Poaching creates ecological traps within an iconic protected area

Abstract

Ecological traps occur when areas preferentially selected by a species harbour an unknown increased mortality risk or reduced fitness for the individuals utilizing them. If animals continue to utilize these habitats, rapid declines may result that threaten the persistence of the population. Both black and white rhinoceroses are plagued by severe, targeted rhino poaching in South Africa that may have population and species‐level consequences in the long term. Poaching can rapidly increase mortality and may create habitats that function as ecological traps for protected populations. We used spatially explicit data of live rhino and poached rhino carcasses in the Kruger National Park, South Africa, to define high‐ and low‐risk states for both black and white rhino species. The proportion of area functioning as ecological trap was similar for both species (black: 37.73%, white: 35.51%), while the proportion of safe harbour was considerably lower for black rhino (black rhino: 32.01%, white rhino: 44.74%). Species‐specific risk areas were condensed into management categories that reflect the actions most likely to be effective for overall rhino protection in those areas. ‘Threat’ area, representing ecological traps for both species, comprised 32.48% of southern Kruger; this represents the highest priority for anti‐poaching interventions. A further 31.03% was identified as ‘haven’, representing safe harbours for both species, which may benefit most from continued rhino monitoring and surveillance. Using these categories, authorities can prioritize the distribution of limited resources and tailor anti‐poaching and biological management actions according to the needs of each area for the concurrent protection of both rhino species. This work illustrates how the conservation of multiple species or taxa within a system can be simultaneously prioritized in vast areas where resources and/or capacity may be insufficient to undertake species‐specific approaches.

Full text

Poaching creates ecological traps within an iconic
protected area
N. leRoex
1,2
, C. Dreyer
3
& S. M. Ferreira
1
1 Scientific Services, South African National Parks, Skukuza, South Africa
2 Institute for Communities and Wildlife in Africa (iCWild), Department of Biological Sciences, University of Cape Town, Cape Town, South
Africa
3 Ranger Services, South African National Parks, Skukuza, South Africa
Keywords
rhino; illegal killing; habitat selection; species
distribution; wildlife management; poaching;
ecological trap; Kruger National Park.
Correspondence
Nikki le Roex, Scientific Services, South
African National Parks, Skukuza 1350, South
Africa.
Email: nikki.leroex@sanparks.org
Editor: Gurutzeta Guillera-Arroita
Associate Editor: Vincenzo Penteriani
Received 20 March 2019; accepted 30 July
2019
doi:10.1111/acv.12532
Abstract
Ecological traps occur when areas preferentially selected by a species harbour an
unknown increased mortality risk or reduced fitness for the individuals utilizing them.
If animals continue to utilize these habitats, rapid declines may result that threaten the
persistence of the population. Both black and white rhinoceroses are plagued by sev-
ere, targeted rhino poaching in South Africa that may have population and species-
level consequences in the long term. Poaching can rapidly increase mortality and may
create habitats that function as ecological traps for protected populations. We used
spatially explicit data of live rhino and poached rhino carcasses in the Kruger National
Park, South Africa, to define high- and low-risk states for both black and white rhino
species. The proportion of area functioning as ecological trap was similar for both spe-
cies (black: 37.73%, white: 35.51%), while the proportion of safe harbour was consid-
erably lower for black rhino (black rhino: 32.01%, white rhino: 44.74%). Species-
specific risk areas were condensed into management categories that reflect the actions
most likely to be effective for overall rhino protection in those areas. ‘Threat’area,
representing ecological traps for both species, comprised 32.48% of southern Kruger;
this represents the highest priority for anti-poaching interventions. A further 31.03%
was identified as ‘haven’, representing safe harbours for both species, which may ben-
efit most from continued rhino monitoring and surveillance. Using these categories,
authorities can prioritize the distribution of limited resources and tailor anti-poaching
and biological management actions according to the needs of each area for the concur-
rent protection of both rhino species. This work illustrates how the conservation of
multiple species or taxa within a system can be simultaneously prioritized in vast areas
where resources and/or capacity may be insufficient to undertake species-specific
approaches.
Introduction
Understanding the drivers of species occurrence and habi-
tat selection contributes to the conservation and manage-
ment of wild populations (Johnson & Gillingham, 2005).
This knowledge can provide insight into the complex nat-
ure of species-environment interactions, ecosystem function
and population dynamics, and can often be used to
explain or predict species performance (Morris, 2003).
Habitat selection by wild herbivores reflects a combination
of climatic, dietary and vegetation preferences, and is
commonly influenced by factors such as surface water
availability and vegetation quality (Fritz et al., 1996;
Muposhi et al., 2016). Competition and predation risk
may also impact herbivore distribution, particularly at a
local scale (Pulliam & Danielson, 1991; Valeix et al.,
2009). These factors create variability in habitat quality,
with high-quality habitat conferring fitness advantages
through increased reproductive potential and survival and
low-quality habitat conferring fitness costs through reduced
reproductive potential and survival (Delibes, Gaona, &
Ferreras, 2001). In addition to ecological factors, modern
anthropogenic disturbances such as fire, habitat degrada-
tion and infrastructure development can significantly affect
habitat selection in particular areas (Abrams et al., 2012;
Robertson, Rehage, & Sih, 2013).
Within protected areas, the drivers of habitat selection
should be dominated by ecological processes rather than
modern anthropogenic disturbances, with high-quality habitat
and the associated increased survival available for most
Animal Conservation  (2019) – ª2019 The Zoological Society of London 1
Animal Conservation. Print ISSN 1367-9430

species. However, in recent years, the rapid growth of the
illegal wildlife trade and corresponding increase in poaching
within protected areas in many countries have placed enor-
mous pressure on certain wildlife species. Poaching can
rapidly increase mortality within a population and may be
temporally or physically localized within a protected area
(Woodroffe & Ginsberg, 2008). If poaching is persistently
localized, and the increased mortality risk cannot be detected
by animals, poaching may create habitats that function as
ecological traps for protected populations (Abrams et al.,
2012). Ecological traps are areas or habitats that are equally
or more attractive than others, but harbour an increased mor-
tality risk or reduced fitness for the individuals utilizing them
(Schlaepfer, Runge, & Sherman, 2002). Ecological traps may
form when habitat quality cues remain unchanged but sur-
vival or fitness in an area is reduced, or habitat cues are dis-
torted and no longer reflect the true habitat quality
(Robertson & Hutto, 2006). If animals are unaware of the
increased risk and continue to utilize these habitats, this can
lead to rapid declines that threaten the persistence of the
population, and ultimately, species (Abrams et al., 2012;
Fletcher, Orrock, & Robertson, 2012). Human-induced mor-
tality such as poaching is an example of such a risk, and
may act as a substantial population-level threat if it persists
over long periods. From a conservation perspective, under-
standing the spatial patterns of habitat selection and human-
induced mortality in protected populations of endangered
species can enable the identification of high and low threat
areas, the design of innovative management approaches and
the optimal allocation of limited capacity and resources.
Recent work in this field has predominantly focused on
large carnivores or omnivores with the identification of eco-
logical traps for species such as brown bears Ursus arctos
(Falcucci et al., 2009; Northrup, Stenhouse, & Boyce, 2012;
Lamb et al., 2017; Penteriani et al., 2018), Andean bears
Tremarctos ornatus (S
anchez-Mercado et al., 2014), jaguars
Panthera onca (Romero-Mu~
noz et al., 2018) and leopards
Panthera pardus (Pitman et al., 2015). These studies have
focused on individual species where conservation strategies
are needed to offset conflict-related deaths, primarily in
human-modified landscapes. While methodologies and data
types differ between studies, all make use of animal loca-
tions and mortality data to predict areas of high risk for the
species in question. Poaching data, in conjunction with spe-
cies occurrence data, have been used to identify ecological
traps for Andean bears (S
anchez-Mercado et al., 2014) and
savanna elephants (Roever, van Aarde, & Chase, 2013) both
inside and outside of the protected areas and rank areas for
additional protection. It is not always possible, however, for
management authorities to prioritize a single species within a
protected area when others are also at risk, particularly when
the threat is severe and/or requires significant resources to
address.
Increasing demand for rhino horn to supply the illegal
wildlife trade has resulted in relentless poaching of rhino-
ceroses in the last decade, particularly in South Africa
(Emslie et al., 2016). The Kruger National Park, South
Africa, hosts the largest populations of two species: the
south-eastern black rhinoceros Diceros bicornis minor; here-
after black rhino and southern white rhinoceros Cera-
totherium simum simum; hereafter white rhino, and is one of
the few remaining free-ranging, ‘natural’populations in the
world. However, Kruger is plagued by severe, targeted rhino
poaching which is likely to have population and species-
level consequences in the long term (Ferreira, Greaver, &
Knight, 2011; Ferreira et al., 2018). The black rhinoceros is
critically endangered, with less than 6000 animals remaining
globally (Emslie et al., 2016). Black rhino are elusive and
difficult to monitor, and have been identified as a species of
special concern within Kruger by South African National
Parks (SANParks), due to their low numbers, declining
growth and poaching losses experienced. The white rhino is
currently listed as ‘near threatened’on the IUCN Red List of
Threatened Species (Emslie, 2012), with a global population
of approximately 20 000 animals (Emslie et al., 2016).
Despite their large population size, white rhino have suffered
the majority of the annual poaching fatalities over the last
5 years in South Africa (Emslie et al., 2016), and the Kruger
population shows a net annual decline (Ferreira et al., 2018).
Management within Kruger therefore need to utilize limited
resources in a manner that protects both rhino species simul-
taneously. Identifying high and low threat areas common
across both species may be the most practical way to ensure
maximum impact in rhino protection and management.
Spatially explicit data of live rhino and poached rhino car-
casses provide an opportunity to determine whether areas in
the southern Kruger National Park function as ecological
traps for rhinos. In this study, we identify areas with high
and low probability of poaching (Nielsen et al., 2006;
Roever et al., 2013; Sanchez-Mercado et al., 2014) and
relate these areas to rhino occurrence within the Kruger land-
scape. We expect that environmental variables will exert the
primary influence on live rhino occurrence, and anthro-
pogenic variables will be the key drivers of carcass occur-
rence. By overlaying these results, we classify high threat
areas (ecological traps) and low threat areas (safe harbours)
within southern Kruger according to the dual occurrence
probability and mortality risk for each rhino species. Finally,
we combine the species-specific risk areas into categories to
inform innovative management interventions that would
enable the optimal allocation of limited resources that maxi-
mizes the protection of both rhino species.
Materials and methods
Study area
The Kruger National Park (24°0041″S, 31°2907″E) is the
largest protected area in South Africa and encompasses
19 485 km
2
of low-lying savanna. The landscape varies
across the park, with landscape types classified according to
vegetation type, soil and geological characteristics (Gerten-
bach, 1983). Black rhinos are predominantly found south of
the Olifants river (Kruger, Reilly, & Whyte, 2008), most
likely as a result of their reintroduction locations in southern
Kruger. White rhino have a wider distribution across the
2Animal Conservation  (2019) – ª2019 The Zoological Society of London
Poaching creates ecological traps in a protected area N. le Roex et al.

park, but occur at their highest densities in the southern
region (Ferreira et al., 2018).
Within the southern region, an Intensive Protection Zone
(IPZ; Fig. 1) was established in 2014 to prioritize the secu-
rity of rhino in high-density areas using advanced technol-
ogy, equipment and infrastructure. The IPZ covers
approximately 4000 km
2
and is comprised of nine landscape
types, summarized as follows: Acacia thicket, Sabie/Croco-
dile thicket, Lowveld sour bushveld, mountain bushveld,
mixed Combretum/Terminalia woodland, Combretum wood-
land, Acacia savanna, Lebombo south and thornveld (Gerten-
bach, 1983). Southern Kruger also suffers the greatest
human incursion and poaching rates, likely as a result of
high rhino densities, intensive human settlement along the
boundary, international access and rudimentary fencing.
Rhino data
We used 5 years of rhino sightings data collected during
annual aerial rhino surveys from 2013 to 2017 in the IPZ of
the Kruger National Park. These surveys were conducted by
Jet Bell Ranger helicopter in August and September each
year, following a block-count approach; details can be seen
in Ferreira et al. (2015). Approximately, 50% of the avail-
able 3 93 km blocks were surveyed each year and animal
age, sex and GPS locations were recorded. Block selection
was randomly distributed across each landscape type. By
combining the annual data, we generated a spatially explicit
dataset of 723 black rhino sightings and 12 921 white rhino
sightings within the study area over a 5-year period. For
poached rhino, we extracted carcass locations of animals
killed between 2013 and 2017 from the large mammal car-
cass database maintained by the SANParks Environmental
Crime Investigative (ECI) unit. Only rhino deaths recorded
as ‘shot’were used in the analyses; unknown and natural
deaths were removed. This resulted in poached carcass data-
sets of 81 black rhino and 1366 white rhino.
Live rhino and carcass occurrence
We used a resource selection function (RSF) approach and
ran generalized linear models (GLMs) with a binomial error
structure to investigate the environmental and anthropogenic
variables influencing rhino occurrence and mortality in south-
ern Kruger. In addition to our live rhino locations, we gener-
ated an equal number of random points within the census
blocks to represent locations available to rhino. The binomial
response variable was thus 0 (random point within the sur-
veyed area) or 1 (rhino location). The predictor variables
included in the models were: distance to main rivers, land-
scape type (Gertenbach, 1983); woody cover (for black
rhino; Bucini, Saatchi, Hanan, Boone, & Smit, 2009) or
herbaceous biomass (for white rhino; Smit, 2011); terrain
ruggedness, distance to human activity (represented by ran-
ger stations), distance to roads and distance to fence. Woody
cover represents the percentage of tree and shrub cover
(Bucini et al., 2009), and herbaceous biomass is a co-kriged
interpolated surface representing average forage quantity
(Smit et al., 2011). Terrain ruggedness is represented by the
topographic position index (TPI) of each cell, calculated
from a 90-m digital elevation model (DEM). Ranger stations
house the rangers of each section and represent areas of per-
manent human activity. Distances were extracted from dis-
tance rasters produced using the ‘raster’package (Hijmans,
2019) in R v3.5.1 (R Core Team, 2018). All variables were
re-sampled to a 200 m 9200 m resolution. Prior to model
selection, predictor variables were tested for collinearity
using Pearson’s Correlation; no pairs of predictors showed
Figure 1 The Intensive Protection Zone (IPZ) within the southern Kruger National Park and surrounds.
Animal Conservation  (2019) – ª2019 The Zoological Society of London 3
N. le Roex et al. Poaching creates ecological traps in a protected area

r>0.6 and thus none were considered collinear. Models
were run using the ‘glmulti’package (Calcagno, 2019) in R
v3.5.1 (R Core Team, 2018), and ranked by Akaike Informa-
tion Criterion adjusted for sample size (AICc). Predictor vari-
ables remaining in the top models were tested for non-linear
fit using quadratic terms or log transformation. Similarly, we
used GLMs with a binomial error structure to test the influ-
ence of variables on poached rhino locations. Landscape type
was, however, excluded from the mortality analyses as small
clusters of carcasses greatly distorted landscape significance
and collapsed the models. Detection probability of both live
rhino and carcass occurrence was considered comparable as
both are performed through extensive aerial surveillance in
southern Kruger.
Landscape preference
We used Ivlev’s Selectivity Index (Krebs, 1999) to investi-
gate the degree to which the landscape preferences of black
and white rhino have been influenced by the removal of ani-
mals by poaching. Ivlev’s index represents a measure of the
use of a landscape in relation to its availability. Values >0
indicate that the landscape is used proportionately more rela-
tive to the amount of that landscape available (preference),
and values <0 indicate that a landscape is used proportion-
ately less than it is available (avoidance). For each species,
we compared the impacted population (consisting of live
rhino sightings only) to a representation of the original popu-
lation without the impact of poaching, that is, without ani-
mals removed (combined live rhino and carcass sightings). If
poaching has removed a significant proportion of animals
from a particular landscape and that landscape has not yet
been recolonized to the same degree, the index of selectivity
using only live sightings may not be a historic reflection of
habitat preference. This may be particularly true for black
rhino as they are typically slow to colonize new or empty
habitat (Linklater & Hutcheson, 2010); the rapid removal of
individuals from a localized area may leave that area devoid
of animals for some time, thus appearing to be ‘avoided’.
Conversely, white rhino move large distances and recolonize
new habitats relatively easily (Norman Owen-Smith, 1983),
thus the landscape preferences seen in the impacted popula-
tion may be more similar to those of the original population.
We used the comparisons between the original and impacted
populations to further understand the likely impacts of eco-
logical traps for each species.
Identifying risk areas
Based on the selected models, we generated probabilities for
both live rhino and carcass occurrence in 200 9200 m pix-
els across the study area. Following the framework estab-
lished by Nielsen et al. (2006), we binned the resulting
probabilities into 10 ordinal bins from low (1) to high (10)
and used the combination of these categories to define five
risk states for each rhino species: primary and secondary
ecological trap, primary and secondary safe harbour and
non-critical habitat (Fig. 2). We set cut-off points determined
by the data to define the boundaries between risk classes,
with 90% of live rhino and 80% of carcass occurrence
describing high use and high risk respectively. High use was
further divided in half to represent primary and secondary
occurrence. Ecological traps exhibit the highest risk for
rhino, with high use and high poaching risk; these areas can
be divided into primary and secondary traps in line with the
probability of live rhino occurrence. Safe harbours represent
the least risky areas for rhino, with high probability of use
but low poaching risk; these areas can similarly be divided
into primary and secondary safe harbours in line with live
rhino occurrence. Non-critical habitat represents the area
inhabited by less than 10% of the population.
To delineate areas for the simultaneous management of
both rhino species, we overlaid and collapsed the black and
white rhino risk areas into management categories: (1) threat
(where both species were primary/secondary trap, or one spe-
cies was primary/secondary trap and the other species was
non-critical), (2) haven (where both species were primary/
secondary safe harbour, or one species was primary/sec-
ondary safe harbour and the other species was non-critical),
(3) species contrast (where one species was primary/sec-
ondary trap and the other was primary/secondary safe har-
bour) and (4) non-critical (non-critical habitat for both
species). These categories were defined in terms of manage-
ment actions that might be most effective. For example, both
primary and secondary ecological trap areas may benefit
most from increased anti-poaching actions, while both pri-
mary and secondary safe harbour areas may benefit from
continued surveillance. Likewise, biological management
strategies would distinguish along similar lines, with
removal/rescue strategies most appropriate for the ecological
trap areas, and individual-based rhino monitoring most
appropriate for safe harbour areas. This combined output is
therefore likely to be the most relevant for prioritizing lim-
ited resource distribution and determining anti-poaching and
biological management activities by management authorities.
Results
Live rhino and carcass occurrence
Black rhino
The top-ranked model for live black rhino occurrence
included all predictors except woody cover (Supporting
Information Appendix S1). Black rhino occurrence decreased
with increasing distance to fence for approximately 17 km,
after which probability increased. Black rhino occurrence
increased with distance from main rivers for the first 5 km.
Occurrence and distance to roads displayed a similar rela-
tionship, with probability of black rhino occurrence increas-
ing for the first 3 km, followed by a plateau. Conversely, as
distance from human activity (ranger stations) and terrain
ruggedness increased, the probability of black rhino occur-
rence decreased. Modelled against the Lowveld sour bush-
veld, black rhino showed comparative preference for
mountain bushveld, Acacia thicket and Acacia savanna.
4Animal Conservation  (2019) – ª2019 The Zoological Society of London
Poaching creates ecological traps in a protected area N. le Roex et al.

Black rhino showed significant preference for Combretum
woodland, mixed Combretum/Terminalia woodland, Sabie/
Crocodile thicket and thornveld and significant avoidance of
the Lebombo south plains.
The top-ranked model for the occurrence of black rhino
carcasses included distance to fence, main rivers and roads,
terrain ruggedness and woody cover (Supporting Information
Appendix S1). The probability of carcass occurrence
decreased with increasing distance from the fence and terrain
ruggedness. Carcasses occurred more frequently closer to riv-
ers and with increasing woody cover.
White rhino
The top-ranked model for live white rhino occurrence included
all predictors (Supporting Information Appendix S1). Live
white rhino occurrence increased with distance from fence for
the first 10 km, then decreased steadily. Similarly, white rhino
occurrence increased for approximately 7 km from main rivers,
after which occurrence decreased. White rhino occurrence
increased with distance to human activity and roads for an ini-
tial period (humans: 6 km; roads: 4 km) and then plateaued.
White rhinos decreased with increasing terrain ruggedness, and
occurred most frequently at intermediate herbaceous biomass
levels. Compared to the Lowveld sour bushveld, white rhino
showed significant preference for mountain bushveld, Combre-
tum woodland, mixed Combretum/Terminalia woodland,
Sabie/Crocodile thicket, Acacia savanna and thornveld and sig-
nificant avoidance of Acacia thicket and Lebombo south.
The top-ranked model for the occurrence of white rhino
carcasses also included all predictors (Supporting Information
Appendix S1). White rhino carcass occurrence decreased
with distance from fence and main rivers, but increased with
distance from roads for approximately 7 km, then decreased.
Carcass occurrence increased with herbaceous biomass and
distance to human activity, and decreased with terrain
ruggedness.
Landscape preference
The impacted and original black rhino populations showed a
difference in landscape preference primarily in Acacia thicket
(Fig. 3a). Small differences in Acacia savanna and Lowveld
sour bushveld were seen, but these were not sufficient to
change the overall interpretation of black rhino landscape
preference. For Acacia thicket, however, when poached black
rhino were added to the dataset, the result changed from
avoidance to a neutral preference (Ivlev’s index =0.33 and
0.00 for the impacted and original black rhino populations
respectively).
Comparing the impacted and original white rhino popula-
tions showed the largest difference in preference for Lowveld
sour bushveld, but no differences that were sufficient to
change the overall interpretation of white rhino landscape
preference (Fig. 3b). Both black and white rhino showed
preference for Combretum woodland, mixed Combretum/Ter-
minalia woodland and thornveld, and avoidance of Acacia
savanna, Lebombo south and Lowveld sour bushveld.
Figure 2 Example of risk states classified according to the probability of live rhino and carcass occurrence, from bin 1 (low) to 10 (high). Fig-
ure adapted from Nielsen et al. (2006).
Animal Conservation  (2019) – ª2019 The Zoological Society of London 5
N. le Roex et al. Poaching creates ecological traps in a protected area

Contrasting species preferences were shown for Sabie/Croco-
dile thicket (black rhino: selected, white rhino: avoided),
mountain bushveld (black rhino: avoided, white rhino:
selected) and Acacia thicket (black rhino: neutral, white
rhino: avoided).
Identifying risk areas
Rhino occurrence and mortality probabilities were projected
across the entire IPZ at 200 9200 m resolution. Binning and
overlaying these probabilities resulted in a map of risk areas in
southern Kruger for each species. Black rhino showed a lower
proportion of primary trap area but a higher proportion of sec-
ondary trap compared to white rhino (Table 1). Together, the
ecological trap proportions were similar for both species (black:
37.73%, white: 35.51%). Black rhino showed lower propor-
tions of both primary and secondary safe harbours than white
rhino; together, the safe harbour areas were considerably lower
for black rhino (black rhino: 32.01%, white rhino: 44.74%).
Non-critical habitat was substantially greater for black rhino
(30.27%) compared to white rhino (19.74%).
When combining and collapsing the risk areas for both
rhino species into management categories, 32.48% of the
IPZ were classified as threat area, 31.03% as haven, 21.21%
as species contrast area and 15.28% as non-critical habitat
(Fig. 4a). A high resolution, spatially explicit map is not
shown for security reasons, but has been distributed to man-
agement authorities.
Discussion
The majority of covariates predicting live rhino and carcass
occurrence reflected similar relationships in both analyses,
suggesting that poaching risk is predominantly related to
ease of access in areas where rhinos are likely to be encoun-
tered in southern Kruger. Black and white rhino showed
similar proportions of ecological trap, with fine-scale differ-
ences likely related to species-specific dispersal behaviour.
Management categories identified approximately one third of
southern Kruger as threat area. Management actions that
apply interventions to different categories in accordance with
their classification would be optimal for the protection of
both rhino species.
Predicting live rhino and carcass
occurrence
Distance to roads, terrain ruggedness, herbaceous biomass
(white rhino), distance to fence and distance to rivers
reflected similar relationships with live rhino and carcass
occurrence. Interestingly, woody cover showed no
Figure 3 Landscape preference based on Ivlev’s selectivity index by (a) black rhino and (b) white rhino, for the impacted and original popula-
tions.
Table 1. The area (km
2
) and proportion (%) represented by each
risk area for black and white rhino across the IPZ, Kruger National
Park, predicted at 200 m 9200 m resolution
Risk area
Area (km
2
) Proportion of IPZ (%)
Black
rhino
White
rhino
Black
rhino
White
rhino
Primary trap 838.80 962.16 20.49 23.50
Secondary trap 705.68 491.80 17.24 12.01
Primary safe
harbour
407.28 689.60 9.95 16.84
Secondary safe
harbour
903.04 1142.24 22.06 27.90
Non-critical habitat 1239.08 808.08 30.27 19.74
6Animal Conservation  (2019) – ª2019 The Zoological Society of London
Poaching creates ecological traps in a protected area N. le Roex et al.

relationship with live black rhino but a positive relationship
with poached carcasses, suggesting that poachers actively
select areas with increased cover. The absence of effect on
live black rhinos may be a consequence of scale, with
woody cover influencing within-home range movement
rather than landscape-level distribution. Live black rhino
were less likely to be close to rivers, while carcasses were
more likely to occur in close proximity. This may be a result
of the diel behaviour of black rhino, avoiding riparian areas
during the day (when the aerial census occurred) and travel-
ling to water to drink at night (Lent & Fike, 2003). No
effect of distance to human activity was seen with black
rhino carcasses, but the inverse relationship was seen with
live black rhino occurrence; this may reflect a correlation
with another unknown habitat variable. Whether the relation-
ship between the likelihood of being poached and the likeli-
hood of live rhino occurrence reflects behavioural
understanding by poachers or simply the greater probability
of encountering rhino in particular areas cannot, unfortu-
nately, be further elucidated in this study.
Our analyses are, however, constrained by a number of
factors. The live rhino models utilized census data collected
during the dry season; thus these models represent covariates
predicting habitat selection specifically during the dry season.
This is in contrast to the carcass data which were collected
throughout the year; the carcass models thus reflect covari-
ates influencing mortality at all times of the year. In addi-
tion, no existing water dataset fully captured all permanent
surface water in southern Kruger. Main rivers are the most
consistent source of water at this scale and time of year, but
it is possible that additional permanent water sources
contribute to animal locations. Finally, there are likely to be
potentially crucial variables influencing poacher behaviour
and movement, such as the locations of permanent or tempo-
rary access points and insider information regarding capture
risk in different areas, which could not be incorporated. The
main purpose of this work, however, was to provide the best
predictive power with the data that were available.
Risk areas and dispersal behaviour
Ecological traps occur as a result of distorted cues of habitat
quality, a change in habitat quality despite the presence of
the original cues or a combination of both methods (Robert-
son & Hutto, 2006). We propose that areas that exhibit both
high probability of rhino occurrence and high mortality risk
function as ecological traps, as these areas convey reduced
survival probability as a result of poaching. Furthermore, the
cues of habitat quality may also be distorted as high-quality
habitat with low rhino density likely to exhibit increased
attractiveness for individual dispersal. Differences seen in the
proportions of risk areas between black and white rhino may
be partly a consequence of species-specific dispersal beha-
viour rather than ecological specificity or poaching suscepti-
bility. Black rhino are known to be poor dispersers and may
take long periods of time to colonize or recolonize vacant
habitat (Linklater & Hutcheson, 2010). Thus following the
large-scale removal of animals within a particular area, it
may take some time (or incentive) for black rhino to move
back into the area. White rhino, however, often move large
distances into unoccupied or low-density areas (Norman
Owen-Smith, 1983).
Figure 4 (a) Proportions of different management categories for both rhino species combined across the IPZ, Kruger National Park. (b) Flow
chart depicting management categories and circumstances or actions that would convert one category to another.
Animal Conservation  (2019) – ª2019 The Zoological Society of London 7
N. le Roex et al. Poaching creates ecological traps in a protected area

The lower numbers of black rhino combined with poor
recolonization of empty habitat would result in primary traps
converting to secondary traps (as some black rhinos are
removed by poaching) and then to non-critical habitat (as
most or all are removed by poaching) far more readily than
for white rhino. This is reflected in the comparative statistics
for these risk areas: primary trap proportions are lower for
black rhino than white rhino, secondary traps are higher and
non-critical areas are higher still. If poaching remains local-
ized for a substantial time period, however, the poor recolo-
nization behaviour of black rhino may, in fact, confer a
‘protective’effect of sorts; once poaching has extirpated a
local population, there are simply no remaining black rhino
to poach within that area. White rhino, however, would not
enjoy the same protection. Continued movement and disper-
sal into vacant habitat would perpetuate the cycle and eco-
logical trap areas would continue to function as population
sinks (Pulliam & Danielson, 1991) for white rhino. This
behaviour is, in fact, actively utilized by park authorities in
other reserves for white rhino population management
(Owen-Smith & Shrader, 2002). If this is the case, black
rhino may experience less severe consequences from ecologi-
cal traps than white rhino.
The ‘avoidance’of Acacia thicket by the impacted black
rhino population and ‘neutral preference’of the original
population support the local extirpation and poor recolo-
nization hypothesis. The Acacia thicket landscape represents
the smallest physical area of the nine IPZ landscape types,
and hosted the highest number of poached black rhino car-
casses per unit area over the study period (data not
shown). The extent of the selectivity change is also likely
to be under-representative, as each carcass was included
only once in the original population dataset, while there
would likely be multiple sightings of individual live rhino
over the 5-year period. The population size of black rhino
in southern Kruger is most likely below the level at which
density-dependent social pressures would increase dispersal;
this may have resulted in this landscape remaining vacant
even if it was a preferred type. Alternatively, the high
mortality per unit area may be perceived by black rhino as
dangerous, reducing the quality of that landscape and
resulting in active avoidance. While bolder or more aggres-
sive animals typically need stronger cues to avoid danger
(Robertson et al, 2013), the high mortality per unit area in
the Acacia thicket may be sufficient to be perceived as
such by black rhino.
The probability of extinction of a population as a result of
an ecological trap increases with the severity of the trap and
the fraction of the population that is trapped (Robertson,
Rehage, & Sih, 2013). The severity of a trap for a particular
species is often determined by the type of population sink
that the ecological trap represents (Robertson, Rehage, &
Sih, 2013). A severe trap is more attractive than other habi-
tats and draws animals in, thus perpetuating the cycle, while
an equal-preference trap exhibits the same draw as other
areas, and typically has less severe consequences for the
population (Robertson et al., 2006). Identifying the type of
trap experienced by black and white rhino is therefore
relevant to the long-term prognosis for each species. Suscep-
tibility to traps can also be related to specific age and sex
classes (Robertson et al., 2013); for example, young dispers-
ing males may be drawn into a trap area due to a lower den-
sity of dominant bulls (severe trap), but other classes may
exhibit a neutral preference for the trap area (equal-prefer-
ence trap). The results of this study suggest that ecological
traps within Kruger may act more as equal-preference traps
for black rhino (particularly while the population remains at
low density) and severe traps for white rhino; future work
identifying whether these designations are specific to particu-
lar age/sex classes would be beneficial for individual species
predictions and management.
Management implications
Collapsed management categories for black and white rhino
resulted in 32.58% threat, 31.03% haven, 21.21% species
contrast and 15.28% non-critical habitat. Grouping both spe-
cies’primary and secondary ecological traps into one ‘threat’
category, and both primary and secondary safe harbours into
one ‘haven’category is a relatively crude approach, but one
that enabled the simple classification of just four categories
that are likely to benefit from different management actions.
We surmise that this type of output is the most practical and
likely to be implemented by conservation authorities. Man-
agement interventions that (1) convert threat areas and non-
critical habitat (where possible) to havens, and (2) maintain
current havens, would likely make the best contribution to
rhino protection in southern Kruger with limited capacity
and resources (Fig. 4b). These may include heavy anti-
poaching interventions in threat areas, increased rhino moni-
toring and biological management in havens and possibly
animal translocations back into non-critical habitat under
specific circumstances. Noteworthy is the absence of havens
around the Kruger boundary; the area adjacent to the fence
is almost entirely threat, species contrast or non-critical area.
This suggests that poachers are still predominantly operating
in easily accessible areas and that increased focus on access
points, fence improvements and boundary protection is still
likely to be highly effective.
Conclusion
Black and white rhinos are under severe poaching pressure
in South Africa, which threatens them at the local population
and species level. Until the illegal demand for rhino horn
decreases, active protection in the form of anti-poaching and
optimal biological management to recover populations will
continue to be crucial for species survival. We classified the
IPZ, a high-priority area in southern Kruger, into manage-
ment categories that reflect the actions most likely to be
effective for rhino protection in those areas. The IPZ com-
prised 32.58% threat, 31.03% haven, 21.21% species contrast
and 15.58% non-critical habitat. For protected areas, the size
of Kruger, prioritized distribution of resources and man-
power are needed, as it is impossible to uniformly cover
such an expanse at the intensity level that is currently
8Animal Conservation  (2019) – ª2019 The Zoological Society of London
Poaching creates ecological traps in a protected area N. le Roex et al.

required. Management actions associated with converting
threat areas to havens and maintaining the current havens
would provide optimal protection for both rhino species
simultaneously with limited resources. This work illustrates
how the conservation of multiple species or taxa within a
system can be simultaneously prioritized in vast areas where
resources and/or capacity are insufficient to undertake spe-
cies-specific approaches.
Acknowledgements
We thank Chenay Simms and Izak Smit for GIS guidance,
Cathy Greaver for coordinating the annual rhino censuses
and the census observers and pilots for their crucial role in
data collection. We are grateful to the SANParks pilots, ran-
gers and ECI team for the identification of carcasses and car-
cass data collection under difficult circumstances. Special
thanks also go to Ross Pitman and Guy Balme for statistical
and conceptual discussions and support.
References
Abrams, P.A., Ruokolainen, L., Shuter, B.J. & McCann, K.S.
(2012). Harvesting creates ecological traps: Consequences of
invisible mortality risks in predator-prey metacommunities.
Ecology 93, 281–293.
Bucini, G., Saatchi, S., Hanan, N., Boone, R.B. & Smit, I.
(2009). Woody cover and heterogeneity in the savannas of
the Kruger National Park, South Africa. IEEE Int. Geosci.
Remote Sens. Symp. 4, 334–337.
Calcagno, V. (2019). glmulti: model selection and multimodel
inference made easy. R package version 1.0.7.1. https://
CRAN.R-project.org/package=glmulti
Delibes, M., Gaona, P. & Ferreras, P. (2001). Effects of an
attractive sink leading into maladaptive habitat selection.
Am. Nat. 158, 277–285.
Emslie, R. (2012). Ceratotherium simum. IUCN Red List
Threat. Species e.T4185A16980466.
Emslie, R., Milliken, T., Talukdar, B., Ellis, S., Adcock, K. &
Knight, M. (2016). African and Asian rhinoceroses –status
, conservation and trade. A Rep. from IUCN Species
Surviv. Comm. (IUCN SSC) African Asian Rhino Spec.
Groups TRAFFIC to CITES Secr. Purs. to Resolut. Conf.
9.14 (Rev. CoP15). Vol. 5.
Falcucci, A., Ciucci, P., Maiorano, L., Gentile, L. & Boitani,
L. (2009). Assessing habitat quality for conservation using
an integrated occurrence-mortality model. J. Appl. Ecol. 46,
600–609.
Ferreira, S.M., Greaver, C.C. & Knight, M.H. (2011).
Assessing the population performance of the black
rhinoceros in Kruger National Park. S. Afr. J. Wildl. Res.
41, 192–204.
Ferreira, S.M., Greaver, C., Knight, G.A., Knight, M.H., Smit,
I.P.J. & Pienaar, D. (2015). Disruption of rhino demography
by poachers may lead to population declines in Kruger
National Park, South Africa. PLoS ONE 10,1–18.
Ferreira, S., Greaver, C., Nhleko, Z. & Simms, C. (2018).
Realization of poaching effects on rhinoceroses in Kruger
National Park. S. Afr. J. Wildl. Res. 48, 013001.
Fletcher, R.J., Orrock, J.L. & Robertson, B.A. (2012). How
the type of anthropogenic change alters the consequences of
ecological traps. Proc. R. Soc. B Biol. Sci. 279, 2546–2552.
Fritz, H., Garine-wichatitsky, M.D.E. & Letessier, G. (1996).
Habitat use by sympatric wild and domestic herbivores in
an African savanna woodland: the influence of cattle spatial
behaviour. J. Appl. Ecol. 33, 589–598.
Gertenbach, W.P.D. (1983). Landscapes of the Kruger
National Park. Koedoe 26,9–121.
Hijmans, R.J. (2019). raster: Geographic Data Analysis and
Modeling. R package version 2.8-19. https://CRAN.R-projec
t.org/package=raster
Johnson, C.J. & Gillingham, M.P. (2005). An evaluation of
mapped species distribution models used for conservation
planning. Environ. Conserv. 32, 117–128.
Krebs, C.J. (1999). Ecological methodology. 2nd edn. San
Francisco: Benjamin Cummings.
Kruger, J.M., Reilly, B.K. & Whyte, I.J. (2008). Application
of distance sampling to estimate population densities of
large herbivores in Kruger National Park. Wildl. Res. 35,
371–376.
Lamb, C.T., Mowat, G., McLellan, B.N., Nielsen, S.E. &
Boutin, S. (2017). Forbidden fruit: human settlement and
abundant fruit create an ecological trap for an apex
omnivore. J. Anim. Ecol. 86,55–65.
Lent, P.C. & Fike, B. (2003). Home ranges, movements and
spatial relationships in an expanding population of black
rhinoceros in the Great Fish River Reserve, South Africa. S.
Afr. J. Wildl. Res. 33, 109–118.
Linklater, W.L. & Hutcheson, I.R. (2010). Black rhinoceros
are slow to colonize a harvested neighbour’s range. S. Afr.
J. Wildl. Res. 40,58–63.
Morris, D.W. (2003). Toward an ecological synthesis: a case
for habitat selection. Oecologia 136,1–13.
Muposhi, V.K., Gandiwa, E., Chemura, A., Bartels, P.,
Makuza, S.M. & Madiri, T.H. (2016). Habitat heterogeneity
variably influences habitat selection by wild herbivores in a
semi-arid tropical savanna ecosystem. PLoS ONE 11,
e0163084.
Nielsen, S.E., Stenhouse, G.B. & Boyce, M.S. (2006). A
habitat-based framework for grizzly bear conservation in
Alberta. Biol. Conserv. 130, 217–229.
Northrup, J.M., Stenhouse, G.B. & Boyce, M.S. (2012).
Agricultural lands as ecological traps for grizzly bears.
Anim. Conserv. 15, 369–377.
Owen-Smith, Norman. (1983). Dispersal and the dynamics of
large herbivores in enclosed areas: implications for
management. In Manag. large Mamm. African Conserv.
areas: 127–143Owen-Smith, N. (Ed.). Pretoria: Haum
Educational.
Owen-Smith, N. & Shrader, A.M. (2002). The role of
companionship in the dispersal of white rhinoceroses
Animal Conservation  (2019) – ª2019 The Zoological Society of London 9
N. le Roex et al. Poaching creates ecological traps in a protected area

(Ceratotherium simum). Behav. Ecol. Sociobiol. 52, 255–
261.
Penteriani, V., Delgado, M.D.M., Krofel, M., Jerina, K.,
Ordiz, A., Dalerum, F., Zarzo-Arias, A. & Bombieri, G.
(2018). Evolutionary and ecological traps for brown bears
Ursus arctos in human-modified landscapes. Mamm. Rev.
48, 180–193.
Pitman, R.T., Swanepoel, L.H., Hunter, L., Slotow, R. &
Balme, G.A. (2015). The importance of refugia, ecological
traps and scale for large carnivore management. Biodivers.
Conserv. 24, 1975–1987.
Pulliam, H.R. & Danielson, B.J. (1991). Sources, sinks, and
habitat selection: a landscape perspective on population
dynamics. Am. Nat. 137, S50–S66.
Robertson, B.A. & Hutto, R.L. (2006). A framework for
understanding ecological traps and an evaluation of existing
evidence. Ecol. Soc. Am. 87, 1075–1085.
Robertson, B.A., Rehage, J.S. & Sih, A. (2013). Ecological
novelty and the emergence of evolutionary traps. Trends
Ecol. Evol. 28, 552–560.
Roever, C.L., van Aarde, R.J. & Chase, M.J. (2013).
Incorporating mortality into habitat selection to identify
secure and risky habitats for savannah elephants. Biol.
Conserv. 164,98–106.
Romero-Mu~
noz, A., Torres, R., Noss, A.J., Giordano, A.J.,
Quiroga, V., Thompson, J.J., Baumann, M., Altrichter, M.,
McBride, R., Velilla, M., Arispe, R. & Kuemmerle, T.
(2018). Habitat loss and overhunting synergistically drive
the extirpation of jaguars from the Gran Chaco. Divers.
Distrib. 25, 176–190.
R Core Team (2018). R: A language and environment for
statistical computing. Vienna: R Foundation for Statistical
Computing. https://www.R-project.org/.
S
anchez-Mercado, A., Ferrer-Paris, J.R., Garc
ıa-Rangel, S.,
Yerena, E., Robertson, B.A. & Rodr

ıguez-Clark, K.M.
(2014). Combining threat and occurrence models to predict
potential ecological traps for Andean bears in the Cordillera
de M
erida, Venezuela. Anim. Conserv. 17, 388–398.
Schlaepfer, M.A., Runge, M.C. & Sherman, P.W. (2002).
Ecological and evolutionary traps. Trends Ecol. Evol. 17,
2995–3013.
Smit, I.P.J. (2011). Resources driving landscape-scale
distribution patterns of grazers in an African savanna.
Ecography 34,67–74.
Valeix, M., Loveridge, A.J., Chamaill
e-Jammes, S., Davidson,
Z., Murindagomo, F., Fritz, H. & Macdonald, D.W. (2009).
Behavioral adjustments of African herbivores to predation
risk by lions: spatiotemporal variations influence habitat use.
Ecology 90,23–30.
Woodroffe, R. & Ginsberg, J.R. (2008). Edge effects and the
extinction of populations inside protected areas. Science
280, 2126–2128.
Supporting information
Additional supporting information may be found online in
the Supporting Information section at the end of the article.
Appendix S1. Top-supported resource selection functions for
live rhino and carcass occurrence.
10 Animal Conservation  (2019) – ª2019 The Zoological Society of London
Poaching creates ecological traps in a protected area N. le Roex et al.