Content uploaded by Julien Fattebert
Author content
All content in this area was uploaded by Julien Fattebert on Dec 17, 2018
Content may be subject to copyright.
Ecological Applications, 25(7), 2015, pp. 1911 –1921
Ó2015 by the Ecological Society of America
Structural habitat predicts functional dispersal habitat of a large
carnivore: how leopards change spots
JULIEN FATTEBERT,
1,2,4
HUGH S. ROBINSON,
1
GUY BALME,
1,3
ROB SLOTOW,
2
AND LUKE HUNTER
1,2
1
Panthera, 8 W 40th Street, 18th Floor, New York, New York 10018 USA
2
School of Life Sciences, Westville Campus, University of KwaZulu-Natal, Durban 4000 South Africa
3
Department of Biological Sciences, University of Cape Town, Cape Town 7701 South Africa
Abstract. Natal dispersal promotes inter-population linkage, and is key to spatial
distribution of populations. Degradation of suitable landscape structures beyond the specific
threshold of an individual’s ability to disperse can therefore lead to disruption of functional
landscape connectivity and impact metapopulation function. Because it ignores behavioral
responses of individuals, structural connectivity is easier to assess than functional connectivity
and is often used as a surrogate for landscape connectivity modeling. However using structural
resource selection models as surrogate for modeling functional connectivity through dispersal
could be erroneous. We tested how well a second-order resource selection function (RSF)
models (structural connectivity), based on GPS telemetry data from resident adult leopard
(Panthera pardus L.), could predict subadult habitat use during dispersal (functional
connectivity). We created eight non-exclusive subsets of the subadult data based on differing
definitions of dispersal to assess the predictive ability of our adult-based RSF model
extrapolated over a broader landscape. Dispersing leopards used habitats in accordance with
adult selection patterns, regardless of the definition of dispersal considered. We demonstrate
that, for a wide-ranging apex carnivore, functional connectivity through natal dispersal
corresponds to structural connectivity as modeled by a second-order RSF. Mapping of the
adult-based habitat classes provides direct visualization of the potential linkages between
populations, without the need to model paths between a priori starting and destination points.
The use of such landscape scale RSFs may provide insight into predicting suitable dispersal
habitat peninsulas in human-dominated landscapes where mitigation of human–wildlife
conflict should be focused. We recommend the use of second-order RSFs for landscape
conservation planning and propose a similar approach to the conservation of other wide-
ranging large carnivore species where landscape-scale resource selection data already exist.
Key words: conservation planning; functional connectivity; natal dispersal; Panthera pardus; resource
selection functions; South Africa; structural connectivity.
INTRODUCTION
Natal dispersal, the emigration of an animal from its
natal area to a new area where it settles and breeds
(Howard 1960), promotes demographic and genetic flow
among populations, and is key to the spatial structuring
of metapopulations (Pulliam 1988, Hanski and Simber-
loff 1997). Landscape heterogeneity can affect the
likelihood of emigration, movement, and survival of
dispersing individuals (Wiens 2001). In fragmented
landscapes, animal populations can be restricted to
discrete patches of suitable habitat (Fahrig 2003, Di
Minin et al. 2013). Degradation of landscape structure
beyond a specific threshold of an individual’s inter-patch
dispersal ability can therefore impact metapopulation
dynamics by reducing connectivity (Ricketts 2001,
Dolrenry et al. 2014). Because of the persistent and
pervasive threat of habitat fragmentation, there is a high
conservation value in maintaining and restoring con-
nectivity (Crooks and Sanjayan 2006).
Landscape connectivity is defined in both structural
and functional terms. Structural connectivity is based on
metrics of the spatial arrangement of suitable habitat
patches on the landscape (Taylor et al. 2006). Func-
tional connectivity is more complex and combines
habitat structure, species behavior, and an individual’s
dispersal ability (Tischendorf and Fahrig 2000, Baguette
and Dyck 2007). Because it ignores the behavioral
responses of individuals, structural connectivity is easier
to assess than functional connectivity and is often used
as a surrogate for functional connectivity modeling
(Taylor et al. 2006 ). Empirical data are needed to
understand and validate the relationship between
structural and functional connectivity (Ruckelshaus et
al. 1997).
Features of connectivity vary widely depending on
species, thus there is no simple, all-purpose method of
modeling connectivity (Fagan and Calabrese 2006,
Manuscript received 26 August 2014; revised 19 December
2014; accepted 16 January 2015. Corresponding Editor: N. T.
Hobbs.
4
E-mail: julien.fattebert@gmail.com
1911
Kadoya 2009, Zeller et al. 2012). Field observation of
habitat use by dispersing individuals has been suggested
as the best measure of functional connectivity (Beier and
Noss 1998, Elliot et al. 2014). Methods such as step-
selection functions (Fortin et al. 2005, Thurfjell et al.
2014) address habitat selection along movement path-
ways explicitly and have been used to model and map
dispersal movement between habitats (e.g., Squires et al.
2013). However, the wide-ranging movements of dis-
persers present significant challenges in monitoring such
movements, and meaningful samples are logistically
difficult to maintain since dispersers usually make up a
small proportion of the population (Waser et al. 2001).
The difficulty in studying dispersal movements
empirically has led to the use of numerous indirect
techniques to model functional landscape connectivity,
such as expert derived least-cost paths (LaRue and
Nielsen 2008), inverse resource selection function (RSF)
least-cost paths (Chetkiewicz and Boyce 2009), graph
theory (Fall et al. 2007), and circuit theory (McRae et al.
2008). Least-cost methods in general, and RSF-based
methods in particular, have been criticized for not
separating resource selection from movement locations
and thus confusing structural for functional dispersal
habitat. An extensive review of connectivity resistance
modeling found no published attempts to address this
issue (Zeller et al. 2012). Here we address this research
gap by empirically testing the relationship between
structural connectivity as predicted by resident adult
habitat use and functional connectivity, habitat use of
dispersing subadults.
We used leopard (Panthera pardus L.), a territorial
large carnivore capable of long-dispersal movements
(Fattebert et al. 2013), as our model species. We
hypothesized that, for this large carnivore, functional
connectivity corresponds to structural connectivity, and
predicted that dispersing subadult individuals would
therefore favor habitats used by non-dispersing resident
adults (Stamps 2001, Davis and Stamps 2004). We
tested how an RSF model (Manly et al. 2002) could
inform functional landscape connectivity through natal
dispersal. Specifically, (1) we developed a second-order
(Johnson 1980) RSF based on resident adult leopard
telemetry data and (2) assessed the ability of the adult-
based RSF model (structural connectivity) to predict
habitat use by dispersing subadult individuals over a
broader landscape (functional connectivity), under
different definitions of dispersal.
MATERIALS AND METHODS
Study area
We conducted this study in the Phinda-Mkhuze
Complex (PMC; 278330–27855 0S; 328060–32826 0E; 591
km
2
), South Africa located within the Maputaland
ecoregion. The prevailing vegetation types in the study
area are grasslands, woodland–savannah mosaics and
forests interspersed with several freshwater and estuary
systems on the Indian Ocean coast (Fattebert et al.
2013). Public (21%of the area) and private (5%)
protected areas in the region are embedded in a mosaic
of non-protected areas. In South Africa, most protected
areas and ranches harboring resident leopard popula-
tions are fenced. However, leopards move freely through
these fences and between neighboring properties, and
are exposed to increased anthropogenic mortality risks
in unprotected areas (Balme et al. 2009).
Telemetry Data
We captured leopards in the protected PMC between
2002 and 2011, following Balme et al. (2007). We
classified leopards into three age classes according to
typical morphological cues (Stander 1997, Balme et al.
2012): cubs were ,1 year old; subadults were between 1–
3 years old; and adults were .3 years old. We fitted
leopards with very high frequency (VHF) (250 g;
Sirtrack, Havelock North, New Zealand) or GPS collars
(420 g; Vectronic-Aerospace, Berlin, Germany) based on
body mass and expected movement ranges. Additional-
ly, we equipped collars fitted to subadult males with a
drop-off mechanism (;50 g; Sirtrack) programmed to
shed automatically 6–12 months after capture (Fattebert
et al. 2013). We located VHF-collared leopards every
three days on average from the ground using the
homing-in method or radio triangulation to the nearest
100 m. We programmed GPS collars to acquire two to
six fixes daily (fix success rate: 0.795 60.035 [mean 6
SE], n¼28 collars). We screened GPS data for
potentially large locational errors by removing three-
dimensional fixes with positional dilution of precision
(PDOP) .15 and two-dimensional fixes with PDOP .
5 (Lewis et al. 2007).
Adult leopard RSF training
To model structural landscape use by adult resident
leopards, we built an RSF (Manly et al. 2002) in a used-
available design at the second-order scale (Johnson
1980). We chose this scale as we believe that dispersal
and the establishment of a home range is by definition a
landscape scale, second-order process (Johnson 1980).
We defined the area available to adult leopards as the
minimum convex polygon (MCP) of combined locations
for all adult individuals plus a buffer corresponding to
the 3 km radius of an average female home range
(Rauset et al. 2013, Fattebert 2014). We used GPS
telemetry data for model training, and reserved VHF
data for model validation. To sample resource avail-
ability, we generated random pseudo-absences within
the buffered MCP (1559 km
2
) at a 1:1 ratio of used and
available locations (DeCesare et al. 2012). We sampled
landscape variables at each used and available point
using the package raster (Hijmans 2013) in the R
environment (R Core Team 2013; Fig. 1a).
We chose candidate explanatory variables based on
previous leopard habitat selection studies (Table 1).
Leopards are strongly associated with vegetation cover
and prey density (Balme et al. 2007, Edgaonkar 2008,
JULIEN FATTEBERT ET AL.1912 Ecological Applications
Vol. 25, No. 7
Simcharoen et al. 2008, Hebblewhite et al. 2011, Mondal
et al. 2013), and tend to select areas close to rivers and
other abundant water sources (Simcharoen et al. 2008,
Mondal et al. 2013). Leopards tend to avoid areas with
human disturbance (Edgaonkar 2008, Hebblewhite et al.
2011, Di Minin et al. 2013, Swanepoel et al. 2013).
Association with topography is less clear, some studies
report selection for terrain ruggedness (Edgaonkar
2008), while others identified selection for flat terrain
(Simcharoen et al. 2008).
We screened variables for collinearity using a cut-off
of jrj¼0.5. For continuous variables, we tested for both
linear and nonlinear (quadratic) responses. We com-
bined all uncorrelated covariates with a univariate P,
FIG. 1. Leopard second-order resource selection function (RSF) model training and validation steps: (a) extraction of
environmental variables to each used adult leopard GPS location and available random pseudo-absences at a 1:1 ratio within a
polygon encompassing all adult locations (for clarity, only locations for adult male M60 are shown); (b) external validation of the
model with out-of-sample adult leopard very high frequency (VHF) telemetry data (blue) within the training area; (c) external
validation of the extrapolated model using subadult leopard telemetry data. Inset map in panel (a) shows position of the study area
within South Africa.
October 2015 1913LEOPARD DISPERSAL HABITAT
0.25 in a multivariate fixed-effects model. We then
conducted a manual backward-stepwise model selection
procedure removing all nonsignificant variables from the
multivariate model until the effects of all remaining
variables were significant P0.05 (Hosmer and Leshow
2000). We fit the final best fixed-effects model as both a
two- and a three-level mixed-effects model, where
individual and sex were modeled as random intercepts
(GLLAMM; Rabe-Hesketh et al. 2004) using STATA
11 (Stata, College Station, Texas, USA.). We selected
our final model from among the fixed-effect, two- and
three-level GLLAMM models using AIC.
Adult leopard RSF validation
We validated our final model using both internal and
external methods. Internal validation consisted of tests
of sensitivity using classification success, pseudo-r
2
, and
the area under the receiver operating curve (AUC;
Hosmer and Leshow 2000). We then validated the model
externally using an independent, out-of-sample adult
leopard VHF data set. We first projected the predicted
relative probability of adult leopard use w
(x)
across the
study area, following Manly et al. (2002)
wðxÞ¼expðb1x1þb2x2þ...þbnxnÞð1Þ
where b
i
is the coefficient of variable x
i
. We reclassified
the modeled landscape into 10 equal area bins by
sampling the predicted RSF values with 10 000 random
locations (Boyce et al. 2002). Essentially, random
locations were used to find delineation points at which
the RSF is divided into equal percentiles. Validation
points were then overlaid onto this evenly distributed
landscape to test the hypothesis of preferential use of the
higher model values. We then used a one-tailed
Spearman rank correlation to compare the frequencies
of our reserved, out-of-sample adult leopard VHF
locations in each bin to the RSF bin’s rank (Fig. 1b).
A significant positive correlation between the RSF
ranking and the number of locations from our
validation data set was considered an indication of the
predictive ability of the RSF model (Boyce et al. 2002).
We further assessed the predictive ability of our final
model, following Johnson et al. (2006). We first
determined the median raw RSF scores for each ordinal
bin. We determined the utilization U(x
i
) value for each
bin, using
UðxiÞ¼wðxiÞAðxiÞ=X
j
wðxjÞAðxjÞð2Þ
where w(x
i
) is the midpoint RSF of bin iand A(x
i
) the
area of bin i. We estimated the expected number of
validation observations within each bin (N
i
) using
Ni ¼N3UðxiÞð3Þ
where Nis the total number of observed validation
locations used and U(x
i
) the utilization function (from
Eq. 2). We then used linear regression to compare the
expected number N
i
(from Eq. 3) to the observed
number of validation data in each bin. We assessed the
slope of the regression line for a significant difference
from a slope of zero, and we assessed the fit of the
regression using R
2
. A slope of zero indicates that the
model is not different from that of a random model
where use equals availability.
TABLE 1. Candidate environmental variables used in training a resource selection function (RSF)
for adult leopards in the Phinda-Mkhuze Complex, South Africa, 2002–2012.
Variable Data set resolution Type of variable
Distance to water 30 m continuous
Distance to human settlements 30 m continuous
Topography 30 m
Elevation continuous
Roughness continuous
Slope continuous
Aspect categorical
Enhanced vegetation index, EVI 537 m, resampled at 30 m
10-yr minimum continuous
10-yr maximum continuous
South African national landcover 30 m
Closed canopycategorical
Thicket categorical
Grassland categorical
Bare land categorical
Commercial agricultureàcategorical
Subsistence agriculture§ categorical
Eucalyptus plantation categorical
Sugarcane plantation categorical
Human infrastructures}categorical
Water bodies categorical
Wetland categorical
Forest and woodland classes combined.
àPermanent and temporary commercial agriculture classes combined.
§ Permanent and temporary subsistence agriculture classes combined.
}All built up classes combined.
JULIEN FATTEBERT ET AL.1914 Ecological Applications
Vol. 25, No. 7
RSF predictive ability of subadult habitat use
To assess how adult structural habitat predicted
functional landscape connectivity through natal dispers-
al, we assessed the ability of our adult-leopard RSF to
predict subadult leopard habitat use. We first extrapo-
lated our validated model to the area encompassing all
subadult telemetry locations (16 961 km
2
) using Eq. 1.
We then repeated the external validation procedures
described above using all out-of-sample subadult GPS
and VHF telemetry data (Fig. 1c). Distinction between
different movement patterns during different behaviors
is becoming central in resource selection studies
(Wilmers et al. 2013, Zeller et al. 2014 ). We therefore
evaluated the predictive ability of our model using eight
separate but non-exclusive subsets of the subadult
leopard data based on different definitions of dispersal
(i.e., differing movement types or spatial locations).
Leopards become independent within their natal
range before dispersing and settle in temporary home-
ranges during dispersal (Fattebert et al. 2013). There-
fore, we assessed subadult leopard habitat use with
regards to differing movement patterns during dispersal
using net displacement curves. Net displacement is the
Euclidean distance between a given location and the first
location of a trajectory. We identified breakpoints in
each subadult leopard’s net displacement curve using a
piece-wise regression model (PWR) in the R package
‘‘segmented’’ (Muggeo 2008). We used Bayesian infor-
mation criteria (BIC) and a threshold gap coefficient t
value ,2 to validate breakpoints against a linear
regression fit (Birkett et al. 2012). Our first subsets of
subadult dispersal locations were based on these break-
points in the time series, that distinguish between linear
movement during the transfer phase of dispersal and
sedentary, home-range-like movements (Fig. 2; Ciucci et
al. 2009).
Effective dispersal is defined as the number of home-
ranges an organism moves away from its natal range
before settling (Shields 1987). We then used a definition
of dispersal based on distance travelled, regardless of
movement patterns. We used net displacement to divide
all subadult locations into two categories: all locations
within (philopatry), or all locations beyond (effective
dispersal) one average female home-range diameter (6
km; Fattebert 2014), and repeated the validation process
for each of these subsets. As habitat selection patterns
by dispersing individuals can be constrained by habitat
cues gained in their natal area (Stamps 2001), these
subsets assessed the ability of our adult-based RSF
model to predict subadult leopard habitat use in familiar
and unfamiliar landscapes within and outside their natal
range, respectively.
In order to maximize the potential conservation
benefits of the entire matrix, there is a need to
understand landscape connectivity between patches of
suitable habitat. We also tested for the predictive ability
of the adult-based RSF on dispersing subadult long-
distance movements between patches within the unpro-
tected matrix. We extracted the protection status of the
land at each subadult leopard location, and we built two
validation subsets; one with all locations within, and one
with all locations outside of protected areas. Because
our adult telemetry data set (i.e., model training) was
centered on a protected area (Fig. 1), these subsets tested
for subadult selection in familiar (i.e., protected) and
non-familiar (i.e., unprotected) landscapes.
Finally, because we extrapolated the adult-based RSF
model beyond the training area to encompass all
subadult locations, we tested for the ability of the
adult-based RSF to predict subadult leopard habitat use
in the broader landscape only. We classified the subadult
leopard locations according to whether they were within
or outside the model training buffered-MCP, and we
built two validation subsets; one with all locations
within, and one with all locations outside, the adult-
based RSF training study area to test for selection in a
landscape completely novel to the dispersing animal.
RESULTS
Leopard captures and tracking
We captured and collared 74 leopards (45 males and
29 females). We obtained 29 290 used GPS locations
from 11 adult male and 6 adult female leopards for
model training. We collected 5489 VHF locations from
16 adult male and 18 adult female leopards for external
validation of the adult RSF. We collected a total of 3827
GPS and 3906 VHF locations for 41 individuals during
dispersal age (25 males and 16 females). Piecewise
regression on the net displacement curves showed that
12 subadult male and 3 subadult female leopards
displayed linear movement (n¼1153 locations) during
the dispersal phase (Appendix). The remaining 26
subadult leopards only displayed home-range-like,
sedentary movement (n¼6580).
FIG. 2. Example movement phases during dispersal of a
subadult male leopard (M68) identified using a piecewise
regression on the net displacement time series. Gray shaded
areas show linear dispersal movement phases, and white areas
show sedentary, home-range-like, movement phases.
October 2015 1915LEOPARD DISPERSAL HABITAT
Adult leopard RSF
Adding a random intercept for sex to the GLLAMM
(three-level) did not improve the model over a two-level
GLLAMM with a random intercept for each individual,
suggesting little variation between males and females
(Table 2). At the second order of selection, probability
of adult leopard use increased with closed woody
vegetation, thicket, and grassland. Adult leopard use
declined in wetlands, patches of subsistence agriculture,
and human infrastructure. The inclusion of a quadratic
response for level of roughness, distance to water, and
primary productivity suggested maximum use at inter-
mediate levels of these variables (Table 3). The adult-
based RSF validated well internally, with 22 098 of the
29 290 (75.4%) adult leopard locations correctly classi-
fied. The model accounted for 16.9%of variation in
leopard spatial use (pseudo-R
2
), and the area under the
ROC curve showed good discrimination (AUC ¼0.766).
External validation using 5489 adult leopard VHF
locations (Fig. 3a) showed high predictive ability of
the final model using both bin ranks correlation with
number of observation and regression between expected
vs. observed number of observations (Table 4 ).
Predictive ability of subadult habitat use
The extrapolated final adult-based RSF model had
high predictive ability of the 7733 subadult leopard
telemetry locations (Fig. 3b; Table 4). When using
subsets of the subadult data, the model had strong
predictive ability of the linear movement data (Fig. 3c),
the sedentary movement data (Fig. 3d), subadult habitat
use beyond (Fig. 3e), and within (Fig. 3f ) one average
female home-range diameter, subadult habitat use
within (Fig. 3g) and outside (Fig. 3h) of protected areas,
as well as subadult habitat use within the training area
(Fig. 3i; Table 4). The predictive ability of the adult-
based RSF model of subadult habitat use outside of the
training area (Fig. 3j) was moderate using rank
correlation, and did not differ from a random model
when tested using Johnson et al.’s (2006) regression
method (Table 4).
DISCUSSION
Our findings suggest that for a wide ranging apex
carnivore, functional connectivity through natal disper-
sal corresponds to structural connectivity as modeled by
an adult-based second-order RSF. Subadult leopards
favored the most suitable habitats for both dispersal and
non-dispersal movement through the landscape. There-
fore, second-order resource selection models based on
resident adult data collected in protected areas may
provide insight into likely dispersal routes and connec-
tivity between populations, including through the matrix
of unprotected areas.
TABLE 2. Results of model selection for an adult leopard second-order RSF, Phinda-Mkhuze
Complex, South Africa, 2002–2012.
Model Log-likelihood df AIC DAIC
Mixed effect, individual 33 192.37 13 66 410.7 0.0
Mixed effect, individual/sex 33 194.58 14 66 417.2 6.5
Fixed effect 33 773.79 12 67 571.6 1160.9
Notes: The final fixed-effect model was compared to mixed-effect models that included a random
effect for individual, or a random effect for individual nested into sex. Selection was based on
model parsimony Aikake’s information criteria (AIC).
TABLE 3. Fixed-effect coefficients of the final adult leopard second-order RSF model, Phinda-
Mkhuze Complex, South Africa, 2002–2012.
Coefficient bSE
95%CI
Lower Upper
Intercept, b
0
8.7 0.3 9.2 8.1
Roughness 0.006 0.001 0.003 0.008
Distance to water0.2 0.02 0.2 0.3
Distance to waterà0.04 0.002 0.04 0.03
Minimum EVI0.006 0.0002 0.005 0.006
Minimum EVIà0.000001 0.00000006 0.000001 0.0000007
Closed canopy 0.4 0.04 0.3 0.5
Thicket 0.8 0.04 0.7 0.9
Grassland 1.3 0.04 1.2 1.4
Wetland 1.9 0.1 2.0 1.7
Subsistence agriculture 3.2 0.1 3.4 3.0
Infrastructures 2.0 0.2 2.4 1.5
Notes: The constant b
0
includes as the reference category the landcover classes presented in
Table 1 that do not have a significant effect in the model. All values are significant at P,0.01.
Linear term.
àQuadratic term.
JULIEN FATTEBERT ET AL.1916 Ecological Applications
Vol. 25, No. 7
Resource selection patterns depend on factors other
than resource availability, and animals need to balance
the trade-off between the benefits of exploiting resources
and the costs associated with accessing them (Brown et
al. 1999). Large carnivores face selection trade-offs
between abundance and vulnerability of their prey
(Balme et al. 2007), avoidance of competitors or
predators of their own (Vanak et al. 2013), and the
avoidance of anthropogenic disturbance in human-
dominated landscapes (Martin et al. 2010). Optimizing
FIG. 3. Proportion of the validation telemetry location subsets in ordinal bins of adult leopard second-order RSF model,
Phinda-Mkhuze Complex, South Africa, 2002–2012. Panels show the distribution of (a) adult leopard VHF locations, (b) all
subadult leopard telemetry locations, (c) subadult leopard linear movement locations, (d) subadult leopard sedentary movement
locations, (e) subadult leopard location beyond one average adult female home-range diameter, (f ) subadult leopard locations
within one average adult female home range, (g) subadult leopard locations within protected areas, (h) subadult leopard locations
outside protected areas, (i ) subadult leopard locations within the RSF model training area, and ( j) subadult leopard locations
outside of the RSF model training area. Number of locations is indicated for each validation subset.
October 2015 1917LEOPARD DISPERSAL HABITAT
such trade-offs is a hierarchical process, and resource
selection patterns can differ depending on spatial or
temporal scale (Manly et al. 2002, DeCesare et al. 2012),
life stage (Naves et al. 2003), sex (Conde et al. 2010),
behavior (Wilmers et al. 2013), or movement pattern
(Zeller et al. 2014).
At the second order of selection (Johnson 1980),
breeding, territorial adult individuals are expected to
monopolize the best habitat patches in the landscape
(‘‘ideal despotic distribution’’; Fretwell and Lucas 1970)
and can force dispersers into suboptimal habitats
(Palomares et al. 2000, Mosser et al. 2009). One obstacle
to quantifying dispersal habitat is therefore the percep-
tion that empirical data from actual dispersers must be
used to model functional landscape connectivity (e.g.,
Beier and Noss 1998), and that using structural resource
selection models as surrogate could be erroneous (Zeller
et al. 2012). However, given that .20 individuals and
approximately .50 locations per individual are required
to accurately assess resource use in large mammals
(Leban et al. 2001), many telemetry based studies avoid
collaring potential dispersers, or are unable to, in favor
of collecting data on resident animals (e.g., Chetkiewicz
and Boyce 2009, Squires et al. 2013). Here we
demonstrated that a simple second-order used-available
design RSF based on adult telemetry data predicts
habitat use by subadult individuals, regardless of the
definition of dispersal considered. Even outside our
study area completely, and where our model predicted
leopard dispersal habitat most poorly (Table 4),
approximately 70%of the subadult locations fell in the
five highest-ranking adult-based habitat classes (Fig. 3j).
External validation of our RSF model using out-of-
sample adult VHF locations indicated a highly
predictive surface, suggesting little bias of sampling
availability in the training area (Northrup et al. 2013).
However, because we extrapolated our model to the
wider landscape where the range of some variables
might exceed that found in the training area, our
prediction of habitat use by dispersing animals could
be biased (Hirzel and Le Lay 2008). If further
validation is required, e.g., species for which sufficient
telemetry and habitat selection data do not exist, fecal
DNA sampling over a broad landscape could yield an
assessment of effective linkages ( i.e., implying breeding
and gene flow) between population patches (Stenglein
et al. 2010, Shafer et al. 2012). Nonetheless, as we
validated the model predictions using naive subsets of
subadult leopards during the transfer phase of dispers-
al, we believe that we achieved a balanced trade-off of
the model’s generality, reality, and precision (Levins
1966).
Dispersing leopards used habitats in accordance to
adult selection patterns. These findings are consistent
with habitat cueing, with dispersers selecting for
habitat similar to their natal range (Stamps 2001).
Habitat cueing is favored in species where habitat
assessment and dispersal is accompanied by a high risk
of mortality and when they show a high level of home-
range fidelity following settlement (Davis and Stamps
2004). Therefore functional dispersal habitat mapping
using adult-based second-order RSF may be well
suited to several species of large carnivore that display
both traits (e.g., grizzly bear Ursus arctos and puma
Puma concolor [Chetkiewicz and Boyce 2009], grey
wolf Canis lupus and Eurasian lynx Lynx lynx [Huck et
al. 2010], African wild dog Lycaon pictus [Davies-
Mostertetal.2012],lionPanthera leo [Elliot et al.
2014]).
Indirect methods of modeling connectivity, such as
least-cost methods, have been criticized for being too
reliant on expert opinion and, while showing the most
favorable path between pairs of points in the landscape
(e.g., Elliot et al. 2014), may have little ability to
actually determine paths most likely taken by dispers-
ers between patches (Sawyer et al. 2011, Zeller et al.
2012). Here we provide a simple approach to model
functional connectivity that only requires structural
habitat selection data, the focus of most telemetry-
based studies. Furthermore, mapping of the adult-
TABLE 4. External validation statistics of a second-order RSF for leopards, Phinda-Mkhuze
Complex, South Africa, 2002–2012.
Validation data set
No.
locations
Spearman rank
correlation
Observed vs. expected
regression
r
S
PbPR
2
Adult VHF locations 5489 0.99 ,0.001 0.88 ,0.001 0.78
Subadult locations
All 7733 0.96 ,0.001 0.82 0.001 0.77
Linear movement 1153 0.96 ,0.001 0.89 ,0.001 0.83
Sedentary movement 6580 0.95 ,0.001 0.81 0.001 0.75
.1 Adult female home range 2180 0.9 ,0.001 0.55 0.033 0.38
,1 Adult female home range 5553 1 ,0.001 0.99 0.001 0.73
Inside protected areas 7175 0.96 ,0.001 0.85 0.001 0.76
Outside protected areas 558 0.79 0.005 0.47 ,0.001 0.83
Inside the RSF training area 7229 0.96 ,0.001 0.87 ,0.001 0.78
Outside the RSF training area 504 0.77 0.007 0.17 0.314 0.02
Note: The parameter bis the slope coefficient of the regression.
P,0.05 indicates a slope significantly different from zero.
JULIEN FATTEBERT ET AL.1918 Ecological Applications
Vol. 25, No. 7
based habitat classes provides direct visualization of
the potential linkages between populations, without the
need to model paths between a priori starting and
destination points (Fig. 1c).
Identifying habitats where mitigation of human–
wildlife conflicts should be focused in order to maintain
functional inter-population connectivity is an essential
step towards landscape conservation planning of large
carnivores. Increased anthropogenic mortality risks in
otherwise structurally suitable patches can lead to the
loss of landscape functionality in attractive sinks less
permeable to dispersal movements (Gundersen et al.
2001). Peninsulas of high quality habitat may draw
dispersers into human-dominated landscapes, not only
frustrating dispersal, but increasing the probability of
human conflicts (Maehr et al. 2002, Balme et al. 2010).
Conservation strategies should therefore aim at miti-
gating conflict and increased anthropogenic mortality
risk in suitable patches outside of protected areas
(Balme et al. 2009). The use of a landscape scale RSF
may provide insight into predicting such high conflict
areas (i.e., dispersal habitat peninsulas) where mitiga-
tion could be focused. In conjunction with source
population demographic and socio-spatial stability that
promotes dispersal (Fattebert 2014), the maintenance
of structurally suitable habitat tracts throughout the
matrix of unprotected land has high potential to
increase the likelihood of persistence of leopard
populations in the region. Therefore, this method could
be used as a basis for a range-wide assessment of
leopard habitat including in the matrix of unprotected
areas.
Ultimately, landscape connectivity models that in-
clude the requirements of different carnivore species
could inform multi-species conservation and matrix
management planning (Brodie et al. 2015). In a free-
ranging large carnivore, dispersal habitat (functional
connectivity) can be modeled using resource selection
information based on adult breeding individuals (struc-
tural connectivity). Therefore, such landscape-scale
conservation planning could be implemented swiftly in
areas where landscape-scale resource selection data
already exist for adults of the target species. Fortunate-
ly, such data exist for many imperiled large carnivores
although the utilization of many datasets has rarely
expanded beyond a site-specific focus (Ray et al. 2005).
Our approach provides a mechanism to utilize under-
used data and, in some cases, obviate the need to repeat
expensive and protracted telemetry studies in similar
scenarios.
ACKNOWLEDGMENTS
The animal handling procedures for this study were
approved by the Animal Ethics Subcommittee of the University
of KwaZulu-Natal Ethics Committee (approval 051/12/Ani-
mal). Panthera, Albert and Didy Hartog, the Timbo Founda-
tion, and UKZN funded this study. We thank &Beyond,
EKZNW, and the neighboring land-owners for allowing us to
conduct research on their land and are grateful to everyone who
assisted with fieldwork, particularly T. Dickerson, V. Mitchell,
K. Pretorius, J. Mattheus, and S. Naylor.
LITERATURE CITED
Baguette, M., and H. Dyck. 2007. Landscape connectivity and
animal behavior: functional grain as a key determinant for
dispersal. Landscape Ecology 22:1117–1129.
Balme, G. A., L. Hunter, and A. R. Braczkowski. 2012.
Applicability of age-based hunting regulations for African
leopards. PLoS ONE 7:e35209.
Balme, G., L. Hunter, and R. Slotow. 2007. Feeding habitat
selection by hunting leopards Panthera pardus in a woodland
savanna: prey catchability versus abundance. Animal Behav-
iour 74:589–598.
Balme, G., R. Slotow, and L. T. B. Hunter. 2009. Impact of
conservation interventions on the dynamics and persistence
of a persecuted leopard (Panthera pardus) population.
Biological Conservation 142:2681–2690.
Balme, G. A., R. Slotow, and L. T. B. Hunter. 2010. Edge
effects and the impact of non-protected areas in carnivore
conservation: leopards in the Phinda-Mkhuze Complex,
South Africa. Animal Conservation 13:315–323.
Beier, P., and R. F. Noss. 1998. Do habitat corridors provide
connectivity? Conservation Biology 12:1241–1252.
Birkett, P. J., A. T. Vanak, V. M. R. Muggeo, S. M. Ferreira,
and R. Slotow. 2012. Animal perception of seasonal
thresholds: changes in elephant movement in relation to
rainfall patterns. PLoS ONE 7:e38363.
Boyce, M. S., P. R. Vernier, S. E. Nielsen, and F. K. A.
Schmiegelow. 2002. Evaluating resource selection functions.
Ecological modeling 157:281–300.
Brodie, J. F., A. J. Giordano, B. Dickson, M. Hebblewhite, H.
Bernard, J. Mohd-Azlan, J. Anderson, and L. Ambus. 2015.
Evaluating multispecies landscape connectivity in a threat-
ened tropical mammal community. Conservation Biology
29:122–132.
Brown, J. S., J. W. Laundre
´, and M. Gurung. 1999. The
ecology of fear: optimal foraging, game theory, and trophic
interactions. Journal of Mammalogy 80:385–399.
Chetkiewicz, C.-L. B., and M. S. Boyce. 2009. Use of resource
selection functions to identify conservation corridors. Journal
of Applied Ecology 46:1036–1047.
Ciucci, P., W. Reggioni, L. Maiorano, and L. Boitani. 2009.
Long-distance dispersal of a rescued wolf from the northern
Apennines to the western Alps. Journal of Wildlife Manage-
ment 73:1300–1306.
Conde, D. A., F. Colchero, H. Zarza, N. L. Christensen, Jr.,
J. O. Sexton, C. Manterola, C. Cha
´vez, A. Rivera, D.
Azuara, and G. Ceballos. 2010. Sex matters: modeling male
and female habitat differences for jaguar conservation.
Biological Conservation 143:1980–1988.
Crooks, K. R., and M. Sanjayan. 2006. Connectivity conser-
vation: maintaining connections for nature. Pages 1–19 in
K. R. Crooks and M. Sanjayan, editors. Connectivity
conservation. Cambridge University Press, Cambridge, UK.
Davies-Mostert, H. T., J. F. Kamler, G. Michael, L. Mills,
C. R. Jackson, G. S. A. Rasmussen, R. J. Groom, and D. W.
Macdonald. 2012. Long-distance transboundary dispersal of
African wild dogs among protected areas in southern Africa.
African Journal of Ecology 50:500–506.
Davis, J. M., and J. A. Stamps. 2004. The effect of natal
experience on habitat preferences. Trends in Ecology &
Evolution 19:411–416.
DeCesare, N. J., et al. 2012. Transcending scale dependence in
identifying habitat with resource selection functions. Ecolog-
ical Applications 22:1068–1083.
Di Minin, E., L. T. B. Hunter, G. A. Balme, R. J. Smith, P. S.
Goodman, and R. Slotow. 2013. Creating larger and better
connected protected areas enhances the persistence of big
game species in the Maputaland-Pondoland-Albany biodi-
versity hotspot. PLoS ONE 8:e71788.
October 2015 1919LEOPARD DISPERSAL HABITAT
Dolrenry, S., J. Stenglein, L. Hazzah, R. S. Lutz, and L. Frank.
2014. A metapopulation approach to African lion (Panthera
leo) conservation. PLoS ONE 9:e88081.
Edgaonkar, A. 2008. Ecology of the leopard (Panthera pardus)
in Bori Wildlife Sanctuary and Satpura National Park, India.
Dissertation. University of Florida, Gainesville, Florida,
USA.
Elliot, N. B., S. A. Cushman, D. W. Macdonald, and A. J.
Loveridge. 2014. The devil is in the dispersers: predictions of
landscape connectivity change with demography. Journal of
Applied Ecology 51:1169–1178.
Fagan, W. F., and J. M. Calabrese. 2006. Quantifying
connectivity: balancing metric performance with data re-
quirements. Pages 297–317 in E. K. R. Crooks and M.
Sanjayan, editors. Connectivity conservation. Cambridge
University Press, Cambridge, UK.
Fahrig, L. 2003. Effects of habitat fragmentation on biodiver-
sity. Annual Review of Ecology, Evolution, and Systematics
34:487–515.
Fall, A., M.-J. Fortin, M. Manseau, and D. O’Brien. 2007.
Spatial graphs: principles and applications for habitat
connectivity. Ecosystems 10:448–461.
Fattebert, J. 2014. Spatial ecology of a leopard population
recovering from over-harvest. Dissertation. University of
KwaZulu-Natal, Durban, South Africa.
Fattebert, J., T. Dickerson, G. Balme, R. Slotow, and L.
Hunter. 2013. Long-distance natal dispersal in leopard
reveals potential for a three-country metapopulation. South
African Journal of Wildlife Research 43:61–67.
Fortin, D., H. L. Beyer, M. S. Boyce, D. W. Smith, T.
Duchesne, and J. S. Mao. 2005. Wolves influence elk
movements: behavior shapes a trophic cascade in Yellow-
stone National Park. Ecology 86:1320–1330.
Fretwell, S. D., and H. L. Lucas. 1970. On territorial behavior
and other factors influencing habitat distribution in birds. I.
Theoretical development. Acta Biotheoretica 19:16–36.
Gundersen, G., E. Johannesen, H. P. Andreassen, and R. A.
Ims. 2001. Source–sink dynamics: how sinks affect demog-
raphy of sources. Ecology Letters 4:14 –21.
Hanski, I., and D. Simberloff. 1997. The metapopulation
approach, its history, conceptual domain and application to
conservation. Pages 5–26 in I. Hanski and M. E. Gilpin,
editors. Metapopulation biology: ecology, genetics and
evolution. Academic Press, San Diego, California, USA.
Hebblewhite, M., D. G. Miquelle, A. A. Murzin, V. V.
Aramilev, and D. G. Pikunov. 2011. Predicting potential
habitat and population size for reintroduction of the Far
Eastern leopards in the Russian Far East. Biological
Conservation 144:2403–2413.
Hijmans, R. J. 2013. raster: geographic data analysis and
modeling. R package version 2.1-49. http://cran.r-project.
org/web/packages/raster/index.html
Hirzel, A. H., and G. Le Lay. 2008. Habitat suitability
modeling and niche theory. Journal of Applied Ecology
45:1372–1381.
Hosmer, D. W., and S. Leshow. 2000. Applied logistic
regression. Second edition. John Wiley and Sons, New York,
New York, USA.
Howard, W. E. 1960. Innate and environmental dispersal of
individual vertebrates. American Midland Naturalist 63:152–
161.
Huck, M., W. Jedrzejewski, T. Borowik, M. M. Osz-Cielma,
K. Schmidt, B. Jedrzejewska, S. Nowak, and R. W.
Mysajek. 2010. Habitat suitability, corridors and dispersal
barriers for large carnivores in Poland. Acta Theriologica
55:177–192.
Johnson, C. J., S. E. Nielsen, E. H. Merrill, T. L. McDonald,
and M. S. Boyce. 2006. Resource selection functions based on
use–availability data: theoretical motivation and evaluation
methods. Journal of Wildlife Management 70:347–357.
Johnson, D. H. 1980. The comparison of usage and availability
measurements for evaluating resource preference. Ecology
61:65–71.
Kadoya, T. 2009. Assessing functional connectivity using
empirical data. Population Ecology 51:5–15.
LaRue, M. A., and C. K. Nielsen. 2008. modeling potential
dispersal corridors for cougars in midwestern North America
using least-cost path methods. Ecological modeling 212:372–
381.
Leban, F. A., M. J. Wisdom, E. O. Garton, B. K. Johnson, and
J. G. Kie. 2001. Effect of sample size on the performance of
resource selection analyses. Pages 293–307 in J. J. Millspaugh
and J. M. Marzluff, editors. Radio tracking and animal
populations. Academic Press, San Diego, California, USA.
Levins, R. 1966. The strategy of model building in population
biology. American Scientist 54:421–431.
Lewis, J. S., J. L. Rachlow, E. O. Garton, and L. A. Vierling.
2007. Effects of habitat on GPS collar performance: using
data screening to reduce location error. Journal of Applied
Ecology 44:663–671.
Maehr, D. S., E. D. Land, D. B. Shindle, O. L. Bass, and T. S.
Hoctor. 2002. Florida panther dispersal and conservation.
Biological Conservation 106:187–197.
Manly, B. F. J., L. L. McDonald, D. L. Thomas, T. L.
McDonald, and W. P. Erickson. 2002. Resource selection by
animals. Statistical design and analysis for field studies.
Kluwer Academic, Dordrecht, The Netherlands.
Martin, J., M. Basille, B. V. Moorter, J. Kindberg, and J. E.
Swenson. 2010. Coping with human disturbance: spatial and
temporal tactics of the brown bear (Ursus arctos). Canadian
Journal of Zoology 88:875–883.
McRae, B. H., B. G. Dickson, T. H. Keitt, and V. B. Shah.
2008. Using circuit theory to model connectivity in ecology,
evolution, and conservation. Ecology 89:2712–2724.
Mondal, K., K. Sankar, and Q. Qureshi. 2013. Factors
influencing the distribution of leopard in a semiarid
landscape of Western India. Acta Theriologica 58:179–187.
Mosser, A., J. M. Fryxell, L. Eberly, and C. Packer. 2009.
Serengeti real estate: density vs. fitness-based indicators of
lion habitat quality. Ecology Letters 12:1050–1060.
Muggeo, V. M. R. 2008. segmented: an R package to fit
regression models with broken-line relationships. R News
8:20–25.
Naves, J., T. Wiegand, E. Revilla, and M. Delibes. 2003.
Endangered species constrained by natural and human
factors: the case of brown bears in northern Spain.
Conservation Biology 17:1276–1289.
Northrup, J. M., M. B. Hooten, C. R. Anderson, and G.
Wittemyer. 2013. Practical guidance on characterizing
availability in resource selection functions under a use
availability design. Ecology 94:1456–1463.
Palomares, F., M. Delibes, P. Ferreras, J. M. Fedriani, J.
Calzada, and E. Revilla. 2000. Iberian lynx in a fragmented
landscape: predispersal, dispersal, and postdispersal habitats.
Conservation Biology 14:809–818.
Pulliam, H. R. 1988. Sources, sinks, and population regulation.
American Naturalist 132:652–661.
R Core Team. 2013. R: a language and environment for
statistical computing. Foundation for Statistical Computing,
Vienna, Austria. http://www.R-project.org/
Rabe-Hesketh, S., A. Skrondal, and A. Pickles. 2004. Gener-
alized multilevel structural equation modeling. Psychometri-
ka 69:167–190.
Rauset, G., J. Mattisson, H. Andre
´n, G. Chapron, and J.
Persson. 2013. When species-ranges meet: assessing differ-
ences in habitat selection between sympatric large carnivores.
Oecologia 172:701–711.
Ray, J. C., L. T. B. Hunter, and J. Zigouris. 2005. Setting
conservation and research priorities for larger African
carnivores. Wildlife Conservation Society, New York, New
York, USA.
JULIEN FATTEBERT ET AL.1920 Ecological Applications
Vol. 25, No. 7
Ricketts, T. H. 2001. The matrix matters: effective isolation in
fragmented landscapes. American Naturalist 158:87–99.
Ruckelshaus, M., C. Hartway, and P. Kareiva. 1997. Assessing
the data requirements of spatially explicit dispersal models.
Conservation Biology 11:1298–1306.
Sawyer, S. C., C. W. Epps, and J. S. Brashares. 2011. Placing
linkages among fragmented habitats: do least-cost models
reflect how animals use landscapes? Journal of Applied
Ecology 48:668–678.
Shafer, A. B. A., J. M. Northrup, K. S. White, M. S. Boyce,
S. D. Cˆ
ote
´, and D. W. Coltman. 2012. Habitat selection
predicts genetic relatedness in an alpine ungulate. Ecology
93:1317–1329.
Shields, W. M. 1987. Dispersal and mating systems: investigat-
ing their causal connections. Pages 3–24 in B. D. Chepko-
Sade and Z. T. Halpin, editors. Mammalian dispersal
patterns: the effects of social structure on population
genetics. University of Chicago Press, Chicago, Illinois, USA.
Simcharoen, S., A. C. D. Barlow, A. Simcharoen, and J. L. D.
Smith. 2008. Home range size and daytime habitat selection
of leopards in Huai Kha Khaeng Wildlife Sanctuary,
Thailand. Biological Conservation 141:2242–2250.
Squires, J. R., N. J. DeCesare, L. E. Olson, J. A. Kolbe, M.
Hebblewhite, and S. A. Parks. 2013. Combining resource
selection and movement behavior to predict corridors for
Canada lynx at their southern range periphery. Biological
Conservation 157:187–195.
Stamps, J. A. 2001. Habitat selection by dispersers: integrating
proximate and ultimate approaches. Pages 230–242 in J.
Clobert, E. Danchin, A. A. Dhondt, and J. D. Nichols,
editors. Dispersal: individual, population, and community.
Oxford University Press, Oxford, UK.
Stander, P. E. 1997. Field age determination of leopards by
tooth wear. African Journal of Ecology 35:156–161.
Stenglein, J. L., M. De Barba, D. E. Ausband, and L. P. Waits.
2010. Impacts of sampling location within a faeces on DNA
quality in two carnivore species. Molecular Ecology Re-
sources 10:109–114.
Swanepoel, L. H., P. Lindsey, M. J. Somers, W. van Hoven,
and F. Dalerum. 2013. Extent and fragmentation of suitable
leopard habitat in South Africa. Animal Conservation 16:41–
50.
Taylor, P. D., L. Fahrig, and K. A. With. 2006. Landscape
connectivity: a return to the basics. Pages 29– 43 in K. R.
Crook and M. Sanjayan, editors. Connectivity conservation.
Cambridge University Press, Cambridge, UK.
Thurfjell, H., S. Ciuti, and M. Boyce. 2014. Applications of
step-selection functions in ecology and conservation. Move-
ment Ecology 2:4.
Tischendorf, L., and L. Fahrig. 2000. On the usage and
measurement of landscape connectivity. Oikos 90:7–19.
Vanak, A. T., D. Fortin, M. Thaker, M. B. Ogden, C. R. Owen,
S. Greatwood, and R. Slotow. 2013. Moving to stay in place:
behavioral mechanisms for coexistence of African large
carnivores. Ecology 94:2619–2631.
Waser, P. M., C. Strobeck, and D. Paetkau. 2001. Estimating
interpopulation dispersal rates. Pages 484 – 497 in J. L.
Gittleman, S. M. Funk, D. W. Macdonald, and R. K.
Wayne, editors. Carnivore conservation. Cambridge Univer-
sity Press, Cambridge, UK.
Wiens, J. A. 2001. The landscape context of dispersal. Pages
96–109 in J. Clobert, E. Danchin, A. A. Dhondt, and J. D.
Nichols, editors. Dispersal: individual, population, and
community. Oxford University Press, Oxford, UK.
Wilmers, C. C., Y. Wang, B. Nickel, P. Houghtaling, Y.
Shakeri, M. L. Allen, J. Kermish-Wells, V. Yovovich, and T.
Williams. 2013. Scale dependent behavioral responses to
human development by a large predator, the puma. PLoS
ONE 8:e60590.
Zeller, K., K. McGarigal, P. Beier, S. Cushman, T. W. Vickers,
and W. Boyce. 2014. Sensitivity of landscape resistance
estimates based on point selection functions to scale and
behavioral state: pumas as a case study. Landscape Ecology
29:541–557.
Zeller, K., K. McGarigal, and A. Whiteley. 2012. Estimating
landscape resistance to movement: a review. Landscape
Ecology 27:777–797.
SUPPLEMENTAL MATERIAL
Ecological Archives
The Appendix is available online: http://dx.doi.org/10.1890/14-1631.1.sm
Data Availability
Data associated with this paper have been deposited in Dryad: http://dx.doi.org/10.5061/dryad.fk2nh
October 2015 1921LEOPARD DISPERSAL HABITAT
