Journal of Mammalogy, xx(x):1–11, 2016

DOI:10.1093/jmammal/gyw184

Living in extreme environments: modeling habitat suitability for

jaguars, pumas, and their prey in a semiarid habitat
Samuel Astete,* Jader Marinho-Filho, Ricardo B. Machado, Bárbara Zimbres,
Anah T. A. Jácomo, Rahel Sollmann, Natália M. Tôrres, and Leandro Silveira
Programa de Pós-Graduação em Ecologia, Universidade de Brasília, Brasília 70910-900, Brazil (SA)
Departamento de Zoologia, Universidade de Brasília, Brasíllia 70910-900, Brazil (JMF, RBM)
Jaguar Conservation Fund/Instituto Onça-Pintada, Mineiros, Goiás 75830-000, Brazil (ATAJ, RS, NMT, LS)
Department of Wildlife, Fish, and Conservation Biology, University of California Davis, Davis, CA 95616, USA (RS)
Programa de Pós-Graduação em Zoologia, Universidade de Brasília, Brasília 70910-900, Brazil (BZ)
Instituto de Biologia, Universidade Federal de Uberlândia, Uberlândia, MG 38400-902, Brazil (NMT)
* Correspondent: samastete@yahoo.com

Carnivores and their prey occurring at the extremes of their ecological niches face particular constraints in
terms of habitat suitability (HS). We combined the use of camera traps and Ecological Niche Factor Analysis
(ENFA) to calculate HS for jaguars (Panthera onca), pumas (Puma concolor), and 4 prey species (3-banded
armadillos, Tolypeutes tricinctus; 9-banded armadillos, Dasypus novemcinctus; collared peccaries, Pecari tajacu;
and gray brocket deer, Mazama gouazoubira) in Serra da Capivara National Park (SCNP), located in the semiarid
Caatinga biome of Brazil. We also evaluated HS in relation to water management applied in the Park since 1994.
Considering the physiological limitations of the Caatinga’s species, we used an “edge of niche” habitat-suitability
algorithm. For all species examined, distance to waterholes was the ecogeographical variable with the greatest
contribution for the ENFA. Water management implemented in SCNP helps to ensure the persistence of predators
and prey species, since the combination of topographic characteristics and location of waterholes explained most
of HS for the analyzed species.

Os carnívoros e suas presas que ocorrem nos extremos dos seus nichos ecológicos enfrentam restrições particulares
en termos de adequabilidade de habitat (AH). Neste estudo, combinamos o uso de armadilhas-fotográficas e a
Análise Fatorial de Nicho Ecologico (ENFA- Ecological Niche Factor Analysis) para calcular a AH para a onça-
pintada (Panthera onca), a onça-parda (Puma concolor) e 4 espécies presa (3 o tatu-bola, Tolypeutes tricintus;
o tatu-galinha, Dasypus novemcinctus; o cateto, Pecari tajacu; e o veado-catingueiro, Mazama gouazoubira)
no Parque Nacional Serra da Capivara (PNSC), localizado no bioma semiárido da Caatinga do Brasil. Também
avaliamos os resultados da AH em termos de uma política de manejo de água aplicada no parque desde 1994.
Considerando as limitações fisiológicas das espécies na Caatinga, utilizamos um algoritmo de AH de “beira de
nicho”. Para todas as espécies analisadas, a distância a poços de água foi a variavel eco-geográfica com a maior
contribuição para o ENFA. A política de manejo de água aplicada no PNSC contribui a asegurar a persitência
dos predadores e presas, uma vez que a combinação de características topográficas e poços de água explicaram a
maior parte da AH para as espécies analisadas.

Key words:  Caatinga, ENFA, gray brocket deer, jaguar, puma, water management

Acquiring reliable information about species distributions with humans for space and food and the alteration of habitats
and building models that can explain and estimate species has led to escalated human–carnivore conflicts (Treves and
occurrences is a major concern for conservation efforts Karanth 2003). As carnivores depend on their prey, reliable
(Araújo and Williams 2000). Carnivores worldwide compete information on distribution and habitat suitability (HS) for

© 2016 American Society of Mammalogists, www.mammalogy.org

1
2 JOURNAL OF MAMMALOGY

both predator and prey is crucial for carnivore conservation areas of the Caatinga biome (MMA 2005). Temperature can
efforts. vary from 12°C to 50°C in the dry season, with a rainy sea-
HS models that predict the spatial distribution of species (Guisan son from October to mid-April (but it can begin later or end
and Zimmermann 2000; Pearce and Boyce 2006) based on their earlier—Emperaire 1984; Figueiredo and Puccioni 2006), and
ecological requirements present a useful tool for conservation and mean annual precipitation of 689 mm, with a maximum, over
management of endangered species (Palma et al. 1999; Rondinini the past 70 years, of 1,269 mm and a minimum of 250 mm.
et al. 2005). HS models relate a set of field records of the target Those values are, however, much lower than the 1,462.4 mm
species to a group of ecological variables that are presumed to yearly potential evapotranspiration (FUMDHAM 1998). There
reflect some key factors of the species’ niche (Hirzel et al. 2001). are 8 habitat types recognized in SCNP (from open to dense
Models based on presence-only or presence–pseudo-absence data arboreal Caatinga vegetation), the predominant habitat being
are the most standard approach to habitat modeling (Hirzel et al. tall, shrubby vegetation 6–10 m high (Emperaire 1984). The
2001; Segurado and Araujo 2004; Chefaoui and Lobo 2008). elevation varies between 280 and 600 m a.s.l. and the topogra-
The semiarid Caatinga biome (seasonal dry tropical forest) phy consists of a main plateau bounded by 50- to 200-m-high
covers an area of almost 750,000 km2 in northeastern Brazil cliffs cut by canyons and valleys (Emperaire 1984). SCNP has
(MMA 2005). Little attention has been given to its conservation no natural perennial water bodies, and a series of permanent
(Silva et al. 2004) and fully protected areas cover only 1.21% artificial waterholes has been installed since 1994 as part of the
of its territory (Capobianco 2002). The facts that there is low park management (SMAPR 1994). SCNP has been declared a
endemism and absence of physiological adaptations of mam- World Heritage Site by UNESCO due to its archeological leg-
mals to the arid conditions in the Caatinga (Mares et al. 1985) acy. To facilitate access and research, the park has an extensive
raise important questions for conservation: how do mammals network of dirt roads and trails. Our study in SCNP was autho-
that also occur in other regions cope with the conditions in this rized by the Brazilian environmental agency ICMBio (Instituto
severe biome, what landscape components do they select (e.g., Chico Mendes de Conservação da Biodiversidade).
topography, position of natural or anthropogenic ecosystems, Species data.—We created presence-only distribution maps
etc.), and how do species interact with each other? of the target species based on records from 3 camera-trap sur-
Semiarid habitats, such as the Caatinga, are particularly veys implemented in SCNP in 2007, 2009–2010, and 2010–
vulnerable to global climate change, because in such habi- 2011, covering both the dry and rainy seasons in each survey
tats changes in water availability and climate may have dis- with an equivalent effort in each season. Each camera-trap sta-
proportionate effects on biodiversity (Millennium Ecosystem tion was on average 3.1 km away from its nearest neighbor. We
Assessment 2005). In Africa, to avoid these effects inside pro- used 24 stations in 2007 and 70 stations from 2009 to 2011. The
tected areas, water management has been applied as a way to last 2 camera-trap surveys covered the entire park, with cam-
recover herbivores (Pienaar 1985; Glaylard et al. 2003). eras placed proportionally in the types of vegetation present at
Using camera-trapping data in combination with Ecological SCNP. Cameras were placed along roads throughout the park
Niche Factor Analysis (ENFA), a presence-only-based HS (Fig. 1), because the dense vegetation impeded placing cameras
model (Hirzel et al. 2002), we investigated HS for jaguars far from roads. We defined a record as the register of a species
(Panthera onca), pumas (Puma concolor), and 4 potential prey by a camera trap.
species—3-banded armadillos (Tolypeutes tricinctus), 9-banded Species maps were produced for 2 predators, jaguars and
armadillos (Dasypus novemcinctus), collared peccaries (Pecari pumas, and for 4 potential prey species for which we had
tajacu), and gray brocket deer (Mazama gouazoubira)—in enough data: gray brocket deer, collared peccaries, 3-banded
the Serra da Capivara National Park (SCNP), the 2nd largest armadillos, and 9-banded armadillos.
protected area in the Caatinga. Due to its size, the SCNP is Environmental variables.—In ENFA, ecogeographical vari-
extremely important for the conservation of biodiversity in this ables (EGVs) are variables assumed to reflect some key fac-
biome, particularly of wide-ranging mammals. We used the so- tors of the species’ niche. We created EGVs using the programs
called “edge of niche” algorithms, which are recommended for ArcGIS 9.31 (ESRI 2009) and Idrisi Taiga (Eastman 2009). We
a species occurring at the margin of its distribution (Braunisch used 6 EGVs including proximity to artificial waterholes, a dig-
et al. 2008), as is the case for most of the species in this study. ital elevation model (DEM), slope, normalized difference veg-
We expected that for most species, scarce resources, such as etation index (NDVI) for each season, distance to paved roads,
water or places to thermoregulate, would be the most important and distance to human settlements around SCNP. NDVI is
factors influencing suitability. As carnivores depend on their derived from remote-sensing data on visible and near-infrared
prey, and the prey are subject to the same climatic stressors, we bands of the electromagnetic spectrum, varies from −1.0 to 1.0,
also expected a high overlap between the HS models of jaguars and is positively related to cover by photosynthetic biomass.
and pumas and some of their prey species. All maps excluding NDVI were provided by the Fundação
Museu do Homem Americano (FUMDHAM), which adminis-
ters the park. We considered these EGVs to summarize charac-
Materials and Methods
teristics of the physical environment and vegetation, as well as
Study area.—The SCNP is located in the state of Piauí, north- human influence in SCNP. Because cameras were placed along
eastern Brazil. With 1,291.4 km2, it is one of the largest protected dirt roads, we did not include a map of these roads as an EGV.
 ASTETE ET AL.—ENFA, JAGUAR, PUMA, PREY, SEMIARID 3

Fig. 1.—Location of the Serra da Capivara National Park (SCNP) in the Caatinga biome of Brazil (dashed area), and spatial distribution of artifi-
cial waterholes and camera traps in the park.

We tested for correlation among EGVs and since there were distribution to that of the focal species’ distribution. The inverse
no significant associations (see Supplementary Data SD1), we of specialization is defined as the species tolerance and repre-
used all of them in our analysis. The values of each EGV at sents a measurement of the focal species’ selectiveness within
each sampling point can be found in Supplementary Data SD2. the available conditions in the area determined by the variables.
Since some of the EGV maps were derived from Landsat A global marginality factor close to 1 means that the species
5 TM images (such as NDVI), we produced all of them with lives in a very particular habitat relative to the reference set,
30-m spatial resolution, the same resolution as the DEM and whereas a tolerance value < 1 indicates some degree of special-
slope maps (Valeriano 2008). The Landsat images were from ization (Hirzel et al. 2002). An advantage of ENFA is that its
2010 and came from the INPE (www.inpe.br) database. We results allow a direct comparison between multiple species that
used Box-Cox or square-root transformations to normalize inhabit the same region, by interpreting the parameters from the
EGVs (Hirzel et al. 2002). As pointed out by Pettorelli et al. perspective of the ecological niche (i.e., marginality).
(2009), the coefficients computed by ENFA for each EGV To estimate HS scores from ENFA factors (Hirzel et al.
should be interpreted with care in the case of distance maps 2002), we used the software BIOMAPPER 4.0 (Hirzel et al.
(e.g., distance to waterholes, roads, and human settlements): a 2008). The program uses a set of algorithms which assume that
high positive value indicates species avoidance, whereas a high the environmental conditions are optimal where the species is
negative value indicates preference for the EGV considered. most frequently found. The default algorithm of BIOMAPPER
ENFA analysis.—ENFA is an approach that uses pres- is the median algorithm (M), which considers that the opti-
ence-only data to create an HS model of a species, based on mum condition is represented by the median of the species
Hutchinson’s niche concept (Hutchinson 1957), referred to here frequency distribution, assumed to be unimodal and asymmet-
as an area in the ecogeographical space where a species has a rical. Considering that HS model results could be biased when
reasonable chance to occur (Hirzel et al. 2002). ENFA uses a environmental conditions in the study area represent a marginal
multivariate factor approach to aggregate information for the part of the species’ fundamental niche, and that this situation
species, aiming to identify general patterns in habitat use based occurs mostly at the margins of a species geographic distribu-
on 2 indices: marginality and specialization. Marginality, rang- tion, we chose an “edge of niche” algorithm, specifically the
ing from 0 to 1, maximizes the multivariate distance between “area-adjusted median algorithm” (Ma), which considers the
the cells occupied by a species and the average global distribu- relative availability of habitat conditions, comparing habitat
tion of each variable. The aggregation of the remaining axes, use to habitat availability to evaluate how species might select
called specialization, is the ratio of the variance of the global a habitat (Braunisch et al. 2008).
4 JOURNAL OF MAMMALOGY

We ran 3 different models for each species, 1 considering HS maps as similar. We did the analyses considering HS maps
only dry season records, 1 considering only wet season records, derived from the total and seasonal records.
and 1 with the combined set of records. The maximum number As jaguars and pumas are sympatric predators (Iriarte et al.
of records for the models was the total number of camera-trap 1990), we looked for information on niche breadth and poten-
sites. We assumed that the dry season might be the limiting sea- tial overlap. Following the methodology of Sattler et al. (2007),
son in terms of resources and by splitting the data into seasonal we used BIOMAPPER Discriminant Analysis to estimate niche
records we tried to understand seasonal differences. breadth and overlap indices (De Angelo et al. 2011), such as
To evaluate the resulting HS models, we used a 10-fold traditional niche breadth index (Levin’s standardized index)
jack-knife cross-validation, the default option in BIOMAPPER and overlap indices (Pianka’s overlap index and Lloyd’s asym-
(Hirzel et al. 2006). We computed 2 presence-only evaluation metric overlap index—Colwell and Futuyama 1971; Hurlbert
measures: AVI (Absolute Validation Index) and the Continuous 1978). Lloyd’s asymmetric overlap index is a measure of direc-
Boyce Index (CBI—Hirzel et al. 2006). AVI ranges from 0 to 1 tional niche overlap: Z12 represents the density of species 2
and indicates how well the model discriminates high-suitability encountered, on average, by species 1; Z21 is the reciprocal.
from low-suitability areas. The CBI, a threshold-independent The index is also called measure of interspecific crowding, so
modification of the Boyce Index (Boyce et al. 2002), is calcu- that the ratio of interspecific crowding values for 2 species is
lated as the Spearman correlation coefficient between the ratio the reciprocal of the ratio of their abundances (Hurlbert 1978).
of the predicted over expected frequency of evaluation points
and the habitat suitability index (HSI). It varies from −1 to 1:
−1 for an inverse model to 0 for a random model to 1 for a
Results
perfect model. In BIOMAPPER, the default option for the 10 We accumulated a total effort of 19,938 camera-trap days:
k-fold model evaluation (predicted-to-expected ratio) in CBI is 3,833 during a 9-month sample in 2007 (February to October)
a sliding window size of 20 HSI units, thus discarding data sets and 16,105 in 2009–2011 (2 samples, from September to
resulting in less than 20 records. We established a CBI thresh- January each). The 2007 sampling included 3 months from the
old of 0.200, and models with a CBI < 0.200 were considered rainy season and 6 from the dry season. The samplings in 2009
having little support and therefore were discarded for further and 2010 included 3 months of the rainy season and 2 months
analysis (Pettorelli et al. 2009). of the dry season every year. For all species, the EGV “Distance
ENFA deals with spatial autocorrelation between sampling to Waterholes” had the greatest coefficient value (Table 1). For
points with a 10-fold jack-knife cross-validation, where the jaguars, pumas, and gray brocket deer, for all models consid-
data set is divided into 10 k-independent partitions, using k − 1 ering total or seasonal records, altitude (DEM) was the 2nd
of them to calibrate the model, and computing the evaluator strongest EGV. Of those 3 species, jaguars had the lowest tol-
on the left-out partition (Hirzel et al. 2001). This procedure is erance value in all models. Considering models derived from
repeated k times, each time leaving out another partition. This total records, the marginality of jaguars was greater than that of
produces k estimations of the evaluator, allowing assessment of pumas, and tolerance was lower. In seasonal models, tolerance
its central tendency and variance (Hirzel et al. 2001). Spatial of pumas was greater than that of jaguars, especially in the dry
autocorrelation may have a particularly bad effect in the cross- season, where marginality of pumas was also greater (Table 1).
validation. If the data set is randomly 10-fold partitioned and The models of some species were classified by the AVI as
the presence points are aggregated, then the assumedly inde- consistent, although they were clearly classified as random
pendent partition that is being cross-validating will not be or with little support by the CBI threshold (i.e., for 3-banded
independent and will produce over-confidence in the model armadillos considering dry season records AVI = 0.533 and
accuracy. In BIOMAPPER, the cross-validation partitioning is CBI = −0.040; Table 2). We obtained 9 models with a CBI
by default geographical: the data partitions for model validation value over the 0.200 threshold (Table 2).
are partitioned evenly but randomly (Hirzel and Arlettaz 2003; Considering only CBI-approved models, we compared HS
Sattler et al. 2007), so that they do not overlap geographically, models derived from total records between species of predators,
minimizing the potential for spatial autocorrelation (Unger and between predators and potential prey. We also compared
et al. 2008). models derived from seasonal records within the same spe-
Species and model comparison.—We looked for interspe- cies between seasons (i.e., rainy versus dry seasons), between
cific differences in HS models through marginality and toler- predators within seasons, and between predators and potential
ance. We also examined differences among species and seasons prey within seasons. We found significant differences between
between the resulting HS maps that passed the CBI threshold. jaguars and pumas based on total records (P < 0.0001), and
Using Hawth’s Tools (Beyer 2004) for ArcGIS, we created 500 between seasons for jaguars (P < 0.0001). Differences between
random points inside our study area and extracted their HS jaguars and pumas in the dry season were not significant (P
values. Points with zero value for both HS maps being com- > 0.05), meaning that HS maps were similar. Comparing HS
pared were discarded. For all remaining points, their values maps between predators and prey, we found significant differ-
were compared with a paired Mann–Whitney test, using the ences between pumas and 9-banded armadillos based on total
statistical software R (R Development Core Team 2012). If the records (P < 0.005), jaguars and gray brocket deer in the rainy
results showed nonsignificant differences, we considered both season (P < 0.0001), and jaguars and 9-banded armadillos in


Table 1.—Coefficients of the ecogeographical variables under Ecological Niche Factor Analysis for the marginality factor for 6 mammalian species in the Serra da Capivara National
Park, Caatinga, Brazil (jaguar = jaguar; puma = puma; graydeer = gray brocket deer; collared = white-collared peccary; threearm = 3-banded armadillo; ninearm = 9-banded armadillo), for
habitat suitability models considering the total (tot) number of records, as well as records in the rainy (rai) and dry (dry) seasons. The number of records from which the models were created
is denominated as “n.” NDVI = normalized difference vegetation index.

Model name Distance human Distance to Slope Distance to Digital NDVI Marginality Tolerance n Number % % explained % explained by
settlements paved roads waterholes elevation rainy/ of factors information by marginality the 1st axis of
model dry season retained explained specialization
jaguar_tot −0.051 0.197 0.016 −0.712 0.556 0.376 0.705 0.514 81 5 0.987 0.589 0.200
puma_tot −0.070 0.194 −0.032 −0.712 0.529 0.410 0.681 0.560 71 5 0.985 0.507 0.217
graydeer_tot −0.037 0.197 0.090 −0.808 0.429 0.338 0.672 0.550 62 5 0.988 0.536 0.197
collared_tot −0.092 0.151 −0.085 −0.770 0.445 0.413 0.664 0.525 30 5 0.992 0.543 0.219
threearm_tot 0.232 0.362 −0.119 −0.717 0.394 0.363 0.641 0.525 35 5 0.991 0.431 0.244
ninearm_tot 0.111 0.307 0.017 −0.751 0.401 0.411 0.610 0.577 48 5 0.988 0.441 0.238
jaguar_rai −0.010 0.152 −0.075 −0.699 0.570 0.396 0.676 0.510 65 5 0.988 0.588 0.186
puma_rai 0.057 0.240 −0.104 −0.727 0.546 0.319 0.663 0.511 48 5 0.987 0.562 0.182
graydeer_rai −0.007 0.191 0.084 −0.780 0.439 0.394 0.685 0.523 43 5 0.988 0.588 0.177
collared_rai −0.167 0.065 0.005 −0.731 0.417 0.509 0.729 0.494 20 5 0.994 0.479 0.295
threearm_rai −0.059 0.387 −0.141 −0.594 0.501 0.471 0.737 0.479 21 5 0.990 0.545 0.225
ninearm_rai 0.108 0.298 0.016 −0.729 0.390 0.464 0.627 0.543 48 5 0.990 0.530 0.189
jaguar_dry 0.016 0.233 −0.071 −0.691 0.539 0.417 0.709 0.481 64 5 0.990 0.619 0.165
puma_dry −0.158 0.227 −0.077 −0.673 0.565 0.381 0.715 0.551 58 5 0.981 0.483 0.230
graydeer_dry −0.033 0.223 0.069 −0.813 0.422 0.325 0.676 0.548 51 5 0.989 0.474 0.214
collared_dry 0.105 0.267 −0.137 −0.707 0.420 0.472 0.707 0.312 15 5 0.998 0.763 0.120
threearm_dry 0.337 0.380 −0.194 −0.669 0.390 0.324 0.625 0.453 21 5 0.994 0.569 0.162
ninearm_dry −0.068 0.238 −0.006 −0.843 0.343 0.326 0.592 0.506 13 5 0.993 0.293 0.310
ASTETE ET AL.—ENFA, JAGUAR, PUMA, PREY, SEMIARID 5
6 JOURNAL OF MAMMALOGY

Table 2.—Ecological Niche Factor Analysis evaluation indices

(AVI = Absolute Validation Index and CBI = Continuous Boyce Index)
for habitat suitability models of 6 mammalian species (jaguar = jag-
uar; puma = puma; graydeer = gray brocket deer; collared = white-
collared peccary; threearm = 3-banded armadillo; ninearm = 9-banded
armadillo) in the Serra da Capivara National Park, Brazil. The mod-
els considered the total (tot) number of records, as well as records in
the rainy (rai) and dry (dry) seasons. Results are from the Ma (area-
adjusted median) algorithm. Models marked in bold were considered
as approved by the CBI (i.e., CBI > 0.2).

Model name AVI CBI

jaguar_tot 0.486 ± 0.315 0.468 ± 0.377
puma_tot 0.498 ± 0.306 0.417 ± 0.484
graydeer_tot 0.502 ± 0.418 0.111 ± 0.544
collared_tot 0.467 ± 0.450 0.035 ± 0.585
threearm_tot 0.517 ± 0.285 0.152 ± 0.478
ninearm_tot 0.513 ± 0.274 0.259 ± 0.446
jaguar_rai 0.494 ± 0.326 0.320 ± 0.464
uma_rai 0.513 ± 0.346 0.195 ± 0.519
graydeer_rai 0.482 ± 0.300 0.258 ± 0.542
collared_rai 0.450 ± 0.438 −0.020 ± 0.581
threearm_rai 0.450 ± 0.438 −0.059 ± 0.587
ninearm_rai 0.487 ± 0.274 0.244 ± 0.509
jaguar_dry 0.512 ± 0.232 0.448 ± 0.460
puma_dry 0.508 ± 0.273 0.344 ± 0.422
graydeer_dry 0.470 ± 0.283 0.205 ± 0.470 Fig. 2.—Habitat suitability maps from Ecological Niche Factor
collared_dry
Analysis based on camera-trapping records for jaguars (Panthera
threearm_dry 0.533 ± 0.341 −0.040 ± 0.447
ninearm_dry
onca) in the rainy (a) and dry season (b) in the Serra da Capivara
National Park, Caatinga, Brazil, resulting from the area-adjusted
median (Ma) algorithm. SCNP = Serra da Capivara National Park.

the rainy season (P < 0.0001). We found similarities in HS

maps between both jaguars and pumas and gray brocket deer
in the dry season, and between jaguars and 9-banded armadil-
los based on total records (all P > 0.05). HS maps also did not
differ significantly between seasons for gray brocket deer (P >
0.05). All HS maps for jaguars, pumas, and gray brocket deer
showed a more defined core area in the central and southern
region of the park (Figs. 2–4).
Considering Levins’s standardized index based on the total
set of records, jaguars had a narrower niche than pumas. For
both species, the index was wider in the dry season than con-
sidering the total set of records (dry and rainy season; Table 3).
Similarly, Lloyd’s asymmetric overlap index showed higher
values for both species in the dry season than considering the
total set of records. Both comparisons registered greater over-
lap between the jaguar’s niche and that of the puma than the
reverse, as showed by the greater Z21 values (Table 3). Those
values decreased for both species in the dry season, especially
for jaguars, by almost one-half. According to Pianka’s niche
overlap index, both niches were close to complete overlap Fig. 3.—Habitat suitability maps from Ecological Niche Factor Analysis
(1.0 indicates complete overlap) considering total records or based on camera-trapping records for pumas (Puma concolor) in the
records from the dry season (Table 3). We did not include mod- dry season in the Serra da Capivara National Park, Caatinga, Brazil,
resulting from the area-adjusted median (Ma) algorithm. HSI = habitat
els from the rainy season because the HS model for pumas in
suitability index; SCNP = Serra da Capivara National Park.
the rainy season was not approved by the CBI threshold. In
the Discriminant Analysis, DEM was the EGV with the larg-
est weight for the discriminant factor, which summarizes the Discussion
resources used by both jaguars and pumas (see Supplementary One of the greatest advantages of ENFA is its ability to make
Data SD3). an analysis of multiple species at the same time using the same
 ASTETE ET AL.—ENFA, JAGUAR, PUMA, PREY, SEMIARID 7

set of ecological factors. We found that the different species HS for jaguars in the Caatinga, Morato et al. (2014) found that
in this arid biome—where none of them are endemic—show elevation was the variable that most influenced presence of
preferences for the same resource: artificial waterholes. Three jaguars in this biome. The authors suggested 2 explanations:
of the 6 (jaguars, pumas, gray brocket deer) also showed the first, that higher elevations are somewhat sheltered from human
same preference for a 2nd factor: elevation. activity, and second, that the vegetation types in the elevated
The fact that distance from artificial waterholes was the EGV areas favor jaguar presence. We found that distance to human
with the highest coefficient for the marginality factor for some settlements was of little relevance to HS of these species. We
species was not surprising. In an ENFA modeling effort for jag- offer an alternative explanation related to thermoregulation: in
uars across Mexico, Rodríguez-Soto et al. (2011) found that a study of natural formations of SCNP, such as caves used as
arid vegetation was characterized by low suitability values. In refuges by prehistoric humans, Figueiredo and Puccioni (2006)
the present study, the park’s southern region holds the highest found that during the dry season, while the temperature in the
suitability for both jaguars and pumas (see Figs. 2 and 3). This hottest moments of the day could reach up to 50°C outside the
is SCNP’s most elevated region, with the highest concentration refuges, inside them it remained under 30°C. The extreme tem-
of waterholes as well as canyons, rocky formations, and deep peratures during the dry season could represent a constraint
forested valleys, in contrast to the central plateau of the park. for felids such as jaguars and pumas, which, unlike canids,
Those formations could be used by jaguars and pumas as ref- do not thermoregulate by panting and radiating heat from the
uges or resting places in the warmest hours of the day. Modeling skin (West 2005). In a typical Caatinga scenario with high
temperatures and deciduous vegetation, those animals would
most likely prefer places where they can thermoregulate during
daytime, such as waterholes and refuges. In semiarid regions
of Africa, waterholes are used by felids, and where available,
leopards (Panthera pardus) make use of caves as a strategy to
escape high daytime temperatures in hot climates and to reduce
water loss (Bothma 1998). This is supported by the mostly noc-
turnal activity patterns of jaguars and pumas at SCNP (Astete
et al. 2008; Foster et al. 2013).
In the rainy season, SCNP turns into a green landscape
and rainwater stores in natural formations on the rocks
(FUMDHAM 1998). Considering this seasonal change in veg-
etation and availability of water, we expected seasonal differ-
ences between the HS maps for all the species, but this trend
was only shown by jaguars. However, when Levin’s standard-
ized index was compared between jaguars and pumas, for both
species it was wider in the dry season than the combination
of dry and rainy season, suggesting that both species had to
explore a broader set of resources during the harsh dry season.
Food habits of jaguars and pumas are similar in the semiarid
Chaco (Taber et al. 1997), a habitat similar to the Caatinga,
and the high value of Pianka’s overlap index for these species
based on our data indicates the potential for competition. Those
results coincide with the similarities of HS maps between those
Fig. 4.—Habitat suitability maps from Ecological Niche Factor 2 species in the dry season. Competition between large preda-
Analysis based on camera-trapping records for gray brocket deer tors and mesopredators has been widely reported (Ritchie and
(Mazama gouazoubira) in the rainy (a) and dry season (b) in the Johnson 2009), with the mesopredator restricting its habitat
Serra da Capivara National Park, Caatinga, Brazil, resulting from the use and altering its foraging behavior to avoid the large preda-
area-adjusted median (Ma) algorithm. HSI = habitat suitability index; tor. The higher asymmetric niche overlap index Z21 and its
SCNP = Serra da Capivara National Park. decrease for both species in the dry season (more notable for

Table 3.—Results of the Discriminant Analysis and Niche Overlap between jaguars (Panthera onca) and pumas (Puma concolor) at Serra da
Capivara National Park, Brazil, considering the total (tot) number of records, as well as records in the dry (dry) season. LSI = Levins’ standardized
index; Pianka’s_O = Pianka’s niche overlap index; Jaguar-Puma(Z12) and Puma-Jaguar(Z21) = Lloyd’s asymmetric overlap index, a measure
of directional niche overlap: Z12 represents the density of species 2 (puma) encountered, on average, by species 1 (jaguar), whereas Z21 is the
reciprocal.

Model name Jaguar_LSI Puma_LSI Pianka’s_O Jaguar-Puma(Z12) Puma-Jaguar(Z21)

Jaguar_Puma_tot 0.163 0.180 0.922 6.963 7.944
Jaguar_Puma_dry 0.260 0.250 0.941 4.047 4.466
8 JOURNAL OF MAMMALOGY

the jaguar) could suggest that: 1) both species try to avoid each on medium-sized mammals like armadillos (Dasypus spp.) and
other more actively during the harsh dry season, or 2) that jag- anteaters (Tamandua tetradactyla), all of them weighing less
uar presence decreases everywhere in the dry season due to its than 15 kg, with birds, reptiles, and other species representing
lower tolerance, concentrating in critical regions most suitable less than 10% of their diet (Olmos 1993; Wolff 2001). Another
for the species, where it has a higher overlap with the puma’s aspect favorable for gray brocket deer is that deer are predomi-
niche. This last supposition is also supported by the seasonal nantly diurnal (Astete 2008), whereas jaguars and pumas are
HS maps for jaguars and pumas. predominantly nocturnal, so the chances to be depredated at
Previous studies of jaguars in SCNP, which sampled mostly waterholes are reduced.
the southern area (Silveira et al. 2009; Sollmann et al. 2013), Camera trapping has been effective for surveying medium-
found that jaguars occur at a density of 1.57/100 km2—a value and large-sized mammals (Karanth 1995; O’Connell et al.
unexpectedly high considering the harsh habitat of the study 2011), providing reliable identification of target species. Our
area (Sollmann et al. 2013), but still one of the lowest com- camera stations were designed to estimate abundance of jag-
pared to other habitats (Astete et al. 2008). The sampling area uars (Silveira et al. 2009) and therefore to maximize the chance
for those studies coincided with SCNP’s region with the high- of detecting jaguars and mid- to large-sized animals. The cam-
est HS values for jaguars and other species, which, assuming a era stations were fixed in place during the sample period, so
positive relationship between HS and abundance (VanDerWal that over time their chances to detect target species were higher,
et al. 2009), could explain the unexpected density value. It is but as a trade-off, the number of locations we could sample was
therefore conceivable that future population surveys in SCNP limited.
covering the park’s total area will result in lower density We focused our discussion on only 3 of the 6 species because
estimates. only they showed robust ENFA models, considering the total
Space use of solitary carnivores such as jaguars and pumas set of data and the seasonal separation of records. All cameras
is not only influenced by the distribution of habitats but also were placed on roads, which are thought to be used preferen-
by the distribution and abundance of their prey (Sandell 1989). tially by felids and other mid- to large-sized mammals (Tobler
SCNP’s southern region combines the greatest concentration of et al. 2008; Harmsen et al. 2010). Small animals, such as arma-
waterholes and a topography that favors the ambush by pred- dillos, are more likely to pass in front of a camera without being
ators, which seem to prefer prey availability more than prey photographed (Tobler et al. 2008) and other species, such as
vulnerability (Hopcraft et al. 2005). As expected, in the dry sea- collared peccaries, may have a tendency for crossing trails
son, the models predicted significant similarities between HS more than following them (Harmsen et al. 2010), leading to
maps of both felids and deer. sample sizes that are too small to build reliable ENFA models.
For gray brocket deer, as for jaguars and pumas, the distance Like other ENFA studies focused on mid- to large-sized
to waterholes also had the highest EGV coefficient for marginal- mammals (Pettorelli et al. 2009; Durant et al. 2010), we noted
ity. Previous research in SCNP observed a correlation between that the less conservative AVI classified the models as more
the spatial distribution of deer and the presence of waterholes robust than the CBI. The index values obtained in those stud-
(Wolff 2001). A study on herbivores in an African semiarid ies are in most cases similar to ours, approving models with a
reserve without permanent sources of water, but with artificial CBI with values over 0.2 (Pettorelli et al. 2009). As the areas
waterholes (Valeix et al. 2008), showed that when natural water sampled by Pettorelli et al. (2009) included semiarid regions,
abundance decreases, artificial waterholes become more cru- we also decided to accept a CBI threshold of 0.2, indicating
cial. In Kruger National Park, another protected area with water that the accuracy of our models was relatively low. This is a
management, Smit et al. (2007) found that surface water avail- general feature with species having a widespread distribution
ability is more important for water-dependent mixed feeders (Segurado and Araujo 2004) and has been proposed to explain
(grazer and browsers) than for less water-dependent browsers. the low predictive power of their models for carnivores, a rela-
Seemingly, gray brocket deer—a mixed feeder (Black-Décima tively generalist taxon with a broad ecological niche and less
et al. 2010)—is following the same water-dependent pattern. habitat specificity than other taxa (Durant et al. 2010). Similar
We suggest that the deer’s preference for higher elevation could to the carnivores, the prey species found in SCNP also have rela-
be explained because high-elevation areas in SCNP (especially tively wide geographic distributions, occupying different habi-
the central and southern region) are covered by woody vegeta- tats throughout South America. The exception is the 3-banded
tion (Emperaire 1984; FUMDHAM 1998), which is seemingly armadillo, endemic to the semiarid Caatinga biome and adjacent
preferred by the species (Serbent et al. 2011). For the deer, the Cerrado vegetation associations (Marinho-Filho et al. 1997).
EGV NDVI (vegetation) was also the 3rd most important factor. Spatial autocorrelation is always problematic with the use of
Artificial waterholes are cited as preferred places for preda- geographical space (Legendre 1993), although there are some
tors to stalk ungulate prey, something that could not only affect options to solve the problem (Diniz-Filho and Telles 2002;
prey distribution (Valeix et al. 2009) but also the local com- Segurado et al. 2006). We do not believe this issue is relevant
munity structure of the mammal assemblage (Owen-Smith and in our study because the applied method (ENFA) has a special
Mills 2006; De Boer et al. 2010). However, we do not think way to deal with the problem. ENFA, a presence-only algo-
waterholes represent a menace for gray brocket deer in SCNP. rithm, is a kind of principal component analysis (PCA—Hirzel
The diet of jaguars and pumas in SCNP is well studied, and et al. 2001), where points of occurrence are crossed with envi-
both species seem to share the same prey items, mostly feeding ronmental variables to produce different maps that correspond
 ASTETE ET AL.—ENFA, JAGUAR, PUMA, PREY, SEMIARID 9

to the factors in a PCA: the marginality and the specialization. Supplementary Data SD3.—Weights of the ecogeographical
As a PCA, the analysis extracts most of the data variance and variables (EGVs) in the discriminant factor, which summarizes
allocates it into uncorrelated factors that contain ecological the resources used by both jaguars and pumas in the different
information (Hirzel et al. 2004). The marginality component models, using the total (tot) set or records and the ones in the
would have a very low importance if the original data had a dry (dry) season.
strong autocorrelation. Consequently, the CBI would classify
the model as inadequate, which was not the case.
Due to our limitation to place cameras only in roads, there Literature Cited
were areas of SCNP that were not surveyed. However, those Araújo, M. B., and P. H. Williams. 2000. Selecting areas for spe-
areas did not have different conditions from those from which cies persistence using occurrence data. Biological Conservation
the models were derived, as cameras were placed proportion- 96:331–345.
Astete, S. 2008. Ecologia da onça-pintada nos Parques Nacionais
ally in the different vegetation and landscape types in SCNP.
Serra da Capivara e Serra das Confusões, Piauí. M.S. thesis,
Thus, our results were not biased by sampling the least Universidade de Brasília, Brasília, Brazil.
favored areas less than other areas, which could lead to spatial Astete, S., R. Sollmann, and L. Silveira. 2008. Comparative ecol-
autocorrelation. ogy of jaguars in Brazil. Cat News 42:9–14.
The semiarid Caatinga does not seem to be the most suit- Beyer, H. L. 2004. Hawth’s analysis tools for ArcGIS. http://www.
able habitat for predators like jaguars and pumas, or for some spatialecology.com/htools. Accessed February 2015.
of their prey, considering their physiological limitations. Using Black-Décima, P., et al. 2010. Brown brocket deer Mazama gouazou-
ENFA, we determined that the region with the combination of bira (Fischer, 1814). Pp. 190–201 in Neotropical cervidology:
rocky areas and the highest abundance of artificial waterholes biology and medicine of Latin American deer (J. M. B. Duarte and
in SCNP explained most of HS for gray brocket deer, jaguars, S. Gonzáles, eds.). Funep/IUCN, Jaboticabal, Brazil.
and pumas, especially during the hot, dry season. To our knowl- Bothma, J. du P. 1998. Family Felidae. Pp. 43–85 in Carnivore ecol-
ogy in arid lands (Adaptation of desert organisms) (J. L. Cloudsley-
edge, this is the 1st study to show the positive effects of water
Thompson, ed.). Springer-Verlag, Berlin, Germany.
management policies in protected areas in Brazil or South
Boyce, M. S., P. R. Vernier, S. E. Nielsen, and F. K. A. Schmiegelow.
America. This combination of characteristics has the potential 2002. Evaluating resource selection functions. Ecological
to make SCNP a key area in the maintenance of jaguars, pumas, Modeling 157:281–300.
and gray brocket deer in the Caatinga. Braunisch, V., K. Bollmann, R. F. Graf, and A. H. Hirzel. 2008.
Living on the edge, modeling habitat suitability for species at
the edge of their fundamental niche. Ecological Modeling 214:
153–167.
Acknowledgements Capobianco, J. P. R. 2002. Artigo base sobre os biomas brasileiros.
Pp. 117–155 in Meio mbiente Brasil, avanços e obstáculos pós-
We thank V. Braunisch for her valuable help with the ENFA
Rio-92 (A. Camargo, J. P. R. Capobianco, and J. A. P. de Oliveira,
analyses. We thank the Editor and 2 anonymous reviewers for orgs.). Estação Liberdade, São Paulo, Brazil.
helpful comments on earlier versions of this manuscript. We Chefaoui, R. M., and J. M. Lobo. 2008. Assessing the effects of
also thank IBAMA (Brazilian Environmental Agency) for the pseudo-absences on predictive distribution model performance.
research license and the FUMDHAM for logistical support in Ecological Modeling 210:478–486.
SCNP during these years, as well as for the DEM, slope, and Colwell, R. K., and D. Futuyama. 1971. Measurement of niche
waterholes GIS maps. GIS analysis was possible due to the sup- breadth and overlap. Ecology 52:567–576.
port of the RedeCom Cerrado/MCT, based on the Conservation De Angelo, C., A. Paviolo, and M. Di Bitetti. 2011. Differential
Planning Lab at the Zoology Department, University of Brasilia. impact of landscape transformation on pumas (Puma concolor)
SA received doctoral grants from CAPES and CNPq, and and jaguars (Panthera onca) in the Upper Paraná Atlantic Forest.
Diversity & Distributions 17:422–436.
post-doctoral grants from CNPq, Brazilian Science Agencies.
De Boer, W. F., et  al. 2010. Spatial distribution of lion kills deter-
Financial support was provided by CNPq to JMF and RBM, and
mined by the water dependency of prey species. Journal of
by FAPDF to JMF. This study is part of the JCF’s jaguar long- Mammalogy 91:1280–1286.
term monitoring program in the Caatinga. We thank all field Diniz-Filho, J. A. F., and M. P. C. Telles. 2002. Spatial autocor-
assistants for help with camera trapping, especially G. Penido, relation analysis and the identification of operational units for the
P. de Podestá, L. Gomes, G. Santoro, T. Boscarato, and J. Leite. conservation in continuous populations. Conservation Biology
16:924–935.
Durant, S. M., et  al. 2010. Does size matter? An investigation
Supplementary Data of habitat use across a carnivore assemblage in the Serengeti,
Tanzania. Journal of Animal Ecology 79:1012–1022.
Supplementary Data SD1.—Correlation matrix of the eco- Eastman, J. R. 2009. IDRISI Taiga. Clark University, Worcester,
geographical variables (EGVs) used in the Ecological Niche Massachusetts.
Factor Analysis. ESRI. 2009. ArcGIS Desktop: release 9.31. Environmental Systems
Supplementary Data SD2.—Values for all ecogeographical Research Institute Inc., Readlands, California.
variables at each camera-trap sampling point in the Serra da Emperaire, L. 1984. A região da Serra da Capivara (Sudeste do Piauí)
Capivara National Park, Caatinga, Brazil. e sua vegetação. Brasil Florestal 14:5–21.
10 JOURNAL OF MAMMALOGY

Figueiredo, D., and S. Puccioni (orgs.). 2006. Consolidação estru- three banded armadillo in the Cerrado of Central Brazil. Edentata
tural da Toca da Entrada do Pajaú: diagnóstico e proposta de inter- 3:11–13.
venção. IPHAN – Instituto do Patrimônio Histórico e Artístico Millennium Ecosystem Assessment. 2005. Ecosystems and human
Nacional, Teresina, Piauí, Brazil. well-being: biodiversity synthesis. World Resources Institute,
Fundação Museu Do Homem Americano (Fumdham). 1998. Parque Washington, D.C.
Nacional Serra da Capivara. Alínea Publicações Editora-Lage Ministério do Meio Ambiente (MMA). 2005. Análise das varia-
Design, Teresina, Brazil. ções da biodiversidade do bioma Caatinga: suporte a estratégias
Foster, V. C., et  al. 2013. Jaguar and puma activity patterns and regionais de conservação (F. S. Araújo, M. J. N. Rodal, and M.
predator-prey interactions in four Brazilian biomes. Biotropica R. V. Barbosa, orgs.). Ministério do Meio Ambiente, Brasília,
45:373–379. Brazil.
Glaylard, A., N. Owen-Smith, and J. F. Redfern. 2003. Surface Morato, R. G., K. M. P. M. de Barros, R. C. de Paula, and C. B. de
water availability: implications for heterogeneity and ecosys- Campos. 2014. Identification of priority corridors for jaguars in the
tem processes. Pp. 171–188 in The Kruger experience: ecology Caatinga biome, Brazil. PLoS One 9:e92950.
and management of Savanna heterogeneity (J. T. du Troit, K. H. O’Connell, A. F., J. D. Nichols, and K. U. Karanth (eds.). 2011.
Rogers, and H. C. Biggs, eds.). Island Press, Washington, D.C. Camera traps in animal ecology. Methods and analyses. Springer,
Guisan, A., and N. E. Zimmermann. 2000. Predictive habitat distribu- New York.
tion models in ecology. Ecological Modeling 135:147–186. Olmos, F. 1993. Notes on the food habits of Brazilian “Caatinga”
Harmsen, B. J., R. J. Foster, S. Silver, L. Ostro, and C. P. Doncaster. carnivores. Mammalia 7:126–130.
2010. Differential use of trail by forest mammals and the implica- Owen-Smith, N., and M. G. L. Mills. 2006. Manifold interactive
tions for camera-trap studies: a case study from Belize. Biotropica influences on the population dynamics of a multispecies ungulate
42:126–133. assemblage. Ecological Monographs 76:73–92.
Hirzel, A., and R. Arlettaz. 2003. Modeling habitat suitability for Palma, L., P. Beja, and M. Rodrigues. 1999. The use of sighting data
complex species distributions by environmental-distance geomet- to analyze Iberian lynx habitat and distribution. Journal of Applied
ric mean. Environmental Management 32:614–623. Ecology 36:812–824.
Hirzel, A. H., J. Hausser, D. Chessel, and N. Perrin. 2002. Pearce, J., and M. Boyce. 2006. Modeling distribution and abun-
Ecological-niche factor analysis: how to compute habitat-suitabil- dance with presence-only data. Journal of Applied Ecology
ity maps without absence data? Ecology 83:2027–2036. 43:405–412.
Hirzel, A. H., J. Hausser, N. Perrin, G. Le Lay, and V. Braunisch. Pettorelli, N., A. L. Lobora, M. J. Msuha, C. Foley, and S.
2008. Biomapper 4.07.303. Laboratory for Conservation Biology, M.  Durant. 2009. Carnivore biodiversity in Tanzania: revealing
University of Lausanne, Lausanne, Switzerland. the distribution patterns of secretive mammals using camera traps.
Hirzel, A. H., V. Helfer, and F. Métral. 2001. Assessing habitat- Animal Conservation 13:131–139.
suitability models with a virtual species. Ecological Modeling Pienaar, U. de V. 1985. Indications of progressive desiccation of the
145:111–121. Transvaal Lowveld over the past 100 years, and implications for
Hirzel, A. H., G. Le Lay, V. Helfer, C. Randin, and A. Guisan. 2006. the water stabilization programme in the Kruger National Park.
Evaluating the ability of habitat suitability models to predict spe- Koedoe 28:93–165.
cies presences. Ecological Modeling 199:142–152. R Development Core Team. 2012. R: a language and environment
Hirzel, A. H., B. Posse, P. A. Oggier, Y. C. Glenz, and R. Arlettaz. for statistical computing. R Foundation for Statistical Computing,
2004. Ecological requirements of a reintroduced species, with Vienna, Austria. http://www.R-project.org/. Accessed December
implications for release policy: the bearded vulture recolonizing 2014
the Alps. Journal of Applied Ecology 41:1103–1116. Ritchie, E. G., and C. N. Johnson. 2009. Predator interactions, meso-
Hopcraft, J. G. C., A. R. E. Sinclair, and C. Packer. 2005. Planning predator release and biodiversity conservation. Ecology Letters
for success: Serengeti lions seek prey accessibility rather than 12:982–998.
abundance. Journal of Animal Ecology 74:559–566. Rodríguez-Soto, C., et  al. 2011. Predicting potential distribu-
Hurlbert, S. H. 1978. The measurement of niche overlap and some tion of the jaguar (Panthera onca) in Mexico: identification
relatives. Ecology 59:67–77. of priority areas for conservation. Diversity & Distributions
Hutchinson, G. E. 1957. Concluding remarks. Cold Spring Harbour 17:350–361.
Symposium on Quantitative Biology 22:415–427. Rondinini, C., S. Stuart, and L. Boitani. 2005. Habitat suitability
Iriarte, J. A., W. L. Franklin, W. E. Johnson, and K. H. Redford. models and the shortfall in conservation planning for African ver-
1990. Biogeographic variation of food habits and body size of the tebrates. Conservation Biology 19:1488–1497.
America puma. Oecologia 85:185–190. Sandell, M. 1989. The mating tactics and spacing patterns of soli-
Karanth, K. U. 1995. Estimating tiger (Panthera tigris) populations tary carnivores. Pp. 164–182 in Carnivore behavior, ecology and
from camera-trap data using capture-recapture models. Biological evolution (J. L. Gittleman, ed.). Cornell University Press, Ithaca,
Conservation 71:333–338. New York.
Legendre, P. 1993. Spatial autocorrelation: trouble or new paradigm? Sattler, T., F. Bontadina, A. H. Hirzel, and R. Arlettaz. 2007.
Ecology 74:1659–1673. Ecological niche modeling of two cryptic bat species calls for
Mares, A. M., M. R. Willing, and T. E. Lacher Jr. 1985. The Brazilian a reassessment of their conservation status. Journal of Applied
Caatinga in South American zoogeography: tropical mammals in a Ecology 44:1188–1199.
dry region. Journal of Biogeography 12:57–69. Secretaria Do Meio Ambiente Da Presidência Da República (Smapr).
Marinho-Filho, J., M. M. Guimarães, M. L. Reis, F. H. G. Rodrigues, 1994. plano de manejo parque nacional serra da capivara. instituto
O. Torres, and G. Almeida. 1997. The discovery of the Brazilian brasileiro do meio ambiente, brasília, brazil.
 ASTETE ET AL.—ENFA, JAGUAR, PUMA, PREY, SEMIARID 11

Segurado, P., and M. B. Araujo. 2004. An evaluation of methods Treves, A., and U. Karanth. 2003. Human-carnivore conflict and
for modelling species distributions. Journal of Biogeography perspectives on carnivore management worldwide. Conservation
31:1555–1568. Biology 17:1491–1499.
Segurado, P., M. B. Araujo, and W. E. Kunin. 2006. Consequences of Unger, D. E., S. Fei, and D. S. Maehr. 2008. Ecological Niche Factor
spatial autocorrelation for niche-based models. Journal of Applied Analysis to determine habitat suitability of a recolonizing carni-
Ecology 43:433–444. vore. Pp. 237–250 in Proceedings of the 6th southern forestry and
Serbent, M. P., M. E. Periago, and G. C. Leyaud. 2011. Mazama natural resources GIS conference (P. Bettinger, K. Merry, S. Fei,
gouazoubira (Cervidae) diet during the dry season in the arid J. Drake, N. Nibbelink, and J. Hepinstall, eds.). Warnell School
Chaco of Córdoba (Argentina). Journal of Arid Environments of Forestry and Natural Resources, University of Georgia, Athens.
75:87–90. Valeix, M., et  al. 2009. Behavioral adjustments of African herbi-
Silva, J. M. C., M., Tabarelli, M. T. Fonseca, and L. Lins (eds.). 2004. vores to predation risk by lions: spatiotemporal variations influence
Biodiversidade da Caatinga: áreas e ações prioritárias para a con- habitat use. Ecology 90:23–30
servação. Ministério do Meio Ambiente, Brasília, Brazil. Valeix, M., H. Fritz, R. Matsika, F. Matsvimbo, and H. Madzikanda.
Silveira, L., et  al. 2009. Density of the near threatened jaguar 2008. The role of water abundance, thermoregulation, per-
Panthera onca in the Caatinga of north-eastern Brazil. Oryx ceived predation risk and interference competition in water
44:104–109. access by African herbivores. African Journal of Ecology 46:
Smit, I. P. J., C. C. Grant, and B. J. Devereux. 2007. Do artifi- 402–410.
cial waterholes influence the way herbivores use the landscape? Valeriano, M. M. 2008. TOPODATA: Guia para utilização de dados
Herbivore distribution patterns around rivers and artificial sur- geomorfológicos locais. Instituto Nacional de PesquisasEspaciais –
face water sources in a large African savanna park. Biological INPE. http://www.dsr.inpe.br/topodata. Accessed 10 Dezember
Conservation 136:85–99. 2014.
Sollmann, R., et  al. 2013. Combining camera-trapping noninvasive VanDerWal, J., L. P. Shoo, C. N. Johnson, and S. E. Williams. 2009.
genetic data in spatial capture-recapture framework improves den- Abundance and the environmental niche: environmental suitability
sity estimates for the jaguar. Biological Conservation 167:242–247. estimated from niche models predicts the upper limit of local abun-
Stockwell, D. R. B., and A. T. Peterson. 2002. Effects of sample size dance. American Naturalist 174:282–291.
on accuracy of species distribution models. Ecological Modeling West, P. 2005. The lion’s mane. American Scientist 93:226–235.
148:1–13. Wolff, F. 2001. Vertebrate ecology in Caatinga: A. Distribution of
Taber, A. B., A. J. Novaro, N. Neris, and F. H. Colman. 1997. The wildlife in relation to water. B. Diet of pumas (P. concolor) and rel-
food habits of sympatric jaguar and puma in the Paraguayan Chaco. ative abundance of felids. M.S. thesis, University of Missouri-St.
Biotropica 29:204–213. Louis, St. Louis.
Tobler, M. W., S. E. Carrillo-Percastegui, R. Leite Pitman,
R. Mares, and G. Powell. 2008. An evaluation of camera traps for Submitted 10 November 2015. Accepted 8 November 2016.
inventorying large- and medium-sized terrestrial rainforest mam-
mals. Animal Conservation 11:169–178. Associate Editor was Marcus Vieira.