Introduction
The leopard Panthera pardus is categorized as Vulnerable on the IUCN Red List. Its risk of extinction is particularly high in fragmented landscapes because of low densities, large spatial requirements and potential for conflict with humans (Nowell & Jackson, Reference Nowell and Jackson1996; Balme et al., Reference Balme, Slotow and Hunter2010). Leopard populations in Africa are increasingly threatened by increasing anthropogenic pressures, leading to concern for the conservation of the species and calls for reliable population estimates to inform conservation management (Jacobson et al., Reference Jacobson, Gerngross, Lemeris, Schoonover, Anco and Breitenmoser-Würsten2016). In the absence of robust population estimates, management decisions often rely on expert opinion rather than being based on sound evidence, making it difficult to identify areas of concern, prioritize conservation investments, or evaluate the effectiveness of interventions (Gray & Prum, Reference Gray and Prum2012; Balme et al., Reference Balme, Lindsey, Swanepoel and Hunter2014).
Density estimation, such as with capture–recapture modelling, has become a key process in wildlife ecology, conservation and management (Gray & Prum, Reference Gray and Prum2012). Initially, capture–recapture techniques estimated abundance rather than density, and relied on estimating the survey's effective sampled area to obtain the latter. However, no theoretical basis exists for this process, and the reliability of this approach is therefore questionable (Efford, Reference Efford2004; Borchers & Efford, Reference Borchers and Efford2008; Royle et al., Reference Royle, Nichols and Karanth2009). Recently developed methodologies, known as spatially explicit capture–recapture, overcome these issues by estimating density directly as an explicit parameter (Efford, Reference Efford2004; Royle et al., Reference Royle, Nichols and Karanth2009).
Since their first application to tiger populations in India (Karanth, Reference Karanth1995), capture–recapture techniques have been employed to obtain density estimates of most large carnivores, including leopards in several African countries, such as South Africa (Balme et al., Reference Balme, Hunter and Slotow2009; Chase Grey et al., Reference Chase Grey, Kent and Hill2013; Swanepoel et al., Reference Swanepoel, Somers and Dalerm2015), Gabon (Henschel et al., Reference Henschel, Hunter, Coad, Abernethy and Muhlenberg2011) and Namibia (Stein et al., Reference Stein, Andreas and Aschernborn2011). Nevertheless, there is still a paucity of such data across much of the continent, precluding effective conservation management (Balme et al., Reference Balme, Lindsey, Swanepoel and Hunter2014). In Mozambique armed conflicts during much of the latter half of the 20th century have contributed to significant declines in wildlife populations (Hatton et al., Reference Hatton, Couto and Oglethorpe2001) and have hindered conservation, and there has been little research conducted on the status, distribution or ecology of the leopard.
Leopards can be legally hunted for trophies in several locations in Mozambique, with the current annual quota set at 120 permits (CITES, 2007). This quota is based on an estimation of the overall abundance of leopards in Mozambique by Martin & de Meulenaer (Reference Martin and De Meulenaer1988), who employed a predictive model, estimating a population of 37,542 leopards in the country based on a mean density of 0.10/km2 (10 leopards/100 km2). This estimate has been widely criticized because the model omitted important factors such as anthropogenic mortality and prey availability and assumed that leopards occur at maximum potential densities in all available habitats (Jackson, Reference Jackson1989; Balme et al., Reference Balme, Slotow and Hunter2010). Nevertheless, it formed the basis of the most recent increase in the trophy hunting quota, mainly because alternative estimates of population densities are not available (CITES, 2007). A more accurate assessment of leopard populations in Mozambique is needed to determine the reliability of the methods currently employed to set the hunting quota, and to ensure that future changes are based on robust data.
Here we use closed-population spatially explicit capture–recapture methodology to estimate the density of leopards in Xonghile Game Reserve in southern Mozambique. The aim of the study was to obtain the first empirical density estimate for a leopard population in southern Mozambique and present information to guide management and provide a baseline for the assessment of conservation interventions, and to explore the implications of our findings for trophy hunting.
Study area
Xonghile Game Reserve (Fig. 1) is a 450 km2, unfenced, legally protected area in southern Mozambique. Its northern border is c. 13 km south of Limpopo National Park, the country's largest national park. It borders South Africa's Kruger National Park to the west and unprotected land to the north, east and south, and is part of the Greater Limpopo Transfrontier Conservation Area, a transboundary initiative linking protected parks and reserves in Mozambique, South Africa and Zimbabwe via non-protected areas. The predominant habitat in the Reserve consists of sand plains (sandveld) characterized by low woodlands and thickets on deep sandy soils, and short-grass pans (seasonally flooded depressions). Although populations of large mammals in the region were severely depleted during the 1964–1992 armed conflicts (Hanks, Reference Hanks, Entwistle and Dunstone2000), the progressive removal of fencing along the border of Kruger National Park since 2005 has provided opportunities for wildlife to move into the area. No human population permanently resides in the Reserve, with the main anthropogenic impacts coming from relatively low levels of poaching for bushmeat, anti-poaching efforts, and low levels of tourism (LA & KTE, unpubl. data). Trophy hunting does not currently take place in the Reserve.
Fig. 1 Xonghile Game Reserve, with camera locations and intensive trapping area within a 10 km buffer zone, as required by the spatially explicit capture–recapture models. Inset map: the Reserve in the context of the wider Greater Limpopo Transfrontier conservation initiative, comprising protected (grey) and non-protected areas (dotted).
Methods
Camera trapping
Twenty-nine digital motion-activated cameras of various models (HC500, Reconyx, Holmen, USA; Tiny W-2, Spy Point, Victoriaville, Canada; Trophy Cam, Bushnell, Overland Park, USA) were deployed at 26 stations over c. 300 km2 in the Xonghile Game Reserve (Fig. 1) during 24 August–23 November 2012. Twenty-three stations were equipped with a single camera and three stations with two cameras each.
The majority of stations (23) were located 0.5–3 km apart, ensuring that multiple cameras were likely to be present in an individual leopard's home range. Three stations were placed 5–6 km from the nearest station. There was therefore a possibility that an individual's home range did not contain a station, but this is unlikely given the low prey densities in the study area (LA & KTE, unpubl. data), and spatially explicit capture–recapture models allow for the presence of such gaps in the trap array when estimating density (Borchers & Efford, Reference Borchers and Efford2008). Cameras were set on trees along roads and game trails at a height of 35 cm. The survey duration was 92 days, which was considered adequate for assuming demographic closure and is consistent with previous studies of large felids (Karanth, Reference Karanth1995; Alexander et al., Reference Alexander, Gopalaswamy, Shi and Riordan2015; Boron et al., Reference Boron, Tzanopoulos, Gallo, Barragan, Jaimes-Rodriguez, Schaller and Payán2016).
Density estimation
Density was modelled in a spatially explicit capture–recapture framework, using the package secr (Efford, Reference Efford2015) in R v. 3.2.3 (R Development Core Team, 2015). A maximum-likelihood framework was chosen over a Bayesian one to make results comparable with other studies (Noss et al., Reference Noss, Polisar, Maffei, García and Silver2013; Tobler & Powell, Reference Tobler and Powell2013) and because computation times are shorter (Efford, Reference Efford2015).
Leopards were identified from their pelage patterns and sexed by visual inspection of external genitalia. We chose the flank with the greater number of captures (left) for identification of individual leopards. Individual spatial capture and trap effort histories were developed following recommended procedures (Efford, Reference Efford2015), with each day (24 hours) treated as a separate sampling occasion (Goldberg et al., Reference Goldberg, Tempa, Norbu, Hebblewhite, Mills, Wangchuk and Lukacs2015). Information on varying effort from different camera stations (the number of days each camera was active) was included to improve estimates of detection probability. We increased the buffer width around the trapping grid until density estimates stabilized, ensuring that no individual outside the buffer area could be captured, and fitted a half-normal detection function to the distance between the centre of the home range and the camera station. This is the most commonly used function in spatial capture–recapture analyses (Efford, Reference Efford2004; Boron et al., Reference Boron, Tzanopoulos, Gallo, Barragan, Jaimes-Rodriguez, Schaller and Payán2016) and describes the probability of capture (P) of an individual i at a trap j as a function of distance d from the activity centre of the individual to the trap, as follows: Pij = g0 exp (-dij 2/(2σ 2), where g0 is the probability of capture at the exact centre of the home range, and σ is a spatial parameter related to home range size (Efford, Reference Efford2004). We fitted a Bernoulli or binomial encounter model to the data because this is most relevant to camera trapping studies; under this model an individual can be recorded at different camera stations during one sampling occasion, but only once at each station (Royle et al., Reference Royle, Nichols and Karanth2009; Noss et al., Reference Noss, Polisar, Maffei, García and Silver2013).
Given that male and female leopards have different ranging patterns (Bailey, Reference Bailey1993; Kittle et al., Reference Kittle, Watson and Fernando2017), with a potential impact on capture parameters, sex was modelled as a covariate (Sollmann et al., Reference Sollmann, Furtado, Gardner, Hofer, Jacomo, Torres and Silveira2011; Tobler & Powell, Reference Tobler and Powell2013; Goldberg et al., Reference Goldberg, Tempa, Norbu, Hebblewhite, Mills, Wangchuk and Lukacs2015). This was achieved by fitting a hybrid mixture model, which accommodates individuals of unknown sex (Efford, Reference Efford2015). The impact of sex on both parameters g0 and σ was tested through the comparison of four alternative models using the Akaike Information Criterion, adjusted for small sample size (AICc; Burnham & Anderson, Reference Burnham and Anderson2002): secr.0 (null model), secr.sex.g0 (g0 varies between males and females), secr.sex.σ (σ varies between males and females), and secr.sex (both g0 and σ vary between males and females; Efford, Reference Efford2015; Boron et al., Reference Boron, Tzanopoulos, Gallo, Barragan, Jaimes-Rodriguez, Schaller and Payán2016).
Results
Sampling effort and capture success
A total sampling effort of 1,021 trap nights by 26 stations (mean trap nights per camera = 39.3) yielded 57 leopard capture events. Of these, 31 (54%) were used to identify nine individual leopards (five males, two females, two unsexed); the remaining 26 events (46%) were not suitable for identification (because of poor image quality and/or the wrong flank being captured) and were therefore discarded. Capture frequencies were 9, 3, 2, 2 and 1 for the five males; 6 and 4 for the two females; and 3 and 1 for the two unsexed individuals.
Density estimation
The best model (AICc = 360.36) did not allow g0 or σ to vary with sex (secr.0), and received significantly more support than the next best alternative (secr.sex.σ, ΔAICc = 10.47; Table 1).
Table 1 Model selection parameters for spatially explicit capture–recapture models in R package secr.
1Akaike Information Criterion, adjusted for small sample size.
2Difference from best ranking (lowest AIC) model.
3Number of model parameters.
4Spatial parameter related to home range size.
5Probability of capture at the home range centre.
The leopard density estimate of the best-fitted model (secr.0) was 2.60 ± SE 0.96 adults/100 km2. Capture probability at the centre of the home range (g0) was estimated to be 0.043 ± SE 0.013, and the spatial parameter (σ) to be 1,936 ± SE 279 m (Table 2). Buffer width stabilized at 10,000 m, as reported by similar leopard density studies (Gray & Prum, Reference Gray and Prum2012; Borah et al., Reference Borah, Sharma, Das, Rabha, Kakati and Basumatary2013).
Table 2 Parameters and density estimated by the best model (secr.0).
1g0, probability of capture at the home range centre; σ, spatial parameter related to home range size.
Discussion
Leopard density
Our study provides a baseline leopard density estimate for a relatively well-protected area and the first empirical estimate for a leopard population in Mozambique. Using spatially explicit capture–recapture models we estimated leopard density in the Xonghile Game Reserve to be 2.60 ± SE 0.96 adults/100 km2. Although this is low compared to studies elsewhere in sub-Saharan Africa, it is higher than estimates from other protected areas in southern Africa: 0.60 leopards/100 km2 in the dry savannahs of the Kalahari Gemsbok National Park (South Africa; Bothma & Le Riche, Reference Bothma, du and Le Riche1984), 0.62/100 km2 in the savannah/woodland Cederberg Wilderness Area (South Africa; Martins & Harris, Reference Martins and Harris2013) and 1.50/100 km2 in the savannah habitat of the Kaudom Game Reserve, Namibia (Stander et al., Reference Stander, Haden and Kaqece1997).
In South Africa's Kruger National Park, contiguous with the Reserve's western border, high leopard densities of 30.3/100 km2 were reported for the Sabie riverine area (Bailey, Reference Bailey1993), and 12.7/100 km2 in the N'wantesi concession (Maputla et al., Reference Maputla, Chimimba and Ferreira2013). We believe the observed differences are probably a reflection of different habitats, and consequently prey availability, between sites. Whereas density estimates from Kruger National Park came from highly productive riverine forests (Bailey, Reference Bailey1993) and savannah woodlands (Maputla et al., Reference Maputla, Chimimba and Ferreira2013), Xonghile Game Reserve predominantly comprises nutrient-poor, lower-productivity sandveld, which sustains lower animal densities (Redfern et al., Reference Redfern, Grant, Biggs and Getz2003; Scholes et al., Reference Scholes, Bond and Eckhardt2003). The relatively high level of protection in the Reserve (LA & KTE, unpubl. data) suggests that the low prey densities are not a result of human hunting activities.
Nonetheless, the Reserve could be acting as a population sink for leopards dispersing from Kruger National Park, through anthropogenic mortalities occurring in the adjacent non-protected areas. Estimates for anthropogenic leopard mortalities in the area are not available, but it is possible that individuals venturing into the non-protected areas adjoining the Reserve could be suffering relatively high anthropogenic mortality rates, which could lower population densities and attract leopards from surrounding areas (e.g. Kruger National Park). This vacuum effect has been documented for large carnivores and leopards in particular, and it may cause edge effects that affect the interior of even large protected areas (Loveridge et al., Reference Loveridge, Searle, Murindagomo and Macdonald2007). Longer term camera trapping or tracking using collars equipped with global positioning system units, combined with social surveys targeting the communities outside the Reserve, are necessary to ascertain whether this affects leopards in and around Xonghile Game Reserve.
Methodological considerations and sex-specific parameters
The majority of our stations had only one camera trap, rather than the recommended two-camera set-up (one for each flank; Karanth, Reference Karanth1995), allowing us to survey a larger area and thus increase the number of captured individuals, with limited resources. Although we believe that this was the best approach in our case, the trade-offs between surveying a larger area and obtaining higher identification rates should be considered on a case-by-case basis.
Although males commonly occupy territories overlapping with those of 2–4 females (Bailey, Reference Bailey1993), more males (n = 5) than females (n = 2) were captured during our study (the sex of two individuals could not be determined). Maputla et al. (Reference Maputla, Chimimba and Ferreira2013) also recorded a male-bias in capture rates and cited several potential reasons for this, including heterogeneity in behaviour between sexes in the vicinity of the trap, and in tendencies to use specific trap locations, such as roads (Krebs, Reference Krebs1999).
The model with the highest support was that in which sex did not influence the detection (g0) or scale (σ) parameters. However, rather than indicating the absence of widely described sex-dependent heterogeneity in behaviour and ranging patterns (Bailey, Reference Bailey1993; Kittle et al., Reference Kittle, Watson and Fernando2017), we believe that the relatively small sample size did not provide enough data to facilitate the inference and modelling of sex-specific differences in detectability and ranging patterns.
Implications for conservation policy in Mozambique
Trophy hunting has the potential to benefit the conservation of large carnivores (Lindsey et al., Reference Lindsey, Roulet and Romanach2007; Loveridge et al., Reference Loveridge, Searle, Murindagomo and Macdonald2007), and it is estimated that each leopard hunted in Mozambique could contribute c. USD 24,000 to the local and national economy (Jorge et al., Reference Jorge, Vanak, Thaker, Begg and Slotow2013). However, if hunting is poorly managed or adds to other sources of anthropogenic mortality, it can reduce numbers to such an extent that a population is no longer viable in the long term. Demonstrating that hunting practices, including quotas, are biologically sustainable is therefore essential for trophy hunting to be an effective tool in the management and conservation of large African carnivores (Swanepoel et al., Reference Swanepoel, Lindsey, Somers, Van Hoven and Dalerum2014; Braczkowski et al., Reference Braczkowski, Balme, Dickman, Macdonald, Fattebert and Dickerson2015).
Our results lead us to question the reliability of the estimates employed to set quotas for hunting leopards in Mozambique. The study by Martin & de Meulenaer (Reference Martin and De Meulenaer1988), quoted as the primary justification for a recent increase of the trophy export quota in Mozambique (from 60 to 120 individuals per annum; CITES, 2007), states that up to 80% of the country supports leopard densities of 3–10 individuals/100 km2. It also suggests that only 3% of the country's total land area should have leopard population densities lower than that found in our study. However, our estimate of 2.60 leopards/100 km2 in Xonghile Game Reserve, one of the better protected areas in the country, suggests that it is unlikely that many areas in Mozambique support the densities cited in the application for a revision of the hunting quota. Although some landscapes will have higher primary productivity levels than the Reserve, it is likely that high levels of anthropogenic disturbances in large parts of the country (Hatton et al., Reference Hatton, Couto and Oglethorpe2001) would more than counteract this. Thus, although we appreciate that trophy hunting has not taken place in the Reserve for nearly 10 years (LA & KTE, unpubl. data) and we acknowledge the limitations of our study in terms of the number of individuals encountered relative to the overall range and total size of the population, we believe it is unlikely that leopard densities as high as those cited in the application for a quota increase are common in areas where hunting currently takes place.
We therefore recommend further assessments of leopard population status and densities across different habitats and land-use types across the country, in both hunting and protected areas. This would be an important step towards the development of a sustainable and empirical quota allocation system, similar to that currently being developed for South Africa, which includes hunting regulations based on leopards’ age, adaptive management strategies, and dynamic, evidence-based quota systems (Department of Environmental Affairs, 2017). Camera trapping surveys are a rapid method for obtaining robust estimates of leopard numbers at a moderate cost (Balme et al., Reference Balme, Hunter and Slotow2009) and, if followed by effective management interventions, could play an important role in the species’ recovery and conservation in many post-conflict landscapes across the country.
The conservation challenges we have identified are not exclusive to Mozambique, with Tanzania and Namibia also employing the density estimates of Martin & de Meulenaer (Reference Martin and De Meulenaer1988) for justifying modifications, approved by CITES (CITES, 2002, 2004), of quotas for hunting leopards. Our results reinforce the need for caution when setting hunting quotas for leopards, and the importance of reliable population estimates across the species' range. We recommend similar research be carried out in other regions where such estimates are used to set harvest quotas, to support a shift towards evidence-based guidance of management and policy.
Acknowledgements
We thank the Director of National Conservation Areas Mozambique for granting KE and LA research permits (005-2011/003-2012), the shareholders of Xonghile Game Reserve for allowing access, the Centre for Wildlife Management, University of Pretoria, and Wilderness Trust for financial support, and Valeria Boron and Charlotte E. Searle for their help with the analyses.
Author contributions
Data collection: LA and KE; analyses: PS; writing: all authors.
Conflicts of interest
None.
Ethical standards
All research carried out complied with the Oryx Code of Conduct.
