Alleviating human–wildlife conflicts: identifying the causes and mapping the risk of illegal poisoning of wild fauna



1. Illegal human behaviour such as those affecting natural resource use or resulting from human–wildlife conflicts threaten the sustainable management of ecosystems and the conservation of biodiversity worldwide. However, the frequently scarce and incomplete data owing to the sensitive nature of illegal activities pose a challenge to developing tools to properly understand and prevent those activities.

2. We used species distribution models to identify factors related to a prominent illegal activity, wildlife poisoning, and to produce detailed, spatially explicit maps of the risk of occurrence in NW Spain. We alleviated the constraints of imperfect information and occurrence of absences by using presence-only methods, that is, maximum entropy modelling (MaxEnt). To our knowledge, this is the first time that this method has been used in the context of illegal activities affecting wildlife.

3. A total of 112 poisoning events involving 228 individuals of 25 different species were reported in the study area from 2000 to 2010. Most of the reported deaths (90·8%) were birds of prey (52·6%) and mammalian carnivores (38·2%), of which 95·2% were scavengers. Illegal poisoning affected eleven species classified as endangered at national and/or global level.

4. Our models highlighted the perceived risk of livestock predation by wolves Canis lupus, although not by bears Ursus arctos, as a major motivation for poisoning. The existence of protected areas was positively correlated to this illegal practice, while socioeconomic factors had less influence on predicting its occurrence. Over 56% of the study area was predicted to be under risk of illegal poisoning.

5.Synthesis and applications. We demonstrate a new use for presence-only models, illustrated using MaxEnt, to assist conservation managers dealing with illegal activities. This approach allows the main causes of an illegal practice to be identified and generates spatially explicit risk maps. Managers can take advantage of this modelling approach to allocate the scarce resources available in conservation to key sectors and locations. In our study system, actions against illegal poisoning should aim to resolve the potential conflict existing between cattle-farming and wolves, especially in protected areas.


Illegal human behaviour affecting natural resource use (e.g. overfishing, illegal hunting, poaching) or resulting from human–wildlife conflict (e.g. retaliatory killing) threatens the sustainable management of ecosystems and the conservation of biodiversity worldwide (Gavin, Solomon & Blank 2010; St John et al. 2011). Besides severe environmental damage, this illegal behaviour has important consequences at economic (e.g. revenue losses) and social (e.g. food supplies) levels (Gavin, Solomon & Blank 2010). The explicit incorporation of illegal behaviour in management systems improves the outcome of conservation and natural resource management (Bunnefeld, Hoshino & Milner-Gulland 2011). Among different, illegal human activities (e.g. shooting, trapping), the deliberate poisoning of wildlife is a serious threat to conservation efforts (Newton 1979; Berny 2007; Guitart et al. 2010).

Poisoning has been used for centuries in many regions of the world to control species that threaten human lives, lifestyles or livelihoods (e.g. crops, livestock, game; Newton 1979; Woodroffe, Thirgood & Rabinowitz 2005). Formerly used as a legally sanctioned tool, the widespread use of poisoning compounds for controlling wildlife has been heavily restricted in many countries during the last few decades (e.g. European Council Directive EC 91/414; United States Framework for Ecological Risk Assessment EPA/630/R-92/001). Accordingly, legal poisoning compounds are now mostly restricted to pesticides authorized to control agricultural pests (i.e. insecticides and rodenticides; Berny 2007). However, despite an increasing awareness of the lethal effects of pesticides on wildlife, their use still causes thousands of unintentionally poisoned, non-target animals every year worldwide (e.g. 6–38% of all poisoned animals in some European countries; Berny 2007; Olea et al. 2009). In addition to this accidental mortality, these compounds are sometimes used illegally to intentionally poison wildlife (e.g. 25–100% of all poisoned animals in some European countries; Berny 2007). Moreover, some banned compounds (e.g. strychnine and endrin in the European Union) are also used with the same aim (Martínez-Haro et al. 2008).

Deliberate poisoning continues illegally in many regions of the world mainly to eliminate predators (Wobeser et al. 2004; Kissui 2008; WWF/Adena 2008; RSPB 2009; Kalaivanan et al. 2011). As the commonest way to use poison is by placing poisoned baits (e.g. meat, carcasses, eggs; WWF/Adena. 2008; RSPB 2009) in the open, the illegal use of poison affects not only the target species but also many other animal species, including domestic pets and even humans (Wobeser et al. 2004; Berny et al. 2010). Thousands of animals of many different species are killed yearly around the world by the illegal use of poison (e.g. 60 000 birds in northern Argentina in 1997; 80 eagles in western Canada in 1993–2002, 433 birds of prey in Scotland in 1998–2008, 7261 animals in Spain in 1990–2003; Goldstein et al. 1999; Wobeser et al. 2004; WWF/Adena. 2008; RSPB 2009). Despite the large numbers of animals confirmed to have been poisoned, only a small percentage of the total illegal poisoning events are detected (e.g. 3–15% of the total in Spain; WWF/Adena. 2008). Moreover, not all of the detected events can actually be confirmed as poisoning events for various reasons (e.g. insufficient sample, victims too decomposed to analyse; Wobeser et al. 2004; Berny 2007). Thus, the real dimension of this threat is expected to be much bigger than that represented by the data (i.e. 18–68% of all suspected poisoning cases depending on the country; Berny 2007).

Decline in population and reduction in range are consequences of the use of illegal poisoning in a range of different species (Newton 1979; Whitfield et al. 2003). Poison use is thought to be the main cause of the local extinction of several species (e.g. Griffon vultures Gyps fulvus in Romania in the 1970s, the grey wolf Canis lupus in many regions of America, Asia and Europe in the 20th century; Newton 1979; IUCN 2010). Poisoning is also considered to be one of the causes of the global extinction of the Guadalupe caracara Caracara lutosa in 1900 and the Javan tiger Panthera tigris sondaica in the 1980s (IUCN 2010). Today, illegal poisoning challenges the conservation of several highly threatened species both at local (e.g. the brown bear Ursus arctos in Spain, the hen harrier Circus cyaneus in the UK; Etheridge, Summers & Green 1997; WWF/Adena. 2008) and global scales (e.g. Egyptian vulture Neophron percnopterus, imperial eagle Aquila adalberti; red kite Milvus milvus; IUCN 2010). Moreover, the impacts of illegal poisoning may extend beyond populations to affect entire ecosystems, through altering trophic cascades, for example (Woodroffe, Thirgood & Rabinowitz 2005).

In a world where reducing the rate of biodiversity loss has become a key target (incorporated into the Millennium Development Goals), the deliberate poisoning of wildlife is a serious threat that needs to be adequately addressed. Successful strategies to combat illegal poisoning include legal prosecution of the poisoners and educational activities to foster a change of attitudes against illegal poisoning among the general public, but especially in farmers and hunters (i.e. the two sectors most likely to use poison; WWF/Adena. 2008; RSPB 2009). However, the resources available for the adequate implementation of actions against illegal activities are frequently limited (Keane et al. 2008). Accordingly, there is an urgent need for the efficient allocation of such scarce resources to maximize the efficiency of these actions (Keane et al. 2008). The identification of those areas most at risk of being illegally poisoned as well as identifying the main factors behind this practice would greatly facilitate this process. In fact, the use of maps of risk has been widely recognized as a fundamental tool to combat illegal poisoning (CNPN 2004). Moreover, model visualization has been highlighted as a powerful tool in effective management of natural resources (Bunnefeld, Hoshino & Milner-Gulland 2011).

Species distribution models (SDMs) are numerical techniques that relate observations of species’ occurrence with environmental and/or spatial information to predict the species’ distribution across a landscape (Elith & Leathwick 2009). One type of SDM uses presence-only data, which compares presence records with background or pseudoabsence points (Elith & Leathwick 2009). Among these techniques, the maximum entropy model (MaxEnt; Phillips, Anderson & Schapire 2006) has become one of the most extensively used forms of SDM, usually outperforming other similar tools (Elith & Graham 2009). Applications of MaxEnt range across many disciplines (e.g. ecology, evolution, conservation; Elith et al. 2011), and it has proven to be a highly useful tool for analysing sparse and/or limited-in-coverage data (e.g. museum databases; Elith et al. 2011). To date, MaxEnt has mainly focused on species distributions (but see Parisien & Moritz 2009), but it could also be applied to conservation scenarios where information is scarce or incomplete. Data on illegal wildlife poisoning is suitable for statistical treatment with MaxEnt, as information on poisoning events is frequently incomplete. This is because of some cases not being located, and the impossibility of confirming certain cases as poisoning events, all of which constrains the use of reliable absence data (Elith et al. 2011).

We utilize a database of poisoning events resulting from an unprecedented collaboration between NGOs, public administrations and specialized laboratories to: (i) examine the socioeconomic and environmental factors underlying the occurrence of illegal poisoning of wild fauna and (ii) produce a spatially explicit map of risk of illegal wildlife poisoning using SDMs.

Materials and methods

Study Area

The study area covers 24 639 km2 in NW Spain, corresponding mainly to the Cantabrian Mountains. The region is recognized for its high biodiversity, holding several protected areas (i.e. 10 Biosphere reserves, 18 sites included within the Natura 2000 network of protected areas of the European Union, one National park and eight Regional parks; Fig. 1). The landscape consists of a wide variety of habitats, including oak Quercus spp. and beech Fagus sylvatica woodlands, scrublands and pastures. The area supports several species of conservation concern such as the highly endangered Cantabrian brown bear U. arctos, with a population of around 190 individuals (FOP 2011), and the globally endangered Egyptian vulture N. percnopterus (c. 175 breeding pairs; Mateo-Tomás, Olea & Fombellida 2010). The main carnivore species within the study area are the brown bear and the Iberian wolf C. lupus signatus, whose population are estimated to be at least 200 individuals (Blanco 2003). Extensive livestock rearing is the main activity in most of the area, including the protected areas. Livestock, mainly cows, stay in the field for 6–7 months per year (Olea & Mateo-Tomás 2009). Hunting, focused on large wild ungulates, is another important activity in the region (Mateo-Tomás & Olea 2010).

Figure 1.

 The study area in NW Spain contains many protected areas of regional, national and international importance. Note that many of them overlap (i.e. dark green: national and regional and European, orange: European and global, brown: national and regional, European and global). White lines show administrative boundaries.

The use of poison to kill wild predators has been banned in Spain since 1989 (Ley 4/1989). Offences are handled according to the Ley Orgánica 15/2003 and can attract a fine of up to €288 000, up to 2 years in prison and always result in a ban from hunting or fishing for at least 1 year (CNPN 2004). Additionally, the Spanish law (Real Decreto 3349/1983) restricts the use of toxic formulations of pesticides to authorized personnel (Martínez-Haro et al. 2008). However, illegal poisoning of wildlife has been reported in the study area, mainly by leaving meat or carcasses laced with banned compounds (i.e. strychnine, aldicarb and carbofuran; Sánchez-Barbudo 2010) in the open.

Data Acquisition

Data on illegal use of poison within the study area were obtained for the period 2000–2010 from specialized veterinary laboratories (i.e. Instituto de Investigación en Recursos Cinegéticos, IREC), official databases (i.e. SEPRONA, Junta de Castilla y León) and the Antidote Program. This program was created in 1997 by eight Spanish NGOs with the main objective of fighting against the illegal use of poison in Spain (BVCF 2011). The program maintains, manages and publicizes a free hotline to report any events suspected to be wildlife poisoning occurrences.

To minimize those potential drawbacks related to differences in sampling intensity between the local administrations existing in the study area, we obtained data from other sources such as unpublished data from laboratories, universities, private institutions, NGOs and press releases. The data from different sources were scrutinized carefully. We considered a poisoning event to have taken place when dead or sick animals or baits laced with poison were found (WWF/Adena. 2008; RSPB 2009). Records that involved several animals killed by the same poisoned bait (i.e. same toxic and similar spatiotemporal location, within 1 km and 1 month) were considered as a single event to avoid duplicating records (Whitfield et al. 2003). We only considered those data with reliable and complete toxicoepidemiologic information (i.e. illegal poison use confirmed by analysis performed in a specialized veterinary laboratory revealing, for example, the use of banned compounds; Martínez-Haro et al. 2008). Most poisoning events (98%; N = 112) involved poisoned animals; the rest (2%) were illegal baits using banned compounds. We used chi-square tests of homogeneity to test the null hypothesis that those species with scavenging habits were affected by illegal poisoning in the same way as non-scavenging species.

From all the poisoning events under consideration (N = 112), those with a precise location (e.g. coordinates, precise description and/or name of the place; N = 45) were selected to perform the MaxEnt models with an adequate resolution (i.e. 1-km pixel). This sample size was sufficient for use in MaxEnt, as it has shown good performance on small sample sizes (i.e. N < 10) (Phillips & Dudik 2008).

To model the risk of illegal use of poison within the study area, we considered 16 variables (predictors) potentially related to the existence of poisoning events (Table 1). The potential existence of human–wildlife conflicts within the study area was considered by including variables on the distribution of both the main predators (i.e. wolf and brown bear) and the human population and its main economic activities (e.g. livestock rearing, hunting). The presence of protected areas was also considered as an important factor influencing the risk of illegal poison use through either enhancing human–wildlife conflicts or persuading people to acting illegally (Woodroffe, Thirgood & Rabinowitz 2005). We included vegetation coverage and topography as additional variables to account for the influence of habitat structure on illegal poison use. Whenever possible, we used information from 2005 as representative of the entire period (i.e. 2000–2010), using the data available from other years within this period when information from 2005 was lacking. All the variables were transformed to a 1-km-pixel final grid. This resolution was considered sufficient to provide a detailed picture of the study area according to the modelled issue (i.e. illegal use of poison) and to characterize the habitat features evaluated by a potential user of illegal poison. At the same time, this resolution accommodated a marginal error of the poison locations in terms of both the measurement technique (e.g. ±15–100 m for GPS; Kowoma 2011) and movement of the poisoned animals from the bait location (Whitfield et al. 2003).

Table 1.  Main variables considered for modelling the illegal use of poison in the study area. Those variables finally used to run the models after reducing multicollinearity (i.e. high correlation between variables, rS > |0·5|) are shown in bold
  1. *Spanish Institute of Geography, ING.

  2. †III Spanish Forest Inventory.

  3. ‡Official databases of Gobierno de Cantabria, Junta de Castilla y León, Principado de Asturias and Xunta de Galicia.

  4. §National Institute of Statistics, INE.

  5. ¶Data obtained at municipality level.

Habitat characteristics
 Slope*Slope of the terrain (degrees)Rough terrain can make access difficult preventing the use/detection of poison.
 Elevation*Elevation of the terrain (m.a.s.l.)Very high elevations within the study area correspond to bare rock and areas of difficult access.
 ForestPercentage of surface covered by forestHabitat structure can determine both livestock, game and predators presence, conditioning the location of poison.
 Pasture†Percentage of surface covered by pasture
 ShrubPercentage of surface covered by shrubs
 Urban†Percentage of urbanized surface
 Protected areasDegree of environmental protection of the territory. (0: non-protected, 1: locally protected, 2: regionally protected, 3: nationally protected, 4: more than one protected area overlapping)Protected areas could persuade people to illegally use poison because of closer surveillance and/or higher fines. Legal protection often creates a feeling of disenfranchisement among local people, which can make them hostile to conservation efforts.
 BearPresence of brown bear Ursus arctosBear and wolf are the main predators in the study area, feeding on both game and livestock.
 WolfPresence of Iberian wolf Canis lupus signatus
Human pulation*
 Inhabitants§Density of inhabitantsHuman presence can increase the illegal use of poison as conflicts with ‘undesired’ species are expected to increase.
 Mean age§Mean age of the populationOlder people could consider illegal poison use as a common way to control wildlife, as it was more frequently used in the recent past (e.g. poisoning of wildlife was authorized in Spain up to 1983).
Human activities¶
 Economic activity§Percentage of population working on industrial activitiesChanges from agricultural to industrial or technological activities could reduce wildlife persecution by humans.
 Hunting‡Big (coded as 1) or small (coded as 2) game huntingSmall game hunting is more frequently related to illegal poison use.
 LU‡Density of livestock units (LU km−2). 1 cow = 5 LUs, 1 sheep or goat = 1 LU.Livestock depredation is the most common cause of human–wildlife conflict worldwide and, accordingly, it is frequently related to illegal use of poison.
 CowsDensity of cows (animals km−2)
 Sheep and goatsDensity of sheep and goats (animals km−2)

Model Development and Validation

Environmental suitability to illegal wildlife poisoning was modelled using maximum entropy modelling (MaxEnt), a machine-learning process that uses presence-only data (Phillips, Anderson & Schapire 2006). MaxEnt gives insight about what predictors are important and estimates the relative suitability of one place vs. another as well as the probability of presence (Elith et al. 2011).

We used version 3.3.0 of the software available for free download ( We accepted recommended default values for convergence threshold (1025), maximum iterations (500) and background points (10 000), as our data set was similar to those used to calculate the tuned settings (i.e. number of environmental variables and number of presence sites) (Phillips & Dudik 2008). Similarly, we also used the regularization value (to reduce overfitting) and the combination of feature classes automatically selected by the program.

The location of the study area within four different administrative regions could result in differences in detection of poisoning events because of, for example, differences in the resources that they invest in fighting against illegal poisoning (WWF/Adena 2011). If this bias is not accounted for, the resulting distribution could be closer to a model of the survey effort than to a model of the true distribution of the modelled target (Elith et al. 2011). To assess the existence and potential effect of a possible bias because of different sampling intensities between regions, we evaluated two alternative backgrounds: a sample of 10 000 points randomly distributed through the study area (RANDOM model) and a sample of 10 000 points distributed according to the number of poisoning events located within each region of the study area (e.g. the 20% of the background points were located within the region with the 20% of the poisoning events; BALANCED model). Moreover, stricter surveillance within the protected areas of the study area (Fig. 1) could give rise to a higher detection of poisoning events within these places and, therefore, to an overrepresentation of the influence of protected areas in the final models (Phillips et al. 2009). To account for this potential bias, we used chi-square tests to test the null hypothesis that there were not significantly more poisoning events within protected areas than outside them in each one of the administrative regions of the study area. For those regions where significantly more poisoning events occurred within protected areas, we accounted for this potential bias by including ranger density km−2 as a bias grid in MaxEnt (Phillips et al. 2011).

To reduce multicollinearity between variables, we calculated the Spearman’s correlation coefficients (rs). From each pair of variables that were highly correlated (rs > 0·5), we selected the variables with the highest contribution to the final model (Table 1; see Table S1, Supporting information). As all the categorical variables (i.e. protected areas and hunting) were discrete ordinal (i.e. quantify the degree of some property; Table 1), we considered them as continuous (Phillips & Dudik 2008).

We fit the models on the full data set (i.e. all presences as training points), but also used 10-fold cross-validation to estimate errors around fitted functions and predictive performance on held-out data (i.e. model robustness; Elith et al. 2011). Cross-validation was performed by randomly dividing the presence data set in 10 equal-sized groups; the data of each group were used only once as test points to validate the model, using the remaining nine groups as training points (i.e. presences used to train the model; Fouquet et al. 2010). As a result, we finally obtained 10 cross-validated models. The area under the receiver operating characteristic curve (i.e. AUC) was used as a threshold-independent method to validate the resulting models (i.e. the one performed with the full dataset and the 10 cross-validated models). As the importance of using more than one metric to assess model performance has been previously highlighted (Elith & Graham 2009), the correct classification rates of both training and test presences were also used as a threshold-dependent method. We used the 10th percentile training presence as a suitability threshold (i.e. we considered that a pixel is under risk of illegal poison use if its score is greater than the 10th percentile of training presence points; Fouquet et al. 2010). This threshold minimized the omission error and had the lowest P-value for the null hypothesis that test points are predicted no better than by a random prediction with the same fractional predicted area. Finally, we calculated Pearson’s pair-wise correlations of the predictions of the models (i.e. habitat suitability values) obtained from the 10 cross-validated models and also using the full data set, as a measure of similarity among predictions obtained using different data sets (Fouquet et al. 2010).


In the study period (i.e. 2000–2010), a total of 112 poisoning events involving 228 individuals of 25 different species were reported (Fig. 2). Three of these species (46 individuals, 20·2%) were included within the IUCN threat categories (IUCN 2010), and 11 species (129 individuals, 56·6%) were classified within the threat categories considered at a national level (BOE 2011; Fig. 2). Most of the reported deaths (90·8%) corresponded to birds of prey (120 individuals, 52·6%) and mammalian carnivores (87 individuals, 38·2%), with 95·2% of the dead animals having scavenging habits (χ2 = 186·1, d.f. = 1, P < 0·0001).

Figure 2.

 Number of poisoned animals by species recorded within the study area in 2000–2010. Only those species with greater than three individuals are shown, the rest are grouped within categories (i.e. livestock, other mammals and other birds). Letters over the bars indicate the species’ threat category according to the IUCN Red List (before the slash: LC, least concern; NT, near threatened; EN, endangered) (IUCN 2010) and to the Spanish legislation (after the slash: SP, under special protection; VU, vulnerable; ER, under extinction risk) (BOE 2011). No letter means no inclusion on either list.

We considered 45 reliable, deliberate poisoning events, involving 94 individual animals, which had the precision necessary to be used for modelling (Fig. 3). After reducing multicollinearity, the final models included 10 variables (Table 1; Table S1, Supporting information). The other six were excluded as they were correlated with variables with a higher contribution to the final models (see Appendix S1, Supporting information).

Figure 3.

 Risk maps of illegal poisoning according to our MaxEnt models: RANDOM and BALANCED (see text for further explanation). Black dots correspond to poisoning events used in the modelling process. Black lines show administrative boundaries.

The results of the RANDOM model (RANDOM hereafter) were very similar to those of the BALANCED model (BALANCED hereafter). The likelihood of illegal use of poison was higher in protected areas (42·0% and 23·5% of contribution for RANDOM and BALANCED, respectively; Fig. 4), in areas with medium densities of extensive grazing cows (i.e. c. 22 cows km−2; 17·4% and 7·1% respectively) and in areas where wolves were present (13·0% and 39·7%). Low shrub coverage was also positively related with illegal poisoning (9·4% and 8·5%). Socioeconomic factors had less influence (Fig. 4). Considering the 10th percentile presence value as a threshold (i.e. 0·28 and 0·37, respectively for RANDOM and BALANCED) (see Materials and methods), the area predicted by the models as under risk of illegal poison use (i.e. habitat suitability above the considered threshold) was 57·2% and 56·3% (respectively, for RANDOM and BALANCED) of the total study area (Fig. 3). Both maps, derived from the predictions of their respective models, showed a higher risk of illegal poisoning through the mountainous range at the centre of the study area, where most of the protected areas occur (Fig. 1). However, only in one of the four administrative regions of our study area (i.e. Asturias) were there significantly more poisoning events within protected areas than outside them (i.e. 20 poisoning events within and nine outside; χ2 = 5·83, d.f. = 1, P = 0·016). Once this potential bias was accounted for (i.e. by including ranger density km−2 as a bias grid in MaxEnt) (Phillips et al. 2011), the resulting model still highlighted the existence of protected areas as the most important factor (26·7% of contribution) related to illegal poisoning, suggesting that protected areas supported more poisoning events than expected regardless of search effort.

Figure 4.

 Importance of the environmental variables included in the two models performed. Black bars correspond to the percentage contribution of each variable to the model (left axis). Grey bars represent the jackknife results for models without that variable, and white bars correspond to the jackknife results for models with only that variable (right axis). Mean values were calculated from the 10-fold cross-validation resulting models. Error bars correspond to standard deviation.

The detailed evaluation of the resulting models highlighted good prediction ability as demonstrated by their AUCs (0·78 and 0·77, for RANDOM and BALANCED, respectively) (Phillips & Dudik 2008). Moreover, the resulting maps derived from the predictions of both models were very similar (Fig. 3). The robustness of the MaxEnt models was good as suggested by 10-fold cross-validation (Table 2). In all 10 of the cross-validations for RANDOM, both the test and training AUC indicated a prediction better than random (i.e. >0·5; Table 2). BALANCED had slightly lower values than those of RANDOM for both the 10-fold cross-validated test and training AUCs. Additionally, all cross-validated models but one (for BALANCED) satisfactorily predicted both the training and the test cases (i.e. correct classification rate >0·5; Table 2), therefore showing a very similar predicted risk of illegal poisoning across the study area. Similarly, the predicted risk of illegal poisoning of the cross-validated models was very similar to that of the model considering all the presence points (i.e. full dataset; Table 2), indicating again the robustness of the models (Fouquet et al. 2010).

Table 2.  Evaluation statistics obtained by 10-fold cross-validation (CV) for the two models considered
ModelAUC mean (range)Correct classification rate
mean (range)
Pearson’s pair-wise correlation of the predictions
mean (range)
TrainingTestTrainingTest10-fold CVFull dataset


Our results identified both the main factors underlying illegal wildlife poisoning, as well as those areas under higher risk of illegal poisoning (i.e. risk map; Fig. 3) in an area of high conservation concern. The illegal use of poison was highly related to the presence of cattle, wolves and protected areas. The relationship of poisoning and cattle could be attributed to the increasing occupation of pastureland within the study area by cattle (Olea & Mateo-Tomás 2009) and to the nature of their management (BOCYL 2011). Unlike sheep, cows mainly graze in the field without the protection of shepherds and dogs, leading to an increase in the probability of attack by predators (Woodroffe, Thirgood & Rabinowitz 2005; BOCYL 2011). Livestock predation is the most common cause of human–wildlife conflict worldwide, inducing the appearance of retaliatory actions such as the illegal use of poison within the farming sector (Newton 1979; Woodroffe, Thirgood & Rabinowitz 2005). Although livestock loss to predators appears to be lower than loss from other factors such as disease (Woodroffe, Thirgood & Rabinowitz 2005), farmers generally see predators as very harmful to their business (Blanco 2003; Woodroffe, Thirgood & Rabinowitz 2005; St John et al. 2011). In fact, according to our models, the presence of a top predator (i.e. wolf) was clearly related to the illegal use of poison. As a result, the northern limit observed in the distribution of the risk of illegal poisoning shown by our maps (Fig. 3) coincided closely with the distribution of wolves in the study area (Llaneza et al. 2005). Additionally, the annual number of wolf attacks on livestock officially reported in the northern part of the study area (i.e. 60% of the study area’s surface) between 1991 and 1999 ranged from 959 to 1179 (Blanco 2003). Therefore, this relationship between wolves, cattle and the illegal use of poison is probably shaped by the fact that wolves frequently prey on livestock in our study area (Blanco 2003).

The importance of low shrub coverage as a factor related to the illegal use of poison within the study area could be because of livestock avoiding areas with high vegetation coverage. As the amount of shrub within the study area was negatively correlated with the amount of pasture, a lower risk of conflict with wolves would be expected in the scrublands, decreasing the risk of illegal poisoning.

Protected areas were identified as another important factor enhancing the illegal use of poison. Although the existence of a protected area may be expected to act as a deterrent to poisoners (e.g. stricter surveillance, more complete campaigns to raise public awareness), legal protection can often make local people hostile to conservation efforts (Woodroffe, Thirgood & Rabinowitz 2005). In fact, a certain feeling of disenfranchisement has been reported among the inhabitants of some protected areas within the study area where retaliatory actions such as hunting are not allowed (Rescia, Fungairiño & Dover 2010), which could exacerbate human–wildlife conflicts leading to the illegal use of poison (Woodroffe, Thirgood & Rabinowitz 2005; BOCYL 2011). Moreover, most of the protected areas within the study area correspond to mountainous habitat (Fig. 1), where attacks on livestock by wolves are expected to be more frequent because of less surveillance of livestock by shepherds and dogs (Blanco 2003). Alternatively, although the presence of stricter surveillance within protected areas did not seem to over-represent the influence of protected areas in the final models (see Results), other factors more difficult to deal with, such as a higher number of tourists or better attitudes with regard to nature conservation, could contribute to the detection of more poisoning events within protected areas. Efforts to control all the possible biases on detection of illegal poisoning events are needed (Elith et al. 2011).

Other factors such as cultural, historical and socioeconomic characteristics have been linked to the human persecution of wildlife (Newton 1979; Martínez-Abraín et al. 2008; BirdLife International 2010; Lozano, Virgós & Mangas 2010; Majic et al. 2011). For example, many vultures were poisoned in South Africa during 2010 because of the mistaken belief that smoking dried vulture brains would confer supernatural powers upon gamblers enabling them to predict match results from the football World Cup (BirdLife International 2010). Examples of the influence of tradition on the illegal use of poison can also be found in more developed countries such as Spain, where poison is used against carnivores in the belief that all these species feed preferentially upon valuable small game species (Lozano, Virgós & Mangas 2010). Additionally, economic changes (i.e. moving from agriculture to industrial and technological activities) could contribute to the abandonment of the illegal use of poison, as it has already contributed to the decrease in other forms of wildlife persecution by humans (i.e. illegal shooting; Martínez-Abraín et al. 2008). In fact, although our results did not highlight any socioeconomic variable as a main factor related to illegal poison use in the study area, a slightly negative relationship between industrialization and the illegal use of poison was shown by our models (Fig. 4). No significant influence of age on the illegal use of poison was detected in our models, despite the fact that younger generations seem to have more positive attitudes towards environmental conservation (Majic et al. 2011).

Management Implications

Illegal human behaviours threaten biodiversity worldwide, but research on these illegal activities is difficult owing to their sensitive nature (Gavin, Solomon & Blank 2010; St John et al. 2011). New practical tools that help to better understand and prevent illegal activities are therefore needed to increase the effectiveness of conservation actions (Keane et al. 2008; Gavin, Solomon & Blank 2010). In this context, our approach allows one to identify the main factors related to illegal wildlife poisoning, providing also spatially explicit information of its distribution. This kind of information can help guide the effective allocation of the frequently scarce resources available in conservation (Keane et al. 2008). For example, actions such as public awareness campaigns to change behaviour or increasing surveillance could be oriented to target sectors and locations, thus increasing their effectiveness (St John et al. 2011). In our study system, conservation interventions should aim to resolve the potential conflict existing between cattle-farmers and wolves within protected areas (Majic et al. 2011). Moreover, the involvement of farmers and other stakeholders in wolf management could considerably improve the results of the actions against illegal poisoning (Majic et al. 2011). Nonetheless, further work is necessary, for example, to explore the usefulness of incorporating indicators of illegal behaviour (e.g. attitude, estimates of peer behaviour) (St John et al. 2011) into our spatially explicit models. In fact, the effectiveness of conservation actions can be improved by including data on rule breakers (Keane et al. 2008). As obtaining such detailed information is both time consuming and expensive, careful planning is needed (Keane et al. 2008). Our approach can guide the collection of data on illegal human behaviours by identifying those areas of special concern where reaching a high degree of compliance is required (e.g. by considering individual behaviour) (Keane et al. 2008; Bunnefeld, Hoshino & Milner-Gulland 2011). Our approach can be applied to other areas of the world and to other illegal activities (e.g. poaching, illegal fishing; Woodroffe, Thirgood & Rabinowitz 2005) if precise spatial data are available and local conditions are taken into account.

Illegal poisoning can jeopardize populations of threatened species and even cause their extinction (see Introduction). For example, bearded vultures Gypaetus barbatus have gone extinct in many places all over the world mainly as a consequence of illegal poisoning (Schaub et al. 2009). This includes our study area, where a reintroduction programme has been recently launched (FCQ 2011). Our data show that illegal poisoning continues to kill endangered species, especially scavengers. Information on the spatial risk of illegal poisoning and their causes, as shown here, is essential to avoid negating conservation and reintroduction efforts for many endangered species (Schaub et al. 2009). This information could also be useful to refine existing SDMs for different species for which poisoning is considered to be an important threat constraining their ranges (e.g. Egyptian vulture N. percnopterus, red kite M. milvus; Hernández & Margalida 2009; Mateo-Tomás, Olea & Fombellida 2010; Smart et al. 2010). In addition, the risk maps might be useful when classifying possible wildlife poisoning events in which toxicological analysis cannot provide conclusive results (e.g. insufficient sample, decomposed carcasses; Guitart et al. 2010), helping to quantify this conservation threat (Berny 2007).

To our knowledge, this is the first time that an application of MaxEnt has been used to address an illegal activity, widening its usefulness in biodiversity conservation (Parisien & Moritz 2009; Elith et al. 2011). In fact, the friendly interface of this software and the relatively simple modelling procedure developed here help to reduce one of the major concerns associated with the use of models to deal with illegal activities, that is, their limited use by some conservation managers because of high levels of analytical sophistication (Gavin, Solomon & Blank 2010). The spatial visualization of the model results also facilitates their use by conservation managers.


We thank C. Cano, L. Varona, R. Menéndez, Fundación para la Conservación del Quebrantahuesos, Junta de Castilla y León and SEPRONA for information on poisoning events. J. Viñuela, two anonymous reviewers and editors provided useful comments on the manuscript. Toxicologic analyses performed at IREC were funded by Principado de Asturias. P.M.T. was supported by a postdoctoral grant funded by Junta de Comunidades de Castilla-La Mancha and Fondo Social Europeo.