Landscape-scale sexual segregation in the dry season distribution and resource utilization of elephants in Kruger National Park, South Africa
ABSTRACT
This study compared the dry season distribution of elephant bull groups and mixed herds and the resources driving these distributions within Kruger National Park, South Africa. It is important to understand what resources drive the distribution of elephants as this may be of relevance to understanding and managing their impact. It is also important to distinguish between resource use by bull groups and mixed herds because their impact on the habitat may differ. Our results indicated that sexual segregation, both in space and in resource selection, did occur in Kruger Park. Bulls roamed more widely in the park, and although their distribution and resource use overlapped with mixed herds, they also occurred in areas that mixed herds apparently did not, or could not, utilize in the dry season. This gave rise to areas used exclusively by bulls but no areas used exclusively by mixed herds. Lower collective feeding requirements as a result of smaller group size, wider habitat tolerance, and increased mobility as a result of bigger body size, as well as conflict avoidance with musth bulls in areas with mixed herds, might have been some of the reasons for bull groups roaming more widely and for the establishment of separate bull areas. Rivers were an important resource driving both the distribution of the mixed herds and bull groups, but with the mixed herds occurring closer to these resource hot spots than the bull groups. Tree cover proved important for mixed herds, probably because of the importance of shade and the higher nutritional requirements of the smaller-sized cows and calves. Artificial waterholes might have opened up previously unutilized areas for bulls in the dry season, especially on the grassy basaltic plains in the north of the park. However, the distribution of the mixed herds suggested that they did not occur in higher densities in areas surrounding waterholes.
INTRODUCTION
Although elephants (Loxodonta africana Blumenbach, 1797) have less stringent habitat requirements than many other smaller mammals (Laws, 1970), they do not use habitats randomly (e.g. Leuthold, 1977; Western & Lindsay, 1984; Viljoen, 1989; Verlinden & Gavor, 1998). This is especially true in the dry season when resources are less abundant and occur in hot spots (Ntumi et al., 2005). Stokke & Du Toit (2000) argue that investigations into sex- and age-related differences in elephant feeding patterns and habitat use, coupled with surveys of elephant population structure, would be logical precursors to meaningful quantifications of the impacts of elephant populations on savanna vegetation. In this paper, we investigate sexual segregation in space and resource use between mixed herds and bull groups in Kruger National Park (KNP), South Africa.
Various studies have shown that larger herbivores can tolerate a wider range of forage quality than smaller herbivores can (e.g. Bell, 1971; Geist, 1974; Jarman, 1974; Du Toit & Owen-Smith, 1989; Illius & Gordon, 1992; Wilmshurst et al., 2000). This phenomena is mostly ascribed to the absolute metabolic needs and retention time increasing with body size, and mass-specific metabolic rates decreasing, with gut capacity as fraction of body mass remaining almost constant. This means that larger herbivores can satisfy their nutritional requirements even when ingesting forage of lower quality, because the cell walls are better digested as a result of increased ingesta retention time (Demment & van der Soest, 1985). Therefore, critical forage quality thresholds are higher for small than large animals and critical forage quantity thresholds are higher for large animals than small animals. In the context of the present study, the theory of allometric scaling of dietary tolerance therefore implies that female and juvenile elephants will have higher nutritional requirements per unit body size than male elephants because of their smaller body size (Du Toit & Owen-Smith, 1989; Du Toit & Cumming, 1999). Considering this, together with the increased reproductive demands on cows (pregnancy and lactation), we can expect bulls to be less selective and more tolerant of lower-quality food than cows and juveniles (Stokke, 1999; Stokke & Du Toit, 2000). This would result in a wider use of habitats and thus a wider spatial distribution of bull groups compared to mixed herds. This forms the first hypothesis we addressed in this paper.
Although bulls have greater total water requirements (Young, 1970), cows and juveniles have higher relative rates of water turnover because of their higher rates of evaporative and respiratory water loss (Gordon, 1977; Beuchat, 1990). Furthermore, mixed herds with juveniles and neonates are less mobile than bull groups. As a consequence, mixed herds will have to drink more often and minimize their distance to the closest water source. The second hypothesis we tested is that mixed herds occur closer to surface water sources than bull groups. This hypothesis has also been proposed by Stokke & Du Toit (2002).
In this paper, we examined dry season sexual segregation in the KNP elephant population at the landscape scale. In particular, we considered differences between mixed herds and bull groups with regard to (1) spatial distribution, (2) forage and habitat selection (tree cover and herbaceous biomass), and (3) ranging distance from rivers and artificial waterholes. We expect bull groups to (1) have a wider spatial distribution, (2) be less specific in their forage and habitat requirements, and (3) occur farther from surface water than mixed herds.
METHODS
Study area
This study was conducted on a landscape scale and included the whole of KNP. The park covers a total area of nearly 20,000 km2 in the eastern Lowveld region of South Africa, adjoining the Mozambique border in the east and Zimbabwe in the north. Elevations range from 260 to 839 m a.s.l. and mean annual rainfall varies from around 750 mm in the south to around 440 mm in the north, but with large variations between years (Gertenbach, 1980; Venter et al., 2003). The west of the park is dominated by granites and their erosion products, whereas the east is dominated by basalts. The Lebombo Mountains, primarily rhyolite, run from north to south on the eastern boundary.
Four major vegetation types dominate the KNP landscape (Eckhardt et al., 2000). The south-west are well-wooded, with accumulation of low-quality grazing because of the nutrient-poor granitic soils underlying this area. Important trees include the red bushwillow (Combretum apiculatum Sond.), knobthorn (Acacia nigrescens Oliv.), tamboti (Spirostachys africana Sonder) and marula (Sclerocarya birrea [A. Rich.] Hochst.). In the south-east, on the basalts, the grasses are more palatable and tend to be more heavily grazed. Important trees include knobthorn, leadwood (Combretum imberbe Wawra) and marula. The north-west granites are dominated by mopane (Colophospermum mopane [Kirk ex benth.] Kirk ex J. Leonard) and red bushwillow and are poorly grassed. The north-east basalts are dominated by homogeneous 1–2 m high, multi-stemmed mopane shrubs, with more grass cover than the north-west granites.
Six perennial rivers cross the park from west to east. Furthermore, more than 300 boreholes were drilled between the 1930s and the 1980s as part of the water provision programme to ensure a reliable year-round network of surface water throughout the park (Gaylard et al., 2003). In more recent years, the desirability of such a high density of surface water, minimizing the influence of temporal variability in rainfall, has been questioned (e.g. Walker et al., 1987; Owen-Smith, 1996; Harrington et al., 1999; Owen-Smith & Mills, 2006). Consequently, the water provision policy was revised, resulting in the closure of more than half the artificial waterholes (Pienaar et al., 1997). The history of the water provision programme and the revision thereof is described in detail, respectively, by Pienaar (1985) and Pienaar et al. (1997).
Spatial framework
A common spatial framework to which all the Geographical Information System (GIS) and satellite data layers could be summarized had to be enforced. This is a common approach in ecological studies faced with data from different sources and with different underlying spatial resolutions (e.g. Tognelli & Kelt, 2004). Considering the spatial resolution of the different data sets, it was decided to summarize all variables to a common 1-km grid. This resulted in 19,269 grid cells covering the whole of the park. As could be expected, some cells on the boundaries were smaller than 1 km2, and as a result all the different data layers were summarized to the reference grid by averaging within the grid, rather than by adding within the grid (i.e. standardizing by area).
Elephant distribution data
Two KNP elephant distribution data sets were used. Data from the annual mega-herbivore census (a helicopter count of elephants, buffaloes and rhinos) were used for testing whether the elephant sexes exhibit spatial and resource segregation. These data were also used for calibrating the multivariate autologistic regression models that identify significant predictors of elephant distribution. The second data set came from the annual ecological aerial survey (EAS), which used a fixed-wing aeroplane to count all the large herbivores in the park. The EAS data set was used only for verifying the predictions of the autologistic models.
Helicopter mega-herbivore census
A helicopter was flown at an average height of 150 m and at a speed of 100 knots, with four observers on board, counting and noting all elephants, buffaloes and rhinos. These annual censuses were conducted over the whole park in the late dry season (August), using a total count on a pattern that follows all of the drainage lines (Whyte, 2001a). When a big elephant herd was sighted, the helicopter would circle lower until the observers reached consensus on the herd size. The number and location was plotted on a 1 : 100,000 map for later digitization. For each sighting, it was noted whether it was a mixed herd or a bull group. Mixed herds were defined to include breeding herds with a matriarch and other adult females with their offspring, as well as family units attended by adult males. Bull groups comprised of adult males only, occurring independently from family units. Individual elephant bulls were included in the bull group category.
We used data from 1985 to 1996 in this study, with census totals ranging from 6887 to 8371 individuals over this period. No accurate locations were recorded for the sightings prior to 1985. In 1997, the water provision policy in the park was revised and many artificial waterholes were closed (Pienaar et al., 1997). This caused a change in the resource distribution and hence data from post 1996 were excluded from the present analysis.
In order to get a better idea of elephant distribution, all sightings over the 12-year period were combined into a single GIS data layer. This approach of combining snap-shot data from various periods is common in ecological studies (e.g. Osborne et al., 2001; Lichstein et al., 2002; Silva et al., 2002; Tognelli & Kelt, 2004; Olivera-Gomez & Mellink, 2005). The data were then transformed into absence/presence data for each 1-km grid cell.
Fixed-wing EAS
The fixed-wing EAS were used as an independent data set for validating the model predications. The EAS was conducted annually over most of the park in the dry season (May–August), using a total area count on 800-m wide strip transects (Viljoen, 1996). A fixed-wing aircraft was flown at a height of 65–70 m at a speed of 95–100 knots, with four observers on board. The location and herd size of the key herbivore species were recorded. Most of the park was covered from 1981 to 1993, after which the methodology was changed and only sections or specific transects in the park were flown. We used data from 1985 to 1996, covering the same period as the mega-herbivore census data used in this study. Again, all the sightings were combined into a single GIS layer and transformed into absence/presence data for each 1-km grid cell.
The annual counting of elephants in both the mega-herbivore census and EAS may appear like duplication in effort. It is known that the EAS data contain an under-count bias (Redfern et al., 2002) and because mega-herbivores (elephant, buffalo and rhino) are of particular concern for managers of confined conservation areas, like the KNP, the helicopter census was designed to provide better estimates and distributions of these heavy, bulk feeders (Joubert, 1983). The greater manoeuvrability of the helicopter compared to the fixed-wing aeroplane improves data collection, resulting in better-quality data. However, even though the EAS data represent an under-count of the elephant population, it still provides independent and spatially accurate distribution data suitable for validating the regression models.
Resource variables
Table 1 lists the resource variables included in the study.
| Variable measured | Resource | Units | Data source |
|---|---|---|---|
| UTM northing | Forage* (broad-scale spatial trend) | m | GIS layer |
| Herbaceous biomass | Forage | kg/ha | Values estimated from co-kriging using field collected biomass data points and AVHRR derived NDVI metrics for interpolation |
| Tree cover | Forage and habitat (shade) | % | MODIS product (MOD44B: vegetation continuous fields) |
| Distance to closest main river | Surface water, forage and habitat (shade) | m | Rivers digitized from 1 : 50,000 maps and distance from rivers calculated using standard GIS procedures |
| Distance to closest artificial waterhole | Surface water | m | Waterhole locations GPS’ed or digitized from 1 : 50,000 maps, with distance from waterholes calculated using standard GIS procedures |
- * UTM northing is used as a crude surrogate for the rainfall gradient from the wetter south to the drier north.
Northing
There is a trend of decreasing annual rainfall from south to north, except in the extreme north-west and south-east, where topography influences rainfall (Venter et al., 2003; Redfern et al., 2005). The northing Universal Transverse Mercator (UTM) coordinate will therefore give a crude estimate of the rainfall gradient. The rainfall gradient is related to forage quantity, with more vegetative biomass in the south.
Average herbaceous biomass
Every May since 1989, herbaceous biomass has been estimated at more than 500 field sites scattered throughout the park. These sites were selected to be proportionally representative of the different landscapes in the park (Trollope et al., 1989). Field measurements were taken with a disc pasture meter at these sites and the herbaceous biomass calculated from a simple regression equation calibrated for Kruger. The method is fully described in Trollope & Potgieter (1986). Variograms of the average herbaceous biomass at each site revealed that the sites exhibited strong spatial structure, suggesting that geostatistical techniques may be useful for interpolation across the park. Co-kriging is one such geostatistical interpolation technique that minimizes the variance of the estimation error by exploiting the cross-correlation between the variable to be interpolated and another variable (Isaaks & Srivistava, 1989; Webster & Oliver, 2002). We established that advanced very high-resolution radiometer (AVHRR) derived normalized difference vegetation index (NDVI) metrics had a reasonable cross-correlation with biomass and therefore we used it as a secondary variable in the co-kriging. Various other studies have also shown that annual metrics derived from remote sensing can be useful for estimating vegetation cover (e.g. Hansen et al., 2002a; Schwarz & Zimmerman, 2005). We found that co-kriging, using the satellite-derived data described previously as secondary variable, gave better interpolation results than ordinary kriging without the satellite data. The data and method are fully described in Smit et al. (in preparation). The interpolation was carried out on a 500-m grid, before being averaged to the 1-km reference grid.
Percentage tree cover
These data were derived from the moderate resolution imaging spectroradiometer (MODIS). The product we used is referred to as MOD44B (vegetation continuous fields) and can be found on the internet (http://glcf.umiacs.umd.edu/data/modis/vcf/description.shtml). This product estimates for each 500 m pixel, the percentage of the pixel covered by (1) trees (2) bare soil and (3) herbaceous layer. The dependent data were derived by classifying high-resolution imagery and then aggregating it to the coarser MODIS scale. The independent data were derived from multitemporal metrics based on a full year of coarse resolution MODIS data. A regression tree algorithm was then used to predict the dependent variable of tree cover based on signatures from the multi-temporal metrics. The method is fully described in Hansen et al. (2002a) and verified for an African woodland in Hansen et al. (2002b).
We included tree canopy cover in our analysis, averaging it to the 1-km reference grid.
Distance to nearest main river
The distance from the centroid of each 1-km grid cell to the closest main river was calculated using standard GIS procedures. Rivers were classified as ‘main’ rivers if they were perennial or seasonal with reliable pools.
Distance to nearest artificial waterhole
The period studied was during the height of the water stabilization programme, with more than 300 artificial waterholes providing surface water for animals throughout the park (Gaylard et al., 2003). The distance from the centroid of each 1-km grid cell to the closest artificial waterhole was calculated.
Data analysis
Extent of bull group and mixed-herd distribution
As a first step to establish whether bulls have a wider occurrence in the park, we simply compared the proportion of the total number of cells within which bulls were observed to the proportion of cells within which mixed herds were observed. The McNemar test (McNemar, 1947) was used to test if the probability is the same to detect a bull group or a mixed herd in any given cell.
Resource segregation between bull groups and mixed herds
Each grid cell was classified into one of four categories depending on the sightings over the 12-year study period: (1) only bull group(s) observed, (2) both mixed herd(s) and bull group(s) observed, (3) only mixed herd(s) observed or (4) no elephants observed. Analysis of variance (anova), followed by Tukey's pairwise comparisons, was used to test for sexual segregation of the resources listed in Table 1, using the categories listed previously. Before conducting the analysis, histograms of all the variables were inspected for normality and the data transformed if necessary.
Significance of resource variables in predicting mixed herd and bull group distributions
To explore the multivariate significance of each resource variable in explaining the presence of mixed herds and bull groups, respectively, logistic regression models were used. However, conventional logistic regression ignores spatial autocorrelation in the residuals. This often inappropriately inflates the statistical importance of variables (Lichstein et al., 2002; Knapp et al., 2003). To overcome this, we adopted the approach proposed by Augustine et al. (1996), by incorporating an autologistic term in the models. The autologistic term was calculated as the probability of occurrence based on kernel density in a neighbourhood around the cell of interest. This means that cells with many herds in close proximity will have bigger autologistic values (higher probabilities) than cells with no herds in close proximity (lower probabilities), using a Gaussian distance decay function. After investigating the indicator variograms (Isaaks & Srivastava, 1989) for mixed herds and bull groups, it was decided to define the neighbourhood as 15 km, representing the range of the spatial dependence. Although calculated slightly differently, this approach of using a distance decay function is similar with that used in other studies (Augustine et al., 1996; Osborne et al., 2001; Lichstein et al., 2002; Tognelli & Kelt, 2004).
To select among multivariate model terms, backward selection procedures were used. The full model was run with all explanatory variables and the term with the highest P value was removed before re-running the model. Terms were removed until the biggest P value was less than 0.05.
The predicted surfaces from the regression models were cross-validated using the elephant distribution data collected in an independent survey (EAS). We compared if elephants indeed occurred more frequently in cells with higher predicted probabilities using a different data set than the one used for calibrating the models.
Statistical analyses were conducted using minitab (version 14) and all GIS analyses were conducted using arcview (version 3.2).
RESULTS
Extent of bull-group and mixed-herd distribution
Over the 12-year period of this study, bulls were observed in significantly more grid cells (24.89%) than the mixed herds (17.50%) (McNemar χ2 = 354.8164, d.f. = 1, P < 0.001).
Additionally, 25.08% of the cells within which bulls were observed occurred more than 2 km from the nearest cell within which a mixed herd was recorded, whereas only 3.96% of mixed herd cells were farther than 2 km from the nearest cells where bulls were observed. Figure 1 confirms this visually. Note the wide distribution of bulls and the exclusive bull areas as well as the apparent lack of such exclusive areas for mixed herds.
Dry season distribution of bull groups and mixed herds in the northern KNP (based on aerial surveys 1985–1996). Note the polygons representing areas dominated by the presence of bulls and the apparent lack of equivalent areas for mixed herds. Furthermore, note that the bull areas often occur in areas with less tree cover (tree cover is represented by a 7 × 7 moving average of MOD44B — see text for detail).
Resource segregation between bull groups and mixed herds
Figure 2 illustrates the anova results. Note that both distance to closest artificial waterhole and distance to closest river were normalized by the square-root transform. Northing was relatively uniformly distributed and could not be transformed to a normal distribution using the transformations usually employed, so care should be taken when interpreting the northing results.
Mean and 95% confidence intervals of (a) UTM northing, (b) herbaceous biomass, (c) percentage tree cover, (d) square root of the distance to closest main river, and (e) square root of the distance to closest artificial waterhole for cells within which (i) only bull groups, (ii) both mixed herds and bull groups, (iii) only mixed herds, or (iv) no elephants were observed. Upper case letters above the bottom line ticks indicate significant statistical relationships among the categories, based on the Tukey multiple comparison test. Categories with the same letter are not significantly different (P > 0.05).
Forage resources
Cells within which only bulls were observed were more north than cells within any of the other categories (Fig. 2a: F = 31.28, d.f. = 3, P < 0.001). The cells with no elephants were, however, more north than both the cells within which mixed herds and bull groups occurred concurrently, as well as the cells within which only mixed herds occurred. However, the northing of the cells within which mixed herds and bulls occurred simultaneously were not significantly different from the northing of cells occupied exclusively by mixed herds.
The categories differed concerning herbaceous biomass (Fig. 2b: F = 41.30, d.f. = 3, P < 0.001). Bull groups occurred, on average, in cells with more herbaceous biomass than cells within which no elephants were encountered. Furthermore, cells within which no elephants were observed had more herbaceous biomass than cells within which both mixed herds and bull groups occurred simultaneously. The latter did not differ significantly from cells within which only mixed herds occurred.
The tree cover was highest in the cells where mixed herds occurred and lowest in cells within which only bulls were encountered (Fig. 2c: F = 63.7, d.f. = 3, P < 0.001).
Water resources
The cells where elephants occurred were closer to the rivers than cells within which no elephants were observed (Fig. 2d: F = 112.27, d.f. = 3, P < 0.001). Cells within which both mixed herds and bull groups were observed were closest to the rivers, but this was not significantly closer to rivers than cells where only mixed herds occurred. Cells within which bull groups occurred exclusively were farther from the rivers than cells where mixed herds occurred exclusively, but it was still closer than cells where no elephants were observed.
The categories differed concerning distance to the closest artificial waterhole (Fig. 2e: F = 37.75, d.f. = 3, P < 0.001). The multiple comparison test showed that cells within which bull groups occurred were significantly closer to the artificial waterholes than cells within which no elephants were observed. However, cells within which both mixed herds and bull groups occurred as well as cells within which mixed herds occurred exclusively were farther from the waterholes.
Significance of resource variables in predicting mixed herd and bull group distributions
Autologistic model calibration
Collinearity between predictor variables may confound their independent effects. Therefore, prior to logistic regression analysis, we calculated the Pearson correlation coefficients (r) for all pairwise combinations of the independent variables. The correlations between the resource variables were moderate to low (| r | < 0.3) and therefore we did not consider collinearity to be of major concern (Berry & Feldman, 1985).
Table 2 summarizes the variables that were included in the models after backward selection. Note that both mixed herds and bull groups are attracted to rivers. Additionally, mixed herds prefer areas with higher tree coverage. Bull groups seem to be significantly associated with waterholes and although they occur throughout the park, their distribution is biased towards the north.
| Predictor variable | Mixed herds | Bull groups |
|---|---|---|
| Northing | Not significant | P < 0.05 (+) |
| Herbaceous biomass | Not significant | Not significant |
| % tree cover | P < 0.05 (+) | Not significant |
| Distance to closest artificial waterhole | Not significant | P < 0.05 (−) |
| Distance to closest main river | P < 0.05 (−) | P < 0.05 (−) |
| Autologistic term | P < 0.05 (+) | P < 0.05 (+) |
Autologistic model cross-validation
The autologistic models performed better than a random model in predicting mixed-herd and bull-group dry season distribution of an independently collected data set (Fig. 3). The cells were sorted from small to large predicted probabilities and based on this, divided into four classes with equal number of cells within each category (i.e. quartiles). The percentages of the total number of mixed herds or bull groups occurring within each of these categories are depicted in Fig. 3(a–b), respectively.
Results of the cross-validation procedure for the (a) mixed herds and (b) bull groups on an independent data set. The histograms represent the percentage of the elephant sightings of the independently collected ecological aerial surveys occurring in categories of increasing predicted probabilities. The categories are based on the quartiles. For example, 12% of all the mixed herds occurred in the 25% cells with the lowest predicted probabilities (Q1) and 40.5% of all the mixed herds were recorded in the 25% cells with the highest probabilities (Q4). A random model with no predictive ability would result in approximately 25% of all herds recorded in each quartile. The increasing trend indicates the predictive ability of the respective models (after Knapp et al., 2003).
The upward trend in both Fig. 3(a–b) indicates that the models are useful in predicting the distribution of mixed herds and bull groups, with more elephant herds occurring in cells with higher predicted probabilities. Note, however, that there are still considerable sightings in the lower probability classes. Considering (1) that the models were based on aggregating snapshot data from different years (2) the inherent inaccuracies associated with data collection and recording (3) that elephants may have been recorded as they moved across non-preferred habitat between feeding hot spots (Bayley et al., 1998) (4) the uncertainty associated with interpolation techniques (herbaceous biomass), (5) the inaccuracies associated with remote sensing data (tree cover), and (6) the fact that the various data layers were collected on different spatial scales, the models performed reasonably well. However, these models are probably more valuable in an exploratory way, indicating which variables drive the landscape scale and sexually distinct dry season distribution patterns of elephants, rather than being accurate models for predicting fine-scale localized patterns.
DISCUSSION
This study indicates differences between the elephant dry season distribution patterns of mixed herds and bull groups in the KNP. Compared to mixed herds, elephant bull groups had a wider scatter across the park. This provides support for the hypothesis that elephant bull groups have less specific habitat and food requirements (Stokke, 1999; Stokke & Du Toit, 2002), and therefore they can roam more extensively across a wider range of habitats than mixed herds. This can be expected because the larger-bodied bulls can tolerate a wider range of forage quality, are more mobile to move between feeding hot spots, and because of smaller group sizes, have lower collective feeding requirements than the bigger mixed herds. This also agrees with work carried out in Mozambique that found bulls’ home range to be larger than that of females (Ntumi et al., 2005). As resources are more widely available during the wet season, niche differentiation may be weaker during the abundant season when competition is less intense. For example, Gordon & Illius (1989) and Voeten & Prins (1999) have shown that the distributions of herbivores overlap more during the abundant season than during the lean season. Ntumi et al. (2005) also found that elephants occur in hot spots during the dry season when resources are less abundant. Therefore, although it seems as if the bull areas remain intact during the wet season (I Whyte, pers. obs.), the sexual segregation boundaries may become softer during the rainy season because of a wider availability of resources, and musth bulls seeking more mating opportunities.
Over and above the wider distribution of bull groups compared to mixed herds, our results demonstrate inter-sexual differences between the resources in the grid cells frequented by the mixed herds and bull groups, respectively. The univariate non-spatial analysis suggests that the average of resources in grid cells where mixed herds occurred exclusively differed significantly from the average of resources in grid cells where bull groups roamed exclusively. In addition, it was of particular significance that the resources in the cells within which both mixed herds and bull groups occurred concurrently were more similar to the resources in the cells within which only mixed herds occurred than the cells within which only bulls occurred. This illustrates that the bulls’ distribution and resource use overlaps with that of mixed herds, but not vice versa. One possible reason is that bulls can utilize resources that the mixed herds utilize, but mixed herds cannot utilize resources/areas that the bulls use (i.e. bull areas). Furthermore, the bulls’ distribution may overlap with mixed herds, as a result of musth bulls associating with the family units for mating opportunities. However, the time of the surveys (dry season) is when very few females are in oestrus (Smuts, 1975) and mating possibilities would be limited. It is also likely that the overlap is the result of young males that have left their natal herd, but still occur in the vicinity (Moss, 1988). We will leave this as an untested hypothesis because no data on size, age and social relationships of the bulls were collected. Note, however, that this hypothesis of young bulls leaving their natal herd and staying in the area close to the herd is in line with the principle of allometric scaling. Because elephants grow for their entire lifetime (Laws, 1966; Hanks, 1972), the young mature bulls leaving the mixed herds will be smaller than old bulls and may therefore have habitat and forage requirements similar to that of the smaller individuals in mixed herds (hence the observed resource overlap). However, as the bulls grow older and bigger, they can move out of these mixed herd areas into bull areas, exploiting resources that will not meet the requirements of the smaller cows, calves and new-generation young bulls.
Mixed herds were expected to occur closer to surface water sources than the bull groups because of higher rates of water turnover (Gordon, 1977; Beuchat, 1990) and lower mobility. This was indeed observed around the rivers (Fig. 2d), with mixed herds occurring closer to the rivers than the bull groups. Stokke & Du Toit (2002) observed the same pattern in Chobe, Botswana, with bull groups roaming farther from the rivers than the mixed herds during the dry season. Laws (1970) also noted that bulls could survive for longer periods away from water. A point of interest in Kruger, however, concerns the pattern around the artificial waterholes, with bull groups occurring closer and mixed herds being neutral to these artificial water sources (Table 2). We hypothesize that this preference for rivers by the mixed herds is because rivers not only act as water sources, but additionally as a habitat hot spot with shade and higher-quality browse. Many of the waterholes, especially in the north of the park, are located farther from the main rivers, sometimes in open, grassy areas with little tree cover. As seen in this paper, this is suboptimal dry season habitat for the mixed herds. Therefore, surface water availability did not seem to provide a strong enough incentive to lure large numbers of mixed herds away from the rivers and the areas with a denser tree canopy where nutritional, water and shade requirements could be met. However, our results suggest that waterholes may be more important for bulls. Visual inspection of Fig. 1 seems to further support this notion. However, it must be remembered that because of the data collection method, the results of this paper show that mixed herds are not aggregating around the waterholes at the time of the survey and do not necessarily imply that the mixed herds do not utilize the waterholes from time to time, at night, during other seasons or when moving across otherwise waterless areas.
The multivariate spatial analysis suggests that tree cover and distance to rivers are significant predictors for mixed herds’ distribution, with mixed herds preferring areas close to rivers and with higher tree coverage. This agrees with the work of Ntumi et al. (2005) in Mozambique who found that elephants prefer habitat types with a relatively closed canopy as well as riverine thickets. This may be expected, as the mixed herds, with their requirements for shade (Laws & Parker, 1968; Laws, 1970; Barnes, 1993) and higher-quality forage (Miquelle et al., 1992; Du Toit, 1985; Weckerley, 1993; Stokke, 1999) will select areas with adequate tree cover. Furthermore, many studies in Africa have proven that elephants prefer to browse rather than graze during the dry season, providing further support for the importance of trees (Laws, 1970; Beekman & Prins, 1989; Kabigumila, 1993). The bark and roots of trees also form important high-quality resources utilized by elephants.
Both distance to rivers and distance to waterholes, as well as UTM northing, significantly predict the bull's distribution (Table 2). It therefore seems as if it is primarily water resources, and not forage (tree or herbaceous cover) determining bull dry season distribution. This also agrees with our hypothesis that bulls, because of their size, can utilize a wider range of low-quality forage. Stokke (1999) found similar results in a study conducted at a feeding patch scale. It was found that family units discriminated between patches in their surroundings and selected patches offering the highest density of palatable species, whereas males were apparently less influenced by the distribution of resources in their environment and browsed in patches containing the same amount and combination of species as surrounding areas. In Kruger, bulls are often seen during the aerial censuses to stand fully exposed on grassy plains or under the few large trees on grassy plains, whereas it is extremely uncommon to see mixed herds in these open areas. We propose that bulls can occur on the open grassy plains in the dry season because of their wider tolerance for low-quality food, with grass being less nutritious than browse in the dry season. Also, individual or small groups of bulls need fewer trees to satisfy their shade and browse requirements (compared to the larger mixed herds, which need more shade, cover and browse, especially if young elephants are present).
Considering all the grid cells, the average distance to the closest artificial waterhole and closest main river were 4.4 km and 6.1 km, respectively, for the study period. This is relatively short distances that should not pose major difficulty for elephants to cross. Therefore, very little of the park was really significantly far from surface water and hence we expect that a full expression of how elephants relate to surface water could not be expressed in the Kruger context. Only if many or all the waterholes are closed or rivers dry up, may elephants alter their distribution with regard to surface water.
Management implications
The elephant population of the Kruger National Park (KNP) increased at a mean intrinsic rate of 7.5% per year between 1967 and 2000 (Whyte et al., 2003). The culling of elephants from 1966 to 1994 kept the population between 7000 and 8500 individuals. However, after 1994, a moratorium was placed on culling (Whyte et al., 1999) and the elephant population increased to 12,467 (0.66/km2) in 2005 (Whyte, 2005). Cumming et al. (1997) stated that an elephant density of 0.5/km2 could convert a savanna woodland to a shrubland. Various other studies have also proven that high densities of elephants modify and change animal and plant communities (e.g. Western & Gichohi, 1989; Moolman & Cowling, 1994; Johnson et al., 1999; Lombard et al., 2001; Western & Maitumo, 2004) and homogenize habitats. The core mission of the KNP is the preservation of biodiversity (Mabunda et al., 2003), and a loss of biodiversity as a result of the effects of a large elephant population is therefore unacceptable. However, whether Kruger has long reached, are about to reach or will ever reach an elephant population density that will cause structural homogenization or cause a loss of biodiversity, are highly debated topics (Anon, 2004) and hence elephant management, or the lack thereof, remains controversial (Whyte et al., 1998; Whyte, 2001a,b; Gillson & Lindsay, 2003; Skinner, 2005; Whyte & Fayrer-Hosken, in press). But not only is the need for elephant management in Kruger, and indeed many southern African countries, controversial, but also the options available to managers. Culling, immuno-contraception, translocations and metapopulations with movement corridors are some of the options often considered (Whyte, 2001b; Anon, 2004). Another non-intrusive management tool that has been proposed is the management of artificial water sources (Owen-Smith, 1996; Gillson & Lindsay, 2003). Considering (1) the results presented here (preference of mixed herds for areas close to rivers as opposed to areas close to artificial waterholes during the dry season) (2) the fact that the population growth rate did not decrease post-1997 when more than half of the waterholes in Kruger were permanently closed (Pienaar et al., 1997; Whyte, 2005) (3) the abundance of natural surface water in Kruger (Redfern et al., 2005), and (4) the abundance of elephants in parks with no or little artificial water provision (Barnes et al., 1999), it is unlikely that closure of artificial waterholes will affect the elephant population in the KNP significantly. In fact, smaller less mobile species with more specific habitat requirements will possibly be more affected if waterholes are closed within their home range (Smit et al., in press).
CONCLUSIONS
It is important to distinguish between resource use by elephant bulls and mixed herds because their impact on the habitat may differ. In this study, we compared the dry season distribution of bull groups and mixed herds, highlighting spatial and resource segregation between them in the Kruger National Park (KNP), South Africa.
It was found that bulls roamed more widely across the park and were more tolerant in their habitat requirements than mixed herds. It was of particular significance that resources in areas frequented by both mixed herds and bull groups were more similar to resources in areas where only mixed herds occurred, but differed from resources in areas that were used only by bulls. Also, it was possible to delineate areas exclusively roamed by bulls in the dry season, but no comparable exclusive mixed herds area could be found. This implies that bulls share space with mixed herds, but additionally have exclusive access to other areas and resources that are not available for mixed herds to utilize. These ‘bull areas’ occur in almost every studied elephant population (e.g. Moss, 1988) and we have shown this to be the case in Kruger as well. Bulls spend most of their adult lives living in peaceful harmony with other bulls, and only move out into the breeding herd areas when in musth.
During the dry season, mixed herds occurred more frequently in areas with abundant tree cover and close to rivers. Although bulls were also attracted to rivers, they occurred, on average, farther from them than mixed herds. Furthermore, artificial waterholes might potentially have opened up previously unutilized areas for bulls in the dry season, especially on the grassy basaltic plains in the north of the park. Elephant bulls show a distinct preference for drinking at artificial waterholes’ reservoirs, apparently preferring the clean water that they can easily access with their trunks. However, natural surface water did not seem to be a scarce enough resource for artificial waterholes to lure mixed herds away from the rivers where they could meet their forage, water and habitat requirements. The feeding and water requirements hypotheses proposed and examined in this study should be further tested in the field, tracking individual herds for extended periods and over different seasons.
ACKNOWLEDGEMENTS
We wish to thank the South African National Parks and the Scientific Services Division of the Kruger National Park for permission to use the census data. Dr Ruth Kerry, Brigham Young University, USA, is thanked for assistance with the herbaceous biomass interpolation. The Ernest Oppenheimer Memorial Trust, Skye Foundation and South African National Research Foundation (NRF), supported I.P.J. Smit's research. Prof Chris Smit (Department of Statistics, University of Pretoria), Dr Bernard Devereux (Unit for Landscape Modelling, University of Cambridge), Magdalene College (University of Cambridge) and the Department of Geography (University of Cambridge) are also thanked for general support. Four anonymous reviewers are acknowledged for their contributions to the manuscript.
