Investigating the effects of management practice on mammalian co-occurrence along the West Coast of South Africa

Study area

This study took place in the largely nutrient-poor Fynbos Biome (Mucina & Rutherford, 2012), along the west coast of South Africa, where annual average rainfall varies between 152 mm in the north and 265 mm in the south. The predominant vegetation type of the region is Strandveld which is dominated by sclerophyllous, broad-leaved shrubs that form communities of medium density to closed shrublands (Mucina & Rutherford, 2012).

We defined three study areas (scenarios) by their different management practices where the ‘stocking rate’ and species of managed ungulates at each scenario acted as a proxy for ‘management practice’. Managed ungulates were defined as those species that require the intervention of people to manage the populations e.g., removals, introductions, feed supplementation and census. Ungulates were classified into different size classes according to weight (Table 1), based on the natural segregation of weight ranges between the managed ungulates, where animals <25 kg were classified as small, between 26 and 200 kg were classified as medium and anything >201 kg was classified as a large. No managed ungulate was classified in the small class. Two scenarios were within a protected area (Postberg and Langebaan sections of the West Coast National Park) and the third was on a research farm (Lamberts Bay, hereafter referred to as the farm). The park is located approximately 100 km north-west of Cape Town, South Africa, was proclaimed in 1985 and since then it has expanded to its current size of approximately 47,000 ha (SANParks, 2013). The farm is located outside of the town Lamberts Bay which is approximately 100 km north of the park (Fig. 1). The farm was selected due to its being of comparable size (the majority of commercial farms along the west coast are used for crop production with little livestock and are therefore not of a comparable size to the two areas within the park) while still being within the bioregion and having comparable vegetation types.

Postberg

Postberg has historically been characterised by intensive management of large and medium-sized herbivores in a small area (1,800 ha). The land was originally acquired in the early 1800s by a group of farmers and was used primarily for winter grazing, but the land was also ploughed. Postberg was proclaimed as a private nature reserve in the 1960s, after which it was contractually included into the park in 1987. Many indigenous and extra-limital large and medium-sized ungulate species were introduced to Postberg since the 1960s which resulted in overgrazing of the small, fenced area. Managed ungulates at this site were estimated to occur at ± 11.3 animals/km² (Dataset S2) and included bontebok (Damaliscus pygargus pygargus), Cape mountain zebra (Equus zebra zebra), red hartebeest (Alcelaphus buselaphus caama), eland (Taurotragus oryx), blue wildebeest (Connochaetes taurinus), kudu (Tragelaphus strepsiceros), springbok (Antidorcas marsupialis) and gemsbok (Oryx gazelle). Although there have been consistent removal efforts of extra-limital and other ungulates since Postberg’s inclusion into the park, November 2016 (6 months prior to the study) saw a significant removal of these species from the area, resulting in slightly lower densities during this study. Due to slow rates of ungulate reproduction and vegetation recovery, the past overabundance of large herbivores in the Postberg section was nonetheless expected to have a legacy effect of potential disturbance on vegetation and small antelope that would extend through the duration of this research.

Langebaan

In the Langebaan scenario, large herbivores occur at lower densities and are not confined to a small area. Historically this area was also used for agriculture, which included livestock and crop production and different portions were proclaimed as part of the park in 1989 and 1996. Managed ungulates within the Langebaan scenario were eland, bontebok, and red hartebeest and occurred at around 7.6 animals/km² (C Cowell, SANParks, 2015, unpublished data).

Lamberts Bay (the farm)

The farm scenario made use of the Nortier research farm that falls under the management of the Department of Agriculture (Elsenburg), Western Cape Government and is 2,780 ha in size. Christie Rheeder, the manager of the farm, granted approval to conduct research at the farm on behalf of the Department of Agriculture (Elsenburg). Several resource flocks and herds are kept on the farm for the Directorate: Animal Sciences, while it is also the site of veld rehabilitation projects run by the Directorate: Plant Sciences. The livestock present include sheep (Ovis airies), comprising three breeds (Namaqua Afrikaner, Dorper, and SA Mutton Merino), Bonsmara beef cattle (Bos taurus) and ostriches (Struthio camelus). The ostrich flock was restricted to camps that were not surveyed. In addition to the livestock, Impala (Aepyceros melampus) and Nyala (Tragelaphus angasii) are also present on the farm and were considered as managed ungulates, which were collectively estimated to occur at around 7.6 animals/km² (C Rheeder, research farm manager, pers. comm., 2018). The research farm is located in a matrix of other farms where predator control is known to take place.

Survey design and in-field methods

The three scenarios were surveyed over the winter of 2017 using 18 Cuddeback®, model C3 blackflash (Non-Typical, De Pere, WI) camera traps. Postberg was surveyed between May 31st and July 3rd, followed by Langebaan between July 4th and August 7th and the farm from August 8th to October 16th 2017. We overlaid each area with a grid of 1 km² cells in ArcGIS (ESRI, 2012). The centroid of each of the 18 resultant cells served as the camera location. Once in the field, we used a handheld GPS to navigate to the centroid, after which we walked outward in a spiral fashion for up to 120 m from the centroid, seeking the first location where two or more signs of animal activity were detected (Colyn, Radloff & O’Riain, 2018). The camera was set at this point. Cameras were mounted approximately 40–50 cm above ground level, onto a wooden stake and faced in a southerly direction, away from the sun, to prevent false triggers and overexposure (Glen et al., 2014). The cameras were programmed to capture three burst photographs when triggered, with a 30-second delay between photographs.

Vegetation height and cover were measured at each site according to the protocol described in Colyn (2016), i.e., by taking a measurement at one and two-meter distances from the camera trap location in a North, South, East and West direction (eight measurements in total). Vegetation height was recorded at the prescribed 1 m and 2 m distances with a measuring tape (short vegetation height allowed for this method) and percentage cover were also measured at these points using a densitometer (Li, Peavey & Lane, 2000). Height and cover were averaged across the eight measurements for use in analyses. Elevation was extracted at site level from the CGIAR-CSI SRTM 90 m Digital Elevation Data (Jarvis et al., 2008) using the spatial analyst tool within ArcGIS (ESRI, 2012). Similarly, slope was calculated using a digital slope model from the same DEM using the spatial analyst tool within ArcGIS (ESRI, 2012).

We predicted that the higher managed ungulate abundance at Posterg and the farm would result in increased competition between small antelope and managed ungulates. Further, we expected a lower overall vegetation height and abundance of small antelope at both Postberg and the farm due to the higher abundance of managed ungulates and we expected a lower abundance of mesopredators at the farm due to the surrounding predator suppression. Specific predictions in relation to each covariate are provided in Table 2.

Data analysis

Camera trap images were downloaded and image data were processed and captured using the TimeLapse2 Image Analyzer software (Saul Greenberg, University of Calgary, Calgary, Alberta) and then exported to excel per camera station. Image databases of camera stations were merged for each management scenario. Non-animal photographs were removed from the database and when more than one species was captured in a single photograph, entries were duplicated and edited to capture the number of individuals of each species as separate records. If a photographed animal was not recognisable at species level but could be classified as either a small antelope or managed ungulate, then it was captured as such. Image data were binned into independent capture events using a loop in R (R Development Core Team, 2015) which grouped all captures of a particular species, at a particular location. It then calculated the time difference between each picture and partitioned photographs of the same species at the same location into 30 min interval groups. The photograph with the highest number of individuals of that species was selected from each group and appended to the analysis database as an independent capture. We used the tidyverse and ggplot2 packages (Wickham, 2016; Wickham, 2017) in R to manipulate databases and produce plots and summaries where necessary. Ungulate species diversity per area was calculated using the Shannon-Wiener index in the vegan package in R (Oksanen et al., 2018).

Occupancy

Since Postberg had the shortest survey duration (32 days), the data from the farm and Langebaan were partitioned to use only the data collected in the first 32 days, which also ensured that only data collected over the wet season (31 May–10 September 2017), were used for the occupancy analyses. Postberg however also had one camera failure which resulted in 542 trap nights for Postberg, compared to 582 for Langebaan and 579 for the farm. To assess how the presence of managed ungulates influenced the occurrence of small antelope, we conducted single season, two-species occupancy analyses in PRESENCE (Hines, 2006) and ran single season, single-species occupancy models in R (R Development Core Team, 2015) using the unmarked package (Fiske & Chandler, 2011).

Single-species occupancy

Site-specific covariates (Table 2) were captured in a separate site database and Pearson correlations assessed for those variables with numeric/continuous values. There was a strong positive correlation between vegetation height and cover (r = 0.75) and therefore we only used vegetation height in models as a proxy for vegetation structure.

We used the camtrapR package (Niedballa et al., 2016) in R to create a camera operation matrix and a detection history for individual species/or suites of species of interest (e.g., managed species). Temporal replication was defined per species by dividing the camera survey into sampling occasions which can range from 1 to 15 days (Kok, 2016). Occasion length varied for different species depending on their detectability. Shorter occasion lengths are better for assessing occupancy of species that are frequently detected, while longer occasion lengths are better for assessing occupancy of species that are less abundant and that have low detection probabilities. We experimented with different occasion lengths and settled on 5 days for caracal detection, 2 days for managed ungulate abundance and 7 days for steenbok and duiker occupancy.

Occupancy and abundance were analysed using the unmarked package in R (R Development Core Team, 2015) using the detection history of the species of interest, observation level, and site-specific covariates. Large and medium ungulate abundance per site was estimated using the abundance-induced heterogeneity model, occuRN function, (Royle & Nichols, 2003) in the unmarked package in R (R Development Core Team, 2015). Data were pooled for all three scenarios; effort was used to explain detection probability and scenario to explain occupancy. Small antelope occupancy and caracal detection were estimated using the single season, single species occupancy model implemented by the occu function (MacKenzie et al., 2002). Data across scenarios were pooled to estimate small antelope occurrence and scenario was included to explain occupancy, while effort was included to explain detection for all species. The managed ungulate species’ abundance estimates along with other site-specific covariates were used as predictors of small antelope occupancy. The best models were selected using the modsel and fitlist functions which produce a table of AIC and R² values. Models were assessed for relative goodness of fit by scrutinizing the AIC, delta and R² values in the model summaries (Burnham, Anderson & Huyvaert, 2011).

The “best” model is considered the model which produces the lowest AIC value. We also considered the delta value, which is the difference in AIC value between a model and the model with the lowest AIC. Earlier literature suggested that models with a delta value of >2 were poor, however recent evidence suggests that models with a delta value in the range of 2–7 should also be considered (Burnham, Anderson & Huyvaert, 2011). We made predictions based on the top models and inspected the 95% confidence intervals of the predictions (Table 3). Models that had no predictive power (i.e., produced lower and upper occupancy confidence estimates ranging between 0, i.e., complete absence, and 1, i.e., 100% occupancy) were not considered to be informative. If there were no covariates that provided strong predictive power, we made predictions based on either the null model or a model that used effort to explain detection and scenario to explain occupancy, depending on comparative AIC values.

Because many of the occupancy models did not have sufficient data or data variation to converge, we also tested particular hypotheses directly. We compared abundance, occupancy, vegetation height, detections and slope between each scenario. All these data were non-parametric (tested using the Shapiro–Wilk test of normality) and thus we used a Dunn’s Test that accounts for tied ranks. We also compared variability of vegetation height between scenarios, using an asymptotic test in the cvequality package in R (Marwick & Krishnamoorthy, 2018). We hypothesised that managed ungulates would show a preference for fallow lands and as such would be detected more often, with a shorter time between captures than recorded elsewhere. Conversely, we expected small antelope to avoid these areas, and thus their time between captures on fallow lands would be higher (fewer detections). To assess this we calculated the number of detections for small, medium and large ungulates at each vegetated site and compared the number of detections on fallow versus natural lands across areas for an equal number of days (32). We also compared the average time between independent captures for species from each group (small/medium/large) between fallow and natural lands in Postberg and Langebaan. These analyses were not conducted for the farm as no fallow lands are present there.

Temporal activity and overlap

Activity level and overlap of small antelope and managed ungulates were assessed using the overlap (Ridout & Linkie, 2009) and activity (Rowcliffe, 2016) packages in R (R Development Core Team, 2015). Prior to activity and overlap analyses, time was converted to decimal numbers. The time of sunrise and sunset was calculated using the StreamMetabolism package (Sefick Jr, 2016) and stored for each record based on the date and GPS location of the record. To increase the available sample size for the temporal activity and overlap analyses, we lumped the data generated by the systematic survey described above with camera trap data that was opportunistically collected in the Postberg and Langebaan sections of the park between June 2016 and February 2017, where cameras were set on management tracks using the same camera settings as above as well as data generated from the full 60 days for which the farm was surveyed. This resulted in 1,150, 1,159 and 1,213 trap nights for the Postberg, Langebaan and the farm respectively. We did not have enough time-of-day observations for caracal at the farm and Postberg to assess temporal activity and overlap with small antelope, so this analysis was restricted to the Langebaan site.

The analysis of circular data is specialised, and standard statistical measures such as mean and variance, or regression are not appropriate (Ridout & Linkie, 2009). Time was converted to radians, a requirement of the overlap package. Activity was broadly depicted by non-parametrically estimating activity patterns using kernel density estimation with the bandwidth concentration parameter set at a maximum of 3 (Ridout & Linkie, 2009). This was further multiplied and adjusted by 1.5 as per Rowcliffe et al. (2014) who noted that bandwidth adjustment of 1.5 gave the most robust and minimally biased activity level estimations. The degree of overlap in temporal use of different species groups was estimated using the coefficient of overlap, Δ, where 0 = complete separation and 1 = complete overlap. If there were less than 75 observations for one of the species, the Dhat1 overlap estimator was applied whereas if the sample size was greater than 75 for both species, Dhat4 was used (Ridout & Linkie, 2009). We compared the activity overlap of the species of interest across scenarios first to assess whether they displayed any difference in activity patterns in the different areas. Following this, we compared the overlap between managed ungulates and small antelope at each scenario to determine if there is any evidence of temporal niche partitioning between species groups. Data were bootstrapped and resampled 500 times for each overlap estimate to generate 95% confidence intervals. To test whether activity patterns differed between species, we applied the Watson-Wheeler test of homogeneity for circular data to non-bootstrapped data using the circular package in R (Agostinelli & Lund, 2017; Taşdan & Yeniay, 2014). Activity levels and overlap were plotted using the overlap package.

Species richness and abundance

Sixteen ungulate species were detected across the three areas during the study (see Table 1 for species; two species were not included in analyses because of single detections and/or detections in spring only). Highest ungulate species richness (nine) and diversity was recorded in Postberg (Shannon-Weiner diversity = 1.7), compared to Langebaan and the farm (Shannon-Weiner diversity = 1.15 and 1.14 respectively).

Abundance models for large ungulates had the best fit (R² = 0.63, Table 3). Model results reflected naïve occupancy (occupancy estimated without detection accounted for) results (naïve occupancy for large herbivores was 0.88, 0.941 and 0.05 for Postberg, Langebaan and the farm respectively, and 0.29, 0.33 and 0.72 for medium ungulates). Significant differences were detected in managed species abundance across sites. Medium sized ungulate abundance was significantly higher at the farm when compared to Postberg (p < 0.0017) and Langebaan (p < 0.001) with a mean site abundance of 1.62 individuals per site Fig. 2B. There was no difference in site abundance of managed medium-sized ungulates between Postberg and Langebaan (p = 0.054) with a mean of 0.511 and 0.371 individuals per site respectively. Large managed ungulates had a mean abundance of 0.074, 2.69, and 3.30 individuals per site for the farm, Langebaan, and Postberg respectively. As such, abundance was significantly lower at the farm compared to both park scenarios (p =  < 0.001), while there was no difference between Langebaan and Postberg (p = 0.19; Fig. 2A). However, when large and medium-sized ungulates were pooled as ‘managed ungulates’ Postberg was estimated to have higher abundance than the farm (p = 0.0461) areas (Fig. 2C).

Detection probability

Differences in detection probability were also detected across areas and between the focal species groups (p < 0.001, Fig. 3). Caracal detection was highest at Langebaan (ρ = 0.0520), followed by Postberg (ρ = 0.0398) and was lowest at the farm (ρ = 0.0253). Common duiker, steenbok and small antelope (common duiker and steenbok lumped) detection differed significantly between all scenarios (p < 0.001). Detections were highest at the farm (0.778, 0.416, 0.867 respectively) followed by Langebaan (ρ = 0.671, 0.384, 0.761 respectively) and Postberg (ρ = 0.194, 0.0974, 0.256 respectively). Conversely, large managed ungulates were detected most frequently at Postberg (ρ = 0.275) followed by Langebaan (ρ = 0.202) and were detected least frequently at the farm (ρ = 0.105, p < 0.001). While large ungulates had the lowest detection probability at the farm, medium and managed (medium and large lumped) ungulates had higher detection probabilities there (ρ = 0.324, 0.328 respectively) compared to Postberg (ρ = 0.261, 0.317 respectively) and Langebaan (ρ = 0.113, 0.231 respectively).

Occupancy

We found that when detection probability was <0.1, occupancy could not be estimated, and therefore we could not make confident predictions for caracal occupancy for Postberg or the farm (Lamberts Bay). Estimates of naïve occupancy ranged from 0.11 at the farm to 0.16 and 0.17 at Langebaan and Postberg.

The results from the two species occupancy models were not particularly informative. Data were insufficient to run two-species occupancy models in Postberg, but models from the other scenarios suggest that ungulate species in different size classes occur independently of one another (see Supplemental Information).

Occupancy models for common duiker provided better goodness of fit compared to steenbok occupancy models (R² = 0.55 vs. 0.25, Table 3). Common duiker occupancy (ψ) differed significantly (p < 0.001) across scenarios with the farm having the highest occupancy of 1 (±0 SE), followed by Langebaan (ψ = 0.889 ± 0.052 SE) and Postberg (ψ = 0.473 ± 0.098 SE). These estimates were very close to naïve occupancy: 1, 0.88 and 0.47 respectively (Fig. 4A). Steenbok occupancy was significantly different (p < 0.001) across all scenarios with the farm having the highest probability of occurrence (ψ = 0.841 ± 0.222 SE; naïve ψ = 0.77) followed by Langebaan (ψ = 0.486 ± 0.013 SE; naïve ψ = 0.44) and then Postberg (ψ = 0.192 ± 0.012 SE; naïve ψ = 0.17) (Fig. 4B). Models for small antelope (common duiker and steenbok lumped) produced similar results to that of common duiker. Vegetation height was the strongest predictive variable for common duiker occurrence. While medium ungulate abundance produced the most reliable predictions for steenbok occurrence (Table 3), slope appeared to have a weak influence as it appeared in two of the top ten models. Estimates for the farm, however, remained uninformative due to 100% occupancy across sites.

Large managed ungulates were captured significantly more often on fallow lands (p = 0.02, Wilcox test), which were visited on average every 2.8 days, while natural lands were visited every 8.2 days across areas. This pattern was particularly prevalent in Langebaan, where most detections took place on or near fallow sites (Fig. 5A). Overall, there was no difference in small antelope detections between fallow and non-fallow lands however, the sample size of fallow sites was small and 32 detections at one fallow land site in Langebaan obscured potential patterns (Fig. 5A). There were no fallow land detections for small antelope at any of the five other fallow sites, inclusive of all fallow sites at Postberg.

Temporal activity and overlap

Activity patterns were typical for the species of interest. Managed ungulates were primarily diurnal, small antelope crepuscular (Fig. 6) and caracals mostly nocturnal (Fig. S1). We assessed temporal overlap between caracal and small antelope independently as well as pooled (common duiker and steenbok) at the Langebaan site. The overall trend across the overlap analyses was for small antelope activity to peak at periods of low caracal activity (Fig. S1). Activity overlap between managed ungulates and small antelope was also assessed separately for each scenario (Fig. 6). As hypothesised, activity overlap was highest at Postberg (86%, CI [0.77–0.93]) followed by Langebaan (79%, CI [0.73–0.84]) and the farm (74%, CI [0.68–0.79]). At the latter two sites temporal activity was deemed to be significantly different (p < 0.001).

Small antelope occurrence

As observed in other parts of Africa (Caro et al., 1998), management practice had an obvious impact on the occurrence of both steenbok and common duiker. Postberg consistently had the lowest probability of occurrence, with a much higher occupancy outside of the protected area at the farm (Fig. 4). Although few of our measured covariates had strong predictive power, vegetation structure was a good predictor for common duiker occurrence (Fig. S2), while medium ungulate abundance had a weak negative effect (Table 3) and slope a weak positive effect on steenbok occurrence (see results section; Heydenrych, 1995; Arsenault & Owen-Smith, 2002). This suggests that habitat and interactions with managed ungulates may be important drivers for the occurrence of small antelope within our study area. Therefore, we specifically explored interactions (i.e., competition and facilitation, habitat and predation) as possible mechanisms driving small antelope occurrence.

Competition versus facilitation

The lower occurrence of small antelope at Postberg is unlikely to be due to medium ungulates since medium ungulate abundance was comparatively lower at Postberg than the farm. However, managed ungulates at the farm were primarily livestock restricted to camps, therefore, the potential influence on small antelope would be restricted within and not widespread across the area. We also considered the potential of managed ungulates facilitating forage opportunities for small antelope. However, facilitation is considered unlikely due to the nutrient-poor soils and slow growth rates of the vegetation which encourages competition between ungulates (Fritz et al., 2002). In addition, small antelope occurrence is lowest in the areas with the highest densities of managed ungulates, which may indicate competition (Fig. 5). For example, elimination of buffalo (Syncerus caffer) from the Serengeti National Park is suggested to have driven increased abundance of small antelope as a result of competition release (Arsenault & Owen-Smith, 2002). While there seems to be little interspecific competition at the local level, this might not be the case at a larger scale. For example, roe deer and wild boar have been seen to occur independently of cattle at the habitat level in China but displayed segregation at the landscape level (Wang et al., 2018). Further, research in the Kruger National Park suggests that interspecific interactions may have effects on the distribution of African megafauna, but that this may not be evident at the local scale (Ryan & Ladau, 2017). The two-species occupancy results (see Supplementary Material) as well as the high overlap in activity (Fig. 6) suggest that small antelope and managed ungulates occur independently of one another at the local scale. However, given that the small area assessed in this study represents the full area under management (for Postberg at least), larger landscape level analyses are not possible in this context.

Habitat, forage and cover availability

Ecological theory suggests that for sympatric species to co-exist, subordinate species need to be able to exploit a resource which is not available to dominant species (Gordon & Illius, 1989). It is therefore likely that small antelope and managed ungulates have segregated food resources at our study site. This segregation may be due to the difference in body size that dictates forage requirements and forage availability on the vertical and horizontal planes (Cromsigt & Olff, 2006). Small antelope would be able to access new growth that is located at a low level or within the shrub itself, whereas the large ungulates are likely utilising the outer and higher parts of the shrub component. Larger ungulates naturally have more access along the horizontal plane as they are not restricted to home ranges (Jarman, 1974) and therefore can move greater distances between suitable foraging patches (Venter et al., 2015). Additionally, common duiker and steenbok are both known to prefer habitats where shrubs provide adequate cover (Heydenrych, 1995).

Vegetation height was found to be more variable in the park, compared to the farm. This was a result of the presence of former cultivated areas (fallow lands). These fallow lands appear to have been maintained by grazing pressure from large-bodied ungulates. Although managed ungulates were detected at a high proportion of sites, they may be transient at many of them and merely passing through. This is supported by the much higher detection of managed ungulates at fallow sites when compared to the more vegetated sites within the park. Research has also suggested that habitat heterogeneity, such as observed in Postberg and Langebaan, may drive movement patterns of larger herbivores as they move between suitable foraging patches (Venter et al., 2015).

Predation risk

Large managed ungulates were less active at night indicating that they were likely meeting their foraging requirements during the day since there is no predation risk for them. In contrast small antelope were more active at night. This likely stems from trade-offs between foraging and vigilance for predators (not required at the current site for larger species) that results in small antelope not meeting their foraging requirements during the day (Owen-Smith & Goodall, 2014; Rowcliffe et al., 2014). Caracal activity in the park is typical of that observed in other felids (Ramesh & Downs, 2015; Reilly et al., 2017), being predominantly nocturnal, with a distinct drop in activity between sunrise and midday after which activity picked up again. Small antelope showed a largely inverse, crepuscular pattern. This is indicative of anti-predator behavior portrayed by small antelope. However, there is a degree of temporal overlap between the species (delta ∼0.5) and this might suggest that caracals do not present a significant risk as a predator, which is corroborated by previous research which found that caracal diet within the park consists primarily of rodents (84% occurrence in scat), while small antelope were less important with a 6.5% occurrence in scat (Avenant & Nel, 2002).

Caveats

Ecological systems are complex and interactions can be influenced by multiple biotic and abiotic characteristics. The capability of our models was limited due to low variation in the data, such as detection of small antelope at all sites on the farm, which made elucidating patterns of interest difficult (e.g., Ramesh & Downs, 2015) did not even attempt occupancy models where naïve occupancy was 1). Although camera traps are excellent tools for monitoring animal communities, they still have some limitations as they only monitor a fixed point in the landscape. For example, dense vegetation restricts the detection zone, which could have influenced detection of species at vegetated sites. Therefore detection probability is a particularly important consideration when conducting camera surveys, especially surveys that focus on communities rather than an individual species as different species often have specific habitat requirements and are thus detected more frequently in areas where other species might not be detected.

Recommendations

While landowners have considered predator control as a possible means of bolstering small antelope populations, the results provide no clear indication that management intervention is required for the caracal population. Park management have commenced removal of large numbers of game, with the focus on phasing out extralimital species and to reduce the negative impact on vegetation. We expect that this process will allow for vegetation recovery and over time, an associated increase in small antelope. As such, we plan to do follow-up surveys in the future. The current survey is however very important for providing a snapshot of species occupancy and co-occurrence around the time of these removals.

There were many areas where either managed or small ungulates had naïve occupancy close to one. Given the small area of the park sections, there was limited opportunity for additional spatial sampling within Postberg or on a comparable vegetation type within Langebaan. Thus we suggest replicating the experiment in an area that allows greater spatial sampling, though no comparable reserve currently exists in the region. The alternative is to repeat the survey across the whole park (including larger areas and other vegetation types), to better understand requirements for existence of individual species as well as co-existence, which will benefit the management of small protected areas. Should recovery of small antelope to levels observed outside of the park not be observed over time, we recommend deploying GPS collars on managed ungulates and small antelope Langebaan and Postberg to provide a finer scale understanding of spatial and temporal resource partitioning and co-occurrence of these sympatric species.