Austria. ISBN 3-900051-07-0,http://www.R-project.org.

Vimal,R.,Pluvinet,P.,Sacca,C.,Mazagol,P.-O.,Etlicher,B.,Thompson,J.D.,2011. Exploringspatialpatternsofvulnerabilityfordiversebiodiversitydescriptorsin regionalconservationplanning.Environ.Manage.95,9–16.

Wilkinson,D.M., 1999.Thedisturbinghistoryofintermediatedisturbance.Oikos84, 145–147.

Zuur,A.F.,Ieno,E.N.,Walker,N.J.,Saveliev,A.A.,Smith,G.M.,2009.MixedEffects ModelsandExtensionsinEcologywithR.Springer,NewYork,NY,USA.