7
Population Ecology Leopard Density in Post-Conflict Landscape, Cambodia: Evidence From Spatially Explicit Capture–Recapture THOMAS NEILL EDWARD GRAY, 1 World Wide Fund for Nature Greater Mekong Cambodia Country Program, House #54, Street 352, Boeung Keng Kang I, Phnom Penh, Cambodia SOVANNA PRUM, World Wide Fund for Nature Greater Mekong Cambodia Country Program, House #54, Street 352, Boeung Keng Kang I, Phnom Penh, Cambodia and Mondulkiri Forestry Administration Cantonment, Forestry Administration, Ministry of Agriculture Forestry and Fisheries, Phnom Penh, Cambodia ABSTRACT Effective conservation of large carnivores requires reliable estimates of population density, often obtained through capture–recapture analysis, in order to prioritize investments and assess conservation intervention effectiveness. Recent statistical advances and development of user-friendly software for spatially explicit capture–recapture (SECR) circumvent the difficulties in estimating effective survey area, and hence density, from capture–recapture data. We conducted a camera-trapping study on leopards (Panthera pardus) in Mondulkiri Protected Forest, Cambodia. We compared density estimates using SECR with those obtained from conventional approaches in which the effective survey area is estimated using a boundary strip width based on observed animal movements. Density estimates from Chao heterogeneity models (3.8 SE 1.9 individuals/100 km 2 ) and Pledger heterogeneity models and models accounting for gender- specific capture and recapture rates (model-averaged density 3.9 SE 2.9 individuals/100 km 2 ) were similar to those from SECR in program DENSITY (3.6 SE 1.0/100 km 2 ) but higher than estimates from Jack-knife heterogeneity models (2.9 SE 0.9 individuals/100 km 2 ). Capture probabilities differed between male and female leopards probably resulting from differences in the use of human-made trails between sexes. Given that there are a number of biologically plausible reasons to expect gender-specific variation in capture probabilities of large carnivores, we recommend exploratory analysis of data using models in which gender can be included as a covariate affecting capture probabilities particularly given the demographic importance of breeding females for population recovery of threatened carnivores. ß 2011 The Wildlife Society. KEY WORDS Cambodia, camera-trap, capture–recapture, density estimation, leopard, population recovery. Armed conflict and political instability are major threats to global biodiversity (Hanson et al. 2009). Negative effects of conflict upon biodiversity include increased hunting and natural resource extraction, lack of investment in conserva- tion, and displacement of people (Hart et al. 1997, McNeely 2003, Loucks et al. 2008). All of these factors affected large areas of eastern and northern Cambodia throughout the majority of the latter half of the 20th century (Chandler 2000). During this time large declines in the population and distribution of large mammal species including tiger (Panthera tigris), leopard (P. pardus), Asian elephant (Elephas maximus), banteng (Bos javanicus), gaur (B. gaurus), and Eld’s deer (Cervus eldii) have been docu- mented (Loucks et al. 2008). These declines were associated with a proliferation of firearms, the development of an external market for wildlife products, and, particularly during the Khmer Rouge-era, government-sponsored hunting (Loucks et al. 2008). Densities of large cats were particularly depressed due to both direct mortality from hunting and the collapse of prey populations. Effective conservation and planning for the recovery of remnant populations of threatened large carnivores requires reliable estimates of population density in order to prioritize conservation investments and assess the effects of conserva- tion interventions. Capture–recapture estimates of survey plot abundance, based on non-invasive sampling obtained through genetic analysis and camera-traps, are widely used to estimate densities of cats and other large carnivores (Karanth and Nichols 1998, Boulanger et al. 2002, Harihar et al. 2008, Mondol et al. 2009). These methodologies are key compo- nents of programmatic monitoring for tiger and other large carnivores and are regarded as best-practice techniques (Nichols and Karanth 2002, Karanth and Nichols 2010). However, conventional capture–recapture techniques esti- mate abundance as opposed to density. Therefore with re- spect to sampling area (e.g., a camera-trap grid) survey plot abundance is likely to be overestimated as animals with only part of their home range within the sampling area are Received: 29 June 2010; Accepted: 7 April 2011; Published: 16 September 2011 1 E-mail: [email protected] The Journal of Wildlife Management 76(1):163–169; 2012; DOI: 10.1002/jwmg.230 Gray and Prum Leopard Density in Cambodia 163

Leopard density in post-conflict landscape, Cambodia: Evidence from spatially explicit capture–recapture

Embed Size (px)

Citation preview

Page 1: Leopard density in post-conflict landscape, Cambodia: Evidence from spatially explicit capture–recapture

Population Ecology

Leopard Density in Post-Conflict Landscape,Cambodia: Evidence From Spatially ExplicitCapture–Recapture

THOMAS NEILL EDWARD GRAY,1 World Wide Fund for Nature Greater Mekong Cambodia Country Program, House #54, Street 352,Boeung Keng Kang I, Phnom Penh, Cambodia

SOVANNA PRUM, World Wide Fund for Nature Greater Mekong Cambodia Country Program, House #54, Street 352, Boeung Keng Kang I,Phnom Penh, Cambodia and Mondulkiri Forestry Administration Cantonment, Forestry Administration,Ministry of Agriculture Forestry and Fisheries, Phnom Penh, Cambodia

ABSTRACT Effective conservation of large carnivores requires reliable estimates of population density,often obtained through capture–recapture analysis, in order to prioritize investments and assess conservationintervention effectiveness. Recent statistical advances and development of user-friendly software for spatiallyexplicit capture–recapture (SECR) circumvent the difficulties in estimating effective survey area, and hencedensity, from capture–recapture data. We conducted a camera-trapping study on leopards (Panthera pardus)in Mondulkiri Protected Forest, Cambodia. We compared density estimates using SECR with thoseobtained from conventional approaches in which the effective survey area is estimated using a boundarystrip width based on observed animal movements. Density estimates from Chao heterogeneity models(3.8 � SE 1.9 individuals/100 km2) and Pledger heterogeneity models and models accounting for gender-specific capture and recapture rates (model-averaged density 3.9 � SE 2.9 individuals/100 km2) weresimilar to those from SECR in program DENSITY (3.6 � SE 1.0/100 km2) but higher than estimatesfrom Jack-knife heterogeneity models (2.9 � SE 0.9 individuals/100 km2). Capture probabilities differedbetween male and female leopards probably resulting from differences in the use of human-made trailsbetween sexes. Given that there are a number of biologically plausible reasons to expect gender-specificvariation in capture probabilities of large carnivores, we recommend exploratory analysis of data using modelsin which gender can be included as a covariate affecting capture probabilities particularly given thedemographic importance of breeding females for population recovery of threatened carnivores. � 2011The Wildlife Society.

KEY WORDS Cambodia, camera-trap, capture–recapture, density estimation, leopard, population recovery.

Armed conflict and political instability are major threats toglobal biodiversity (Hanson et al. 2009). Negative effects ofconflict upon biodiversity include increased hunting andnatural resource extraction, lack of investment in conserva-tion, and displacement of people (Hart et al. 1997, McNeely2003, Loucks et al. 2008). All of these factors affected largeareas of eastern and northern Cambodia throughout themajority of the latter half of the 20th century (Chandler2000). During this time large declines in the population anddistribution of large mammal species including tiger(Panthera tigris), leopard (P. pardus), Asian elephant(Elephas maximus), banteng (Bos javanicus), gaur(B. gaurus), and Eld’s deer (Cervus eldii) have been docu-mented (Loucks et al. 2008). These declines were associatedwith a proliferation of firearms, the development of anexternal market for wildlife products, and, particularly duringthe Khmer Rouge-era, government-sponsored hunting

(Loucks et al. 2008). Densities of large cats were particularlydepressed due to both direct mortality from hunting and thecollapse of prey populations.

Effective conservation and planning for the recovery ofremnant populations of threatened large carnivores requiresreliable estimates of population density in order to prioritizeconservation investments and assess the effects of conserva-tion interventions. Capture–recapture estimates of surveyplot abundance, based on non-invasive sampling obtainedthrough genetic analysis and camera-traps, are widely used toestimate densities of cats and other large carnivores (Karanthand Nichols 1998, Boulanger et al. 2002, Harihar et al. 2008,Mondol et al. 2009). These methodologies are key compo-nents of programmatic monitoring for tiger and other largecarnivores and are regarded as best-practice techniques(Nichols and Karanth 2002, Karanth and Nichols 2010).

However, conventional capture–recapture techniques esti-mate abundance as opposed to density. Therefore with re-spect to sampling area (e.g., a camera-trap grid) survey plotabundance is likely to be overestimated as animals with onlypart of their home range within the sampling area are

Received: 29 June 2010; Accepted: 7 April 2011;Published: 16 September 2011

1E-mail: [email protected]

The Journal of Wildlife Management 76(1):163–169; 2012; DOI: 10.1002/jwmg.230

Gray and Prum � Leopard Density in Cambodia 163

Page 2: Leopard density in post-conflict landscape, Cambodia: Evidence from spatially explicit capture–recapture

available for capture (Parmenter et al. 2003). Traditionaldensity estimation methods account for this by adding aboundary strip width around the sampling area to determinethe effective trapping area (Wilson and Anderson 1985).This strip should approximate the distance animals availablefor capture move away from the sampling area during normalmovements (White et al. 1982, Parmenter et al. 2003). Themost widely used boundary strip width in capture–recapturecamera-trap studies is half the mean maximum distance moved(HMMDM; Karanth and Nichols 1998). However, this essen-tially ad hoc manner for estimating effective survey area has beenidentified as the weak link in this widely used methodology(Borchers and Efford 2008, Karanth and Nichols 2010).

A recently developed alternative approach, which does notassume geographic closure or estimate the area sampled, islikelihood-based spatially explicit capture–recapture (SECR;Borchers and Efford 2008). Spatially explicit capture–recap-ture models use the locations at which each animal isdetected within a likelihood-based framework to fit a spatialmodel of the detection process. These models combinecapture–recapture with distance sampling methods to esti-mate each animals range and center of activity and model theprobability density functions for detections of animals basedon distance from activity centers (Borchers and Efford 2008,Borchers 2011). In addition to uncertainties regarding theeffective trapping area, the choice of estimator for abundanceis likely to strongly affect density estimates. Although thishas been acknowledged in a number of taxa (e.g., grizzly bear[Ursus arctos]; Boulanger et al. 2002), few studies on Asiancats have examined the effects of different closed populationabundance estimators on density estimates.

We used closed population capture–recapture methodolo-gies on leopard camera-trapping data from within the corearea of Mondulkiri Protected Forest, eastern Cambodia. Wecompared estimates using a number of abundance estimatorsin software CAPTURE (Rexstad and Burnham 1991) andMARK (White 2008) and calculated densities using both thetraditional HMMDM buffer and newer SECR techniques.Our objectives were also to obtain the first density estimatesfor any large carnivore from Cambodia and provide infor-mation on the current status of leopard in MondulkiriProtected Forest therefore providing a baseline against whichthe effectiveness of future conservation activities can bemeasured.

STUDY AREA

Mondulkiri Protected Forest (approx. location 128080N,1068050E) is located in eastern Cambodia and forms partof a protected area complex of over 13,000 km2 includingYok Don National Park in Dak Lak province, Vietnam(Fig. 1). The core area of Mondulkiri Protected Forest isapproximately 1,500 km2 and consists of a matrix of domi-nant flatland deciduous dipterocarp forest with smallerpatches of semi-evergreen and mixed deciduous forest.Until the Paris Peace Accords of 1992, the area was heavilyharvested for high-value large mammal species (e.g.,elephant, wild cattle, large cats, Sunda pangolin [Manisjavanica]) partly to fund Khmer Rouge guerrilla activities.

The first peace-time surveys of the site in 1996 and 2000,while documenting the presence of Asian elephant and tiger,reported extremely low encounter rates of signs of largecarnivores and considerable numbers of traps for large catswere observed (Desai and Lic 1996, Long et al. 2000). Since2005, Mondulkiri Protected Forest has been managed by theForestry Administration of the government of Cambodiawith financial and technical assistance from the World WideFund for Nature (WWF). The protected area and the adja-cent forest landscape has received more than $5 million ininternational conservation investment during this period.This investment has supported protected area infrastructuredevelopment; the recruitment, training, and operation ofranger enforcement patrols; improved judiciary response;and alternative livelihoods work with communities adjacentto the protected area.

METHODS

Camera-Trap MethodsWe set-out camera-trap pairs in a grid pattern following theprotocols of Nichols and Karanth (2002) for closed popula-tion capture–recapture studies on large carnivores. Fiftycamera-trap (Reconyx RapidFire Professional PC90;Reconyx, Inc., Holmen, WI) pairs operated simultaneouslywithin 211 km2 of the core area of Mondulkiri ProtectedForest (Fig. 1). Camera-trap pairs were located either side ofroutes (i.e., motorbike trails, dry-river beds, and ridgelines)designed to maximize encounters with large carnivores andwere spaced approximately 2–3 km apart, thereby ensuringthat no individuals had a non-zero capture probability (i.e.,camera-trap spacing sufficiently small to ensure no homeranges between cameras). The camera-trap grid was opera-tional for 67 nights (8 Apr 2009–13 Jun 2009) with eachsampling occasion a single 24-hr period. This period, justover 2 months, is well within the range of periods assumed tosatisfy the assumption of demographic closure in similarstudies on Asian large cats (Karanth et al. 2004, Hariharet al. 2008, Wang and Macdonald, 2009).

Analytical MethodsEstimating leopard abundance.—We identified individual

leopards based on unique pelage patterns. To reduce thepossibilities of individual misidentification, we only usedencounters when both flanks were clearly photographed.We rejected unclear photographs (36 encounters). One ob-server identified all individuals (T.N.E. Gray). We devel-oped individual capture histories in a standard X-matrixformat with rows representing the capture histories ofeach captured individual and columns representing captureson each sampling occasion. We estimated leopard populationsize based on closed population models implemented insoftware CAPTURE (Rexstad and Burnham 1991) andMARK (White 2008). We formally tested population clo-sure using program CloseTest (Stanley and Burnham 1999).

Program CAPTURE allowed us to compare probabilisticmodels of the underlying capture–recapture process that werelikely to have generated the observed capture histories

164 The Journal of Wildlife Management � 76(1)

Page 3: Leopard density in post-conflict landscape, Cambodia: Evidence from spatially explicit capture–recapture

Figure 1. Location of Mondulkiri Protected Forest (MPF), eastern Cambodia and camera-trap grid used for density estimation of leopard (Panthera pardus)April to June 2009. Half Mean Maximum Distance Moved buffer (HMMDM) and buffer width used for spatially explicit capture–recapture analysis inDENSITY indicated.

Gray and Prum � Leopard Density in Cambodia 165

Page 4: Leopard density in post-conflict landscape, Cambodia: Evidence from spatially explicit capture–recapture

(White et al. 1982). Jack-knife population estimates frommodel Mh (heterogeneous capture probabilities but no trapor time response) are generally reported for Asian largecarnivores (Karanth et al. 2004, Maffei et al. 2004,Simcharoen et al. 2007, Wang and Macdonald, 2009).However, Chao heterogeneity models (Mh Chao; Chao1989) are theoretically more robust to heterogeneity in cap-ture probabilities when data are sparse, and therefore may bemore appropriate for smaller datasets (Boulanger et al. 2002).

Closed population Pledger models (Pledger 2000), whereheterogeneity is handled using a finite number of mixtures,and Huggins models (Chao and Huggins 2005), which canincorporate covariates affecting capture and recapture prob-abilities, can be run in Program MARK (White 2008). Thesemodels may be more resilient at low sample sizes and arebetter able to deal with heterogeneity in capture probabilitiesthan simpler models in CAPTURE (Boulanger et al. 2002).

We estimated leopard abundance using Mh (Jack-knife)and Mh (Chao) models in CAPTURE and Full ClosedCaptures with Heterogeneity models in MARK implement-ing Pledger mixture models with 2 mixtures of capture andrecapture probabilities (White 2008; henceforth Pledgermodels). We also obtained abundance estimates inMARK with leopard gender (male and female) coded asan attribute group affecting capture and recapture probabili-ties (henceforth gender models). This is analogous to usinggender as a covariate in Huggins models with the advantagethat, as abundance is calculated in the likelihood (Cooch andWhite 1999), direct comparisons with estimates fromPledger models are valid (Williams et al. 2002). We ranmodels M0, Mh, Mb, and Mbh (sensu White 2008 respec-tively: the null model with all capture and recapture proba-bilities equal, the heterogeneity model with 2 mixtures ofidentical capture and recapture probabilities, the behavioralmodel with different capture and recapture probabilities, andthe behavior and heterogeneity model with 2 mixtures ofcapture and recapture probabilities and a behavioral response)and model-averaged abundance estimates (Burnham andAnderson 2002), based on Akaike Information Crtieriaweights corrected for small sample size (AICc), acrossboth Pledger and gender models.Estimating leopard density.—Leopard densities (per

100 km2) based on our abundance estimates fromCAPTURE and MARK were calculated using theHMMDM method, in which a buffer of half of the meanmaximum distance moved for all individuals captured atmore than 1 camera-trap location is added to the trap-grid polygon (Karanth and Nichols 1998). We calculatedvariance in densities following Karanth and Nichols (1998).We also obtained leopard density estimates using full maxi-mum likelihood SECR in Program DENSITY (DENSITY4.4, www.otago.ac.nz/density/, accessed 20 Oct 2009). Thisdid not rely upon closed population estimates fromCAPTURE or MARK. We coded leopard gender (maleor female) as a session, analogous to attribute groups inMARK (M. Efford, University of Otago, personal commu-nication), affecting estimates of g0 (capture probability at thecenter of an individual’s home range) and sigma (a function

of the scale of animal movements; Efford 2004). We assumeda half-normal spatial capture probability function and aPoisson distribution of home-range centers. We set a bufferwidth around the trapping grid at 10,000 m.

RESULTS

We photographed 12 individually identified leopards, 5 maleand 7 female, during the 3,738 camera-trap night surveyperiod. The number of encounters with individually identi-fied leopard was 60; the number of sampling occasions (24-hrperiods) in which each individual was encountered variedbetween 1 and 23 (mean 5 � SE 1.9). Capture frequencieswere 23, 10, 9, 8, and 1 for the 5 males and 2, 2, 1, 1, 1, 1, and1 for the 7 females.

The program CloseTest supported the assumption of pop-ulation closure during the 67-day survey period (x2 ¼ 1.5,P ¼ 0.27). In Program CAPTURE, tests for heterogeneityin trapping probabilities (x2

1 ¼ 10:5, P < 0.01) and the af-fect of a behavioral response (x2

1 ¼ 3:4, P ¼ 0.07) supportedthe appropriateness of these models in comparison withplausible alternatives (all P < 0.5). The overall model selec-tion test in CAPTURE ranked Mh (incorporating individualheterogeneity in capture probabilities) as the best model(rated 1) followed by Mbh (incorporating both a behavioralresponse and individual heterogeneity; rated 0.8). Average P-hat (the probability of detecting an individual on at least onesampling occasion) was 0.19 and the population estimateusing Mh with the jack-knife estimator was 16 individuals(�SE 4.1) and 21 (�SE 10.2) using the Chao estimator.

In Program MARK, abundance estimates varied between12 and 22.4 (�SE 10.8) individual leopards (Table 1).Pledger models Mh and Mbh were by far the most stronglysupported, based on AICc scores, with a model-averagedleopard abundance estimate of 21.6 (�SE 15.7) individuals(Table 1). Model-averaged initial capture probability (P) was0.15 � SE 0.02 (group 1 individuals/males) and 0.009 �SE 0.005 (group 2 individuals/females) and recapture prob-ability (c) 0.19 � SE 0.04 (group 1 individuals/males) and0.008 � SE 0.002 (group 2 individuals/females).

Leopard Densities

We captured 6 individual leopards, 4 male and 2 female, atdifferent locations. Maximum distance moved between cap-tures were between 4.8 km and 16.8 km (mean 10.3 � SE2.1); total study area, based on HMMDM, was therefore560 km2 � SE 77.12. Leopard density estimates were2.9 � SE 0.9 individuals/100 km2 based on model Mh(Jack-knife) in CAPTURE, 3.8 � SE 1.9 individuals/100 km2 based on model Mh (Chao) in CAPTURE, and3.9 � SE 2.9 individuals/100 km2 based on model-averaged abundance estimates from MARK. In SECR anal-ysis in DENSITY, we estimated leopard density at 3.6 � SE1.0 individuals/100 km2. Capture probability at home-rangecenter (G0) was estimated at 0.04 � SE 0.01 (males) and<0.01 � SE <0.01 (females) and sigma (a function ofmovement) was 4,700 m � SE 650 m (males) and4,900 m � SE 1,050 m (females).

166 The Journal of Wildlife Management � 76(1)

Page 5: Leopard density in post-conflict landscape, Cambodia: Evidence from spatially explicit capture–recapture

DISCUSSION

We have presented the first robust density estimate for anylarge carnivore from Cambodia. This has indicated a sub-stantial leopard population within Mondulkiri ProtectedForest. Our leopard density estimates using Chao heteroge-neity models in CAPTURE, closed population capture–recapture modeling in MARK and SECR in DENSITY,with effective survey area for the former 2 calculated usingthe Half Mean Maximum Distance Moved (HMMDM)buffer, were between 3.6 � SE 1.0 and 3.9 � SE 2.9 indi-viduals/100 km2. Although true population size is unknown,and therefore we cannot infer bias of estimates, these 3methodologies gave similar density estimates. Howevermodel Mh Jack-knife in CAPTURE, traditionally themost widely used in capture–recapture studies of Asiancats (e.g., Karanth et al. 2004, Johnson et al. 2006, Wangand Macdonald, 2009, Gopal et al. 2010), gave the lowestdensity estimate; more than 20% lower than from the othermethods. The Jack-knife estimator is known to bias low insituations with small sample size and low capture probabili-ties (Chao 1989) a scenario invariably encountered duringsurveys of elusive and rare large carnivores. The continuedused of the Jack-knife estimator for analysis of camera-trapstudies of large carnivores therefore seems inappropriate.

Our density estimates from all 4 methodologies used were,however, bounded by large confidence intervals. These arelikely to have resulted from the relatively small sample size ofphoto-captured individuals and the low number of recapturesparticularly for female leopards. If our density estimates fromMh Chao, MARK, and SECR are extrapolated across theentire 1,500 km2 core area of Mondulkiri Protected Forest,the overall population of leopards, within 95% confidenceintervals, lies between 16 individuals and 113 individuals.Even accounting for the high level of uncertainty around theestimate, this represents one of the largest leopard popula-tions in South-east Asia. The wide confidence intervalshowever highlight that when using camera-traps for survey-ing rare large carnivores maximizing both the sample size ofphoto-captured individuals and individual capture and re-capture probabilities is important for meaningful monitoring.

Salom-Perez et al. (2007) suggested that low capture prob-abilities for female jaguar (Panthera onca) may arise fromdifferences in the use of human-made trails between thesexes with females being more timid and socially subordinate,and thus less likely to use human-made trails. We proposethis is the underlying reason for the low capture probabilityof female leopards in our study. Although only 50% ofcamera-traps were located on human-made trails, theseaccounted for >80% of all leopard encounters. Identifyingadditional routes used by large carnivores, particularly theless socially dominant females who may avoid the human-made trails regularly patrolled by males, was difficult. Thismay particularly be the case in remote landscapes such asMondulkiri Protected Forest where access is difficult andknowledge of the landscape imperfect; in such scenariosdifficulties in identifying locations away from human-made trails to deploy camera-traps may lead to low captureand recapture probabilities and thus greater uncertaintyaround density estimates.

Great care is also needed when extrapolating densities fromour study area, which represented approximately 15% of thecore area of Mondulkiri Protected Forest, to the entireprotected area complex. However, levels of law enforcementpatrolling activity and the prey base (T. N. E. Gray and R.Singh, WWF, unpublished data) are similar throughout thecore of Mondulkiri Protected Forest, as are vegetationpatterns and topography. Camera-trap encounter rates, stan-dardized for effort, are also similar within our study area andthe core area of the adjacent Phnom Prich Wildlife Sanctuary(T. N. E. Gray and C. Phan, WWF, unpublished data).Additional camera-trapping in the buffer and communityuse zones of these 2 protected areas, where leopard signs arewidespread, will be valuable further work in order to betterunderstand the value of the landscape for leopard conserva-tion and the effective size of the protected areas.

A distinctive characteristic of our dataset were skewedcapture histories with high heterogeneity in capture proba-bilities between male and female leopards. Given gender andage-specific differences in ranging patterns of large carni-vores (Woods et al. 1999, Boulanger and McLellan 2001,Salom-Perez et al. 2007, Crespin et al. 2008, Sharma et al.

Table 1. Estimates of leopard abundance (�SE), from Pledger and Gender closed population models in MARK, within Mondulkiri Protected Forest, EasternCambodia. Models ranked based on Akaike Information Criteria corrected for small sample size (AICc).

Model specificationa Model type AICc Delta AICc AICc weight Parameters Abundance estimate (�SE)

Mh Pledger 335.9 0 0.573 4 22.4 (�10.7)Mbh Pledger 336.5 0.6 0.418 6 20.5 (�22.6)Mb (male) M0 (female) Gender 345.6 9.7 0.004 4 19.8 (�8.6)Mb (male) Mb (female) Gender 345.7 9.9 0.004 4 19.8 (�8.6)M0 (male) M0 (female) Gender 350.5 14.6 <0.001 3 20.5 (�9.6)M0 (male) Mb (female) Gender 352.5 16.6 <0.001 4 21.9 (�19.3)Mb Pledger 379.4 43.5 <0.001 3 13.9 (�3.6)M0 Gender 388.9 53.0 <0.001 1 12 (�0)M0 Pledger 402.2 66.3 <0.001 1 12 (�0)

a Model definitions: M0 (null model with all capture and recapture probabilities equal), Mb (behavioral response with different capture and recaptureprobabilities), Mh (2 mixture heterogeneous capture and recapture probabilities); Mbh (2 mixture heterogeneous capture and recapture probabilities with abehavioral response, i.e., capture and recapture probabilities different). In gender models capture and recapture probabilities different between male andfemale leopards either with (Mb) or without (M0) a behavioral response.

Gray and Prum � Leopard Density in Cambodia 167

Page 6: Leopard density in post-conflict landscape, Cambodia: Evidence from spatially explicit capture–recapture

2009), gender-specific variation in capture probabilities,resulting from a disparity in the exposure of different gendersto traps, is likely to be a feature of many capture–recapturedatasets on large carnivores. Given, that gender may stronglyinfluence detection probabilities, it may be important tomodel its effect on capture probabilities and hence abun-dance estimates. Therefore, we recommend exploring datausing models in which gender can be included as a covariateaffecting capture probabilities. However, in our study suchgender models were less supported than Pledger heteroge-neity models.

A widely acknowledged disadvantage of using abundanceestimates from closed population capture–recapture modelsfor monitoring large carnivores is that translating abundanceto density is a largely ad hoc process based on observedanimal movements (Borchers and Efford 2008, Karanthand Nichols 2010). Although recent developments ofSECR intrinsically estimate density, as opposed to abun-dance, they remain constrained by a number of assumptionsconcerning spatial use and animal distributions (Efford2004). How SECR performs when the underlying assump-tions are violated is yet to be extensively explored and theassumptions need to be considered, and related to the biologyof the study species and characteristics of the study area,before uncritically using SECR. Spatially explicit capture–recapture models in DENSITY assume that all animals havecircular home ranges with a Poisson (i.e., random) distribu-tion of home-range centers, that animal movements are bestapproximated by a bivariate normal distribution, and that theeffective survey area (e.g., camera-trap grid) is completelyopen (Efford 2004). In the homogenous flatland deciduousdipterocarp dominated forests of Mondulkiri, with few nat-ural barriers to animal movements, such as lakes and moun-tains, these assumptions are likely to be valid for a habitatgeneralist such as the leopard. However, in smaller isolatedprotected areas, such as many in the Indian subcontinent, orother study species, such as Asian elephants where capture-mark-recapture techniques are being used on fecal DNA,many of the assumptions of SECR in DENSITY may be lessstrongly supported.

MANAGEMENT IMPLICATIONS

Baseline density estimates for large carnivores are essentialfor monitoring effectiveness of conservation activities.However, due to the nature of sampling rare large carnivores,low sample size and low probabilities of capture and recap-ture may lead to uncertainty around abundance estimates.Although our estimates were bounded by high confidenceintervals, we provide some evidence that leopard densitiescomparable to those in well protected national parks inThailand (e.g., Simcharoen et al. 2008) have been reachedfollowing conservation activities in Cambodian lowland for-est. Detectability (i.e., capture and recapture rates) was lowand appeared to differ considerably between male and femaleleopards. Such variation may be inherent in large carnivorestudies given gender differences in behavior and naturalhistory (Woods et al. 1999, Boulanger and McLellan2001, Salom-Perez et al. 2007, Sharma et al. 2009).

Although in our dataset, models which can account forindividual heterogeneity (Pledger and Chao models) weremore strongly supported than those using gender as a covar-iate affecting detectability (gender models), we recommendexploring data using more complex models in MARK par-ticularly given the demographic importance of breedingfemales for population recovery of threatened carnivores.

ACKNOWLEDGMENTS

We conducted this study as part of WWF Greater MekongCambodia Program’s Eastern Plains Landscape project. Projectfunding was provided by WWF-US, WWF-Sweden, andHumanscale. Work in Mondulkiri Protected Forest is sup-ported by the Forestry Administration of the Ministry ofForestry Fisheries and Agriculture, Cambodia and we thankHis Excellency S. Ty, P. Men, and S. Keo. K. Lien, N. Lien,T. Mel, C. Me, C. Phan, and O. Khaev assisted in camera-trapping. K. Huy produced the figure. S. Barber-Meyer,P. Doggerty, 3 anonymous reviewers, and K. McKelvey pro-vided valuable comments on previous drafts of this study.A. Harihar, M. Efford, and www.phidot.org forum providedstatistical advice and C. Bruce, N. Cox, B. Long, B. Pandav,and T. Seng assisted in project planning and logistics.

LITERATURE CITEDBorchers, D. L. 2011. A non-technical overview of spatially explicit capture

recapture models. Journal of Ornithology 152: in press.

Borchers, D. L., and M. G. Efford. 2008. Spatially explicit maximumlikelihood methods for capture–recapture studies. Biometrics 64:377–385.

Boulanger, J., and B. N. McLellan. 2001. Closure violation in DNA-basedmark-recapture estimation of grizzly bear populations. Canadian Journalof Zoology 79:642–651.

Boulanger, J., G. C. White, B. N. McLellan, J. Woods, M. Proctor, and S.Himmer. 2002. A meta-analysis of grizzly bear DNA mark-recaptureprojects in British Colombia, Canada. Ursus 13:137–152.

Burnham, K. P., and D. R. Anderson. 2002. Model selection and multi-model inference. Springer, New York, New York, USA.

Chandler, D. 2000. A history of Cambodia. Third edition. Westview Press,Boulder, Colorado, USA.

Chao, A. 1989. Population size for sparse data in capture–recapture experi-ments. Biometrics 45:427–438.

Chao, A., and R. M. Huggins. 2005. Modern closed-population capture–recapture models. Pages 58–87 in S. C. Amstrup, T. L. McDonald, and B.F. J. Manly, editors. Handbook of capture. Princeton University Press,Oxford, United Kingdom.

Cooch, E., and G. G. White. 1999. MARK Program Mark. A gentleintroduction. 8th edition. <http://www.phidot.org/software/mark/index.html>. Accessed Dec 2009.

Crespin, L., R. Choquet, M. Lima, J. Merrit, and R. Pradel. 2008.Is heterogeneity of catchability in capture–recapture studies a mere sam-pling artifact or a biologically relevant feature of the population?Population Ecology 50:247–256.

Desai, A. A., and V. Lic. 1996. Status and distribution of large mammals ineastern Cambodia: results of the first foot surveys in Mondulkiri andRatanakiri. IUCN/FFI/WWF, Phnom Penh, Cambodia.

Efford, M. 2004. Density estimation in live-trapping studies. Oikos 106:598–610.

Gopal, R., Q. Qureshi, M. Bhardwaj, R. K. J. Singh, and Y. V. Jhala. 2010.Evaluation the status of the endangered Tiger Panthera tigris and its preyin Panna Tiger Reserve, Madhya Pradesh, India. Oryx 44:383–389l.

Hanson, T., T. M. Brooks, G. A. B. da Fonseca, M. Hoffmann, J. F.Lamoreux, G. Machlis, C. G. Mittermeier, R. A. Mittermeier, and J. D.Pilgrim. 2009. Warfare in biodiversity hotspots. Conservation Biology23:578–587.

168 The Journal of Wildlife Management � 76(1)

Page 7: Leopard density in post-conflict landscape, Cambodia: Evidence from spatially explicit capture–recapture

Harihar, A., B. Pandav, and S. P. Goyal. 2008. Responses of tiger (Pantheratigris) and their prey to removal of anthropogenic influences in RajajiNational Park, India. European Journal of Wildlife Research 55:97–105.

Hart, T., J. Hart, C. Fimbel, and R. Fimbel. 1997. Conservation and civilstrife: two perspectives from Central Africa. Conservation Biology11:308–310.

Johnson, A., C. Vongkhamheng, M. Hedemark, and T. Saithongdam. 2006.Effects of human-carnivore conflict on tiger (Panthera tigris) and preypopulations in Lao PDR. Animal Conservation 9:421–430.

Karanth, K. U., R. S. Chundawat, J. D. Nichols, and N. Samba Kumar.2004. Estimation of tiger densities in the tropical dry forests of Panna,Central India, using photographic capture–recapture sampling. AnimalConservation 7:285–290.

Karanth, K. U., and J. D. Nichols. 1998. Estimation of tiger densities inIndia using photographic capture and recaptures. Ecology 79:2852–2862.

Karanth, K. U., and J. D. Nichols. 2010. Non-invasive survey methods forassessing tiger populations. Pages 461–481 in R. Tilson and P. Nyhus,editors. Tigers of the world: the science, politics and conservation ofPanthera tigris. Second edition. Elsevier, New York, New York, USA.

Long, B., S. Swan, and M. Kry. 2000. Biological surveys in NortheastCambodia, April 2000. FFI and the Wildlife Protection Office, Hanoi& Phnom, Penh, Cambodia.

Loucks, L., M. B. Mascia, A. Maxwell, K. Huy, K. Duoung, N. Chea, B.Long, N. Cox, and T. Seng. 2008. Wildlife decline in Cambodia, 1953–2005: exploring the legacy of armed conflict. Conservation Letters 2:82–92.

Maffei, L., E. E Cuellar, and A. Noss. 2004. One thousand jaguars(Panthera onca) in Bolivia’s Chaco? Camera trapping in the Kaa-IyaNational Park. Journal of Zoological Society London 262:295–304.

McNeely, J. A. 2003. Conserving forest biodiversity in times of violentconflict. Oryx 37:142–152.

Mondol, S., K. U. Karanth, A. M. Gopalaswarmy, A. Andheria, and U.Ramakrishnan. 2009. Evaluation of non-invasive genetic sampling meth-ods for estimating tiger population size. Biological Conservation142:2350–2360.

Nichols, J. D., and K. U. Karanth. 2002. Statistical concepts: estimatingabsolute densities of tigers using capture–recapture sampling. Pages 124–137 in K. U. Karanth and J. D. Nichols, editors. A manual for researchers,managers and conservationists in tropical Asia. Centre for WildlifeStudies, Bangalore, India.

Parmenter, R. R., T. L. Yates, D. R. Anderson, K. P. Burnham, J. L.Dunnum, A. B. Franklin, M. T. Friggens, B. C. Lubow, M. Miller, andG. S. Olson. 2003. Small-mammal density estimation: a field comparisonof grid-based vs. web-based density estimators. Ecological Monographs73:1–26.

Pledger, S. 2000. Unified maximum likelihood estimates for closed modelsusing mixtures. Biometrics 56:434–442.

Rexstad, E., and K. P. Burnham. 1991. User’s guide for interactive programCAPTURE abundance estimation of closed animal populations. ColoradoState University, Fort Collins, USA.

Salom-Perez, R., E. Carrillo, J. C. Saenz, and J. M. Mora. 2007. Criticalcondition of the jaguar Panthera onca in Corcovado National Park, CostaRica. Oryx 41:1–7.

Sharma, R. K., Y. Jhala, Q. Qureshi, J. Vattakaven, R. Gopal, and K. Nayak.2009. Evaluating capture–recapture population and density estimation oftigers in a population with known parameters. Animal Conservation 13:94–103.

Simcharoen, S., A. C. D. Barlow, A. Simcharoen, and J. L. D. Smith. 2008.Home range size and daytime habitat selection of leopards in Huai KhaKhaeng Wildlife Sanctuary, Thailand. Biological Conservation 141:2242–2250.

Simcharoen, S., A. Pattanvibool, K. U. Karanth, J. D. Nichols, and N.Sambar Kumar. 2007. How many tigers Panthera tigris are there in HuaiKha Khaeng Wildlife Sanctuary, Thailand? An estimate using photo-graphic capture–recapture sampling. Oryx 41:447–453.

Stanley, T. R., and K. P. Burnham. 1999. A closure test for time-specificcapture–recapture data. Environmental Ecological Statistics 6:197–209.

Wang, S. M., and D. W. MacDonald. 2009. The use of camera traps forestimating tiger and leopard populations in the high altitude mountains ofBhutan. Biological Conservation 142:606–613.

White, G. C. 2008. Closed population estimation models and their exten-sions in Program MARK. Environmental and Ecological Statistics 15:89–99.

White, G. C., D. R. Anderson, K. P. Burnham, and D. L. Otis. 1982.Capture–recapture and removal methods for sampling closed populations.Los Alamos National Laboratory Rep. LA-8787-NERP, Los Alamos,New Mexico, USA.

Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Analysis andmanagement of animal populations. Academic Press, San Diego,California, USA.

Wilson, K. R., and D. R. Anderson. 1985. Evaluation of two densityestimators of small mammal population size. Journal of Mammalogy66:13–21.

Woods, J. G., D. Paetkau, D. D Lewis, B. N. McLellan, M. Proctor, andC. Strobeck. 1999. Genetic tagging free-ranging black and brown bears.Wildlife Society Bulletin 27:616–627.

Associate Editor: Kevin McKelvey.

Gray and Prum � Leopard Density in Cambodia 169