Factors Influencing Nest Site Selection, Breeding Density and Breeding Success in the Bearded

Factorsinfluencingnestsiteselection,breeding densityandbreedingsuccessinthebeardedvulture (Gypaetusbarbatus)

J.A.DONAZAR,F.HIRALDOand J.BUSTAMANTE*t

EstaciOn BiológicadeDoñana, (CSIC),PabellóndelPeru,Avda M’LuisaSn, 41013Sevilla,Spain;and

*csIRo,DivisionofWildlife Ecology, P0 Box84Lyneham,ACT 2602,Australia

Summary

1. We examined the nest site selection,breeding density and breeding success in the bearded vulture Gypaetus barbatus in relation to physiography, climate, land-use and degree ofhuman disturbance.The study area wasinthe Pyrenean Cordillera,Spain, where thelargest Europeanpopulationofthisspeciesoccurs. UnivariateanalysesandGeneralizedLinear Modelswereemployed.

2. Modelscorrectly classifiedthe78%ofthecliffsanalysed (occupied bybearded vultures,andselectedatrandom).Theprobabilityofoccupationofacliffbybearded vultureswasdirectlyrelated totheruggednessofthetopography,altitude,distance tothenearest bearded vulture occupied nest, anddistance tothenearest village.

3.Breedingdensitywaspositivelycorrelated withaltitudeandruggednessofthe topography andnegativelycorrelatedwithsnowprecipitation.Open areasseemed alsotohavepositiveeffects,probablybyincreasingtheavailabilityoffood,although itseffectswerenotseparablefrom thatoftherelief, asthetwofactors covary.

4. Beardedvultures showed lower breeding successinareas withhigh potential

human disturbance(density ofpaved roads). The existence ofabrupt and open landsmighthaveapositiveeffectonbreeding successbyreducing accessibilityto humans, andperhaps byincreasing foodavailability.

Key-words:nest-siteselection,breedingdensity,breedingsuccess,bearded vulture, conservation.

504

Introduction

The bearded vulture (Gypaetus barbatus L.) is a cliff-nestingaccipitridvulture inhabitingOld Worldmountainrangesandfeeding onbones, predominantlyof medium-sizedungulates(Hiraldo etal.1979;Brown1988;BrownPlug1990).Its breedingdistributionhasbeengreatlyreducedin Europesincethelastdecadesofthe nineteenth century andisrestricted nowadaystothePyrenees, SouthernBalkans and the islands of Corsica and Crete (Hiraldoetal. 1979). The European popu­ lationisestimatedtobec.120breedingpairs(Elosegi

1989; R. Heredia,personal communication).The biggestpopulation isthatofthePyrenees with72 occupied territories(52ontheSpanish sideand20 ontheFrenchside(R. Heredia,personalcommuni­ cation)).The species isconsidered asendangered

Presentaddress:Nationalparkverwaltung,Doktorberg

6,8240Berchtesgaden, Germany.

bothinSpain(ICONA1986)andinEurope (Conseil del’Europe1981),andisincludedintheAnnexIof theDirective79/409/EEC asaspeciessensitiveto habitat alterations.

Thedeclineofthisspecieshasbeen attributedto

several causes. The cooling of the climate could havebeenresponsibleforthedeclineintheAlps during the nineteenth century (Haller 1983) but direct persecution,killingofadults and robbery of eggs and chicks, and indirect mortality caused by poison baiting of carnivores are the more widely accepted causes for the decline of the species in Europe during the twentieth century. Nowadays these factors seem tohaveonlyalimited influence on the populationof bearded vultures, at least in Spain (B. Heredia1991). As a consequence,the populationin the Pyrenees has increased steadily during the last two decades (R. Heredia 1991a). Nowadays,mountain habitatsarebeingtransformed with the change in human production systems. There is a reduction of areas used for extensive

grazing of livestock and an increase in tourism. Ithasbeensuggestedthatchangesintheuse of mountain areas canreduce thecarryingcapacity of theenvironment andthebreedingsuccessofthe bearded vulture (B. Heredia 1991;Terrasse 1991).

Quantitativestudiesofhabitat selection, asaway topredictspeciesrequirements,arefrequentlyused to design strategies for the conservation of nd­ angeredspecies(Morris1980;BednarzDinsmore

1981; Newton, Davis Moss 1981; Andrew Mosher1982;Peterson1986;Gonzalez,Bustamante

Hiraldo1992). Thiskindofanalysisforthebearded vulturecanbeusefultodayas therehasbeena reintroduction projectintheAlpssince1978,the releaseofbirdshavingstarted in1986.Itisexpected thatthisandother projectswill beextended toother mountain rangesin thecontinent includingSpain (FAPAS1991;AMA—CSIC 1991). Because these reintroduction projects are expensive (Patchlatko

1991)itisimportanttoevaluate thesuitabilityofan area before thebirdsarereleased.

The present study analyses nest-site selection,

densityof breedingpairsandbreedingsuccessin relation tovariables for topography,climate, food availabilityandhumandisturbanceinthepopulation of bearded vultures in the Spanish Pyrenees.The aims were two. First, to identify possible relation­ ships between variables measured and bearded vulturedistribution, densityandbreedingsuccess. Secondly,toobtainmodelsto evalulatewhethera proposed areaforreintroductionhadadequatecliffs asnestingsitesfor the species, and what might be the expected productivity and breeding density in thearea. Although itis difficulttoevaluate the confidence in the values predicted by the models once they are used inan area different from that fromwhichtheywerecalculated,thesemodelsarea stepforward whencompared toasubjective assess­ mentofsuitabilityofproposed reintroductionareas.

Studyareaandmethods

The bearded vulture populationstudied isdistrib­ uted allalong the southernslopes ofthe Pyrenees butis moredenseintheareaknownasCentral Pyrenees. Thebearded vulture populationofthe Spanish Pyrenees hasbeen studied since1977and alloccupied territoriesare known (n=52in1991) (seeR.Heredia 1991a).

DATA

Nestsiteselection

Thirteenvariablesrepresentingphysiography, land- useanddegreeofhumandisturbanceweremeasured on111cliffswith bearded vulture nests (Table 1). Thesenestsbelongedto37 differentbreeding ter­ ritories. Theexistenceorexactlocation ofthenests wasunknownintheremaining15 territories. Another

111 cliffs without nests were selected at random and used to estimate the nesting habitat available for the species. Random points were selected by pseudo-random generation of coordinates with a calculator,andchoosingthe nearest pointonacliff without a nest. Since, in the Pyrenees, most of the nests were found near the middle of the cliff (R. Heredia, unpublished),we selected a similar location (half of the cliff height) for the random points. To avoid bias due to different breeding densities, random sampling wasstratified and the numberofrandom cliffssampledoneachmapsheet where thespecieswasbreeding (‘L’series1:50000 topographicmap ofSpain, each sheet covering an area of26712km2) wasmade equal tothe number ofpreviouslysampled nestingcliffs.Variables were measured on topographic maps of the Spanish CartographicService and Land-use maps of the

Spanish Ministry of Agriculture. We also used

Table 1.Variablesusedtocharacterizebeardedvulturenestingcliffsandrandomcliffs.Forrandomlyselectedcliffs,

variables weremeasured fromapointinthecentre ofthecliff

RELIEF:topographicirregularityindex.Totalnumberof20-rncontourlines,cutbyfour1-kmlinesstarting fromthenestin directions N,S,E andW.

ALTITUDE:altitude ofthenestabove thesea-level (rn).

CLIFF:cliffheight,measured asthenumber of20-rncontours cutbya50-rnlineperpendiculartotheclifffaceatnestlevel. ORIENTATION:orientationoftheclifffaceatthelevelofthenest.Orientationswerescored inincreasingshelter fromcold humidwindsfromtheNWwhicharedominantinthearea: 1=NW,2=NorW,3=NEorSW,4=EorS,5 =SE.

FOREST: extension (%)offorested areas ina1000-mradius around thenest. DISTANCEVILLAGE:distance tothenearest inhabited village(km). INHABITANTS:numberofinhabitantsinthenearestvillage.

ICILOMETRES ROADS: kilometresofpavedandunpavedroadsina1000-mradiusofthenest.

DISTANCEPAVED ROAD:shortestlineardistancebetweenthenestandtheclosestpaved road(km).

DISTANCEROAD:shortest linear distance between thenestandtheclosestroad, pavedorunpaved (km).

HEIGHTPAVEDROAD:altitudinal difference between thenestandtheclosestpavedroad, measured atthepointtheroadis closertothenest(rn).Ifthenestislowerthantheroadanegativevalueisobtained.

HEIGHTROAD:altitudinaldifference between thenestandtheclosestroad, pavedorunpaved, measured atthepointis closertothenest(m). Ifthenestislowerthantheroadanegativevalueisobtained.

NEARESTNEIGHBOUR:linear distance between thenestand theclosestnestofthenearest neighbour (km).

the reports on human populationcensus of 1981 (INE 1984).

Breedingdensity

We used the distance ofthe most frequently used nestofabreeding pairtothenearest nestbelonging toanotherbeardedvulturepairasaninversemeasure of breedingdensityinthearea.Nearestneighbour distances are currently employed as evaluatorsof raptor breeding density (see e.g. Newton 1979). Nearestneighbourdistances,however,are not independent. Thisisparticularly clearwhenthe distance between two pairs iscounted twice, but evenwhenthisdoesnotoccur,independence is unlikelyasdensityisaglobalmeasure related tothe situation ofevery pair. Although our analysisdoes notsolvecompletelytheproblemof lackofinde­ pendence ofnearest neighbour distances, measures that wererepeatedwereconsideredonlyonce.

Wecharacterizedtheareasurrounding28bearded vulture breeding pairs in the Central Pyrenees (58% oftheSpanish population).Weselected this

area because itisknown that there the number of

breedingpairshasremained almoststablesince1977 (R. Heredia,unpublished),and this suggests that the population isinequilibrium with the environ­ mentandwe couldexpecttodetecttheecological factors thatlimitbreeding density.

Foreachpairwe quantifiedthetopography, vegetation, land-use, climate,foodavailabilityand degree of human disturbance in a 15km radius around themostfrequentlyusednest-site(707km2) (Table 2). According to Brown (1988) bearded vultures inSouth Africa forage in acircular area of 300—700km2around the nest, so we assumed that the variables measured in a circle of 15km radiuswouldbeanadequate description ofthemain foragingareaof abreedingpair.Thevaluesofthe variables wereobtained from thesame mapsasfor nest-siteselection, andalsofromtheClimaticAtlas ofSpain(InstitutoNacionaldeMeteorologIa 1983), livestockcensusfrom1986 (SpanishMinistryof Agriculture,unpublished)andChamois(Rupicapra rupicapraL.)censuscarriedby theAutonomous Communities ofNavarra, Aragon andCatalufla (unpublished).

Table 2.Variablesusedtocharacterizethemain foragingareas(acircleof15-kmradiusaroundthemostfrequently usednest)

RELIEF:topographicirregularityindex.Numberof100-rncontourscutbyfour15-kmlinesstartingfromthenestin

directionsN,S,E andW.

MAXIMUMALTITUDE: maximumaltitudeinthemainforagingarea.

MINIMUMALTITUDE: minimumaltitudein themainforagingarea.

AVERAGEALTiTUDE:averagealtitudeinthemainforagingarea=(maximumaltitude+ minimumaltitude)/2.

ALTITUDINALDIFFERENCE: maximumaltitude— minimumaltitude.

AREAOVER1600m:percentageofthemainforagingareaover1600m.

TEMpERATURE: averageannualtemperature.

SUNSHINE:averageannualnumberofhoursofsunshine.

RAINFALL: averageannualrainfall(mm).

DAYSWITH RAIN: averageannualnumberofdayswithrain.

DAYSWITHSNOW:averageannualnumberofdayswithsnow.

WINTERRAiNFALL:averagerainfallinDecernber,JanuaryandFebruary,duringthecourtshipandlayingperiodofthe beardedvulture. V

SPRING RAINFALL:averagerainfallinMarchandApril,duringtheincubationandhatchingofthebeardedvulture.

CULTIVATEDLANDS: percentageofthemainforagingareacoveredbycultivatedlands.

FORESTS: percentageofthemainforagingareacoveredbyforests.

PASTURE LANDS: percentageofthemainforagingareacoveredbypasturelands.

HIGHMOUNTAIN: percentageofthemainforagingareacoveredbyunproductivehighmountainterrain(rockyoutcrops, snowpatches,screes).

SCRUBLAND: percentageofthemainforagingareacoveredbyscrubland.

OPEN LAND: sumofpasturelands,highmountainandscrubland.

DISTANCETOCAPITAL:lineardistancefromthenesttothenearestprovincialcapital.

VILLAGES:numberofpermanentlyinhabitedvillagesinthemainforagingarea.

VILLAGES OVER1000INHABITANTs: numberofvillageswithmore than1000inhabitants in themainforagingarea.

INHABITANTS: totalnumber ofinhabitantsinthemainforagingarea.

KILOMETRESOF PAVEDROADS: kilometresof pavedroadsin themainforagingarea. KIL0METRESOFUNPAVEDROADS: kilometresof unpavedroadsin themainforagingarea. TouRIsM: numberof hotelbedsandcamping placesinthemainforagingarea.

KILOMETRESOFELECTRICPOWER LINES:kilometresof hightension electric powerlinesinthemainforagingarea.

LIVESTOCK: numberofsheep andgoatperkm2inthemainforagingarea. Weassumed thatthenumberofsheep andgoat from amunicipality inamainforagingareawas proportionalto thepercentageofthatmunicipalityinsidethemain foragingarea.

CHAMoIS:number ofchamois(Rupicaprarupicapra)perkm2inthemainforagingarea.

TOTALUNGULATES: livestock+chamois.

One appropriate link function for a binomial distributionisthelogisticfunction. Thismeansthat

Toevaluatebreedingsuccesswe usedtheaverage productivity of each breeding pair (defined as: numberof fledglingsraisedpernumberofyears monitored).Productivity values were available for

25 breeding pairs from all the Spanish Pyrenees. These pairswere monitored for more than 5years (mean 1O2 years, SD 38).Productivity values werecompared withthevariablescharacterizing the nest-site(themostfrequently usednestofeachpair) andthemainforaging area.

STATISTiCAL ANALYSiS

First we made a univariate analysis of the data. Meanvaluesfornestingcliffsandrandomcliffswere compared usingt-tests. Nearest neighbour distance ofthe pairsinthe Central Pyrenees wascompared withthe variables characterizingthe mainforaging area, and productivity values were compared with the variables characterizingthe most usednest-site andthemainforaging area.

Secondly, weusedGeneralizedLinearModels,or GLM (Nelder Wedderburn1972;Dobson 1983; McCullaghNelder1983),tomakeamathematical description of the nest-site selection, breeding densityandbreedingsuccessofthebearded vulture. GeneralizedLinear Modelsare aclassofmodels from whichthelinear regression formsaparticular case. GLM permit a wider range of relationships between theresponse andtheexplanatoryvariables and the useofother error formulationswhen the normal error for the traditional regression is not

applicable.

ThreecomponentshavetobedefinedforaGLM: alinearpredictor,anerror function andalinkfunc­ tion. Alinear predictor (LP) isdefined asthesum oftheeffectsofthepredictor variables asfollows,

LP=a+bx1+cx2+ eqn1 where a, b,c,...are parametersto be estimated

from theobserved data andx1,x2,...theexplana­ toryvariables. These parametersdefinetheeffectof thevariables ontheLP.

The error function willdepend on the nature of thedata. Forbinaryresponsevariables thebinomial distributionisanadequateerror function. We assumed a binomial distribution of errors in the models ofnest-siteselection,inwhichtheresponse variable hadthevalue1(cliff selected asanest-site) or 0 (cliff not selected asa nest-site),and in the modelsofproductivityinwhichtheresponsevariable hadthevalue1(whenachickhadbeenfledgedfrom aterritory a certainyear)or0 (whennochickhad fledged). Nearest neighbour distances seemed to belog-normally distributed, sovalueswerelog- transformedandanormaldistributionoferrors was assumedforthemodels.

theprobability ofacliffbeingselected asanestsite orofachickfledginginaterritory acertainyearisa logistic,s-shaped function whenthelinearpredictor isafirst-order polynomial orabell-shaped function forsecond-orderpolynomials. Thelogisticfunction canbeexpressed as:

p=(e’’)/(1+eLP), eqn2 wherepistheprobability ofobtaining apositivere­

sponseandeisthebaseofthenaturallogarithm.This expression canbetransformedtoalinearfunction:

ln[p!(1— p)] LP, eqn3

wherelnisthenatural logarithm. Equations1and3 define theGLMfornest-siteselectionandbreeding successofthebearded vulture.

To model the nearest neighbour distance, or

breeding density, weusedanidentity link. Inthese casesthemodelsdonotdifferfromamultiplelinear regressionwiththedependent variable(nearest neighbour distance) log-transformed.

ANALYTiCAL PROCEDURE

For nest-site selection we dividedeachpredictor variable intosixclassesandgraphically represented thenumberandpercentageofnestingcliffsinrelation tototalnumber ofcliffs ineachclass.Average productivityandnearest neighbourdistanceforeach of the 25and 28 breeding territories respectively wereplottedagainstthevaluesof eachpredictor variable.Visual inspectionofthesegraphsrevealed whichshapeofresponse couldbeexpected foreach predictorvariable(linearor curvedresponse)and whethera transformationofthepredictorvariable couldberecommended.

Wefittedeachexplanatoryvariabletotheobserved

data using the program GUM(Baker Nelder

1978)following amodification ofatraditionalfor­ wardstepwise procedure.Each variable wastested forsignificanceinturn. Thevariablecontributingto the largest significant change indeviance from the null model was then selected and fitted. Once a variable wasfitted to the model we tested if the addition ofasecondvariablesignificantlyimproved the model. As we were using a large number of variables we chose a 1% level of significance to includeavariable inamodel.

Ifthe initial bivariate graphs suggested acurved

response aquadraticfunction was tested initially, and a cubic term was then tested to ensure that a higher order polynomial was not necessary to improvethemodel.Squarerootand logarithmic transformationofallthe predictor variables involv­ ingdistanceswerealsotestedasthegraphsuggested they had more a multiplicative than an additive effect.

es

Table3.Mean(SD)ofthevariablescharacterizingnestingandrandomcliffs

Recent papers havecriticizedautomaticstepwise proceduresastheyarenotnecessarilyabletoselect the mostinfluential variable from asubset ofvari­ ables(James McCulloch1990).Ourmodification ofastepwise modelling procedure involved testing thealternativemodelsthatwereobtained whenthe secondorthethirdmost significantvariablewas included(provided itwassignificantatthe1%level) instead ofthe firstmost significant one ateach of thesteps.Thisbranchingprocedure couldeventually produce aset ofdifferent models, but inmost in­ stances itconverged into asinglemodelortoaset of models from which similar causal relationships couldbeinferred.

In addition,a residual analysis was undertaken

forthebestmodelora setofbestmodels.Three diagnosticmeasures wereusedtoevaluate thefitof the models tothe data: ameasure ofthe residual lackoffit,thepotentialinfluence,andthecoefficient of sensitivityofeachobservation(Pregibon1981; Nicholls1989).Standardizedresiduals wereplotted against fitted value for possible deviations of the initial assumptions of the model. Observations withhighpotential influence werere-examined lookingforerrorsin thedataorpossibleoutliers. Observationswith a high coefficient ofsensitivity were excluded and models refitted evaluating the effect these observations had on the parameters ofthemodels.

The robustness of the nest-site selection model

was alsotestedwithajack-knifeprocedure.Each observationwasomitted inturnandtheparameters ofthemodelswererecalculated with theremaining observations. Weobtained fromthismodelthe probability theexcludedobservationhadofbeinga nestingcliff.Percentages ofcorrect classification obtained by thejack-knifeprocedure werethen compared tothoseoftheinitialmodel.

Results

NE5T SITE SELECTiON

Anaverageofthree nestsperbreeding territory was knowninthe37territorieswithknown nestingsites intheSpanish Pyrenees (range1—6nests,SD=3). Theaveragedistancefromanesttothenearest nest in the same territory was992m (range 50—8450, SD=1443,n=104)andtothefarthest nest2696m (range 50—10150,SD=2467,n=104).

Thereweresignificantdifferencesbetweennesting cliffsandrandom cliffsinvariables related torelief andcliffheight. Neststended tobelocated inareas moreruggedandathighercliffs than theaverage available.Also, nesting cliffstended to be farther from the nearest occupied nest ofbearded vulture thanfromrandom cliffs(Table3).

Observations of nesting cliffs (1) and random cliffs (0)werefittedtoaGEMmodelassuminga binomial distributionoferrors and usingalogistic link(equivalent toalogisticregression).

The best model obtained (the one withsmaller

residualdeviance) includedthevariables: relief,dis­ tancetothenearestoccupiednest(log-transformed), altitude (quadraticfunction) and distance to the nearestvillage(log-transformed)(Table4).Itshowed that the bearded vultures selected asnesting cliffs those inareas withthe mostirregular topography, farfromother breeding pairs,atanaveragealtitude (avoiding cliffsat high or low altitudes) and not closetovillages.

Other alternative models only differed in the

order inwhich the different explanatoryvariables wereincluded andfinallyconverged withthissame model. Cliffheights (whichhadahighlysignificant correlation with topography, r=0674, df=220, P0001)were included in some of the initial modelsbuttheyhadabiggerresidualdeviance than those whichincluded reliefinstead,andcliffheight did notsignificantlyimprove the model once relief

Table4.GLMmodelfornest-siteselection, usingbinomial

errorand logisticlink. Distance tothenearest neighbour

and distance tovillagearelog-transformed

(negativecorrelation) andaltitudinal difference (negativecorrelation) inthemainforagingrange (Table 5).

JParameterestimate Standarderror

Constant —33•93 5452

RELIEF 009058 001480

NEARESTNEIGHBOUR 1644 03405

ALTITUDE 0009867 0003l74

(ALTITUDE)2 —4024x10 ll46x10

DISTANCEVILLAGE 0945l O•2917

Residual deviance 190•12

df 216

had been included. The log-transformationofdis­ tance tothe nearest occupied nest anddistance to thenearestvillagesignificantlyimproved themodels compared withtheuntransformed variables.The coefficientsofthesevariables showthat distance to avillageisonlyimportantwhencliffsareveryclose to villages (coefficient <1), but distance to the nearest occupiednestisimportantovertherangeof distancesmeasured (almostallcliffsina8-kmradius of abreeding pairareunavailableforotherpairs) indicating that nests tend to be regularly spaced inthearea.

Fitted values ofthe GLM models can be inter­ preted as the estimated probability (P) of a cliff beinganestingcliff.ThoseobservationswithP>05 were considered classified as nesting cliffs and thosewithP05asrandom cliffs.Thefinalmodel classifiedcorrectly 793%ofthe nesting cliffsand

76.6% of the random cliffs. This classification is

56%betterthanrandom(Kappa=0559,Z=8337, P<0.001). The jack-knife classification showed the robustness ofthe model; 78.4% ofthe nesting cliffsand757%oftherandom cliffswerecorrectly classified. On average, the jack-knife procedure onlymisclassified1%moreobservations thanthe complete model.

Thesensitivityanalysisofthemodeldidnotshow anyoutliers.Allobservations withhighpotential influenceinthemodelwerecheckedanddatavalues werefoundtobecorrectandreasonable.Theobser­ vationwiththehighestcoefficientofsensitivitywas omitted and parametersrefitted. Thechangeinthe parameterswaslessthan 4%.

BREEDING DENSITY

Theaveragedistancetothenearest neighbour inthe SpanishPyreneeswas110Mm(range=2125—28000, n=51). In the core area selected for thestudy of breeding density,theCentral Pyrenees, thecorre­ sponding estimate was8813m (range 2125—19500, n=28).

There were significant correlations between nearest neighbour distance and the extent ofopen areas (pastures, scrubland and high mountain)

Nearest neighbour distance (an inverse measure of breeding density) was log-transformed and a GLM model with an identity link and assuming normal error,wasfitted (equivalentto a multiple linear regression).

The stepwise branching procedure produced three similarly significant models. Each included oneofthree variablesrelated toaltitude (maximum altitude,altitudinal difference,average altitude, all significantly correlated P0.001) and number of dayswithsnowasa secondvariable(Table6).All modelsindicated thatdensityofbreeding pairs increased withaltitude whichmeansthatwithinthe rangeof altitudesatwhichnestswerefound, the distance between nestsisgreater athigheraltitudes. All themodelsshowedalsothatbreedingdensity decreased withthenumber ofdayswithsnow.

Theextensionofopenareasin theforagingrange(a variablepositivelycorrelated withaltitude, r=0825, df=26,P<0001)was the first variable to come intothemodel,butno othervariablesignificantly improved this model (although number of days

Table5.Correlationbetweennearestneighbourdistance and variablescharacteriringthemainforagingareainthe Central Pyrenees(df=26). *Correlations that remain significant(P0.05)afterBonferroni sequentialcorrection (Rice1989)

Variable r P

Table6.GLMmodelsfornearestneighbourdistance(log-transformed), usingnormalerrorandidentitylink

constant

MAXIMUMALTITUDE DAYSWiTHSNOW

Residual SS

df

F(2,25)

Constant

ALTITUDINALDIFFERENCE DAYSWITHSNOW

ResidualSS

df

F(2,25)

Constant

AVERAGEALTITUDL DAYSWITHSNOW

Residual SS

df

F(2,25)

Parameterestimate

9-963

—8-025xio—

0-03458

2-1987

9-681

—8-022xio

0-02782

2-2033

10-20

—0-00153

0-04005

2-3330

Standarderror

0-1982

1-154xiO

0-006570

25

23-67(P<0-01)

0-1703

1-156x1o4

0-005909

25

23-67(P<0-01)

0-2343

2-339xio—

0-007448

25

22-89(P<0-01)

NullModelSS=6-4580df=27.

WithSnOWWas nearly significant, 001<P005). Although thereduction indeviance byextension of open areas wasSlightlygreater,itwanot signifi­ cantlydifferent from thevariables related Withalti­ tude. Once number ofdaysWithsnOwwasincluded in themodeltheinclusionofanaltitudinal variable improved the model more than the inclusion of extension ofopen areas. Thealtitudinalvariable of themodelcouldalsobesubstitutedbyrelief.

Inspectionof theresidualsshowedthatthelog- transformation of the nearest neighbour distance, thenormalerror andtheidentitylink,werereason­ ableassumptionsforthemodel.No outlierswere detected fromthepotential influencemeasure. Removaloftheobservation withthehighestcoef­ ficientof sensitivityproduced a 12%changeinthe parameters,whichseemed reasonable giventhe number ofobservations(n=28).

BREEDING SUCCESS

Ofthevariablescharacterizing thenestingcliff, productivity wassignificantly correlatedonly with the relief (Table 7). Pairs nesting incliffsinmore rugged terrain hadhigherproductivity.Ofthevari­ ablescharacterizingthemainforagingarea, number ofinhabitantsand kilometres ofpaved roads were significantlynegativelycorrelated withproductivity whiletheextent ofopen areas (pastures,scrubland andhighmountain) was significantlypositively correlated.

Weconsidered thenumber ofchicksfledgedina

territory (a maximum ofone chick isfledged per year)asavariable withabinomialdistribution,and use it as binomial denominatorfor the models in whichbreeding successofthe territory wasknown foranumber ofyears.

Thebestmodelincludedonly thevariable‘kilo­ metres of paved roads’ inthe main foraging area, andshowedthatbreedingsuccesswaslowerinareas withahigh densityofpavedroads(Table8).The residual deviance ofthe model wasstillquite large butnoother variablesignificantlyimproved it.

Other variables (extent ofopen land, number of

inhabitants,reliefofthe mainforaging area, relief ofthenestingcliff,maximumaltitudein themain foragingarea, andnestingcliff height)alsosignifi­ cantlydecreased thedevianceinrelation tothenull model. Allthese variables weresignificantlycorre­ lated with kilometers of paved roads, but models withthesevariablesstillhadasignificant(P<001) ornearlysignificant(P<0.05)decrease indeviance whenkilometres ofpavedroadswasincluded.

The sensitivity analysis pointed out some ter­

ritorieswith highpotentialinfluence.Thesecor­ responded toterritorieswithvalus closetothe maximumandminimumofthevariables(numberof chicksfledged,andkilometres ofpavedroads), and withahighbinomialdenominator(number ofyears thebreedingsuccesswas known).Therewereno reasons toconsider them asoutliers. Omitting the pairwiththehighestcoefficientofsensitivityinthe model produced a10% change inthecoefficientof kilometres ofpaved roads.

Discussion

NEST SITE SELECTION

The results show that the bearded vulture has a strong selectionforcertain nestingcliffsfrom those available. The78%correct classificationofcliffsby theGLMmodelcanbeconsidered good,giventhat

Table 7. Correlation between productivity (Average numberoffledglingsperyear)andvariablescharacterizing nesting cliff and the main foraging area (df=23).

*Correlations that remain significant (P005) after

Bonferronisequential correction (Rice1989)

distance to villages are, in this order, the main factors conditioning the selection of a cliff as a nesting site. Relief can be considered the main factor, as it alone classifiescorrectly 69% of the

cliffs.There arethreereasonswhybearded vultures

Nestingcliff RELIEF ALTITUDE CLIFF ORIENTATION

FOREST

DISTANCEVILLAGE INHABITANTS KILOMETRES ROAOS DISTANCEPAVEDROAD DISTANCEROAD