People, predators and perceptions: patterns of livestock depredation by snow leopards and wolves

Authors


Correspondence author. E-mail: kulbhushan@ncf-india.org

Summary

  1. Livestock depredation by large carnivores is an important conservation and economic concern and conservation management would benefit from a better understanding of spatial variation and underlying causes of depredation events. Focusing on the endangered snow leopard Panthera uncia and the wolf Canis lupus, we identify the ecological factors that predispose areas within a landscape to livestock depredation. We also examine the potential mismatch between reality and human perceptions of livestock depredation by these carnivores whose survival is threatened due to persecution by pastoralists.
  2. We assessed the distribution of the snow leopard, wolf and wild ungulate prey through field surveys in the 4000 km2 Upper Spiti Landscape of trans-Himalayan India. We interviewed local people in all 25 villages to assess the distribution of livestock and peoples' perceptions of the risk to livestock from these carnivores. We monitored village-level livestock mortality over a 2-year period to assess the actual level of livestock depredation. We quantified several possibly influential independent variables that together captured variation in topography, carnivore abundance and abundance and other attributes of livestock. We identified the key variables influencing livestock depredation using multiple logistic regressions and hierarchical partitioning.
  3. Our results revealed notable differences in livestock selectivity and ecological correlates of livestock depredation – both perceived and actual – by snow leopards and wolves. Stocking density of large-bodied free-ranging livestock (yaks and horses) best explained people's threat perception of livestock depredation by snow leopards, while actual livestock depredation was explained by the relative abundance of snow leopards and wild prey. In the case of wolves, peoples' perception was best explained by abundance of wolves, while actual depredation by wolves was explained by habitat structure.
  4. Synthesis and applications. Our results show that (i) human perceptions can be at odds with actual patterns of livestock depredation, (ii) increases in wild prey populations will intensify livestock depredation by snow leopards, and prey recovery programmes must be accompanied by measures to protect livestock, (iii) compensation or insurance programmes should target large-bodied livestock in snow leopard habitats and (iv) sustained awareness programmes are much needed, especially for the wolf.

Introduction

Livestock depredation by large carnivores and their retaliatory or preventive killing is a world-wide conservation concern (Madhusudan & Mishra 2003; Woodroffe, Thirgood & Rabinowitz 2005; Treves et al. 2006). Persecution of carnivores over livestock depredation, together with the desire to conserve them, leads to situations referred to as human–carnivore conflict (Woodroffe, Thirgood & Rabinowitz 2005). The conflict arises because farmers' interests are compromised, as are conservation goals.

These conflicts have two important dimensions – the reality of damage caused by wildlife to humans, and the perceptions and psyche of humans who suffer wildlife-caused damage. Domestic animals are particularly vulnerable to wild carnivore depredation because a decreased risk of predation in a human-mediated environment has led to a degeneration of their antipredatory abilities (Zohary, Tchernov & Horwitz 1998; Madhusudan & Mishra 2003). Inadequate management of livestock is another important cause of livestock depredation (Woodroffe et al. 2007). On the other hand, people's tolerance for large carnivores varies, depending on several factors, including their religious beliefs, income, education level, characteristics of carnivores and cultural factors (Mishra 1997; Liu et al. 2011). Often, livestock losses attributed to wild carnivores tend to get exaggerated, either mistakenly or deliberately (Mishra 1997; Rigg et al. 2011). These perceptions can have strong emotional and political consequences, ultimately resulting in persecution of carnivores (Kellert et al. 1996). An understanding of the nature of conflicts along both these dimensions – actual damage and people's perceptions – is important if conflicts are to be managed effectively, because together they influence human responses to these losses.

Conservation and the management of human–wildlife conflict would therefore benefit from a better understanding of the extent, causes and correlates of actual livestock damage caused by wild carnivores, and the threat of damage that affected people perceive. While actual levels of livestock depredation are likely to be influenced by the behavioural ecology of the carnivore and livestock-rearing practices, the perceived threat is likely to be influenced by a suit of individual, family, socio-economic and cultural factors such as the cultural and economic value of the livestock killed, the cultural significance of the carnivore, general awareness of the local communities, the presence of good conservation models (unlike livelihood threatening exclusionary conservation actions), and the physical and behavioural characteristics of the carnivore.

The wolf Canis lupus and the endangered snow leopard Panthera uncia occur across the mountain ranges of Central Asia where they live alongside large numbers of livestock (Mishra et al. 2003). Persecution of these two carnivores by pastoralists over livestock depredation threatens their survival across their range (Mishra 1997; Mishra et al. 2003; Namgail, Bhatnagar & Fox 2007). Studies report 3–12% annual losses of livestock holdings to snow leopards and wolves in high conflict areas (Mishra 1997; Hussain 2000; Mishra et al. 2003; Jackson & Wangchuk 2004; Namgail, Bhatnagar & Fox 2007). High losses sometimes create such levels of intolerance that local communities view carnivore extermination as the only solution (Oli, Taylor & Rogers 1994). Managing human–carnivore conflict is of utmost priority for the continued survival of these carnivores in Central Asia (McCarthy & Chapron 2003).

Research on human–carnivore conflicts in Central Asia has focused on documenting the extent of livestock depredation and peoples' attitudes in target villages (Oli, Taylor & Rogers 1994; Mishra 1997; Hussain 2003; Ikeda 2004; Namgail, Bhatnagar & Fox 2007; Jackson et al. 2010). Researchers have recognized the existence of conflict ‘hotspots’ or sites predisposed to livestock depredation, but they have not explored the ecological causes or correlates of livestock depredation or conflict hotspots. Studies have relied on interviews of affected people to understand patterns in livestock depredation. Interviews are likely to reflect peoples’ perceptions of the conflict, which may be at odds with the reality of actual livestock depredation. This potential dichotomy between human perceptions and the actual levels of livestock depredation has not been explored. At an even more fundamental level, many studies have failed to distinguish between livestock damage caused by snow leopards from that caused by wolves (e.g. Mishra 1997; Hussain 2003). These carnivores have different hunting strategies. Being a stalker, the snow leopard is expected to prefer structurally complex habitat with adequate ambush cover for hunting, while the pack-living wolf is expected to hunt in more open habitats (Caro & Fitzgibbon 1992). The difference in their hunting strategies is expected to lead to different patterns of livestock depredation, which may require different conflict management approaches. Similarly, people may relate to these two carnivores differently, which would affect their perceptions of each species in conflict situations.

What are the factors that predispose certain areas or villages within a landscape to livestock depredation by snow leopards or wolves? How similar are the perceptions of local people compared with the actual patterns of livestock depredation? In this study, we focus on identifying (i) the landscape-level ecological factors influencing livestock depredation by snow leopards and wolves and (ii) the correlates of human perceptions of livestock depredation. Our data come from extensive field surveys of carnivore and prey distribution, interviews of local people to understand conflict perceptions and monitoring actual livestock damage over a 2-year period in a 4000 km2 trans-Himalayan landscape. To explain the spatial variation in livestock depredation by each carnivore, we examine the influence of key explanatory factors that together capture variation in habitat topography, abundance of carnivores and wild prey and abundance and other attributes of livestock.

Materials and methods

Study Area

The Upper Spiti Landscape (USL; lat 32°00′–32°42′N; Long 77°37′–78°30′E) in trans-Himalayan India is a high elevation (3500–6700 m) region. Annual temperatures range from −40 °C in peak winter to c. 30 °C in summer. Vegetation is classified as ‘Alpine scrub’ or ‘dry alpine steppe’ (Champion & Seth 1968). Large mammalian fauna of USL includes bharal Pseudois nayaur, ibex Capra ibex and their predators, the snow leopard and the wolf.

Agro-pastoralist communities, currently of Buddhist denomination, have inhabited this region for two to three millennia. Presently, the population of USL is 4908 people in 25 permanent villages and a township. The livestock assemblage includes sheep, goat, donkey, cow, cow–yak hybrid, horse and yak. Livestock graze in the pastures except during extreme winter when they are stall-fed. Based on herding practices, livestock can be classified as (i) Large-bodied free ranging, henceforth referred to as large stock (yaks and horses), and, (ii) Medium and small-bodied herded, henceforth referred to as herded stock (cow, donkey, cow–yak hybrid, goat and sheep). Although large stock of each village graze in separate pastures with exclusive village rights, two or more neighbouring villages sometimes share large pastures for few months. Herded stock is shepherded to the pastures every morning and brought back to the stocking pens inside the villages in the evening. Every family has a separate corral. In summer, the herded stock is kept in well-secured corrals adjoining the house. These corrals often have a separate section covered with wire mesh for smaller animals such as sheep and goats. During winters, all the livestock are penned inside the houses in a section on the ground floor. Every family takes turns at shepherding the entire village's herded stock along with a designated village-shepherd. Families mostly own small agricultural land holdings (1–2 ha). Primary crops are barley Hordeum vulgare, green pea Pisum sativum and a local variety of green pea called black pea.

Field Surveys for Distribution and Relative Abundance of Carnivores and Prey

We divided the USL into 32 sampling blocks comprising catchments of smaller valleys formed by the tributaries of Spiti River (Fig. 1). Block sizes (after excluding areas above 5200 m elevation, where wildlife populations are absent) ranged from 50 to 160 km2 with the mean around 80 km2. We uniformly sampled for wildlife occurrence in all the blocks along three 5-km transects in each block. Each transect was subdivided into five sections of 1-km. All transects were located along microhabitats used by carnivores for social marking as well as good vantage locations such as ridgelines and cliff tops. Ridgelines and cliffs are common features of all blocks in this landscape which allowed consistent placement of transects across blocks. Within each block, we preferred this sampling design focusing on carnivore marking habitats over randomly laid transects as it maximized our chances of detecting carnivore signs. The vantage locations, offering high visibility, provided extensive coverage of habitats within each block for locating evidences of prey. Transects were separated from each other by at least one kilometre. The surveys were conducted by a team of 10 trained people from September 2008–January 2009. Sightings of bharal and ibex groups while surveying the blocks were recorded. The location, group size and composition were noted. Snow leopard and wolf signs (pug marks and scrape marks for snow leopard and only pug marks for wolves) were recorded along every section of each transect. Snow leopard pug marks and scrape signs were unambiguous as this is the only large felid in USL. Wolf signs could be confused with those of village dogs. Therefore, we cross-checked wolf pug marks with secondary reports of wolf sightings in the sampling block within the past month by asking key informants in the nearest village. A total of 412 key informants were interviewed to cross-check wolf presence and gather additional information about presence of bharal and ibex. Of the 96 potential transects, we could survey 75 transects in 30 blocks. The others could not be surveyed due to difficulty in accessing the areas.

Figure 1.

The distribution of the bharal Pseudios nayaur and ibex Capra sibirica, key snow leopard prey, in the Upper Spiti Landscape, India. Encounter rates of bharal and ibex was calculated as number of individuals sighted per km along transects.

Interview Surveys for Perceived Threat to Livestock

To identify areas where people perceived snow leopards and/or wolves to be a threat to livestock, we conducted group interviews in all the villages (n = 25) between October 2008 and January 2009. Interviewees comprised village elders, village heads and the youth. We collected village-level information on livestock holding, land holding and human population. Resource uses such as the livestock grazing cycle, pastures, location of livestock holding pens, agricultural fields, roads and streams were mapped and later digitized at 1:50000 scale (manifold version 8.0; Manifold Net Ltd., Carson City, NV, USA).

We asked the interviewees to list the problems they faced in rearing livestock, without prompting them about livestock damage by carnivores. Interviewees who reported livestock depredation by wild carnivores were also questioned about the species (snow leopard and/or wolf). Each village was accordingly scored one or zero for livestock depredation by snow leopard and by wolf. Unlike many other parts of the snow leopard's range, the stocking pens in our study area are better protected, and all livestock depredation instances in the last 5 years were reported to occur in the pastures. There was just one case of killing of over 20 goats in the stocking pen of a hamlet with just one house. Thus, the explanatory variables we used in the analysis were related to the pastures where the livestock of each village grazed.

Actual Livestock Damage

During the interview surveys of 2008, we requested the professional herders of each village to maintain a record of all livestock mortality on a monthly basis. The herders maintained records of the species and age of livestock, month, cause of mortality and the total number of livestock in each village throughout January 2009 to December 2010. Identity of the predator was confirmed through direct sighting or signs around the kill. By cross-checking their knowledge, the team ensured that the herders keeping records were experienced at identifying signs of snow leopards and wolves. A local representative maintained contact with each herder every 4–6 months and recorded all the livestock mortalities. We expect the herder's records to be accurate, as they also had to report any mortality to the livestock owner.

We examined seven explanatory variables that potentially determined the occurrence of livestock depredation by wolves and snow leopards (Table 1). They captured variation in topography, relative use by carnivores and prey, and abundance and other attributes of livestock. Encounter rates of carnivore signs were used as indices of relative use of each block. To ensure that our indices were robust, we compared naïve occupancy estimate with estimates using software presence 2.0 wherever necessary (Hines 2004). Stocking density was the number of livestock per square kilometre. Mean altitude of pastures was estimated using a 90 × 90 m SRTM digital elevation model (http://srtm.csi.cgiar.org/SELECTION/inputCoord.asp). Ruggedness was calculated using a moving circular window of five-pixel radius. Ruggedness of the central pixel of the window was calculated as the standard deviation of altitude of all the other pixels within the window. Ruggedness values of all pixels were then averaged for each pasture.

Table 1. Explanatory variables considered as potential landscape-level predictors of the extent of livestock depredation by snow leopards and wolves
Variable nameVariable description
Wild-prey encounter rateEncounter rate of wild prey (bharal and ibex) calculated as number of individuals sighted per km along transects
AltitudeMean altitude of pastures grazed by village livestock
Density of herded stockMean stocking density of small-bodied herded stock (cow, cow–yak hybrid, donkey, goat and sheep) weighted by the number of months that the animals were stocked at a particular density
Density of large stockMean stocking density of large-bodied free-ranging livestock (horse and yak) weighted by the number of months that the animals are stocked at a particular density
Ruggedness of herded stock pastureMean ruggedness of pastures grazed by small and medium-bodied herded stock. Ruggedness was estimated using a Digital Elevation Model (MODIS 250 × 250 m). Ruggedness was calculated using a moving circular window of radius five pixels. Ruggedness of the central pixel of the window was calculated as the standard deviation of altitude of all the pixels within the window. Ruggedness values of all pixels were then averaged for each pasture. ARC GIS was used to calculate ruggedness
Ruggedness of large stock pastureMean ruggedness of pastures grazed by large-bodied free-ranging stock
Wolf encounter rateEncounter rate of wolf signs. Encounter rate of wolf presence signs was calculated as the proportion of transect sections in which wolf signs were recorded
Snow leopard encounter rateEncounter rate of snow leopard presence signs calculated as in the case of wolf

We considered encounter rates of the snow leopard and wolf signs and wild-prey sightings as potential explanatory variables. Higher encounter rate of carnivore signs was assumed to reflect relative abundance or greater use of an area by the carnivores. Encounter rate was calculated for each sampling block as the proportion of transect sections with carnivore signs. Encounter rate of wild prey was calculated as the number of animals seen per kilometre. Higher encounter rate of bharal and ibex sightings was assumed to reflect greater availability of wild prey. For each village, encounter rate was calculated using data from all the blocks over which its livestock ranged.

Data Analysis

Prey and carnivore distribution

Site occupancy for each species was estimated as the proportion of blocks in which it was detected. A species was considered present if it was detected in any of the transect sections within a block. For the wolf, the data were used to estimate detection probability and ‘corrected’ site occupancy using the programme ‘presence 2.0’ (MacKenzie et al. 2002). This analysis was not deemed necessary for the snow leopard as the species was detected in all blocks. Site occupancy for the snow leopard, bharal and ibex was calculated as the proportion of blocks with the species presence sign (for snow leopard) or sighting (for bharal and ibex). We chose to rely on naïve site occupancy for the prey species because their sightings in different transects were not always independent, owing to high visibility in our study area.

Livestock Selectivity

We assessed the selectivity of snow leopard and wolf for various livestock species following Vanderploeg & Scavia (1979). Electivity index (E*) was calculated as: Ei* = [Wi − (1/n)]/[Wi + (1/n)], where n is the total number of livestock species. Wi = (ri/pi)/Σ(ri/pi), where the proportion of the ith livestock species in all the livestock killed by the carnivore is denoted by ri, and the proportion of the ith livestock species in the total population is denoted by pi.

Variables Influencing Livestock Depredation and Perceived Conflict

We had two Poisson-distributed variables: number of livestock killed by snow leopard and by wolves. We also had two binary response variables: villages where people perceived a threat to livestock from the snow leopard and from the wolf. We used Moran's I coefficient, based on straight-line distances between villages to test whether or not the response variables in village close together were spatially autocorrelated.

We evaluated 16 multiple logistic regression models and ranked them using Akaike's information criterion adjusted for small samples (AICc; Burnham & Anderson 2002). Although multiple regressions allow inferences from observational data when there are several competing hypotheses, ultimately the inferences are based on the models included in the candidate set. The exclusion of the best model or inclusion of a meaningless model can mislead the inferences (Johnson & Omland 2004). Also, for inferring the relative importance of predictor variables, it is necessary to examine the entire set of candidate models. We therefore used two complementary analyses. We calculated cumulative Akaike weight for each explanatory variable by summing model weights for all models containing that variable (Burnham & Anderson 2002). The variables with the highest cumulative AICc weight have the greatest relative influence on the response variable, allowing the explanatory variables to be ranked from most important to least important. Secondly, we used hierarchical partitioning with Root Mean Square Error as the measure of goodness-of-fit to determine the relative importance of each explanatory variable (Chevan & Sutherland 1991; Mac Nally 2002; Mac Nally & Walsh 2004). This method addresses multicollinearity between different explanatory variables (Mac Nally 2000). We followed Mac Nally (2002) to identify statistically significant variables by randomizing the data matrix 1000 times. We used Poisson and logistic regression models in the hier.part package in the software package r (version 2.13: http://www.r-project.org) to estimate the independent and joint contribution of each variable. All statistical analyses were performed using the software r (version 2.9: http://www.r-project.org).

Mac Nally (2000) argues that these two methods should agree for the inferences to be significant. We compared the explanatory variables identified by both methods to identify the factors having the largest independent effect on villages affected by livestock depredation by either carnivore.

Results

Distribution of Carnivores and Prey

Site occupancy was highest for snow leopard with their signs being detected in all the blocks (Fig. 2). Thus, we do not expect under-estimation of site occupancy due to pseudoabsences and we used encounter rates of snow leopard sign as predictor variable in explaining actual and perceived livestock depredation by this carnivore. Bharal site occupancy was 0·65 (Fig. 1). Bharal presence could be confirmed through direct sightings (192 bharal in 26 groups) in all the blocks identified as bharal areas by key informants. Two blocks identified as bharal areas by key informants could not be surveyed due to logistic difficulties. Site occupancy for ibex was 0·36. We could not confirm the ibex presence through direct sightings in two grid cells identified as ibex areas by key informants (Fig. 1). We sighted 120 ibex in nine groups during this survey. Only field sign survey data were used for estimating site occupancy. Cumulative site occupancy for ibex and bharal was 0·94. Only 10% of the blocks were occupied by both bharal and ibex.

Figure 2.

Distribution and encounter rate of snow leopard Panthera uncia, and villages where they were perceived as a threat to livestock (open circles) in the Upper Spiti Landscape, India. Encounter rate of snow leopard signs was calculated as the proportion of transect sections in which snow leopard signs were recorded. These villages were not spatially autocorrelated with the observed Moran's I score being −0·05 (SE 0·03) against the expected score of −0·04 (P = 0·58).

Detection probability estimated for the wolf using software presence 2.0 was 0·45 (SE = 0·15). The site occupancy estimated (using presence) was 0·32 (SE = 0·12). The corrected site occupancy estimate was not significantly different from the naïve estimate of 0·25 (z-score 0·58; = 0·56). Naïve estimate is the proportion of sites where wolf presence was actually detected. Wolf distribution seems to be limited to the blocks where we actually detected their presence (Fig. 3). Thus, we used encounter rates as predictor variables in explaining actual and perceived livestock depredation by the wolf.

Figure 3.

Distribution and encounter rate of the wolf Canis lupus, and villages where they were perceived as a threat to livestock (open circles) in the Upper Spiti Landscape, India. Encounter rate of wolf signs was calculated as the proportion of transect sections in which wolf signs were recorded. These villages were not spatially autocorrelated with the observed Moran's I score being −0·07 (SE = 0·03) against the expected score of −0·04 (P = 0·25).

Perceived Threat to Livestock

In 13 villages, people did not perceive any threat to livestock from either snow leopard or wolf. In five villages, people perceived both snow leopard and wolf as a threat to livestock. Only the snow leopard was perceived as a threat in three villages, and only the wolf in four. Thus, in eight villages, people perceived livestock damage by snow leopard as a cause of concern for livestock rearing. These were not spatially autocorrelated, with the observed Moran's I score for them being -0·05 (SE 0·03) and expected being -0·04 (= 0·58).

The best model, identified as the one with least AICc, in explaining and positively associated with villages where snow leopards were perceived as a threat to livestock included just one parameter: Density of large stock (R2 = 0·42). Coefficient estimate was 0·22 (SE = 0·08). Burnham & Anderson (2002) suggest considering all models within two ∆AICc units of the best model. We did not have any other model within two ∆AICc units of the best model. The next best model had a ∆AICc of 2·04 and included two parameters: density of large stock and density of herded stock (R2 = 0·44; Table 2). Density of large stock had the highest cumulative Akaike weight (0·99) followed by density of herded stock (0·25). All other variables had cumulative AICc of <0·2 (Table 3). Hierarchical partitioning again showed that density of large stock had the highest explanatory power for livestock depredation by snow leopards, with independent, statistically significant (< 0·005) contribution of 82·38% (Fig. 4a). Although ruggedness of large stock pastures was next in explanatory power, its independent contribution was only 6·48% and it was not statistically significant (= 0·9).

Table 2. Structure of models used in a multiple logistic regression framework with ∆AICc as model selection criteria to identify the variables of villages affected by human–snow leopard and human–wolf conflict
Model StructureKLivestock damage by snow leopardLivestock damage by wolfPerceived threat of snow leopardPerceived threat of wolf
AICcResidual devianceDelta AICcAICc weightAICcResidual devianceDelta AICcAICc weightAICcResidual devianceDelta AICcAICc weightAICcResidual devianceDelta AICcAICc weight
  1. Abbreviations indicate snow leopard encounter rate (sl.enc), wolf encounter rate (w.enc), wild-prey encounter rate (wp.enc), density of herded stock (den.hs), density of large stock (den.ls), altitude (alt), ruggedness of herded stock pasture (rugg.hs) and ruggedness of large stock pasture (rugg.ls). Variables are defined in Table 1. K is the number of estimable parameters including the intercept. AICc, Akaike's information criterion.

carnivore.enc2331·0271·5105·50·0557·0518·1278·60·035·631·112·90·026·922·370·00·4
den.hs2407·6348·1182·10·0579·3540·4300·90·034·630·111·90·035·130·528·20·0
den.ls2434·6375·1209·10·0589·6550·7311·20·022·718·160·00·531·627·014·60·0
wp.enc2329·6270·1104·10·0595·2556·3316·80·035·931·3313·20·036·231·659·30·0
den.hs + den.ls3410·1348184·60·0575·9534·4297·50·024·717·552·00·233·025·836·10·0
den.ls + rugg.ls + alt4372·5307·5147·00·0301·6257·323·20·027·417·414·70·036·426·359·40·0
den.hs + rugg.hs + alt4330·9265·9105·40·0278·42340·00·938·628·5815·90·039·329·2812·40·0
den.hs + den.ls + rugg.hs + rugg.ls5338·6270·4113·00·0283·62365·20·130·517·377·80·039·025·8112·10·0
carnivore.enc + wp.enc3225·5163·40·01·0554·2512·8275·80·038·231·0915·50·028·421·31·50·2
carnivore.enc + den.ls3328·1266102·60·0559·1517·6280·70·025·318·122·60·129·021·862·10·1
carnivore.enc + den.hs3328·0265·9102·50·0559·0517·5280·60·037·230·0914·50·029·121·992·20·1
carnivore.enc + den.ls + rugg.ls4281·4216·455·90·0354·6310·276·20·026·816·774·10·131·821·844·90·0
carnivore.enc + den.hs + rugg.hs4244·3179·318·80·0310·9266·632·50·040·030·0317·30·032·021·995·10·0
rugg.hs + alt3330·4268·4104·90·0289·1247·710·70·036·229·0913·50·037·330·1210·30·0
rugg.ls + alt3375·3313·2149·80·0298·8257·320·40·034·727·612·00·036·729·529·70·0
wp.enc + den.ls +den.hs4333·5268·6108·00·0577·9533·5299·50·027·117·064·40·133·523·496·60·0
Table 3. Cumulative Akaike's information criterion (AICc) weight of explanatory variables from all models of actual livestock damage by snow leopards and wolves, and the threat to livestock perceived by local people from snow leopards and wolves
VariablesSnow leopard damageWolf damagePerceived threat snow leopardPerceived threat wolf
Snow leopard encounter rate1·00NA0·20NA
Wolf encounter rateNA0·00NA0·91
Density herded stock0·000·990·250·20
Density large stock0·000·071·000·25
Altitude0·000·930·050·01
Ruggedness herded stock pasture0·001·000·010·04
Ruggedness large stock pasture0·000·070·120·04
Wild-prey encounter rate1·000·000·060·20
Figure 4.

Independent contribution of the seven explanatory variables to (a) actual livestock damage by snow leopard, and, (b) villages where people perceived a threat to livestock from the snow leopard, determined by hierarchical partitioning. Asterisk denotes statistical significance. Abbreviations indicate snow leopard encounter rate (sl.enc), wild-prey encounter rate (wp.enc), density of herded stock (den.hs), density of large stock (den.ls), altitude (alt), ruggedness of herded stock pasture (rugg.hs) and ruggedness of large stock pasture (rugg.ls). Variables are defined in Table 1.

People in nine villages perceived the wolf as a threat to livestock rearing. These villages were not spatially autocorrelated. The observed Moran's I score for these villages was −0·07 (SE = 0·03) against the expected score of −0·04 (= 0·25). The model with least AICc explaining perceived threat by wolves included only one parameter: Wolf encounter rate that was positively related to the villages where people perceived wolves to be responsible for livestock depredation (R2 = 0·32; AICc = 26·91) (Table 2). The coefficient estimate was 24·64 (SE = 9·59). The only other model within two ∆AICc units included the variables wolf encounter rate and wild-prey encounter rate. The cumulative AICc weight of wolf encounter rate was 0·91 followed by density of large stock (0·25), density of herded stock (0·20) and wild-prey encounter rate (0·20). In the hierarchical partitioning analysis, wolf encounter rate had significantly high independent explanatory power (< 0·05) with an independent contribution of 41%. Density of large stock was the next best variable with independent contribution of 19·86% (z-score = 0·75; = 0·45), but it was not statistically significant (Fig. 5a).

Figure 5.

Independent contribution of the seven explanatory variables to (a) actual livestock damage by the wolf, and, (b) the villages where people perceived a threat to livestock from the wolf, determined by hierarchical partitioning. Asterisk denotes statistical significance. Abbreviations indicate wolf encounter rate (w.enc), wild-prey encounter rate (wp.enc), density of herded stock (den.hs), density of large stock (den.ls), altitude (alt), ruggedness of herded stock pasture (rugg.hs), ruggedness of large stock pasture (rugg.ls). Variables are defined in Table 1.

Actual Livestock Depredation

Over 2 years of monitoring, livestock damage by snow leopard was recorded in 14 villages. This included seven of the eight villages where people in 2008 had reported a perceived threat from the snow leopard to livestock. These villages were not autocorrelated in space. The observed Moran's I was −0·03 (SE 0·02) against an expected −0·04 with a nonsignificant = 0·86. Interestingly, in one village where people had not perceived snow leopard as a threat to livestock in 2008, 51 goats were killed by snow leopards in 2010. The other five villages where people had not perceived the snow leopard to be a threat together recorded snow leopard caused mortality of 12 livestock over 2 years.

Actual livestock damage by wolves was recorded from eight villages during the study period. This included only five of the nine villages where people had perceived wolves to be a threat to livestock at the beginning of our study, and three additional villages where people had not perceived wolves to be a threat. These villages were not autocorrelated in space. The observed Moran's I score was −0·02 (SE 0·02) against and expected −0·04 with a nonsignificant = 0·39. Two villages where people did not perceive wolf as a threat to livestock in 2008 recorded wolf-caused mortality of five sheep and three donkeys over the next 2 years. A third village where people had not perceived the wolf to be a threat to livestock in 2008 recorded mortality of an unprecedented 66 sheep, 11 donkeys and two horses in 12 independent events in 2009.

In total, we recorded 809 livestock mortalities (330 events) in the 25 villages over a period of 2 years (Table 4). Surprisingly, the biggest cause of mortality was predation by feral dogs (338 livestock heads). Snow leopards and wolves killed 194 and 173 livestock, respectively. Disease accounted for 52 animals and another 52 either went missing or the cause of mortality was uncertain. Over 90% (170) of the livestock damage by snow leopard occurred during the 5 months between May and September. Livestock damage by wolves, on the other hand, was spread throughout the year.

Table 4. Number of livestock deaths due to various causes between January 2009 and December 2010 in the 25 villages of the Upper Spiti Landscape, India
Cause of mortalityCowDonkeyDzoGoatSheepHorseYakTotal
Snow leopard51163233467194
Wolf5411308826173
Feral Dog65412020030338
Disease65211174752
Missing426371932
Unknown1221130120

The electivity index (E*) of snow leopards was positive for horse and yak (0·62 and 0·18, respectively), the two large-bodied free-ranging livestock types. Of the total yaks (67) and horses (34) killed by snow leopards, over 90% and 80%, respectively, were young animals (<1 year old). The selection of snow leopards for all herded livestock including cow, yak–cow hybrids, donkey, goat and sheep was negative in relation to their proportional abundance (−0·74, −0·96, −0·95, −0·09 and −0·66, respectively). E* index of wolf was positive for donkey and sheep (0·4 and 0·35, respectively). E* indices of wolf for cow, dzo, goat, horse and yak were −0·43, −0·87, 0·03, −0·18 and −0·48, respectively. Three of the six yaks killed by wolves were adult females and the other three were young males.

Factors Influencing Actual Livestock Depredation

Of the models explaining livestock depredation by snow leopard, the one with the least AICc score included snow leopard encounter rate and wild-prey encounter rate (R2 = 0·56; Table 2). The coefficients of snow leopard encounter rate and wild-prey encounter rate were 6·6 (SE 0·7) and 7·4 (SE 0·9), respectively. No other model was within two or even four AICc units of the best model. Snow leopard encounter rate and wild-prey encounter rate had a cumulative AICc weight of 1·0. Hierarchical partitioning also showed snow leopard encounter rate and wild-prey encounter rate as the most important variables with independent effects of 43·8 and 25·0%, respectively (Fig. 4b). Both the terms were significant (< 0·0005 and < 0·05, respectively).

In the case of the wolf, the model with the least AICc score included density of herded stock, ruggedness of herded stock pastures and altitude (R2 = 0·58; Table 2). The coefficients for density of herded stock, ruggedness of herded stock pasture and altitude were −0·003 (SE 0·001), −0·05 (SE 0·006) and 0·0003 (SE 0·0008), respectively. No other model was within two or even four AICc units of the best model. The cumulative AICc weight of density of herded stock, ruggedness of herded stock pasture and altitude were 0·99, 1·0 and 0·93, respectively. However, hierarchical partitioning analysis suggested that only the ruggedness of herded stock pasture with an independent contribution of 34% was significant at < 0·05. Both density of herded stock and altitude had an independent effect of only 10·6 and 13·9%, respectively (Fig. 5b). These contributions were not statistically significant (= 0·96 & 0·78, respectively).

Discussion

We recorded notable differences between the ecological correlates of livestock depredation, both perceived and actual, by snow leopards and wolves. Although snow leopards were ubiquitous in our study area, people perceived them to be a threat to livestock in only one-third of the villages, especially those where large-bodied, free-ranging livestock (yaks and horses) were economically important. In contrast, wolves occurred only in one-thirds of our study area but were perceived as a threat to livestock in all the villages in those areas. The occurrence of wolves, by itself, was the most important predictor of peoples' perception of the threat posed by this canid to livestock.

Snow leopards preferred horses and yaks over all other species of livestock. Case studies of snow leopard diet have also reported snow leopards to attack horses disproportionately compared with their abundance (Mishra 1997; Bagchi & Mishra 2006; Namgail, Bhatnagar & Fox 2007). Although people in only eight villages had perceived snow leopards to be a threat to livestock in the beginning of our study, many more (14 villages) recorded instances of livestock depredation by snow leopards over the following 2 years. In contrast, although people in nine villages had perceived wolves to be a threat, fewer (eight villages) actually recorded depredation by the wolf. Our results show that compared with snow leopards, threat perceptions are disproportionately biased against the wolf.

Peoples' perception of a carnivore in conflict is likely to be influenced by (and in turn influence) their attitude towards the species in question. A strong cultural bias against the wolf is reflected in strong negative attitudes and relatively high persecution compared with other sympatric carnivores such as the mountain lion Puma concolor in North America (Kleiven, Bjerke & Kaltenborn 2004). The attitudes towards a species are reportedly influenced by the physical and behavioural characteristics of the carnivore, and their cultural and historical associations (Kellert et al. 1996; Kleiven, Bjerke & Kaltenborn 2004). Although causal relationships are not clear, factors such as greater visibility, perceived threat and conspicuous behaviours such as howling and group living, ease in detection of denning sites etc., may accentuate the threat perception and generate greater negative attitudes towards the wolf (Kellert et al. 1996).

Actual livestock depredation by snow leopards was best explained by its own encounter rate and of its wild prey. In the case of wolves, the ruggedness of the pastures used by herded stock was the most important factors influencing actual livestock depredation. This relationship between structural complexity of the habitat and the extent of livestock depredation by wolves was negative. This is consistent with the expectation that a cursorial carnivore would prefer structurally less complex habitat. Surprisingly, we did not find any association between livestock depredation by snow leopards and ruggedness of the pasture. It appears that structural complexity may be more important for snow leopards at a finer scale of hunting site rather than at the scale of a grazing pasture.

Implications

Our study provides two main insights into human–carnivore conflicts. First, human perceptions can be at considerable odds with actual patterns of livestock depredation, and, second, livestock depredation by snow leopards and wolves show rather different patterns in prey selectivity and ecological determinants. This suggests that while interviews of local people, which have been commonly employed to study livestock depredation conflicts, could yield accurate information on peoples' perception of a conflict situation, the reality of livestock depredation must be measured additionally and independently.

More importantly, the insights from our study have implications for conflict management programmes and help gauge the future of human–carnivore conflicts in Central Asia. The relationship between livestock depredation by snow leopards and the relative abundance of wild prey suggests that human–snow leopard conflicts are likely to get more intense if successful conservation programmes lead to increases in wild-prey abundance from the low densities typical of multiple use, livestock-grazed landscapes (Mishra et al. 2004). This is in contrast to our own earlier work where we have indicated that livestock depredation by snow leopards may decline with increase in wild-prey abundance (Bagchi & Mishra 2006). In fact, this premise has guided an important aspect of conflict management, viz., facilitating wild-prey population recovery (Mishra et al. 2003). Our present study, however, suggests that an increase in wild prey – a highly desirable conservation outcome considering their own endangerment and their functional role – would, in fact, lead to an increase in livestock depredation by snow leopards, presumably by supporting a greater abundance of the cat. We therefore hypothesize that the relationship between snow leopard depredation of livestock and wild-prey abundance may be bimodal. We expect livestock depredation to increase as wild-prey populations increase or decline beyond certain thresholds that are influenced by the carrying capacity of the prey, predator and availability of other resources such as denning and resting sites. This means that conservation initiatives aimed at facilitating the recovery of wild-prey populations must also be accompanied by measures to better protect livestock. Furthermore, based on livestock preference patterns, we suggest that conflict management programs should especially target large-bodied livestock in snow leopard habitats where they form an important part of the peoples' economy.

Our results suggest that the extent of livestock depredation by wolves may not be similarly affected by changes in wild-prey abundance. However, the wolf is likely to face even more intense persecution in the future, considering the on-going socio-economic changes in Central Asia where the global demand for cashmere is leading to an increase in livestock population, and replacement of larger bodied livestock with smaller bodied, cashmere-producing goats (Schaller 1998; Namgail, Bhatnagar & Fox 2007). Increasing abundance of goats, especially in relatively flatter parts of Central Asia such as the Tibetan Plateau and the northern steppes, will likely intensify human conflicts with wolves much more compared with snow leopards. Apart from measures to better protect livestock, sustained education and awareness towards the importance of the conservation of these carnivore species to increase the social carrying capacity will be needed, especially for the wolf.

Acknowledgements

We thank the Forest Department of Himachal Pradesh for financial support and permissions. BBC Wildlife Fund, Whitley Fund for Nature, and Conservation Leadership program for funding. We thank Rashid Raza, Koustubh Sharma, Umesh Srinivasan for comments. Sushil ‘Tandup Dorje’, Tenzin Thillay, Dorge Tsewang, Chudim, Rinchen Tobgay, Kalzang Pulzor, Kalzang Gurmet, Palden Rabgay, Thukten provided valuable contribution to fieldwork. We thank the 412 key informants for invaluable information.

Ancillary