Human-mediated changes in habitat structure may disturb predator–prey relationships.
We investigated the influence of perimeter fences on the diet of a reintroduced population of African wild dogs Lycaon pictus Temminck 1820 in a 316 km2, fenced reserve in South Africa, by tracking radio-collared individuals during hunting periods to determine dietary composition from observed kills.
Nutritional status of impala Aepyceros melampus and kudu Tragelaphus strepsiceros prey, as measured by the percentage of femur marrow fat, was significantly lower than that of unselectively culled individuals. This supports the hypothesis that wild dog predation is at least partially compensatory.
Fence-impeded kills (those for which escape was deemed to be compromised by the fence) comprised 40·5% of kills (n = 316), and 54·1% of all edible biomass consumed. Compared with fence-unimpeded kills, fence-impeded kills comprised larger species (32·9 vs. 25·0 kg, W = 25667·0, P ≪ 0·001), older age classes for one prey category (female kudu: Fisher's exact test, P =0·02, n = 65) and animals in better condition for adult impala males (Mann–Whitney, W = 111·0, P =0·012, n = 28).
Fence-impeded kills also provided greater catch per unit hunting effort (27·3 vs. 12·2 kg km−1; χ2 = 7·89, P =0·005), resulting in longer interkill intervals. Movement of the pack towards the fence at the start of each hunting period suggested a decision to exploit the advantage that fences conferred for capturing prey.
Synthesis and applications. By enabling coursing predators to capture prey that would otherwise have escaped, fences may reduce the compensatory nature of predation, causing shifts in predator–prey dynamics that could influence the ability of small reserves to support such predators. The establishment of larger conservation areas to reduce perimeter-to-area ratios should be encouraged to limit the undesired effects of fences on predator–prey dynamics.
Patterns of prey selection by large terrestrial carnivores are shaped by a range of factors including energetic benefits and costs (Griffiths 1975; Creel & Creel 2002), mechanisms of selection including search images or prey vulnerability and habitat characteristics related to hunting or escape (Husseman et al. 2003). Human-mediated changes in habitat structure could disturb predator–prey relationships (Sweitzer, Jenkins & Berger 1997; Wittmer et al. 2007). Such changes include habitat alteration due to agricultural practices (Sweitzer, Jenkins & Berger 1997) and habitat fragmentation (Crooks & Soulé 1999).
The recent proliferation of small (<1000 km2), fenced reserves in South Africa has led to numerous large carnivore reintroductions (Hunter 1998). For example, African wild dogs Lycaon pictus have been reintroduced into several 50–900 km2 reserves (Davies-Mostert, Mills & Macdonald 2009), some of which are smaller than estimates of wild-dog home ranges under natural conditions in similar habitat (537 km2: Mills & Gorman 1997), and provide an opportunity to examine the effects of range constriction on predator–prey dynamics. Here, we used data from a reintroduced population of wild dogs to determine the effects of perimeter fences on hunting and prey selection.
Wild dogs are social, cursorial hunting carnivores (Fanshawe, Frame & Ginsberg 1991), with packs ranging from two to >20 adults and young (Fuller et al. 1992). As communal hunters, they increase the proportion of larger prey than would be taken if hunting alone (Estes & Goddard 1967; Creel & Creel 1995, 2002). In large conservation areas, they prey mainly on medium-sized ungulates ranging from 15 to 200 kg (Reich 1981a; Mills & Biggs 1993; Creel & Creel 1995, 2002; Mills & Gorman 1997). These are usually the most abundant prey species available (Hayward et al. 2006). Impala Aepyceros melampus are the principal prey in most southern African reserves, comprising 54–89% of prey by frequency (Reich 1981a; Creel & Creel 2002), with greater kudu Tragelaphus strepsiceros (Mills & Gorman 1997; Pole et al. 2004), common duiker Sylvicapra grimmia (Mills & Gorman 1997; Pole et al. 2004) and/or nyala Tragelaphus angasi (Kruger, Lawes & Maddock 1999) also frequently killed. Locally, wild dogs appear to avoid areas of high prey and competitor density (Mills & Biggs 1993; Mills & Gorman 1997; Creel & Creel 1998) although their densities are positively correlated to prey biomass at a landscape scale (Hayward, O'Brien & Kerley 2007).
As wild dogs are coursing predators, kills tend to occur when prey become exhausted (Fanshawe & Fitzgibbon 1993). Mean chase distances in the Selous Game Reserve were 0·57 km (range: 0·05–4·60 km; Creel & Creel 1995), but in Kruger were only 0·14 km (Reich 1981a). Coursing predators are likely to exert higher selection for animals in poorer condition than ambush predators (Estes & Goddard 1967; Husseman et al. 2003). In southern Zimbabwe, this was the case: wild dog prey were in poorer condition than unselectively culled animals. Wild dogs also selected juveniles of both impala and kudu (Pole et al. 2004).
All wild dog reintroduction sites in South Africa are small (<1000 km2) and by law are surrounded by ‘predator-proof’ fencing, which limits movements of predators and ungulates. High perimeter-to-area ratios increase the likelihood of encounters with fences, and this varies according to reserve size and shape. It has been proposed that wild dogs modify their hunting strategies to utilize fences to their advantage (Van Dyk & Slotow 2003; Rhodes & Rhodes 2004). Fence hunting is likely to influence the impacts of wild dogs on prey populations, enabling capture of different species, age classes, sex and condition, than would otherwise be taken (Van Dyk & Slotow 2003). Fences might also contribute to hunting success by enabling capture of individuals against the fence that might otherwise have escaped.
We investigated the influence of perimeter fences on prey selection at one reintroduction site. Attributes of prey were examined to test the hypothesis that fences confer a hunting advantage to wild dogs, such that: (i) fence-impeded kills would comprise a greater proportion of larger prey species than those not fence impeded, (ii) physical condition of prey would be better among fence-impeded kills and (iii) wild dogs would spend more time hunting close to fences to maximize the advantage they confer. Results are discussed in the context of large carnivore reintroductions into fenced habitat fragments and the implications of altered predation patterns for the sustainability of prey populations.
Materials and methods
The De Beers Venetia Limpopo Nature Reserve (VLNR) is a 316 km² privately owned reserve in Limpopo Province, South Africa (22°15′–22°30′ south, 029°12′–029°18′ east). Altitude ranges from 560 to 790 m, and the semi-arid climate is characterized by wet, hot summers and dry, mild winters. Mean monthly minimum and maximum temperatures are 20·3 and 32·0 °C for summer (December) and 7·2 and 24·7 °C for winter (June). Mean annual rainfall is ~350 mm, falling primarily from October to March. The reserve falls within the ‘Mopane Bushveld (Veld Type 10)’ of the savanna biome (Bredenkamp, Granger & van Rooyen 1996).
Herbivores include impala, kudu, blue wildebeest Connochaetes taurinus, gemsbok Oryx gazella, Burchell's zebra Equus zebra and waterbuck Kobus ellipsiprymnus. The reserve holds naturally occurring populations of lion Panthera leo, spotted hyaena Crocuta crocuta, brown hyaena Hyaena brunnea, cheetah Acinonyx jubatus and leopard Panthera pardus. Population sizes of hyaena, cheetah and leopard were low but stable (W.S. Davies-Mostert, pers. obs.). Lion numbers ranged between 12 and 30 (Davies-Mostert 2010). The reserve is enclosed by ‘predator-proof’ perimeter fencing, which mostly contains the large mammals.
A pack of 16 wild dogs (nine adults, seven pups) was reintroduced in January 2002 (Davies-Mostert, Mills & Macdonald 2009). Prior to release, five individuals were immobilized by tranquilizer darts loaded with Zoletil (Zoletil-Virbac, BP 27—06511 Carros, France) at a dosage of 0·4 mg kg−1 and each fitted with a VHF radiocollar. Interventions were conducted under permit (Northern Province Department of Agriculture and Environment permit no.: AG10610). Each study pack contained 1–5 VHF radio-collared individuals at any time, and collars were replaced before expiry.
Pack fission occurred in May 2002, when an adult male and female formed the splinter Pack 2, producing six pups in August 2002. A lion killed the alpha female in January 2003, and no pups reached adulthood: the male persisted as a solitary individual (Pack 3). Pack 1 fluctuated from 11 to 25 individuals, with peaks each year following the denning season, followed by attrition from mortality and dispersal.
Wild Dog Diet
We obtained dietary information from direct observations of kills, located by tracking radio-collared individuals during periods of hunting activity (dawn and dusk) and occasionally through the night. We were able to leave roads to locate kills, so our observations were unbiased except perhaps for very small prey. We supplemented long-term direct observations (3–10 consecutive days) with short-term observations (activity periods between two consecutive resting locations). When pack members became separated during hunts, we focused observations on a predetermined radio-collared individual.
Variables recorded for each kill included date, time, coordinates, species, sex, age, femur marrow condition, estimated edible biomass, size class, distance from fence, carnivore presence and whether the kill was fence impeded. We classified all carcasses within 20 m of a fence as fence impeded; all carcasses 20–100 m from the fence as uncategorized; and carcasses >100 m from the fence as not fence impeded.
Sex of prey was usually easy to determine as males of most species bear horns, while females do not (Packer 1983). We did not assign sex to juveniles due to the underdevelopment of horns in this age class. We assigned prey to age classes based on eruption and wear of teeth in the lower jaw (Mills 1991). We separated impala into nine age classes: Classes 1–4 included juvenile and subadult impala (<26 months old), and Class 5 was the youngest adult age class (29 months old) when adult dentition is complete. We did not attribute age to the remaining adult classes (6–9). Based on tooth eruption, kudu were categorized into juvenile (molars unemerged or still erupting), subadult (permanent premolars erupting) or adult classes (full permanent adult dentition).
We assessed impala and kudu body condition from the fat content of the femur marrow of wild dog kills (Sinclair & Duncan 1972; Reich 1981b; Pole et al. 2004). Femur marrow is a sensitive measure of body condition as the fat is mobilized after other bones (Brooks, Hanks & Ludbrook 1977). Following Pole et al. (2004), we extracted 2–10 g portions of marrow from the centre of the femur and froze these in airtight containers. Samples were weighed to the nearest 0·1 g and oven-dried until there was no further weight change. We estimated the percentage of marrow fat as:% marrow fat = % dry weight − 7, where 7 is a constant representing the proportion of nonfat residue left after drying (Sinclair & Duncan 1972). In some cases, due to field limitations, we could not dry the marrow samples, and therefore, we also assigned each sample to five colour and five texture classes. Colour classes ranged from Class 5 (white: good condition) to Class 1 (clear: poor condition), and texture classes ranged from Class 5 (hard: good condition) to Class 1 (liquid: poor condition).
We also determined the condition of impala and kudu harvested by recreational hunters. Although they tended to select males, they were not expected to be influenced by body condition. However, trophy-hunted animals were excluded to avoid biases through hunter selection for prime males, which may be of poorer condition than the rest of the population (Smithers 1983). Recreational and trophy hunting operations took place in completely separate meat handling facilities.
We obtained mean carcass weights from the literature and converted these to edible biomass (61% of body mass; Blumenschine & Caro 1986) to estimate edible biomass (kg) from each kill. This calculation assumes that all edible biomass was consumed. We grouped prey into three size classes irrespective of species: small (<25 kg), medium (25–90 kg) and large (>90 kg).
The Influence of Fences
We compared species composition, age, sex, size class and body condition of fence-impeded vs. unimpeded kills to determine the extent to which fences influence prey selection. Thick vegetation limited observations of hunts. We therefore used the distance of the kill site from the pack's previous resting location as a proxy measure of catch per unit effort to determine the comparative efficiency of fence-impeded hunts. Although not a direct measure of energetic costs, it provided a comparative measure of effort expended for successful hunts. Catch per unit effort (C, kg km−1) was calculated as: C = B/DR→K, where B is estimated edible biomass of kill (kg) and DR→K is the straight-line distance (km) from resting to kill. We predicted that if fences provided an advantage to the pack, C would be higher for kills that were fence-impeded than those that were not. Using a subsample of 57 activity periods, we assessed how kill biomass affected future hunting behaviour. For each activity period, we related total prey biomass to the number of periods elapsing before the next kill.
We used Duncan's (1983) index of preference (PI) to measure whether the pack selected areas close to the fence for hunting. To ensure independence of activity locations, we used only one location from each period, recorded 30 min after activity commenced and analysed morning and evening locations separately. We assigned activity locations to five zones according to distance from the fence (0–1, 1–2, 2–3·5, 3·5–5 and >5 km) and compared the proportional frequency at which we observed wild dogs in each zone to the proportional area covered by that zone, such that PI = (Uh/Ut)/(Ah/At), where Uh is the number of observations in a zone; Ut, the total number of observations in all zones; Ah, the area of a zone; and At, the total area of all zones. By excluding denning season kills, we avoided bias arising from proximity of den sites to the fence. PI was log-transformed (log10[PI + 1]) to give a normalized PI largely overcoming the problem of compressing values for zones that are not preferred relative to those that are (Duncan 1983). We also compared activity locations to pack daily resting locations to determine whether the wild dogs were spending more active time on or close to the perimeter fence.
We used aerial census data collected in October 2005 (for census methods, see Davies-Mostert 2010) to assess whether kill distribution patterns were explained by prey distributions. As above, Duncan's normalized preference index was calculated for each of the five zones, for six prey species.
Use of Data and Statistical Analysis
We recognized four seasons: denning (mid-June–mid-September) when the pack is at a den site and pups are immobile; postdenning (mid-September–mid-December) when the pack has left the den but movements are limited due to the presence of small pups; lambing (mid-December–mid-March) when main prey species are lambing; and predenning (mid-March–mid-June) when pups move with the pack.
We used Pearson chi-square analysis or, for small samples, Fisher's exact test, to test for differences in prey age or sex class selection and prey profiles between fence-impeded and not fence-impeded kills. We used the binomial z-ratio to test for differences between proportions.
We normalized percentage femur marrow fat values using the arcsine square root transformation (Krebs 1999) to resolve the issues of the percentage values being bounded by 0 and 1 and used a Generalized Linear Model to determine the relationships between colour and texture classes and percentage marrow fat, enabling the assignment of transformed values to unprocessed samples. We used Student's t-tests on the normalized data to test for differences in transformed marrow fat between wild dog kills and culled animals and differences in nutritional quality between kills made on and off the fence.
Using binary logistic regression, we identified factors characterizing kills assigned as fence-impeded vs. those that were not (Hosmer & Lemeshow 2000). We performed separate analyses using: (i) all kills and (ii) a subset of kills made outside of the denning season (when the wild dogs are mobile) to eliminate the potential effect of denning season bias on parameter estimation. As species, age, sex and size class were strongly correlated to edible biomass, we used the latter as a proxy measure of these characteristics. We evaluated and retained parameters in a forward stepwise manner. We compared models using Akaike's Information Criterion with small sample adjustment (AICc), where variable significance was expressed as the difference in AICc between each model and the model with the lowest AICc value.
Of 316 kills where we could determine use of fences, 128 (40·5%) were fence impeded (Table 1). A greater proportion of large prey animals was caught on the fence than off the fence (χ2 = 33·05, P ≪ 0·001). The median mass of fence-impeded kills was larger than those considered to be unimpeded (32·9 vs. 25·0 kg, W = 25667·0, P ≪ 0·001), constituting 54·1% of captured edible biomass. Kudu were twice as likely to be captured on the fence than away from the fence (43·0% and 16·0% of kills, respectively; Z =−5·27, P <0·001) and comprised 69% of the biomass from fence-impeded kills, but only 36% of the biomass of kills that were not fence impeded.
Table 1. The number and proportion of fence-unimpeded and fence-impeded wild dog kills at Venetia Limpopo Nature Reserve, 2002–2004
Ratio: proportion of fence-impeded kills/proportion of unimpeded kills.
Test for the difference between two proportions: H0, proportion of unimpeded kills = proportion fence-impeded kills; H1, proportions not equal; significance level P =0·05; > indicates higher proportion of unimpeded kills, < indicates higher proportion of fence-impeded kills,= indicates proportions equal. LS = tests not conducted for species with low sample sizes.
The proportion of males and females killed on the fence vs. off the fence did not differ for either of the two main prey species (impala: χ2 = 0·13, P =0·72, n =123; kudu: Fisher's exact test, P =0·12, n =78). Adult impala comprised a higher proportion of fence-impeded kills than those that were not fence impeded (0·71 vs. 0·56; Z =1·84, P =0·03). Conversely, juvenile impala were caught less frequently on the fence than expected (0·13 vs. 0·31; Z =−2·30, P =0·01). Adult female kudu were more likely to be caught on the fence than were younger females (Fisher's exact test, P =0·02, n =65). We could not conduct this comparison for male kudu due to small sample sizes.
Logistic regression revealed that edible biomass, followed by year, was the most important factor distinguishing whether a kill was fence impeded or not (Table 2). As biomass increased, so did the odds of a kill being fence impeded and the probability of a kill being fence impeded decreased between 2002 and 2004. When we excluded denning season kills, edible biomass became the most significant variable in predicting whether a kill was fence impeded or not. Edible biomass coefficients changed minimally between models which included, or excluded, denning season kills, indicating a robust relationship between prey mass and the odds of a kill being fence impeded. Both models held up to additional model-fit tests (all kills: H-L stat = 6·522, d.f. = 7, P =0·480; excluding denning season: H-L stat = 4·042, d.f. = 5, P =0·543).
Table 2. Logistic regression models of variables predicting the likelihood of wild-dog kills at Venetia Limpopo Nature Reserve, 2002–2004, being fence impeded vs. fence unimpeded, where fence-impeded kills are the reference cell
Variable significance is expressed as the difference in the Akaike's Information Criterion with small sample adjustment (∆AICc) between each model and the model with the lowest AICc value.
Continuous variable (i.e. each increase in edible biomass of 1 kg increases the odds of the kill being fence impeded by 1·02).
Categorical variable, design coded with 2002 as reference, x1 = 2003, x2 = 2004 (i.e. kills made in 2004 are less than half as likely to be fence impeded than those in 2002).
We collected bone marrow samples for 120 impala (56 wild dog kills, 64 shot) and 56 kudu (47 kills, nine shot) between May 2002 and September 2004. Among adult impala, wild dog prey were older than shot animals (χ2 = 12·19, P =0·002; Fig. 1). Transformed% marrow fat values (AMF) and texture classes (T) correlated strongly for both species, and we used the following equations
to assign transformed% marrow fat values to an additional 32 impala and 24 kudu wild dog kills for which we could not determine% marrow fat.
Mean transformed% marrow fat was lower for wild dog kills than culled animals for both impala (t1−tailed = 2·14, d.f. = 134, P =0·017) and kudu (Mann–Whitney, W = 546·6, P =0·003). For impala, this difference was most pronounced in the predenning season (Fig. 2). We could not conduct seasonal comparisons for kudu due to the small number culled (n =9).
Adult impala killed on the fence were in better condition than those killed away from the fence (Mann–Whitney, W = 882·0, P =0·026, n =63), particularly for fence-killed males vs. those killed away from the fence (Mann–Whitney, W = 111·0, P =0·012, n =28). There was no difference in condition of female kudu killed on or away from the fence (t = −0·02, P =0·56, d.f. = 13). Kudu males killed on the fence had higher percentage marrow fat than any other category, but as only one male was killed away from the fence, statistical comparison was impossible.
Catch Per Unit Effort
Mean distance travelled from resting to kill locations was only slightly lower for fence-impeded kills than unimpeded kills (2·5 and 2·9 km, respectively; Mann–Whitney test, P =0·716). However, fence-impeded kills comprised greater biomass and provided a greater median catch per unit effort (27·3 vs. 12·2 kg km−1; χ2 = 7·89, P =0·005).
Median distance from den site to kill site increased with time, from 2·2 km in 2002 to 5·9 km in 2004 (Kruskal–Wallis, P =0·004), likely to be due to increased distance of the den from the perimeter fence each year (mean weighted distance was 218 m in 2002 vs. 1·94 km in 2004) or to decreasing herbivore densities (Davies-Mostert 2010).
Biomass consumed per activity period had a strong influence on time elapsed between kills. Median interkill windows were two periods for large kills but only one period for medium and small kills (Kruskal–Wallis, H = 6·94, P =0·031). Longer interkill windows followed periods with greater biomass consumption, with a median consumption of 117·8 kg for interkill windows of more than two activity periods vs. 41·0 kg for windows of 1–2 activity periods (Mann–Whitney, W = 620·0, P =0·016, n =57).
Wild Dog and Prey Zone Use
Normalized zone use preference indices correlated negatively to distance from the fence (Spearman's rank correlation coefficient, rs = −1), suggesting strong selection for areas closer to the fence (Table 3). The pack was more likely to be found closer to the fence when active than when resting (χ2 = 7·94, P =0·047). Conversely, impala, kudu, gemsbok and waterbuck displayed strong preference for zones >2 km from the perimeter fence (Table 4).
Table 3. Zone selection by wild dogs at Venetia Limpopo Nature Reserve, 2002–2004, as measured by indices of preference (PI) Normalized indices > 0·3 presented in bold font indicate selection for a given zone
Zone (distance from fence, km)
Proportion of reserve area
Number of active locations
Proportion of active locations
Preference index (PI)
Normalized preference index (log10[PI + 1])
Table 4. Zone selection by wild dog prey during an aerial census at Venetia Limpopo Nature Reserve, October 2005, as measured by normalized indices of preference (log10[PI + 1]) for five zones. Indices presented in bold font indicate selection for a given zone
Count in each zone
Proportion in each zone
Normalized preference index in each zone (log10[PI + 1])
Our results indicate that human-mediated changes in habitat structure can lead to quantitative and qualitative shifts in prey selection patterns by wild dogs. The former refers to the differences in species composition among fence-impeded and unimpeded kills (e.g. fences allow more kudus to be captured), the latter to changes in selection parameters for different age, sex and body condition categories. These shifts provide benefits to wild dogs by increasing hunt efficiency and subsequently reducing the inherent risks associated with hunting.
The Influence of Fences on Diet and Prey Selection
The thick bush and flightiness of the herbivores precluded an accurate ongoing assessment of habitat-specific prey densities (e.g. strip counts, Caughley 1977; distance sampling, Buckland et al. 1993). Similarly, low visibility prevented measurement of prey encounter rates, the initiation and duration of hunts and hunting success – all of which are important for prey selection (Greene 1986; Creel & Creel 2002). Despite this, the large number of recovered prey carcasses enabled a thorough investigation into the influence of perimeter fences on dietary composition of wild dogs.
The diet of wild dogs at VLNR is similar to what has been found in other areas in the region (Van Dyk & Slotow 2003; Pole et al. 2004; Rhodes & Rhodes 2004) and is largely composed of impala and kudu. Fence-impeded kills comprised larger species and individuals than unimpeded kills, suggesting that fences facilitate the capture of large prey that would otherwise have escaped. Capturing larger prey was advantageous as it provided more biomass per unit effort. Furthermore, wild dogs at VLNR were usually able to completely consume prey carcasses and typically became fully engorged following a large kill. Spotted hyaenas were only once observed kleptoparasitizing wild dogs. The full use of carcasses made longer interkill intervals possible after larger kills. Fence hunting therefore conferred a double advantage by reducing both the time expended hunting and the overall number of hunts.
Seasonal variation in body condition of impala coincided with that described by Pole et al. (2004) for south-eastern Zimbabwe, with a peak in condition at the end of the wet season (March–May) and the poorest condition at the end of the dry season (September–November). Mean percentage marrow fat of wild dog prey was lower than that of animals culled unselectively from the population, lending support to the hypothesis that wild dogs select compromised prey (Pole et al. 2004). For impala, this difference was most pronounced in the predenning season (March–May), when the body condition of impala was highest. This contrasts with the findings of Pole et al. (2004) who found that selection for impala in poor condition is most pronounced between October and February when impala females are pregnant and lactating. However, our results support Temple's (1987) proposition that selection of animals in poor condition becomes most critical when prey are in generally good condition and therefore harder to capture. This is supported by the fact that impala males, which are generally difficult to capture (Creel & Creel 2002), were of better condition when killed on the fence than away from the fence. A similar result was not obtained for impala or kudu females.
Several processes influence predator diet, including predator–prey encounter rates (Pulliam 1974), decisions to initiate a hunt and the ability to capture the prey once a chase has begun (Greene 1986). Patterns of encounter are influenced by prey abundance, but other factors such as habitat selection, selective searching and active avoidance by prey are also important (Mech 1970). In Hluhluwe–iMfolozi Park, encounter rates between wild dogs and impala were strongly influenced by the dogs' selection of woodland over more open habitats (Kruger, Lawes & Maddock 1999). Mobile predators may structure their behaviour to maximize encounters with profitable prey (Scheel 1993) as shown by the wild dogs in this study preferentially hunting close to the fence, even though these areas were not preferred by prey.
For coursing predators, the ability to capture prey once a chase has been initiated depends largely on individual prey characteristics (e.g. defensive tactics, age, body condition), but is also influenced by habitat requirements related to hunting or escape. Husseman et al. (2003) found that wolf Canis lupus kills were more likely to occur in areas where vegetation was denser and hence prey flight ability inhibited. Predators do not choose to hunt every prey item they encounter. In Selous, wild dogs hunted only 37% of herds encountered and hunting decisions appeared to be based on the vulnerability of prey (Creel & Creel 2002). Hunts were also more likely to be successful when they ended in thick habitats (Creel & Creel 2002), which may be considered analogous to the perimeter fences in this study.
Implications for Prey Populations
Predation by coursing predators is at least partially compensatory: they tend to take prey with an apparently high risk of mortality from other causes. Modelling elk Cervus elaphus dynamics in Yellowstone National Park, Vucetich, Smith & Stahler (2005) found that the observed variance in elk populations was best explained by climate change and harvest rate and that wolf predation was a less important influence. Husseman et al. (2003) found that the probability of a kill being made by wolves increased as per cent femur fat decreased and described this as partially compensatory predation. In Save Valley Conservancy, wild dogs tended to take weaker individuals from prey populations (Pole et al. 2004) as we found in VLNR.
We provide strong evidence that perimeter fences increase the vulnerability of several prey species, size and condition categories. It is less clear how this affects prey populations, and impacts are likely to vary among species. The age and sex distributions of fence-impeded impala kills closely matched those of unimpeded kills: the only discernible influence of perimeter fences was manifested in the capture of higher condition adult males. In contrast, fences appear to have a strong influence on both the overall number and selection of kudu in the wild dog diet. Adult kudu females were almost three times as likely to be caught on than off the fence, and this proportion was even higher for kudu bulls. These results support earlier studies (Van Dyk & Slotow 2003; Rhodes & Rhodes 2004). Minimum area requirements for reintroduced wild dog packs are positively correlated to the proportion of kudu in the diet (Lindsey, du Toit & Mills 2004). As perimeter fences increase the proportion of kudu kills, they are likely to increase the impacts of predation on populations of such large-bodied prey species. Conversely, by enabling wild dogs to capture larger prey (thereby reducing the number of kills required to sustain a pack), perimeter fences may buffer populations of smaller prey.
Inverse density-dependent predation acts to limit but not to regulate prey populations at high densities, but at low prey densities may cause the extinction of prey (Sinclair & Pech 1996). Herbivores might be particularly vulnerable in a system such as VLNR where both ambush and coursing predators are present, where human harvest compounds predation effects and where it is fenced. We suggest that reintroduced wild dog packs are able to extend their prey profile using fence-impeded techniques to capture prey, but further monitoring of this and other reintroduced populations is necessary to establish the consequences of this.
Consequences for Wild Dog Conservation Management
Earlier studies of wild dogs in small, fenced reserves incorporated opportunistic kill data to assess the influence of perimeter fences on diet (Van Dyk & Slotow 2003; Rhodes & Rhodes 2004). This study, which utilized directly observed kills, revealed that perimeter fences may reduce the compensatory nature of wild dog predation by enabling them to capture prey of certain species, size, age and condition categories that might otherwise have escaped. The effects of these shifts will vary among species and may have important implications for the sustainability of fenced reserves for predators like wild dogs. The establishment of larger conservation areas to reduce perimeter-to-area ratios should be encouraged to limit the undesired effects of fences on predator–prey dynamics.
Thanks to Melanie Boshoff, Lynda Hedges, Herta van Helsdingen, Helen Trotman, Magriet van der Walt and De Beers staff for data collection. Pat Fletcher provided support and Mariette Wheeler, editorial assistance. H.D.M. was supported through grants to D.W.M. from Fauna & Flora International and Siren. Fieldwork was supported by the Endangered Wildlife Trust, De Beers Consolidated Mines and Jaguar Land Rover South Africa.