Abstract
Carnivores that exhibit fission–fusion social organization can adapt group sizes to prevailing social and ecological conditions. This study focuses on social organization of African lions (Panthera leo) in the Okavango Delta, a seasonally flooded wetland. We used generalized estimating equations and generalized linear mixed models to estimate the effects of flooding, as well as prey availability and intraspecific competition on group sizes of lions. During years of high flood, total lion pride sizes as well as reproductive rates declined. Prides showed extensive overlap in annual home ranges, likely as a result of habitat saturation at high densities, and pride sizes were not limited by prey availability. At the subgroup level, the number of attending cubs was the most consistent predictor of subgroup size of adult females. For subgroups without cubs, higher numbers of neighbors in adjacent, competing prides resulted in larger subgroups in focal prides, likely to maintain numerical advantage in inter-pride encounters. Larger subgroups were also formed in response to greater availability of large prey, as successfully hunting large prey requires a greater degree of cooperation. In the southwestern Okavango Delta, competition for space resulting from changing flooding regimes is a greater limiting factor for total pride sizes than food availability.
Many social carnivores live in fission–fusion societies, where individuals share a common territory, but may be found alone or in small subgroups (Smith et al. 2008). This fluid social organization allows for flexibility, where individuals can maximize individual fitness by remaining with or leaving a group based on environmental or social circumstances (Gittleman 1989; Wrangham et al. 1993; Chapman et al. 1995). The sizes of these groups are influenced intrinsically by social factors (Van Orsdol et al. 1985; Higashi and Yamamura 1993; Smith et al. 2008) as well as extrinsically by biotic and abiotic factors in the environment (Mills 1978; Bowen 1981; Feore and Montgomery 1999). Because social systems can influence population responses to ecological changes and influence the genetic makeup of populations (Spong 2002; Packer et al. 2005), understanding how carnivores adapt their social organization to local conditions can reveal the degree to which populations can adapt to changing environmental and anthropogenic influences (Patterson 2007).
African lions (Panthera leo; hereafter, lions) are the only social felids and live in fission–fusion social groups known as prides, which may fragment into smaller subgroups (Schaller 1972). At both levels of social organization, group sizes are influenced by a combination of extrinsic and intrinsic factors (Caraco and Wolf 1975; Turner and Antón 1997; Mosser and Packer 2009). Initially, it was believed that the benefit gained through cooperative hunting was the main driver behind lion social organization (Schaller 1972; Macdonald 1983; Turner and Antón 1997). However, although total pride size may be related to prey availability (Bertram 1973; Hanby and Bygott 1979; Van Orsdol et al. 1985), sizes of subgroups are often larger than would be predicted based on maximizing individual food intake (Caraco and Wolf 1975; Packer 1986). Success of hunting large and difficult-to-capture prey increases with higher numbers of participants (Schaller 1972; Packer and Ruttan 1988); thus, the availability of large prey relative to smaller prey may influence sizes of subgroups to some extent.
More recently, social factors, such as the protection of cubs against unfamiliar males or territorial defense, have been posited as the main drivers of group living in lions (Packer et al. 1990; Mosser and Packer 2009). Because adult females crèche cubs and cooperate to rear them, the protection of cubs in subgroups is often central to grouping patterns within a pride (Pusey and Packer 1994). Grouping patterns also may be influenced by intraspecific competition, where larger subgroups form in order to have a numerical advantage in territorial disputes involving neighboring prides (Packer et al. 1990; Grinnell et al. 1995). However, an increasing number of lions in neighboring prides may limit overall pride size by negatively influencing reproductive rates (Mosser and Packer 2009; Miller and Funston 2014).
Aside from biotic factors, abiotic factors such as habitat structure and even landscape characteristics exert considerable influence on the social organization and ecology of lions (Mosser 2008; Celesia et al. 2009). For example, there is a clear link between dry season precipitation and the number of associating adult females over a wide range of habitats (Celesia et al. 2009), and in savannas, prides with year-round access to surface water are often larger and more successful (Mosser and Packer 2009). In Hluhluwe-iMfolozi Park, South Africa, pride size is dependent on vegetation structure across the landscape; smaller prides occur in thick vegetation, which is more conducive to ambush hunting and effective concealment of cubs, while the opposite occurs on the open plains (Trinkel et al. 2007).
These studies, however, focus on arid and mesic environments, whose ecological functioning is largely driven by rainfall and surface water availability (Elliot and McTaggart Cowan 1978; Celesia et al. 2009; Valeix et al. 2010). In contrast, in wetlands such as the Okavango Delta, Botswana, the annual flood pulse is the main ecological driver of change (Ramberg et al. 2006; Murray-Hudson et al. 2014). This flood pulse exerts an extensive influence on the landscape and its inhabitants, firstly, by reducing the area of available dry land, thereby confining both predators and prey to small islands, and secondly, by providing water and a green flush of grazing at the height of the dry season (Crawshaw and Quigley 1991; Fynn et al. 2015). Lions in wetlands therefore must continually adapt to varying flooding patterns to maximize individual fitness, which may influence several aspects of their ecology and behavior, including social organization, and ultimately, population size.
Flooding patterns in wetlands offer a unique set of circumstances in which habitat naturally expands and contracts on a cyclical basis. Our aim, using the Okavango Delta as a case study, was to estimate the extent to which flooding cycles influenced social organization of lion prides, while accounting for conventional social and ecological influences on group sizes of lions. We used detailed information collected on lion group size and composition, spanning a 7-year period from low to high flood cycles in the Okavango Delta, Botswana, to assess the effects of: 1) the maximum annual flood extent, 2) social factors, and 3) prey availability on lion pride sizes, subgroup sizes, and reproductive rates. We expected that 1) higher flood levels would lead to increased competition for space, resulting in smaller pride sizes and subgroup sizes, and lower reproductive rates; 2) intraspecific competition, in the form of number of adult females from neighboring prides with adjacent home ranges, would negatively affect pride size by decreasing reproductive rates, but encourage larger subgroup sizes for territorial defense where cubs are absent; 3) number of cubs would have a positive relationship with the number of adult females associating in subgroups due to crêching behavior; and 4) prey abundance would positively influence pride size, but be inversely related to subgroup size.
Materials and Methods
Study area
The Okavango Delta, situated in northwestern Botswana, is one of the biggest freshwater deltas in the world, covering approximately 14,000 km2 (McCarthy and Ellery 1998; Fig. 1). Each year, seasonal rains, which fall between October and April in the Angolan highlands, flow down through the Caprivi Strip of Namibia and into Botswana, terminating in a unimodal, annual flood pulse (Ramberg et al. 2006; Murray-Hudson 2009). Flood waters typically arrive from early April and reach their full extent in late August or September (Gumbricht et al. 2004). Besides the rainfall patterns in Angola, the extent and duration of the annual pulse are variable and are dependent on a number of other factors, including local seasonal rainfall and the attributes of the previous flood (Murray-Hudson 2009). These conditions vary not only annually, but on a multi-decadal scale that can span 30–40 years of high flood or low flood conditions (Murray-Hudson et al. 2014).
Study area situated in the southwestern Okavango Delta, Botswana. The southwestern floodplains are characterized by high variability in interannual flooding. All 5 African lion (Panthera leo) prides were followed in this area.
The study area is situated in wildlife management concessions NG29 and NG30, between 19°33′S and 19°53′S, and 22°48′E and 23°06′E, southwest of Moremi Game Reserve. This area is characterized by highly variable flooding patterns, with considerable fluctuations in flood levels with each pulse and large interannual variation in flood extent (Wolski and Murray-Hudson 2006a). From 1997, the Xudum channel, which flows through the center of the study area, started receiving a sudden increase in inflow, most likely as a result of permanent vegetation changes upstream (Wolski and Murray-Hudson 2006a). As a result of the redirected flow, increases in water levels in the Xudum distributary were accompanied by a simultaneous decrease in flow in other nearby distributaries (Wolski and Murray-Hudson 2006a). Over this same period, the Okavango Delta entered into a “wetting” phase, characterized by periodic higher volumes of inflow. As a result, this area became increasingly flooded over the study period (Wolski and Murray-Hudson 2006a; Murray-Hudson et al. 2014).
Data collection
Detailed data on lion social organization were collected for 5 prides in the study area between 1997 and 2004. During the study period, at least 1 lioness in each pride was fitted with a VHF (very high frequency) radiocollar. Immobilizations were conducted by registered veterinarians and all protocols adhered to American Society of Mammalogists guidelines for study of live animals (Sikes et al. 2016). Collars were necessary to allow for easier and more consistent follow-up on prides, and an attempt was made to locate each pride once per week.
Prides were defined as “groups of resident lionesses with their cubs, as well as the attending males, which share a pride area and interact peacefully” (Schaller 1972). Individuals in each pride were identified using whisker spot patterns (Pennycuick and Rudnai 1970) and assigned pride memberships based on Schaller’s (1972) description. Subgroups of prides were defined as the number of individuals of the same pride encountered together at any one time, separated by a distance of 200 m or less. This distance of separation is based on the assumption that all pride members present would participate in group activities such as hunting or feeding and thus have some degree of direct social interaction (Cavalcanti and Gese 2009). Data collected during observations included pride identity, Global Positioning System (GPS) location, and group size and composition (sex and age of individuals present). For older lions not born during the study period, ages were estimated using body size, nose color, and tooth wear (Smuts 1982; Whitman and Packer 2007). Individuals were classified as cubs if younger than 2 years old, subadults if between 2 and 4 years old, and adults if older than 4 years (Smuts 1982). Only observations where all members of the subgroup were identified by sex and categorized into an age group were used in the analysis.
Surveys to estimate prey availability were conducted along transects on established roads in the wildlife management concessions NG30 and NG29. Transects were driven in the early mornings and late afternoons, when large herbivores are most active. During surveys along transects, potential herbivore prey animals were counted on either side of the road for a perpendicular distance of up to 200 m, and the species, herd size, GPS location, and time of day were recorded. Data were collected in the dry season between August and November, when flood water had receded enough to allow access to survey areas. Data on flooding extent were obtained from the Okavango Research Institute, Maun, Botswana.
Data analysis
Pride sizes were estimated on an annual basis, in 2 different ways. The first measure included the number of adult females only, because adult females represent the most stable estimation of pride size (Bertram 1973). The second was total pride size, which included both adult females and all subadults, because subadults contribute to pride success by assisting with hunting and compete for resources within the pride. Subgroup sizes were also classified in 2 ways: adult females only, and adult females plus subadults (total subgroup size). While crèching of cubs is the most important factor in determining subgroup size of adult females in the Serengeti (Pusey and Packer 1994), the tolerance of subadults by adults can be more dependent on other variables, such as prey availability, carcass size, or group territoriality (Van Orsdol et al. 1985; Trinkel et al. 2007). Therefore, for subgroup size including adult females only, communal rearing (expressed as the number of dependent cubs) was used as an additional predictor, and for subgroup analysis including adults and subadults, all observations including cubs were removed from the data to exclude the effect that cubs would have on the number of associating adults.
We modeled the effects of 2 ecological and 1 social covariate on lion pride size and subgroup size. Per capita prey availability, maximum annual flood extent and an index of intraspecific competition were determined for each of the 5 prides for years 1997–2002 and 2004. No data were collected in 2003.
To account for differences in covariates between prides, home ranges of prides were calculated and the boundaries were used as cutoff points to divide herbivore transects between each pride, and to estimate how much dry land was available to each pride at the maximum annual flood extent. Home ranges were calculated using all GPS location data collected from lion sightings. Due to the limitations imposed by using VHF tracking to locate animals, location data were pooled from all of the years in the study period to construct long-term home ranges. This is a more robust representation of area used per pride, as although ranges vary seasonally, there is strong annual fidelity in home range use (Schaller 1972). In Kafue National Park, Zambia, interannual site fidelity was high despite seasonal shifts in range use in a similarly seasonally flooded environment (Midlane 2013), and so we are confident that range use was represented accurately in this study. Home ranges were calculated in ArcView GIS 3.2 (ESRI 2002), using 90% kernel density estimates (Börger et al. 2006; Loveridge et al. 2009; Midlane 2013) and a href smoothing factor (Hemson et al. 2005).
We estimated abundance of preferred lion prey per pride home range with the kilometric abundance index (KAI—Vincent et al. 1991). Units of the KAI were number of observed animals per kilometer of transect (Vincent et al. 1991). This index can reliably detect trends in animal abundance from 1 year to the next (Vincent et al. 1991; Maillard et al. 2001) and was used as an indication of the rate at which lions come across prey while moving through their home range (Kiffner et al. 2009; Loveridge et al. 2009). Transects were apportioned to each study pride using the boundaries of lion home ranges as cutoff points.
Given the relationship between biomass of preferred prey and lion numbers (Hayward et al. 2007), data collected during the study period on successful and attempted hunts (n = 129) were used to assess local prey preference of the study prides. Animals that made up ≤ 1% of the diet were excluded from all analyses. To account for the influence of various prey sizes on group behavior, prey species were categorized into small (< 100 kg), medium (100–350 kg), and large (350–1,000 kg) prey. KAIs were calculated for each of the preferred prey species, and then converted to kilometric biomasses by multiplying the KAI by the average weight of females for each species (Loveridge et al. 2009). Each species’ kilometric biomass was then multiplied by 0.75 to account for young and subadult animals in the population (Schaller 1972; Hayward et al. 2007; Midlane 2013). The kilometric biomasses for each species were then added to calculate total biomass of each of the prey sizes available to lions in their specific home ranges each year per kilometer, resulting in an index of total kilometric prey abundance. Lastly, the total kilometric prey abundances for each prey size were divided by total pride size to reach a per capita prey abundance index. In group-living carnivores, individuals compete for both live prey and for their share of carcasses (Watts and Holekamp 2009). A per capita prey abundance index is therefore a more accurate indication of prey abundance when dealing with groups of social carnivores (Watts and Holekamp 2009).
Intraspecific competition was defined as the total number of adult female lions from neighboring prides whose home ranges overlapped with the focal pride’s home range (Mosser and Packer 2009). Van der Waal et al. (2009) considered neighbors to be the number of prides whose core territories (50% kernels) were within 3 km of one another. This holds true for the 5 prides in the present study. For simplicity, only adult females were considered, because these were the members of the pride most likely to engage in territorial conflict or take over kills (Heinsohn et al. 1996; Mosser and Packer 2009). Because male coalitions often had tenure over more than 1 pride at the same time during the study period, they were excluded from the analysis.
To determine the area of dry land available at maximum flood extent each year (hereafter, land availability), satellite thematic mapping TM4 and TM+ imagery depicting maximum flood levels for the years 1997–2002 were obtained from the Okavango Research Institute (Wolski and Murray-Hudson. 2006b). Each image consists of 28 × 28 m pixels, which were classified as either flooded or dry based on methods of Wolski and Murray-Hudson (2006b). Home ranges of lions were superimposed on flood maps using ArcMap 10.1 (ESRI 2012), and the number of dry pixels in each home range was used to delineate land availability for each year. For the year 2004, no satellite imagery was available, and consequently MODIS imagery depicting flooding extent was used as a substitute. Pixel size is larger (250 × 250 m), but Murray-Hudson et al. (2014) show that MODIS imagery can be used as a comparable substitute for finer-scale TM imagery.
Statistical analysis
Prior to statistical modeling, variance inflation factors (VIFs) were calculated for each covariate used in the study. All VIFs were < 5, indicating no collinearity between covariates; thus, all covariates were retained in the global models. Because of differences in order of magnitude, and to make results directly comparable, covariates were standardized prior to modeling by centering and dividing by 2 SDs (Schielzeth 2010). Generalized estimating equations (GEEs) with a Poisson error distribution were used to estimate the relationships between per capita prey abundance, competition with neighbors, land availability, and group sizes (Liang and Zeger 1986; Ballinger 2004). GEEs are well suited for analyzing longitudinal data, and account for dependence between repeated observations of the same prides over time by including a correlation structure in the model (Ballinger 2004). To account for nondependence between observations of group sizes, an autoregressive correlation structure was specified for each model on group sizes, as observations closer in time are likely to be more correlated than those farther apart (Zuur et al. 2012).
To examine the effects of social and ecological factors on size of lion prides, modeling was broken down into 3 steps. Firstly, we examined the effects of the ecological factors prey availability and land availability on number of adult females and total pride sizes. However, as survival of cubs and recruitment of adults and subadults occurs in 2-year time steps due to the selected age divisions, group sizes may be influenced not only by extant flood conditions, but by preceding flood conditions. We thus first examined different combinations of extant and time-lagged flood conditions on pride size by constructing the global model and replacing land availability in the current year with land availability indices from preceding years, and land availability averaged over the current and preceding years. We then compared these models using Akaike’s information criterion (Burnham and Anderson 2002) and chose the flood condition index in the best performing model for use in all subsequent models. Secondly, once the appropriate flood index (land availability) was chosen for models of pride size including only adult females and total pride size, respectively, we then ran these global models, which included per capita prey abundance of medium-sized and large prey, and land availability as fixed effects. While we predict an increase in group size with increases in both medium and large prey, we included both prey size categories separately to examine the differences in effect size in the models. Thirdly, we used a generalized linear mixed model (GLMM) to determine the effects of number of competing neighbors, as well as land availability and prey availability, on reproductive rate, which was defined as the number of cubs born to each pride in each year. For prey availability, small, medium, and large prey were pooled, as mothers with cubs are more likely to be solitary hunters or hunt in small groups. Pride identity and year were included as random effects, to incorporate variation within each pride and each year (Bertram 1975; Bolker 2008). This was done to treat each pride and year as samples from a larger population rather than factors of interest per se (Bolker 2008).
For analyses of subgroup sizes of adult females, global models included number of competing neighbors, land availability, and prey availability as fixed effects. Extant land availability was maintained as the most appropriate index of flooding since subgroup sizes are determined by fission–fusion decisions influenced by extant ecological conditions. For subgroup sizes of adult females only, number of cubs was included as an additional fixed effect. For examining the effect of prey availability on lion subgroup sizes, we included both medium and large prey separately as predictors of subgroup size. Due to differences in pride size, which put an upper limit on subgroup size, pride size was included as an offset in all subgroup models.
Generalized estimating equations were calculated using the geepack package (Højsgaard et al. 2006), and generalized linear models were constructed using package lme4 (Bates et al. 2014) in R 3.1.0 (R Core Development Team 2014). Global and nested models were compared using the MuMIn package (Barton 2013). For GEEs, models were ranked based on quasi-likelihood (QIC) under the independence model criterion as recommended by Pan (2001), and for GLMMs, models were ranked based on Aikaike’s information criterion for small sample sizes (AICc—Burnham and Anderson 2002). Models that differed from the top-ranked model by < 2 AICc or QIC values were included as equally parsimonious and taken to represent the best models of the candidates considered to describe the data (Burnham and Anderson 2002). Where no clear model was selected as the best model, models with ΔAICc < 2 and ΔQIC < 2 were averaged to account for model uncertainty, and coefficients calculated using the natural average method (Nakagawa and Freckleton 2010; Grueber et al. 2011).
Results
Over the study period, the average pride size including only adult females was 6.14 ± 1.07 (mean ± SD; range 2–14), and the average pride size with subadults included was 8.51 ± 1.44 (range 4–18; Fig. 2). From year to year, the number of adult females in each pride remained relatively close to the mean, with the median group size varying between 4 and 6 adults (Fig. 2). Average pride size, when considering only adult females, increased slightly over the study period, reached a peak in 2002, and then dropped in 2004 (Fig. 1). When subadults were included in pride size, however, there was an increase in pride size toward 1999, and thereafter a general decline toward 2004 (Fig. 2). Subgroup sizes for groups of adult females, and total subgroup sizes, showed similar trends throughout the study period. Subgroup sizes decreased toward the middle of the study period around 2000, and then increased again slightly before dropping in 2004 (Fig. 3).
Average pride sizes (± 1 SD) for African lions (Panthera leo) from 5 prides in the southwestern Okavango Delta, Botswana, from 1997 to 2004. Two aspects of pride size were measured: A) adult female pride size, and B) total pride size (adult females and subadults).
Average subgroup sizes (± 1 SD) for 5 African lion (Panthera leo) prides in the southwestern Okavango Delta, Botswana, from 1997 to 2004. Subgroups were analyzed in 2 ways: A) number of adult females in subgroups, including subgroups where young cubs (0–2 years old) are present, and B) total subgroup size where adult females and all subadults are included, excluding groups in which cubs are present.
There was a large degree of overlap between long-term home ranges of all 5 prides (Fig. 4). There was also a general decrease in land availability for each pride between 1997 and 2004 (Fig. 4), coinciding with the redistribution of flow down the western side of the Okavango Delta and down the Xudum system, as well as to increased inflow and longer inundation periods for the Okavango Delta overall. The average area of home range to remain dry during peak flood decreased from 80% in 1997 to only 25% in 2004 (Fig. 4). Overall prey availability declined over the study period, primarily as a result of the decline in medium-sized prey. However, when separated into medium-sized and large prey, medium-sized prey showed large declines, whereas the availability of large prey increased slightly over the study period (Fig. 5).
Home ranges (90% kernel) of 5 African lion (Panthera leo) prides in the southwestern Okavango Delta, Botswana. Panels illustrate the changes in maximum annual flood extent in home range areas from low to high floods during the study period from 1997 to 2004.
Kilometric biomass, in kilograms, of medium-sized and large prey encountered in lion home ranges, averaged over the 5 African lion (Panthera leo) prides in the southwestern Okavango Delta, Botswana, between 1997 and 2004.
For pride size including adult females only, the model containing extant flood conditions was selected as the best model, and so land availability in the current year was retained in all subsequent models (Table 1). Only 2 models proved parsimonious in explaining pride size of adult females; the first model included land availability and availability of medium-sized prey, and the second was the global model that included availability of large prey as an additional factor (Table 2). However, after model averaging, only land availability was retained as being marginally significant, and was positively related to pride size (Table 5).
Selection of global models with the most appropriate land availability indices related to pride size based on number of adult females only and total pride size (including subadults) of African lions (Panthera leo) in the southwestern Okavango Delta, Botswana. Aside from the land availability index, all models also included per capita availability of medium-sized and large prey. The top model was selected as the global model for all subsequent analyses. Land availability represents the amount of dry land available in each pride home range, each year during maximum flood extent. Land(t0) = land availability of current year, Land(t−1) = land availability for the previous year, Land(t−2) = land availability from 2 years prior, Land((t0+t−1)/2) = average land availability between current and previous year, Land((t0+t−1+t−2)/3) = land availability averaged over the current year and previous 2 years. ΔQIC = difference in information criterion for quasi-likelihood estimates between the current model and the top-ranked model, ω = QIC model weight, k = number of parameters in the model, and qLik = quasi-likelihood of the model.
| Model | ΔQIC | ω | k | qLik | |
|---|---|---|---|---|---|
| Pride size, adult females only | |||||
| Prey indices | Land availability indices | ||||
| PreyL + PreyM | Land(t0) | 0.00 | 0.349 | 3 | 179 |
| PreyL + PreyM | Land avg((t0+t−1+t−2)/3) | 0.37 | 0.290 | 3 | 179 |
| PreyL + PreyM | Land(t−2) | 1.01 | 0.211 | 3 | 178 |
| PreyL + PreyM | Land avg((t0+t−1)/2) | 2.14 | 0.120 | 3 | 178 |
| PreyL + PreyM | Land(t−1) | 4.88 | 0.030 | 3 | 173 |
| Total pride size | |||||
| Prey indices | Land availability indices | ||||
| PreyL + PreyM | Land(t−1) | 0 | 0.422 | 3 | 358 |
| PreyL + PreyM | Land avg((t0+t−1)/2) | 1.45 | 0.204 | 3 | 357 |
| PreyL + PreyM | Land avg((t0+t−1+t−2)/3) | 1.66 | 0.184 | 3 | 357 |
| PreyL + PreyM | Land(t−2) | 1.82 | 0.170 | 3 | 357 |
| PreyL + PreyM | Land(t0) | 6.07 | 0.020 | 3 | 353 |
| Model | ΔQIC | ω | k | qLik | |
|---|---|---|---|---|---|
| Pride size, adult females only | |||||
| Prey indices | Land availability indices | ||||
| PreyL + PreyM | Land(t0) | 0.00 | 0.349 | 3 | 179 |
| PreyL + PreyM | Land avg((t0+t−1+t−2)/3) | 0.37 | 0.290 | 3 | 179 |
| PreyL + PreyM | Land(t−2) | 1.01 | 0.211 | 3 | 178 |
| PreyL + PreyM | Land avg((t0+t−1)/2) | 2.14 | 0.120 | 3 | 178 |
| PreyL + PreyM | Land(t−1) | 4.88 | 0.030 | 3 | 173 |
| Total pride size | |||||
| Prey indices | Land availability indices | ||||
| PreyL + PreyM | Land(t−1) | 0 | 0.422 | 3 | 358 |
| PreyL + PreyM | Land avg((t0+t−1)/2) | 1.45 | 0.204 | 3 | 357 |
| PreyL + PreyM | Land avg((t0+t−1+t−2)/3) | 1.66 | 0.184 | 3 | 357 |
| PreyL + PreyM | Land(t−2) | 1.82 | 0.170 | 3 | 357 |
| PreyL + PreyM | Land(t0) | 6.07 | 0.020 | 3 | 353 |
Selection of global models with the most appropriate land availability indices related to pride size based on number of adult females only and total pride size (including subadults) of African lions (Panthera leo) in the southwestern Okavango Delta, Botswana. Aside from the land availability index, all models also included per capita availability of medium-sized and large prey. The top model was selected as the global model for all subsequent analyses. Land availability represents the amount of dry land available in each pride home range, each year during maximum flood extent. Land(t0) = land availability of current year, Land(t−1) = land availability for the previous year, Land(t−2) = land availability from 2 years prior, Land((t0+t−1)/2) = average land availability between current and previous year, Land((t0+t−1+t−2)/3) = land availability averaged over the current year and previous 2 years. ΔQIC = difference in information criterion for quasi-likelihood estimates between the current model and the top-ranked model, ω = QIC model weight, k = number of parameters in the model, and qLik = quasi-likelihood of the model.
| Model | ΔQIC | ω | k | qLik | |
|---|---|---|---|---|---|
| Pride size, adult females only | |||||
| Prey indices | Land availability indices | ||||
| PreyL + PreyM | Land(t0) | 0.00 | 0.349 | 3 | 179 |
| PreyL + PreyM | Land avg((t0+t−1+t−2)/3) | 0.37 | 0.290 | 3 | 179 |
| PreyL + PreyM | Land(t−2) | 1.01 | 0.211 | 3 | 178 |
| PreyL + PreyM | Land avg((t0+t−1)/2) | 2.14 | 0.120 | 3 | 178 |
| PreyL + PreyM | Land(t−1) | 4.88 | 0.030 | 3 | 173 |
| Total pride size | |||||
| Prey indices | Land availability indices | ||||
| PreyL + PreyM | Land(t−1) | 0 | 0.422 | 3 | 358 |
| PreyL + PreyM | Land avg((t0+t−1)/2) | 1.45 | 0.204 | 3 | 357 |
| PreyL + PreyM | Land avg((t0+t−1+t−2)/3) | 1.66 | 0.184 | 3 | 357 |
| PreyL + PreyM | Land(t−2) | 1.82 | 0.170 | 3 | 357 |
| PreyL + PreyM | Land(t0) | 6.07 | 0.020 | 3 | 353 |
| Model | ΔQIC | ω | k | qLik | |
|---|---|---|---|---|---|
| Pride size, adult females only | |||||
| Prey indices | Land availability indices | ||||
| PreyL + PreyM | Land(t0) | 0.00 | 0.349 | 3 | 179 |
| PreyL + PreyM | Land avg((t0+t−1+t−2)/3) | 0.37 | 0.290 | 3 | 179 |
| PreyL + PreyM | Land(t−2) | 1.01 | 0.211 | 3 | 178 |
| PreyL + PreyM | Land avg((t0+t−1)/2) | 2.14 | 0.120 | 3 | 178 |
| PreyL + PreyM | Land(t−1) | 4.88 | 0.030 | 3 | 173 |
| Total pride size | |||||
| Prey indices | Land availability indices | ||||
| PreyL + PreyM | Land(t−1) | 0 | 0.422 | 3 | 358 |
| PreyL + PreyM | Land avg((t0+t−1)/2) | 1.45 | 0.204 | 3 | 357 |
| PreyL + PreyM | Land avg((t0+t−1+t−2)/3) | 1.66 | 0.184 | 3 | 357 |
| PreyL + PreyM | Land(t−2) | 1.82 | 0.170 | 3 | 357 |
| PreyL + PreyM | Land(t0) | 6.07 | 0.020 | 3 | 353 |
Generalized estimating equations depicting predictor variables present in the best models (ΔQIC < 2) for pride sizes of African lions (Panthera leo) in the southwestern Okavango Delta, Botswana. Covariates in models include an estimate of kilometric biomass per capita prey availability for medium-sized prey (PreyM), an estimate of kilometric biomass per capita prey availability for large prey (PreyL), and the amount of dry land available at maximum flood extent. Two responses of group sizes were measured at the pride level: number of adult females, and total group sizes which included adult females and all subadults. ΔQIC = difference in information criterion for quasi-likelihood estimates between the current model and the top-ranked model, ω = QIC model weight, k = number of parameters in the model, and qLik = quasi-likelihood of the model.
| Model | ΔQIC | ω | k | qLik |
|---|---|---|---|---|
| Pride size, adult females only | ||||
| Land + PreyM | 0.00 | 0.463 | 4 | 179 |
| Land + PreyL + PreyM | 0.45 | 0.369 | 5 | 179 |
| Total pride size | ||||
| Land(t−1) + PreyL + PreyM | 0.00 | 0.983 | 5 | 358 |
| Model | ΔQIC | ω | k | qLik |
|---|---|---|---|---|
| Pride size, adult females only | ||||
| Land + PreyM | 0.00 | 0.463 | 4 | 179 |
| Land + PreyL + PreyM | 0.45 | 0.369 | 5 | 179 |
| Total pride size | ||||
| Land(t−1) + PreyL + PreyM | 0.00 | 0.983 | 5 | 358 |
Generalized estimating equations depicting predictor variables present in the best models (ΔQIC < 2) for pride sizes of African lions (Panthera leo) in the southwestern Okavango Delta, Botswana. Covariates in models include an estimate of kilometric biomass per capita prey availability for medium-sized prey (PreyM), an estimate of kilometric biomass per capita prey availability for large prey (PreyL), and the amount of dry land available at maximum flood extent. Two responses of group sizes were measured at the pride level: number of adult females, and total group sizes which included adult females and all subadults. ΔQIC = difference in information criterion for quasi-likelihood estimates between the current model and the top-ranked model, ω = QIC model weight, k = number of parameters in the model, and qLik = quasi-likelihood of the model.
| Model | ΔQIC | ω | k | qLik |
|---|---|---|---|---|
| Pride size, adult females only | ||||
| Land + PreyM | 0.00 | 0.463 | 4 | 179 |
| Land + PreyL + PreyM | 0.45 | 0.369 | 5 | 179 |
| Total pride size | ||||
| Land(t−1) + PreyL + PreyM | 0.00 | 0.983 | 5 | 358 |
| Model | ΔQIC | ω | k | qLik |
|---|---|---|---|---|
| Pride size, adult females only | ||||
| Land + PreyM | 0.00 | 0.463 | 4 | 179 |
| Land + PreyL + PreyM | 0.45 | 0.369 | 5 | 179 |
| Total pride size | ||||
| Land(t−1) + PreyL + PreyM | 0.00 | 0.983 | 5 | 358 |
For total pride size, which included adult females and all subadults, the top model examining extant and time-lagged flood effects included land availability from the previous year (Table 1). The amount of land available during peak flood of the previous year was retained in all further models. During model selection, the global model, which included land availability from the previous year, availability of medium-sized prey, and availability of large prey, was selected as the best model with a model weight of 0.983 (Table 2). This model indicated that pride sizes generally increase with an increase in land availability, and, interestingly, with a decline in availability of medium-sized and large prey.
At the pride level, 2 models were selected as top models explaining variation in reproductive rates; the first model included only competition with neighbors, while the second model included competition with neighbors as well as land availability (Table 3). After model averaging, both of these covariates were retained as having a significant effect on reproductive rate (Table 5). Reproductive rates were positively influenced by land availability, and negatively influenced by the number of competing neighbors. Across all prides, there was a general decrease in reproductive rate, measured as the number of cubs born in each pride, over the study period (Fig. 6).
Generalized linear models depicting predictor variables present in top models (ΔAICc < 2) for reproductive rate of adult female African lions (Panthera leo) in the southwestern Okavango Delta, Botswana. Reproductive rate was measured per pride as the number of cubs born in each pride each year. The explanatory variables selected include competition, which represents the number of neighboring adult females, and land availability, which represents the dry land available in each pride home range during maximum flood extent of each year. ΔAICc = difference in Aikaike’s Information Criterion for small sample sizes between the current model and the top-ranked model, ω = AICc model weight, k = number of parameters in the model, and logLik = log-likelihood of the model.
| Model | ΔAICc | ω | k | logLik |
|---|---|---|---|---|
| Competition | 0 | 0.72 | 4 | −74.56 |
| Competition + Land | 1.93 | 0.28 | 5 | −74.11 |
| Model | ΔAICc | ω | k | logLik |
|---|---|---|---|---|
| Competition | 0 | 0.72 | 4 | −74.56 |
| Competition + Land | 1.93 | 0.28 | 5 | −74.11 |
Generalized linear models depicting predictor variables present in top models (ΔAICc < 2) for reproductive rate of adult female African lions (Panthera leo) in the southwestern Okavango Delta, Botswana. Reproductive rate was measured per pride as the number of cubs born in each pride each year. The explanatory variables selected include competition, which represents the number of neighboring adult females, and land availability, which represents the dry land available in each pride home range during maximum flood extent of each year. ΔAICc = difference in Aikaike’s Information Criterion for small sample sizes between the current model and the top-ranked model, ω = AICc model weight, k = number of parameters in the model, and logLik = log-likelihood of the model.
| Model | ΔAICc | ω | k | logLik |
|---|---|---|---|---|
| Competition | 0 | 0.72 | 4 | −74.56 |
| Competition + Land | 1.93 | 0.28 | 5 | −74.11 |
| Model | ΔAICc | ω | k | logLik |
|---|---|---|---|---|
| Competition | 0 | 0.72 | 4 | −74.56 |
| Competition + Land | 1.93 | 0.28 | 5 | −74.11 |
Reproductive rates, determined as the number of cubs born each year, summed across 5 African lion (Panthera leo) prides, over the study period from 1997 to 2004 in the southwestern Okavango Delta, Botswana.
Top models for subgroup sizes of adult females included all social and ecological variables (Table 4). However, after model averaging, only the number of cubs had a significant influence on the response variable, and positively influenced group size (Table 5). For total subgroup size, all 4 explanatory variables were retained in the top models (Table 4). Model averaging revealed that only competition with neighbors and availability of large prey were significant, and both factors had a positive relationship with subgroup size, as well as a similar effect size (Table 5). While availability of medium-sized prey did not prove to be significant after model averaging, it had a negative relationship with subgroup size.
Generalized estimating equations depicting predictor variables present in the top models (ΔQIC < 2) for subgroup sizes of African lion (Panthera leo) prides in the southwestern Okavango Delta, Botswana. Covariates include an estimate of kilometric biomass per capita prey availability for medium-sized prey (PreyM), an estimate of kilometric biomass per capita prey availability for large prey (PreyL), the number of competing adult female neighbors as an index of competition, the amount of dry land available at maximum flood extent, and the number of cubs present in subgroups of adult females. Two responses of group sizes were measured at subgroup level: number of adult females, and total group sizes, which included adult females and all subadults. ΔQIC = difference in information criterion for quasi-likelihood estimates between the current model and the top-ranked model, ω = QIC model weight, k = number of parameters in the model, and qLik = quasi-likelihood of the model.
| Model | ΔQIC | ω | k | qLik |
|---|---|---|---|---|
| Subgroup size, adult females only | ||||
| Land + No. cubs | 0.00 | 0.132 | 5 | 8.51 |
| Land + PreyL + No. cubs | 0.41 | 0.108 | 6 | 9.19 |
| Competition + Land + No. cubs | 0.63 | 0.097 | 6 | 8.53 |
| Competition + Land + PreyL + No. cubs | 0.80 | 0.089 | 7 | 9.20 |
| PreyL | 1.20 | 0.073 | 4 | 8.64 |
| Land + PreyL + PreyM | 1.52 | 0.062 | 6 | 9.20 |
| Competition + Land + PreyL + PreyM + No. cubs | 1.63 | 0.059 | 8 | 9.21 |
| Competition + PreyL + No. cubs | 1.64 | 0.058 | 6 | 8.65 |
| PreyL + No. cubs | 1.80 | 0.054 | 5 | 8.62 |
| Land + PreyM + No. cubs | 1.82 | 0.053 | 6 | 8.68 |
| Total subgroup size (adult females and subadults) | ||||
| Competition + PreyL | 0.00 | 0.172 | 5 | 15.9 |
| Competition + Land + PreyL | 0.19 | 0.157 | 6 | 16.0 |
| Competition + PreyL + PreyM | 0.35 | 0.145 | 6 | 16.0 |
| Competition + Land + PreyL + PreyM | 0.68 | 0.122 | 7 | 16.0 |
| Competition + Land | 1.62 | 0.077 | 5 | 14.9 |
| Model | ΔQIC | ω | k | qLik |
|---|---|---|---|---|
| Subgroup size, adult females only | ||||
| Land + No. cubs | 0.00 | 0.132 | 5 | 8.51 |
| Land + PreyL + No. cubs | 0.41 | 0.108 | 6 | 9.19 |
| Competition + Land + No. cubs | 0.63 | 0.097 | 6 | 8.53 |
| Competition + Land + PreyL + No. cubs | 0.80 | 0.089 | 7 | 9.20 |
| PreyL | 1.20 | 0.073 | 4 | 8.64 |
| Land + PreyL + PreyM | 1.52 | 0.062 | 6 | 9.20 |
| Competition + Land + PreyL + PreyM + No. cubs | 1.63 | 0.059 | 8 | 9.21 |
| Competition + PreyL + No. cubs | 1.64 | 0.058 | 6 | 8.65 |
| PreyL + No. cubs | 1.80 | 0.054 | 5 | 8.62 |
| Land + PreyM + No. cubs | 1.82 | 0.053 | 6 | 8.68 |
| Total subgroup size (adult females and subadults) | ||||
| Competition + PreyL | 0.00 | 0.172 | 5 | 15.9 |
| Competition + Land + PreyL | 0.19 | 0.157 | 6 | 16.0 |
| Competition + PreyL + PreyM | 0.35 | 0.145 | 6 | 16.0 |
| Competition + Land + PreyL + PreyM | 0.68 | 0.122 | 7 | 16.0 |
| Competition + Land | 1.62 | 0.077 | 5 | 14.9 |
Generalized estimating equations depicting predictor variables present in the top models (ΔQIC < 2) for subgroup sizes of African lion (Panthera leo) prides in the southwestern Okavango Delta, Botswana. Covariates include an estimate of kilometric biomass per capita prey availability for medium-sized prey (PreyM), an estimate of kilometric biomass per capita prey availability for large prey (PreyL), the number of competing adult female neighbors as an index of competition, the amount of dry land available at maximum flood extent, and the number of cubs present in subgroups of adult females. Two responses of group sizes were measured at subgroup level: number of adult females, and total group sizes, which included adult females and all subadults. ΔQIC = difference in information criterion for quasi-likelihood estimates between the current model and the top-ranked model, ω = QIC model weight, k = number of parameters in the model, and qLik = quasi-likelihood of the model.
| Model | ΔQIC | ω | k | qLik |
|---|---|---|---|---|
| Subgroup size, adult females only | ||||
| Land + No. cubs | 0.00 | 0.132 | 5 | 8.51 |
| Land + PreyL + No. cubs | 0.41 | 0.108 | 6 | 9.19 |
| Competition + Land + No. cubs | 0.63 | 0.097 | 6 | 8.53 |
| Competition + Land + PreyL + No. cubs | 0.80 | 0.089 | 7 | 9.20 |
| PreyL | 1.20 | 0.073 | 4 | 8.64 |
| Land + PreyL + PreyM | 1.52 | 0.062 | 6 | 9.20 |
| Competition + Land + PreyL + PreyM + No. cubs | 1.63 | 0.059 | 8 | 9.21 |
| Competition + PreyL + No. cubs | 1.64 | 0.058 | 6 | 8.65 |
| PreyL + No. cubs | 1.80 | 0.054 | 5 | 8.62 |
| Land + PreyM + No. cubs | 1.82 | 0.053 | 6 | 8.68 |
| Total subgroup size (adult females and subadults) | ||||
| Competition + PreyL | 0.00 | 0.172 | 5 | 15.9 |
| Competition + Land + PreyL | 0.19 | 0.157 | 6 | 16.0 |
| Competition + PreyL + PreyM | 0.35 | 0.145 | 6 | 16.0 |
| Competition + Land + PreyL + PreyM | 0.68 | 0.122 | 7 | 16.0 |
| Competition + Land | 1.62 | 0.077 | 5 | 14.9 |
| Model | ΔQIC | ω | k | qLik |
|---|---|---|---|---|
| Subgroup size, adult females only | ||||
| Land + No. cubs | 0.00 | 0.132 | 5 | 8.51 |
| Land + PreyL + No. cubs | 0.41 | 0.108 | 6 | 9.19 |
| Competition + Land + No. cubs | 0.63 | 0.097 | 6 | 8.53 |
| Competition + Land + PreyL + No. cubs | 0.80 | 0.089 | 7 | 9.20 |
| PreyL | 1.20 | 0.073 | 4 | 8.64 |
| Land + PreyL + PreyM | 1.52 | 0.062 | 6 | 9.20 |
| Competition + Land + PreyL + PreyM + No. cubs | 1.63 | 0.059 | 8 | 9.21 |
| Competition + PreyL + No. cubs | 1.64 | 0.058 | 6 | 8.65 |
| PreyL + No. cubs | 1.80 | 0.054 | 5 | 8.62 |
| Land + PreyM + No. cubs | 1.82 | 0.053 | 6 | 8.68 |
| Total subgroup size (adult females and subadults) | ||||
| Competition + PreyL | 0.00 | 0.172 | 5 | 15.9 |
| Competition + Land + PreyL | 0.19 | 0.157 | 6 | 16.0 |
| Competition + PreyL + PreyM | 0.35 | 0.145 | 6 | 16.0 |
| Competition + Land + PreyL + PreyM | 0.68 | 0.122 | 7 | 16.0 |
| Competition + Land | 1.62 | 0.077 | 5 | 14.9 |
Model-averaged estimates for top models investigating the effects of social and ecological factors on pride size, subgroup sizes, and reproductive rate of 5 African lion (Panthera leo) prides in the southwestern Okavango Delta, Botswana. Total group size includes adult females and all subadults, and total subgroup size excludes subgroups in which cubs are present. Reproductive rate is defined as the number of cubs born per annum. Prey represents an estimate of kilometric biomass per capita prey availability for medium-sized (PreyM) and large prey (PreyL), respectively; competition represents the number of competing adult female neighbors; land represents the amount of dry land available at maximum flood extent; and number of cubs represents the number of cubs present in subgroups of adult females. Predictor variables were standardized for all of the models to make results directly comparable. For group sizes, generalized estimating equations were used, and all models within 2 QIC units of the top model were used for model averaging. For reproductive rates, a generalized linear mixed model was used, and all models within 2 AICc units of the model were used for model averaging. Estimate represents the magnitude of the effect, and SE represents unconditional standard error. CIs that did not cross zero are highlighted in bold and indicate variables that were considered influential in determining group sizes or reproductive rate. *Marginally significant.
| Variable | Estimate | SE | CI |
|---|---|---|---|
| Pride size, adult females only | |||
| (Intercept) | 1.7807 | 0.1414 | (1.5036, 2.0580) |
| Land | 0.1574 | 0.0922 | (−0.0233, 0.3380)* |
| PreyM | −0.0372 | 0.0863 | (−0.2064, 0.1320) |
| PreyL | −0.0470 | 0.0775 | (−0.1989, 0.1050) |
| Subgroup size, adult females only | |||
| (Intercept) | −1.06536 | 0.0722 | (−1.2070, −0.9237) |
| Land | −0.25514 | 0.1993 | (−0.6459, 0.1356) |
| No. cubs | 0.41414 | 0.0574 | (0.3017, 0.5265) |
| PreyL | 0.24537 | 0.2239 | (−0.1934, 0.6841) |
| Competition | −0.00545 | 0.0631 | (−0.1291, 0.1182) |
| PreyM | −0.00674 | 0.0859 | (−0.1750, 0.1615) |
| Total subgroup size | |||
| (Intercept) | −0.9856 | 0.0897 | (−1.1614, −0.8098) |
| Competition | 0.2459 | 0.1134 | (0.0237, 0.4680) |
| PreyL | 0.2640 | 0.1235 | (0.0219, 0.5062) |
| Land | −0.0044 | 0.0976 | (−0.1957, 0.1869) |
| PreyM | −0.0219 | 0.0480 | (−0.1160, 0.0722) |
| Reproductive rate | |||
| (Intercept) | 0.4579 | 0.2817 | (−0.1194, 1.0352) |
| Competition | −1.1895 | 0.3011 | (−1.8016, −0.5774) |
| Land | 1.1908 | 0.3123 | (0.5501, 1.8315) |
| Variable | Estimate | SE | CI |
|---|---|---|---|
| Pride size, adult females only | |||
| (Intercept) | 1.7807 | 0.1414 | (1.5036, 2.0580) |
| Land | 0.1574 | 0.0922 | (−0.0233, 0.3380)* |
| PreyM | −0.0372 | 0.0863 | (−0.2064, 0.1320) |
| PreyL | −0.0470 | 0.0775 | (−0.1989, 0.1050) |
| Subgroup size, adult females only | |||
| (Intercept) | −1.06536 | 0.0722 | (−1.2070, −0.9237) |
| Land | −0.25514 | 0.1993 | (−0.6459, 0.1356) |
| No. cubs | 0.41414 | 0.0574 | (0.3017, 0.5265) |
| PreyL | 0.24537 | 0.2239 | (−0.1934, 0.6841) |
| Competition | −0.00545 | 0.0631 | (−0.1291, 0.1182) |
| PreyM | −0.00674 | 0.0859 | (−0.1750, 0.1615) |
| Total subgroup size | |||
| (Intercept) | −0.9856 | 0.0897 | (−1.1614, −0.8098) |
| Competition | 0.2459 | 0.1134 | (0.0237, 0.4680) |
| PreyL | 0.2640 | 0.1235 | (0.0219, 0.5062) |
| Land | −0.0044 | 0.0976 | (−0.1957, 0.1869) |
| PreyM | −0.0219 | 0.0480 | (−0.1160, 0.0722) |
| Reproductive rate | |||
| (Intercept) | 0.4579 | 0.2817 | (−0.1194, 1.0352) |
| Competition | −1.1895 | 0.3011 | (−1.8016, −0.5774) |
| Land | 1.1908 | 0.3123 | (0.5501, 1.8315) |
Model-averaged estimates for top models investigating the effects of social and ecological factors on pride size, subgroup sizes, and reproductive rate of 5 African lion (Panthera leo) prides in the southwestern Okavango Delta, Botswana. Total group size includes adult females and all subadults, and total subgroup size excludes subgroups in which cubs are present. Reproductive rate is defined as the number of cubs born per annum. Prey represents an estimate of kilometric biomass per capita prey availability for medium-sized (PreyM) and large prey (PreyL), respectively; competition represents the number of competing adult female neighbors; land represents the amount of dry land available at maximum flood extent; and number of cubs represents the number of cubs present in subgroups of adult females. Predictor variables were standardized for all of the models to make results directly comparable. For group sizes, generalized estimating equations were used, and all models within 2 QIC units of the top model were used for model averaging. For reproductive rates, a generalized linear mixed model was used, and all models within 2 AICc units of the model were used for model averaging. Estimate represents the magnitude of the effect, and SE represents unconditional standard error. CIs that did not cross zero are highlighted in bold and indicate variables that were considered influential in determining group sizes or reproductive rate. *Marginally significant.
| Variable | Estimate | SE | CI |
|---|---|---|---|
| Pride size, adult females only | |||
| (Intercept) | 1.7807 | 0.1414 | (1.5036, 2.0580) |
| Land | 0.1574 | 0.0922 | (−0.0233, 0.3380)* |
| PreyM | −0.0372 | 0.0863 | (−0.2064, 0.1320) |
| PreyL | −0.0470 | 0.0775 | (−0.1989, 0.1050) |
| Subgroup size, adult females only | |||
| (Intercept) | −1.06536 | 0.0722 | (−1.2070, −0.9237) |
| Land | −0.25514 | 0.1993 | (−0.6459, 0.1356) |
| No. cubs | 0.41414 | 0.0574 | (0.3017, 0.5265) |
| PreyL | 0.24537 | 0.2239 | (−0.1934, 0.6841) |
| Competition | −0.00545 | 0.0631 | (−0.1291, 0.1182) |
| PreyM | −0.00674 | 0.0859 | (−0.1750, 0.1615) |
| Total subgroup size | |||
| (Intercept) | −0.9856 | 0.0897 | (−1.1614, −0.8098) |
| Competition | 0.2459 | 0.1134 | (0.0237, 0.4680) |
| PreyL | 0.2640 | 0.1235 | (0.0219, 0.5062) |
| Land | −0.0044 | 0.0976 | (−0.1957, 0.1869) |
| PreyM | −0.0219 | 0.0480 | (−0.1160, 0.0722) |
| Reproductive rate | |||
| (Intercept) | 0.4579 | 0.2817 | (−0.1194, 1.0352) |
| Competition | −1.1895 | 0.3011 | (−1.8016, −0.5774) |
| Land | 1.1908 | 0.3123 | (0.5501, 1.8315) |
| Variable | Estimate | SE | CI |
|---|---|---|---|
| Pride size, adult females only | |||
| (Intercept) | 1.7807 | 0.1414 | (1.5036, 2.0580) |
| Land | 0.1574 | 0.0922 | (−0.0233, 0.3380)* |
| PreyM | −0.0372 | 0.0863 | (−0.2064, 0.1320) |
| PreyL | −0.0470 | 0.0775 | (−0.1989, 0.1050) |
| Subgroup size, adult females only | |||
| (Intercept) | −1.06536 | 0.0722 | (−1.2070, −0.9237) |
| Land | −0.25514 | 0.1993 | (−0.6459, 0.1356) |
| No. cubs | 0.41414 | 0.0574 | (0.3017, 0.5265) |
| PreyL | 0.24537 | 0.2239 | (−0.1934, 0.6841) |
| Competition | −0.00545 | 0.0631 | (−0.1291, 0.1182) |
| PreyM | −0.00674 | 0.0859 | (−0.1750, 0.1615) |
| Total subgroup size | |||
| (Intercept) | −0.9856 | 0.0897 | (−1.1614, −0.8098) |
| Competition | 0.2459 | 0.1134 | (0.0237, 0.4680) |
| PreyL | 0.2640 | 0.1235 | (0.0219, 0.5062) |
| Land | −0.0044 | 0.0976 | (−0.1957, 0.1869) |
| PreyM | −0.0219 | 0.0480 | (−0.1160, 0.0722) |
| Reproductive rate | |||
| (Intercept) | 0.4579 | 0.2817 | (−0.1194, 1.0352) |
| Competition | −1.1895 | 0.3011 | (−1.8016, −0.5774) |
| Land | 1.1908 | 0.3123 | (0.5501, 1.8315) |
Discussion
Unraveling the drivers behind group sizes in social animals is complex (Packer et al. 1990), and the relative roles of social and ecological factors in determining group sizes change at different tiers of social organization. Our results indicate that flooding patterns influenced total pride sizes of lion prides in the Okavango Delta, and that as flood levels increased, total pride sizes decreased. This was likely a result of a decline in reproductive rates during years of higher flood. Intraspecific competition, in the form of number of neighboring adult females, had a similar negative effect on reproductive rates. Contrary to our expectations, prey availability did not appear to limit pride sizes in the Okavango Delta, which is often the case in drier areas (Hanby and Bygott 1979; Van Orsdol et al. 1985; Stander 1991). However, prey availability did influence subgroup behavior, with larger subgroups forming as numbers of large prey increased. Subgroups of adults and subadults without cubs increased in size as the number of neighbors increased, likely to allow for numerical advantage in inter-pride encounters (Mosser and Packer 2009). Lastly, sizes of subgroups of adult females were positively related to the number of attending cubs, simply suggesting that crèching of cubs is the most consistent predictor of subgroup behavior of adult females (Mosser and Packer 2009).
Under favorable ecological circumstances, the survival of large cohorts of cubs in multiple prides may lead to a sudden increase in the lion population (Packer et al. 2005). In 1997 and 1998, reproductive rates were relatively high. These were the driest years of the study period, following the lowest recorded flood levels in the Okavango Delta in 1996 (Murray-Hudson et al. 2014). The recruitment of these cubs as subadults resulted in the general increase in total pride sizes up to 1999, and their recruitment as adults is reflected in the general increase in adult females toward 2001 and 2002. As flood levels steadily increased toward the latter part of the study period, total pride sizes showed a general decline. Given the longevity of lions, there is likely to be a significant time lag between current ecological circumstances and changes in the population, particularly when examining the number of adult females in the population. The lack of a clear significant response between adult female pride sizes and flood levels may thus be a consequence of this time lag.
The time lag in response between flooding and group size changes is more evident in the relationship between total pride size and flooding, where flood conditions of the previous year were more predictive of total pride size than the extant flood levels. The negative relationship between flooding and total pride size, however, supported our hypotheses that total pride sizes would decrease as the Okavango Delta goes through high flood phases. We posit that these changes could be related to inflated local densities, resulting from the contraction of available habitat rather than an increase in population size. Ultimately, these inflated densities could lead to a number of cascading demographic responses during extended periods of high flood, such as lower cub survival and increased wounding and mortality from fighting (Packer et al. 1988; Mosser and Packer 2009), or as is shown in this study, a decline in reproductive rates. As available dry land declines with higher floods, there may be increasing contact with neighboring prides, resulting in higher risk of infanticide. Under such circumstances, females may thus delay breeding until conditions become more favorable (Wolff 1997). In Kafue National Park, Zambia, pride sizes and survival of young cubs were posited to be similarly negatively affected by seasonal flooding (Midlane 2013). Flooding patterns therefore exert significant influence on the lion population in wetland environments.
In contrast to our expectations, prey abundance was negatively rather than positively related to total pride size. In mesic systems, such as Serengeti National Park and Ngorongoro Crater, Tanzania, higher lean-season prey abundance is associated with larger prides (Bertram 1973; Hanby and Bygott 1979; Van Orsdol et al. 1985). However, while group size in carnivores is often related to resource availability (e.g., coyotes, Canis latrans—Bowen 1981; golden jackals, Canis aureus—Macdonald 1979; lynx, Lynx canadensis—O’Donoghue et al. 1998), this study shows that where resources are not limiting, other proximate social and ecological factors may become more important in influencing social organization (Messier 1994; Trinkel et al. 2007). In South Africa, for example, lions reintroduced to areas with abundant prey increase quickly until they become limited by social rather than ecological factors (Hunter 1998; Kilian 2003; Trinkel et al. 2010). Alternatively, time lags between prey decline and population responses by lions could explain the lack of association between pride size and prey availability in the same year. For example, for wolves (Canis lupus) on Isle Royale, population responses to declining moose (Alces alces) lagged by 3–5 years (Peterson and Page 1988). As a result of this time lag, lion numbers may not have reached equilibrium with prey numbers within the study period (Fuller and Sievert 2001; Packer et al. 2005).
In systems where predator populations are not limited by prey, they may first be regulated by declining space or intraspecific competition (Holling 1965; Messier 1994). At low population densities, populations are regulated through territorial or spacing behavior, as individuals leave natal areas to enter adjacent territories (Hestbeck 1982). However, as densities increase, large numbers of neighbors may reduce emigration by increasing the costs of dispersal (Hestbeck 1982). In lion prides in the Serengeti, for example, dispersal by subadult females decreases because of increasing numbers of neighbors and lack of vacant territories (Packer et al. 2005; Van der Waal et al. 2009). Given the large overlap in home range areas in our study area, it is likely that the habitat was saturated, which would account for the recruitment of subadult females into their natal prides over the study period. This would account for the incremental increase in adult females in the population, despite declining trends in prey availability and increases in flood levels. The resulting increase in within-group competition may be another explanation for why reproductive rates decline during periods of high flood.
The negative relationship between number of competing neighbors and reproductive rates provides further evidence for density-dependent regulation in the Okavango Delta, and high numbers of neighbors may thus limit group sizes over time (Fuller and Sievert 2001). Studies of lions over a range of ecosystems suggest that reproductive rates in lions are density-dependent and decrease with increasing density of conspecifics (Smuts 1976; Packer et al. 1988; Trinkel et al. 2010). In the Serengeti, reproductive success of females, which was defined as the number of cubs per adult female surviving to 12 months, declined as the number of neighbors increased (Mosser and Packer 2009). The decline in reproductive rates with increasing density also has been well documented in South Africa, where lions have been reintroduced onto small reserves (Trinkel et al. 2010; Miller and Funston 2014). Reduced reproductive rates in response to intraspecific competition have similarly been observed in other carnivores (see Wallach et al. 2015). For example, in brown bears (Ursus arctos), reproductive suppression of females was caused by neighboring females with cubs (Ordiz et al. 2008), and in protected populations of gray wolves, densities are higher and reproductive rates lower than in hunted populations (Haber 1996).
While the extent of home range overlap observed in our study is likely the result of high lion densities and declining space, it also may be a function of a wetland environment. In more xeric environments, maintaining access to limited surface water has been shown to have important fitness consequences for lion prides (Mosser 2008; Valeix et al. 2010, 2012), and the ability to monopolize such resources through cooperation may have acted to increase territorial exclusivity (Grinnell et al. 1995; Heinsohn and Packer 1995; Valeix et al. 2012). This territorial behavior also has been observed in other carnivores; where resources are clumped or scarce, maintaining territorial exclusivity conveys fitness advantages, and group sizes may be larger in order to outcompete smaller groups (golden jackal, C. aureus—Macdonald 1979; striped hyena, Hyaena vulgaris—Kruuk 1976; Macdonald 1978; spotted hyena, Crocuta crocuta—Kruuk 1972). In wetlands, however, water is more abundant, particularly as the flooding occurs in the dry season, making the distribution of prey and availability of dry land more unpredictable (Davidson et al. 2013; Midlane 2013). Furthermore, flooding might alter territorial boundaries by washing away scent, making the costs of maintaining exclusive territories much higher (Crawshaw and Quigley 1991; McLoughlin et al. 2001). Where resource distribution is unpredictable and where territoriality is costly, species can exhibit home range sharing (brown bears—McLoughlin et al. 2001; red foxes, Vulpes vulpes—Macdonald et al. 1999), as is the case in this study. The same pattern of extensive home range overlap in wetlands in contrast to drier areas has been observed in female jaguars (Panthera onca) in the Pantanal, Brazil, a seasonally flooded wetland (Crawshaw and Quigley 1991).
With a decline in land availability, and thus increasing competition for space, inter-pride interactions likely increased over the study period. Communal rearing of cubs would thus likely prove vital in ensuring reproductive success (Mosser and Packer 2009). In this study, the number of associating adult females in subgroups was simply related to the number of cubs present, suggesting that crèching behavior alone is the main reason for associations between adult females with cubs. This pattern may also stem from the fact that subadults and adult females without cubs actively avoid crèches due to lower individual food intake (Packer 1986; Packer et al. 1990). These findings are consistent with other studies on lions in the Serengeti (Pusey and Packer 1994) as well as in the Kgalagadi Transfrontier Park (Funston and Hermann 2001). Increased reproductive success from communal rearing of cubs is an important benefit of sociality in lion society (Bertram 1975). The presence of cubs, therefore, remains the most consistent predictor of subgrouping behavior of adult females across a variety of habitats, as shown in Mosser and Packer (2009).
In subgroups where cubs were absent, larger subgroups formed as the numbers of adult females in competing prides increased, which was consistent with our predictions. In the Serengeti, subgroup sizes were larger in large prides when in areas of shared territory, presumably to ensure numerical advantage in interterritorial disputes which may result in gained territory (Mosser and Packer 2009). As the costs of fighting through mortality and wounding are high in lions (Schaller 1972; Mosser and Packer 2009), maintaining larger subgroup sizes in areas of shared territory is particularly beneficial. Alternatively, larger subgroup sizes may have the added benefit of better defense of kills against scavengers, or in this study area, an increase in the number of neighboring lions (Elliot and McTaggart Cowan 1978; Packer 1986; Cooper 1991). This may become even more significant as lions hunt larger prey, where carcasses may need to be defended for several days. These results are substantiated by several other studies which show that territorial defense and defense of kills may result in larger subgroup sizes than would be predicted by maximizing food intake alone (Caraco and Wolf 1975; Packer 1986; Mosser and Packer 2009).
While prey availability did not influence pride sizes, the availability of large prey had a significant influence on sizes of subgroups including adult females and subadults. While both large and medium-sized prey fall within the lion’s preferred weight range, lions are expected to form larger subgroups when hunting large prey in order to increase their chance of success (Packer and Ruttan 1988; Scheel and Packer 1991). If there is sufficient medium-sized prey, however, which are easier to hunt in smaller groups or as solitary individuals, subgroups may be smaller to maximize individual food intake (Packer et al. 1990). In agreement with this difference in hunting strategy, our results indicate that medium-sized prey and large-sized prey had opposite effects on subgroup size. The availability of large prey in particular resulted in significant increases in subgroup sizes. In our study area, medium-sized prey declined throughout the study period in comparison to large herbivores, and these results are consistent with patterns observed throughout the greater Okavango Delta system (Chase 2011). Consequently, lions might have been forced to hunt larger, more challenging prey such as giraffe (Giraffa camelopardis) and buffalo (Syncerus caffer), which would require larger hunting parties (Packer 1986; Packer et al. 1990; Stander 1991).
While in solitary species, population changes are incremental in response to ecological change, the social organization of the lion results in complex population responses to a changing environment (Packer et al. 2005). Understanding how social organization responds to ecological change may thus give us important insight into how population trends may be affected over time. In wetlands, available dry land is an important resource, and changes in land availability exert significant influence over lion social organization and reproductive rates. In dry years, the simultaneous survival of large cohorts of cubs in multiple prides can lead to an increase in population growth, and the establishment of a new population equilibrium (Packer et al. 2005). Years of higher flood, however, resulted in increased competition for space, and total group sizes started to decrease as reproductive rates declined. In the southwestern Okavango Delta, pride sizes are thus more limited by competition for space rather than food availability. During extended high flood phases, the reduction in space will likely result in the establishment of a new, lower population equilibrium, with smaller total pride sizes, until the next drying phase and increase in dry land results in a subsequent population increase.
Acknowledgments
We thank the President of Botswana, the Ministry of Environment, Wildlife and Tourism, and the Botswana Department of Wildlife and National Parks for making this research possible (permit reference numbers: OP 46/1 LXVIII (133) and EWT 3/3/8 XXIX (50)). We are grateful to the numerous volunteers at Lion Camp for their data collection efforts, and also to andBeyond for permission to access their concessions. Funding for the research as well as logistical support was provided by Safari South (Pty) Ltd and Rann Safaris (Pty) Ltd, whose involvement in no way affected the results. Finally, we thank the Okavango Research Institute and M. Murray-Hudson, P. Wolski, and K. Thito for providing satellite imagery depicting flooding patterns in the study area.
