6
Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar Rahel Sollmann a,b,, Natália Mundim Tôrres a,c , Mariana Malzoni Furtado a,d , Anah Tereza de Almeida Jácomo a , Francisco Palomares e , Severine Roques e , Leandro Silveira a a Jaguar Conservation Fund/Instituto Onça-Pintada, Mineiros, Goiás, Brazil b North Carolina State University, Department of Forestry and Environmental Resources, Raleigh, NC, USA c Universidade Federal de Uberlândia, Instituto de Biologia, Uberlândia, Minas Gerais, Brazil d Universidade de São Paulo, Faculdade de Medicina Veterinária e Zootecnia, São Paulo, Brazil e Department of Conservation Biology, Estación Biológica de Doñana (CSIC), Sevilla, Spain article info Article history: Received 10 May 2013 Received in revised form 26 July 2013 Accepted 1 August 2013 Keywords: Brazil Caatinga Carnivores Estimator performance Panthera onca Population estimation Scat survey abstract Abundance and density are key pieces of information for questions related to ecology and conservation. These quantities, however, are difficult to obtain for rare and elusive species, where even intensive sam- pling effort can yield sparse data. Here, we combine data from camera-trapping and noninvasive genetic sampling (scat surveys) of a jaguar population in the Caatinga of northeastern Brazil, where the species is threatened and little studied. We analyze data of both survey types separately and jointly in the frame- work of spatial capture–recapture. Density estimates were 1.45 (±0.46) for the camera-trap data alone, 2.03 (±0.77) for the genetic data alone, and 1.57 (±0.43) and 2.45 (±0.70) for the two methods, respec- tively, in the joint analysis. Density and other parameters were estimated more precisely in the joint model. Particularly the differences in movement between males and females were estimated much more precisely when combining both data sources, especially compared to the genetic data set alone. When compared to a previous non-spatial capture–recapture approach, present density estimates were more precise, demonstrating the superior statistical performance of spatial over non-spatial capture recapture models. The ability to combine different surveys into a single analysis with shared parameter allows for more precise population estimates, while at the same time enabling researchers to employ complemen- tary survey techniques in the study of little known species. Ó 2013 Elsevier Ltd. All rights reserved. 1. Introduction Knowledge of population abundance and density is of funda- mental importance for many questions related to ecology, conser- vation and management. For most organisms, but especially for rare and cryptic species, we cannot census entire populations but have to rely on estimation techniques that account for our imper- fect ability to detect individuals (Williams et al., 2002). Capture– recapture models have been a standard tool in estimating wildlife population parameters for decades. More recently, traditional cap- ture–recapture models have been extended to spatial capture–re- capture (SCR) models. SCR models combine a detection process describing how animals are detected by an array of traps, condi- tional on an ecological process describing individual distribution and movement in space (Efford, 2004; Efford et al., 2004; Gardner et al., 2009; Royle and Young, 2008). Contrary to traditional cap- ture–recapture, these models fully account for heterogeneity in detection probability resulting from different exposure of individ- uals to the trap array. Further, they explicitly link abundance to a specific area, so that density estimation is straight forward, whereas traditional capture–recapture relies on ad hoc approaches to determine the area sampled and transform abundance into den- sity estimates. Rare and cryptic species, though often of particular conserva- tion concern, are generally challenging to study, and even consid- erable survey effort may only yield sparse data. Sparse data limit our ability to obtain precise parameter estimates and to include potentially important covariates into an analytical model. Ap- proaches that allow us to combine data from multiple surveys are therefore of prime importance in the study of rare and elusive species. In the context of SCR, combining data from several surveys using the same technique (for example, camera-trapping) either at several sites, or repeatedly over time, has enabled researchers to estimate density of little studied species such as leopard cats (Mohamed et al., 2013) and Sunda clouded leopards (Wilting 0006-3207/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.biocon.2013.08.003 Corresponding author. Address: North Carolina State University, Department of Forestry and Environmental Resources, Turner House, Campus Box 7646, Raleigh, NC 27695-7646, USA. Tel.: +1 919 706 3658. E-mail address: [email protected] (R. Sollmann). Biological Conservation 167 (2013) 242–247 Contents lists available at ScienceDirect Biological Conservation journal homepage: www.elsevier.com/locate/biocon

Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar

  • Upload
    leandro

  • View
    214

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar

Biological Conservation 167 (2013) 242–247

Contents lists available at ScienceDirect

Biological Conservation

journal homepage: www.elsevier .com/locate /b iocon

Combining camera-trapping and noninvasive genetic data in a spatialcapture–recapture framework improves density estimates for the jaguar

0006-3207/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.biocon.2013.08.003

⇑ Corresponding author. Address: North Carolina State University, Department ofForestry and Environmental Resources, Turner House, Campus Box 7646, Raleigh,NC 27695-7646, USA. Tel.: +1 919 706 3658.

E-mail address: [email protected] (R. Sollmann).

Rahel Sollmann a,b,⇑, Natália Mundim Tôrres a,c, Mariana Malzoni Furtado a,d,Anah Tereza de Almeida Jácomo a, Francisco Palomares e, Severine Roques e, Leandro Silveira a

a Jaguar Conservation Fund/Instituto Onça-Pintada, Mineiros, Goiás, Brazilb North Carolina State University, Department of Forestry and Environmental Resources, Raleigh, NC, USAc Universidade Federal de Uberlândia, Instituto de Biologia, Uberlândia, Minas Gerais, Brazild Universidade de São Paulo, Faculdade de Medicina Veterinária e Zootecnia, São Paulo, Brazile Department of Conservation Biology, Estación Biológica de Doñana (CSIC), Sevilla, Spain

a r t i c l e i n f o

Article history:Received 10 May 2013Received in revised form 26 July 2013Accepted 1 August 2013

Keywords:BrazilCaatingaCarnivoresEstimator performancePanthera oncaPopulation estimationScat survey

a b s t r a c t

Abundance and density are key pieces of information for questions related to ecology and conservation.These quantities, however, are difficult to obtain for rare and elusive species, where even intensive sam-pling effort can yield sparse data. Here, we combine data from camera-trapping and noninvasive geneticsampling (scat surveys) of a jaguar population in the Caatinga of northeastern Brazil, where the species isthreatened and little studied. We analyze data of both survey types separately and jointly in the frame-work of spatial capture–recapture. Density estimates were 1.45 (±0.46) for the camera-trap data alone,2.03 (±0.77) for the genetic data alone, and 1.57 (±0.43) and 2.45 (±0.70) for the two methods, respec-tively, in the joint analysis. Density and other parameters were estimated more precisely in the jointmodel. Particularly the differences in movement between males and females were estimated much moreprecisely when combining both data sources, especially compared to the genetic data set alone. Whencompared to a previous non-spatial capture–recapture approach, present density estimates were moreprecise, demonstrating the superior statistical performance of spatial over non-spatial capture recapturemodels. The ability to combine different surveys into a single analysis with shared parameter allows formore precise population estimates, while at the same time enabling researchers to employ complemen-tary survey techniques in the study of little known species.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Knowledge of population abundance and density is of funda-mental importance for many questions related to ecology, conser-vation and management. For most organisms, but especially forrare and cryptic species, we cannot census entire populations buthave to rely on estimation techniques that account for our imper-fect ability to detect individuals (Williams et al., 2002). Capture–recapture models have been a standard tool in estimating wildlifepopulation parameters for decades. More recently, traditional cap-ture–recapture models have been extended to spatial capture–re-capture (SCR) models. SCR models combine a detection processdescribing how animals are detected by an array of traps, condi-tional on an ecological process describing individual distributionand movement in space (Efford, 2004; Efford et al., 2004; Gardner

et al., 2009; Royle and Young, 2008). Contrary to traditional cap-ture–recapture, these models fully account for heterogeneity indetection probability resulting from different exposure of individ-uals to the trap array. Further, they explicitly link abundance to aspecific area, so that density estimation is straight forward,whereas traditional capture–recapture relies on ad hoc approachesto determine the area sampled and transform abundance into den-sity estimates.

Rare and cryptic species, though often of particular conserva-tion concern, are generally challenging to study, and even consid-erable survey effort may only yield sparse data. Sparse data limitour ability to obtain precise parameter estimates and to includepotentially important covariates into an analytical model. Ap-proaches that allow us to combine data from multiple surveysare therefore of prime importance in the study of rare and elusivespecies. In the context of SCR, combining data from several surveysusing the same technique (for example, camera-trapping) either atseveral sites, or repeatedly over time, has enabled researchers toestimate density of little studied species such as leopard cats(Mohamed et al., 2013) and Sunda clouded leopards (Wilting

Page 2: Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar

Fig. 1. Camera-traps (grey circles) operated in 2007 and scat surveys (dashed lines)walked in 2008 to study the jaguar population of the Serra da Capivara NationalPark (cross) in the Caatinga (grey), northeastern Brazil (inset map); triangles markthe locations of jaguar scats collected in 2008.

R. Sollmann et al. / Biological Conservation 167 (2013) 242–247 243

et al., 2012). Recently, Gopalaswamy et al. (2012) developed an SCRmodel that combines data from a camera-trapping and a scat sur-vey of a tiger population and showed that using both data sourcesled to a more precise estimate of population density.

Monitoring populations using different techniques allowsresearchers and managers to draw upon the specific advantagesof each approach. For example, camera-traps are indisputably themost efficient tool to monitor populations of individually identifi-able large cats such as tigers or jaguars (e.g., Karanth and Nichols,1998; Wallace et al., 2003): camera-traps can be operated overlarge areas and extended periods of time with comparably little ef-fort and work continuously once they are set up. Surveys can easilybe standardized and plenty of ancillary data on the local mammalcommunity is collected without additional effort (Silveira et al.,2003). On the other hand, the collection and subsequent geneticanalysis of scats can provide insight into feeding ecology, geneticstructure and diversity and aspects of health of the population un-der study (Kohn and Wayne, 1997; Schwartz et al., 2007; Waitsand Paetkau, 2005). Clearly, both methods complement each otherin studying a species’ ecology.

The jaguar remains one of the less studied large felids (Brodie,2009). It is the largest cat of the Americas and with a loss ofapproximately 40% of its original range in the past century (Zeller,2007) it is one of the large mammals with the largest absoluterange contraction in the past 500 years (Morrison et al., 2007).Although the species’ stronghold is the Amazon rainforest (Sander-son et al., 2002), it occurs in a variety of habitats, including thesemi-arid Caatinga biome of northeastern Brazil. A recent regionalevaluation of the jaguar’s conservation status in the Caatingaplaced it at Critcally Endangered; habitat loss and especially highdegrees of habitat fragmentation threaten the species’ persistencein this biome (Paula et al., 2012).

The Serra da Capivara National Park (SCNP) holds an importantjaguar population for the Caatinga (Paula et al., 2012). Using cam-era-trapping and non-spatial capture–recapture models, Silveiraet al. (2009) estimated jaguar density in SCNP at 2.67 individu-als/100 km2. In the year following this camera trap effort, a jaguarscat collection survey was carried out at the SCNP. Here, we modifythe approach by Gopalaswamy et al. (2012) to estimate separatepopulation densities of jaguars in the SCNP in two subsequentyears, by combining data from both the camera-trapping and thescat collection survey within a single SCR model. Using an im-proved analytical tool, the density estimates we produce are morereliable than earlier estimates from non-spatial capture–recapturemodels. Further, we demonstrate how combining data from com-plementary survey methods can improve inference about popula-tion density and other ecological parameters. Since the ability todetect changes in populations depends strongly on the ability toestimate population size precisely, the approach we demonstrateis useful for many monitoring programs of rare and elusive species.

2. Methods

2.1. Study site

The Serra da Capivara National Park (SCNP) is located in thesouth of the state of Piauí, north-east Brazil (Fig. 1). It protects129,140 ha of the Caatinga, a semi-arid biome contained entirelywithin Brazil. Temperatures can reach as high as 45 �C, the rainyseason is from October to April (Emperaire, 1985), and mean totalannual precipitation is 644 mm (SMAPR, 1994). A 6 to 10-m-tallshrubby vegetation predominates the vegetation of the park(Emperaire, 1985). Altitude ranges from 280 to 600 m and thetopography consists of a main plateau bounded by 50- to 200-mcliffs and dissected by valleys and canyons. There is no natural per-

manent water within the Park but a system of artificial waterholeshas been constructed over the past 15 years.

2.2. Field surveys and data preparation

From August to October 2007, we set up 20 camera-trap sta-tions along the roads of the southern and central part of SCNP, cov-ering an area of approximately 205 km2 (Fig. 1), as described indetail by Silveira et al. (2009). Each station consisted of two pas-sive-sensor Camtrakker (CamTrack South Inc., Watkinsville, USA)camera traps, model Original 35 mm, facing each other to captureboth flanks of a passing animal. Stations were spaced at approxi-mately 3 km and checked every 15 days over 92 days. From thephotographs we constructed individual and trap station specificdetection histories, containing the number of times each individualwas photographed at each trap. To account for occasional cameratrap failure we also recorded the number of days each camera trapstation was functional over the total survey period.

Between November 10 and December 3 2008 we collected scatsin SCNP with the help of scat detector dogs (Long et al., 2007;Smith et al., 2001), trained to search for scats from jaguar andpuma. One dog-handler team walked 180.94 km during 21 tran-sects along the roads of SCNP, covering an area of approximately506 km2. The scat survey area overlapped largely with the cameratrap survey area (Fig. 1). The largest portion of each scat was storedfrozen for dietary analysis, and a small portion was stored in 96%ethanol for subsequent genetic analysis. DNA was extracted usingprotocols based on the GuSCN/silica method (Boom et al., 1990;Höss and Pääbo, 1993; Frantz et al., 2003). Posterior species iden-tification was carried out using an optimized PCR-based protocolas described in Roques et al. (2011). Sex was identified using themethod proposed by Pilgrim et al. (2005) based on the size

Page 3: Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar

244 R. Sollmann et al. / Biological Conservation 167 (2013) 242–247

difference between the male and female Amelogenin gene, butusing newly design primers (Roques et al., accepted for publica-tion). For individual genotyping, an optimized set of 11 microsatel-lite markers (Menotti-Raymond et al.. 1999) was used as describedin Roques et al. (accepted for publication).

To bring the dataset of both methods into a comparable format,we followed the approach by Russell et al. (2012) for unstructuredspatial sampling (as opposed to having a fixed grid of sampling de-vices as in camera-trapping): We divided the study area into a gridwith cell size of 2 � 2 km and treated the center of each grid cell asa ‘trap’. We assigned each scat to the center coordinate of the cell itwas found in and then constructed an individual and grid cell spe-cific encounter history containing the number of scats that werefound for each individual in each grid cell. We tested other grid cellsizes (1 � 1 km and 3 � 3 km) but found no effect of grid resolutionon density estimates. To account for varying spatial sampling effortwe calculated the distance walked by the dog-handler team in eachgrid cell.

2.3. Model

To analyze data, we formulated a joint spatial capture–recap-ture (SCR) model as described in Gopalaswamy et al. (2012). SCRmodels assume that each individual i has an activity center si, char-acterized by a pair of coordinates. All s are located within the state-space S, an area encompassing the trapping grid, which has to bechosen large enough so as to include all individuals that could havebeen exposed to the trapping grid. In our analysis we buffered themost extreme coordinates of the scat locations and the camera trapgrid by 30 km to define S.

In the following, the superscript C denotes data and parameterspertaining to camera trapping, G denotes the scat survey, and B de-notes parameters that are shared by both sampling types. We as-sume that for both methods the observed encounter history forindividual i at trap/grid cell j, yij, are mutually independent out-comes of a Poisson random variable,

yijC � PoissonðkCijÞ and yG

ij � PoissonðkGij Þ;

where variation in kij is modeled as a function of the distance oftrap/grid cell j to si, dij. Particularly, we use a Gaussian kernel (alsoreferred to as half-normal function) of the form:

kCij ¼ kC

0sex � ð�dC2ij =2 � rB2

sexÞ � eff Cj

kGij ¼ kG

0sex � ð�dG2ij =2 � rB2

sexÞ � eff Gj

The k0sex are the respective baseline encounter rates (the ex-pected encounter rate of an individual with a trap or at a grid cellcenter if that trap/cell center were located precisely at the individ-ual’s activity center) and we allowed this parameter to be sex-spe-cific. Since the detection process differs between methods, weassigned each data type a separate baseline encounter rate. Be-cause we cannot associate photographic images of an individualjaguar with its scat samples, and because home ranges and there-fore activity centers of individuals can shift over time, we allowedfor independent individual activity centers across the two surveys,hence we have two sets of distances, dC

ij and dGij . The parameter rsex

(units of the trapping grid, here km) is the scale parameter of theGaussian kernel and is related to animal home range radius (Rep-pucci et al., 2011). We expect that home ranges differ betweenthe sexes and thus allow males and females to have separate r.If we assume that movement remains constant across the two sur-veys, we can model r as a shared parameter of both detectionmodels for camera trapping and scat sampling, so that it is in-formed by both types of data. The variable effj is the sampling effort

at site j, for the camera traps the number of days the station wasfunctional, for the scat survey the distance walked by the dog-han-dler team in cell j.

We estimate N (the number of activity centers in S) for each sur-vey separately to account for possible changes in population sizebetween the two surveys. To do so, we used a Bayesian analysisby data augmentation of the model (Royle and Dorazio, 2012; Roy-le et al., 2007). In data augmentation, we let M be a number that islarger than the largest possible population size N in S, and n be thenumber of detected individuals. We assume a prior distribution forN that is discrete-uniform over the interval (0, M) and augment theobserved data set with M - n individuals that were never detectedand thus have encounter histories that are all zero. For all M poten-tial individuals we introduce an auxiliary variable, zi, which is 1 ifthe animal is part of the population and 0 if it is not. N is then thesum over all z. M is adequately large when estimates of N are notlimited by M. Density D can be derived by dividing N by the areaof S. Sex is coded as a binary variable, with 1 for males and 0 forfemales. Sex for the augmented individuals is treated as missingdata and estimated as a Bernoulli random variable with probabilityp, which represents the proportion of males in the population(Gardner et al., 2010; Sollmann et al., 2011). Again, it seems biolog-ically reasonable to assume that the sex ratio of a long lived speciesdoes not change significantly over the course of a year, so that wecan estimate p from the joint camera trapping and scat survey dataset.

We ran separate models for only the camera-trapping and onlythe scat data, as well as a joint model for the combined data. Weimplemented the models in the software WinBUGS (Gilks et al.,1994) accessed through R 2.13.0 (R Development Core Team,2011) using the package R2WinBUGS (Sturtz et al., 2005). We ran3 parallel Markov chains with 30,000 iterations each, a burn-in of10,000 iterations and a thinning rate of 3. For all parameters we re-port posterior means and standard deviations, as well as 2.5 and97.5 percentiles, which mark the 95% Bayesian Credible Intervals(BCI). We consider parameters to be significantly different whentheir 95% BCI do not overlap.

3. Results

3.1. Data

Camera traps accumulated 1450 trap days, during which we ob-tained 72 photographs of 13 jaguars (4 females and 6 males; 3unidentified). Individual photo-frequencies ranged from 1 to 17.Eighty-two potential jaguar scats were collected, and for 57 ofthem species could be identified (4 and 53 were puma and jaguarscats, respectively). Forty-seven jaguar scats were assigned to 16individuals (6 females and 10 males), with individual scat frequen-cies ranging from 1 to 11.

3.2. Model results

Analysis of the camera-trapping data set yielded an estimate ofjaguar density of 1.45 (±0.46) individuals/100 km2. Baselineencounter rate kC

0 was higher for females (0.09 ± 0.04) than formales (0.05 ± 0.09), but 95% BCI for both parameters overlapped.The scale parameter r was significantly larger for males(5.98 ± 1.21) than for females (2.67 ± 0.46).

Jaguar density estimated from the scat survey was higher, with2.03 (±0.77) individuals/100 km2, but BCI of density estimates forboth methods overlapped. Baseline detection rate for the scat sur-vey was significantly higher for males (0.21 ± 0.06) than for fe-males (0.04 ± 0.03), while r was higher for females than for

Page 4: Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar

R. Sollmann et al. / Biological Conservation 167 (2013) 242–247 245

males (6.92 ± 4.47 and 4.52 ± 0.63, respectively), but not signifi-cantly so.

The combined model lead to similar but slightly increased den-sity estimates for both methods (1.57 ± 0.43 and 2.45 ± 0.70 for thecamera-trap and scat survey, respectively). k0 for both years andsexes remained similar to results from the separate analyses. Thecombined r was significantly smaller for females (2.78 ± 0.44) thanfor males (5.10 ± 0.55).

All models indicated a mean sex ratio slightly skewed towardsfemales, but in all models 95% BCI of p overlapped 0.5, suggestingthat the sex ratio was not significantly different from 1:1. Posteriorsummaries of model parameters are listed in Table 1.

4. Discussion

Combining data and sharing information across surveys can im-prove inference from sparse data sets such as those generally ob-tained for elusive species. Traditional non-spatial capture–recapture models only have one fundamental detection parameter,the capture probability, p (which can vary with different covari-ates). Data from subsequent surveys using the same detectionmethodology (i.e., only camera-trapping, or only scat surveys)can be combined into a single capture–recapture model and pcan be shared across surveys if it can safely be assumed to remainconstant. But detection methods as distinct as camera-trappingand scat surveys are very unlikely to share detection probability– the process of finding scats with the help of dogs surely is differ-ent from a camera recording an image of a passing jaguar. Contraryto non-spatial capture recapture, the detection model in SCR hastwo parameters (in fact, it can have more than two, dependingon the detection-by-distance model used), one of which, r, is re-lated to animal home range radius. Home range size is an ecologi-cal quantity and does not depend on the detection method. Thus,even two methods with different detection processes can sharethis parameter when an SCR framework is used.

Table 1Posterior summaries of SCR model parameters for camera trapping data, scat data andcombined data; we present the posterior mean, the standard deviation (SD) and the2.5%, 50% and 97.5% intervals for the baseline encounter rate (k0), the half normalscale parameter (r), the probability of being a male (p) and density (D).

Mean SD 2.5% 50% 97.50%

Camerask0

a, female 0.09 0.04 0.04 0.08 0.18k0

a, male 0.05 0.09 0.02 0.04 0.17r, female [km] 2.67 0.46 1.97 2.61 3.74r, male [km] 5.98 1.21 4.35 5.73 9.15D [individuals/100 km2] 1.45 0.46 0.72 1.39 2.51p 0.35 0.14 0.11 0.34 0.66

Scatsk0

a, female 0.04 0.03 0.01 0.03 0.12k0

a, male 0.21 0.06 0.17 0.20 0.34r, female [km] 6.92 4.74 3.70 5.23 20.89r, male [km] 4.52 0.63 4.07 4.44 5.98p 0.47 0.18 0.33 0.46 0.84D [individuals/100 km2] 2.03 0.77 0.83 1.92 3.70

Both methodsk0

a, cameras, female 0.09 0.04 0.04 0.08 0.18k0

a, cameras, male 0.04 0.01 0.02 0.04 0.08k0

a, scats, female 0.07 0.04 0.02 0.06 0.17k0

a, scats, male 0.18 0.05 0.11 0.18 0.28r, female [km] 2.78 0.44 2.11 2.72 3.81r, male [km] 5.10 0.55 4.71 5.04 6.35p 0.35 0.10 0.18 0.34 0.56D, camera trapping [individuals/

100 km2]1.57 0.43 0.85 1.54 2.52

D, scat survey [individuals/100 km2] 2.45 0.70 1.29 2.37 3.95

a Expected number of encounters at an animal’s activity center per camera-trapday for camera-traps, and per km walked for scat survey.

In the present combined analysis of a camera-trap and a geneticdata set, both data sets share the movement parameter r and thesex ratio p. The shared model parameters, but also densities, wereestimated with higher precision in the combined model, especiallywhen compared to estimates from the genetic data set only. Forexample, r for females estimated from the genetic data set alonehad a coefficient of variation (CV; SE/mean) of 68%; its CV droppedto 16% in the combined analysis. The parameter r is related to ani-mal movement and larger values of r reflect larger movementranges over the course of the study. Male jaguars tend to have lar-ger home ranges than females (Astete et al., 2008) and accountingfor the resulting larger movements in SCR models is important(Sollmann et al., 2011; Tobler et al., 2013; Tobler and Powell,2013). The extreme uncertainty about this parameter when esti-mated from genetic data alone most likely arises from a lack of spa-tial recaptures of females in this particular data set; thus, it is not aconsequence of the survey type but of the resulting data. Combin-ing data from both surveys, however, allowed us to include sex asan individual covariate and estimate movement parameters forboth sexes with good precision. The results show that male jaguarsin the SCNP have larger movement ranges than females. Further,including sex in the model, we found some evidence that the sexratio of the population was biased towards females (estimates ofthe probability of being a male, p, were less than 0.5, though notsignificantly so; Table 1). These patterns have been observed be-fore in other parts of the species’ distribution (e.g., Sollmannet al., 2011; Tobler et al., 2013).

Baseline camera-trap encounter rates for females were almostdouble that of males, although due to the high uncertainty in fe-male kC

0, this difference was not deemed significant (Table 1). High-er baseline encounter rates from camera-trapping are generallyobserved for male jaguars (e.g., Sollmann et al., 2011; Tobleret al., 2013), possibly because males exhibit higher movementrates. It is unclear what may have caused higher female baselineencounter rates in the present study. Contrary, for the scat surveythe baseline encounter rate of males was considerably higher thanthat of females (95BCI for kS

0 of males did not include the estimatefor females, and vice versa). Male jaguars are thought to demarcatetheir territories and are thus probably more prone to leave theirscats in more conspicuous sites such as roads and trails. Althoughto our knowledge, there are no studies on the differential scatdeposition behavior in male and female jaguars, Palomares et al.(2012) observed high proportions of male feces in several samplesfrom jaguar populations and attributed it to a lower detectability ofscats from females.

Gopalaswamy et al. (2012) combined camera-trapping and scatsurvey data to obtain a single tiger density estimate from both datasets. As in that case both surveys were carried out in close tempo-ral proximity, the population could be assumed to be demograph-ically closed over the course of both surveys. Contrary, in thepresent case, the two surveys were carried out in subsequent years.As a consequence, we let density vary between surveys. This re-flects a simplistic form of an open population model, one that onlyconsiders changes in density without estimating underlying demo-graphic parameters such as survival or recruitment. Although thedensity estimate for 2008 was higher than for 2007, the BCI forboth estimates overlapped considerably, indicating that therewas no significant change in population density between the twoyears. Since both surveys covered essentially the same area withinSCNP (Fig. 1) and the jaguar is a long-lived and slowly reproducingspecies, this outcome was not surprising. Still, our approach high-lights the possibility of monitoring a population using alternatingmethodologies.

Silveira et al. (2009) analyzed the same photographic data setwith non-spatial capture–recapture models. Comparing the pres-ent density estimate for 2007 of 1.57 (±0.43) individuals/100 km2

Page 5: Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar

246 R. Sollmann et al. / Biological Conservation 167 (2013) 242–247

to the non-spatial estimate by Silveira et al. (2009), we see that thespatial model returns a density estimate similar to the non-spatialmodel with the full mean maximum distance moved (MMDM;1.28 ± 0.62 individuals per 100 km2). This confirms findings ofother studies that report the full MMDM to be a better representa-tion of jaguar movement that the half MMDM (e.g., Soisalo andCavalcanti, 2006; Tobler and Powell, 2013). Jaguar density in theSCNP compares with estimates from other parts of the species dis-tribution, which range from under 1 individuals per 100 km2 inareas of central Brazil and the Atlantic Forest (Paviolo et al.,2008; Sollmann et al., 2011) to 7–8 individuals per 100 km2 inthe Brazilian Pantanal (Soisalo and Cavalcanti, 2006) and the Beliz-ean rainforest (Silver et al., 2004). Considering that the semi-aridCaatinga represents somewhat marginal habitat for the jaguar,density estimates in the SCNP are surprisingly high. This may bea result of the presence of permanent artificial waterholes in thePark, in combination with strong and effective poaching controlsthat have been implemented since 2000 (Silveira et al., 2009).These management measures likely helped prey populations to re-cover from previously low numbers (Wolff, 2001), thereby alsobenefitting the local jaguar population. As such, SCNP appears tobe an important regional refuge for the Critically Endangered Caat-inga jaguar.

Although in the present case density estimates were similaracross modeling approaches, a major disadvantage of non-spatialcapture–recapture models, especially when used for wide-rangingspecies, is the need to use ad hoc approaches like the MMDM toestimate the area surveyed. These estimates depend heavily ontrap spacing and the size of the trap array relative to animal homerange size, and ideally, the trap array should cover at least fourtimes the average home range (e.g., Bondrup-Nielsen, 1983; Dillonand Kelly, 2007; Maffei and Noss, 2008). SCR models are muchmore robust to these issues of spatial study design (Sollmannet al., 2012). Our analysis showed that SCR models produced moreprecise density estimates, for the combined data set, but also whenapplied to the camera-trap data set alone: the CV for the non-spa-tial estimate was 37% (Silveira et al., 2009), whereas for the spatialcameras-only estimate is was 32%, and 27% for the combined anal-ysis. Precision in estimating density is important when monitoringpopulations, as it determines our ability to detect changes in thepopulation of interest.

4.1. Conclusions

Camera-traps are a standard tool in monitoring populations ofindividually marked species such as jaguars. With the advent ofspatially explicit capture–recapture (Borchers, 2012; Efford,2004; Royle and Young, 2008) more appropriate models exist forthe analysis of resulting individual detection data. We showed thatthe SCR framework we adopt here results in more precise esti-mates of density than previously published non-spatial estimates(Silveira et al., 2009) for the same data set. Better statistical perfor-mance of SCR over traditional capture–recapture models has beenshown before in a simulation study (Blanc et al., 2013). Moreover,the spatial framework allows combining and sharing parametersbetween different data types, which further improved populationestimates. But advantages of combining different survey types inthe monitoring of populations go beyond better estimates of den-sity. Scat surveys also provide us with information that camera-traps cannot collect. Specifically, dietary analysis of scats showedthat the six-banded armadillo Euphractus sexcinctus and the col-lared anteater Tamandua tetradactyla comprised approximately80% of the jaguar’s diet in SCNP (JCF, unpublished data). Geneticanalyses revealed the signature of a recent population bottleneck,coupled with signs of reduced gene flow between SCNP and theother regions (Roques et al., in preparation). All these pieces of

information are extremely important for our understanding of jag-uar ecology in the Caatinga, an extreme and supposedly untypicalhabitat for a species that is usually associated with rivers or wet-land habitat (e.g., Crawshaw and Quigley, 1991; Emmons, 1987;Mondolfi and Hoogesteijn, 1986), and where the species remainsextremely little studied.

Acknowledgments

We appreciate the support provided by the Fundação Museu doHomem Americano and the park administration of the InstitutoChico Mendes de Conservação da Biodiversidade, which made thisstudy possible. The study was funded by the Jaguar ConservationFund (JCF), The Memphis Zoo, and the projects BIOCON 05-100/06 of Fundación BBVA and CGL2010-16902 of the Spanish Ministryof Science and Innovation. We thank all field assistants for helpwith camera trapping and scat collection and analysis, especiallySamuel Astete, Raphael Lucas de Almeida, Grasiela Porfírio and Tia-go Boscarato.

References

Astete, S., Sollmann, R., Silveira, L., 2008. Comparative ecology of jaguars in Brazil.CAT News Special Issue 4, 9–14.

Blanc, L., Marboutin, E., Gatti, S., Gimenez, O., 2013. Abundance of rare and elusivespecies: Empirical investigation of closed versus spatially explicit capture–recapture models with lynx as a case study. J. Wildl. Manage. 77, 372–378.

Bondrup-Nielsen, S., 1983. Density estimation as a function of live-trapping gridand home range size. Can. J. Zool. 61, 2361–2365.

Boom, R., Sol, C.J., Salimans, M.M., Jansen, C.L., Wertheim-van Dillen, P.M., Van derNoordaa, J., 1990. Rapid and simple method for purification of nucleic acids. J.Clin. Microbiol. 28, 495–503.

Borchers, D., 2012. A non-technical overview of spatially explicit capture–recapturemodels. J. Ornithol. 152, 435–444.

Brodie, J.F., 2009. Is research effort allocated efficiently for conservation? Felidae asa global case study. Biodivers. Conserv. 18, 2927–2939.

Crawshaw Jr., P.G., Quigley, H.B., 1991. Jaguar spacing, activity and habitat use in aseasonally flooded environment in Brazil. J. Zool. 223, 357–370.

Dillon, A., Kelly, M.J., 2007. Ocelot Leopardus pardalis in Belize: the impact of trapspacing and distance moved on density estimates. Oryx 41, 469–477.

Efford, M., 2004. Density estimation in live-trapping studies. Oikos 106, 598–610.Efford, M.G., Dawson, D.K., Robbins, C.S., 2004. DENSITY: software for analysing

capture–recapture data from passive detector arrays. Anim. Biodivers. Conserv.27, 217–228.

Emmons, L.H., 1987. Comparative feeding ecology of felids in a neotropicalrainforest. Behav. Ecol. Sociobiol. 20, 271–283.

Emperaire, L., 1985. A região da Serra da Capivara (Sudeste do Piauí) e suavegetação. Bras. Florestal 14, 5–21.

Frantz, A.C., Pope, L.C., Carpenter, P.J., Roper, T.J., Wilson, G.J., Delahay, R.J., Burke, T.,2003. Reliable microsatellite genotyping of the Eurasian badger (Meles meles)using faecal DNA. Mol. Ecol. 12, 1649–1661.

Gardner, B., Royle, J.A., Wegan, M.T., Rainbolt, R.E., Curtis, P.D., 2010. Estimatingblack bear density using DNA data from hair snares. J. Wildl. Manage. 74, 318–325.

Gardner, B., Royle, J.A., Wegan, M.T., 2009. Hierarchical models for estimatingdensity from DNA mark-recapture studies. Ecology 90, 1106–1115.

Gilks, W.R., Thomas, A., Spiegelhalter, D.J., 1994. A language and program forcomplex bayesian modelling. J. R. Stat. Soc. Ser. D (Stat.) 43, 169–177.

Gopalaswamy, A.M., Royle, J.A., Delampady, M., Nichols, J.D., Karanth, K.U.,Macdonald, D.W., 2012. Density estimation in tiger populations: combininginformation for strong inference. Ecology 93, 1741–1751.

Höss, M., Pääbo, S., 1993. DNA extraction from Pleistocene bones by a silica-basedpurification method. Nucleic Acids Res. 21, 3913–3914.

Karanth, K. Ullas, Nichols, James D., 1998. Estimation of tiger densities in India usingphotographic captures and recaptures. Ecology 79, 2852–2862.

Kohn, M.H., Wayne, R.K., 1997. Facts from feces revisited. Trends Ecol. Evol. 12, 223–227.

Long, R.A., Donovan, T.M., Mackay, P., Zielinski, W.J., Buzas, J.S., 2007. Comparingscat detection dogs, cameras, and hair snares for surveying carnivores. J. Wildl.Manage. 71, 2018–2025.

Maffei, L., Noss, A.J., 2008. How small is too small? Camera trap survey areas anddensity estimates for ocelots in the Bolivian Chaco. Biotropica 40, 71–75.

Menotti-Raymond, M., David, V.A., Lyons, L.A., Schäffer, A.A., Tomlin, J.F., Hutton,M.K., O’Brien, S.J., 1999. A genetic linkage map of microsatellites in theDomestic Cat (Felis catus). Genomics 57, 9–23.

Mohamed, A., Sollmann, R., Bernard, H., Ambu, L.N., Lagan, P., Mannan, S., Hofer, H.,Wilting, A., 2013. Density and habitat use of the leopard cat (Prionailurusbengalensis) in three commercial forest reserves in Sabah, Malaysian Borneo. J.Mammal. 94, 82–89.

Page 6: Combining camera-trapping and noninvasive genetic data in a spatial capture–recapture framework improves density estimates for the jaguar

R. Sollmann et al. / Biological Conservation 167 (2013) 242–247 247

Mondolfi, E., Hoogesteijn, R., 1986. Notes on the biology and status of the jaguar inVenezuela. In: Cats of the World, Biology, Conservation and Management, pp.85–123.

Morrison, J.C., Sechrest, W., Dinerstein, E., Wilcove, D.S., Lamoreux, J.F., 2007.Persistence of large mammal faunas as indicators of global human impacts. J.Mammal. 88, 1363–1380.

Palomares, Francisco, Roques, Séverine, Chávez, C., Silveira, Leandro, Keller, Claudia,Sollmann, Rahel, do Prado, D.M., Torres, P.C., Adrados, Begoña, Godoy, J.A.,Jácomo,, A.T.A., Tôrres, Natália Mundim, Furtado, Mariana Malzoni, López-Bao,J.V., 2012. High proportion of male faeces in jaguar populations. PLoS ONE 7,e52923.

Paula, R., Campos, C., Oliveira, T.G., 2012. Red List assessment for the jaguar in theCaatinga biome. Cat News Special Issue 7, 19–24.

Paviolo, A., De Angelo, C.D., Di Blanco, Y.E., Di Bitetti, M.S., 2008. Jaguar Pantheraonca population decline in the upper Paraná Atlantic forest of Argentina andBrazil. Oryx 42, 554–561.

Pilgrim, K.L., McKelvey, K.S., Riddle, A.E., Schwartz, M.K., 2005. Felid sexidentification based on noninvasive genetic samples. Mol. Ecol. Notes 5, 60–61.

R Development Core Team, 2011. R: A Language and Environment for StatisticalComputing. R Foundation for Statistical Computing, Vienna, Austria.

Reppucci, J., Gardner, Beth., Lucherini, M., 2011. Estimating detection and density ofthe Andean cat in the high Andes. J. Mammal. 92, 140–147.

Roques, S., Adrados, B., Chavez, C., Keller, C., Magnusson, W., Palomares, F., Godoy, J.,2011. Identification of neotropical felid faeces using RCP–PCR. Mol. Ecol. Res. 11,171–175.

Roques, S., Furtado, M.M., Jácomo, A.T.A., Sollmann, R., Tôrres, N.M., Godoy, J.,Palomares, F. Monitoring jaguar populations (Panthera onca) with non-invasivegenetics: a pilot study in Brazilian ecosystems. Oryx (accepted for publication).

Roques, S., Sollmann, R., Jácomo, A.T.A., Tôrres, N.M., Silveira, L., Chavez, C., Keller, C.,Mello do Prado, D., Torres, P.C., Jorge dos Santos, C., Bernardes Garcia da Luz, X.,Magnusson, W.E., Godoy, J.A., Ceballos, G., Palomares, F. Effects of habitatdeterioration on the genetics of wide-ranging carnivores: the case of the jaguarin the Neotropics. In preparation.

Royle, J.A., Young, K.V., 2008. A hierarchical model for spatial capture–recapturedata. Ecology 89, 2281–2289.

Royle, J.Andrew., Dorazio, Robert.M., 2012. Parameter-expanded data augmentationfor Bayesian analysis of capture–recapture models. J. Ornithol. 152, 521–537.

Royle, J.A., Dorazio, R.M., Link, W.A., 2007. Analysis of multinomial models withunknown index using data augmentation. J. Comput. Graph. Stat. 16, 67–85.

Russell, R.E., Royle, J.A., Desimone, R., Schwartz, M.K., Edwards, V.L., Pilgrim, K.P.,Mckelvey, K.S., 2012. Estimating abundance of mountain lions fromunstructured spatial sampling. J. Wildl. Manage. 76, 1551–1561.

Sanderson, E.W., Redford, K.H., Chetkiewicz, C.L., Medellin, R.A., Rabinowitz, A.R.,Robinson, J.G., Taber, A.B., 2002. Planning to save a species: the jaguar as amodel. Conserv. Biol. 16, 58–72.

Schwartz, Michael.K., Luikart, G., Waples, R.S., 2007. Genetic monitoring as apromising tool for conservation and management. Trends Ecol. Evol. 22, 25–33.

Silveira, L., Jácomo, A.T.A., Astete, S., Sollmann, R., Tôrres, N.M., Furtado, M.M.,Marinho-Filho, J., 2009. Density of the near threatened jaguar Panthera onca inthe caatinga of north-eastern Brazil. Oryx 44, 104–109.

Silveira, L., Jácomo, A.T.A., Diniz-Filho, J.A., 2003. Camera trap, line transect censusand track surveys: a comparative evaluation. Biol. Conserv. 114, 351–355.

Silver, S.C., Ostro, L.E., Marsh, L.K., Maffei, L., Noss, A.J., Kelly, M.J., Wallace, R.B.,Gómez, H., Ayala, G., 2004. The use of camera traps for estimating jaguarPanthera onca abundance and density using capture/recapture analysis. Oryx38, 148–154.

SMAPR, 1994. Plano de Manejo Parque Nacional Serra da Capivara. Secretaria doMeio Ambiente da Presidência da República, Brasília, Brazil.

Smith, D.A., Ralls, K., Davenport, B., Adams, B., Maldonado, J.E., 2001. Canineassistants for conservationists. Science 291, 435.

Soisalo, M.K., Cavalcanti, S., 2006. Estimating the density of a jaguar population inthe Brazilian pantanal using camera-traps and capture–recapture sampling incombination with GPS radio-telemetry. Biol. Conserv. 129, 487–496.

Sollmann, R., Furtado, M.M., Gardner, B., Hofer, H., Jácomo, A.T.A., Tôrres, N.M.,Silveira, L., 2011. Improving density estimates for elusive carnivores:accounting for sex-specific detection and movements using spatial capture–recapture models for jaguars in central Brazil. Biol. Conserv. 144, 1017–1024.

Sollmann, R., Gardner, B., Belant, J.L., 2012. How does spatial study design influencedensity estimates from spatial capture–recapture models? PLoS ONE 7, e34575.

Sturtz, S., Ligges, U., Gelman, A., 2005. R2WinBUGS: a package for running WinBUGSfrom R. J. Stat. Softw. 12, 1–16.

Tobler, M.W., Carrillo-Percastegui, S.E., Zúñiga Hartley, A., Powell, G.V.N., 2013. Highjaguar densities and large population sizes in the core habitat of thesouthwestern Amazon. Biol. Conserv. 159, 375–381.

Tobler, M.W., Powell, G.V.N., 2013. Estimating jaguar densities with camera traps:problems with current designs and recommendations for future studies. Biol.Conserv. 159, 109–118.

Waits, L., Paetkau, D., 2005. Noninvasive genetic sampling tools for wildlifebiologists: a review of applications and recommendations for accurate datacollection. J. Wildl. Manage. 69, 1419–1433.

Wallace, R.B., Gomez, H., Ayala, G., Espinoza, F., 2003. Camera trapping for jaguar(Panthera onca) in the Tuichi Valley, Bolivia. Mastozoología Neotrop. 10, 133–139.

Williams, B.K., Nichols, James.D., Conroy, M.J., 2002. Analysis and Management ofAnimal Populations: Modeling, Estimation, and Decision Making. AcademicPress, San Diego, USA and London, UK.

Wilting, A., Mohamed, A., Ambu, L.N., Lagan, P., Mannan, S., Hofer, H., Sollmann, R.,2012. Density of the Vulnerable Sunda clouded leopard Neofelis diardi in twocommercial forest reserves in Sabah, Malaysian Borneo. Oryx 46, 423–426.

Wolff, F., 2001. Vertebrate Ecology in Caatinga: A. Distribution of Wildlife inRelation to Water. B. Diet of Pumas (P. concolor) and Relative Abundance ofFelids. (Master’s Thesis). Univerist of Missouri, St. Louis, USA.

Zeller, K., 2007. Jaguars in the New Millennium Data Set Update: The State of theJaguar in 2006. Wildlife Conservation Society, Bronx, New York.