Journal ofthe Bombay Natural HistorySociety, 105(1),Jan-Apr2008 49-54 GAPANALYSIS OF INDIAN FOX CONSERVATION USING ECOLOGICALNICHE MODELLING AbiTamim Vanak1 Mohammed Irfan-Ullah2andA.TownsendPeterson3 , 'WildlifeConservationSociety-lndiaProgram,GraduateResearchAssistant,DepartmentofFisheriesandWildlifeSciences, UniversityofMissouri,Columbia,MO65211,USA, Email: [email protected] 2AshokaTrustforResearchinEcologyandtheEnvironment(ATREE),Bengaluru 560024, Karnataka,India. PresentAddress: RMSIPvtLtd,A7Sec 16.Noida201 301,UttarPradesh,India. Email: [email protected] NaturalHistoryMuseumandBiodiversityResearchCenter,UniversityofKansas,Lawrence,Kansas66045,USA. Email: [email protected] We used ecological niche modelling to predict the geographic distribution ofthe Indian Fox, acanid endemic to the Indian subcontinent. This little known canid, while not yet endangered, is threatened due to rapid habitat loss and poaching throughout its range. We analysed 58 known occurrence locations from survey data collected from three states in peninsular India using the software Desktop GARP. We created an ecological niche model for India using vegetation and topographic data and furtherrefined it by including 18 additional bioclimatic data sets. Basedon the ecologicalnichemodellingresults,agapanalysisofprotectionofferedtopotentialFoxhabitatintwostatesofsouthern India was conducted by overlaying existing protected areaboundaries on the refined distribution and calculating the extent ofprotection. Our analysis showed that the Indian Fox habitat consists primarily of low elevation semi-arid grassland, scrubandthornforests,whichrankamongthemostvulnerable in Indiaowingtoconversiontoagriculture, industryandurbanareas.Thegapanalysisshowedthatalittleover 1%ofpredictedFoxdistributioniscoveredbythe protectedareanetwork.Theunderrepresentationofthesehabitatsisdeleterious notonly totheIndianFoxbutalsoto arange ofotherspecies, such as the endangeredGreat Indian Bustard, Indian Grey Wolfand Blackbuck. Key words: Indian Fox, Vulpesbengalensis, ecological niche modelling, gap analysis, distribution, protected areas INTRODUCTION terrain,tallgrasslandsandtruedesert(Prater 1980;Johnsingh and Jhala 2004). Recent surveys indicate that though the The Indian Fox Vulpes bengalensis, a canid endemic species is widespread, it is not common through most ofits to the Indian subcontinent, is widespread, ranging from the rangeandisencounteredathighestfrequencies inprotected foothills ofthe Himalaya in the north to the southern tip of semi-arid shortgrasslands anddry scrub areas(Vanak 2003, the Indian peninsula, and from Sindh Province of Pakistan 2005; Vanak and Gompper 2007). Although these habitats easttoBangladesh(JohnsinghandJhala2004;Gompperand rank among the most endangered in India (Rahmani 1989) Vanak 2006). Even though this species is believed to be and are subject to constant human encroachment, they are common (Johnsingh and Jhala 2004), little is known about rarely the focus ofconservation attention. Moreover, these its ecology or the details of its geographic distribution, or habitats have been categorised as wastelands by various land populationstatus.TheIUCNCanidSpecialistGroupclasses management agencies and are subject to intense pressure to thisspeciesas ‘LeastConcern,’ (JohnsinghandJhala2004); betransformedintoagriculturalandpastorallandscapes(http:/ itislistedunderScheduleIIoftheIndianWildlife(Protection) /dolr.nic.in/wasteland.htm accessedon January 28, 2008). Act,1972 (as amended up to 2002), which prohibits hunting In the southern Indian states, Indian Fox habitats are ofthis species (Anonymous 2002). Despite this protection, decreasing, and the continued survival of the species is Indian Fox populations are declining owing to habitat loss seriously threatened (Johnsingh and Jhala 2004). Despite a from conversion to intensive agriculture, industry and reported presence in some protected areas (PAs), most developmentprojects (Johnsingh and Jhala2004). populations of this species remain outside the PA network TheIndianFoxisfoundinsemi-arid,flatorundulating (Vanak2005). Inthe absence ofsystematicproactiveefforts terrain in biogeographic zones 3, 4 and 6 ofIndia (Rodgers for its conservation, this species might suffer a substantial etal. 2000), which are typically drier biomes characterised reduction in potential habitat, affecting its future survival. bylowrainfall,scrub,thorn,ordrydeciduousforestsorshort Wethusdevelopedagapanalysisoftheprotectioncurrently grasslands (Manakadan and Rahmani 2000; Rodgers et al. affordedtoFoxdistributionareasinthesouthernIndianstates 2000; Vanak 2005). Indian Foxes avoid dense forest, steep ofKarnatakaandAndhra Pradesh. GAPANALYSIS OF INDIAN FOX CONSERVATION Gapanalysisisaproactiveapproachtoplanningtimely overamongrules)tomaximisepredictability.Thepredictive action for species conservation that focuses on evaluating accuracy is then evaluated based on 1,250 pointsresampled the degree to which native species are represented in PAs. with replacement from the intrinsic testing data and SpeciesnotadequatelyrepresentedintheexistingPAnetwork 1,250 points sampled randomly from the study region as a constitute ‘gaps’ in theconservationprogram. Gapanalysisis whole to represent pseudo-absences. GARP is designed to intendedtopreventadditionalspeciesfrombecomingthreatened work based on presence-only data; missing information is orendangered,andinthissenseisproactive,ratherthanreactive included in the modelling via sampling of pseudo-absence (Scott etal. 1993; Flatheretal. 1997; Davisetal. 1998). points from the set ofpixels where the species has not been Mapping the distribution of a species exhaustively detected (Stockwell and Peters 1999). The change in throughon-groundsurveyswouldbeprohibitivelyexpensive predictive accuracy from one iteration to the next is used to andtime-consuming,ifnotsimplyimpossible.Thealternative evaluatewhetheraparticularruleshouldbeincorporatedinto used here is that occurrence data available from regions the model, and the algorithm runs either 1,000 iterations or sampledindetailcanbeusedtoreconstructthespecies’overall until convergence. distributions using Ecological Niche Modelling (ENM) Weused58uniquepointoccurrencesoftheIndianFox (Peterson and Kluza 2003; Peterson 2005). The ecological sampled from the states of Karnataka, Andhra Pradesh and niche of a species can be defined as the set of ecological Maharashtra, in 2003 and 2005 (Vanak 2003, 2005) (Fig. 1) conditions within which it is able to maintain populations foranalysis.Weused ‘monthly’compositesofthemaximum withoutimmigration(Grinnell 1917;HoltandGaines 1992). NormalisedDifferenceVegetationIndex(NDVI)imagesfrom Several approaches have been used to approximate species’ the Advanced Very High Resolution Radiometer (AVHRR) ecologicalniches(Austinetal. 1990;WalkerandCocks 1991; satellite (Eidenshink and Faundeen 1994) for 2003, as well Scott et al. 1996; Scott et al. 2002); of these, one that has aselevation,slope,aspectandCompoundTopographicIndex seenconsiderable testing is the GeneticAlgorithm forRule- (CTI)fromtheHydro-IKdataset(USGS 2001)andaglobal set Prediction (GARP), which includes several inferential landcover coverage (Hansen et al. 1998, 2000). All approaches in an iterative, evolutionary computing environmental datasets were resampled to pixels of about environment (Stockwell and Peters 1999). All modelling in 1 km x 1 km foranalysis. this study was carried out on a desktop implementation of the GARP algorithm (Stockwell and Noble 1992; http:// Table 1: Bioclimaticvariablesusedforrefiningthedistribution ofthe Indian Foxinthetwosouthernstates www.lifemapper.org/desktopgarp). Sr. No. PredictorVariables Source MATERIALAND METHODS 1. Meanmonthlyprecipitation WorldClim 2. Meanmonthlytemperature WorldClim 3. Maximummonthlytemperature WorldClim Ecological Niche Modelling (ENM) has been used in 4. Minimummonthlytemperature WorldClim numerous applications and subjected to various tests, based 5. Annualmeantemperature WorldClim on diverse analytical approaches (Miller 1994; Csuti 1996; 6. Meandiurnal range WorldClim (meanofmonthlymaxtemp-mintemp) Tuckeretal. 1997;Gottfriedetal. 1999;Maneletal. 1999a,b). 7. Isothermality WorldClim The particular approach to modelling species’ ecological (Febprecipitation/Julyprecipitation)* 100 niches and predicting geographic distributions used here 8. Temperatureseasonality WorldClim (standarddeviation*100) (summarised below) is described in detail elsewhere 9. Maxtemperatureofwarmestmonth WorldClim (Stockwell and Peters 1999; Peterson etal. 2002). Previous 10. Mintemperatureofcoldestmonth WorldClim tests ofthe predictive powerofthis modellingtechnique for 11. Temperatureannualrange WorldClim 12. Meantemperatureofwettestquarter WorldClim diverse phenomena in various regions have been recorded 13. Meantemperatureofdriestquarter WorldClim elsewhere (Peterson 2001; Peterson etal. 1999; Peterson et 14. Meantemperatureofwarmestquarter WorldClim al. 2002; Peterson andVieglais 2001;Anderson etal. 2002, 15. Meantemperatureofcoldestquarter WorldClim 16. Annualprecipitation WorldClim 2003; Stockwell and Peterson 2002). 17. Precipitationofwettestmonth WorldClim GARP works in an iterative process ofrule selection, 18. Precipitationofdriestmonth WorldClim evaluation, testing and incorporation or rejection: first, a 19. Precipitationseasonality WorldClim (coefficientofvariation) method is chosen from a set of possibilities (e.g. logistic 20. Precipitationofwettestquarter WorldClim regression, bioclimatic rules), and is then applied to the 21. Precipitationofdriestquarter WorldClim trainingdata, and arule is developed; rules mayevolveby a 22. Precipitationofwarmestquarter WorldClim 23. Precipitationofcoldestquarter WorldClim number of means (e.g. truncation, point changes, crossing- 50 J. Bombay Nat. Hist. Soc., 105 (1), Jan-Apr 2008 GAPANALYSIS OF INDIAN FOX CONSERVATION estimates, following recent recommendations (Anderson etal. 2003). First, 20 models with 0% omission errors were chosen,andofthese 10modelswithinthecentral50%ofthe commission values were selected as the best models. To provide an independent validation of model performance, we randomly created four independent replicates of40 point locations each from the original data set(n=58).Theremaining 18pointsofeachofthesereplicates were set aside for an independent test of the predictive accuracyforeachreplicate.Coincidencebetweenindependent testing points and model predictions for each replicate was usedasameasureofmodelpredictiveability. Binomialtests were used to compare the observed predictive success with that expected underrandom (null) models ofno association between predictions and test points. The test results are in the formofa ‘ramp’ ofmodel agreementfrom0 (all models predict absence of 18 validation points) to 10 (all models predictpresenceof18validationpoints).Therefore,foreach replicate, wecalculatedbinomial probabilities ateachofthe Fig. 1: PredicteddistributionofIndian Fox Vulpesbengalensisin 10 predictive levels (Anderson etal. 2003). Indiabasedon NDVIandtopographicvariables To characterise modelled distributions further, based Note:The Modelover-predictsdistribution in someareas, ontheexistingknowledgeofthespecies’ habitatpreferences, suchasnorth east Indiaand Kashmir, wherethe we overlaid the predicted Fox distribution on a global speciesisnot reportedtooccur landcover data set (Hansen et al. 1998, 2000) for all of An arbitrary setof 100 model runs was developed for mainlandIndia andcalculatedproportionsoflandcovertypes each analysis. In general, 25% (c. 14 points) of available within the predicted distribution. occurrences were used for rule development (training data) Since the potential distribution of the Indian Fox is and 25% (c. 14 points) for rule selection and refinement limited in southern India and is decreasing, given growing (intrinsictestingdata), andtheremaining50% (c. 29points) urbanisation,changeinlandusepatternsandhuman-induced point locations were set aside for an independent test and disturbance (JohnsinghandJhala2004),wedevelopedagap filterofbestqualitymodels(extrinsictestingdata).Tochoose analysisforthespecieswithrespecttothePAnetworkwithin the best models from among the 100 replicates, we filtered thestatesofAndhraPradeshandKarnataka.WeobtainedPA models on the basis of omission and commission error boundaries for the two states and updated/corrected them usingtopographic maps.Werepeatedthe modelling process by restricting ourselves only to the geographic limits of Table2: PredicteddistributionofIndian Fox Vulpesbengalensis acrosslandcovertypesin India Karnataka and Andhra Pradesh to limit overprediction. We additionallyincorporated 19 ‘bioclimatic’ variables(Hijmans Landcover Location Landcover Foxdistributional et al. 2004) in the analysis to improve the algorithm’s type equivalent typeswithin rangeacross Foxdistributional eachland resolving power and to obtained a refined estimate of the range(%) covertype% Fox distribution within the two states. We then overlaid existingPAboundariestodeterminegapsintheprotectionof Water Water 0.2 2 Woodland Deciduous 0.3 1 Indian Fox habitat. The reason we have done this for only forest these two states is that this is where the majority ofthe data Wooded Open scrub 50.0 33 comes from, and this allows us to better represent Fox gCrlaossseldand fTorroepsitcalthorn 23.6 62 distribution withinasmallergeographicarea.Webelievethe shrubland Forest coarser analysis at a larger scale allows us to delimit the Open Grassland 23.7 29 broader distribution of the species at a countrywide level, shrubland/ grassland whiletherefinedanalysisallowsustoovercometheinherent Cropland Cropland 2.1 3 over-prediction ofGARPdistribution forgapanalysis atthe Bare-ground Bare-ground 0.2 1 state level. J. Bombay Nat. Hist. Soc., 105 (1), Jan-Apr 2008 51 GAPANALYSIS OF INDIAN FOX CONSERVATION Fig. 2: PredicteddistributionofIndian Fox Vulpesbengatensisin Indiabasedon NDVI andtopographicvariables RESULTS ofthe species was furtherrefined to4,70,951 sq. km (9%) of the total areaofthe states (Table 3, Fig. 2). We used model stacking to combine the results ofthe Overlayingtheprotectedareanetworkofthetwostates 10 best models for predicting the distribution of the Indian ontherefineddistributionmaprevealedthatonly sevenPAs Fox. Independentvalidationofthepredictivesuccessofthese Table3:AreastatisticsofIndian Fox Vulpesbengatensis modelsrangedfrom 11 to 18correctlypredictedoccurrences distribution inthestatesofKarnatakaandAndhra Pradesh, India out of the 18 testing points for each of the four replicates. Binomial probabilities at each ofthe 10 predictive levels in Area % of % of %of (sq. km) refined generic state all cases (4 replicate tests x 10 predictive levels eac«h) were distribution distribution geographic significantlybetterthanrandom(binomialtests,allP 0.05). area This success in predicting independent tests of occurrence Karnataka (Geographicarea-1,92,493sq. km) data gave confidence in the model accuracy; as aresult, we used all available points to develop final models (Fig. 1). Totalgeneric 1,11,403 58 Furtherexplorationofthesefinalmodelsillustratedthe distribution Total refined 21,324 19 11 species’potentialdistributioninecological dimensions.The distribution Indian Fox occupies an elevation range of 100-900 m with Totalarea (2 PAs) 92.13 0.43 0.08 0.05 low rainfall (500-1,000 mm) and moderate annual mean protected temperatures (25-30 °C). Peak NDVI values of the post- Andhra Pradesh(Geographicarea-2,78,458sq. km) monsoonseason(0.5-0.8)correspondcloselytoareasholding wooded grasslands, scrub and thorn forest systems. Cross- Totalgeneric 1,04,270 37 distribution tabulating the predicted distribution with landcover data Totalrefined 21,833 21 8 showed that ‘open scrub forest' (50%) and ‘grassland and distribution tropicalthornforest’ (47%)werethedominantrepresentative Totalarea(5 PAs) 495.51 2.27 0.48 0.18 protected landcover types within the species’ distributional area Combinedtotal 587.64 1.36 0.27 0.12 (Table 2). Focusing within Andhra Pradesh and Karnataka areaprotectedfor andaddinginthebioclimaticvariables,thepredicteddistribution bothstates 52 J. Bombay Nat. Hist. Soc., 105 (1), Jan-Apr 2008 1 . GAPANALYSIS OF INDIAN FOX CONSERVATION coincided with some part ofthe predicted range, protecting These habitats also rank among the least represented in the approximately 588 sq. km (c. 1%) of the species’ potential IndianPAnetwork(Rodgersetal.2000).Adeardemonstration distribution. of this under-representation is that only a little over 1% of the species’ distribution potential in the two states is within DISCUSSION the PAsystem. ItisthereforeclearthatkeyhabitatsfortheIndianFox The predictedrange ofthe Indian Fox developed here are inadequately represented in the PAnetwork ofthe states agreeswellwithcurrentknowledgeofthespecies’ distribution of Karnataka and Andhra Pradesh. A greater representation (Johnsingh and Jhala 2004). It excludes regions such as the ofdry-landbiomesinthePAnetworkwouldbepositive, not Himalaya, thedesertsofRajasthan andthehill rangesofthe only for the Indian Fox but also for other obligate dry WesternandEasternGhats,fromwherethespecieshasnever grassland species such as the endangered Indian Bustard been reported (Gompper and Vanak 2006) (Fig. 1). Despite Ardeotis nigriceps, Indian Grey Wolf Canis lupuspallipes, the geographically limited sampling (limited areas in three andBlackbuckAntilopecervicapra.Wesuggestthatthiskind states from peninsular India), the model performed well in ofpredictivedistributionalmodellingbeusedbyconservation capturing the species’ ecological niche, as well as its plannerstoidentifycrucialhabitatsfortheprotectionofthese geographicdistribution,acrossamuchbroaderregion.Studies endangeredspecies. elsewherehavedemonstratedasimilarpredictiveperformance ofthe GARP algorithm based on small numbers oftraining ACKNOWLEDGEMENTS locations(Peterson2001;Andersonetal. 2003; Petersonand Kluza 2003). The first author was supported by a Wildlife Ouroriginalmodels(NDVIandtopography predicted Conservation Society-India Program Small Grant, and ) about46%areaofKarnatakaandAndhraPradeshassuitable logistic support was provided by the Centre for Wildlife (Table 3) while theanalysis using anadditional 19 variables Studies, Bangalore. The second author acknowledges the permittedustorefinedistributionalestimatesforthisspecies, support of the Ministry of Environment and Forests, reducing it to 20% of the total extent of the states as the GovernmentofIndia,andtheAndhraPradeshandKarnataka potentially suitable range. This analysis confirmed that low state forest departments for permits. The Ford Foundation elevation grasslands, open scrub forest and tropical thorn and the Indo-US Science and Technology Forum provided forest constitute the bulk ofthe distribution ofthis species. support. Anderson,R.R,D.Lew&A.T.Peterson(2003): Evaluatingpredictive Flather, C.H., K.R. Wilson, D.J. Dean & W.C. McComb (1997): models ofspecies distributions: criteria for selecting optimal Identifying gaps in conservation networks: Of indicators and models. EcologicalModelling 162: 211-232. uncertainty in geographic-based analyses. Ecological Anderson, R.P.,A.T. Peterson & M. Gomez-Laverde(2002): Using Applications 7: 531-542. niche-based GIS modeling to test geographic predictions of Gompper, M. &A.T. Vanak (2006): Vulpes bengalensis. Mammalian competitive exclusion and competitive release in South species 795: 1-5. Americanpocket mice. Oikos 98: 3-16. Gottfried, M., H. Pauli, K. Reiter & G. Grabherr (1999): A fine- Anonymous (2002): Indian Wildlife (Protection)Act(1972 amended scaled predictive model for changes in species distribution 1991): GovernmentofIndia. patterns ofhigh mountain plants induced by climate warming. Austin,M.P.,A.O. Nicholls&C.R. Margules (1990): Measurement DiversityandDistributions5: 241-251 of the realized qualitative niche: Environmental niches Grinnell, J. (1917): Field tests of theories concerning distributional of five Eucalyptus species. Ecological Monographs 60: control.AmericanNaturalist51: 115-128. 161-177. Hansen, M., R. DeFries, J.R.G. Townsend & R. Sohlberg (1998): Csuti,B.( 996):MappinganimaldistributionareasforGapAnalysis. I km land cover classification deuved from AVHRR. Version Pp. 135-145. In: Scott, J.M., T.H. Tear & F.W. Davis (Eds): 1.0.In:TheGlobalLandCoverFacility.UniversityofMaryland., GapAnalysis:Alandscapeapproachtolandmanagementissues. College ParkMaryland. American Society of Photogrammetry and Remote Sensing, Hansen,M.,R.DeFries,J.R.GTownsend&R.Sohlberg(2000):Global Bethesda, Maryland. landcoverclassificationat 1 km resolutionusingadecisiontree Davis,F.W.,D.M.Stoms,A.D. Hollander,K.A.Thomas,PA.Stine, classifier.InternationalJournalofRemoteSensing21: 133L1365. D. Odion, M.l. Borchert, J.H. Thorne, M.V. Gray, Hijmans,R.J.,S.E.Cameron,J.L.Parra,P.GJones&A.Jarvis(2004): R.E.Walker, K.War—ner & J. Graae (1998): The California TheWorldCliminterpolatedglobalterrestrialclimatesurfaces. GapAnalysisProject Final Report.UniversityofCalifornia. Version 1.3. In: (http://biogeo.berkeley.edu/). Eidenshink, J.C. & J.L. Faundeen (1994): The 1 km AVHRR global Holt, R.D. & M.S. Gaines (1992): Analysis of adaptation in land data set: First stages in implementation. International heterogeneous landscapes: Implications for the evolution of JournalofRemoteSensing 15: 3443-3462. fundamental niches. EvolutionaryEcology 6: 433-447. J. Bombay Nat. Hist. Soc., 105 (1), Jan-Apr 2008 53 GAPANALYSIS OF INDIAN FOX CONSERVATION Johnsingh,A.J.T.&Y.V.Jhala(2004):IndianFoxVitlpesbengalensis. InstituteofIndia. Pp. 219-222. In: Sillero-Zubiri, C., M. Hoffman & Scott,J.M., F. Davis, B. Csuti, R. Noss, B. Butterfield, C. Groves, M.W. Macdonald (Eds): Canids: Foxes, Wolves, Jackals and H. Anderson, S. Caicco, F. D’Erchia, T.C. Edwards Jr., Dogs. IUCN/SSC Canid Specialist Group, IUCN, Gland, J. Ullman & R.G. Wright (1993): Gap analysis: ageographic SwitzerlandandCambridge, UK. approach to protection of biological diversity. Wildlife Manakadan, R. & A.R. Rahmani (2000): Population andecology of Monographs 123: 1-41. the Indian Fox Vulpes bengalensis at Rollapadu Wildlife Scott,J.M.,P.J.Heglund,J.B.Haufler,M. Morrison,M.G.Raphael, Sanctuary, Andhra Pradesh, India. J. Bombay Nat. Hist. Soc. W.B. Wall & F.-Samson (Eds.) (2002): Predicting species 97: 3-14. occurrences: Issues of accuracy and scale. Island Press, Manel,S.,J.M.Dias&S.J.Ormerod(1999a):Comparingdiscriminant Washington DC. analysis,neuralnetworksandlogisticregressionforpredicting Scott, J.M., T.H. Tear & F.W. Davis (Eds.) (1996): Gap Analysis: speciesdistributions:acasestudywithaHimalayanriverbird. A Landscape Approach to Biodiversity Planning. American EcologicalModelling 120: 337-347. Society for Photogrammetry and Remote Sensing, Bethesda, Manel, S., J.M. Dias, S.T. Buckton & S.J. Ormerod (1999b): Maryland. Alternative methods for predicting species distribution: an Stockwell,D.R.B.&I.R.Noble(1992):Inductionofsetsofrulesfrom illustration with Himalayan river birds. Journal ofApplied animal distribution data: a robust and informative method of Ecology36: 734-747. data analysis. Mathematical and Computer Simulation 33: Miller,R.I. (Ed)(1994): Mappingthe DiversityofNature.Chapman 385-390. & Hall, Fondon. Stockwell,D.R.B.& D.Peters(1999):TheGARPmodellingsystem: Peterson, A.T. (2001): Predicting species geographic distributions problems and solutions to automated spatial prediction. basedon ecological niche modelling. Condor 103: 599-605. InternationalJournalofGeographicalInformationScience13: Peterson, A.T. (2005): Kansas Gap Analysis: the importance of 143-158. validating distributional models before using them. Stockwell, D.R.B. &A.T. Peterson(2002):Effectsofsamplesizeon Southwestern Naturalist50: 230-236. accuracy ofspecies distribution models. EcologicalModelling Peterson,A.T. & D.A. Kluza (2003): New distributional modelling 148: 1-13. approaches forgap analysis.Animal Conservation 6: 47-54. Tucker, K., S.P Rushton, R.A. Sanderson, E.B. Marti—n & Peterson, A T., J. Soberon & V. Sanchez-Cordero (1999): J. Blaiklock (1997): Modelling bird distributions a Conservatismofecologicalnichesinevolutionarytime.Science combined GIS and Bayesian rule-based approach. Landscape 285: 1265-1267. Ecology 12: 77-93. Peterson,A.T.,D.R.B.Stockwell&D.A.Kluza(2002):Distributional USGS (2001): HYDRO!k elevation derivative database. U.S. prediction based on ecological niche modelling of primary Geological Survey, http://edcdaac.usgs.gov/gtopo30/hydro/, occurrence data. Pp. 617-623. In: Scott, J.M., WashingtonDC. P.J. Heglund & M.L. Morrison (Eds): Predicting species Vanak, A.T. (2003): Preliminary survey of Indian Fox Vulpes occurrences: Issues of scale and accuracy. Island Press, bengalensisdistributioninsouthernIndia.WildlifeConservation WashingtonDC. Society. Peterson,A.T. & D.A.Vieglais (2001): Predictingspeciesinvasions Vanak,A.T. (2005): Distribution and statusofthe Indian Fox Vulpes usingecological niche modelling. BioScience 51: 363-371. bengalensisinsouthern India. CanidNews8. Prater,S.H. (1980):The BookofIndianMammals. Bombay Natural Vanak, A.T. & M.E. Gompper (2007): Effectiveness ofnon-invasive History Society, Mumbai, India, pp. 316. techniquesforsurveyingtheactivityandhabitatuseoftheIndian Rahmani,A.R.(1989):TheGreatIndianBustard(FinalReport).Bombay Fox in southern India. WildlifeBiology 13: 219-224. Natural History Society. Walker,PA.&K.D.Cocks(1991):Habitat:Aprocedureformodelling Rodgers,W.A.,H.S.Panwar&V.B.Mathur(2000):Wildlifeprotected adisjointenvironmentalenvelopeforaplantoranimalspecies. area network in India: areview (executive summary). Wildlife GlobalEcologyandBiogeographyLetters 1: 108-118. 54 J. Bombay Nat. Hist. Soc., 105 (1), Jan-Apr 2008