Spotted hyaena space use in relation to human infrastructure inside a protected area

Study area

The KNP is situated in the north eastern corner of South Africa and covers almost two million hectares (Fig. 1). This study took place between May 2007 and March 2010 in a 5,000 km² southern portion of the park. Vegetation in the study area is characterised by woodland with basalt soils dominated by Clerocarya caffra and Acacia nigrescens, with Combretum species on granite soils (Ogutu & Owen-Smith, 2003). Rainfall is seasonal with the majority falling between October and March, with a peak in January and February (Venter, Scholes & Eckhardt, 2003). Average annual rainfall is approximately 650 mm for the Southern section (Venter & Gertenbach, 1986). Mean monthly temperatures range from 7 to 32°C for this area (Venter & Gertenbach, 1986). KNP hosts a diverse array of herbivorous and carnivorous mammals. Prey available for hyaenas in the Southern section of the park include, along with small mammals; impala (Aepyceros melampus), blue wildebeest (Connochaetes taurinus), Burchell’s zebra (Equus burchelli), greater kudu (Tragelaphus strepsiceros), common warthog (Phacochoerus africanus), imbabala bushbuck (Tragelaphus sylvaticus), nyala (Nyala angasii), common reedbuck (Redunca arundinum), waterbuck (Kobus ellipsiprymnus), steenbok (Raphicerus campestris), common duiker (Sylvicapra grimmia) and Cape buffalo (Syncerus caffer). Other megaherbivores such as African elephant (Loxodonta africana), white rhinoceros (Ceratotherium simum), black rhinoceros (Diceros bicornis), and giraffe (Giraffa camelopardalis) are also available, presumably most often as carrion. Impala in particular constitutes a large part of the hyaena diet in KNP (Henschel & Skinner, 1990; Ryan, 2007). Four large carnivores live sympatrically with hyaenas in KNP; African lion (Panthera leo), leopard (Panthera pardus), cheetah (Acinonyx jubatus), and African wild dog (Lycaon pictus).

Data were collected in two areas with contrasting levels of human activity. The Skukuza area included the Skukuza rest camp and staff village area (31°59′E, 25°00′S). Skukuza is the largest rest camp in KNP and hosts up to 300 visitors. It is also the administrative hub for the whole of KNP with a large staff village. In Skukuza, hyaenas had free access to the unfenced staff village consisting of 250 houses, an enclosed staff compound, a golf course, a shop, communal areas, and administrative buildings beside an enclosed area with tourist accommodation (rest camp). The staff area combined with the rest camp covers 4.3 km² and houses approximately 2,300 staff (Foxcroft, Richardson & Wilson, 2008). Fences around both individual houses and the compound prevented easy access to household rubbish bins. However rubbish bins in communal areas and larger waste collection sites were unfenced. Open gates and damaged fencing also allowed for opportunistic access to other rubbish bins. Hyaenas also had access to the unfenced car park of a picnic site, which contained rubbish bins, and they were able to walk along the perimeter fence of the tourist rest camp. Visitors are required to return to a camp or leave the park by a specific time that varies throughout the year to coincide with sunset and members of staff do not walk in unfenced areas after dark. Animals were therefore able to use unfenced anthropogenic areas after dark with minimal disturbance. In contrast, we also collected data in a neighbouring area (Doispane; 31°25′E, 25°01′S) approximately 20 km away that had limited levels of human activity and the only permanent infrastructure was a building that occasionally was used for short stays by park staff or guests. The Doispane area was at the border of the park and had similar to the Skukuza area access to permanent water. Vegetation in this region of the Kruger National Park is homogeneous (Rutherford et al., 2006). Water access, which is one of the main drivers behind herbivore distribution within the park (Redfern et al., 2003; Smit, Grant & Devereux, 2007), was similar between the two areas and prey densities are relatively homogeneous throughout this southern section of the Kruger National Park (Seydack et al., 2012). Therefore, the main differences between the two areas in terms of resource availability for spotted hyaenas are likely related to the elevated human presence in Skukuza caused by the Skukuza village complex.

Study animals and instrumentation

Each area (e.g., Skukuza and Doispane) was inhabited by one spotted hyaena clan. The clan in Skukuza consisted of 5 adult females, 1 adult male and up to 9 subadult or young adult males and 7 subadult or young adult females. The Doispane clan was substantially smaller and consisted of 3 adult females, 2 adult males and up to 2 subadult or young adult males and 1 subadult or young adult female. Both clans had juveniles present during the duration of the time they were monitored. We monitored the clan in Skukuza from May 2007 to December 2009 and the one at Doispane from May 2007 until August 2008. Monitoring was primarily done at the den locations but also when animals were opportunistically encountered. Observations were partly done for a concurrent study on the influence of human activity on hyaena behaviour and ecology. We monitored the locations of den sites throughout the study by visiting the clans. These visits varied in frequency from daily to once every second week. When a den was not located within sight of a road, we used clusters of relocations from marked animals (see below) at dawn and dusk to indentify likely den locations, which were confirmed by direct visitations. In each clan we had all animals individually recognized based on general characteristics and spot patterns. We scored rank relationships from the outcome of pair-wise interactions.

We fitted one animal in each clan with a collar mounted GPS unit that was tasked to download data through the GSM network (African Wildlife Tracking, Pretoria, South Africa). We selected the dominant female from each clan to create a reliable comparison (Boydston et al., 2003). The social rank was confirmed through behavioural observations of aggressive interactions between clan members. The animals were immobilised from a vehicle by a veterinarian from South Africa National Parks’ Veterinary Wildlife Service department. Both animals were first baited with three pieces of meat, each containing 2 × 15 mg midazolam tablets to enable safer darting. A combination of 4 mg medetomidine hydrochloride and 60 mg Zoletil was then delivered via a CO₂-powered dart rifle. An intramuscular injection of atipamezole was administered to reverse the effects of the medetomidine and animals were kept under observation whilst recovering. The female in Skukuza was fitted with her first collar on the 20th October 2007. This stopped working 5th July 2008 and was replaced 24th April 2009. The second collar stopped working on the 19th November 2009 and could not be removed. The collar on the female in Doispane was fitted on the 20th November 2007 and removed July 2011, although we only had access to data from this collar until 6 March 2010. We therefore collected spatial data on the female in Skukuza during the periods October 2007–5 July 2008 and 24 April–19 November 2009 and on the female in Doispane during the period 20 November 2007–6 March 2010. Hence, we collected data on both clans simultaneously for the majority of the time the Skukuza clan had an active collar, and we additionally collected data on an extended time period for the Doispane clan. Although sample size may bias home range size estimates, we have retained our full data record in the analyses to improve the accuracy of the estimated home ranges for Doispane. With the complete set of locations we had sufficient samples sizes in both Skukuza and Doispane to accurately estimate seasonal home range sizes (Supplementary Information 1), so that the uneven sample sizes should not influence any differences between the clans in terms of home range size. Both females were nursing during the time for which each clan were observed, i.e., May 2007 to December 2009 for Skukuza and May 2007 to August 2008 for Doispane.

Research was approved by the University of Pretoria Animal Use and Care Committee (protocol number EC010-07) and the Kruger National Park Animal Use and Care Committee, and was additionally carried out under a research permit from the South African National Parks Board for the project “Impact of human habitation on population dynamics of spotted hyaenas”.

Data collection, classification and analyses

The collars were set to take readings on an 11 h schedule. This schedule provided temporally independent points that covered all hours of the day. Each relocation point was classed as night, day or den. We regarded the time between one hour before sunset and one hour after sunrise as night time and times outside of these hours were as day time as it correspond to spotted hyaena activity patterns (Henschel, 1986; Kolowski et al., 2007). Although we acknowledge that we did not have direct measurement of activity during each of these time periods, our observations confirmed that activity within both clans were principally nocturnal, suggesting that most locations during the night were of active animals and locations during the day were resting locations. Data on sunrise and sunset times for the local area were retrieved from a weather service internet site (http://www.timeanddate.com). However, any relocation that occurred within 30 m of an identified den site was labelled as den points regardless of the time of day. In addition to these three classes of locations, we also grouped relocations by season. Following Venter, Scholes & Eckhardt (2003) we defined all relocations between October and March as having occurred during the wet season and the other relocations as having occurred during the dry season.

We used 95% Minimum Convex Polygons (MCP’s: Mohr, 1947) to estimate home range sizes for each animal. We used MCP’s to quantify home range sizes because they are relatively robust to possible temporal autocorrelation among data and they do not rely on arbitrarily chosen smoothing parameters or spatial resolutions of the underlying reference grid, which could influence the resulting space use estimates (Swihart & Slade, 1985; Row & Blouin-Demers, 2006; Boyle et al., 2009). MCP estimates are also repeatable across different software programs and therefore provide results that are directly comparable with those of other studies (Harris et al., 1990; Larkin & Halkin, 1994; Lawson & Rodgers, 1997). During October 2008, the clan at Doispane shifted its home range to the west with only a small overlap with the previous home-range. This shift included a shift in den locations, and our observations confirmed that all clan members appear to have shifted their movement patterns along with the marked female. We have therefore treated these two areas as separate home ranges for our analyses. Due to their highly clustered nature, we removed den site locations from all home range size estimations, but we have included them in the visual representation of the home ranges because den location potentially can influence home range patterns. For each home range, we created three size estimates, one including all relocations, one for the wet season and one for the dry season. We based our home range estimates on 745 locations (470 in the dry and 275 in the wet season) for the Skukuza female, 269 locations (138 dry and 131 wet season) for the Doispane female in the initial home range and 558 (195 dry and 363 wet season) locations in the subsequent home range.

We used two metrics to evaluate the spatial patterns of utilization within each home range. First, we quantified the utilization of the home ranges during the night, i.e., when we regarded the animals to have been active, using a normalized Shannon spatial diversity index (Payne et al., 2005). This index provides a measure of the evenness of home range utilization and varies from 0, which indicates a completely clustered utilization, to 1, which indicates a completely even utilization of the home range. The index is a quantification of continuous use of space, albeit sampled at discrete points in time. We selected this index for the night time locations because we regard them to be instantaneous point samples of a continuous movement process, and hence this index to be more appropriate than indices that explicitly evaluate patterns of discrete spatial points. To calculate this index, we first created a grid where the cells corresponded to 1% of respective home range, and calculated the number of relocations within each cell. The grid was confined within each respective estimated home range border. We selected this grid resolution as it provided a sufficient number of cells for calculations while avoiding an excessive number of empty cells. We calculated the index H′ as:

${\displaystyle H^{\prime }=\frac{\sum _{i=1}^N{P}_i\times ln\left(P_i\right)}{ln\left(N\right)}}$ where N is the total number of cells in each home range and P_i is the proportion of relocations in each given cell i. We calculated indices for all relocations combined as well as one index for each season. Second, we used the nearest neighbour index to quantify the spatial distributions of day time locations, i.e., when we regarded animals to have been resting (Clark & Evans, 1954). We opted for a separate index for the day time locations because it is explicitly quantifying the spatial distribution of discrete spatial events, which we believe was appropriate for the distribution of locations when animals were assumed to have been stationary. The nearest neighbour index (R) ranges from 0 (totally clustered distribution) to 2.15 (completely even distribution), and is scaled so that a value of 1 indicates a random distribution, values >1 indicate an over-dispersed distribution and values >1 indicate a clustered distribution. For both indices, we evaluated if the observed values deviated from expectations based on a random spatial distribution of points by generating 1000 random point data sets for each home range, each constrained within the home range border and with the same number of locations as the real datasets, and then calculated the index values for each of these random data sets. A random utilization is a sensible expectation to have under the null hypothesis of no preference for features or areas within a home range (Samuel, Pierce & Garton, 1985). We evaluated how likely the observed index values were under random expectations using a z-score transformation based on values from the randomly generated data (Baddeley, Rubak & Turner, 2015). As a heuristic way of comparing the spatial distribution of night and day time locations between the Skukuza and the Doispane clans, we subtracted the observed value from those calculated from the random data sets (Manly, 2007), and used these deviations from random expectations to compare the Shannon and nearest neighbour index between the Skukuza and each of the Doispane home ranges using two-sample permutation tests. We did one comparison for each pair of home ranges (i.e., Skukuza and each of the two Doispane ranges) for both seasons combined as well as one for each season.

We evaluated the utilization of the urban village area in Skukuza at two separate scales. First, we outlined the whole village area using satellite images retrieved from Google Earth (www.google.com/earth/), supplemented with GPS data collected in the field. We quantified the number of night time (i.e., active) and day time (resting) locations within and outside this area. Second, we described the utilization of different land use types within the village area. For this quantification, we similarly created a map that delineated four different types of land use in the area; highly modified areas—unfenced area with high levels of human use that are unfenced, administration areas—unfenced areas containing business buildings and their surrounding car parks with no fences, intermediately modified areas—areas that have been altered from their natural state but are without buildings or facilities, e.g., golf course and a cricket pitch, and unmodified areas—unaltered habitat inside the village boundary. We then scored each location in the village area to belong to each of these four classes. For both scales, we used a simple resource selection function to determine whether areas were preferred or avoided during night and day, i.e., while active or resting. Following Manly, Mcdonald & Thomas (1993), we calculated the selection indices β_i as:

${\displaystyle B_i=\frac{W_i}{\sum _{i=1}^H{W}_i}}$ where w_i is the selection ratios for each land use class i (i.e., the proportion of locations within each class divided by the proportion of available land that each class was covering) and H is the total number of land use classes. For ease of interpretation, we scaled each index so that a value of zero indicates that a class has been used in relation to its availability, a negative value suggests avoidance and a positive value suggests selection (Dalerum, Boutin & Dunford, 2007). We evaluated whether the utilization of the different habitat classes (i.e., within or outside of the village area or the four land use types within the village area) deviated significantly from a utilization based on availability using chi-square tests.