Correlates of survival rates for 10 African ungulate populations: density, rainfall and predation
Summary
- 1
Through reconciling census totals with population structure, annual survival rates were estimated for the juvenile, yearling and adult stages of 10 ungulate species over 14 years or longer in South Africa's Kruger National Park. During this period four species maintained high abundance levels, while six species declined progressively in abundance.
- 2
Multiple regression models fitted to these estimates indicated that juvenile survival was sensitive to annual variability in rainfall for most of these species, especially in the dry season component, but with no density feedback apparent. Rainfall components affected adult survival in several of the declining species, while negative density dependence in adult survival was evident for three of the four species that maintained high abundance. A negative effect of past prey availability, indexing putative changes in predator abundance, on adult survival was more strongly supported statistically among the declining species than the lagged effect of prior rainfall, potentially affecting herbaceous vegetation cover and composition.
- 3
The high sensitivity of juvenile survival to environmental variability among these ungulate species was consistent with the general pattern identified for large mammalian herbivores, although the absence of any survival response counteracting the density declines was surprising. The susceptibility of adult survival to environmental influences for the declining species appeared unusual and probably reflected an interaction between nutritional shortfalls and a numerical increase in lions, preying largely upon the adult segment of these species. The ungulate species that persisted at high abundance seemed resistant to effects of rainfall on food resources and evidently drove the changes in predator abundance. The sharp density effect on adult survival among these species could indicate prey switching by lions following changes in their relative availability.
- 4
Findings extend past generalizations about the demographic processes underlying the population dynamics of large mammalian herbivores and reveal how the survival rates of particular population segments respond differently to environmental influences. Demographic patterns help reveal the interplay of changing resource supplies, predation pressure and population abundance on population changes.
Introduction
Population dynamics is the outcome of the changing survival rates, reproductive success and movements of the animals constituting the population. Understanding of the environmental influences on these vital rates in large mammal populations has been advanced considerably by long-term studies of red deer Cervus elaphus L. on the Island of Rum in Scotland (Albon et al. 2000; Clutton-Brock & Coulson 2002), Soay sheep Ovis aries L. on the Island of Hirta in the Scottish Hebrides (Clutton-Brock et al. 1997; Coulson et al. 2001), and roe deer Capreolus capreolus in France (Gaillard et al. 1993), using individually marked or otherwise recognizable animals. Reviews of these and other studies have led to the conclusion that, among large mammalian herbivores, juvenile survival fluctuates widely in response to both stochastic environmental variation and changing population density, while adult survival tends to be resistant to such influences (Gaillard, Festa-Bianchet & Yoccoz 1998; Gaillard et al. 2000). Eberhardt (1977, 2002) suggested furthermore that, for such long-lived vertebrates, density-dependent feedbacks first affect juvenile survival, then age at first breeding, thereafter adult fecundity, and only lastly adult survival.
However, these generalizations have been derived largely from studies on temperate-zone ungulates, mostly conducted in environments lacking large predators. Separating the additive impact of predation from environmental influences affecting susceptibility to predation constitutes a difficult challenge. In many situations, predators merely consume prey that are predisposed to die because of nutritional deficiencies or old age (Scheel 1993). Predators can alter their prey selection in response to shifts in the relative availability of alternative prey species (Mills, Biggs & Whyte 1995). Vegetation changes in response to climatic variation may affect vulnerability to predation, as well as the nutritional status of animals (Smuts 1978). Numerical responses in predator abundance to changes in prey availability (Vucetich & Peterson 2003; Grange et al. 2004) commonly entail time lags, thereby tending to promote coupled population oscillations through feedback loops (Berryman 1999; Turchin 2003). The extent to which survival rates in different population segments are affected can provide important clues towards the mechanisms involved. The consequences of malnutrition seem to affect survival particularly through the juvenile stage (Owen-Smith 1990; Gaillard et al. 2000). In contrast, predation can fall differentially on either the adult or juvenile segments, depending on the predator concerned and the relative size of the prey species (Pienaar 1969; Mills & Biggs 1993; Kunkel & Pletscher 1999).
Owen-Smith & Mason (2005) found that the changed population trends shown by a set of ungulate species in South Africa's Kruger National Park (KNP) after 1986 were brought about mostly largely by reduced survival in the adult segment. This implicated elevated predation, specifically by lion Panthera leo L., as centrally involved. However, the population declines shown by many species were associated with a period during which rainfall received during the dry season months remained extremely low for several successive years (Ogutu & Owen-Smith 2003). The resultant lack of sufficiently nutritious forage during the dry season could have reduced survival rates in the adult segment either directly through malnutrition, or by making them more vulnerable to predation.
In this paper, we seek to establish the extent to which changes in adult vs. juvenile survival in these ungulate populations were associated with changes in environmental factors potentially influencing these rates. In African savanna environments, vegetation growth and hence food production for herbivores depends strongly on rainfall received during a wet season (Rutherford 1980). Rain falling during the dry season months promotes the retention of green foliage, and hence improves the nutritional quality of herbivore diets during this bottleneck period (Mduma, Sinclair, & Hilborn 1999). Rainfall also influences the composition of the herbaceous layer, and hence its capacity to produce forage of a suitable quality, but with delays in plant population responses (Kennedy, Biggs & Zambatis 2003). Habitat conditions in the form of grass cover and height furthermore affect the vulnerability of certain ungulate species to being predated by lions (Smuts 1978). The lion population also responds to changes in prey availability, leading to long-term oscillations in the relative abundance of this principal predator and its main prey species (Stevenson-Hamilton 1947). The prevailing density of the ungulate population additionally modifies the effective food availability (Owen-Smith 1990), and influences selection for alternative prey species by predators (Mills et al. 1995).
However, available data were restricted to annual aerial counts of large herbivore populations over an extended period, ground surveys of the sex and age structure of these populations, and rainfall records. No direct information on changes in predator abundance was available, because these species are not readily visible from the air. Accordingly, we devised proxy indices or statistical probes (Turchin 2003) to assess the weight of evidence provided by these data for the alternative mechanisms. Recognizing that statistical correlations do not necessarily indicate causal connections, we approach this analysis in the spirit of ‘ecological detective-work’ (Hilborn & Mangel 1997) or ‘adaptive inference’ (Holling & Allen 2002).
Materials and methods
the count data
The aerial census procedure has been described fully elsewhere (Viljoen & Retief 1994; Owen-Smith & Mason 2005). Briefly, surveys by fixed wing aircraft, involving four observers together with the pilot and a recorder, covered almost the entire 20 000 km2 extent of the KNP annually between 1977 and 1996. They were conducted during the dry season between May and August, when juveniles born during the preceding wet season were mostly about 0·5 years of age. A boundary fence precluded dispersal outside the park until 1993, when a section separating the KNP from private wildlife reserves was removed. The 10 ungulate species considered, order of abundance with mean census totals, were impala Aepyceros melampus (Lichtenstein; 110 000), Burchell's zebra Equus burchelli (Gray; 30 000), blue wildebeest Connochaetes taurinus (Burchell; 12 000), greater kudu Tragelaphus strepsiceros (Pallas; 7000), giraffe Giraffa camelopardalis L. (5000), common waterbuck Kobus ellipsiprymnus (Ogilby; 3000), warthog Phacochoerus aethiopicus (Pallas; 2500), sable antelope Hippotragus niger (Harris; 1600), tsessebe Damaliscus lunatus (Burchell; 750), and roan antelope Hippotragus equinus (Desmarest; 300). Zebra, wildebeest, impala and giraffe persisted at high abundance over the latter part of the study period, while the remaining six species showed progressive declines. For analysis, the census results shown in annual reports were partitioned among four regions (South, Central, North and Far North), each covering a roughly similar area and separated by major rivers. However, population structure samples for the two northern regions were inadequate for them to be analysed separately, so they were amalgamated into one Northern section of the park.
Sex and age classifications of all ungulate populations were carried out by park-wide ground surveys between August and October from 1983 to 1996. The procedures used to classify animals as adult (> 2 years) male, adult (> 2 years) female, yearling (age 1–2 years) or juvenile (< 1 year) are described elsewhere (Mason 1990; Owen-Smith & Mason 2005). After 1991 the observer changed, samples became smaller, and the distinction between juvenile and yearling zebra unreliable, so that some of the later data were discarded. For wildebeest and kudu, estimates of population structure prior to 1983 were available from studies by Whyte (1985) and Owen-Smith (1990).
Monthly rainfall records were averaged from 3 to 5 recording stations within each region. Wet season rainfall spanned October–March, and dry season rainfall April–September for the year preceding the count. Annual rainfall totals were calculated over the seasonal cycle July–June
data analysis
Annual survival estimates
Survival rates for the juvenile segment were obtained from juvenile/adult female ratios, with an adjustment for the age at which females first reproduced, following Owen-Smith & Mason (2005). Annual survival estimates for the adult segment were derived by balancing juvenile and yearling recruitment, represented by the proportions of these age classes in the population structure samples, against annual changes in population size, from the count totals, with the aid of a stage-structured spreadsheet model. The survival rate of adult males SAm was assumed to be lower than that of adult females SAf according to the relation SAm = (SAf)z, with the power coefficient z adjusted until the adult sex ratio in the modelled population matched the mean adult sex ratio observed for each species. The adult survival rates used in the analysis represent survival beyond the yearling stage, for both sexes combined. Survival rates from the juvenile stage into the yearling class were obtained from the ratio between the number of yearlings in the current year and the number of juveniles in the previous year using the spreadsheet model, thereby factoring out changes in adult female survival between years.
To suppress the effect of variable census errors on the count totals, and hence on the annual survival estimates for the adult segment, the original count totals were transformed using a two-point weighted interpolation: 
= 0·67Nt + 0·33Nt+1, where 
= adjusted population estimate, and Nt = recorded population count, for year t. Because data transformations are statistically contentious, analyses using adult survival estimates obtained directly from the original count totals are presented for comparison.
Survival estimates were made at a regional subpopulation level where the data were deemed adequate. Regions were amalgamated if necessary to ensure that the regional sample size of adult females for estimating adult female/juvenile ratios did not fall below 20 in any one year. Roan antelope were so uncommon that the requirement for annual samples to include at least 20 adult females was waived, and after 1990 this species was no longer encountered during the population classification surveys. For kudu, survival relationships from data obtained during the prior study over 1974–84, based on individually identifiable animals (Owen-Smith 1990), are presented for comparison.
Statistical procedures
Linear multiple regression models were fitted relating annual survival rates to various predictors, i.e. SX,t =b0 + baAt−1 + brRt + blLt–d, where SX,t = annual survival for stage class X into year t, At−1 = preceding population abundance, Rt = rainfall measure for the year t, Lt–d = a lagged predictor operating with a time delay d, and the bs represent the regression coefficients. Survival estimates were not logistically transformed because they were not derived as zero-one proportions, and could exceed 1·0 through sampling errors. Statistical relationships seemed acceptably linear, as judged by residuals and scatter plots, over the observed range in the data.
For large, long-lived mammals, density effects on resource availability for adults, and hence on juvenile recruitment, potentially operate over a period extending from prior to conception through gestation to post-birth survival. Accordingly, density dependence was assessed using a 3-year weighted average of the census totals, = 0·5Nt + 0·25Nt+1 + 0·25Nt−1, with the weighting centred on the year preceding that into which the survival estimate was made. This adjustment also avoided the spurious density dependence that can be generated by sampling errors in annual abundance estimates (Solow 1998). Because populations within different regions fluctuate around different density levels, the adjusted regional population totals were transformed into a relative abundance measure through dividing by the mean count totals within each region over 1978–94. This standardization enabled variation in survival rates within each region to be related to relative variation in predictors. The abundance measure was not log-transformed, because this would be inconsistent with the form of density dependence expected for large mammal populations (Fowler 1987; Saether et al. 2002). However, to correspond with the zero standardized mean for log-transformed predictors, one was subtracted to centre the relative abundance measure on zero.
For rainfall, either the annual rainfall total over the seasonal cycle (July–June) preceding the count, or the additive contribution of the separate wet season (October–March) and dry season (April–September) components, was incorporated into the regression model. Rainfall was standardized by dividing by regional means over 1960–99, and then loge transformed. The log-transformation was applied because it was expected that proportional variation in rainfall would be more relevant than the absolute amount of the variation in rainfall.
The two lagged effects considered were prior rainfall, potentially affecting the state of the vegetation, and past prey availability for lions, putatively affecting the abundance of this major predator. Other possible lagged factors were deemed likely to be less influential. Both of the lagged factors considered would have their effects manifested cumulatively over several years. Accordingly, to index the potential effect of prior rainfall on vegetation cover and composition, the running average of the standardized annual rainfall over the preceding 4 years, loge transformed, was used. The 4-year period was chosen to approximate the expected response time of the herbaceous layer. Extending or contracting the 4-year block by 1 year either way made little difference to the results.
The abundance of predators depends not only on the size of prey populations, but also on the extent to which these populations provided food for the predator concerned. Carcasses of animals dying constitute potential food, irrespective of whether these animals were killed by the predator or scavenged after they had died from other causes. Carcasses uneaten by predators were rarely encountered in KNP, except during severe drought years, or during outbreaks of the disease anthrax (affecting mainly African buffalo Syncerus caffer (Sparrman) and kudu; Bengis, Grant & de Vos 2003). All 10 ungulate species constituted prey for lions (Pienaar 1969), plus buffalo, not otherwise considered because their population was subjected to culling through much of the study period. The count totals for each prey species were transformed into an annual carcass biomass production through multiplying by the estimated mortality losses incurred by the adult, subadult and yearling segments and then by the species body mass. For buffalo, annual population totals (from rigorous counts conducted by helicopter), juvenile proportions and the numbers of animals removed annually by culling were available from 1969 onwards. Mortality losses for other ungulate populations were approximated prior to 1983, based on their status (increasing or decreasing) during these years in comparison with years showing similar trends during the study period. This adjustment accommodated the situation whereby buffalo provided relatively little food for lions, despite their high biomass density, during earlier years when their population was culled, restricting natural mortality losses. The derived index of annual prey carcass mass production relates specifically to changing food availability for lions. Accordingly impala were down-weighted because a large proportion of their deaths were due to other predators. Allowing for a delayed response in the lion population to changing food availability, the prevailing predator abundance was indexed from the prey carcass production over the preceding 4 years, standardized relative to regional indices for 1979 as a starting baseline, then loge transformed. This period was chosen to match that adopted for indexing prior rainfall, while recognizing that the lag in response of a lion population to changing prey availability could actually be somewhat longer. Other large predators, including spotted hyena Crocuta crocuta (Erxleben), leopard Panthera pardus L., cheetah Acinonyx jubatus (Schreber) and African wild dogs Lycaon pictus (Temminck), prey mainly on impala, although to some extent also on kudu, waterbuck and warthog, and on the young of larger ungulates (Pienaar 1969). These species were not expected to vary much in abundance, because the impala population showed little change.
Linear regression models for stage-specific survival rates were fitted by standard least squares regression in systat 8·0 (SPSS 1998) for the juvenile and yearling segments, and in the case of adults for estimates derived from the transformed count totals. For survival estimates derived from the original count data, proc autoreg in SAS 8·2 (SAS Institute 2001) was used in order to correct parameter estimates and associated confidence limits for autoregressive errors, using the Yule–Walker (or generalized least squares) procedure. Without this correction, standard error estimates are biased and the efficiency of parameter estimates reduced where serial autocorrelation in residual errors exists, as was the case for most of these estimates (Gallant & Goebel 1976). The residual autocorrelation in survival estimates was largely eliminated when the count data were transformed prior to making these estimates as described above.
The basic model included relative population abundance and seasonal rainfall components as the two predictors. After basic slope coefficients and their standard errors for these basic predictors had been established, either predation or prior rainfall was added to the model, and the effect of the third predictor assessed. Both lagged factors were not considered together, because they varied contemporaneously, i.e. during the period of the population declines prior rainfall was generally low and indexed past prey availability high. Hence their statistical effect was divided when they were included together in models. Predictors yielding slope coefficients opposite in sign to the hypothesized effects (e.g. a negative rather than positive effect of rainfall) were excluded from models, so as not to distort the effect of the other predictors. Models were fitted using separate survival estimates for each regional subpopulation, where data were adequate to support this subdivision, or otherwise to park-wide totals. Possible regional differences in slopes or intercepts were ignored. Outlying points were deleted if studentized residuals consistently exceeded 3·0 in various models and the data for these years seemed likely to be misleading. For some species yearlings were amalgamated with adults into a broadened ‘adult’ class, in order to maintain an adequate sample size for statistical power.
P-values relative to null hypotheses are not reported, as having little meaning for observational studies of this kind, especially given the uncertain effect of data transformations (cf. Anderson, Burnham & Thompson 2000). Slope coefficients with standard errors less than half of the effect were regarded as indicating strong relationships, and others having standard errors still less than the estimated effect as being worthy of note. The standardization of predictor variables enabled comparisons to be made among ungulate species in the relative values of the regression coefficients. Zero intercepts represented the survival rates expected for conditions with rainfall and population density at their mean values, over the relevant periods spanned, and the past prey abundance index at its 1979 level.
The support from the data for alternative linear regression models incorporating different predictors was then assessed more rigorously. For adult survival, models were fitted to the estimates derived from the interpolated count data, which were judged to be the most neutral for interpreting the relative effects of both direct and lagged predictors. The two-stage procedure proposed by Diggle (1988) and Wolfinger (1993) was followed. First, using restricted maximum likelihood in SAS proc mixed (SAS Institute 2001), it was established that models assuming independent rather than autocorrelated errors consistently received strongest support. Thereafter, alternative models were compared by maximum likelihood using SAS proc mixed, to identify the set of predictor variables that was best supported by the data. Maximum likelihood was used because restricted maximum likelihood filters out the contributions of the predictor variables, and so cannot be used to compare them. The strength of support was evaluated using the corrected Akaike Information Criterion (AICC), calculated as −2LL + 2K +{2K(K + 1)}/(n − K − 1), where LL = log likelihood, n = sample size and K = total number of parameters in the model, including regression coefficients, intercept and covariance parameters. Finally, alternative models were re-fitted by restricted maximum likelihood, to suppress bias due to the number of estimated parameters, which can be substantial if the number of estimated parameters is large and the sample size small. The single-factor or multiple-factor model best supported by the data was identified, together with other models with AICC values differing by less than four units from the best models, following the guidelines presented by Anderson et al. (2000) and Anderson & Burnham (2002). The best-fitting model was that giving the lowest AICC value, and ΔAICC values were calculated as deviations from this best model. As before, predictors with the opposite sign coefficient to that expected from the hypothesized mechanisms were interpreted as spurious outcomes of colinearity with other factors and thus omitted.
Results
temporal trends in predictor variables and survival estimates
Wet season rainfall was prevalently below average through the study period, while dry season rainfall remained extremely low after 1987. Prior rainfall over the preceding 4 years was above the long-term mean at the beginning of the study period, but dropped below average after 1984 (Fig. 1). Past prey abundance for lions over the preceding 4 years had doubled by 1987 compared with the 1979 baseline, and increased further after 1992.
Trends in the standardized indices of annual wet season and dry season rainfall, prior annual rainfall over the preceding 4 years (all standardized relative to long-term means), and past prey abundance for lions over the preceding 4 years (relative to a 1979 baseline).
Among the ungulate species that maintained high abundance levels, zebra and giraffe showed relatively little change in annual adult survival (Fig. 2). Somewhat greater variability in adult survival was evident for impala, while for wildebeest a trend was apparent, with adult survival lower after 1986 than it had been earlier. Juvenile survival varied somewhat more widely among these species than did adult survival, and tended to be lower during the later part of the study period, even for zebra. Yearling survival showed a similar pattern of variability to that of juvenile survival for zebra and impala, while for wildebeest yearling survival increased after 1985.
Temporal trends in annual survival estimates over the entire Kruger Park for the four ungulate species that persisted at high abundance, for the adult (i.e. animals older than 2 years, except for giraffe – squares), yearling (circles) and juvenile (triangles) segments. Also indicated associated changes in population census totals (smoothed lines without symbols).
All six of the declining species showed a substantial decrease in adult survival after 1986 (Fig. 3). A later recovery in adult survival rates was apparent after 1994, associated with a levelling in the population trends. Patterns of yearling and juvenile survival were less consistent. For kudu, juvenile survival was generally higher between 1984 and 1991 than it had been during the prior study up to 1984, while yearling survival was markedly lower than it had been earlier. Tsessebe showed a large and persistent reduction in juvenile survival after 1985. For most of these ungulate species, sharp drops in juvenile survival during the severe drought years of 1983 and 1992–93 were evident.
Comparative trends in annual survival estimates for the six ungulates species that declined substantially in abundance, symbols as in Fig. 1.
Four species were sufficiently abundant and widespread for regional distinctions in survival rates to be estimated, for both adults and juveniles. For impala, no regional differences were apparent. For kudu, survival estimates for both adults and juveniles varied in fairly close synchrony across the park, with juvenile survival depressed everywhere during the 1983 and 1992 droughts (4, 5). Estimates of the annual survival of adult kudu exceeded 1·0 for the South and Central regions during 1984–85, probably as an artefact of the undercount bias associated with the preceding drought period. Influences from movement are unlikely, because kudu are largely sedentary. The sharp drop in adult survival of kudu in Northern in 1991 was associated with an outbreak of anthrax in this region. For zebra, estimated adult survival exceeded 1·0 in Northern through 1984–86 and again in 1994, coupled with depressed adult survival estimates over the same periods in the Central region (Fig. 4). This probably reflected a population redistribution northwards at these times. Annual survival estimates suggest movement from Northern back into the Central region in 1993, and from the Central region into the South in 1992. These periods of apparent relocation were associated with drought conditions. Because of the distortion of adult survival estimates by such movements, data for the northern half of the park were excluded when relating adult survival for zebra statistically to environmental conditions. For wildebeest, adult survival estimates suggest a small population shift northwards over 1985–87, which was apparently reversed during the 1992–93 drought. For zebra, the survival rate of juveniles was significantly lower, and that of yearlings significantly higher, by about 0·1, in the South than elsewhere. For wildebeest, the survival of both juveniles and yearlings was significantly lower in the South, by almost 0·15, than elsewhere. Adult survival did not differ regionally for either of these species.
Differences in adult survival estimates for zebra, wildebeest and kudu between the Central (squares), South (circles) and Northern (triangles) regions of the Kruger Park (dotted line indicates survival rate of 1·0).
Differences in juvenile survival estimates for zebra, wildebeest and kudu between the Central (squares), South (circles) and Northern (triangles) regions of the Kruger Park.
correlations with abundance
A density feedback on adult survival was indicated for all four of the ungulate species that persisted at high abundance (Table 1). However, for wildebeest the statistical support was weak, while for zebra it became strongly supported only after count data were amalgamated across the park, to circumvent the effect of movements. The density influence on adult survival became manifested quite steeply above a threshold abundance level, especially for zebra and giraffe (Fig. 6). There was a weak indication of density dependence in juvenile survival for zebra, but no density relationship for juvenile wildebeest and giraffe, while for impala the density influence on juvenile survival appeared positive (Table 1).
| Species | Stage* | Data† | Regions‡ | n | Intercepts + SE | Slope coefficients + SE | R 2 | ||
|---|---|---|---|---|---|---|---|---|---|
| Relative abundance | Annual or wet season rainfall | Dry season rainfall | |||||||
| Zebra | Ad + SA | Orig | S, C | 22 | 0·910 ± 0·028 | 0·007 | |||
| Ad + SA | Trans | S, C | 22 | 0·927 ± 0·013 | −0·323 ± 0·182 | 0·136 | |||
| Ad + SA | Orig | Park | 13 | 0·941 ± 0·018 | −0·249 ± 0·182 | 0·157 | |||
| Ad + SA | Trans | Park | 13 | 0·953 ± 0·010 | −0·347 ± 0·099 | 0·026 ± 0·016 | 0·593 | ||
| Yearling | S, C, N | 27 | 0·535 ± 0·035 | 0·048 | |||||
| Juvenile | S, C, N | 27 | 0·466 ± 0·020 | −0·248 ± 0·176 | 0·075 | ||||
| Wildebeest | Ad + SA | Orig | S, C, N | 42 | 0·896 ± 0·022 | 0·074 ± 0·028 | 0·151 | ||
| Ad + SA | Trans | S, C, N | 39 | 0·893 ± 0·014 | −0·061 ± 0·050 | 0·046 ± 0·022 | 0·134 | ||
| Yearling | S, C, N | 41 | – | (0·120 ± 0·113) | (−0·118 ± 0·084) | (−0·102 ± 0·051) | – | ||
| Juvenile | S, C, N | 45 | 0·465 ± 0·022 | 0·125 ± 0·037 | 0·211 | ||||
| Impala | Ad + SA + Yrl | Orig | S, C, N | 32 | 0·882 ± 0·025 | −0·456 ± 0·135 | 0·140 ± 0·073 | 0·367 | |
| Ad + SA + Yrl | Trans | S, C, N | 32 | 0·827 ± 0·024 | −0·393 ± 0·149 | 0·235 | |||
| Juvenile | S, C, N | 32 | 0·594 ± 0·038 | (0·918 ± 0·247) | 0·166 ± 0·066 | 0·160 | |||
| Giraffe | Ad + SA | Orig | Park | 14 | 0·928 ± 0·017 | −0·611 ± 0·230 | 0·063 + 0·052 | 0·437 | |
| Ad + SA | Trans | Park | 14 | 0·926 ± 0·020 | −0·745 ± 0·309 | 0·513 | |||
| Juvenile–Yrl | Park | 14 | 0·639 ± 0·056 | (0·185 ± 0·168) | (−0·062 + 0·041) | 0·143 ± 0·052 | 0·386 | ||
- * Ad = Adults, SA = Subadults, Yrl = Yearlings.
- † Orig = original count data, Trans = transformed count data.
- ‡ S = South, C = Central, N = Northern, Park = entire Kruger Park.
Density dependence of adult survival estimates for zebra (park-wide), giraffe (park-wide) and impala (by regions).
Among the declining species, a negative density effect on adult survival was shown only by kudu and waterbuck, and rather weakly in both cases (Table 2). Adult survival appeared to be positively related to population abundance for warthog and sable. Juvenile survival also appeared positively density dependent for kudu, waterbuck and warthog, and unaffected by population abundance in the remaining species. For kudu, this was in contrast to the negative density dependence in juvenile survival that had been shown during the prior study.
| Species | Stage* | Data† | Regions‡ | n | Intercepts + SE | Slope coefficients + SE | R 2 | ||
|---|---|---|---|---|---|---|---|---|---|
| Relative abundance | Annual or wet season rainfall | Dry season rainfall | |||||||
| Kudu | Ad + SA | Orig | S, C, N | 37 | 0·846 ± 0·025 | 0·124 ± 0·078 | 0·118 ± 0·057 | 0·295 | |
| Ad + SA | Trans | S, C, N | 36 | 0·829 ± 0·019 | −0·068 ± 0·061 | 0·058 ± 0·045 | 0·078 ± 0·037 | 0·191 | |
| Ad + SA | Prior | 19 | 0·875 ± 0·017 | −0·121 ± 0·096 | 0·123 ± 0·046 | 0·040 ± 0·047 | 0·341 | ||
| study | |||||||||
| Yearling | S, C, N | 26 | 0·661 ± 0·157 | 0·073 ± 0·079 | 0·072 ± 0·057 | 0·150 | |||
| Yearling | Prior | 17 | 0·847 ± 0·026 | 0·162 ± 0·068 | 0·290 | ||||
| study | |||||||||
| Juvenile | S, C, N | 36 | 0·463 ± 0·024 | (0·236 ± 0·067) | 0·160 ± 0·057 | 0·123 ± 0·036 | 0·407 | ||
| Prior | 21 | 0·447 ± 0·023 | −0·301 ± 0·136 | 0·497 ± 0·066 | 0·162 ± 0·053 | 0·803 | |||
| Juvenile | study | ||||||||
| Waterbuck | Ad + SA + Yrl | Orig | S + C, N | 26 | 0·835 ± 0·062 | 0·123 ± 0·088 | 0·150 ± 0·057 | 0·303 | |
| Ad + SA + Yrl | Trans | S + C, N | 27 | 0·800 ± 0·026 | −0·104 ± 0·072 | 0·074 ± 0·061 | 0·128 ± 0·046 | 0·351 | |
| Juvenile | S + C, N | 28 | 0·505 ± 0·048 | (0·293 ± 0·123) | 0·152 ± 0·076 | 0·177 | |||
| Warthog | Ad + SA + Yrl | Orig | Park | 13 | 0·684 ± 0·128 | (0·494 ± 0·153) | 0·109 | ||
| Ad + SA + Yrl | Trans | Park | 13 | 0·737 ± 0·061 | (0·409 ± 0·099) | 0·112 ± 0·108 | 0·096 ± 0·063 | 0·266 | |
| Juvenile | Park | 12 | 0·509 ± 0·047 | (0·231 ± 0·114) | 0·274 ± 0·143 | 0·325 | |||
| Sable | Ad + SA + Yrl | Orig | Park | 13 | 0·806 ± 0·107 | (0·290 ± 0·135) | 0·107 ± 0·057 | 0·240 | |
| Ad + SA + Yrl | Trans | Park | 13 | 0·794 ± 0·031 | (0·267 ± 0·087) | 0·004 ± 0·051 | 0·092 ± 0·043 | 0·292 | |
| Juvenile | Park | 13 | 0·531 ± 0·060 | 0·166 ± 0·106 | 0·092 ± 0·075 | 0·294 | |||
| Tsessebe | Ad + SA + Yrl | Orig | N | 13 | 0·810 ± 0·037 | (0·195 ± 0·107) | 0·127 ± 0·069 | 0·050 ± 0·049 | 0·326 |
| Ad + SA + Yrl | Trans | N | 13 | 0·802 ± 0·025 | 0·114 ± 0·048 | 0·418 | |||
| Juvenile | N | 13 | 0·405 ± 0·047 | (0·233 ± 0·164) | (−0·105 ± 0·101) | 0·145 ± 0·070 | 0·280 | ||
| Roan | Juvenile | N | 8 | 0·659 ± 0·072 | 0·401 ± 0·141 | 0·644 | |||
- * Ad = Adults, SA = Subadults, Yrl = Yearlings.
- † Orig = original count data, Trans = transformed count data, Prior study = 1974–84 kudu study in two study areas.
- ‡ S = South, C = Central, N = Northern, S + C = South + Central combined, Park = entire Kruger Park.
correlations with seasonal rainfall
Among the four ungulate species that maintained high abundance levels, wet season rainfall had little or no influence on the survival of any population segment (Table 1). The positive correlation apparent for impala when survival beyond the juvenile stage was estimated from the untransformed count totals was probably an artefact of the effect of rainfall on visibility. In contrast, dry season rainfall positively influenced juvenile survival for wildebeest, impala and giraffe, as well as adult survival in the case of wildebeest. The survival of yearling wildebeest was surprisingly unrelated to rainfall. All population segments of zebra seemed impervious to any direct rainfall influence on their survival.
For the six declining species, adult survival was related to both seasonal components of rainfall for kudu, waterbuck and perhaps warthog, only to the dry season component for sable, and only to the wet season component for tsessebe (Table 2). Wet season rainfall appeared to have a stronger effect on adult survival estimated from the original counts than from the transformed data for kudu, waterbuck and tsessebe, all of which are brown species likely to be less visible in dry years. However, warthog, which are also brown, did not show this pattern. The survival of yearling kudu likewise apparently responded to both rainfall components. Juvenile survival was positively influenced by wet season rainfall for kudu, warthog, roan and, more weakly, sable. Dry season rainfall affected juvenile survival for all of these species, except roan antelope.
correlations with prior rainfall and past prey availability
Past prey availability for lions apparently had a stronger or at least similar effect to that of prior rainfall on adult survival for all six of the ungulate species that declined substantially in abundance (Table 3). Juvenile survival showed such a relationship only in the case of tsessebe. Among the stabilizing species, an effect of past conditions on adult survival was evident only for wildebeest, with prior rainfall showing a somewhat stronger effect than past prey abundance. Juvenile survival likewise appeared most clearly related to prior rainfall for wildebeest, giraffe and impala. Notably, the effect of these lagged predictors largely usurped that of dry season rainfall, because the latter remained low during the period when past prey abundance had been high and prior rainfall low. A further indication of the survival reductions associated with increased past prey abundance comes from the change in intercept coefficients when the latter predictor was included in models. Somewhat low survival rates for the adult segment of the declining species under mean rainfall and density conditions are indicated in Table 2, while the intercept values shown in Table 3 represent the projected survival rates with past prey abundance at its 1979 level.
| Species | Stage* | Data† | Regions‡ | n | Slope coefficients + SE | |||
|---|---|---|---|---|---|---|---|---|
| Intercepts + SE | Prior rainfall | Past prey availability | R 2 | |||||
| Stabilizing species | ||||||||
| Wildebeest | Ad + SA | Orig | S, C, N | 41 | 0·909 ± 0·016 | 0·220 ± 0·124 | −0·067 ± 0·042 | 0·219 |
| Ad + SA | Trans | S, C, N | 38 | 0·909 ± 0·013 | 0·260 ± 0·099 | −0·067 ± 0·038 | 0·302 | |
| Juvenile | S, C, N | 41 | 0·490 ± 0·021 | 0·520 ± 0·140 | 0·408 | |||
| Impala | Juvenile | S, C, N | 32 | 0·640 ± 0·052 | 0·450 ± 0·337 | 0·214 | ||
| Giraffe | Juvenile–Yrl | Park | 14 | 0·682 ± 0·058 | 0·943 ± 0·347 | −0·348 ± 0·178 | 0·633 | |
| Declining species | ||||||||
| Kudu | Ad + SA | Trans | S, C, N | 36 | 0·872 ± 0·034 | −0·081 ± 0·053 | 0·247 | |
| Ad + SA | Study + Trans | All | 56 | 0·876 ± 0·017 | 0·106 ± 0·088 | −0·077 ± 0·035 | 0·322 | |
| Yearling | S, C, N | 26 | 0·613 ± 0·027 | 0·366 ± 0·322 | −0·090 ± 0·092 | 0·199 | ||
| Yearling | Study + Trans | All | 46 | 0·785 ± 0·026 | 0·709 ± 0·137 | −0·303 ± 0·061 | 0·462 | |
| Waterbuck | Ad + SA + Yrl | Trans | S + C, N | 27 | 0·867 ± 0·053 | 0·310 ± 0·275 | −0·114 ± 0·080 | 0·406 |
| Warthog | Ad + SA + Yrl | Trans | Park | 13 | 0·737 ± 0·061 | 1·266 ± 0·414 | −0·349 ± 0·142 | 0·593 |
| Sable | Ad + SA+Yrl | Orig | Park | 13 | 1·037 ± 0·084 | 0·750 ± 0·350 | −0·395 ± 0·114 | 0·520 |
| Ad + SA+Yrl | Trans | Park | 13 | 1·001 ± 0·056 | 0·727 ± 0·297 | −0·333 ± 0·080 | 0·760 | |
| Tsessebe | Ad + SA + Yrl | Orig | N | 13 | 1·042 ± 0·104 | 0·446 ± 0·302 | 0·277 ± 0·119 | 0·581 |
| Ad + SA + Yrl | Trans | N | 13 | 0·980 ± 0·117 | 0·341 ± 0·200 | −0·164 ± 0·087 | 0·582 | |
| Juvenile | N | 13 | 0·816 ± 0·109 | 0·774 ± 0·264 | −0·514 ± 0·131 | 0·717 | ||
| Roan | Ad + SA + Yrl | Orig | N | 8 | 1·195 ± 0·304 | 0·625 ± 0·632 | 0·558 ± 0·409 | 0·271 |
| Ad + SA + Yrl | Trans | N | 8 | 1·210 ± 0·189 | 0·666 ± 0·348 | −0·609 ± 0·256 | 0·485 | |
- * Ad = Adults, SA = Subadults, Yrl = Yearlings.
- † Orig = original count data, Trans = transformed count data, Study + Trans = data from prior study over 1974–84 combined with transformed count data after 1984.
- ‡ S = South, C = Central, N = Northern, S + C = South + Central combined.
The lagged influence of past conditions is also evident when survival rates relative to current rainfall are considered separately for the pre-1987 and post-1986 periods (Fig. 7). Survival rates for adult and yearling kudus were significantly lower relative to the product of the standardized seasonal rainfall components after 1986 than they had been earlier (ancova: F54,1 = 13·42, P = 0·001 for adults; F45,1 = 47·96, P < 0·001 for yearlings). For juvenile kudus, the difference between periods was small and insignificant (F56,1 = 2·45, P = 0·124). After 1986, the survival of adult and yearling kudus was unaffected by rainfall variation. The survival intercept for adult kudu was consistent with that observed in the prior study only when the effect of past prey abundance for lions was included in the model (Tables 2 and 3), while the latter made no difference to the survival of juvenile kudu. The survival of yearling kudu was substantially higher before 1985 than subsequently, possibly as a result of the putative increase in lion abundance during the later period (Table 3).
Rainfall dependence of stage-specific survival estimates for kudu, relative to the product of the standardized annual wet season and dry season rainfalls for each region, plotted on a log scale. Regions differentiated are South (circles), Central (squares) and Northern (triangles). Open symbols indicate pre-1987, closed symbols post-1986. Data from the prior study over 1974–84 distinguished by study areas Pretorius Kop (crosses) and Tshokwane (asterisks). Narrow lines indicate linear regression relationships for the pre-1987 period, broad lines linear regression relationships fitting the post-1986 data.
model fitting
Among the species stabilizing at high abundance, relative abundance was supported as the primary influence on adult survival for zebra, impala and giraffe (Table 4). A lower survival among adult zebra in the South than in the Central region was also indicated. Prior rainfall was the primary factor affecting the survival of adult wildebeest, with a possible additional influence from annual or seasonal rainfall. For zebra and wildebeest, the influence of prior rainfall on juvenile survival was more strongly supported than that of current annual or seasonal rainfall, although for zebra the discrimination among models was poor because relationships were statistically weak. For zebra, an effect of prior rainfall on juvenile survival was strongly manifested in the Northern section (the driest region), but with no effect apparent in the Central region where zebra were most abundant. For wildebeest, the effect of prior rainfall on juvenile survival was consistent across regions, despite the regional distinctions in mean survival. An additional influence of dry season rainfall was supported only for Northern. For yearling zebra survival, strongest support was for an effect of past prey availability for lions, with annual rainfall weakly supported as an additional factor, but with the effect of the former barely apparent in the Central region.
| Species | Stage | n | Single predictor | ΔAICC | Two or more predictors | ΔAICC |
|---|---|---|---|---|---|---|
| Stabilizing species | ||||||
| Zebra | Ad + SAd | 22 | Abundance + region | 0 | ||
| Yrl | 27 | Predation | 0 | Predation + annual rain | 2·1 | |
| Annual rain | 2·5 | |||||
| Dry seas. rain | 2·6 | |||||
| Juvenile | 27 | Prior rain * region | 0 | |||
| Abundance* region | 2·0 | |||||
| Wildebeest | Ad + SAd | 39 | Prior rain | 0 | Prior rain + dry seas. rain | 1·2 |
| Prior rain + annual rain | 1·4 | |||||
| Juvenile | 45 | Prior rain + region | 0 | Prior rain + dry seas. rain | 1·1 | |
| Impala | Ad + SAd + Yrl | 32 | Abundance | 0 | Abundance + dry seas. rain | 1·7 |
| Abundance + annual rain | 1·9 | |||||
| Juvenile | 35 | Dry seas. rain | 0 | Dry seas. rain + prior rain | 0·7 | |
| Prior rain | 2·2 | Prior rain + annual rain | 3·4 | |||
| Giraffe | Ad + SAd | 14 | Abundance | 0 | Abundance + annual rain | 3·3 |
| Abundance + predation | 3·3 | |||||
| Juvenile + Yrl | 14 | Dry seas. rain | 3·1 | Dry seas. rain + prior rain | 0 | |
| Prior rain | 3·2 | |||||
| Predation | 3·4 | |||||
| Declining species | ||||||
| Kudu | Ad + SAd | 36 | Annual rain | 0 | Annual rain + predation | 0·8 |
| Dry seas. rain | 0·9 | Wet seas. rain + dry seas. rain | 1·2 | |||
| Predation | 2·9 | Dry seas. rain + abundance + predation | 1·5 | |||
| Annual rain + prior rain | 1·9 | |||||
| Wet seas. rain + dry seas. rain + abundance | 2·6 | |||||
| Yrl | 26 | Annual rain | 0 | Annual rain + prior rain | 0·1 | |
| Dry seas. rain | 2·1 | Annual rain + predation | 1·6 | |||
| Predation | 3·3 | Dry seas. rain + predation | 3·1 | |||
| Wet seas. rain + dry seas. rain | 3·2 | |||||
| Juvenile | 36 | (Dry seas. rain) | (5·1) | Dry seas. rain + wet seas. rain | 0 | |
| Dry seas. rain + wet seas. rain | 2·5 | |||||
| Waterbuck | Ad. + SAd + Yrl | 27 | Dry seas. rain | 1·0 | Dry seas. rain + abundance | 0 |
| Annual rain | 2·8 | Wet seas. rain + dry seas. rain | 0·7 | |||
| Dry seas. rain + abundance + predation | 1·2 | |||||
| Wet seas. rain + dry seas. rain + abundance | 1·4 | |||||
| Dry seas. rain + predation | 2·1 | |||||
| Juvenile | 28 | Dry seas. rain | 0 | Wet seas. rain + dry seas. rain | 2·2 | |
| Dry seas. rain + predation | 2·2 | |||||
| Warthog | Ad + SAd + Yrl | 12 | Predation | 0 | Predation + annual rain | 2·6 |
| Annual rain | 0·7 | Prior rain + annual rain | 3·6 | |||
| Prior rain | 1·0 | |||||
| Juvenile | 13 | Annual rain | 0 | Annual rain + prior rain | 0·7 | |
| Predation | 1·4 | Wet seas. rain + dry seas. rain | 2·0 | |||
| Sable | Ad + SAd + Yrl | 13 | Predation | 0·9 | Predation + dry seas. rain | 0 |
| Juvenile | 13 | Annual rain | 0 | Annual rain + abundance | 2·3 | |
| Wet seas. rain + dry seas. rain | 3·0 | |||||
| Tsessebe | Ad. + SAd + Yrl | 13 | Annual rain | 1·6 | Annual rain + predation | 0 |
| Annual rain + prior rain | 0·7 | |||||
| Juvenile | 13 | Predation | 0·3 | Predation + dry seas. rain | 0 | |
| Prior rain | 3·7 | Predation + annual rain | 3·0 | |||
| Prior rain + annual rain | 3·3 | |||||
| Roan | Ad + SAd + Yrl | 8 | Predation | 0 | ||
| Prior rain | 1·5 | |||||
| Juvenile | 8 | Annual rain | 0 | |||
- Ad = adult, SAd = subadult, Yrl = yearling; seas. = season. *Indicates an interaction with region, i.e. slope coefficients differ regionally.
Among the declining species, rainfall components were supported as the primary influence on adult survival for kudu, waterbuck and tsessebe, while past prey availability for lions appeared to be the predominant influence on the survival of warthog, sable and roan (Table 4). However, judged by the small AICC deviations, past prey availability needed to be recognized as a possible additional factor for tsessebe, kudu and perhaps waterbuck. Dry season rain additionally contributed strongly for sable. Past predator food consistently received stronger support than that of prior rainfall for these declining populations. A density feedback was supported for adult waterbuck and, rather weakly as a third factor, for adult kudu. The best supported models incorporated one or another rainfall component for all six of these species with respect to juvenile survival, as well as for yearling survival among kudu, while past predator food was apparently the main influence on the survival of juvenile tsessebe.
Discussion
Eberhardt (2002) maintained that, among large, long-lived mammals, density dependence first affects juvenile survival and only last adult survival. Gaillard et al. (1998) reported that, for large mammalian herbivores, juvenile survival is generally sensitive to both population density and environmental variability, while adult survival is resistant to and perhaps canalized against such influences (Gaillard & Yoccoz 2003). Our findings seem discrepant with these generalizations in several ways. Density dependence was manifested in adult survival for several of these species, but not in juvenile survival with the possible exception of zebra. Adult survival as well as juvenile survival appeared sensitive to environmental variability, particularly in dry season rainfall. Besides immediate effects, adult but not juvenile survival was subject to additional lagged influences possibly arising from past prey availability for the principal predator, or from the effects of past rainfall on habitat conditions, among most of the declining species. How can the apparent differences in demographic responses of these African ungulate populations to the proxy variables representing possible underlying mechanisms be explained?
juvenile survival
For almost all of the ungulate populations in KNP, juvenile survival seemed to be particularly responsive to rainfall occurring during the dry season months, potentially affecting green leaf retention through this critical period. A similar pattern was reported for the Serengeti wildebeest population (Mduma et al. 1999). However, for the KNP the correlation was with rainfall during the dry season prior to the birth of these juveniles, and thus inferred to occur via the nutritional status of the mother, affecting the birth mass and hence subsequent viability of her offspring (Festa-Bianchet et al. 1997; Gaillard et al. 2000). However, dry season rainfall across the KNP remained consistently low for seven consecutive years after 1987 (Ogutu & Owen-Smith 2003), and the resultant serial correlation in the rainfall does not permit the effect to be ascribed narrowly to any specific year. The lack of a density influence on juvenile survival, despite clear evidence of resource limitation from rainfall relationships, could be because juvenile/adult female ratios were assessed part-way through the dry season, before competition for diminishing resources had become extreme. However, a density effect should then have been manifested in subsequent survival to the yearling stage, but this was not apparent for the three ungulate species with adequate data to analyse yearling survival separately. The surprising feature is that, for the six declining species, there was no compensatory increase in juvenile survival to counteract the reduction in population abundance brought about by reduced adult survival.
Although juvenile survival was apparently unaffected by density variation over the study period, observed survival rates in this stage was generally considerably lower than the maximum rates of about 80% or more theoretically possible under conditions of low density and high food availability. Furthermore, juvenile survival estimated from mother/offspring ratios incorporates an assumption that the death of a mother implies the death also of her calf. The sensitivity of juvenile losses to variability in resource production and quality, affecting both maternal nutrition and post-natal survival of offspring, perhaps overwhelmed any effect of population abundance on intraspecific competition for these resources. Hence, overall the findings from this study are not deeply discordant with previous generalizations about survival rates in the juvenile stage.
adult survival
Gaillard et al. (1998, 2000) emphasized constancy in the survival of prime-aged females in particular, and recognized that ageing females could be more responsive to environmental variability. Our estimates of adult survival were for the entire adult segment, including males and senescent animals of both sexes. Part of the reduction in overall survival within the broad adult class could arise from a shift in age structure towards older animals with lower survival rates, particularly during the phase transition from increasing to stabilizing trends (Festa-Bianchet, Gaillard & Côté 2003). However, for this effect to be substantial requires reduced recruitment, coupled with high survival through the prime years, otherwise ageing adults constitute only a small proportion of the adult class. For kudu, mortality among prime-aged females averaged about 8–9% per year prior to 1985, compared with 20% among old females (Owen-Smith 1990), with additive predation (i.e. deaths uninfluenced by nutrition) being the most likely cause of deaths during the prime stage. Mortality across the entire adult female class averaged 13% per year prior to 1987, but 21% per year thereafter, with similar or greater changes apparent for other declining species (Owen-Smith & Mason 2005). The magnitude of the reduction in adult survival makes it unlikely that the increase in mortality was restricted to the senescent stage.
For both African buffalo and migratory wildebeest in the Serengeti ecosystem, population growth was halted by a density-dependent increase in mortality in the adult segment (Sinclair 1974; Mduma et al. 1999). The most prevalent cause of the adult mortality seemed to be malnutrition among older animals, despite the abundance of predators. Nevertheless, a substantial proportion of the adults found dead were in prime age classes, suggesting additive predation at least in this segment.
Ogutu & Owen-Smith (2003) found that the declines shown by several ungulate populations in the KNP after 1986 were associated primarily with an extreme and persistent reduction in dry season rainfall. Our analysis indicates that adult as well as juvenile survival was adversely affected by low dry season rainfall for some of these species. However, the most pervasive effect on adult survival among these species seemed to be associated with past prey availability for lions. This pattern is consistent with heightened predation risk as a consequence of a numerical increase by this major predator. The effect of the putative increase in lion abundance an adult survival cannot be clearly separated from that of dry season rainfall statistically, because indexed predator abundance was highest during the period when dry season rainfall was lowest. Nevertheless, a similar effect on juvenile survival from both predictors was not observed. The vulnerability of adult ungulates to predation could be explained by risk-prone foraging in circumstances where food shortfalls were experienced, as suggested by Sinclair & Arcese (1995). Furthermore, the risk of predation from minor deficiencies in nutrition is amplified in the presence of predators (Sinclair 1974). Cohort effects arising from conditions in the year of birth (Albon, Clutton-Brock & Guinness 1987, Albon, Clutton-Brock & Langvatn 1992) could possibly have had a further effect on the susceptibility of prime-aged adults to mortality. However, it seems improbable that the more-than-twofold increase in prey availability would not result in some increase in the lion population, and that elevated lion abundance would not bring a rise in the risk of predation, all else being equal.
Smuts (1978) suggested that wildebeest and zebra were less vulnerable to predation under conditions of low rainfall, as a consequence of the effect of rainfall on vegetation cover. For wildebeest, changes in survival among both adults and juveniles appeared to be positively related to prior rainfall conditions, the opposite of this suggested effect. This was a consequence of the decline by this species towards the end of the study period, after rainfall had remained low for several successive years. For zebra, impala and giraffe, the main influence on adult survival was apparently a sharp, density-related reduction in adult survival, curtailing the earlier population increase shown by these populations. The steepness of the density effect seems inconsistent with food limitation. It possibly reflects switching by lions towards these principal prey species following their increase in abundance, coupled with decreased availability of alternative prey species. For the two largest ungulates in the data set, zebra and giraffe, for which lion predation tends to be concentrated somewhat more on young animals than on adults (Pienaar 1969; Mills & Biggs 1993), there were weak indications of an effect of predator abundance on the survival of immature stages.
However, interpretations are restricted by the limited period over which data on population structure were available for all of these species except kudu and wildebeest. Data from the phase of population increase prior to 1987 covered just 4 years, one of them an extreme drought period. Kudu survival rates showed no indication of changing predation pressure during the prior study spanning 1974–84 (Owen-Smith 1990), but this had clearly changed after 1986 (Owen-Smith 2000).
concluding remarks
The demographic patterns strongly implicate heightened predation on the adult segment as a consequence of an increase in the abundance of lions as contributing to the progressive declines shown by several ungulate populations in the KNP after 1986, rather than merely a response to rainfall deficiencies affecting food availability. Additionally, the predation risk probably amplified the vulnerability of these species to mortality as a consequence of nutritional shortfalls. The ungulate species maintaining high abundance over the study period seemed resistant to the effects of rainfall on food supplies, and moreover largely drove the changes in predator abundance through fluctuations in their relative availability as prey. A corresponding effect on juvenile survival was lacking because lions prey largely on the adult segment of most of these species. Hence the survival patterns of different population segments respond in different ways to changing resource supplies, predation pressures and density levels. Findings extend past generalizations about the demographic processes underlying population dynamics, and reveal circumstances in which adult as well as juvenile survival responds to environmental variability, perhaps even for prime-aged females.
Acknowledgements
We are indebted to South African National Parks for making the census and demographic data, and acknowledge the effort that their staff put into collecting the data base than enabled this paper to be written. The manuscript was much improved by critical comments on early drafts by J.-M. Gaillard, M. Festa-Bianchet, M.G.L. Mills and two anonymous reviewers. This paper also benefited during its preparation from other members of the Working Group on ‘Dynamics of large mammalian herbivores in changing environments’, supported by the National Centre for Ecological Analysis and Synthesis with funds from NSF (grant no. DEB-94-21535), the University of California at Santa Barbara, and the State of California.
