Carnivore species across the globe are in decline (Ripple et al., Reference Ripple, Estes, Beschta, Wilmers, Ritchie and Hebblewhite2014). With an increasing human population, the biological traits of carnivores put them at high risk of extinction (Cardillo et al., Reference Cardillo, Purvis, Sechrest, Gittleman, Bielby and Mace2004). Although some smaller species are plentiful and adaptable, many larger carnivores are vulnerable to extinction as a result of their narrow geographical ranges, small and isolated populations, low genetic diversity, specialized niche requirements, large home ranges, and propensity to compete with humans for the apex of shared food webs (Gittleman et al., Reference Gittleman, Funk, Macdonald, Wayne, Gittleman, Funk, Macdonald and Wayne2001; Sillero-Zubiri & Laurenson, Reference Sillero-Zubiri, Laurenson, Gittleman, Funk, Macdonald and Wayne2001). The key factors affecting carnivore survivorship are often interrelated; thus a connected and multifaceted approach is recommended in carnivore conservation (Winterbach et al., Reference Winterbach, Winterbach, Somers and Hayward2013).
Determining which factors influence species occupancy is vital for conservation, but this is often challenging for carnivores because of their secretive behaviour, nocturnal habits, low densities and broad spatial requirements that often extend beyond physical, administrative and political boundaries (Balme et al., Reference Balme, Slotow and Hunter2010; Bischof et al., Reference Bischof, Brøseth and Gimenez2016; Chundawat et al., Reference Chundawat, Sharma, Gogate, Malik and Vanak2016). Camera traps are an increasingly popular tool in carnivore research and management because they overcome many of these obstacles (McCallum, Reference McCallum2013). They are affordable, can be deployed over large areas, and are able to collect continuous data non-invasively on even the most cryptic species. Accordingly, camera traps have the capacity to monitor species and inform conservation strategies at both national and global scales (Ahumada et al., Reference Ahumada, O'Brien, Mugerwa, Hurtado, Rovero and Zimmermann2016).
Despite this potential, few studies have examined carnivore occupancy, and the factors influencing occupancy, beyond the scale of single study sites (but see Karanth et al., Reference Karanth, Gopalaswamy, Kumar, Vaidyanathan, Nichols and MacKenzie2011; Pitman et al., Reference Pitman, Fattebert, Williams, Williams, Hill and Hunter2017a; Miller et al., Reference Miller, Pitman, Mann, Fuller and Balme2018). This localized approach to monitoring generates a limited, and potentially biased, view of species ecology and behaviour. Recent collaborations have shown, however, that robust broad-scale species assessments are possible if researchers combine datasets from several single-site camera-trap surveys (e.g. Linkie et al., Reference Linkie, Dinata, Nugroho and Haidir2007; Miller & Grant, Reference Miller and Grant2015; Tan et al., Reference Tan, Rocha, Clements, Brenes-Mora, Hedges and Kawanishi2017). In addition, although many camera-trap surveys are established to monitor a single species, they also collect data on non-target species (Linkie et al., Reference Linkie, Guillera-Arroita, Smith, Ario, Bertagnolio and Cheong2013). Such bycatch data can contribute towards assessments of non-target species if spatial violations are accounted for (Edwards et al., Reference Edwards, Cooper, Uiseb, Hayward, Wachter and Melzheimer2018). Occupancy modelling is forgiving of non-homogenous survey parameters and imperfect detection, and is therefore ideal for broad-scale collaborative analyses (MacKenzie et al., Reference MacKenzie, Nichols, Royle, Pollock, Bailey and Hines2006). Occupancy modelling estimates the proportion of areas occupied by target species, and assesses which parameters affect utilization, through the examination of detection/non-detection data. Camera traps are ideal for collecting such data (O'Connell & Bailey, Reference O'Connell, Bailey, O'Connell, Nichols and Karanth2011). Compensating for potential false absences is an essential function of occupancy analysis, which distinguishes between a naïve occupancy estimate and the best estimate of true occupancy (MacKenzie et al., Reference MacKenzie, Nichols, Royle, Pollock, Bailey and Hines2006).
A wide range of environmental, biological and anthropogenic factors can affect occupancy. In many cases, parameters associated with human disturbance and intraguild relationships have a greater influence on carnivore occupancy than environmental covariates (Schuette et al., Reference Schuette, Wagner, Wagner and Creel2013; Strampelli et al., Reference Strampelli, Andresen, Everatt, Somers and Rowcliffe2018; Wang et al., Reference Wang, Royle, Smith, Zou, Lü and Li2018). For species that increasingly inhabit human-modified landscapes, a deeper exploration of how biological and anthropogenic factors affect occupancy is required.
We used camera trapping and single-season, single-species occupancy modelling to examine factors affecting brown hyaena Parahyaena brunnea occupancy across its South African range. Humans pose both direct and indirect threats to carnivores through habitat loss and degradation, prey depletion and poaching (Ripple et al., Reference Ripple, Estes, Beschta, Wilmers, Ritchie and Hebblewhite2014). Private land used for farming comprises a large proportion of brown hyaena range, and is vital to their survival (Kent & Hill, Reference Kent and Hill2013). Despite their ability to survive outside protected areas and in close proximity to people (e.g. Kuhn, Reference Kuhn2014), brown hyaenas are subject to anthropogenic pressures, causing the species, which is categorized as Near Threatened, to come close to qualifying as threatened on the IUCN Red List (Wiesel, Reference Wiesel2015). At present, research on brown hyaenas is predominately limited to geographically scattered, small-scale studies. As a nocturnal and cryptic species, there is a paucity of data on the parameters affecting the species occupancy on a broader scale (Yarnell et al., Reference Yarnell, Richmond-Coggan, Bussiere, Williams, Bissett, Welch, Wiesel, Child, Roxburgh, Do Linh San, Raimondo and Davies-Mostert2016). By compiling data from 25 surveys into the largest detection/non-detection dataset collected on the species, our study makes a critical step towards filling this gap in ecological knowledge for protected areas and areas used for ecotourism. We also aim to demonstrate the potential for data initially collected to examine one target species, the leopard Panthera pardus, to provide a broadscale assessment of a bycatch species, the brown hyaena.
We collated data on brown hyaena detection and non-detection from 25 camera-trap surveys conducted across South Africa during 2013–2017 (Fig. 1, Table 1, Supplementary Material 1). The area surveyed totalled 7,705 km2. Camera-trap survey sites within and outside formally protected areas were represented (21 formally protected sites and 4 privately protected sites). Although sites are either designated as protected areas by IUCN (UNEP-WCMC, 2018; Supplementary Table 1) or privately prioritize the conservation of native species, sections of KwaZulu Private Game Reserve and adjacent private properties, Lajuma and adjacent private properties, and Little Karoo are unprotected. All but three sites are fenced (Supplementary Material 1). However, even electrified perimeters are often permeable and brown hyaenas in protected areas may occupy home ranges that also include unprotected areas (Kesch et al., Reference Kesch, Bauer and Loveridge2013, Reference Kesch, Bauer and Loveridge2015; Miller et al., Reference Miller, Pitman, Mann, Fuller and Balme2018).
1 Minimum convex polygon covered by camera-trap stations.
An array of Panthera V-Series camera traps (models V4, V5 and V6; Panthera, New York, USA) monitored all sites except Mountain Zebra National Park, where Cuddeback Attack camera traps (Greenbay, Wisconsin, USA) were used. The location and spacing of camera stations was optimized for estimation of leopard population density using a spatially explicit capture–recapture framework at all sites other than Mountain Zebra National Park. The mean distance between camera-trap stations was 2.04 ± SD 0.58 km and stations were dispersed evenly across survey sites. Camera stations consisted of pairs of camera traps mounted on poles or trees c. 40 cm above the ground, and 2–3 m from a road or trail. Camera stations were visited weekly or fortnightly to download images, change batteries and maintain the cameras. Although most surveys were initially established for leopard population monitoring, camera traps were placed in locations with a reasonable probability of detecting brown hyaenas but without guarantee of detection, thus reducing false absences and adhering to the requirements of occupancy analysis (Mackenzie & Royle, Reference MacKenzie and Royle2005; MacKenzie et al., Reference MacKenzie, Nichols, Royle, Pollock, Bailey and Hines2006). Comparable detection probability for leopards and brown hyaenas was based on a similar shoulder height (Estes, Reference Estes1991), a preference by both species for travelling on roads and trails (Mann et al., Reference Mann, O'Riain and Parker2015; Welch et al., Reference Welch, Tambling, Bissett, Gaylard, Muller and Slater2016), and a high photographic capture rate. We tested photographic capture rate of hyaenas and leopards by placing a Reconyx Hyperfire HC600 no-glow infrared camera trap (Holmen, Wisconsin, USA), which has the fastest trigger speed and recovery time available (Trolliet et al., Reference Trolliet, Huynen, Vermeulen and Hambunkers2014), 2 m from the camera traps at one survey site (Lajuma and adjacent private properties). We detected a similar photographic rate using both models, thus confirming that all animals were detected despite brown hyaenas generally moving faster to cover larger daily distances than leopards (Mills, Reference Mills1990; Swanepoel, Reference Swanepoel2008; Martins & Harris, Reference Martins and Harris2013; Williams, Reference Williams2017).
Four key categories of site covariates (habitat; relative sympatric predator abundance; human activity, e.g. people on foot and vehicles; relative abundance of medium- and large-sized prey species) were quantified from camera-trap and habitat data (Table 2). Habitat classifications from Department of Environmental Affairs Republic of South Africa (2015) were assigned to each camera-trap station using the Point Sampling Tool in QGIS 3.0.1 (QGIS Development Team, 2018). Apex predator abundance was determined individually for cheetah Acinonyx jubatus, spotted hyaena Crocuta crocuta, leopard, lion Panthera leo and African wild dog Lycaon pictus. Prey with an average female weight of 15–50 kg were defined as medium-sized and those > 50 kg were considered large-sized (following Mills & Mills, Reference Mills and Mills1978). Small-bodied species have lower detection frequencies on camera traps (Henschel et al., Reference Henschel, Hunter, Coad, Abernethy and Mühlenberg2011) and do not comprise a large proportion of brown hyaena diet (Williams et al., Reference Williams, Williams, Fitzgerald, Sheppard and Hill2018), hence their exclusion. To ensure statistical independence between photographic events and to create a comparable baseline, consecutive photographs of the same species at the same station within 30 minutes were excluded (O'Brien et al., Reference O'Brien, Kinnaird and Wibisono2003). Species and human activity abundance were calculated using a relative abundance index: the number of single capture events per 100 camera-trap days (Treves et al., Reference Treves, Mwima, Plumptre and Isoke2010; Carter et al., Reference Carter, Shrestha, Karki, Pradhan and Liu2012). Site covariates with continuous relative abundance index values were standardized as z-scores because of the large range of values present (Harihar & Pandav, Reference Harihar and Pandav2012).
1 Predator represents sympatric predator abundance.
2 Department of Environmental Affairs Republic of South Africa (2015).
*We expect this variable to be important, but as it is ordinal it is not meaningful to predict an overall direction.
To avoid multicollinearity disrupting clear outcomes in regards to detection probabilities (MacKenzie & Bailey, Reference MacKenzie and Bailey2004), we tested relationships between each pair of covariates for multicollinearity, in R 3.4.3 (R Development Core Team, 2017), using Spearman's rank correlation tests, and confirmed outcomes using the R package VIF (Lin et al., Reference Lin, Foster and Ungar2011). Any pairwise correlation coefficients with Rho > 0.6 in the same model were considered correlated and one covariate was excluded (Tan et al., Reference Tan, Rocha, Clements, Brenes-Mora, Hedges and Kawanishi2017). Covariates with a variance inflation factor < 3 were retained (Wang et al., Reference Wang, Royle, Smith, Zou, Lü and Li2018).
Brown hyaena detection histories, with values of 0 (non-detection) and 1 (detection), were created to reflect the presence of brown hyaenas at each camera-trap station on each day of the survey. To meet the assumption of a closed population (Rota et al., Reference Rota, Fletcher, Dorazio and Betts2009), data from only the first 40 days of each survey were used. The average number of days per survey was 48 excluding Mountain Zebra National Park (367 days). The global model was tested for goodness-of-fit with detection histories collapsed into intervals of 4–7 days (MacKenzie & Bailey, Reference MacKenzie and Bailey2004). Collapsing detection histories into eight 5-day intervals resulted in the lowest overdispersion statistic (ĉ) value, thus maximizing model fit (Supplementary Table 2). This length of time did not over-compress the statistical power of the data, accurately represented the rarity of the study species, and was in line with other large carnivore occupancy studies (Negrões et al., Reference Negrões, Sarmento, Cruz, Eira, Revilla and Fonseca2010; O'Connell & Bailey, Reference O'Connell, Bailey, O'Connell, Nichols and Karanth2011).
A two-step approach was used to examine factors influencing brown hyaena occupancy. In the first stage, single-season single-species occupancy analysis was conducted using the R package unmarked 0.12-2 (Fiske & Chandler, Reference Fiske and Chandler2011), to test the effect of covariates on occupancy (ψ) and probability of detection (p) without spatial autocorrelation. The effect of survey covariates were modelled on the probability of detection (Long et al., Reference Long, Donovan, MacKay, Zielinski and Buzas2011). During this stage we identified site covariates to test during the second stage, a multivariate analysis with spatial autocorrelation.
Models were ranked using Akaike's information criterion (AIC), with higher-ranking models receiving the lowest AIC values (Burnham & Anderson, Reference Burnham and Anderson1998). The global model was assessed for goodness-of-fit using Pearson χ 2 tests (MacKenzie & Bailey, Reference MacKenzie and Bailey2004) and normal dispersion using ĉ. Sites with an inflated ĉ value were removed to improve the goodness-of-fit before re-running the occupancy analysis (Meredith, Reference Meredith, Gibbs, Hunter and Sterling2008).
Summed model weights for biological and anthropogenic covariates > 0.10 were calculated to determine which covariates should be analysed for spatial autocorrelation in the next stage (Schuette et al., Reference Schuette, Wagner, Wagner and Creel2013; Tan et al., Reference Tan, Rocha, Clements, Brenes-Mora, Hedges and Kawanishi2017). Summed model weights ≥ 0.5 showed a strong response and were retained (Barbieri & Berger, Reference Barbieri and Berger2004).
The retained covariates were modelled for spatial autocorrelation in the second stage. By collating data from multiple sites in which the spacing of camera traps was optimized for another species, our dataset is likely to violate the assumption of independence between sampling sites in occupancy modelling and to suffer from spatial autocorrelation (Legendre, Reference Legendre1993; MacKenzie et al., Reference MacKenzie, Nichols, Lachman, Droege, Royle and Langtimm2002), especially given the tendency of brown hyaenas to travel up to 50 km per night (Mills, Reference Mills1990). Spatial autocorrelation assumes that neighbouring camera stations have a greater likelihood of sharing a characteristic than more distant camera stations, and can produce overestimated precision in occupancy estimates and underestimated standard errors (Latimer et al., Reference Latimer, Wu, Gelfand and Silander2006; Johnson et al., Reference Johnson, Conn, Hooten, Ray and Pond2013).
Spatial autocorrelation was addressed by fitting Bayesian versions of the candidate set using the package stocc 1.2.3 (Johnson, Reference Johnson2015) in R. A restricted spatial regression model was used to assess occupancy on all combinations of covariates with a summed model weight ≥ 0.5 (Broms et al., Reference Broms, Johnson, Altwegg and Conquest2014).
The threshold for distinguishing spatial structure between camera-trap stations was set to 6.15 km, the mean radius of a brown hyaena home range, established from averaged brown hyaena home range estimates taken from 13 collared individuals in five of the survey sites (Supplementary Table 3). The Moran cut was 92.5 (0.1 × number of camera-trap stations, following Hughes & Haran, Reference Hughes and Haran2013). The detection and occupancy processes were assigned flat prior distributions with a Gamma distribution of 0.5 and 0.0005 (Johnson et al., Reference Johnson, Conn, Hooten, Ray and Pond2013; Wang et al., Reference Wang, Royle, Smith, Zou, Lü and Li2018). The Gibbs sampler for each Bayesian model ran for 200,000 Markov-Chain Monte Carlo iterations with a burn-in period of 50,000 and a thinning rate of 10. Covariates with a 95% Bayesian credible interval that did not overlap zero were considered to have a significant association with brown hyaena occupancy (Wang et al., Reference Wang, Royle, Smith, Zou, Lü and Li2018). Model parameter convergence was assessed using Geweke diagnostic statistics and the |Z| < 1.96 scores (Wang et al., Reference Wang, Royle, Smith, Zou, Lü and Li2018). The posterior predictive loss criterion (Gelfand & Ghosh, Reference Gelfand and Ghosh1998) was compared between the Bayesian restricted spatial regression and non-spatial models for positive spatial autocorrelation.
Robustness for the restricted spatial regression model was assessed using the area under the curve (AUC) statistic (Broms et al., Reference Broms, Johnson, Altwegg and Conquest2014; Wang et al., Reference Wang, Royle, Smith, Zou, Lü and Li2018). We inputted the median occurrences and spatially corrected ψ estimates at each camera-trap station with the package ROCR (Sing et al., Reference Sing, Sander, Beerenwinkel and Lengauer2005) in R to determine the AUC statistic.
After removing 19 stations that captured brown hyaenas during every 5-day sampling interval (Supplementary Table 4), to reduce overdispersion, the final occupancy survey included 965 camera-trap stations and totalled 36,999 camera-trap days. Brown hyaenas were recorded at 630 camera-trap stations for a total of 2,862 independent capture events, resulting in an overall naïve occupancy of 0.65.
Effort and survey site affected probability of detection (Table 3). No covariates displayed multicollinearity so all were retained in the subsequent analyses (r < 0.6 and variance inflation factor < 3). Models with ΔAIC < 2 were included in the candidate set (Table 4). The global model fitted the data well (ĉ = 1.21). Site covariates relating to the five sympatric predator, human and vehicle abundance had summed model weights > 0.5, and these were retained for the next stage of analysis (Table 5). Habitat and prey fell below this threshold and were therefore discarded from the second phase.
1 Akaike's information criterion.
2 Difference in AIC between each model and top ranking model.
3 AIC weight.
4 Number of parameters.
5 Twice the negative log-likelihood.
1 Akaike's information criterion.
2 Difference in AIC between each model and top ranking model.
3 AIC weight.
4 Number of parameters.
5 Twice the negative log-likelihood.
The restricted spatial regression model was more parsimonious than the non-spatial model (posterior predictive loss criterion: 2,606.07 vs 2,790.45), confirming that positive spatial autocorrelation influenced the data. Spotted hyaena and human abundance had a negative impact on brown hyaena occupancy, and leopard and vehicle abundance had a positive influence on brown hyaena occupancy (Table 6, Fig. 2).
Brown hyaenas are estimated to occur across 66% of the total sites surveyed (ψ = 0.66 ± SE 0.09) with site occupancy probability estimates of 0.12–0.98 (Supplementary Table 5). The majority of sites in eastern and southern South Africa had a lower mean occupancy estimate than those further north (Fig. 1). The AUC value was 0.74 for the restricted spatial regression model, suggesting reasonable levels of accuracy in our predictions of occupancy.
Our study used a detection/non-detection dataset to identify, with a high level of confidence, a suite of four biological and anthropogenic factors that influence brown hyaena occupancy at a national scale. These findings suggest that camera-trap data focused on one species can provide useful data to determine broadscale occupancy trends of other, non-target species, provided the species share habitats and have similar-scale home range sizes. Our study enables us to make generalizations about variables influencing occupancy of brown hyaenas on a scale not previously possible for the species. The 25 survey sites incorporated a diverse array of environmental, biological, and anthropogenic conditions that are found throughout their international range. Our results therefore have conservation management implications applicable across the global range of the species, especially in protected areas and areas used for ecotourism.
The relative abundance of sympatric apex predators had the strongest influence on brown hyaena occupancy. Spotted hyaena relative abundance had a strong negative impact on brown hyaena occupancy, whereas the relative abundance of leopards showed a positive relationship with brown hyaena occupancy. Sympatric carnivore density estimations were not available as covariates in this study. We encourage future research to test brown hyaena occupancy against predator densities. Although relative abundance indices are not a substitute for density (Jennelle et al., Reference Jennelle, Runge and MacKenzie2002), relationships between brown hyaena occupancy and sympatric predator relative abundance mirror trends found when brown hyaenas cohabit with high densities of dominant competitors. Spotted hyaenas present a competitive threat to the more submissive brown hyaena, through kleptoparasitism and occasionally as a source of mortality (Mills, Reference Mills1990). Mills & Mills (Reference Mills and Mills1982) similarly showed that brown hyaenas avoided areas of high spotted hyaena density in the southern Kalahari, regardless of prey abundance. This negative relationship may explain why brown hyaena mean occupancy was lowest at survey sites in eastern South Africa. Many of the eastern sites support healthy populations of dominant competitors such as spotted hyaenas and lions. High densities of dominant predators can restrict the success of subordinate predators such as African wild dogs, brown hyaenas and cheetahs (Mills & Gorman, Reference Mills and Gorman1997; Marker et al., Reference Marker, Dickman, Mills, Macdonald, Macdonald and Loveridge2010).
The negative interspecific relationship and spatial separation between spotted and brown hyaenas can largely be attributed to environmental adaptability and diet. Brown hyaenas have a catholic and opportunistic diet, a secretive nature and low water requirements, which have enabled the species to survive in areas where less adaptable carnivores cannot persist (Maude, Reference Maude2005). Spotted hyaenas, although also adaptable, have less plasticity in their habitat and prey requirements (Hayward, Reference Hayward2006; Schuette et al., Reference Schuette, Wagner, Wagner and Creel2013) and are highly vulnerable to persecution by humans on unprotected land, especially on livestock farms (Mills & Hofer, Reference Mills and Hofer1998). The brown hyaena's adaptability in diet and habitat requirements probably explains why we found that habitat characteristics and prey abundance had little effect on brown hyaena occupancy, as in other studies (Thorn et al., Reference Thorn, Scott, Green, Bateman and Cameron2009; Williams, Reference Williams2017).
Alternatively, some sympatric predators do not present a direct threat to brown hyaenas and may instead have a positive effect on brown hyaena success. The positive relationship we detected between relative leopard abundance and brown hyaena occupancy may be attributed to the environment leopards inhabit and their potential to provide a food source for scavengers. Leopards prefer areas away from urban development with few competitive apex predators and a high prey abundance (Gavashelishvili & Lukarevskiy, Reference Gavashelishvili and Lukarevskiy2008; Steinmetz et al., Reference Steinmetz, Seuaturien and Chutipong2013; Strampelli et al., Reference Strampelli, Andresen, Everatt, Somers and Rowcliffe2018), and thus their presence may reflect more secure living conditions for brown hyaenas. In addition, the brown hyaena is a poor hunter, and typically depends on scavenging to meet c. 95% of its dietary intake (Mills, Reference Mills1984; Maude & Mills, Reference Maude and Mills2005). It therefore relies on other large carnivores, such as the leopard, to kill larger prey species (Stein et al., Reference Stein, Fuller and Marker2013; Mills, Reference Mills2015). In areas where leopards are present, brown hyaenas have a similar diet (Stein et al., Reference Stein, Fuller and Marker2013; Williams et al., Reference Williams, Williams, Fitzgerald, Sheppard and Hill2018), which can be explained by a high frequency of observed scavenging incidents from leopard kills (76%; Stein et al., Reference Stein, Fuller and Marker2013).
Another factor that influences brown hyaena occupancy is anthropogenic activity. Brown hyaena occupancy was lower in areas with high human foot traffic. A similar trend was detected on private land in northern Limpopo Province, South Africa, where brown hyaena occupancy in unprotected farmland was negatively affected by high disturbance by people on foot (Williams, Reference Williams2017). Striped hyaenas Hyaena hyaena also avoid human disturbance, with higher occupancy in rugged areas devoid of human activity and in areas with the greatest distance from human habitation (Singh et al., Reference Singh, Qureshi, Sankar, Krausman, Goyal and Nicholson2014). This finding fits well with the declines observed in carnivore populations globally as a result of human activity (Ripple et al., Reference Ripple, Estes, Beschta, Wilmers, Ritchie and Hebblewhite2014).
Our finding that vehicle activity positively influenced brown hyaena occupancy was unanticipated, as we expected the abundance of vehicles to affect brown hyaena occupancy in the same manner as the relative abundance of people on foot. People on foot could represent threats to hyaenas such as poaching, legal hunting or control of damage causing animals, in addition to less threatening ecotourism or more general human activity. In contrast, vehicle-based ecotourism is conducted across many of our survey sites, and in these areas vehicles movements predominantly occur during the day, when brown hyaenas are resting, and thus pose a low risk. People in vehicles are drawn to sightings of iconic animals, which could be more likely to occur in areas with preferable conditions for brown hyaenas (e.g. areas where human foot traffic is low and resources are abundant). Vehicle-based ecotourism may also contribute to anti-poaching efforts by providing additional surveillance (Baral, Reference Baral2013).
Despite the large spatial scale of our study, private land used for farming was underrepresented in our dataset. Management of the analysed survey sites, especially in regards to fencing and supplementary feeding, may differ in unprotected land and we recommend further research to examine how these site-specific factors affect brown hyaena occupancy. Large predators such as spotted hyaenas, lions, cheetahs and African wild dogs have been largely extirpated from much unprotected land in southern Africa (Ray et al., Reference Ray, Hunter and Zigouris2005). The absence of these species leaves leopards and brown hyaenas as the only apex predator across much of their shared range. Our results suggest that such conditions promote high brown hyaena occupancy as long as human disturbance is minimized, emphasizing the potential importance of private land for brown hyaena conservation.
Carnivores occurring outside formally protected areas are particularly vulnerable to anthropogenic threats (Balme et al., Reference Balme, Slotow and Hunter2010). With a severe decline in leopards already underway globally (Jacobson et al., Reference Jacobson, Gerngross, Lemeris, Schoonover, Anco and Breitenmoser-Würsten2016), conservation measures need to be established to preserve leopards on private land; our results suggest this will also aid brown hyaena conservation across their range. We thus recommend a holistic, multi-species approach that considers the entire large carnivore guild, rather than a single-species approach. Broader approaches to conservation that encompass multiple species or even entire landscapes are becoming increasingly popular, such as the focal species paradigm (Lindenmayer et al., Reference Lindenmayer, Lane, Westgate, Crane, Michael, Okada and Barton2014), which utilizes a suite of species, each of which is used to define various attributes in a landscape (Lambeck, Reference Lambeck1997).
Areas with low human disturbance are also vital for brown hyaena conservation and should be prioritized for protection. This is easier to facilitate in protected areas, but outside protected areas private land used for wildlife farming tends to have lower levels of human disturbance than land used for livestock farming (Thorn et al., Reference Thorn, Green, Dalerum, Bateman and Scott2012). Wildlife farms provide a plentiful prey base for predators and abundant scavenging opportunities for brown hyaenas, both naturally and through meat discarded after commercial hunts. Scavengers such as brown hyaenas provide important ecosystem services to these areas through their feeding habits (Beasley et al., Reference Beasley, Olson, DeVault, Benbow, Tomberlin and Tarone2015). Despite the potential for wildlife farms to act as refuges for brown hyaenas, intolerance towards predators and subsequent persecution can be high on wildlife farms (Pitman et al., Reference Pitman, Fattebert, Williams, Williams, Hill and Hunter2017b). Greater public education about the value of brown hyaenas and their dietary habits is therefore required.
Survey site and effort influenced brown hyaena detection probability. The effect of site variability on detection can be the product of a multitude of variables, including the density of brown hyaenas, animal behaviour, thickness of vegetation and seasonality (Tan et al., Reference Tan, Rocha, Clements, Brenes-Mora, Hedges and Kawanishi2017). Although there are advantages of broad-scale studies, inconsistences in probability of detection highlight the importance of fine-scale or regional-scale surveys in conservation and incorporating a variety of knowledge sources, especially when constructing locally specific management approaches (Raymond et al., Reference Raymond, Fazey, Reed, Stringer, Robinson and Evely2010).
Finally, our case study is an example of how a collaborative approach that combines small-scale datasets and utilizes bycatch data can be a powerful tool to fit occupancy models at broader spatial scales. This approach can inform conservation management strategies at national or global scale when applied to species with a restricted international area of occupancy such as the brown hyaena. With limits of time, funding and resources challenging ecological research, this recycle and repurpose approach to data extends the potential of camera trapping to inform high-level wildlife management strategies. We suggest the development of better information sharing platforms to enable the collaborations necessary to share camera-trap data and conduct research at this scale. These platforms need to extend beyond the boundaries of protected areas, where research is frequently conducted. Collaborative data sharing from camera-trap owners outside protected areas will not only provide vital distribution and occupancy information but will also strengthen dialogues and relationships between conservation organizations, scientists and private landowners.
We acknowledge the financial assistance of the National Research Foundation and the Durham University COFUND research fellowship programme towards this study. Opinions expressed, and conclusions, are those of the authors and are not necessarily to be attributed to the National Research Foundation. Panthera funded research at 24 survey sites. At these sites we thank the Limpopo Department of Economic Development, Environment and Tourism, Ezemvelo KwaZulu-Natal Wildlife, the numerous reserves, Wildlife ACT, Wildlife and Ecological Investments, and Siyafunda Conservation. This research forms part of Camera CATalogue (cameracatalogue.org), a division of Panthera's integrated data management system. We are grateful to Zooniverse (zooniverse.org), their staff, and the 12,500 citizen scientists for their contributions in classifying camera-trap data. Data collected by JC at Mountain Zebra National Park was funded by the National Research Foundation and Rhodes University. STW is funded by a University of Venda postdoctoral grant.
Study design: KSW, DMP, RTP, GAB, RAH; fieldwork: RTP, GKHM, GW, JC, KSW, STW; data analysis: KSW; writing: all authors.
Conflicts of interest
This research abided by the Oryx guidelines on ethical standards.