Determining the risk of predator attacks around protected areas: the case of Bardia National Park, Nepal

Abstract In this study we determined the probability of predator attacks on livestock around Bardia National Park, Nepal. We conducted semi-structured interviews to explore the patterns and factors affecting livestock losses in four administrative sectors of the Park's buffer zone. We developed models to investigate the overall probability of livestock loss, economic damage caused, and the respondents’ attitudes towards wildlife. The probability of leopard attacks on livestock was much higher (85% of all livestock lost to depredation) than that of tiger attacks (8%), and the northern sector experienced the highest loss of livestock (50% modelled probability of livestock loss) in the buffer zone. Livestock loss was significantly related to the number of livestock owned by respondents, their ethnic group, and village distance to the Park boundary. Economic damage was influenced by buffer zone sector, number of livestock owned, and distance to the Park boundary. Conservation attitudes depended on respondents’ knowledge of wildlife, levels of education and self-sufficiency, and the probability of livestock being killed by leopards. Respondents who were male, highly educated and self-sufficient were most likely to support conservation. Tigers are tolerated based on religious beliefs, and these cultural values, together with the sharing of conservation benefits, facilitate conservation. Leopards, however, are not tolerated in the same way and are the most damaging predators.


Introduction
L arge carnivore populations have declined globally for a variety of reasons, but mostly because of human interventions (Woodroffe & Ginsberg, ; Karanth & Chellam, ). Poaching for traditional medicine and furs, habitat destruction and depletion of their natural prey are major threats (Smith et al., ; Treves & Karanth, ). Trophy hunting and retaliatory killing in response to livestock depredation have also had adverse effects on lion populations in Africa (Bauer & Iongh, ; Croes et al., ).
Several studies have reported raised levels of so-called human-wildlife conflict (Madden, ) in areas where large carnivore populations have started to increase as a result of conservation actions (Saberwal et al., ; Treves & Karanth, ; Inskip & Zimmermann, ; Seidensticker, ; Silwal et al., ). However, the use of the term human-wildlife conflict is misleading because, in reality, the conflict is often between conservation and other human interests (Peterson et al., ; Redpath et al., ; Fisher, ).
Predator attacks on livestock are one of the most serious challenges faced by villagers in and near protected areas, with tigers Panthera tigris and leopards Panthera pardus killing  and  livestock annually in Bardia and Chitwan National Parks, respectively (Lamichhane et al., ). Despite these losses, people living around protected areas in Asia are relatively tolerant of wildlife (Dinerstein et al., ; Karanth & Nepal, ) compared to Indigenous People of other regions because their cultural values and beliefs promote wildlife conservation (Gogoi et al., ). Because the recovery of both leopard and tiger populations depends on the capacity of local communities to coexist with them, conflict mitigation measures need to be implemented in areas where these felids occur in the vicinity of human settlements (Treves et al., ; Woodroffe et al., ; Carter et al., ).
Here we aim to () identify the probability of livestock loss to predator attacks and other factors, such as disease and flooding, in different sectors of the buffer zone of Bardia National Park; () quantify the economic damage to resident communities caused by predator attacks; and () assess the residents' attitudes towards the conservation of tigers, leopards and other wildlife, and the factors underlying these attitudes.

Study area
Bardia National Park (henceforth Bardia, IUCN category II), established in , is located in south-western Nepal (Fig. ). It is the largest national park in the plains (terai) of Nepal ( km  ; DNPWC, ) and one of the prime habitats for tigers and leopards in the country (Walston et al., ). The Park's  km  buffer zone was established in , with an additional  km  of primarily hilly terrain in the Surkhet district included in . According to regulations -% of the revenue generated by the protected area are to be invested in the buffer zone, to minimize the impact of wildlife on local communities (Baral & Heinen, ).
The rapid expansion of human settlements, habitat degradation, and poaching caused tiger numbers to fall to  individuals inside the Park in . However, the tiger population had increased to  individuals in  (Dhakal et al., ). Leopard numbers have not been assessed recently in Bardia, but Wegge et al. () estimated their density to be  individuals/ km  .
The Park has three distinct seasons: winter (late September-mid February), summer (mid February-mid June) and monsoon (mid June-September). The maximum temperature is °C and mean annual rainfall is , mm (Dinerstein, ; DNPWC, ). Flooding in  caused significant loss of human life and livestock (Bhattarai et al., ), but loss of wildlife was not documented.
Indigenous Tharu people and migrants from the hills (Pahade) inhabit the buffer zone of the Park (Bhattarai et al., ). The majority of households are involved in subsistence farming supplemented by livestock grazing in forests and grassland (Thapa Karki, ). Rice and maize are grown mainly in the monsoon, whereas wheat, mustard and lentils are grown in winter, for domestic consumption (Studsrød & Wegge, ). Livestock including cattle, buffalo, sheep, goats, pigs and chickens are kept by villagers primarily for milk, meat, manure and draught power (Thapa Karki, ).

Data collection
We used a semi-structured questionnaire survey (Supplementary Material ; adapted from Sogbohossou et al.,  and Bhattarai & Fischer, ) of  households that were opportunistically selected in the four study sectors, taking into account the size and the total number of households in each sector ( Table ). Interviews with the heads of the selected households took place during May-August  and were conducted in Nepali by native Nepali and Tharu language speakers who also worked as tourist guides in Bardia. They were trained and instructed on the structure and purpose of the questionnaire before the interview, and during the course of the survey regular assessments were performed to limit any bias that could have resulted from their positive attitude towards conservation as professional nature guides.
We divided the surveyed communities into four groups based on their location in Park management sectors. We considered relative densities of livestock and natural prey in these sectors to be important factors contributing to the probability of predator attacks. The Thakurdwara and western sectors were characterized by relatively high densities of both natural prey and livestock, whereas the eastern and northern sectors had relatively low densities of natural prey but high livestock densities.

Data analyses and statistics
We developed logistic models to estimate () overall probability of loss of livestock and poultry (including loss caused by predator attacks and other factors such as disease or natural calamities), () probability of loss to all predators combined, and () probability of loss to leopards in particular. We used a linear model for economic loss (i.e. the cost of livestock losses to predator attacks and other factors).
We further created logistic models for attitude (which was either positive, coded as , or negative, coded as ) towards () wildlife in general, () wildlife conservation in general, () wildlife conservation when family members were affected by wildlife, and () wildlife conservation amongst respondents personally affected by livestock losses to predators. All models were created in R (R Core Team, ). We used a likelihood ratio test to compare models with and without independent variables (Bolker et al., ; see Supplementary Table  for all variables).

Results
Respondents,  men and  women, were - years old (mean  years). The majority (%) were literate.
Twenty-four per cent had completed a basic education, % were educated to primary, % to lower secondary and % to secondary level, and % had completed a higher secondary or university level education. Just over half of respondents (%) were able to sustain their households on their own production for - months, whereas for % this period was ,  months. Interviewees were of several cultural backgrounds, with % Tharus, % Brahmin or Chhetri, % Dalits and % from other ethnic groups. Most (c. %) respondents were farmers. The mean household size was  persons and the mean numbers of livestock kept by respondents were . cattle, . sheep/goats, . buffalo and . pigs. The northern sector contained % of total livestock in the buffer zone, Thakurdwara sector %, and western and eastern sectors  and %, respectively.
Over half (c. %) of respondent households were within  km of the Park boundary, % - km, and % .  km from the Park boundary. In total,  (%) household heads reported loss of livestock and poultry either to predator attacks ( cases, %) or other factors such as disease and flooding ( cases, %). Of the  cases of livestock and poultry depredation,  (%) were caused by leopards, eight (%) by tigers, and six (%) by other wildlife (foxes Vulpes vulpes, jackals Canis aureus, crocodiles Crocodylus palustris, pythons Python bivittatus and eagles Aquila spp.). Leopards killed primarily goats (% of leopard attacks), followed by pigs (%), sheep (%) and cattle (%). Tigers killed seven cattle (.% of tiger attacks) and one buffalo (.%). Crocodiles and foxes killed goats, whereas other wildlife killed poultry only. Predator attacks were more common in summer (%) and winter (%) than in the monsoon (%). The majority (%) of losses occurred inside the village and % outside a settlement, in the forest. Most respondents (c. %) were able to distinguish a tiger from a leopard based on photographs. They were also able to recognize the predators based on pugmarks at the kill site and bite marks on the livestock carcass, which was confirmed by the Park authority when verifying compensation claims.

Probability of loss
The probabilities of livestock loss per household for each study sector, with their respective causes, are shown in  Table ).
In all three models, the probability of livestock loss was highest in the northern sector (c. %), followed by the eastern (c. %) and the Thakurdwara (c. %) sectors (Fig. a), and the probability of loss in the Tharu ethnic group was lower than for other ethnic groups (Fig. b). Approximately % of predator attacks took place when livestock was held inside corals and % occurred at night.

Economic loss
The total costs of livestock lost to predator attacks and other factors was USD , for the surveyed households, of which USD , (.%) was a result of predatory attacks and USD , (.%) because of other factors ( Table ). The mean value of livestock species ranged from USD  to USD  per animal, depending on their size. The linear model suggests that the most important factors contributing to the predation-related economic loss were study sector (P , .) and distance to the National Park (P = .; Table ).

Attitudes towards wildlife
Of the  responses on questions exploring attitudes towards wildlife, % were positive. Most (%) respondents were supportive of wildlife conservation in general, even when family members had suffered wildlife-related losses. Approximately % of the respondents who had personally suffered livestock losses from predator attacks in the past indicated they still support the protection and conservation of wildlife (Table ).
Our logistic model on attitudes towards wildlife showed that respondents from the western sector were most supportive of conservation, followed by Thakurdwara, eastern, and northern sectors, in this order (P = .). Respondents with a higher level of education were more positive towards wildlife in general (P = .) and willing to conserve wildlife (P = .). Respondents who were self-sufficient (i.e. generating crop yields that could sustain their household throughout the year) were more positive about wildlife conservation in general than respondents who were not selfsufficient (P = .). Men were more positive about wildlife conservation than women (P = .; Supplementary Table ). Contrary to our expectations, respondents who had lost livestock to tigers had a positive attitude towards wildlife conservation (P = .).
The model on attitudes towards wildlife conservation shows that self-sufficiency and education level were positively related to a positive attitude (P = . and .), even when family members had lost livestock to predators. There is an indication that the overall probability of loss affects attitudes towards conservation (P = .; Supplementary Table ).
The attitude of respondents who had personally suffered livestock losses varied between study sectors. The majority (c. %) of respondents in the Thakurdwara and western sectors, % in the eastern and only % in the northern sector were positive towards wildlife conservation after having lost livestock to predators ( Table ). The proportion of positive responses increased with educational level (illiterate, %; primary level education, %; P , .).
The overall probability of livestock loss also affected attitudes, with a lower probability of livestock loss being linked to a more positive attitude (P = .).

Discussion
Household surveys provide insight into how people live and interact with wildlife. We found that leopards caused the majority of livestock loss around Bardia National Park, in contrast to Acharya et al. () Tamang & Baral (). Tharu people reported minimal losses, although they owned similar numbers of cattle to people of other ethnic groups. This could be related to the Tharu's long experience of living with wildlife as an Indigenous group and their better livestock husbandry practices, such as more effective enclosures (Kolipaka et al., ). Distance to Park boundary is an important determinant of predator attacks on livestock: attack probability increased at distances below - km in Bardia and in Waza National Park, Cameroon (Studsrød & Wegge, ; Van Bommel et al., ).
Households in the northern sector suffered considerably higher economic damage compared to other sectors, which can be attributed to poor husbandry techniques, such as open sheds. Villagers with a lower socio-economic status lost significantly more livestock compared to villagers who could afford better protection measures and husbandry techniques such as predator-proof corrals (Saberwal et al., ). People from marginalized groups are severely affected by such financial losses (Manral et al., ).
Economic loss caused by predator attacks in the study area was comparable to losses resulting from other causes, specifically two adverse natural events: a flood in , which caused a rise in livestock deaths, and a bird-flu outbreak in July , which led to high losses of poultry. In other areas, studies have shown that even where livestock predation is considered to be a major cause of economic loss, other factors such as disease and theft actually contributed more to overall economic losses than predator attacks (Dar et al., ; Tumenta et al., ).
The difference in attitudes between respondents from the western sector (% had a positive attitude towards wildlife conservation) and those from the northern sector (c. % were positive), is probably a consequence of the difference in numbers of livestock lost between these two sectors. In addition, interviewees from the northern sector were generally dissatisfied because this area was incorporated into the Park's buffer zone more recently (in ) and residents have not benefitted from the buffer zone programme to the same extent as people in other sectors, where -% of Park revenue has been allocated to community development activities. Although people's attitudes towards wildlife can be influenced by predator attacks and other wildlife-related financial losses (Røskaft et al., ), in general people are more tolerant of wildlife if they derive benefits from a protected area (Allendorf et al., ; Baral & Wegge et al., ). Therefore, active involvement of local communities in planning, executing and managing smallscale conservation projects leads to a positive attitude towards conservation (Nepal, ). Despite the differences between the study sectors, c. % and c. % of all respondents had a positive attitude towards conservation even when leopards or tigers, respectively, had killed their livestock. The fact that tigers were regarded more favourably is a reflection of the cultural values prevalent in this region (Bhattarai & Fischer, ; Kolipaka et al., ): people from Bardia believe that tigers are the vehicle of Durga, the goddess of might, and should not be harmed (Bhattarai & Fischer, ), whereas leopards do not benefit from tolerance based on religious beliefs. Respondents educated to a higher level and with a greater degree of self-sufficiency had a more positive attitude towards conservation, which confirms findings of other studies (Allendorf et al., ; Sarker & Røskaft, ; Sogbohossou et al., ; Tumenta et al., ; Bhattarai & Fischer, ). In Bangladesh, wealthy respondents favoured wildlife conservation . times more than their less wealthy peers (Sarker & Røskaft, ). People relying on a single source of income or few resources are more vulnerable to the impact of depredation (Ogra, ; Dickman, ). More educated people are generally less dependent on natural resources for their sustenance than their less educated neighbours (Dickman, ). Education can therefore be an important tool in wildlife conservation at the local scale (Nepal & Weber, ), and conservation authorities could also promote existing Indigenous knowledge and technologies to reduce negative human-wildlife interactions (Kolipaka et al., ). Women had a less positive attitude towards conservation than men, which may be linked to the risks they are exposed to when collecting forest products, making them more vulnerable, and as a result, more vigilant, in the vicinity of wildlife (Allendorf, ; Bhattarai & Fischer, ). However, we cannot generalize this because of the low number of women amongst the respondents in our study. Similar to our results, Carter et al. () reported that women, less educated persons and people from marginalized groups have negative attitudes, specifically towards tigers.
We expect that our findings will be of value to Bardia wildlife managers and other conservation authorities in the region, to help predict where predation by tigers and leopards is most likely to occur, and design intervention strategies that can reduce financial losses caused by predators (Kansky & Knight, ). When planning mitigation measures in and around Bardia, managers should consider the specific behavioural traits of the felid species involved, such as taking cover in dense vegetation and attacking in dark. To reduce the impact of livestock depredation we recommend () improvement in enclosure and herding practices, () reduction in the number of livestock by diversifying the economy, () implementation of a community-based livestock insurance programme, and () establishment of an early warning system such as deployment of motion sensors around corrals and mobile alerts. Chandra Kandel, Ramesh Thapa and Ashok Bhandari for their support and permission to work in Bardia; Rabin Kadariya, Ambika Prasad Khatiwada, Shree Ram Ghimire and Shailendra Yadav from the National Trust for Nature Conservation, Bardia for providing logistical support; the communities residing in the buffer zone for their participation; Hemanta Acharya and his team for conducting interviews; two anonymous reviewers for their critiques; and Barbara Croes for language editing.
Author contributions Conception and study design: all authors; data collection: SKU; statistical models and data analyses: CJM and SKU; writing and revisions: all authors.

Conflict of interest None.
Ethical standards This research complies with the Oryx ethical guidelines. The interview questionnaire was reviewed and approved by the Institute of Environmental Science, Leiden University, and we obtained verbal consent from each respondent before starting the interview.