Professional Documents
Culture Documents
Biological Conservation
journal homepage: www.elsevier.com/locate/biocon
Sierra Nanchititla Biological Station, Faculty of Sciences, Autonomous University of the State of Mexico, Mexico
Instituto literario 100, Centro, Toluca, Estado de Mxico, CP 5000, Mexico
c
Faculty of Sciences, Autonomous University of Baja California, Carretera Tijuana-Ensenada Km 106, 22800 Ensenda, BC, Mexico
b
a r t i c l e
i n f o
Article history:
Received 15 May 2012
Received in revised form 23 October 2012
Accepted 4 November 2012
Available online 21 January 2013
Keywords:
Livestock
Predation
Ecological niche model
Puma
Jaguar
Mxico
a b s t r a c t
Predation on livestock is one of the main factors that cause the felids hunting, in particular for puma and
jaguar this conict with humans is severe. Most studies have assessed the predation impacts on livestock
production; however there is a spatial pattern in attacks occurrence that is feasible to analyze from ecological niche modeling. The objective of this research was to generate a risk model of livestock predation
by puma and jaguar in Mexico based on environmental and livestock management variables, which
allows identication of zones of risk in order to dene mitigation strategies at national level. We produced a geographic ensemble model of risk of predation from three algorithms for jaguar and ve for
puma. The variables most positively related with predation risk by jaguar were vegetative cover percentage, percentage of free grazing animals, and altitude, whereas arid vegetation has a negative inuence on
predation risk. In the case of puma the variables with highest contribution were livestock density, which
negatively inuences on the predation risk, in addition to forest and altitude, both with a positive relation. The ensemble models are an accurate approach to delineating the zones of predation risk by felids;
however at a regional scale the environmental characteristics that favor predation may be different. It is
recommended that researchers carry out studies for each biogeographic province that facilitate the identication of specic patterns and the denition of mitigation strategies most suitable for each one.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
Populations of large carnivores are declining worldwide due to
the loss or fragmentation of their habitat, and by direct hunting of
the carnivores or their prey; currently, 17% of the species are threatened (IUCN, 2011). The hunting of felids takes place in response to
their supposed predation on livestock (Holmern et al., 2007; Kissui,
2008; Gusset et al., 2009); this problem affects 75% of felid species
worldwide, and in particular is severe for puma and jaguar (Inskip
and Zimmermann, 2009). In previous studies, data on felids hunted
in retaliation for predation have been reported, which range from
seven pumas killed per year on average in a ranch in Sao Paulo
to 150 pumas and jaguars in Alta Floresta, Brazil (Inskip and Zimmermann, 2009).
Hence, it is evident that a fundamental topic in large cats conservation is the analysis of the conicts with livestock holders. This
issue has mainly been approached by describing the impact of predation on livestock production and identifying factors related with
Corresponding author at: Instituto literario 100, Centro, Toluca, Estado de
Mxico, CP 5000, Mexico. Tel.: +1 722 296 55 53.
E-mail addresses: martha.zarco.g@gmail.com (M.M. Zarco-Gonzlez), tavomonroyvilchis@gmail.com, omv@uaemex.mx (O. Monroy-Vilchis).
0006-3207/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.biocon.2012.11.007
the predation levels (Patterson et al., 2004; Kolowski and Holekamp, 2006; Azevedo and Murray, 2007; Van et al., 2007; Palmeira
et al., 2008; Iftikhar et al., 2009). Other authors have observed differences in the environmental characteristics of sites with and
without predation (Jackson et al., 1996), and this may be an indicator of conditions inherent to the sites that inuence on the risk
(Stahl et al., 2002). Among the proposed environmental variables
that may be related with the frequency of attacks are proximity
to forest zones (Mazzolli et al., 2002; Stahl et al., 2002; Azevedo
and Murray, 2007; Palmeira et al., 2008), the vegetation type (Rosas-Rosas et al., 2010), altitude (Lui et al., 2006), topography (Stahl
et al., 2002; Michalski et al., 2006; Kissling et al., 2009), density of
livestock and wild prey (Treves et al., 2004; Bagchi and Mishra,
2006; Kolowski and Holekamp, 2006), distance to protected areas,
human settlements, roads, and water sources (Lui et al., 2006; Van
et al., 2007; Gusset et al., 2009; Rosas-Rosas et al., 2010). An approach that may be useful in decreasing the severity of the conict
is to anticipate its spatial location and to propose preventive actions in specic areas, optimizing economic and human resources.
A commonly used tool to plan strategies of wildlife management, which relates ecological variables and spatial processes,
are ecological niche models (Gibson et al., 2007; Zarco-Gonzlez
et al., 2012), the modeling provides estimates of occurrence
81
of the country is above 1000 m above sea level and nearly 47% of
the surface has slopes over 27% (UNAM, 1990; Mittermeier and
Mittermeier, 1992). In conservation importance, Mexico hosts between 8% and 12% of the total of the species on the planet (Mittermeier and Mittermeier, 1992), including six feline species, of which
puma (Puma concolor) and jaguar (Panthera onca) are the largest.
The records of attacks on livestock by puma and jaguar in Mexico were collected in three ways: (1) databases of the Mexican government, Secretariat of Environment and Natural Resources
(SEMARNAT), Pronatura (civil association) and technical reports
(Nez, 2007a); (2) review of scientic literature: theses (BuenoCabrera, 2004; Villordo-Galvn, 2009), books (Brown and Lpez,
2001; Rosas-Rosas and Lpez-Soto, 2002; Caso, 2007; Cruz et al.,
2007; Leyequin and Balvanera, 2007; Lira and Ramos-Fernndez,
2007; Navarro et al., 2007; Nez, 2007b), and papers (Rosas-Rosas et al., 2008; Chvez and Zarza, 2009; Zarco-Gonzlez et al.,
2012); (3) and eld work in different states of the country (states
of Mexico, Chihuahua, Yucatn, Campeche, Baja California and
Guerrero). Approximately 78% of predation data were recorded at
the site of attack with GPS devices and 22% were inferred using
Google Earth and maps and descriptions presented in the original
sources. A database of predation events was produced with the records specifying date, predator species, place of attack, and geographic coordinates (Fig. 1).
Considering the reports of other studies in relation to the inuence of the environmental factors on the predation risk (Stahl et al.,
2002; Treves et al., 2004; Bagchi and Mishra, 2006; Kolowski and
Holekamp, 2006; Lui et al., 2006; Michalski et al., 2006; Van
et al., 2007; Gusset et al., 2009; Kissling et al., 2009; Rosas-Rosas
et al., 2010), the variables used to characterize the sites of attack
were grouped in three types: landscape, livestock management,
82
3. Theory
Machine learning methods are algorithms that are used to learn
the mapping function or classication rules inductively, directly
from the training data (Franklin, 2009).
They include articial neural networks, genetic algorithms,
maximum entropy and support vector machines. These methods
tend to perform well given complex classication problems, however, interpreting the modeled relationships between predictors
and response is not always straightforward in some machine learning implementations.
ANNs consist of many processing elements (articial neurons)
that are interconnected to form a network. ANNs are trained
by repeatedly passing large numbers of known examples of the
problem under consideration through the network. By repeatedly
adjusting the connections between processing elements the difference between the network predictions and the known examples
can be minimized (Pearson et al., 2004). Neural networks can identify non-linear responses to environmental variables, as well as
incorporating multiple types of variables, including qualitative
and quantitative. A disadvantage of this method is that it does
not identify the relative contribution of the different variables
(Gevrey et al., 2003). The details of the classier developed using
neural networks are not obvious or interpretable, it is considered
a black box approach to classication, and so the main advantage
of this method is that it sometimes achieves much higher classication accuracy than other methods in high-dimensional, complex
classication problems (Franklin, 2009).
The GARP software implements an ensemble method that generates a population of prediction rules based on several different
types of models, and then uses a genetic algorithm to select among
them and develop a nal rule set to make predictions. Further, because the output from GARP is stochastic, it is typical to run the
model multiple times, and then average a subset of the best models
(Stockwell and Peters, 1999). GARP is not able to model responses
to categorical predictors and tends to over predict the extent of
species distributions, that is, to have higher commission errors
(Franklin, 2009).
In Maxent 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. The advantages of this method are that it can use both categorical and
continuous data, as well as incorporate interactions between different variables (Phillips et al., 2006), additionally Maxent can
work well with very small samples that have relatively widespread
geographic distributions (Hernandez et al., 2008).
SVM use a functional relationship known as a kernel to map
data onto a new hyperspace in which complicated patterns can
83
Table 2
AUC values of the models for jaguar.
Algorithm
Maxenta
GARP (with best subsets)a
SVMa
ENFA
GARP (single run)
ANN
Maxent (open modeller)
Environmental distance
ENFA (open modeller)
a
AUC
External evaluation
Internal evaluation
0.799
0.784
0.767
0.728
0.728
0.725
0.704
0.626
0.500
0.919
0.792
0.837
0.743
0.714
0.730
0.669
0.923
0.520
Table 3
AUC values of the models for puma.
Algorithm
Maxenta
Environmental distancea
ENFAa
GARP (with best subsets)a
GARP (single run)a
ANN
SVM
Maxent (open modeller)
ENFA (open modeller)
a
AUC
External evaluation
Internal evaluation
0.944
0.875
0.866
0.817
0.767
0.720
0.617
0.503
0.500
0.934
0.961
0.828
0.802
0.790
0.730
0.591
0.505
0.510
Table 4
Importance of variables in the models generated with Maxent for each felid species
(only the most important are shown).
4. Results
We obtained 241 records of felid attacks on livestock in Mexico,
and after data reduction we retained 222 records; 152 predation
events by jaguar and 70 by puma. The retained records are from
1990 to 2010, and are distributed in 19 states (Table 1).
For jaguar, three algorithms were considered suitable according
to the internal evaluation: Maxent, GARP-with best subsets and
SVM (Table 2). For puma ve algorithms were considered: Maxent,
Environmental distance, ENFA, GARP-with best subsets and GARPsingle run (Table 3). For both species the algorithm with the best
performance was Maxent.
According to ENFA, for data from jaguars, we obtained a global
marginality value of 0.793, which indicates that the observed conditions in the sites of attack are different from the average environmental conditions and the range in which predation occurs is very
restricted in relation to the global characteristics. Maxent indicates
that the variables with highest contribution to the model are tree
cover percentage, percentage of animals in free grazing areas,
and altitude. These variables are positively related with the risk
Table 1
Number of data obtained of each source and year of collect.
Source
Number of
data
Year of collect
80
20092010
10
20002007
13
36
26
76
20002006
19912006
19982004
20062010
Variable
Altitude
Forest
Arid vegetation
Cover percentage
Livestock density
Percentage of free gazing livestock
Puma
10.3
3.7
8.5
47
4.5
12.1
12.3
34.1
0.5
6.7
35.2
6.6
of attack by jaguar, whereas arid vegetation has a negative correlation with predation risk (Table 4).
In the case of puma, ENFA indicated a marginality global value
of 1.157, these results imply that the characteristics of predation
sites by puma are very specic, even higher than those of jaguar,
and the general variation in these characteristics is low. Maxent
showed that the variables with highest contribution were livestock
density, which negatively inuences on the predation risk by
puma, in addition to forest and altitude, both with a positive relation (Table 4).
The ensemble model for jaguar had an AUC value of 0.8. The
zones of highest predation risk by this species (probability > 64)
are in the western Sierra Madre Occidental, at the border with
Sonorense province, the Costa del Pacco, western and center of
Cuenca del Balsas, eastern Sierra Madre del Sur, eastern Soconusco,
the northern and southern limits of Altos de Chiapas and Golfo de
Mxico, mainly northern and limits with Sierra Madre del Sur,
Yucatn, Petn and southern Sierra Madre Oriental (Fig. 2).
The ensemble model for puma presented an AUC value of 0.9. In
this case the risk zones (probability > 69) are concentrated on
southern California, western and central Sierra Madre Occidental,
84
Fig. 2. Ensemble model of risk predation by jaguar in Mexico. (1) Sierra Madre Occidental, (2) Sonorense, (3) Costa del Pacco, (4) Cuenca del Balsas, (5) Sierra Madre del Sur,
(6) Soconusco, (7) Altos de Chiapas, (8) Golfo de Mxico, (9) Yucatn, (10) Petn y, (11) Sierra Madre Oriental.
Fig. 3. Ensemble model of risk predation by puma in Mexico. (1) California, (2) Sierra Madre Occidental, (3) Sierra Madre Oriental, (4) Eje Volcnico Transmexicano, (5)
Cuenca del Balsas, (6) Sierra Madre del Sur, (7) Soconusco y, (8) Altos de Chiapas.
northern Sierra Madre Oriental, the southern limit of the Eje Volcnico Transmexicano, the center of Cuenca del Balsas, Sierra Madre
del Sur, Soconusco and the center of Altos de Chiapas (Fig. 3).
5. Discussion
The conict of livestock predation by the two large cats of Mexico has negatively impacted on the populations of both jaguar and
puma. In the north of the country, the illegal hunting of jaguar is
perhaps the main threat for its conservation (Rosas-Rosas et al.,
2008). However, in the case of puma the severity of the conict
has not been assessed, but it is known that in some areas predation
on livestock is the most frequent reason for their hunting by hu-
mans (Zarco-Gonzlez et al., 2012). Despite this situation, systematic data on the topic are scarce, and most of the published studies
are focused on jaguar (Rosas-Rosas et al., 2008, 2010; Chvez and
Zarza, 2009; Villordo-Galvn, 2009). Fewer studies have been carried out on puma (Bueno-Cabrera, 2004; Zarco-Gonzlez et al.,
2012). As a consequence, there are no data where livestock predation occurs, and much less on the impact on livestock production
or on the felids conservation, so the present risk model shows a
spatial panorama of the predation problem in Mexico.
One of the most recent actions to address this issue was carried
out in 2009 by the National Livestock Confederation and the Secretariat of Agriculture, Livestock, Rural Development, Fishing and Alimentation (SAGARPA). They established a program providing
livestock insurance for death due to predators. The general objec-
85
86
87
Phillips, S.J., Anderson, R.P., Schapire, R.E., 2006. Maximum entropy modeling of
species geographic distributions. Ecol. Model. 190, 231259.
Polisar, J., Maxit, I., Scognamillo, D., Farrell, L., Sunquist, M., Eisenberg, J., 2003.
Jaguars, pumas, their prey base, and cattle ranching: ecological interpretations
of a management problem. Biol. Conserv. 109, 297310.
Rodrguez-Soto, C., Monroy-Vilchis, O., Maiorano, L., Boitani, L., Faller, J.C., Briones,
M.A., Nez, R., Rosas-Rosas, O., Ceballos, G., Falcucci, A., 2011. Predicting
potential distribution of the Panthera onca in Mexico: identication of priority
areas for conservation. Divers. Distrib. 17, 350361.
Rosas-Rosas, O., Lpez-Soto, J.H., 2002. Distribucin y estado de conservacin del
jaguar en el estado de Nuevo Len, Mxico. In: Medelln, R. A., Equihua, C.
Chetkiewicz, C.L.B., Crawshaw, P.G., Rabinowitz, A., Redford, K.H., Robinson, J.G.,
Sanderson, E.W., Taber, A.B. (Eds.), El jaguar en el Nuevo milenio, Fondo de
Cultura Econmica, Instituto de Ecologa, Universidad Autnoma de Mxico,
Wildlife Conservation Society, Mxico, pp. 393402.
Rosas-Rosas, O., Bender, L., Valdez, R., 2008. Jaguar and puma predation on cattle
calves in Northeastern Sonora, Mexico. Rangel. Ecol. Manag. 61, 554560.
Rosas-Rosas, O., Bender, L.C., Valdez, R., 2010. Habitat correlates of jaguar kill-sites
of cattle in northeastern Sonora, Mexico. HumanWildlife Interactions 4, 103
111.
Roura-Pascual, N., Brotons, L., Peterson, A.T., Thuiller, W., 2008. Consensual
predictions of potential distributional areas for invasive species: a case study
of Argentine ants in the Iberian Peninsula. Biol. Invasions 11, 10171031.
SAGARPA (Secretara de Agricultura, Ganadera, Desarrollo Rural, Pesca y
Alimentacin), 2006. Situacin actual y perspectiva de la produccin de carne
de bovino en Mxico 2006. Coordinacin General de Ganadera. <http://
www.sagarpa.gob.mx/Dgg> (accessed 20.05.11).
Salvatore, M., Pozzi, F., Ataman, E., Haddleston, B., Bloise, M., 2005. Mapping global
urban and rural population distributions. Food and Agricultural Organizations
of the United Nations. Environmental and Natural Resources Working Paper 24.
FAO, Rome.
Scotts, D., Drielsma, M., 2003. Developing landscape frameworks for regional
conservation planning: an approach integrating fauna spatial distributions and
ecological principles. Pac. Conserv. Biol. 8, 235254.
SEMARNAP (Secretara del Medio Ambiente, Recursos Naturales y Pesca), INEGI
(Instituto Nacional de Estadstica, Geografa e Informtica), UNAM (Universidad
Nacional Autnoma de Mxico), 2001. Inventario Nacional Forestal. Escala
1:250000. Instituto de Geografa, UNAM. Mxico.
Skov, F., Svenning, J.C., 2004. Potential impact of climatic change on the distribution
of forest herbs in Europe. Ecography 27, 366380.
Stahl, P., Vandel, J., Ruette, S., Coat, L., Coat, Y., Balestra, L., 2002. Factors affecting
lynx predation on sheep in the French Jura. J. Appl. Ecol. 39, 204216.
Stockwell, D., Peters, D., 1999. The GARP modelling system: problems and solutions
to automated spatial prediction. Int. J. Geogr. Inf. Sci. 13, 143158.
Thomas, C.D., Cameron, A., Green, R.E., Bakkenes, M., Beaumont, L.J., Collingham,
Y.C., Erasmus, B.F.N., Ferreira, M., Grainger, A., Hannah, L., Hughes, L., Huntley,
B., van Jaarsveld, A.S., Midgley, G.F., Miles, L., Ortega-Huerta, M.A., Townsed, A.,
Phillips, O.L., Williams, S.E., 2004. Extinction risk from climate change. Nature
427, 145148.
Thuiller, W., Lavorel, S., Arajo, M.B., Sykes, M.T., Prentice, I.C., 2005. Climate change
threats to plant diversity in Europe. Proc. Natl. Acad. Sci. USA 102, 82458250.
Treves, A., Naughton-Treves, L., Harper, E., Mladenoff, D., Rose, R., Sickley, T.,
Wydeven, A., 2004. Predicting human-carnivore conict: a spatial model
derived from 25 years of data on wolf predation on livestock. Conserv. Biol.
18, 114125.
Tsoar, A., Allouche, O., Steinitz, O., Rotem, D., Kadmon, R., 2007. A comparative
evaluation of presence only methods for modeling species distribution. Divers.
Distrib. 13, 397405.
UNAM, 1990. Atlas nacional de Mxico. Instituto de Geografa, Mxico.
USGS (United States Geological Survey), 2007. Shuttle radar topography mission
(SRTM) 3-arc second ARTM Format Documentation. USGS/NASA. <http://
www.edc.usgs.gov/products/elevation/srtmbil.html>.
Van, L., Bij de Vaate, M., De Boer, W., De Iongh, H., 2007. Factors affecting livestock
predation by lions in Cameroon. Afr. J. Ecol. 45, 490498.
Vilaboa, J., Daz, P., 2009. Caracterizacin socioeconmica y tecnolgica de los
sistemas ganaderos en siete municipios del estado de Veracruz, Mxico. Zoo.
Tropical 27, 427436.
Villordo-Galvn, J.A., 2009. Distribucin y estado de conservacin del jaguar
(Panthera onca) en San Luis Potos, Mxico. Tesis de Maestra. Colegio de
Postgraduados, Montecillo, Texcoco, Estado de Mxico.
Wang, S., Macdonald, D., 2006. Livestock predation by carnivores in Jigme Singye
Wangchuck National Park. Bhutan. Biol. Conserv. 129, 558565.
Woodroffe, R., Lindsey, P., Romaach, S., Stein, A., ole Ranah, Symon M.K., 2005.
Livestock predation by endangered African wild dogs (Lycaon pictus) in northern
Kenia. Biol. Conserv. 124, 225234.
Zarco-Gonzlez, M., Monroy-Vilchis, O., Rodrguez-Soto, C., Urios, V., 2012. Spatial
factors and management associated with livestock depredation by Puma
concolor in Central Mexico. Hum. Ecol. 40, 631638.