Leopard (Panthera pardus) occupancy in the Chure range of Nepal

Abstract Conservation of large carnivores such as leopards requires large and interconnected habitats. Despite the wide geographic range of the leopard globally, only 17% of their habitat is within protected areas. Leopards are widely distributed in Nepal, but their population status and occupancy are poorly understood. We carried out the sign‐based leopard occupancy survey across the entire Chure range (~19,000 km2) to understand the habitat occupancy along with the covariates affecting their occupancy. Leopard signs were obtained from in 70 out of 223 grids surveyed, with a naïve leopard occupancy of 0.31. The model‐averaged leopard occupancy was estimated to be 0.5732 (SE 0.0082) with a replication‐level detection probability of 0.2554 (SE 0.1142). The top model shows the additive effect of wild boar, ruggedness, presence of livestock, and human population density positively affecting the leopard occupancy. The detection probability of leopard was higher outside the protected areas, less in the high NDVI (normalized difference vegetation index) areas, and higher in the areas with livestock presence. The presence of wild boar was strong predictor of leopard occupancy followed by the presence of livestock, ruggedness, and human population density. Leopard occupancy was higher in west Chure (0.70 ± SE 0.047) having five protected areas compared with east Chure (0.46 ± SE 0.043) with no protected areas. Protected areas and prey species had positive influence on leopard occupancy in west Chure range. Similarly in the east Chure, the leopard occupancy increased with prey, NDVI, and terrain ruggedness. Enhanced law enforcement and mass awareness activities are necessary to reduce poaching/killing of wild ungulates and leopards in the Chure range to increase leopard occupancy. In addition, maintaining the sufficient natural prey base can contribute to minimize the livestock depredation and hence decrease the human–leopard conflict in the Chure range.

The leopard habitat outside protected areas is rapidly declining, and within Pas, they face exploitative and interference competition with the socially dominant large carnivores such as tigers (Panthera tigris) and lions (Panthera leo) in most of their distribution range (Barber-Meyer et al., 2013;McDougal, 1988;Miller et al., 2018;Miquelle et al., 2005;Seidensticker, 1976;Seidensticker et al., 1990).
In Southern lowlands and Himalayan foothills of Nepal, the leopards coexist with tigers in the National Parks and Buffer Zone areas (DNPWC & DoFSC, 2018;Subedi, Bhattarai, et al., 2021;. Further, a recent camera trap study in the Chure range detected tigers in Kapilvastu, Palpa, and Rupandehi districts between Chitwan National Park and Banke National Parks (Subedi, Bhattarai, et al., 2021;. The tiger populations in Nepal have almost doubled since 2010 through tiger-focused conservation activities in and around the tiger bearing PAs (DNPWC & DFSC, 2018;. Thus, the increasing number of tigers may have pushed leopards to marginal habitats with some resource overlapping (Kafley et al., 2019;. A large part of the Chure range falls outside the PAs. The forested areas of the Chure range adjoining the PAs provide habitat for dispersing wildlife population including the leopards ( Figure 1). Tigers are primarily confined to protected areas and connected forest patches, and a large part of Chure is unoccupied by them. Thus, the Chure forest provides an opportunity for leopards to occupy a large area as an apex predator (Thapa & Kelly, 2016, 2017Thapa et al., 2021).
Although Chure range has a potential of being key wildlife habitat for leopards and other associated wildlife, with increasing human pressure, the fragile Chure range has high deforestation rate (FRA/ DFRS, 2014) which will affect in the abundance and distribution of wildlife (GoN-RCTM, 2017). In addition, there is no comprehensive study on the status and distribution of wildlife in the Chure range.
We carried out this study as a part of faunal diversity assessment in forests of Chure range (~70% of the total Chure range) of Nepal to understand the distribution and occupancy of leopards. This study provides information on leopard occupancy and associated covariates in Chure range of Nepal with far-reaching implications for the conservation of leopards in the human-dominated landscapes of Nepal and elsewhere.

| Study area
The study was carried out in the Chure range (18,982 km 2 ) of Nepal.
Chure is the young mountain range consisting of fragile sedimentary rocks such as mudstones, shale, sandstones, siltstones, and conglomerates (Pokhrel, 2013). It extends from east to west in southern Nepal spread in all the seven provinces ( Figure 1). Chure has monsoon-dominated subtropical climate. The average maximum and minimum temperature of this range lies between 15.8 and 31.8°C. The mean annual precipitation is between 1,400 mm and 2,000 mm (FRA/DFRS, 2014;GoN-RCTM, 2017). The Chure range has highly rugged terrain, and the altitudinal variation ranges from 120 to ~2,000m. Over 160 river systems with a different origin flow through this range (Chaudhary & Subedi, 2019;FRA/DFRS, 2014;GoN-RCTM, 2017).
A large part of the Chure range (>70%) is forested and is the potential habitat for various wildlife such as leopards. The range consists of 23.4% of the forests nationally and 3.5% of other woodland covers of Nepal (FRA/DFRS, 2014 Chure is the home for 14% of Nepal's human population, and only 14% of the Chure area is suitable for cultivation (SAWTEE, 2016).
The majority of the people depend on subsistence farming for food crops, and animal husbandry is an integral part of their farm. Livestock grazing is widespread across the Chure forests.

| Study design
The Chure range was divided into 4 blocks (size ~2,200-6,400 km 2 ) for easy organization of the survey. Each block was further divided into grids of size 10 × 10 km 2 and surveyed in two to three shifts successively. We chose 10 × 10 km 2 grid size because it was larger than the home range size of leopards, that is, 6-90 km 2 in lowland Nepal and similar habitats (Norton & Henley, 1987;Odden & Wegge, 2005;Seidensticker, 1976;Simcharoen et al., 2008). We sampled the entire Chure, and thus, results reflect the true occupancy, that is, the proportion of area occupied by leopard at landscape level Thapa et al., 2021). Biologists and wildlife technicians (n = 12) with over 5 years of field experience in wildlife research conducted the survey in the field. The survey team was trained on survey protocols and wildlife sign identification before starting the survey to ensure the quality of the data. Out of 322 grids cells in the entire Chure range, 223 were surveyed which falls in the forested areas. The rest of the grids (n = 109), which either fall entirely outside of the forests or was inaccessible due to undulating steep rugged terrain, were omitted from our study. Each grid was further divided into 16 subgrids of 2.5 × 2.5 km 2 (n = 3,568) for the uniformity to search the presence of leopard sign and associated covariates influencing their occupancy and detection. The survey was conducted between 2016 and 2018. We could not cover the entire Chure range in a single year due to the large area and limited human resources available. We carried out the survey in the same season F I G U R E 1 Chure range: divided into four blocks which were further divided into 10 km × 10 km grids. Each color represents each block and boundary of Chure. The blocks are in the order (from east to west): eastern block (yellow color), central block (light blue color), west block (pink color), and far-west block (blue color). In our analysis, the Chure range was also divided into east Chure (includes eastern block) and west Chure (includes rest three blocks west from the eastern block). The lowland protected areas are shown, which is five in the west Chure and one in the east Chure. Kathmandu is the capital of Nepal, and others are the major cities in the lowland of Nepal (postmonsoon) to avoid the potential bias from surveys in different years.
A 2-km-long continuous random walking transect (defined as search paths; Thapa et al., 2021) with four segments of 500 m was surveyed within a subgrid, with maximum of 32-km search paths within each grid; that is, the encounter occasions limit to 16 spatial replicates of 2 km each. However, these expected survey efforts within each grid differ from actual survey effort in the field due to logistical constraints (Harihar & Pandav, 2012). We targeted the existing trails and dirt roads (where possible) to minimize the likelihood of false absences. We recorded the presence/absence of the tracks, fresh droppings, and other signs (feeding sign, territory marking, etc.) to detect the presence of leopards, tigers, and large (>55 kg), medium (20-55 kg), and small-sized prey species (<20 kg)  at each segment in the standard data format as sample covariates. The leopard pugmark was differentiated from tiger from their smaller sizes such as pad size width (<6.5 cm, tiger = 9-10 cm), front foot width (~9 cm leopards, tiger = 12-14 cm), adult stride length (~90 cm, tiger = >100 cm), and claw-scraping (<25 cm height and <15 cm width; tiger = >35 cm height and >19 cm width). Further, the tiger scat diameter is >2.5 cm (Reddy et al., 2004) and has a lower degree of coiling and a relatively larger gap between two successive constrictions (Andheria et al., 2007;Biswas & Sankar, 2002;Wang & Macdonald, 2009). The prey species were identified through their pellets, and track shape and size. The track size of rhesus (circular forehands ~6cm, elongated hind tracks ~6.5-8 cm), spotted deer (length of male = 5-6.6 cm, female = 3.5-5cm, width = 3.8-4.5 cm, elongated track), barking deer (3.5-4.9 cm length, 3-3.7 cm width, sharp edges cutting deep into the soil), wild boar (5.0-6.5 cm length, two dewclaws may mark on the soil and their anterior section marks deeply), goats (4.5-5.8 cm length, 4.8 cm width), cows (<10 cm length, outer hoof surface is well marked), and buffalo (10-12 cm length, front hoof section marks deep were referred from Menon and Daniel (2009) and Kolipaka (2014)). Similarly, the human pressure as lopping, encroachment, and livestock presence was recorded in each segment.

| Occupancy modeling
The naïve occupancy was calculated by dividing the number of grids with species present/total number of grids surveyed in the block.
We used program PRESENCE (2.12.33) to obtain the true occupancy of leopards of Chure range (MacKenzie et al., 2002). We applied the single species occupancy model with correlated replicate surveys, which explicitly take into account the spatial correlation in detection across the 2-km continuous random walking transect (search paths) within each grid. It is because the leopards can travel greater than the size of our replicate (2 km) per day; hence, the detection of the sign in successive spatial replicates violates the statistical independence required by the standard occupancy model (MacKenzie et al., 2017). The spatial correlation model (Hines et al., 2010) accounts for this correlation in the detection using the Markov spatial dependence approach. For the degree of dependence between the replicated samples, the model uses replicate-level occupancy parameters "θ 0 " and θ 1 , where "θ 0 " = Pr (leopard presence in a replicate/grid occupied and which was absent in the previous replicate) and "θ 1 " = Pr (leopard presence in a replicate/grid occupied and was present in the previous replicate). We also checked the performance of the standard occupancy model (MacKenzie et al., 2002) and spatial correlation model (Hines et al., 2010) without adding any covariates in our data. We compared these models based on the Akaike information criterion (AIC) and chose one with lowest AIC scores (Burnham & Anderson, 2002). It clearly showed the spatial dependencies in sign detection on 2-km long replicates with less AIC value (better performance) for the spatial correlation model compared with the TA B L E 1 Model selection between spatial correlation and standard occupancy model standard occupancy model (Table 1). Hence, all other analyses were performed using spatial correlation model (Hines et al., 2010).
Next, we identified suitable sample and site covariates that could potentially explain any heterogeneity in leopard occupancy. For this, we predicted the effect of covariates in detectability and occupancy of leopards. We priori expected that the prey abundance and human disturbance (lopping and human encroachment) across the grid influence the leopard occupancy positively and negatively, respectively (Harihar & Pandav, 2012;Jhala et al., 2010;Karanth et al., 2011).
Further, we expected the livestock negatively influences the leopard occupancy as it can be considered as a substitute for human impact . Similarly, the increased human population density across the grid raises the human disturbance and hence has negative influence on leopard occupancy. Likewise, we expected the positive influence of management regime on leopard occupancy and detection as the survey grid inside the protected areas has lower disturbance compared with outside. Also, we predicted normalized difference vegetative index (site covariate, NDVI) positively influences the occupancy by providing cover and increasing opportunity for leopard, an ambush hunter, to hunt (Sharma et al., 2015), and negatively influence the detection (thick vegetation and leaf litter reduce the chances of sign detection or direct observation of leopard on the search path). Similarly, the terrain ruggedness positively influences both the occupancy and the detection of leopards as increased ruggedness will be harder for people to access and hence lowers the disturbance (Johnson et al., 2020). We also expected sampling effort (total km of search path in a grid) positively influences the detection of leopards as it may vary between the survey grids due to logistic constraint (Harihar & Pandav, 2012). We prepared a list of nine a priori hypotheses (Appendix 6).
The sample covariates collected from the field survey included prey species, PS = (barking deer, wild boar, chital, and rhesus), human disturbance (HD = lopping, human encroachment), and livestock presence (L). We separated the wild boar (W) from other prey species because many studies reported leopards avoiding the wild boar (Karanth & Sunquist, 1995;Ramakrishnan et al., 1999), and we wanted to know how wild boar affects the presence of a leopard.
Moreover, the occurrence of wild boar was the most widespread among the prey species.
The site covariates were management regime (IO = inside or outside of the national park), vegetation cover measured as NDVI-Normalized Difference Vegetation Index (N), terrain ruggedness index (R), and human population density (PD). If a grid falls more than half inside the national park or buffer zone, it was coded as  (Fujisada et al., 2005). We averaged them across the grid surveyed using the z-statistic in ArcGIS 10.1. We also included sampling effort (Samp_Eff) as a covariate that affects the detection probability. Before adding the covariates in our analysis, we tested the Spearman correlation coefficient (r) using PAST version (4.0) (Hammer et al., 2001) and one was dropped when a set of two covariates have |r| ≥ .7. Among the covariates we used, human disturbance (lopping and encroachment) and livestock were highly correlated (Appendix 7) and we used livestock to obtain the final model (Kandel et al., 2020;Kshettry et al., 2018;Reynaert, 2018).
The data were prepared in an excel sheet via creating detection history for the leopard and their prey and livestock detection across all the grids, having 16 replicates each. On each replicate, the detection of the species was coded 1 and nondetection was coded 0. The site covariates were constant in each grid, and we applied z-transformation to normalize the site covariate data. We defined the global model as follows: We identified the suitable covariates on the basis of ecological importance, a recommendation from previous studies, and simplest explanation of model (parsimony). We used a constant model for replicate-level occupancy parameters (θ 0 and θ 1 ) .
We also could not ignore the possibilities that some of the co-  Figure 1). In the Eastern Block ("east Chure" hereafter, number of PAs = 0, n = 71 grids in Chure; Figure 1), a small protected area named Koshi Tappu Wildlife Reserve (KTWR) occurs with a small portion of its northwest boundary touched to Chure range but not included in the survey grids (Figure 1; DNPWC, 2021). These PAs of west Chure bear the leopard source population, and we assumed that the leopard's occupancy is higher compared with the east Chure. Hence, we also separately estimated the leopard occupancy for the east Chure and west Chure. All the covariates described above were used, except the management regime was dropped in the east Chure as no survey grid falls inside the PA., and tiger presence was added in the west Chure. The tigers occupy the protected areas (and some forests outside) of the west Chure (Eisenberg & Lockhart, 1972;Hayward et al., 2006;Pokheral & Wegge, 2019;Ramakrishnan et al., 1999). We followed all the steps and methods as described above (Appendix 8 and Appendix 9 for correlation coefficient "r" between covariates of east Chure and west Chure). In the east, there was no spatial correlation in detection, while we checked the performance of the standard occupancy model (MacKenzie et al., 2002) over spatial correlation model (Hines et al., 2010). So, all the analysis was performed using standard occupancy model, whereas in the west Chure, spatial correlation model performed better over standard occupancy model (Table 1) and hence was used for the further analysis.  Table 2). The terrain ruggedness and the sampling effort did not influence on the leopard detection. Then, we fixed the top detection TA B L E 2 Role of covariates in determining detection probability of leopard sign (Pt) on 2 km long replicates, based on covariates for probability of occurrence of leopard from the global model,  (Table 6).

| D ISCUSS I ON
This is the first comprehensive survey of leopard occupancy covering the entire Chure range (~19,000 km 2 ) of Nepal. We found the spatial replicate model performed better than the standard occupancy model. Our result showed that more than half of the Chure range was occupied by leopards. Leopard occupancy was higher in the west Chure (0.7) compared with East ( Similarly, all five national parks are the home for tiger, the apex carnivore, but the leopard occupancy in the west Chure range, inside TA B L E 4 Role of covariates in determining probability of leopard occupancy in the Chure range, structured on Pt obtained from Table 2 Model AIC ΔAIC w Model Likelihood K Ψ(WB+PD+R+L) θ 0 (·) θ 1 (·)p(IO+N+L), θ 0 pi(·) 858.33  (Table 2) was modeled as p(IO+N+L); + = covariates modeled additively; (·) = parameters are held constant. Our results did not correspond to our a priori hypothesis that leopard avoids wild boar (Eisenberg & Lockhart, 1972;Hayward et al., 2006;Pokheral & Wegge, 2019;Ramakrishnan et al., 1999)  Our results showed the importance of wild boar as prey species in areas with low prey density for the occurrence of leopard (Figure 3).

TA B L E 5 Model-specific β coefficient estimates for covariates determining leopard occupancy in the Chure range
We also used other prey species (barking deer, rhesus, and chital) as covariates, but their influence in the model was weak. We believe the rarity of prey other than wild boar in the Chure range is the reason for such results in contrast to our expectation of strong relation between predators (leopard) and prey .
The opportunistically placed camera traps along with this survey also photographed poachers with guns in various locations. It indicates the widespread hunting of wild prey species (Subedi, Bhattarai, et al., 2021) which have probably contributed to reducing the prey abundance.
The positive influence of the ruggedness index on leopard occupancy of Chure range indicates the extensive use of rugged Chure hills by leopards. The rugged terrain provides an opportunity for ambush predators to hunt (Sharma et al., 2015). Leopards are excellent climbers, and rugged terrain probably does not limit their movements/use of the habitat. Generally flat and less rugged areas are occupied by human settlements, and the rugged hills are still covered with forest providing habitat for leopards, their prey, and other wildlife. However, we did not find the relation between vegetation cover (NDVI) and leopard occupancy. Instead, as our a priori assumption, the detection probability was inversely related to NDVI as the survey was conducted in the postmonsoon season, the time the leaves start shedding from the deciduous trees. These fallen leaves covering the forest floor reduce the chances of detecting the leopard sign in densely vegetated areas. In intact forests (high NDVI value) generally, there are fewer and less visible animal trails.
Detecting the leopard sign in such a forest is comparatively difficult which reduces the detection probability. Similarly, the detection of the leopard sign was higher in the Chure range that falls outside the protected areas. It may be because the vegetation cover (NDVI) inside the national park is high in comparison with the outside area, and NDVI has negatively influenced the leopard detection of Chure range (P N = −1.29, SE 0.43; Krishna et al., 2008).
We found the positive influence of human population density and livestock on leopard occupancy, oppose to our prediction. The majority of the Nepalese rural community is based on agriculture, and livestock is an integral part of their farm . Livestock was present in ~55% of the surveyed grid and leopard occurred in 19% of the grids with livestock presence.
Leopard can persist in highly modified landscape with high human population density (Athreya et al., 2013(Athreya et al., , 2016Kuhn, 2014). They adopt different ways to minimize the landscape of fear arose with direct contact with humans in the high human-disturbed areas (Kerley et al., 2002). Hence, this positive association should not be taken as coexistence but manifestation of high nexus between the animal, habitat, and communities as present in our agrarian society in the landscape, thus prevailing chances of human-wildlife conflict in the Chure range.
Leopards are specialized solitary hunters primarily hunting wild ungulates, but also kill livestock if opportunity arises (Kandel et al., 2020;Treves & Karanth, 2003). In the presence of the sufficient natural prey base, leopards tend to avoid livestock (Kolowski & Holekamp, 2006). We do not have the data on the density of prey in the Chure range but the low detection of prey signs (except the wild boar) indicates their low abundance (Smallwood & Fitzhugh, 1995;Stander, 1998). In the absence of enough wild prey, leopards shift to livestock for diet (Hussain et al., 2019;Khorozyan et al., 2015).
Different studies have shown livestock contribution in leopard's diet (Aryal & Kreigenhofer, 2009;Deo, 2014;Harihar et al., 2011;Hussain et al., 2019), and in human-use landscape, the livestock biomass contribution was even high (Kshettry et al., 2018). Further, the percentage of livestock consumption was high in leopard's diet compared with tigers, where the detection of leopard was positively influenced by livestock . Also,

| CON CLUS ION
More than half of the Chure range is occupied by leopards. We Sign-based occupancy survey can efficiently assess the spatial distribution of large carnivores such as leopards, providing the direction and effect of covariates governing their presence. Hence, we recommend carrying out the occupancy survey every 5 years across the leopard habitats to understand their status as done for tigers in TAL . In future research, the exploration of the livestock depredation and human-leopard conflict data, assessing prey density in leopard habitat via distance sampling or using camera traps (since camera traps capture poachers and also are used to estimate relative prey abundance), and assessing leopard reproductive success and survival/mortality rate inside and outside of the PAs add value to understanding the dynamics of the conflict.

ACK N OWLED G M ENTS
The field survey was organized by the National Trust for Nature

CO N FLI C T O F I NTE R E S T
No conflict of interest.

A PPE N D I X 2
Role of covariates in determining detection probability of leopard sign (Pt) on 2-km-long replicates of west Chure, based on covariates for probability of occurrence of leopard from the global model, 622.55 7.75 0.011 0.0208 31 622.77 7.97 0.0099 0.0186 30 625.94 11.14 0.002 0.0038 30

APPENDIX 3
Role of covariates in determining probability of leopard occupancy in the east Chure range, structured on Pt obtained from Appendix 1

A PPE N D I X 6
Definition and predicted effect of covariates in detectability and occupancy of leopard of Chure range

Prey abundance
The relative abundance of prey species (barking deer, chital, wild boar, rhesus) across the 2-km continuous random transect Continuous Positive (Ψ, known to be the function of carnivore densities, Karanth et al., 2011) In case of wild boar, we separated it from rest prey as it is often avoided by predators for its aggressive behavior, so we expected negative effect on Ψ

Tiger
The relative abundance of tiger in the west Chure range  (Sharma et al., 2015), negative "p" (our study period was postmonsoon, and the leaves shading from the deciduous tree during this season reduces the chances of sign detection) Arc GIS 10.1 Terrain Ruggedness (R) Averaged across grid cell. Calculated using 90 m ASTER DEM.

Continuous
Positive for Ψ and "p" (increased ruggedness will be harder for people to access and disturbance will be lower)

Continuous
Positive for "p" (survey effort varied from grid to grid due to logistical constraint (Harihar & Pandav, 2012) A PPE N D I X 7 Correlation coefficient (r) value between the covariates in the Chure range. It was calculated using PAST version (4.0). When a set of two covariates have |r| ≥ .7, one was dropped from the analysis. The correlation coefficient between human disturbance (HD) and livestock presence (L) was >0.7 (bold), so HD was dropped off. Other covariates: Samp_Eff=sample effort, IO = management regime (inside and outside of PA), R = ruggedness, PD = human population density, Leo = leopard, PS = prey species, WB = wild boar

A PPE N D I X 8
Correlation coefficient (r) value between the covariates in the east Chure range. It was calculated using PAST version (4.0). When a set of two covariates have |r| ≥ .7, one was dropped from the analysis. The correlation coefficient between human disturbance (HD) and livestock presence (L) was >0.7 (bold), so HD was dropped off. Other covariates: Samp_Eff = sample effort, R = ruggedness, N = NDVI, PD = human population density, Leo = leopard, PS = prey species, WB = wild boar

A PPE N D I X 9
Correlation coefficient (r) value between the covariates in the west Chure range. It was calculated using PAST version (4.0). When a set of two covariates have |r| ≥ .7, one was dropped from the analysis. The correlation coefficient between human disturbance (HD) and livestock presence (L) was >0.7 (bold), so HD was dropped off. Other covariates: Samp_Eff = sample effort, IO = management regime (inside and outside of PA), R = ruggedness, N = NDVI, PD = human population density, Leo = leopard, PS = prey species, WB = wild boar, T = tiger