Relative influence of wild prey and livestock abundance on carnivore‐caused livestock predation

Abstract Conservation conflict over livestock depredation is one of the key drivers of large mammalian carnivore declines worldwide. Mitigating this conflict requires strategies informed by reliable knowledge of factors influencing livestock depredation. Wild prey and livestock abundance are critical factors influencing the extent of livestock depredation. We compared whether the extent of livestock predation by snow leopards Panthera uncia differed in relation to densities of wild prey, livestock, and snow leopards at two sites in Shey Phoksundo National Park, Nepal. We used camera trap‐based spatially explicit capture–recapture models to estimate snow leopard density; double‐observer surveys to estimate the density of their main prey species, the blue sheep Pseudois nayaur; and interview‐based household surveys to estimate livestock population and number of livestock killed by snow leopards. The proportion of livestock lost per household was seven times higher in Upper Dolpa, the site which had higher snow leopard density (2.51 snow leopards per 100 km2) and higher livestock density (17.21 livestock per km2) compared to Lower Dolpa (1.21 snow leopards per 100 km2; 4.5 livestock per km2). The wild prey density was similar across the two sites (1.81 and 1.57 animals per km2 in Upper and Lower Dolpa, respectively). Our results suggest that livestock depredation level may largely be determined by the abundances of the snow leopards and livestock and predation levels on livestock can vary even at similar levels of wild prey density. In large parts of the snow leopard range, livestock production is indispensable to local livelihoods and livestock population is expected to increase to meet the demand of cashmere. Hence, we recommend that any efforts to increase livestock populations or conservation initiatives aimed at recovering or increasing snow leopard population be accompanied by better herding practices (e.g., predator‐proof corrals) to protect livestock from snow leopard.


| INTRODUC TI ON
Large mammalian carnivores are among the most threatened group of species with over 60% of them facing high risk of extinction (Ripple et al., 2014). While habitat loss, fragmentation, poaching, and prey depletion continue to cause their populations to decline (Cardillo et al., 2004;Chapron et al., 2008;Wolf & Ripple, 2016), retaliatory killing over livestock predation is perhaps the most widespread and direct threat to their conservation (Inskip et al., 2014).
Livestock depredation compromises the livelihoods of often marginalized communities and erodes human tolerance of carnivores (Inskip et al., 2016;Mishra, 1997;Thirgood et al., 2005;Treves & Karanth, 2003). It is therefore critical to mitigate the conflicts surrounding livestock depredation for ensuring sustainable livestock production by pastoral communities and continued survival of carnivore populations, especially for wide-ranging species that occur outside protected areas. Reducing livestock depredation by carnivores requires an understanding of factors affecting their predation behavior.
Density of wild herbivore prey is known to be a critical determinant of carnivore density (Carbone & Gittleman, 2002;Karanth et al., 2004;Suryawanshi et al., 2017). However, the role of wild herbivore density in determining the extent of livestock predation by carnivores is debatable (Bagchi & Mishra, 2006;Khorozyan et al., 2015;Meriggi & Lovari, 1996;Soofi et al., 2019;Suryawanshi et al., 2017). Studies investigating the impact of livestock abundance on predation levels have shown higher intensities of depredation in areas of higher livestock densities (Pimenta et al., 2018). Despite a range of field studies and reviews examining patterns of livestock depredation by large carnivores (van Eeden et al., 2018;Inskip & Zimmermann, 2009;Janeiro-Otero et al., 2020;Weise et al., 2018), our knowledge of the relative impact of wild prey and livestock abundance on livestock predation is still limited.
The snow leopard Panthera uncia is listed as Vulnerable in the IUCN Red list of Threatened Species and occurs in 12 countries across the Himalaya and high mountains of Central Asia (McCarthy et al., 2017). Fewer than 4,000 adult snow leopards are believed to occur in the wild and little is known about their population trends (Snow Leopard Working Secretariat, 2013;Suryawanshi et al., 2019). Pastoralism is the dominant form of land use and economy across the snow leopard distribution range in Central and South Asia (Mishra et al., 2009). As its distribution range overlaps extensively with the pastoral production landscapes, livestock predation by snow leopard is ubiquitous and is of high concern for pastoral communities. Among other factors such as habitat loss and decline of prey, livestock depredation has been the key factor driving its endangerment through retaliatory killings while also imposing significant economic costs on marginalized herder communities (Aryal et al., 2014;Hussain, 2003;Ikeda, 2004;Johansson et al., 2015;Li et al., 2013;Mishra, 1997).
Herders have been found to incur high losses, up to 12% of their livestock holdings annually, to snow leopard and sympatric predators, which sometimes amounts up to 50% of the average annual household income (Mishra, 1997;Oli et al., 1994). Livestock loss causes serious hostility among herder communities, often resulting in persecution of the snow leopard (Oli et al., 1994).
A few studies that have tried to identify the causes of livestock predation by snow leopards have found a range of factors influencing the extent of livestock predation such as wild prey density, livestock density, herding practices, and habitat type Bagchi & Mishra, 2006;Bagchi et al., 2020;Chetri et al., 2017Chetri et al., , 2019Jackson et al., 1996;Rashid et al., 2020;Suryawanshi et al., 2013Suryawanshi et al., , 2017. These studies have generated an understanding of spatial and temporal patterns of livestock predation, providing insights into the location and season for prioritizing mitigation measures. Suryawanshi et al. (2017) showed that the extent of predation on livestock could increase with livestock density as well as wild prey density via increased snow leopard density. However, relative influence of wild prey and livestock density on livestock predation by snow leopards is still unclear as livestock predation can often be context-dependent with site specific idiosyncrasies in habitat characteristics and management interventions (Chetri et al., 2019).
The Himalayan and trans-Himalayan habitats of Shey Phoksundo National Park, Nepal, which vary in habitat characteristics and livestock management practices offer an interesting opportunity to examine such effects. Here, we report on the extent of snow leopard predation on livestock in relation to wild prey and livestock density in two sites, Upper Dolpa and Lower Dolpa, that are broadly characterized by the trans-Himalayan and Himalayan habitats, respectively.
We also examined the factors influencing household level variation in livestock depredation by snow leopard.

| Study area
This study was conducted at two sites, the Lower Dolpa and the Upper Dolpa in Shey Phoksundo National Park, Nepal (29°15′-29°45′ N and 83°08′-83°31′ E; Figure 1). Located in western part of Nepal, the park covers an area of 3,555 km 2 with elevation ranging from 2,130 m in Ankhe to 6,883 m at the summit of Kanjirowa Mountain.
The park contains the transition from a monsoon dominated climate with 1,500 mm of annual precipitation in the south (Suligad) to an arid climate with less than 500 mm a year in the northern slopes.

| Camera trapping survey of snow leopard
We  Table 2. To ensure systematic placement of camera traps, we overlaid the study area with 4 × 4 km 2 square grid cells for camera trap deployment. This grid cell size was chosen to be small enough to avoid holes for snow leopards to go undetected and large enough to have multiple spatial recaptures of individuals. Before deploying camera traps, areas with higher probability of encountering snow leopard signs such as human trails, welldefined and narrow ridgelines, valley bottoms, scent marking sites or immediately adjacent to frequently scent-sprayed rocks and scrapes were identified through preliminary sign surveys (Jackson et al., 2006).
This method is analogous to two sample capture-mark-recapture (CMR) technique of animal abundance estimation (Williams et al., 2002).
The logic is that individual animals are difficult to uniquely identify and mark in mountain ungulate species but groups or herds can be identified uniquely based on group characteristics, sighting location and time etc. Hence, traditional CMR can be still used at the group level, the unit of analysis being the group or herd. In this method, the study area is divided into smaller blocks where surveys (combination of trails and observation points) can be done ensuring complete visual coverage.
Two observers or two teams of observers independently search and count the herds and numbers of animals along specific trails or routes separated by about 15-20 min. Both observer teams record sufficient information on each of the ungulate sighting (e.g., herd size, geographic location, time of the sighting, distance to the herd location, age-sex composition of the herd) to allow them to later identify the common (recaptured herds) and unique herds. The key assumptions of this form of double-observer survey are that complete visual coverage of survey block is possible, common groups are not misidentified, and there is no group fission or fusion during the duration of the two surveys (Suryawanshi et al., 2012).
To conduct the field surveys, we first mapped 5-7 blocks of 30-50 km 2 in both Upper and Lower Dolpa. These blocks were delimited using geographic features such as rivers, ridgelines, and watershed boundary. Human-used trails, valley bottoms, and ridgelines were mapped as potential survey routes within these sub-blocks.
Two observers equipped with either binoculars or spotting scope then conducted double-observer surveys along these routes spacing themselves by 15-20 min. Information on herd size, age-sex composition of the herd, sighting time and location and any particular characteristics and composition of the herd (e.g., male only groups) were used for verifying unique and common (recapture) herds and to avoid double counting. Surveys were conducted between 19 February 2018 and 26 April 2018 when blue sheep movement between adjacent blocks was generally low due to high accumulation of snow.

| Livestock depredation surveys
We conducted household level interviews using semi-structured questionnaire forms to gather information on species wise livestock holdings, grazing practices and livestock lost to snow leopards.
A snowball sampling approach was used to sample households, where a respondent was asked to introduce another respondent (Goodman, 1961

| Snow leopard population density
We used the maximum likelihood based spatially explicit capturerecapture (SECR) approach to analyze the spatial capture recapture data of snow leopards obtained from camera traps for density estimation. SECR is a spatially explicit hierarchical modeling process, which combines a state model and an observation model. on their pelage, forehead, and tail pattern Jackson et al., 2006). At least two observ- Each unique individual snow leopard from the field dataset was checked for recapture at the same station and at the other stations and across sites. Each day (24-hr period) was considered as a unique sampling occasion (Karanth & Nichols, 1998). Spatial capture-recapture history of each individual snow leopard was prepared for the first 90 sampling occasions and included for analysis in order to try and meet the closure assumption.
For each study site, data were analyzed separately to estimate the parameters of interest: density (D), capture probability of an individual snow leopard at its activity center (g0), and the spatial scale over which detection probability declines as the distance between an individual's activity center to the camera trap station increases (σ). Variability in sampling effort may negatively bias density estimates and reduce the ability to explain variation in detection probability, so we accounted for variable sampling effort by using the number of days each sampling detector was active (Efford, 2018 (Burnham & Anderson, 2002) and used the best supported model to make inference on density estimates (Tables S1a & S1b). We did not use any site covariates to model spatial variation in density since our primary goal was to obtain robust estimate of density for the two sites rather than spatial variation within them.

| Wild prey abundance and density
The total number of blue sheep herds was estimated using the Chapman's bias-corrected estimator (Chapman, 1951

| Livestock density and livestock predation
Livestock density was calculated by dividing the total livestock population by the total livestock grazing area, which was mapped in field with the help of local herders.

| Snow leopard density
Overall For both sites, the best model based on AICc included a constant capture probability (g0) and movement parameter (σ) with half-normal detection as a detection function (

| Livestock population, density, and depredation
The total livestock population of Upper Dolpa was approximately 13,000 and Lower Dolpa was approximately 1,600 (Table 1).
Livestock density was almost fourfold higher in Upper Dolpa (17.21 per square kilometer) as compared to Lower Dolpa (4.51 animals per km 2 ) (

| Relationship between snow leopard, wild prey and livestock density
The blue sheep density was similar in both sites. The 95% confidence intervals around density estimates for the two sites overlapped. The estimated snow leopard density, livestock number, livestock density, and average livestock holding were all higher in Upper Dolpa in comparison to Lower Dolpa (Figure 2). The proportion of livestock loss per household (total livestock lost to snow leopards divided by total livestock holding owned by a respondent household) and proportion of total livestock holding lost (total livestock lost by all respondents divided by the sum total of livestock holdings of all respondents) in a year was significantly higher in Upper Dolpa (Table 4, Figure 2).

| Factors influencing household level variation in livestock depredation
Respondents who owned higher number of livestock were likely to incur more livestock loss to snow leopard (β = 0.45 ± SE 0.03). in livestock stocking density and also with an increase in wild prey via increased abundance of snow leopards (Chetri et al., 2019;Suryawanshi et al., 2013).
We found household who owned more livestock and a greater proportion of small bodied livestock to lose higher number of livestock to snow leopard predation. Large herd size with higher number of small bodied livestock may be easier to detect and prey upon (Chetri et al., 2019;Mijiddorj et al., 2018 leopard population due to conservation efforts or increase in livestock population due to local and regional demands will result in increased predation on livestock. If unmanaged, this will lead to greater conflict and retaliatory killing of snow leopards. Snow leopard habitats are multi-use landscapes, where a growing human population will continue to depend on pastoralism (Johansson et al., 2016;Mishra et al., 2009). Across snow leopard distribution range in the Himalaya and Central Asia, livestock population is expected to increase, particularly goat population, in response to the international demand for cashmere (Berger et al., 2013).
On the one hand, this increase in livestock population, while being important for local economies, may have negative impact on wild prey population as it been found to depress herbivores prey population (Mishra, Van Wieren, Ketner, Heitkonig, & Prins, 2004, Suryawanshi, Bhatnagar, & Mishra, 2010. The resulting outcomes of depressed wild prey population could result in reduced density of snow leopards. On the other hand, conserving snow leopards, while being positive for conservation, may have repercussions on local economies that rely on livestock production, as higher density of snow leopards may also imply higher livestock predation, and consequently more retaliatory killing of snow leopards. Conservation initiatives aiming to recover snow leopard populations or efforts to increase livestock populations in multi-use landscapes for enhancing local economies therefore must be accompanied by preventive measures to protect livestock predation by snow leopards (e.g., predator-proof corrals), and offsetting economic costs of livestock predation (e.g., compensation payments, community managed livestock insurance programs).

ACK N OWLED G M ENT
We thank the Department of National Park and Wildlife We also thank the Snow Leopard Trust, USA, Nature Conservation Foundation, India (NCF) and WWF Nepal for providing camera traps.
We are grateful to Bhumi Prakash Chaudhary Tharu, Kesang Chunit, Tenjin Thuktang, Nurbu Lama, and national park staff and local field assistants for their support in data collection. The corresponding author would like to thank his MSc classmates for useful discussions, and Jayshree Ratnam, Ajit Kumar, and Chandni Gurusrikar for administrative support.

CO N FLI C T S O F I NTE R E S T
None.