Introduction

Twenty percent of felid species worldwide are endangered or critically endangered (IUCN 2011) due to habitat loss, hunting (Robinson and Bodmer 1999), parts in demand for medicinal uses (Alves et al. 2007), and because they are often killed in response to livestock predation (Michalski et al. 2006; Monroy-Vilchis et al. 2009a; García-Alaniz et al. 2010), particularly around protected areas too small to maintain viable populations (Hoogesteijn 2001; Holmern et al. 2007; Van et al. 2007; Kissui 2008; Gusset et al. 2009). Consequently, effective conservation strategies for large cats depends in part on understanding and resolving conflicts between cats and people, especially those landowners within or near protected areas. Understanding these conflicts requires knowledge about the actual impacts of predation events as landowners’ perceptions of impacts can be biased. It is also important to understand perceptions and tolerance of local people to felids (Marker et al. 2003; Jackson and Wangchuck 2004; Zimmermann et al. 2005; Gusset et al. 2008; García-Alaniz et al. 2010). An understanding of economic impact, as even small losses might be economically significant to an individual landowner, can be important in communities with subsistence economies. Thus, management strategies to reduce felid-human conflicts must be based on empirical knowledge and local experiences.

Economic loss as a consequence of predation of domestic species has been documented worldwide (Stahl et al. 2002; Ogada et al. 2003; Patterson et al. 2004; Wang and Macdonald 2006; Gazzola et al. 2007; Van et al. 2007), including studies of felids in the Americas (Mazzolli et al. 2002; Michalski et al. 2006; Azevedo and Murray 2007; Palmeira et al. 2008; García-Alaniz et al. 2010). However, only one study has been conducted in northern Mexico (Rosas-Rosas et al. 2008). These studies generally quantified the effects of physical condition, health, age, behavior and sex of the predator and the prey as risk factors (Polisar et al. 2003; Inskip and Zimmermann 2009). A few have assessed relationships between livestock management practices and predations (Ogada et al. 2003; Michalski et al. 2006; Wang and Macdonald 2006; Van et al. 2007; Gusset et al. 2009), with even fewer suggesting predation sites were associated with topographic features. The identification of environmental conditions associated with livestock predations could be used to generate predictive spatial models and identify high-risk areas, allowing the implementation of conflict mitigation efforts.

The objectives of our study were: 1) to evaluate the relationships between livestock predation risk and local topographical factors, 2) to assess livestock management practices in this region central Mexico. Specifically, we intended to identify areas prone to attacks, as well as to develop specific strategies to minimize human-felid conflicts.

Study Site

The Balsas River Basin (17°00′–20°00′ N, 97°30′–103°15′ W) is bordered to the north by the Trans-Mexican Volcanic Belt, to the south and west by the Sierra Madre del Sur and to the east by the Sierra Madre de Oaxaca, average elevation 1,000 m (Challenger 1998). This area includes parts of the states of Michoacan, Guerrero, Mexico, Morelos, Puebla and Oaxaca. Our study was conducted in an area of 3,864 km2 along the southern part of the state of Mexico (Fig. 1). This is the most ecologically important zone in the state with 53 registered mammal species (Monroy-Vilchis et al. 2011a). Tropical deciduous forests of oak and pine-oak cover 50 % of the area, with the remainder agricultural and pasture land (SEMARNAP-INEGI-UNAM 2001). The climate is warm, Aw”0(w) (e)g, and semi-warm, A(C)W2(w)(i’)g (Monroy-Vilchis et al. 2011b). Livestock production is mainly cattle (118,700 animals), goats (61,990) and sheep (42,120) (INEGI 2009). Repeated conflicts between pumas (Puma concolor) and local people due to livestock predations in the area have resulted in the killing of 32 pumas through 2006 (Monroy-Vilchis et al. 2009a).

Fig. 1
figure 1

Southern zone of the Mexico State, in the sub-province of Balsas River Basin

Full size image

Data Collection and Analysis

From 2002 to 2008, we conducted interviews with livestock owners in 14 communities. The only criterion for the selection of interviewees was that they had had predation events on their farms. Each respondent was informed of the purpose of the study prior to a single interview. We recorded the data of felids killed in response to predations and verified in the field presumed attacks reported from 2005 to 2008. The carcasses were examined as per Hoogesteijn (2001) and Shaw et al. (2007). We also recorded data on livestock management practices, date of attack, characteristics of dead animals and the sites of attacks, as well as the perceptions of livestock holders.

Carcasses described in interviews or examined in field were included in the analyses if there was evidence of tracks, scat, bites to the throat or neck, or dragging (Azevedo and Murray 2007). We placed cameras (CamTrakker® and Recon outdoors® brands) with passive infrared sensors at sites where attacks were reported to validate information provided during interviews.

We analyzed the effect on the three most frequently predated domestic species (cow, sheep, goats) by recording the relative availability of each species, from which we obtained the expected frequency. By subtracting the expected frequency from that observed in each case we produced a use index to which we applied the G test.

The location and elevation of attack sites were determined using ArcView 3.2 software (Environmental System Research Institute 2000). Using Idrisi Andes software (Clark Labs 2006) we generated maps of distance to human settlements (>300 inhabitants), roads, areas of dense vegetation and cliffs with slope >60º (INEGI 2001, 2005); these variables have been shown to influence puma and jaguar movements (Monroy-Vilchis et al. 2009a). These maps and a digital elevation model were processed in raster format with a pixel size of 30 × 30 m.

We used three modelling techniques: Environmental Niche Factor Analysis (ENFA; Hirzel et al. 2004), Mahalanobis Distance (MD; Clark et al. 1993), and Maximum Entropy (MAXENT; Phillips et al. 2006). ENFA is based on a multivariate description of species occurrence locations. It estimates species niche more explicitly based on the magnitude of the difference between the species mean and the entire range of environmental conditions observed, and thus marginality and specialization. We implemented the ENFA algorithm using BioMapper4 (Hirzel et al. 2004). MD is a measurement of dissimilarity and represents the standard squared distance between a set of sample variables and an ideal habitat as estimated from a set of animal relocations (Clark et al. 1993). We calculated for each pixel within the study area the multivariate distance between the environmental conditions in the pixel and those in the sites of attacks; the MD greatest value is most similar to attack sites and represents the highest risk. We calculated the MD using the algorithm available in Idrisi Andes software (Clark Labs 2006). MAXENT is a machine-learning method; the multivariate distribution of suitable habitat conditions in environmental feature-space is estimated using the best approximation of an unknown distribution with maximum entropy (the most dispersed) subject to known constraints. To avoid model over-fitting, we developed our model using settings outlined in Phillips and Dudík (2008) and employed a logistic output with suitability values ranging from 0 to 1.

The performance of each model was evaluated by calculating the area under the curve (AUC; Hanley and McNeil 1982). We used the Idrisi Andes AUC module in a bootstrap procedure similar to the one implemented in MAXENT (Phillips et al. 2006) to calculate the AUC value. The approach outlined by Niels and ter Steege (2007) was used to test the statistical significance of AUC, considering a total of 1,000 null models.

We chose these modelling techniques because they use presence-only data and produce the most consistently reliable results (Elith et al. 2006; Tsoar et al. 2007; Hernandez et al. 2008; Roura-Pascual et al. 2008). However, several recent studies have investigated the efficacy of different methods for modeling species distributions (Elith et al. 2006; Tsoar et al. 2007); while some methods are more effective at predicting species’ distributions than others, no single modeling method is most effective in all situations (Hernandez et al. 2008). To deal with this variability, one solution is to develop models using multiple methods and to identify common areas of consistent prediction (Anderson et al. 2003; Araújo et al. 2006). These common areas incorporate modeling uncertainties to produce more reliable estimates of species’ potential distributions (Hartley et al. 2006). Consequently, we produced a final ensemble spatial model of predation risk, considering common areas identified from the three individual models (Araújo and New 2007; Roura-Pascual et al. 2008).

To obtain the ensemble model (EM), AUC weights from each model were used in a weighted average (Marmion et al. 2009). AUC was also used to evaluate the EM. To improve interpretation of the final EM, we reclassified the original results (ranging from 0 to 100) into three levels of risk: high (67-100), medium (34-66), and low (0-33).

Linear regressions were performed between risk and each variable to identify which variables best predicted predation risk. Subsequently we also divided each variable into five equal intervals, assigned each record to an interval and calculated a “use index.” Finally we used a G test to compare observed frequencies to expected frequencies considering the availability of each variable at each interval.

Results

From August 2002 to October 2008, we interviewed 156 inhabitants from 14 communities, of which 52 reported livestock losses attributed to felid predation. From evaluating the evidence from the interviews and verified by field inspections, we selected 28 of these cases, all attributed to puma.

We obtained detailed information in 16 confirmed cases on management practices, characteristics of attacks and domestic prey, attack locations and human tolerance toward felids. In 12 of the 16 cases households were within 1,500 m, four had water sources nearby, and all where associated with gullies. Fourteen householder maintain livestock on cliffs and in forested areas far from human settlement. In nine cases animals grazed in large groups (>20 individuals) and were not stabled at night. All households checked livestock weekly or monthly but only seven sheltered offspring in stables near their homes, although only for a week after birth.

We recorded 996 domestic animals (357 cattle, 176 sheep, 359 goats, 67 swine, 17 horses, 12 mules and 8 donkeys) present at the attack sites. The number of animals per livestock holder ranged from 10 to 136, with a mean of 55 (SD ± 44.17). Two hundred and eighteen animals were attacked: 59 cattle, 103 goats, 50 sheep and 6 donkeys. Sixty-seven percent of losses occurred at four ranches where attacks were recurrent. The number of predated animals at these sites ranged between 22 and 68, with a mean of 37 (SD ± 20.94), mostly goats.

Cattle weighed 200–700 kg and 70 % were adult. Sheep and goats weighed 20–50 kg and 72 % were adults. The mean weight of donkeys attacked was 150 kg. None of the prey were reported as sick or in poor physical condition. Goats were predated more often than expected (G = 23.16, d.f. = 2, p < 0.05; Fig. 2). Primary evidence confirming attacks (28) included bites to the throat (60 % of cases) or to the back of the neck (10 %), footprints, sightings and/or scat (30 %); dragging of prey (37 %) and cached carcasses (34 %). We obtained 19 photographs of pumas from cameras near the attack sites.

Fig. 2
figure 2

Use index for domestic species depredated by puma in the southern State of Mexico

Full size image

Generally, livestock holders did not take measures to protect their animals after an attack. Three livestock holders stated killing felids could reduce livestock predations but recognized it as only temporarily effective. Two owners considered enclosing livestock at night and monitoring them during the day effective in reducing predations. Nevertheless, only 50 % of livestock holders experiencing livestock predations considered enclosing livestock an effective technique for reducing predations. The remaining 50 % considered killing predating animals the only effective technique. Consequently, 40 pumas were killed between 1993 and 2008.

We used 28 known sites of attack to model predation risk (Fig. 3). The ensemble model had high AUC value (0.906, p < 0.0001). The highest risk area encompassed about 107 km2 (3% of the study area). It was observed that predation risk was positively associated with altitude and distance to roads or human settlement, whereas it was negatively associated with distances to cliffs and dense vegetation. Distance to cliffs was the most important variable in all three analyses and therefore may be considered the best predictor of risk (Table 1). In addition, attacks occurred with higher frequency at altitudes from 1,700 to 2,000 m. (G = 17.15, d.f. = 4, p < 0.05) and closer (<2 km) to dense vegetation (G = 11.28, d.f. = 4, p < 0.05).

Fig. 3
figure 3

Spatial model of predation risk in the southern State of Mexico

Full size image
Table 1 Variables contributing to the predation risk. Showing r, ENFA and MAXENT values
Full size table

Discussion

Predation on livestock is frequently overestimated (Holmern et al. 2007; Kissui 2008; Gusset et al. 2009). Consequently, it was important to correctly identify the predation events. We achieved this through in-field verification of carcasses and detailed descriptions of evidence provided during interviews. Evidence such as tracks and scat of felids, bites to the throat or the back of the neck, and dragging of prey, are reliable indicators (Azevedo and Murray 2007). Photographs of puma at the attack sites provided further identification. Finally, we did not include in our analyses animals whose carcasses were not found.

A previous study in this region indicated that livestock contributed only 8.2% of the relative biomass of puma prey, mainly remnants of cows and goats (Monroy-Vilchis et al. 2009b), which supports the conclusion that felids in this region kill fewer domestic animals than livestock holders believe. In all confirmed cases, puma was considered the responsible predator, which was supported by evidence found in more than half of the attacks. On the other hand, livestock holders identified this species in all sightings. Notwithstanding, it is important to note that in spite of Panthera onca presence in this area (Monroy-Vilchis et al. 2008), this felid is less abundant than puma by 1:3 (Monroy-Vilchis et al. 2009a; Soria-Díaz et al. 2010). In three cases carcasses were bitten on the back of the neck, which could be attributed to jaguar (Hoogesteijn 2001).

Although most the domestic prey were adults, we noted that livestock holders do not keep records of livestock births and commented that occasionally offspring disappear. It is likely that some of these losses are due to predations and that predation of young may occur more frequently than perceived. None of the prey were reported to be sick, which differs from data recorded in other studies (Azevedo and Murray 2007), suggesting that predation habits respond to the fact that livestock, independently of individual conditions, represent easy prey, probably because the domestication process has inhibited anti-predatory behavior (Linnell et al. 1999). We found predation selection for goats, possibly to reduce risk of injury to pumas.

Free range grazing of livestock is common in central and southern Mexico and throughout Latin America (Challenger 1998), most often associated with the subsistence economies of villagers who do not have the means to guarantee protection and fodder for their livestock. Free range grazing is a way to maintain livestock with minimal investment. In the study zone most animals graze in large groups, often far from human settlement, which implies greater risk to pumas as group size is positively correlated with predation (González-Fernández 1995; Jackson et al. 1996; Treves et al. 2004; Van et al. 2007), with groups >20 individuals most affected (Michalski et al. 2006). The lack of night shelter increases the livestock vulnerability in relation to the nocturnal habits of puma (Mazzolli et al. 2002; Holmern et al. 2007). Although offspring are protected in some cases, this is usually for no longer that a week and thus ineffective against predation. Michalski et al. (2006) note that to be effective protection must last for at least the first 3 months of life for cattle.

The impact of human activity on puma populations is high, as 40 individuals killed in 15 years represents an average of more than two individuals per year. In the study area people frequently poison carcasses to kill pumas, which negatively impacts non-targeted species. Livestock holders also kill other species of carnivores they encounter, which puts small felids at risk, such as ocelots, margays and jaguarondis (Sánchez et al. 2002).

Generally, in regions where livestock predation has been studied, it has been found that the impact is low. In the El Pantanal and El Cerrado regions of Brazil, the percentage of deaths by predation was 0.5 and 0.4 % of the total livestock population, respectively (Azevedo and Murray 2007; Palmeira et al. 2008). The same percentage was found in the Llanos of Venezuela (González-Fernández 1995). Similarly, 1.2 % was reported in the Brazilian Amazon (Michalski et al. 2006), 0.7 % in the province of Arezzo in Italy (Gazzola et al. 2007), 2.3 % in Wangchuck National Park, Bhutan (Wang and Macdonald 2006), 2.4 % in the Tsavo National Park, Kenya (Patterson et al. 2004), 4.5 % in Serengeti National Park, Tanzania (Holmern et al. 2007), and 7.1 % in Nepal (Jackson et al. 1996). We found low percentage of livestock losses at the regional level (SAGARPA 2009): 0.05 % in cattle, 0.26 % in goats, 0.44 % in sheep and 0.12 % for the three species combined. Nonetheless at the local level, losses accounted for 20 % of livestock present in the sites of attacks, representing 17 % of the mean number of animals per owner.

Although the percentage of livestock loss by predation is relatively low, because these are marginalized communities these losses are considerable as they account for about 23,846 US dollars or 17 % of the total value of livestock present in the sites of attacks. This suggests a high local impact associated with subsistence economies, common in rural localities in Central and South America, Asia and Africa. It is important to note that livestock herding in study area is not for commercial purposes and is not the main productive activity. Thus livestock predation directly affects the household economy, which aggravates peoples’ perceptions of felids.

The habitat models used have been generally described as habitat suitability models, with model output interpreted as potential areas of species’ distribution (Knick and Dyer 1997; Corsi et al. 1999; Cuesta et al. 2003; Rodríguez-Soto et al. 2011). This study applies these models to other ecological contexts. With the EM model we obtained a good approximation for predation risk. We developed our model in the framework of ensemble forecasting (Araújo and New 2007), incorporating model uncertainties and producing a more reliable estimate of potential predation risk (Hartley et al. 2006). We based our ensemble model on three models which gave similar results on puma ecology, with distance to cliffs, distance to dense vegetation and distance to roads being the most important variables explaining predation risk.

Our results are consistent with the literature relating environmental characteristics to predation risk (Stahl et al. 2002; Treves et al. 2004). For example, characteristics such as distance to cliffs, provide important areas of shelter for felids because of their inaccessibility to humans. These results differ from other studies where the main variable has been the distance to protected areas (Van et al. 2007; Gusset et al. 2009). This difference might be attributed to previous studies generally being conducted in prairies, with limited altitudinal ranges and little variation of landscape features (Ogada et al. 2003; Patterson et al. 2004; Azevedo and Murray 2007; Gusset et al. 2009). In these studies geographic features are unlikely to be determining factors.

We found a high percentage of losses were concentrated in a few sites. Similar patterns have been reported for other regions where livestock is attacked at a small percentage of farms (Jackson et al.1996; Mazzolli et al. 2002; Stahl et al. 2002; Michalski et al. 2006; Gazzola et al. 2007; Palmeira et al. 2008; Rosas-Rosas et al. 2008). These locations are known as hot spots and their existence suggests a site effect. This study is one of the first to evaluate and to confirm this site effect through the influence of physiographic variables at the sites of attack. The importance of characterizing attack sites using predictive models is clear: modeling allows focusing prevention efforts and mitigation measures in high-risk areas which optimizes resource use. The regional scale analysis can be used for livestock management decisions: the risk of conflict must be a basic criterion in the selection of suitable grazing sites and should be considered in felid management and conservation plans. In addition, the smaller-scale identification of specific sites where predations may occur would be useful for livestock holders.

The most feasible measure to mitigate conflicts would be to emphasize management practices. Immediate efforts should be made at sites identified as “hot spots” which could reduce livestock losses by >60%. We demonstrated that physiographic variables greatly influence predation risk (e.g., presence of gullies and proximity to cliffs), therefore we recommend moving grazing sites at least 2 km from these areas. In addition, it will also be important to promote programs to fund the construction of enclosures near communities to shelter livestock. Technical advice such as keeping livestock away from forested areas and promoting intensive management (e.g., rotational grazing), could improve economic sustainability for livestock holders. Similarly, improved shelter for offspring and veterinary advice on managing livestock health would have an important positive effect on production as losses from disease would be reduced. Although not generally perceived as important, livestock losses from disease may exceed losses from predation.

The success of these measures will depend on taking into account communities’ interests and initiatives, such as the potential to incorporate felids into some of their productive activities, such as ecotourism. Economic stability produced by an improvement in incomes through alternative productive activities should increase community tolerance towards carnivores. In addition, since the educational level in the study region is only an average of three years (Monroy-Vilchis et al. 2011b), ecological information relating to felid conservation should be made readily available, particularly highlighting that predation is not necessarily the principal cause of livestock losses.