Cape buffalo (Syncerus caffer caffer) social dynamics in a flood-pulsed environment

Cape buffalo (Syncerus caffer caffer) social dynamics in a flood-pulsed environment Abstract Fission–fusion social dynamics allow animals to respond to short-term environmental changes by temporarily adjusting group size. The drivers of such complex social dynamics are thought to relate to resource availability, density effects, and social interactions. During 2008–2009, we collared 15 Cape buffalo (Syncerus caffer caffer) cows in different groups in the Okavango Delta, Botswana, to study social dynamics in a flood-pulsed ecosystem. Hierarchical clustering identified 2 subpopulations, 1 migratory and 1 resident. We calculated Utilization Distribution Overlap Index (UDOI) and number and duration of fusion events between buffalo dyads and related them to environmental variables. Number of fusion events and duration of fusion periods did not vary seasonally, total fusion time varied seasonally and annually, and UDOI varied with year and subpopulation. Fission events were more likely in cluttered habitats, but only in the late flood season. There was more open habitat in home range overlap (HRO) areas than home ranges (HR) in most seasons. Pan density in rainy season HROs was lower and higher than in HRs in 2008 and 2009, respectively, and in all flood seasons HRO areas were closer to permanent water than HRs, suggesting that fusion occurred when buffalo congregated on resources. Whereas previous studies described large herds that sometimes split, we identified numerous smaller groups that occasionally fused, indicating a very fluid social system. Our results highlight the need to understand social system flexibility to ensure appropriate management and understand the varying impacts of environmental and anthropogenic effects on subpopulations within the same geographic area. INTRODUCTION Gregarious, group-dwelling herbivores benefit from collective information sharing about resources (Isvaran 2007; Aureli et al. 2008), high levels of vigilance (Kie 1999), and effective antipredator defense strategies (Kelley et al. 2011). However, associating with other individuals also incurs costs such as competition for mates and resources, which are often dependent on density and resource availability (Alexander 1974). Optimal group size is therefore a trade-off between these benefits and costs (Markham et al. 2015), which can vary temporally through seasonal changes in environmental conditions, spatially through structural differences between habitat types, and overall through social interactions (Couzin and Laidre 2009). Herbivores usually occur in larger groups in open habitats with high visibility (Jarman 1974; Isvaran 2007), where they can detect potential predators most readily (Fortin et al. 2009), but also where intraspecific interactions can take place over larger distances. Short-term responses to changing environmental conditions can lead to fission–fusion social dynamics (Aureli et al. 2008), whereby groups split and join periodically, creating a relatively fluid and adaptable society (Sueur et al. 2011), although the individual composition of each core group usually remains the same (Wittemyer et al. 2005). The drivers of fusion events are not fully understood, but they include predation pressure (Kelley et al. 2011), reproductive opportunities (Mourier et al. 2012), congregation on abundant (Isvaran 2007) or limited resources (Holmes et al. 2016), and social attraction (Couzin 2006; Bercovitch and Berry 2010). Fission is thought to occur when the costs of competition for limited resources or mates become too high (Couzin and Laidre 2009), the costs of activity synchrony rise for dimorphic species (Conradt and Roper 2000), knowledge of profitable patches differs between experienced individuals (Ramos et al. 2015) that move apart and are followed by other group members (Merkle et al. 2015), or detection probability by predators increases with herd size, such as in cluttered habitats (Fortin et al. 2009). The most productive season is usually when the costs of social cohesion are at their lowest, and therefore should be associated with the largest group sizes in species exhibiting fission–fusion social dynamics (Wittemyer et al. 2005; Isvaran 2007; Bercovitch and Berry 2010). Fission–fusion dynamics can increase vulnerability of some groups to disturbance effects (Haulsee et al. 2016; Sigaud et al. 2017), but can also increase adaptability by allowing animals to alter their group size in response to environmental conditions (Holmes et al. 2016). Most herbivore species do not defend territories, but occupy home ranges (HRs) that can vary geographically according to seasonal changes in resource abundance and distribution (Owen-Smith et al. 2010). Utilizing HRs allows animals to access familiar locations and, through spatial memory, exploit productive areas and predict temporal and spatial shifts in resource availability (Wolf et al. 2009). Some areas can support groups with overlapping HRs, and the extent of spatial overlap relates to resource availability (Kolbe and Weckerly 2015). Animals may need to share limiting resources, such as water, or be able to coexist in areas with high resource abundance (Chaverri et al. 2007), leading to social interactions between individuals with overlapping HRs (VanderWaal et al. 2014). The extent of home range overlap (HRO) should follow a U-shaped curve in relation to resources: HRO should be high during periods of high and low resource abundance and lowest during periods of intermediate abundance (Zengeya and Murwira 2016). Variation in resource abundance and extent of social interactions can therefore lead to seasonal changes in fission–fusion dynamics within HRO areas. Fission–fusion social dynamics have been identified in a range of species, including numerous primates (Amici et al. 2008), giraffe (Giraffa camelopardalis; Bercovitch and Berry 2010), African elephant (Loxodonta africana; Couzin 2006), and African buffalo (Syncerus caffer; Cross et al. 2005), though the extent of fission–fusion dynamics may vary with subspecies for the last. Forest buffalo (S. c. nanus) occur in stable herds rarely larger than 20 (Melletti et al. 2007; Korte 2008). West African savanna buffalo (S. c. brachyceros) occur in herds of approximately 50 individuals with very little interaction between herds (Cornélis et al. 2011). Cape buffalo (S. c. caffer) has no characteristic herd size, although they can occur in groups of several thousand individuals (Couzin and Laidre 2009). Most studies of Cape buffalo in southern and eastern Africa have identified large herds with permanent members that periodically subdivide (e.g., Prins 1996; Cross et al. 2005) but consistently occupied identifiable HRs (Ryan et al. 2006). The cohesion of such large herds can vary temporally according to geographical region: during the rainy season, groups in Chobe National Park, Botswana were more fragmented (Taolo 2003), whereas groups in the Serengeti National Park, Tanzania, were larger (Sinclair 1977). Cape buffalo social systems therefore appear to vary with environmental conditions, and the definition of a herd can be problematic. Herds are hereafter defined as assemblages of animals with fixed membership and size that spend most of their time together, whereas groups can vary in these characteristics. In 2012, the Okavango Delta, Botswana, had an estimated 50000 Cape buffalo (unpublished aerial survey, Department of Wildlife and National Parks 2012) in a complex mosaic of different habitat types. The Delta has 2 annual influxes of water: rains falling concurrently in the Okavango Delta and further upstream in Angola produce a delayed flood pulse, which takes several months to flow through rivers in the catchment area before reaching the Okavango Delta (McCarthy and Ellery 1998). This creates high levels of spatial and temporal heterogeneity in resource availability (McCarthy and Ellery 1998). During the rainy season, most herbivore species in the Delta exploit nutrient-rich annual grasses growing in well-drained soils (Fynn et al. 2015), which can support large numbers of herbivores. Outside the rainy season, buffalo rely on seasonal floodplains (Bennitt et al. 2014), which provide water and fresh forage growth when the availability of these resources in other habitats declines progressively after the last rainfall. Migratory and resident buffalo subpopulations with distinct home ranges have been identified in the Okavango Delta (Bennitt et al. 2016), but their social structure has not yet been studied. The subpopulations showed divergent migratory behavior (Bennitt et al. 2016), 1 moving approximately 50 km east during the rainy season, and the other remaining in the same HR year round (Figure 1), although the latter displayed seasonal variation in habitat selection (Bennitt et al. 2014). Much of the previous research on fission–fusion in Cape buffalo has relied on observational data and focused on individual associations (Cross et al. 2005). Technological developments allow continuous remote data collection from numerous individuals and can therefore provide new insights into social behavior (Couzin 2006; Farine and Whitehead 2015). We use data from GPS-enabled collars fitted to 15 buffalo cows in different groups in the Okavango Delta in 2008–2009 to test the hypotheses that: 1) seasonal Utilization Distribution Overlap Index (UDOI) can be used to assign groups containing collared individuals to resident or migratory subpopulations, 2) groups with overlapping seasonal HRs displayed fusion, but only within subpopulations, and 3) sociality, represented by HRO and fusion, would be highest during the rainy season, in open habitat types and closer to water. Figure 1 View largeDownload slide Location of subpopulation ranges within the study area in Botswana. Permanently flooded parts of the study area are shaded. Figure 1 View largeDownload slide Location of subpopulation ranges within the study area in Botswana. Permanently flooded parts of the study area are shaded. METHODS Study site The Okavango Delta covers 15000 km2 in northern Botswana, between 22.0°–24.0°E and 18.5°–20.5°S (Heinl et al. 2006). Average rainfall and flooding patterns were used to define 3 seasons according to fixed dates: the rainy season (December to March); the early flood season, when flood waters were rising (April to July); and the late flood season, when flood waters were receding (August to November). The study was conducted in the south-eastern area of the Okavango Delta, and included permanent and temporary floodplains, as well as permanently dry areas (Figure 1). Habitat types included low shrubland and acacia, mopane and riparian woodlands, which were considered cluttered habitats, and floodplains and grassland, which were considered open habitats. A habitat map of the area was developed by digitizing satellite images at a 1:10000 scale using ArcGIS 10.0 (ESRI, Redlands, CA), and ground-truthing showed an accuracy of 88.1% (Bennitt et al. 2014). The vector habitat map was converted to a raster layer with a 50 × 50 m pixel size, which was defined as the grid to which all other layers snapped. Secondary floodplain was the only habitat type that was flooded all year round, and was therefore identified as a permanent water source and used to generate a raster layer representing distance to permanent water for each pixel in the study area. The location of every ephemeral pan in the study area was recorded from Google Earth (Google Inc., Mountain View, CA), and used to generate a pan density layer. Ephemeral pans can be identified throughout the year by their dark soil and their associated gaps in the tree canopy, so photographic capture date did not influence their detection using Google Earth. CAPTURE AND COLLARING We fitted 15 buffalo cows in different groups with Tellus Simplex 4D GPS-enabled satellite collars (Followit, Lindenberg, Sweden), programmed to record 1 GPS fix per hour. We located buffalo groups from the air and darted opportunistically, so we did not know whether groups were in a state of fusion or fission, but they were spatially distinct at the time of darting. The collars emitted a unique VHF signal, allowing them to be tracked using conventional telemetry equipment, and weighed 1.8 kg, 0.4% of the total mass of the smallest collared buffalo (Bennitt et al. 2014). Cows were selected because they represented breeding herds, and male buffalo tend to lose their collars within a few months of deployment (Taolo 2003). Buffalo were collared for periods of 3–16 months between December 2007 and March 2010, although we only used data from December 2007 to November 2009 to allow annual comparisons of seasonal data (Table 1). Table 1 View largeDownload slide Seasons during which 15 Cape buffalo cows in the Okavango Delta, Botswana were collared Table 1 View largeDownload slide Seasons during which 15 Cape buffalo cows in the Okavango Delta, Botswana were collared There were 24 buffalo darting operations: 15 to fit collars, 2 to replace malfunctioning collars, and 7 to remove collars. A helicopter and a vehicle were used for 22 and 2 darting events, respectively. A vehicle was only used to remove collars with functioning VHF so that we could track the darted animal if we lost visual contact. Sudden malfunctions occurred in 2 collars and we were unable to locate the animals to remove them, but all other collars were recovered by the end of the study. Buffalo were immobilized using either 10 mg M99, 40 mg Azaperone and 5 000 i.u. Hyalaze, reversed with 42 mg M5050 (N = 11), or 8 mg A3080, reversed with Naltrexone (N = 13). We only darted adult females in good condition, not visibly pregnant or with a small calf. Every effort was made to minimize stress and immobilization time, and all animals were observed returning to their herds after darting. All darting operations were carried out by 1 of 3 experienced veterinarians registered in Botswana. Darting permits were obtained prior to collar deployment, based on research permits EWT 3/3/8 XXXVII 44 and EWT 8/36/4 IV 62 issued by the Department of Wildlife and National Parks of Botswana. All capture and handling protocols were approved by the University of Bristol Ethics Committee (UIB/08/034), and conformed to the American Society of Mammalogists’ guidelines for the use of mammals in wildlife research (Sikes 2016) and to the Association for the Study of Animal Behaviour’s guidelines for the use of animals (Association for the Study of Animal Behaviour 2012). Subpopulation identification GPS data from collared animals visually separated them into resident and migratory subpopulations (Bennitt et al. 2016), but we used empirical data from seasonal utilization distributions (UDs) to confirm subpopulation membership and to determine the extent of interaction between and within subpopulations. We calculated buffalo seasonal UDs according to the Movement-Based Kernel Density Estimation method (MKDE; Benhamou 2011), using the “adehabitatHR” package (Calenge 2006) in R v. 3.3.2 (The R Foundation for Statistical Computing, Vienna, Austria, http://www.r-project.org). Sequential GPS coordinates recorded by collars were used to incorporate habitat-specific movement rates into the UD calculations, a technique that was developed using African buffalo (Benhamou and Cornélis 2010), and has previously been used for our study population (Bennitt et al. 2014). To allow for annual and seasonal variation in resource availability, and hence in UD location, we separated the data into 6 time periods: the 2008 and 2009 rainy, early flood and late flood seasons. For each seasonal dataset, we used the “adehabitatHR” package in R v. 3.3.2 to calculate the UDOI for each dyad, the best index for quantifying overlap with regards to shared space use (Fieberg and Kochanny 2005). We used the data cloud geometry-tree (DCG-tree) method to identify inherent patterns of association between buffalo dyads, based on UDOI (Fushing et al. 2013). We constructed a similarity matrix based on UDOI and used the “DCG” package in R v 3.3.2 to build dendrograms depicting the social network of collared buffalo (Fushing et al. 2013), which we used to confirm individual subpopulation membership. We ran a generalized linear mixed model with a Gamma distribution and buffalo dyad as a random variable to determine whether UDOI was affected by season, year or subpopulation. Fusion events A combination of observational and GPS data indicated that collared individuals were more frequently in different groups than in the same group, and they were never observed alone. We have therefore assumed that collared individuals represented groups that occasionally displayed fusion, rather than fragments of a larger group in a state of fission. Hereafter we analyzed factors leading to fusion events, identified as the point at which groups containing collared animals joined, with each event leading to a fusion period that lasted until a fission event occurred, when groups containing collared animals split up. We calculated hourly Euclidean distance between dyads of concurrently collared buffalo. GPS fix success rate was 82.0 ± 9.9% (Bennitt 2012); data were excluded when 1 or both collars in a dyad failed to obtain a fix. Cross et al. (2005) identified fusion events when collared buffalo were within 1 km of each other. We refined this definition by calculating the frequency of occurrence of different Euclidean distances < 1 km across all dyads and generating a histogram, which showed a drop in frequency of observations > 300 m (Figure 2). We therefore defined fusion as periods when dyads were separated by < 300 m, which represented 87.4% of all distances between dyads < 1000 m apart. Since this level of proximity could occur by chance in cluttered habitats, we only considered herds to be in a fusion state when there were at least 3 consecutive fixes for which separation was < 300 m. Some fluidity was allowed, so some fusion events included times when dyads were > 300 apart for ≤ 3 h, after which distance between dyads decreased again. Figure 2 View largeDownload slide Histogram showing frequency of occurrence of Euclidean distance < 1000 m between Cape buffalo dyads in the Okavango Delta, Botswana. The arrow marks the threshold distance for identifying fusion between dyads. Figure 2 View largeDownload slide Histogram showing frequency of occurrence of Euclidean distance < 1000 m between Cape buffalo dyads in the Okavango Delta, Botswana. The arrow marks the threshold distance for identifying fusion between dyads. For each season, we recorded the total number of fusion events per dyad, the duration of each fusion event and total fusion time as a percentage of the entire season. We used the “lme4” package (Bates 2010) in R v. 3.3.2 to run a generalized linear mixed model with a Poisson distribution and a log-link function to determine whether season, year and subpopulation affected the number of fusion events. After testing for normality, we ran a general linear mixed model on the normal log of the duration of each fusion event to determine whether season, year, and subpopulation affected fusion duration. We ran a generalized linear mixed model with a binomial distribution to determine whether total fusion time was affected by season, year, or subpopulation. For all models, we included buffalo dyad as the random effect. We also ran a general linear model with a Poisson distribution to determine whether UDOI was related to number of fusion events. Habitat structure We plotted the start and end GPS coordinates of each fusion period onto the habitat map, corresponding to group fusion and fission events, respectively, to determine whether habitat type influenced group cohesion. We extracted UD outlines from the 95% isopleths of the UDs as HR polygons and identified areas of HRO for each dyad. We compared the proportion of fusion and fission events taking place in open and closed habitats to the proportion of open and closed habitats in the seasonal HRO for each dyad as an analogue to a habitat selection analysis. We also compared the proportions of open and closed habitats in seasonal HROs to those in seasonal HRs to determine whether the location of HROs was linked to habitat type. We calculated Manly selection indices (Manly et al. 2002) in the adehabitatHS package in R v. 3.2.3 and used them to identify habitats that were selected and avoided. Selection ratios with both 95% confidence intervals above and below 1 indicated selection and avoidance, respectively (Manly et al. 2002). Water availability We extracted water availability data from the seasonal HRs and HROs of each buffalo dyad. For rainy season data, we used the pan density layer as a measure of water availability because ephemeral pans were filled with water during that season. We calculated the number of pans within 300 m of each pixel, the distance used to define herd membership, such that each pixel represented the number of pans within 0.28 km2. For the early and late flood seasons, we used the distance to permanent water, since most ephemeral pans were dry during those seasons. The relatively small pixel size meant that a high volume of data was generated for each HR and HRO, so we subsampled the datasets, randomly extracting 1000 values from each. For each seasonal dataset, we used generalized linear mixed models with a binomial distribution and buffalo dyad as a random effect to determine whether water availability affected fusion and fission location, and generalized linear mixed models with a binomial distribution and buffalo dyad as a random effect to test whether water availability differed between seasonal HRs and HROs. All models were run with all combinations of variables, including interaction effects. The most parsimonious models were identified from Akaike’s Information Criterion (Akaike 1974). All statistical analyses were run in R v. 3.3.2. RESULTS There were 6, 4, and 6 buffalo collared during the 2008 rainy, early flood and late flood seasons, respectively, and 6, 7, and 8 buffalo collared during the 2009 rainy, early flood and late flood seasons, respectively (Table 1). Subpopulation identification The dendrograms produced by the DCG-tree method identified 2 main clusters in each season, which for the most part corresponded to the observed resident and migratory subpopulations (Figure 3). During the 2008 early flood season, only 4 individuals were collared and the clustering technique requires 2 individuals per cluster, so B5 was grouped with B6, despite their membership of different clusters in all other seasons. UDOI was therefore greater within subpopulations than between them, and these results were consistent across seasons and years. The resident subpopulation included B1, B2, B5, B8, B9, B10, and B11, whereas the migratory subpopulation included B3, B4, B6, B7, B12, B13, B14, and B15. Figure 3 View largeDownload slide Dendrograms showing hierarchical clustering of 15 Cape buffalo cows in the Okavango Delta, Botswana during the rainy season in (a) 2008 and (b) 2009, the early flood season in (c) 2008 and (d) 2009, and the late flood season in (e) 2008 and (f) 2009. Figure 3 View largeDownload slide Dendrograms showing hierarchical clustering of 15 Cape buffalo cows in the Okavango Delta, Botswana during the rainy season in (a) 2008 and (b) 2009, the early flood season in (c) 2008 and (d) 2009, and the late flood season in (e) 2008 and (f) 2009. Seasonal HRO Within the confines of the subpopulation structure, the maximum number of dyads that could have had overlapping HRs was 6, 3, and 7 in the 2008 rainy, early flood and late flood seasons, respectively, and 7, 9, and 12 in the 2009 rainy, early flood and late flood seasons, respectively. In all 2008 seasons, all of the dyads displayed HRO within subpopulations, but in the 2009 rainy, early flood and late flood seasons, only 6, 5, and 9 dyads, respectively, displayed HRO within subpopulations (Figure 4). There were also 2 and 3 dyads in the 2008 early and late flood seasons, respectively, which displayed HRO between subpopulations, but UDOI for most of these was <0.001. Only 1 dyad, B5B7, had a UDOI value comparable to dyads within subpopulations (0.09). We excluded all of these dyads from subsequent analyses. Figure 4 View largeDownload slide Seasonal home ranges of 15 Cape buffalo cows in the Okavango Delta, Botswana during the rainy season in (a) 2008 and (b) 2009, the early flood season in (c) 2008 and (d) 2009, and the late flood season in (e) 2008 and (f) 2009. Cluttered and open habitats are indicated by darker and lighter gray, respectively. Figure 4 View largeDownload slide Seasonal home ranges of 15 Cape buffalo cows in the Okavango Delta, Botswana during the rainy season in (a) 2008 and (b) 2009, the early flood season in (c) 2008 and (d) 2009, and the late flood season in (e) 2008 and (f) 2009. Cluttered and open habitats are indicated by darker and lighter gray, respectively. Mean ± SD UDOI was 0.23 ± 0.19, 0.09 ± 0.07, and 0.09 ± 0.10 during the 2008 rainy, early flood and late flood seasons, respectively, and 0.03 ± 0.02, 0.41 ± 0.74, and 0.32 ± 0.34 during the 2009 rainy, early flood and late flood seasons, respectively (Table 2). The most parsimonious model included the effect of subpopulation only (AICc = −55.26, AICω = 0.55), although the model with the effects of subpopulation and year was also competitive (AICc = −53.90, AICω = 0.28; Figure 5). Table 2 Seasonal UDOI, number of fusion events, and total fusion time as percentage of season duration for Cape buffalo dyads in the Okavango Delta, Botswana Dyad  Subpopulation  Year  UDOI  N fusion events  Total fusion duration (%)  Rainy  Early flood  Late flood  Rainy  Early flood  Late flood  Rainy  Early flood  Late flood  B1B2  Resident  2008  0.02  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B1B5  Resident  2008  0.07  N/A  N/A  2  N/A  N/A  4.58  N/A  N/A  B2B5  Resident  2008  0.55  N/A  N/A  11  N/A  N/A  26.37  N/A  N/A  B5B8  Resident  2008  N/A  N/A  <0.01  N/A  N/A  0  N/A  N/A  0      2009  0.0  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B8B11  Resident  2009  N/A  0.16  0.38  N/A  7  8  N/A  13.28  19.71  B9B10  Resident  2009  N/A  1.73  1.09  N/A  3  12  N/A  17.59  19.83  B3B4  Migratory  2008  0.24  0.16  0.30  5  3  3  10.79  22.44  18.75      2009  0.04  N/A  N/A  1  N/A  N/A  0.03  N/A  N/A  B3B6  Migratory  2008  0.22  0.10  0.13  5  0  0  18.51  0  0      2009  0.05  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B3B7  Migratory  2009  0.02  N/A  N/A  1  N/A  N/A  0.28  N/A  N/A  B4B6  Migratory  2008  0.28  0.02  0.13  2  3  2  0.65  1.37  0.85      2009  <0.01  N/A  N/A  0  N/A  N/A  0.0  N/A  N/A  B4B7  Migratory  2008  N/A  N/A  0.02  N/A  N/A  0  N/A  N/A  0      2009  0.04  N/A  N/A  1  N/A  N/A  0.24  N/A  N/A  B6B7  Migratory  2008  N/A  N/A  0.09  N/A  N/A  3  0  N/A  0.99      2009  <0.01  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B7B12  Migratory  2009  N/A  <0.01  N/A  N/A  0  N/A  N/A  0  0  B7B13  Migratory  2009  N/A  0.14  N/A  N/A  5  N/A  N/A  7.34  0  B12B13  Migratory  2009  N/A  <0.01  0.14  N/A  0  1  N/A  0  1.06  B12B14  Migratory  2009  N/A  N/A  0.15  N/A  N/A  0  N/A  N/A  0  B12B15  Migratory  2009  N/A  N/A  0.25  N/A  N/A  0  N/A  N/A  0  B13B14  Migratory  2009  N/A  N/A  0.04  N/A  N/A  2  0  N/A  1.40  B13B15  Migratory  2009  N/A  N/A  0.11  N/A  N/A  4  0  0  9.66  B14B15  Migratory  2009  N/A  N/A  0.48  N/A  N/A  5  0  N/A  4.77  Dyad  Subpopulation  Year  UDOI  N fusion events  Total fusion duration (%)  Rainy  Early flood  Late flood  Rainy  Early flood  Late flood  Rainy  Early flood  Late flood  B1B2  Resident  2008  0.02  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B1B5  Resident  2008  0.07  N/A  N/A  2  N/A  N/A  4.58  N/A  N/A  B2B5  Resident  2008  0.55  N/A  N/A  11  N/A  N/A  26.37  N/A  N/A  B5B8  Resident  2008  N/A  N/A  <0.01  N/A  N/A  0  N/A  N/A  0      2009  0.0  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B8B11  Resident  2009  N/A  0.16  0.38  N/A  7  8  N/A  13.28  19.71  B9B10  Resident  2009  N/A  1.73  1.09  N/A  3  12  N/A  17.59  19.83  B3B4  Migratory  2008  0.24  0.16  0.30  5  3  3  10.79  22.44  18.75      2009  0.04  N/A  N/A  1  N/A  N/A  0.03  N/A  N/A  B3B6  Migratory  2008  0.22  0.10  0.13  5  0  0  18.51  0  0      2009  0.05  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B3B7  Migratory  2009  0.02  N/A  N/A  1  N/A  N/A  0.28  N/A  N/A  B4B6  Migratory  2008  0.28  0.02  0.13  2  3  2  0.65  1.37  0.85      2009  <0.01  N/A  N/A  0  N/A  N/A  0.0  N/A  N/A  B4B7  Migratory  2008  N/A  N/A  0.02  N/A  N/A  0  N/A  N/A  0      2009  0.04  N/A  N/A  1  N/A  N/A  0.24  N/A  N/A  B6B7  Migratory  2008  N/A  N/A  0.09  N/A  N/A  3  0  N/A  0.99      2009  <0.01  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B7B12  Migratory  2009  N/A  <0.01  N/A  N/A  0  N/A  N/A  0  0  B7B13  Migratory  2009  N/A  0.14  N/A  N/A  5  N/A  N/A  7.34  0  B12B13  Migratory  2009  N/A  <0.01  0.14  N/A  0  1  N/A  0  1.06  B12B14  Migratory  2009  N/A  N/A  0.15  N/A  N/A  0  N/A  N/A  0  B12B15  Migratory  2009  N/A  N/A  0.25  N/A  N/A  0  N/A  N/A  0  B13B14  Migratory  2009  N/A  N/A  0.04  N/A  N/A  2  0  N/A  1.40  B13B15  Migratory  2009  N/A  N/A  0.11  N/A  N/A  4  0  0  9.66  B14B15  Migratory  2009  N/A  N/A  0.48  N/A  N/A  5  0  N/A  4.77  Dyads were not all collared concurrently throughout seasons. N/A = not applicable. View Large Figure 5 View largeDownload slide Effects of year and subpopulation on Utilization Distribution Overlap Index (UDOI) of Cape buffalo dyads in the Okavango Delta, Botswana. The circles denote means and the whiskers denote SEs. Figure 5 View largeDownload slide Effects of year and subpopulation on Utilization Distribution Overlap Index (UDOI) of Cape buffalo dyads in the Okavango Delta, Botswana. The circles denote means and the whiskers denote SEs. Fusion events Despite overlapping HRs, not all dyads displayed fusion: 83.3%, 66.7%, and 42.9% of dyads fused during the 2008 rainy, early flood and late flood seasons, respectively, and 28.6%, 33.3%, and 55.6% of dyads fused during the 2009 rainy, early flood and late flood seasons, respectively. Mean ± SD number of fusion events per dyad was 5.4 ± 3.8, 3.0 ± 0.0, and 2.7 ± 0.6 during the 2008 rainy, early flood, and late flood seasons, respectively, and 1.0 ± 0.0, 5.0 ± 2.0, and 5.5 ± 4.4 during the 2009 rainy, early flood, and late flood seasons, respectively (Table 2). The most parsimonious model included the effect of subpopulation only (AICc = 161.39), but the null model was also competitive (AICc = 162.65) which indicated that none of season, year or subpopulation had any effect on the number of fusion events per dyad. Only 1 dyad containing individuals from different subpopulations, B5B7, displayed fusion, for 155 h during the 2008 late flood season. Mean ± SD duration of fusion events was 63.7 ± 122.0, 74.1 ± 119.6, and 75.4 ± 118.6 h during the 2008 rainy, early flood, and late flood seasons, respectively, and 7.5 ± 0.7, 74.6 ± 126.3, and 75.3 ± 69.8 h during the 2009 rainy, early flood, and late flood seasons, respectively. The most parsimonious model was the null model (AICc = 193.82), indicating that none of season, year or subpopulation had any effect on fusion duration. Mean ± SD total fusion duration as a percentage of each season was 12.2 ± 10.4%, 11.9 ± 14.9%, and 6.8 ± 10.3% during the 2008 rainy, early flood, and late flood seasons, respectively, and 0.3 ± 0.0%, 12.7 ± 5.1%, and 14.6 ± 18.6% during the 2009 rainy, early flood, and late flood seasons, respectively. 1 dyad, B9B10, spent approximately 61 days or 49.8% of the 2009 late flood season in fusion, and the 2nd longest cumulative fusion duration was approximately 32 days or 26.4% of the 2008 rainy season, for B2B5 (Table 2). The most parsimonious model included an interaction effect between season and year, and the fixed effect of subpopulation (AICc = 389.78, AICω = 0.54), but the model with only the interaction effect between season and year was also competitive (AICc = 390.13, AICω = 0.46; Figure 6). Number of fusion events was not affected by UDOI: the null model was the most parsimonious (AICc = 162.65). Figure 6 View largeDownload slide Effects of season and year on total fusion time as a percentage of season duration for Cape buffalo dyads in the Okavango Delta, Botswana. The circles denote means and the whiskers denote SEs. Figure 6 View largeDownload slide Effects of season and year on total fusion time as a percentage of season duration for Cape buffalo dyads in the Okavango Delta, Botswana. The circles denote means and the whiskers denote SEs. There were 2 time periods when 3 collared animals were in the same herd, both during the 2008 rainy season. Fusion was recorded for B3B4 and B3B6, but not for B4B6, over a 35-h period of concurrent fusion for both dyads. During a 405-h fusion event for B3B6, B3B4 showed fusion for 40 h, but again no fusion was recorded for B4B6. It is therefore possible that some fusion events were undetected by the 300 m threshold, but increasing the threshold to an arbitrary distance may have captured false fusion events when dyads were close to each other without being in the same group. Habitat structure The relatively small sample size for dyads displaying fusion periods did not allow the data to be split between years, so we grouped seasonal data to compare habitat types at fission and fusion events with available habitat in HROs. We did not identify any seasons when fusion events were more likely to take place in open or closed habitats (Table 3). Fission events in the rainy and early flood seasons took place in open and cluttered habitats in proportion to their availability, but during the late flood season, fission was more likely in cluttered habitats and less likely in open habitats. Table 3 Habitat selection ratios for locations of fission and fusion events and habitat composition of seasonal home range overlap for Cape buffalo in the Okavango Delta, Botswana   Rainy  Early flood  Late flood  Location  Number  Open  Cluttered  Number  Open  Cluttered  Number  Open  Cluttered  Fission  7  1.12 (−2.29–2.53)  0.96 (0.49–1.43)  5  0.67 (0.34–1.01)  2.17 (−0.30–4.64)  8  0.63 (0.35–0.91)  2.26 (1.13–3.39)  Fusion  7  1.69 (−0.25–3.63)  0.76 (0.30–1.21)  5  0.85 (0.55–1.14)  1.56 (0.15–2.96)  8  0.87 (0.62–1.11)  1.44 (0.47–2.41)  Home range overlap 2008  12  0.91 (0.74–1.08)  1.06 (0.95–1.17)  6  1.15 (1.02–1.28)  0.17 (−0.00–0.35)  12  1.08 (1.01–1.15)  0.39 (0.08–0.71)  Home range overlap 2009  12  1.63 (1.26–2.01)  0.95 (0.93–0.98)  10  1.06 (0.98–1.14)  0.82 (0.54–1.11)  18  1.16 (1.05–1.26)  0.69 (0.50–0.88)    Rainy  Early flood  Late flood  Location  Number  Open  Cluttered  Number  Open  Cluttered  Number  Open  Cluttered  Fission  7  1.12 (−2.29–2.53)  0.96 (0.49–1.43)  5  0.67 (0.34–1.01)  2.17 (−0.30–4.64)  8  0.63 (0.35–0.91)  2.26 (1.13–3.39)  Fusion  7  1.69 (−0.25–3.63)  0.76 (0.30–1.21)  5  0.85 (0.55–1.14)  1.56 (0.15–2.96)  8  0.87 (0.62–1.11)  1.44 (0.47–2.41)  Home range overlap 2008  12  0.91 (0.74–1.08)  1.06 (0.95–1.17)  6  1.15 (1.02–1.28)  0.17 (−0.00–0.35)  12  1.08 (1.01–1.15)  0.39 (0.08–0.71)  Home range overlap 2009  12  1.63 (1.26–2.01)  0.95 (0.93–0.98)  10  1.06 (0.98–1.14)  0.82 (0.54–1.11)  18  1.16 (1.05–1.26)  0.69 (0.50–0.88)  Significant results are in bold. View Large The larger sample sizes for dyads with overlapping HRs allowed us to split the data by year as well as season. There was no difference between habitat composition in the HRs and HROs during the 2008 rainy season and the 2009 early flood season, but there was more open and less cluttered habitat in HROs than HRs during the other seasons (Table 3). Water availability The relatively small sample size for dyads displaying fusion events did not allow the data to be split between years, so we grouped seasonal data to compare water availability at fission and fusion locations. During the rainy season, mean ± SD pan density was 6.8 ± 6.5 and 3.6 ± 4.1 pans/0.28 km2 at fission and fusion locations, respectively. The most parsimonious model included the effect of pan density (AICc = 99.8, AICω = 0.89), indicating that pan density was higher at fission than fusion event locations during the rainy season. During the early flood season, mean ± SD distance to permanent water was 302.7 ± 348.5 m and 223.7 ± 335.2 m at fusion and fission event locations, respectively. The most parsimonious model was the null model (AICc = 62.5), indicating that there was no effect of water availability on fission/fusion during the early flood season. During the late flood season, mean ± SD distance to permanent water was 1275.4 ± 4177.3 m and 1639.2 ± 5264.1 m at fusion and fission locations, respectively. The most parsimonious model was the null model (AICc = 123.4), indicating that there was no effect of water availability on fission/fusion during the late flood season. The larger sample sizes for dyads with overlapping HRs allowed us to split the data by year as well as season. During the 2008 rainy season, mean ± SD pan density was 6.0 ± 8.5 and 6.6 ± 9.0 pans/0.28 km2 in HRs and HROs, respectively. The most parsimonious model included the effect of pan density (AICc = 2783.36, AICω = 1.00), indicating that pan density was higher in HROs than in HRs during the 2008 rainy season. During the 2009 rainy season, mean ± SD pan density was 6.4 ± 9.7 and 5.0 ± 6.8 pans/0.28 km2 in HRs and HROs, respectively. The most parsimonious model included the effect of pan density (AICc = 32942.47, AICω = 1.00), indicating that pan density was higher in HRs than in HROs during the 2009 rainy season. During the 2008 early flood season, mean ± SD distance to permanent water was 1526.5 ± 5438.9 m and 94.0 ± 151.8 m in the HRs and HROs, respectively. The most parsimonious model included the effect of distance to permanent water (AIC = 15480.17, AICω = 1.00), indicating that water availability was lower in HRs than in HROs during the 2008 early flood season. During the 2009 early flood season, mean ± SD distance to permanent water was 1526.6 ± 4364.9 m and 394.6 ± 577.5 m in the HRs and HROs, respectively. The most parsimonious model included the effect of distance to permanent water (AICc = 26834.63, AICω = 1.00), indicating that water availability was lower in HRs than in HROs during the 2009 early flood season. During the 2008 late flood season, mean ± SD distance to permanent water was 778.6 ± 3792.8 m and 77.8 ± 186.6 m in the HRs and HROs, respectively. The most parsimonious model included the effect of distance to permanent water (AICc = 37576.9, AICω = 1.00), indicating that water availability was lower in HRs than in HROs during the 2008 late flood season. During the 2009 late flood season, mean ± SD distance to permanent water was 1585.9 ± 3400.1 m and 1046.7 ± 1447.1 m in the HRs and HROs, respectively. The most parsimonious model included the effect of distance to permanent water (AICc = 62891.4, AICω = 1.00), indicating that water availability was lower in HRs than in HROs during the 2009 late flood season. DISCUSSION Understanding animal social systems provides in-depth knowledge of possible routes for disease transmission (Body et al. 2015), information sharing, and gene flow (Biosa et al. 2015), and allows the identification of key processes and pathways for population connectivity and health (Morales et al. 2010). Previous studies of African buffalo in different parts of their range have recorded a variety of social systems, from steady herds with permanent membership (Melletti et al. 2007) to more fluid groups occupying identifiable home ranges (Ryan et al. 2006). Our results show that, in the Okavango Delta, UDOI could be used to identify subpopulation membership for individual collared animals and fusion between individuals in different subpopulations was rare. Number of fusion events and duration of fusion periods did not vary seasonally, total fusion time varied seasonally and annually, and UDOI varied with year and subpopulation. Fission events were more likely in cluttered habitats, but only in the late flood season, and HROs contained more open than cluttered habitats compared to HRs in most seasons. Pan density was higher at fission than fusion events sites during the rainy season, but water availability had no effect on fission/fusion in other seasons. Pan density in rainy season HROs was lower and higher than in HRs in 2008 and 2009, respectively, and in all flood seasons, HRO areas were closer to permanent water than HRs. Buffalo were not collared concurrently, so we could have deployed a collar in a group that had previously contained a collared animal. However, the high levels of social fluidity and environmental variation suggest that HR locations may have varied between years, which would also have led to variation in HRO locations and levels of interaction with neighbors, as indicated by the different rainy season fusion duration between 2008 and 2009, when data from 4 of the same individuals were used. Hourly GPS fixes provide large datasets, but they are still relatively coarse and cannot be used to identify precise moments when buffalo groups became aware of each other and decided whether to approach and fuse. Equally, the moment of fission is very difficult to identify because consecutive Euclidean distances between collared buffalo could shift from < 300 m to ≥ 1000 m over a 1 h period, and it was not possible to identify exactly when or where fission occurred. Indeed, when 3 collared animals were in the same group, our methods detected concurrent fusion between 2 dyads, but did not identify a fusion pattern for the 3rd. Some fusion events may therefore have gone undetected based on our criteria, perhaps for very large groups in which collared animals could have been >300 m apart, but raising the threshold distance for defining fusion would have been arbitrary and could have led to the inclusion of false fusion events when collared individuals were close to each other without being in the same group. Also, some fission–fusion events could have occurred without being recorded because they involved noncollared animals, but it is unlikely that the behavior of collared animals was substantially different from that of other buffalo within the study area. There were no impermeable barriers between the subpopulations: buffalo could cross channels and resident herds could follow the migration route taken by the migratory subpopulation (Bennitt et al. 2016). The resident subpopulation may have migrated in a different direction prior to the erection of a veterinary fence in 1982 (Mbaiwa and Mbaiwa 2006), and they may not have the spatial memory or cognitive capacity to adjust to changing conditions caused by the erection of a hard boundary. Resident buffalo frequently moved along the fence during the rainy season, suggesting that they would have pushed further if the fence had not been there. Habitats on either side of the fence were similarly dominated by mopane woodlands and grasslands, possibly forming a corridor that buffalo could have followed ancestrally. Competition between the subpopulations could have led to the observed spatial distribution, but the migratory subpopulation left their flood season HR during the rainy season, leaving almost entirely unoccupied habitat that the resident subpopulation could have exploited during the rainy season but did not. The very low levels of HRO between subpopulations indicate that their encounter levels were too low for competition because they rarely converged on shared resources. Competition within subpopulations would have been possible in HRO areas, but our results suggested that groups congregated and fused in areas with key resources, rather than competitively excluding other groups. The subpopulations displayed similar reproductive productivity and body condition, despite smaller resident HRs and group sizes (Bennitt et al. 2016), suggesting that resource availability in the subpopulation HRs was comparable and that there was no strong driver for interaction between subpopulations. This suggests that there may have been social reasons for the disparate migratory strategies, possibly related to spatial information passed down through interaction with other subpopulation members. We hypothesized that the abundance and wide distribution of water and productive forage across the landscape during the rainy season would lead to higher levels of fusion, but the frequency of fusion events and duration of fusion periods did not appear to be affected by environmental factors such as seasonal changes in resources. Group sizes may have increased with males joining breeding groups to exploit mating opportunities during the rainy season, but fusion behavior between collared females was not affected. Buffalo occupied more cluttered habitats during the rainy season (Bennitt et al. 2014), so predation pressure could have reduced optimal group size, leading to fission events (Fortin et al. 2009). When the flood waters recede during the limiting late flood season, nutrient-poor floodplain grasses offer the only fresh forage growth (Murray-Hudson et al. 2006), and this could drive buffalo to congregate in areas of limited resources, thereby leading to group fusion (Bercovitch and Berry 2010). Total time spent in fusion varied with season and year, being lowest during the 2009 rainy season, when there appears to have been a drop in environmental pressures that would have caused fusion. Rainfall in the 2009 rainy season started early (Okavango Research Institute weather monitoring website, http://www.okavangodata.ub.bw/ori/monitoring/rainfall/), possibly leading to rapid growth of productive vegetation that would also have allowed buffalo to be in a state of fission. UDOI varied with subpopulation and, to some extent, with year: UDOI was higher in the resident subpopulation, especially in 2009. The resident subpopulation utilized a smaller area than the migratory subpopulation, which could have caused the higher levels of HRO, but there was also one dyad, B9B10, collared in the resident subpopulation in 2009, which had much higher UDOI than any other dyad, and this could have affected these results. UDOI did not have an effect on the number of fusion events between dyads, even though encounters between dyads with extensive HRO would probably have been more frequent because population density in HRO areas should have been higher (Kolbe and Weckerly 2015). Indeed, elk (Cervus elaphus) encounter rates increased with HRO, although overall population density also had a strong effect (Vander Wal et al. 2014). Some species with high levels of HRO partition the overlap temporally by utilizing shared resources at different times (Markham et al. 2015), but our results suggest that buffalo were utilizing shared resources at the same time, potentially leading to opportunistic fusion events (Bauder et al. 2016). In contrast to expectations, fusion was not more common in open habitats, and there was only 1 season when fission was more frequent in cluttered habitats. However, in 4 of the 6 seasons, the proportion of open habitat was higher in HROs than in HRs. Open habitats included grassland and floodplains, which are all favored grazing habitats (Bennitt et al. 2015), and the latter were identified as sources of permanent water. Therefore these habitats provide key resources and would form vital components of HRs, possibly leading to overlapping patterns of use by 2 or more individuals. The only time of year when there were differences in water availability between fusion and fission sites was during the rainy season, when pan density was higher in the latter. Pan density was also higher in HRs than in HROs during the 2009 rainy season, despite being slightly lower in HRs than in HROs in the 2008 rainy season. Pans are more common in poorly-drained soils where water can accumulate, such as in mopane woodland, but these are often not the most productive soil types in terms of herbaceous resources (Bennitt et al. 2016). Buffalo HRs may therefore have overlapped in areas with high quality or abundant forage, which may also have been areas with lower water availability, as seen in semi-free-ranging cattle (Bos taurus; Zengeya and Murwira 2016). The wide distribution of abundant pans across the landscape would have removed the limiting factor of water availability present during the flood seasons. In contrast, during the early and late flood seasons, HROs were closer to permanent water than the rest of the seasonal HRs. As obligate drinkers, buffalo need access to water every day, and their movements are therefore restricted by water availability outside the rainy season. During the late flood season, forage is senescent in dry habitats, but fresh grasses grow on the floodplains as the floodwaters recede (Murray-Hudson et al. 2006), providing an additional attractant for buffalo. Seasonal variation in forage and water availability in a flood-pulsed ecosystem such as the Okavango Delta may have exerted different pressures on buffalo that required similar levels of spatial and social interaction, such as the availability of abundant but spatially restricted resources. Cape buffalo in the Okavango Delta therefore behaved differently to other populations in eastern (Sinclair 1977) and southern Africa (Taolo 2003), maintaining similar levels of group cohesion throughout the year. We were unable to record group sizes accurately during the rainy season because of low visibility, so groups could have been larger at that time, but there were no differences in group size between the early and late flood seasons, when median group size was 100–200 individuals (Bennitt et al. 2016). CONCLUSIONS Our results show that Cape buffalo in the Okavango Delta form fluid groups within relatively distinct subpopulations that occupy separate HRs and show very little social interaction, and that fusion events within subpopulations occur in HROs located in areas with key environmental resources. Buffalo social systems vary in different parts of their global range, indicating a disparity in social complexity that may be linked to resource availability. Previous studies of buffalo social systems in eastern and southern Africa have described several large herds that subdivide (Prins 1996; Cross et al. 2005), typical of species displaying fission–fusion social dynamics (Sueur and Maire 2014). Our results suggest that buffalo in the Okavango Delta form numerous small groups that occasionally fuse into larger aggregations when they congregate on key resources, which appear to relate to forage during the rainy season and water during the early and late flood seasons. Our results therefore call into question the definition of a “herd”, particularly in reference to animals that are difficult to identify individually and occur in large aggregations, such as Cape buffalo. We suggest that the term “herd” should be used with caution, and that groups of animals with undetermined levels of cohesion and membership permanence should be called “groups”. Previous research has shown a higher level of sociality between dyads with high levels of spatial overlap (Chaverri et al. 2007), but it is difficult to distinguish cause and effect: animals may be attracted to each other, leading them to share space, or they may be attracted to resources, leading to higher encounter rates (Best et al. 2014). Our results suggest that buffalo HRs overlapped in areas with key seasonal resources, which led to fusion events followed by fusion periods when all animals in a herd made the same collective foraging decisions (Ramos-Fernández et al. 2006). Such spatial causes of higher levels of group cohesion and social interaction could be more important than kinship and relatedness (Best et al. 2014), and it is possible that, once fusion had occurred, herds moved together for several hours, allowing social interaction between herd members. Fission could ensue when key individuals differed in their decisions about the next foraging location, and were joined by regular followers, leading to a group split (Ramos-Fernández and Morales 2014). From aerial observations, the resident and migratory subpopulations in our study area were estimated at 5000 and 8000 buffalo, respectively (E.B., Personal observation, 2007–2010), group sizes that are too large to be sustained by clumped resources. Indeed, total seasonal fusion duration showed that collared buffalo were in a fission state for the majority of their time, with only 1 dyad spending almost 50% of 1 season in a fusion state. The lack of interfaces between subpopulations suggests that they are relatively distinct in terms of spatial patterns and social interactions, which could have consequences for gene flow and the transmission of adaptive behavior. These 2 subpopulations, despite occupying HRs with common boundaries, could therefore constitute a meta-population, with minimal contact between subpopulations experiencing different environmental pressures, and further research should be conducted to determine the level of gene flow between subpopulations. Populations displaying fission–fusion social dynamics can be more vulnerable to anthropogenic impacts, as well as being more difficult to count reliably (Haulsee et al. 2016). Dynamic levels of social cohesion could mean that some meta-populations are more at risk, whereas others are safer, decreasing their vulnerability to the effects of stochastic environmental change through increased variability (Holmes et al. 2016). Disruptions from anthropogenic influences or stochastic climatic variation could have large impacts on buffalo in the Okavango Delta through a reduction in access to seasonal ranges (Bennitt et al. 2014), particularly for buffalo on the periphery of the ecosystem. Peripheral subpopulations are also more vulnerable to human-wildlife conflict (Bennitt et al. 2014) and could encounter ecological traps, whereby incomplete knowledge of risks lead herds to utilize resources in areas with higher levels of human activity and hence increased mortality (Sigaud et al. 2017). Cape buffalo in the Okavango Delta should therefore be managed as meta-populations exposed to different ecological pressures and threats rather than 1 continuous population with homogeneous mixing of genetic material (Titcomb et al. 2015). FUNDING This work was supported by Jenny and Martin Bennitt; the Dulverton Trust; Harry Ferguson; Ian Fuhr; Rodney Fuhr; Dane Hawk; Idea Wild; the North of England Zoological Society; the Roberts Fund; and Wilderness Safaris Wildlife Trust. The authors thank the Botswana Ministry of Environment, Wildlife and Tourism for permission to conduct this study; Dane Hawk, Rob Jackson, and Larry Patterson for veterinary support; Laura Atkinson, Janette Baarman, Roz Balen, Shavaughn Davies, and Jennifer Gilbert for assistance in the field; and Guy Lobjoit and Duncan Rowles for logistical support. Data accessibility: Analyses reported in this article can be reproduced using the data provided by Bennitt et al. (2017). REFERENCES Akaike H. 1974. A new look at the statistical model identification. IEEE T Automat Cont . 19: 716– 723. Google Scholar CrossRef Search ADS   Alexander RD. 1974. The evolution of social behavior. Annu Rev Ecol Syst . 5: 325– 383. Google Scholar CrossRef Search ADS   Amici F, Aureli F, Call J. 2008. Fission-fusion dynamics, behavioral flexibility, and inhibitory control in primates. Curr Biol . 18: 1415– 1419. Google Scholar CrossRef Search ADS PubMed  Association for the Study of Animal Behaviour. 2012. Guidelines for the treatment of animals in behavioural research and teaching. Anim Behav . 83: 301– 309. CrossRef Search ADS   Aureli F, Schaffner CM, Boesch C, Bearder SK, Call J, Chapman CA, Connor R, Di Fiore A, Dunbar RIM, Henzi SPet al.   2008. Fission-fusion dynamics: new research frameworks. Curr Anthropol . 49: 627– 654. Bates D. 2010. lme4: mixed-effects modeling with R. www.R-project.org/. Accessed 23 March 2017. Bauder JM, Breininger DR, Bolt MR, Legare ML, Jenkins CL, Rothermel BB, McGarigal K. 2016. The influence of sex and season on conspecific spatial overlap in a large, actively-foraging colubrid snake. PLoS ONE . 11: e0160033. Google Scholar CrossRef Search ADS PubMed  Benhamou S, Cornélis D. 2010. Incorporating movement behavior and barriers to improve kernel home range space use estimates. J Wild Manage . 74: 1353– 1360. Google Scholar CrossRef Search ADS   Benhamou S. 2011. Dynamic approach to space and habitat use based on biased random bridges. PLoS ONE . 6: e14592. Google Scholar CrossRef Search ADS PubMed  Bennitt E. 2012. The ecology of African buffalo (Syncerus caffer) in the Okavango Delta, PhD Thesis , Botswana: University of Bristol. Bennitt E, Bonyongo MC, Harris S. 2014. Habitat selection by African buffalo (Syncerus caffer) in response to landscape-level fluctuations in water availability on two temporal scales. PLoS ONE . 9: e101346. Google Scholar CrossRef Search ADS PubMed  Bennitt E, Bonyongo MC, Harris S. 2015. Behaviour-related scalar habitat use by cape buffalo (Syncerus caffer caffer). PLoS ONE . 10: e0145145. Google Scholar CrossRef Search ADS PubMed  Bennitt E, Bonyongo MC, Harris S. 2016. Effects of divergent migratory strategies on access to resources for cape buffalo (Syncerus caffer caffer). J Mammal . 97: 1682– 1698. Google Scholar CrossRef Search ADS   Bennitt E, Bonyongo C, Harris S. Data from: cape buffalo (Syncerus caffer caffer) social dynamics in a flood-pulsed environment. Dryad Digital Repository . http://dx.doi.org/10.5061/dryad.d8j34 Bercovitch FB, Berry PSM. 2010. Ecological determinants of herd size in the thornicroft’s giraffe of zambia. Afr J Ecol . 48: 962– 971. Google Scholar CrossRef Search ADS   Best EC, Dwyer RG, Seddon JM, Goldizen AW. 2014. Associations are more strongly correlated with space use than kinship in female eastern grey kangaroos. Anim Behav . 89: 1– 10. Google Scholar CrossRef Search ADS   Biosa D, Grignolio S, Sica N, Pagon N, Scandura M, Apollonio M. 2015. Do relatives like to stay closer? spatial organization and genetic relatedness in a mountain roe deer population. J Zool . 296: 30– 37. Google Scholar CrossRef Search ADS   Body G, Weladji RB, Holand Ø, Nieminen M. 2015. Measuring variation in the frequency of group fission and fusion from continuous monitoring of group sizes. J Mammal . 96: 791– 799. Google Scholar CrossRef Search ADS   Calenge C. 2006. The package “adehabitat” for the R software: a tool for the analysis of space and habitat use by animals. Ecol Model . 197: 516– 519. Google Scholar CrossRef Search ADS   Chaverri G, Gambia-Rios M, Kunz TH. 2007. Range overlap and association patterns in the tent-making bat Artibeus watsoni. Anim Behav . 73: 157– 164. Google Scholar CrossRef Search ADS   Conradt L, Roper TJ. 2000. Activity synchrony and social cohesion: a fission-fusion model. Proc Roy Soc B . 267: 2213– 2218. Google Scholar CrossRef Search ADS   Cornélis D, Benhamou S, Janeau G, Morellet N, Ouedraogo M, de Visscher M-N. 2011. Spatiotemporal dynamics of forage and water resources shape space use of West African savanna buffaloes. J Mammal . 92: 1287– 1297. Google Scholar CrossRef Search ADS   Couzin ID. 2006. Behavioral ecology: social organization in fission-fusion societies. Curr Biol . 16: R169– R171. Google Scholar CrossRef Search ADS PubMed  Couzin ID, Laidre ME. 2009. Fission-fusion populations. Curr Biol . 19: R633– R635. Google Scholar CrossRef Search ADS PubMed  Cross PC, Lloyd-Smith JO, Getz WM. 2005. Disentangling association patterns in fission-fusion societies using African buffalo as an example. Anim Behav . 69: 499– 506. Google Scholar CrossRef Search ADS   Farine DR, Whitehead H. 2015. Constructing, conducting and interpreting animal social network analysis. J Anim Ecol . 84: 1144– 1163. Google Scholar CrossRef Search ADS PubMed  Fieberg J, Kochanny CO. 2005. Quantifying home-range overlap: the importance of the utilization distribution. J Wild Manage . 69: 1346– 1359. Google Scholar CrossRef Search ADS   Fortin D, Fortin M-E, Beyer HL, Duchesne T, Courant S, Dancose K. 2009. Group-size-mediated habitat selection and group fusion-fission dynamics of bison under predation risk. Ecology . 90: 2480– 2490. Google Scholar CrossRef Search ADS PubMed  Fushing H, Wang H, VanderWaal K, McCowan B, Koehl P. 2013. Multi-scale clustering by building a robust and self correcting ultrametric topology on data points. PLoS ONE . 8: e56259. Google Scholar CrossRef Search ADS PubMed  Fynn RWS, Murray-Hudson M, Dhliwayo M, Scholte P. 2015. African wetlands and their seasonal use by wild and domestic herbivores. Wetlands Ecol Manage . 23: 559– 581. Google Scholar CrossRef Search ADS   Haulsee DE, Fox DA, Breece MW, Brown LM, Kneebone J, Skomal GB, Oliver MJ. 2016. Social network analysis reveals potential fission-fusion behavior in a shark. Sci Rep . 6: 34087. Heinl M, Neuenschwander A, Silva J, Vanderpost C. 2006. Interactions between fire and flooding in a southern African floodplain system (Okavango Delta, Botswana). Landscape Ecol . 21: 699– 709. Google Scholar CrossRef Search ADS   Holmes SM, Gordon AD, Louis EE, Johnson SE. 2016. Fission-fusion dynamics in black-and-white ruffed lemurs may facilitate both feeding strategies and communal care of infants in a spatially and temporally variable environment. Behav Ecol Sociobiol . 70: 1949– 1960. Google Scholar CrossRef Search ADS   Isvaran K. 2007. Intraspecific variation in group size in the blackbuck antelope: the roles of habitat structure and forage at different spatial scales. Oecologia . 154: 435– 444. Google Scholar CrossRef Search ADS PubMed  Jarman PJ. 1974. The social organisation of antelope in relation to their ecology. Behaviour . 48: 215– 267. Google Scholar CrossRef Search ADS   Kelley JL, Morrell LJ, Inskip C, Krause J, Croft DP. 2011. Predation risk shapes social networks in fission-fusion populations. PLoS ONE . 6: e24280. Google Scholar CrossRef Search ADS PubMed  Kie JG. 1999. Optimal foraging and risk of predation: effects on behavior and social structure in ungulates. J Mammal . 80: 1114– 1129. Google Scholar CrossRef Search ADS   Kolbe NR, Weckerly FW. 2015. Home-range overlap of roosevelt elk herds in the bald hills of redwood national park. Calif Fish Game . 101: 208– 217. Korte LM. 2008. Variation of group size among African buffalo herds in a forest-savanna mosaic landscape. J Zool . 275: 229– 236. Google Scholar CrossRef Search ADS   Manly BFJ, McDonald LL, Thomas DL, McDonald TL, Erickson WP. 2002. Resource selection by animals: statistical design and analysis for field studies . Dordrecht: Kluwer Academic Publishers. Markham AC, Gesquiere LR, Alberts SC, Altmann J. 2015. Optimal group size in a highly social mammal. PNAS . 112: 14882– 14887. Google Scholar CrossRef Search ADS PubMed  Mbaiwa JE, Mbaiwa OI. 2006. The effects of veterinary fences on wildlife populations in the Okavango Delta, Botswana. Int J Wilderness . 12: 17– 41. McCarthy TS, Ellery WN. 1998. The Okavango Delta. Trans R Soc S Afr . 53: 157– 182. Google Scholar CrossRef Search ADS   Melletti M, Penteriani V, Mirabile M, Boitani L. 2007. Some behavioral aspects of forest buffalo (Syncerus caffer nanus): from herd to individual. J Mammal . 88: 1312– 1318. Google Scholar CrossRef Search ADS   Merkle JA, Sigaud M, Fortin D. 2015. To follow or not? how animals in fusion-fission societies handle conflicting information during group decision-making. Ecol Lett . 18: 799– 806. Google Scholar CrossRef Search ADS PubMed  Morales JM, Moorcroft PR, Matthiopoulos J, Frair JL, Kie JG, Powell RA, Merrill EH, Haydon DT. 2010. Building the bridge between animal movement and population dynamcs. Phil Trans R Soc B . 365: 2289– 2301. Google Scholar CrossRef Search ADS PubMed  Mourier J, Vercelloni J, Planes S. 2012. Evidence of social communities in a spatially structured network of a free-ranging shark species. Anim Behav . 83: 389– 401. Google Scholar CrossRef Search ADS   Murray-Hudson M, Wolski P, Ringrose S. 2006. Scenarios of the impact of local and upstream changes in climate and water use on hydro-ecology in the Okavango Delta, Botswana. J Hydrology . 331: 73– 84. Google Scholar CrossRef Search ADS   Owen-Smith N, Fryxell JM, Merrill EH. 2010. Foraging theory upscaled: the behavioural ecology of herbivore movement. Phil Trans R Soc B . 365: 2267– 2278. Google Scholar CrossRef Search ADS PubMed  Prins HHT. 1996. Ecology and behaviour of the African buffalo: social inequality and decision making . London: Chapman & Hall. Google Scholar CrossRef Search ADS   Ramos A, Petit O, Longour P, Pasquaretta C, Sueur C. 2015. Collective decision making during group movements in European bison, Bison bonasus. Anim Behav . 109: 149– 160. Google Scholar CrossRef Search ADS   Ramos-Fernández G, Boyer D, Gómez VP. 2006. A complex social structure with fission-fusion properties can emerge from a simple foraging model. Behav Ecol Sociobiol . 60: 536– 549. Google Scholar CrossRef Search ADS   Ramos-Fernández G, Morales JM. 2014. Unraveling fission-fusion dynamics: how subgroup properties and dyadic interactions influence individual decisions. Behav Ecol Sociobiol . 68: 1225– 1235. Google Scholar CrossRef Search ADS   Ryan SJ, Knechtel CU, Getz WM. 2006. Range and habitat selection of African buffalo in South Africa. J Wildl Manage . 70: 764– 776. Google Scholar CrossRef Search ADS   Sigaud M, Merkle JA, Cherry SG, Fryxell JM, Berdahl A, Fortin D. 2017. Collective decision-making promotes fitness loss in a fusion-fission society. Ecol Lett . 20: 33– 40. Google Scholar CrossRef Search ADS PubMed  Sikes RS; Animal Care and Use Committee of the American Society of Mammalogists. 2016. 2016 guidelines of the american society of Mammalogists for the use of wild mammals in research and education. J Mammal . 97: 663– 688. Google Scholar CrossRef Search ADS   Sinclair ARE. 1977. The African buffalo: a study of resource limitation of populations . Chicago: University of Chicago Press. Sueur C, King AJ, Conradt L, Kerth G, Lusseau D, Mettke-Hofmann C, Schaffner CM, Williams L, Zinner D, Aureli F. 2011. Collective decision-making and fission-fusion dynamics: a conceptual framework. Oikos . 120: 1608– 1617. Google Scholar CrossRef Search ADS   Sueur C, Maire A. 2014. Modelling animal group fission using social network dynamics. PLoS ONE . 9: e97813. Google Scholar CrossRef Search ADS PubMed  Taolo C. 2003. Population ecology, seasonal movement and habitat use of the African buffalo (Syncerus caffer) in Chobe National Park, Botswana . PhD Thesis, Trondheim: Norwegian University of Science and Technology. Titcomb EM, O’Corry-Crowe G, Hartel EF, Mazzoil MS. 2015. Social communities and spatiotemporal dynamics of association patterns in estuarine bottlenose dolphins. Mar Mamm Sci . 31: 1314– 1337. Google Scholar CrossRef Search ADS   Vander Wal E, Laforge MP, McLoughlin PD. 2014. Density dependence in social behaviour: home range overlap and density interacts to affect conspecific encounter rates in a gregarious ungulate. Behav Ecol Sociobiol . 68: 383– 390. Google Scholar CrossRef Search ADS   VanderWaal KL, Wang H, McCowan B, Fushing H, Isbell LA. 2014. Multilevel social organization and space use in reticulated giraffe (Giraffa camelopardalis). Behav Ecol . 25: 17– 26. Google Scholar CrossRef Search ADS   Wittemyer G, Douglas-Hamilton, I, Getz WM. 2005. The socioecology of elephants: analysis of the processes creating multitiered social structures. Anim Behav . 69: 1357– 1371. Google Scholar CrossRef Search ADS   Wolf M, Frair J, Merrill E, Turchin P. 2009. The attraction of the known: the importance of spatial familiarity in habitat selection in wapiti Cervus elaphus. Ecography . 32: 401– 410. Google Scholar CrossRef Search ADS   Zengeya FM, Murwira A. 2016. Intraspecific variations in home range overlaps of a semi-free range herbivore are explained by remotely sensed productivity. Int J Geogr Inf Sci . 30: 5– 19. Google Scholar CrossRef Search ADS   © The Author(s) 2017. Published by Oxford University Press on behalf of the International Society for Behavioral Ecology. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Behavioral Ecology Oxford University Press

Cape buffalo (Syncerus caffer caffer) social dynamics in a flood-pulsed environment

Loading next page...
 
/lp/ou_press/cape-buffalo-syncerus-caffer-caffer-social-dynamics-in-a-flood-pulsed-WEjkarF40A
Publisher
Oxford University Press
Copyright
© The Author(s) 2017. Published by Oxford University Press on behalf of the International Society for Behavioral Ecology. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
1045-2249
eISSN
1465-7279
D.O.I.
10.1093/beheco/arx138
Publisher site
See Article on Publisher Site

Abstract

Abstract Fission–fusion social dynamics allow animals to respond to short-term environmental changes by temporarily adjusting group size. The drivers of such complex social dynamics are thought to relate to resource availability, density effects, and social interactions. During 2008–2009, we collared 15 Cape buffalo (Syncerus caffer caffer) cows in different groups in the Okavango Delta, Botswana, to study social dynamics in a flood-pulsed ecosystem. Hierarchical clustering identified 2 subpopulations, 1 migratory and 1 resident. We calculated Utilization Distribution Overlap Index (UDOI) and number and duration of fusion events between buffalo dyads and related them to environmental variables. Number of fusion events and duration of fusion periods did not vary seasonally, total fusion time varied seasonally and annually, and UDOI varied with year and subpopulation. Fission events were more likely in cluttered habitats, but only in the late flood season. There was more open habitat in home range overlap (HRO) areas than home ranges (HR) in most seasons. Pan density in rainy season HROs was lower and higher than in HRs in 2008 and 2009, respectively, and in all flood seasons HRO areas were closer to permanent water than HRs, suggesting that fusion occurred when buffalo congregated on resources. Whereas previous studies described large herds that sometimes split, we identified numerous smaller groups that occasionally fused, indicating a very fluid social system. Our results highlight the need to understand social system flexibility to ensure appropriate management and understand the varying impacts of environmental and anthropogenic effects on subpopulations within the same geographic area. INTRODUCTION Gregarious, group-dwelling herbivores benefit from collective information sharing about resources (Isvaran 2007; Aureli et al. 2008), high levels of vigilance (Kie 1999), and effective antipredator defense strategies (Kelley et al. 2011). However, associating with other individuals also incurs costs such as competition for mates and resources, which are often dependent on density and resource availability (Alexander 1974). Optimal group size is therefore a trade-off between these benefits and costs (Markham et al. 2015), which can vary temporally through seasonal changes in environmental conditions, spatially through structural differences between habitat types, and overall through social interactions (Couzin and Laidre 2009). Herbivores usually occur in larger groups in open habitats with high visibility (Jarman 1974; Isvaran 2007), where they can detect potential predators most readily (Fortin et al. 2009), but also where intraspecific interactions can take place over larger distances. Short-term responses to changing environmental conditions can lead to fission–fusion social dynamics (Aureli et al. 2008), whereby groups split and join periodically, creating a relatively fluid and adaptable society (Sueur et al. 2011), although the individual composition of each core group usually remains the same (Wittemyer et al. 2005). The drivers of fusion events are not fully understood, but they include predation pressure (Kelley et al. 2011), reproductive opportunities (Mourier et al. 2012), congregation on abundant (Isvaran 2007) or limited resources (Holmes et al. 2016), and social attraction (Couzin 2006; Bercovitch and Berry 2010). Fission is thought to occur when the costs of competition for limited resources or mates become too high (Couzin and Laidre 2009), the costs of activity synchrony rise for dimorphic species (Conradt and Roper 2000), knowledge of profitable patches differs between experienced individuals (Ramos et al. 2015) that move apart and are followed by other group members (Merkle et al. 2015), or detection probability by predators increases with herd size, such as in cluttered habitats (Fortin et al. 2009). The most productive season is usually when the costs of social cohesion are at their lowest, and therefore should be associated with the largest group sizes in species exhibiting fission–fusion social dynamics (Wittemyer et al. 2005; Isvaran 2007; Bercovitch and Berry 2010). Fission–fusion dynamics can increase vulnerability of some groups to disturbance effects (Haulsee et al. 2016; Sigaud et al. 2017), but can also increase adaptability by allowing animals to alter their group size in response to environmental conditions (Holmes et al. 2016). Most herbivore species do not defend territories, but occupy home ranges (HRs) that can vary geographically according to seasonal changes in resource abundance and distribution (Owen-Smith et al. 2010). Utilizing HRs allows animals to access familiar locations and, through spatial memory, exploit productive areas and predict temporal and spatial shifts in resource availability (Wolf et al. 2009). Some areas can support groups with overlapping HRs, and the extent of spatial overlap relates to resource availability (Kolbe and Weckerly 2015). Animals may need to share limiting resources, such as water, or be able to coexist in areas with high resource abundance (Chaverri et al. 2007), leading to social interactions between individuals with overlapping HRs (VanderWaal et al. 2014). The extent of home range overlap (HRO) should follow a U-shaped curve in relation to resources: HRO should be high during periods of high and low resource abundance and lowest during periods of intermediate abundance (Zengeya and Murwira 2016). Variation in resource abundance and extent of social interactions can therefore lead to seasonal changes in fission–fusion dynamics within HRO areas. Fission–fusion social dynamics have been identified in a range of species, including numerous primates (Amici et al. 2008), giraffe (Giraffa camelopardalis; Bercovitch and Berry 2010), African elephant (Loxodonta africana; Couzin 2006), and African buffalo (Syncerus caffer; Cross et al. 2005), though the extent of fission–fusion dynamics may vary with subspecies for the last. Forest buffalo (S. c. nanus) occur in stable herds rarely larger than 20 (Melletti et al. 2007; Korte 2008). West African savanna buffalo (S. c. brachyceros) occur in herds of approximately 50 individuals with very little interaction between herds (Cornélis et al. 2011). Cape buffalo (S. c. caffer) has no characteristic herd size, although they can occur in groups of several thousand individuals (Couzin and Laidre 2009). Most studies of Cape buffalo in southern and eastern Africa have identified large herds with permanent members that periodically subdivide (e.g., Prins 1996; Cross et al. 2005) but consistently occupied identifiable HRs (Ryan et al. 2006). The cohesion of such large herds can vary temporally according to geographical region: during the rainy season, groups in Chobe National Park, Botswana were more fragmented (Taolo 2003), whereas groups in the Serengeti National Park, Tanzania, were larger (Sinclair 1977). Cape buffalo social systems therefore appear to vary with environmental conditions, and the definition of a herd can be problematic. Herds are hereafter defined as assemblages of animals with fixed membership and size that spend most of their time together, whereas groups can vary in these characteristics. In 2012, the Okavango Delta, Botswana, had an estimated 50000 Cape buffalo (unpublished aerial survey, Department of Wildlife and National Parks 2012) in a complex mosaic of different habitat types. The Delta has 2 annual influxes of water: rains falling concurrently in the Okavango Delta and further upstream in Angola produce a delayed flood pulse, which takes several months to flow through rivers in the catchment area before reaching the Okavango Delta (McCarthy and Ellery 1998). This creates high levels of spatial and temporal heterogeneity in resource availability (McCarthy and Ellery 1998). During the rainy season, most herbivore species in the Delta exploit nutrient-rich annual grasses growing in well-drained soils (Fynn et al. 2015), which can support large numbers of herbivores. Outside the rainy season, buffalo rely on seasonal floodplains (Bennitt et al. 2014), which provide water and fresh forage growth when the availability of these resources in other habitats declines progressively after the last rainfall. Migratory and resident buffalo subpopulations with distinct home ranges have been identified in the Okavango Delta (Bennitt et al. 2016), but their social structure has not yet been studied. The subpopulations showed divergent migratory behavior (Bennitt et al. 2016), 1 moving approximately 50 km east during the rainy season, and the other remaining in the same HR year round (Figure 1), although the latter displayed seasonal variation in habitat selection (Bennitt et al. 2014). Much of the previous research on fission–fusion in Cape buffalo has relied on observational data and focused on individual associations (Cross et al. 2005). Technological developments allow continuous remote data collection from numerous individuals and can therefore provide new insights into social behavior (Couzin 2006; Farine and Whitehead 2015). We use data from GPS-enabled collars fitted to 15 buffalo cows in different groups in the Okavango Delta in 2008–2009 to test the hypotheses that: 1) seasonal Utilization Distribution Overlap Index (UDOI) can be used to assign groups containing collared individuals to resident or migratory subpopulations, 2) groups with overlapping seasonal HRs displayed fusion, but only within subpopulations, and 3) sociality, represented by HRO and fusion, would be highest during the rainy season, in open habitat types and closer to water. Figure 1 View largeDownload slide Location of subpopulation ranges within the study area in Botswana. Permanently flooded parts of the study area are shaded. Figure 1 View largeDownload slide Location of subpopulation ranges within the study area in Botswana. Permanently flooded parts of the study area are shaded. METHODS Study site The Okavango Delta covers 15000 km2 in northern Botswana, between 22.0°–24.0°E and 18.5°–20.5°S (Heinl et al. 2006). Average rainfall and flooding patterns were used to define 3 seasons according to fixed dates: the rainy season (December to March); the early flood season, when flood waters were rising (April to July); and the late flood season, when flood waters were receding (August to November). The study was conducted in the south-eastern area of the Okavango Delta, and included permanent and temporary floodplains, as well as permanently dry areas (Figure 1). Habitat types included low shrubland and acacia, mopane and riparian woodlands, which were considered cluttered habitats, and floodplains and grassland, which were considered open habitats. A habitat map of the area was developed by digitizing satellite images at a 1:10000 scale using ArcGIS 10.0 (ESRI, Redlands, CA), and ground-truthing showed an accuracy of 88.1% (Bennitt et al. 2014). The vector habitat map was converted to a raster layer with a 50 × 50 m pixel size, which was defined as the grid to which all other layers snapped. Secondary floodplain was the only habitat type that was flooded all year round, and was therefore identified as a permanent water source and used to generate a raster layer representing distance to permanent water for each pixel in the study area. The location of every ephemeral pan in the study area was recorded from Google Earth (Google Inc., Mountain View, CA), and used to generate a pan density layer. Ephemeral pans can be identified throughout the year by their dark soil and their associated gaps in the tree canopy, so photographic capture date did not influence their detection using Google Earth. CAPTURE AND COLLARING We fitted 15 buffalo cows in different groups with Tellus Simplex 4D GPS-enabled satellite collars (Followit, Lindenberg, Sweden), programmed to record 1 GPS fix per hour. We located buffalo groups from the air and darted opportunistically, so we did not know whether groups were in a state of fusion or fission, but they were spatially distinct at the time of darting. The collars emitted a unique VHF signal, allowing them to be tracked using conventional telemetry equipment, and weighed 1.8 kg, 0.4% of the total mass of the smallest collared buffalo (Bennitt et al. 2014). Cows were selected because they represented breeding herds, and male buffalo tend to lose their collars within a few months of deployment (Taolo 2003). Buffalo were collared for periods of 3–16 months between December 2007 and March 2010, although we only used data from December 2007 to November 2009 to allow annual comparisons of seasonal data (Table 1). Table 1 View largeDownload slide Seasons during which 15 Cape buffalo cows in the Okavango Delta, Botswana were collared Table 1 View largeDownload slide Seasons during which 15 Cape buffalo cows in the Okavango Delta, Botswana were collared There were 24 buffalo darting operations: 15 to fit collars, 2 to replace malfunctioning collars, and 7 to remove collars. A helicopter and a vehicle were used for 22 and 2 darting events, respectively. A vehicle was only used to remove collars with functioning VHF so that we could track the darted animal if we lost visual contact. Sudden malfunctions occurred in 2 collars and we were unable to locate the animals to remove them, but all other collars were recovered by the end of the study. Buffalo were immobilized using either 10 mg M99, 40 mg Azaperone and 5 000 i.u. Hyalaze, reversed with 42 mg M5050 (N = 11), or 8 mg A3080, reversed with Naltrexone (N = 13). We only darted adult females in good condition, not visibly pregnant or with a small calf. Every effort was made to minimize stress and immobilization time, and all animals were observed returning to their herds after darting. All darting operations were carried out by 1 of 3 experienced veterinarians registered in Botswana. Darting permits were obtained prior to collar deployment, based on research permits EWT 3/3/8 XXXVII 44 and EWT 8/36/4 IV 62 issued by the Department of Wildlife and National Parks of Botswana. All capture and handling protocols were approved by the University of Bristol Ethics Committee (UIB/08/034), and conformed to the American Society of Mammalogists’ guidelines for the use of mammals in wildlife research (Sikes 2016) and to the Association for the Study of Animal Behaviour’s guidelines for the use of animals (Association for the Study of Animal Behaviour 2012). Subpopulation identification GPS data from collared animals visually separated them into resident and migratory subpopulations (Bennitt et al. 2016), but we used empirical data from seasonal utilization distributions (UDs) to confirm subpopulation membership and to determine the extent of interaction between and within subpopulations. We calculated buffalo seasonal UDs according to the Movement-Based Kernel Density Estimation method (MKDE; Benhamou 2011), using the “adehabitatHR” package (Calenge 2006) in R v. 3.3.2 (The R Foundation for Statistical Computing, Vienna, Austria, http://www.r-project.org). Sequential GPS coordinates recorded by collars were used to incorporate habitat-specific movement rates into the UD calculations, a technique that was developed using African buffalo (Benhamou and Cornélis 2010), and has previously been used for our study population (Bennitt et al. 2014). To allow for annual and seasonal variation in resource availability, and hence in UD location, we separated the data into 6 time periods: the 2008 and 2009 rainy, early flood and late flood seasons. For each seasonal dataset, we used the “adehabitatHR” package in R v. 3.3.2 to calculate the UDOI for each dyad, the best index for quantifying overlap with regards to shared space use (Fieberg and Kochanny 2005). We used the data cloud geometry-tree (DCG-tree) method to identify inherent patterns of association between buffalo dyads, based on UDOI (Fushing et al. 2013). We constructed a similarity matrix based on UDOI and used the “DCG” package in R v 3.3.2 to build dendrograms depicting the social network of collared buffalo (Fushing et al. 2013), which we used to confirm individual subpopulation membership. We ran a generalized linear mixed model with a Gamma distribution and buffalo dyad as a random variable to determine whether UDOI was affected by season, year or subpopulation. Fusion events A combination of observational and GPS data indicated that collared individuals were more frequently in different groups than in the same group, and they were never observed alone. We have therefore assumed that collared individuals represented groups that occasionally displayed fusion, rather than fragments of a larger group in a state of fission. Hereafter we analyzed factors leading to fusion events, identified as the point at which groups containing collared animals joined, with each event leading to a fusion period that lasted until a fission event occurred, when groups containing collared animals split up. We calculated hourly Euclidean distance between dyads of concurrently collared buffalo. GPS fix success rate was 82.0 ± 9.9% (Bennitt 2012); data were excluded when 1 or both collars in a dyad failed to obtain a fix. Cross et al. (2005) identified fusion events when collared buffalo were within 1 km of each other. We refined this definition by calculating the frequency of occurrence of different Euclidean distances < 1 km across all dyads and generating a histogram, which showed a drop in frequency of observations > 300 m (Figure 2). We therefore defined fusion as periods when dyads were separated by < 300 m, which represented 87.4% of all distances between dyads < 1000 m apart. Since this level of proximity could occur by chance in cluttered habitats, we only considered herds to be in a fusion state when there were at least 3 consecutive fixes for which separation was < 300 m. Some fluidity was allowed, so some fusion events included times when dyads were > 300 apart for ≤ 3 h, after which distance between dyads decreased again. Figure 2 View largeDownload slide Histogram showing frequency of occurrence of Euclidean distance < 1000 m between Cape buffalo dyads in the Okavango Delta, Botswana. The arrow marks the threshold distance for identifying fusion between dyads. Figure 2 View largeDownload slide Histogram showing frequency of occurrence of Euclidean distance < 1000 m between Cape buffalo dyads in the Okavango Delta, Botswana. The arrow marks the threshold distance for identifying fusion between dyads. For each season, we recorded the total number of fusion events per dyad, the duration of each fusion event and total fusion time as a percentage of the entire season. We used the “lme4” package (Bates 2010) in R v. 3.3.2 to run a generalized linear mixed model with a Poisson distribution and a log-link function to determine whether season, year and subpopulation affected the number of fusion events. After testing for normality, we ran a general linear mixed model on the normal log of the duration of each fusion event to determine whether season, year, and subpopulation affected fusion duration. We ran a generalized linear mixed model with a binomial distribution to determine whether total fusion time was affected by season, year, or subpopulation. For all models, we included buffalo dyad as the random effect. We also ran a general linear model with a Poisson distribution to determine whether UDOI was related to number of fusion events. Habitat structure We plotted the start and end GPS coordinates of each fusion period onto the habitat map, corresponding to group fusion and fission events, respectively, to determine whether habitat type influenced group cohesion. We extracted UD outlines from the 95% isopleths of the UDs as HR polygons and identified areas of HRO for each dyad. We compared the proportion of fusion and fission events taking place in open and closed habitats to the proportion of open and closed habitats in the seasonal HRO for each dyad as an analogue to a habitat selection analysis. We also compared the proportions of open and closed habitats in seasonal HROs to those in seasonal HRs to determine whether the location of HROs was linked to habitat type. We calculated Manly selection indices (Manly et al. 2002) in the adehabitatHS package in R v. 3.2.3 and used them to identify habitats that were selected and avoided. Selection ratios with both 95% confidence intervals above and below 1 indicated selection and avoidance, respectively (Manly et al. 2002). Water availability We extracted water availability data from the seasonal HRs and HROs of each buffalo dyad. For rainy season data, we used the pan density layer as a measure of water availability because ephemeral pans were filled with water during that season. We calculated the number of pans within 300 m of each pixel, the distance used to define herd membership, such that each pixel represented the number of pans within 0.28 km2. For the early and late flood seasons, we used the distance to permanent water, since most ephemeral pans were dry during those seasons. The relatively small pixel size meant that a high volume of data was generated for each HR and HRO, so we subsampled the datasets, randomly extracting 1000 values from each. For each seasonal dataset, we used generalized linear mixed models with a binomial distribution and buffalo dyad as a random effect to determine whether water availability affected fusion and fission location, and generalized linear mixed models with a binomial distribution and buffalo dyad as a random effect to test whether water availability differed between seasonal HRs and HROs. All models were run with all combinations of variables, including interaction effects. The most parsimonious models were identified from Akaike’s Information Criterion (Akaike 1974). All statistical analyses were run in R v. 3.3.2. RESULTS There were 6, 4, and 6 buffalo collared during the 2008 rainy, early flood and late flood seasons, respectively, and 6, 7, and 8 buffalo collared during the 2009 rainy, early flood and late flood seasons, respectively (Table 1). Subpopulation identification The dendrograms produced by the DCG-tree method identified 2 main clusters in each season, which for the most part corresponded to the observed resident and migratory subpopulations (Figure 3). During the 2008 early flood season, only 4 individuals were collared and the clustering technique requires 2 individuals per cluster, so B5 was grouped with B6, despite their membership of different clusters in all other seasons. UDOI was therefore greater within subpopulations than between them, and these results were consistent across seasons and years. The resident subpopulation included B1, B2, B5, B8, B9, B10, and B11, whereas the migratory subpopulation included B3, B4, B6, B7, B12, B13, B14, and B15. Figure 3 View largeDownload slide Dendrograms showing hierarchical clustering of 15 Cape buffalo cows in the Okavango Delta, Botswana during the rainy season in (a) 2008 and (b) 2009, the early flood season in (c) 2008 and (d) 2009, and the late flood season in (e) 2008 and (f) 2009. Figure 3 View largeDownload slide Dendrograms showing hierarchical clustering of 15 Cape buffalo cows in the Okavango Delta, Botswana during the rainy season in (a) 2008 and (b) 2009, the early flood season in (c) 2008 and (d) 2009, and the late flood season in (e) 2008 and (f) 2009. Seasonal HRO Within the confines of the subpopulation structure, the maximum number of dyads that could have had overlapping HRs was 6, 3, and 7 in the 2008 rainy, early flood and late flood seasons, respectively, and 7, 9, and 12 in the 2009 rainy, early flood and late flood seasons, respectively. In all 2008 seasons, all of the dyads displayed HRO within subpopulations, but in the 2009 rainy, early flood and late flood seasons, only 6, 5, and 9 dyads, respectively, displayed HRO within subpopulations (Figure 4). There were also 2 and 3 dyads in the 2008 early and late flood seasons, respectively, which displayed HRO between subpopulations, but UDOI for most of these was <0.001. Only 1 dyad, B5B7, had a UDOI value comparable to dyads within subpopulations (0.09). We excluded all of these dyads from subsequent analyses. Figure 4 View largeDownload slide Seasonal home ranges of 15 Cape buffalo cows in the Okavango Delta, Botswana during the rainy season in (a) 2008 and (b) 2009, the early flood season in (c) 2008 and (d) 2009, and the late flood season in (e) 2008 and (f) 2009. Cluttered and open habitats are indicated by darker and lighter gray, respectively. Figure 4 View largeDownload slide Seasonal home ranges of 15 Cape buffalo cows in the Okavango Delta, Botswana during the rainy season in (a) 2008 and (b) 2009, the early flood season in (c) 2008 and (d) 2009, and the late flood season in (e) 2008 and (f) 2009. Cluttered and open habitats are indicated by darker and lighter gray, respectively. Mean ± SD UDOI was 0.23 ± 0.19, 0.09 ± 0.07, and 0.09 ± 0.10 during the 2008 rainy, early flood and late flood seasons, respectively, and 0.03 ± 0.02, 0.41 ± 0.74, and 0.32 ± 0.34 during the 2009 rainy, early flood and late flood seasons, respectively (Table 2). The most parsimonious model included the effect of subpopulation only (AICc = −55.26, AICω = 0.55), although the model with the effects of subpopulation and year was also competitive (AICc = −53.90, AICω = 0.28; Figure 5). Table 2 Seasonal UDOI, number of fusion events, and total fusion time as percentage of season duration for Cape buffalo dyads in the Okavango Delta, Botswana Dyad  Subpopulation  Year  UDOI  N fusion events  Total fusion duration (%)  Rainy  Early flood  Late flood  Rainy  Early flood  Late flood  Rainy  Early flood  Late flood  B1B2  Resident  2008  0.02  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B1B5  Resident  2008  0.07  N/A  N/A  2  N/A  N/A  4.58  N/A  N/A  B2B5  Resident  2008  0.55  N/A  N/A  11  N/A  N/A  26.37  N/A  N/A  B5B8  Resident  2008  N/A  N/A  <0.01  N/A  N/A  0  N/A  N/A  0      2009  0.0  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B8B11  Resident  2009  N/A  0.16  0.38  N/A  7  8  N/A  13.28  19.71  B9B10  Resident  2009  N/A  1.73  1.09  N/A  3  12  N/A  17.59  19.83  B3B4  Migratory  2008  0.24  0.16  0.30  5  3  3  10.79  22.44  18.75      2009  0.04  N/A  N/A  1  N/A  N/A  0.03  N/A  N/A  B3B6  Migratory  2008  0.22  0.10  0.13  5  0  0  18.51  0  0      2009  0.05  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B3B7  Migratory  2009  0.02  N/A  N/A  1  N/A  N/A  0.28  N/A  N/A  B4B6  Migratory  2008  0.28  0.02  0.13  2  3  2  0.65  1.37  0.85      2009  <0.01  N/A  N/A  0  N/A  N/A  0.0  N/A  N/A  B4B7  Migratory  2008  N/A  N/A  0.02  N/A  N/A  0  N/A  N/A  0      2009  0.04  N/A  N/A  1  N/A  N/A  0.24  N/A  N/A  B6B7  Migratory  2008  N/A  N/A  0.09  N/A  N/A  3  0  N/A  0.99      2009  <0.01  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B7B12  Migratory  2009  N/A  <0.01  N/A  N/A  0  N/A  N/A  0  0  B7B13  Migratory  2009  N/A  0.14  N/A  N/A  5  N/A  N/A  7.34  0  B12B13  Migratory  2009  N/A  <0.01  0.14  N/A  0  1  N/A  0  1.06  B12B14  Migratory  2009  N/A  N/A  0.15  N/A  N/A  0  N/A  N/A  0  B12B15  Migratory  2009  N/A  N/A  0.25  N/A  N/A  0  N/A  N/A  0  B13B14  Migratory  2009  N/A  N/A  0.04  N/A  N/A  2  0  N/A  1.40  B13B15  Migratory  2009  N/A  N/A  0.11  N/A  N/A  4  0  0  9.66  B14B15  Migratory  2009  N/A  N/A  0.48  N/A  N/A  5  0  N/A  4.77  Dyad  Subpopulation  Year  UDOI  N fusion events  Total fusion duration (%)  Rainy  Early flood  Late flood  Rainy  Early flood  Late flood  Rainy  Early flood  Late flood  B1B2  Resident  2008  0.02  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B1B5  Resident  2008  0.07  N/A  N/A  2  N/A  N/A  4.58  N/A  N/A  B2B5  Resident  2008  0.55  N/A  N/A  11  N/A  N/A  26.37  N/A  N/A  B5B8  Resident  2008  N/A  N/A  <0.01  N/A  N/A  0  N/A  N/A  0      2009  0.0  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B8B11  Resident  2009  N/A  0.16  0.38  N/A  7  8  N/A  13.28  19.71  B9B10  Resident  2009  N/A  1.73  1.09  N/A  3  12  N/A  17.59  19.83  B3B4  Migratory  2008  0.24  0.16  0.30  5  3  3  10.79  22.44  18.75      2009  0.04  N/A  N/A  1  N/A  N/A  0.03  N/A  N/A  B3B6  Migratory  2008  0.22  0.10  0.13  5  0  0  18.51  0  0      2009  0.05  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B3B7  Migratory  2009  0.02  N/A  N/A  1  N/A  N/A  0.28  N/A  N/A  B4B6  Migratory  2008  0.28  0.02  0.13  2  3  2  0.65  1.37  0.85      2009  <0.01  N/A  N/A  0  N/A  N/A  0.0  N/A  N/A  B4B7  Migratory  2008  N/A  N/A  0.02  N/A  N/A  0  N/A  N/A  0      2009  0.04  N/A  N/A  1  N/A  N/A  0.24  N/A  N/A  B6B7  Migratory  2008  N/A  N/A  0.09  N/A  N/A  3  0  N/A  0.99      2009  <0.01  N/A  N/A  0  N/A  N/A  0  N/A  N/A  B7B12  Migratory  2009  N/A  <0.01  N/A  N/A  0  N/A  N/A  0  0  B7B13  Migratory  2009  N/A  0.14  N/A  N/A  5  N/A  N/A  7.34  0  B12B13  Migratory  2009  N/A  <0.01  0.14  N/A  0  1  N/A  0  1.06  B12B14  Migratory  2009  N/A  N/A  0.15  N/A  N/A  0  N/A  N/A  0  B12B15  Migratory  2009  N/A  N/A  0.25  N/A  N/A  0  N/A  N/A  0  B13B14  Migratory  2009  N/A  N/A  0.04  N/A  N/A  2  0  N/A  1.40  B13B15  Migratory  2009  N/A  N/A  0.11  N/A  N/A  4  0  0  9.66  B14B15  Migratory  2009  N/A  N/A  0.48  N/A  N/A  5  0  N/A  4.77  Dyads were not all collared concurrently throughout seasons. N/A = not applicable. View Large Figure 5 View largeDownload slide Effects of year and subpopulation on Utilization Distribution Overlap Index (UDOI) of Cape buffalo dyads in the Okavango Delta, Botswana. The circles denote means and the whiskers denote SEs. Figure 5 View largeDownload slide Effects of year and subpopulation on Utilization Distribution Overlap Index (UDOI) of Cape buffalo dyads in the Okavango Delta, Botswana. The circles denote means and the whiskers denote SEs. Fusion events Despite overlapping HRs, not all dyads displayed fusion: 83.3%, 66.7%, and 42.9% of dyads fused during the 2008 rainy, early flood and late flood seasons, respectively, and 28.6%, 33.3%, and 55.6% of dyads fused during the 2009 rainy, early flood and late flood seasons, respectively. Mean ± SD number of fusion events per dyad was 5.4 ± 3.8, 3.0 ± 0.0, and 2.7 ± 0.6 during the 2008 rainy, early flood, and late flood seasons, respectively, and 1.0 ± 0.0, 5.0 ± 2.0, and 5.5 ± 4.4 during the 2009 rainy, early flood, and late flood seasons, respectively (Table 2). The most parsimonious model included the effect of subpopulation only (AICc = 161.39), but the null model was also competitive (AICc = 162.65) which indicated that none of season, year or subpopulation had any effect on the number of fusion events per dyad. Only 1 dyad containing individuals from different subpopulations, B5B7, displayed fusion, for 155 h during the 2008 late flood season. Mean ± SD duration of fusion events was 63.7 ± 122.0, 74.1 ± 119.6, and 75.4 ± 118.6 h during the 2008 rainy, early flood, and late flood seasons, respectively, and 7.5 ± 0.7, 74.6 ± 126.3, and 75.3 ± 69.8 h during the 2009 rainy, early flood, and late flood seasons, respectively. The most parsimonious model was the null model (AICc = 193.82), indicating that none of season, year or subpopulation had any effect on fusion duration. Mean ± SD total fusion duration as a percentage of each season was 12.2 ± 10.4%, 11.9 ± 14.9%, and 6.8 ± 10.3% during the 2008 rainy, early flood, and late flood seasons, respectively, and 0.3 ± 0.0%, 12.7 ± 5.1%, and 14.6 ± 18.6% during the 2009 rainy, early flood, and late flood seasons, respectively. 1 dyad, B9B10, spent approximately 61 days or 49.8% of the 2009 late flood season in fusion, and the 2nd longest cumulative fusion duration was approximately 32 days or 26.4% of the 2008 rainy season, for B2B5 (Table 2). The most parsimonious model included an interaction effect between season and year, and the fixed effect of subpopulation (AICc = 389.78, AICω = 0.54), but the model with only the interaction effect between season and year was also competitive (AICc = 390.13, AICω = 0.46; Figure 6). Number of fusion events was not affected by UDOI: the null model was the most parsimonious (AICc = 162.65). Figure 6 View largeDownload slide Effects of season and year on total fusion time as a percentage of season duration for Cape buffalo dyads in the Okavango Delta, Botswana. The circles denote means and the whiskers denote SEs. Figure 6 View largeDownload slide Effects of season and year on total fusion time as a percentage of season duration for Cape buffalo dyads in the Okavango Delta, Botswana. The circles denote means and the whiskers denote SEs. There were 2 time periods when 3 collared animals were in the same herd, both during the 2008 rainy season. Fusion was recorded for B3B4 and B3B6, but not for B4B6, over a 35-h period of concurrent fusion for both dyads. During a 405-h fusion event for B3B6, B3B4 showed fusion for 40 h, but again no fusion was recorded for B4B6. It is therefore possible that some fusion events were undetected by the 300 m threshold, but increasing the threshold to an arbitrary distance may have captured false fusion events when dyads were close to each other without being in the same group. Habitat structure The relatively small sample size for dyads displaying fusion periods did not allow the data to be split between years, so we grouped seasonal data to compare habitat types at fission and fusion events with available habitat in HROs. We did not identify any seasons when fusion events were more likely to take place in open or closed habitats (Table 3). Fission events in the rainy and early flood seasons took place in open and cluttered habitats in proportion to their availability, but during the late flood season, fission was more likely in cluttered habitats and less likely in open habitats. Table 3 Habitat selection ratios for locations of fission and fusion events and habitat composition of seasonal home range overlap for Cape buffalo in the Okavango Delta, Botswana   Rainy  Early flood  Late flood  Location  Number  Open  Cluttered  Number  Open  Cluttered  Number  Open  Cluttered  Fission  7  1.12 (−2.29–2.53)  0.96 (0.49–1.43)  5  0.67 (0.34–1.01)  2.17 (−0.30–4.64)  8  0.63 (0.35–0.91)  2.26 (1.13–3.39)  Fusion  7  1.69 (−0.25–3.63)  0.76 (0.30–1.21)  5  0.85 (0.55–1.14)  1.56 (0.15–2.96)  8  0.87 (0.62–1.11)  1.44 (0.47–2.41)  Home range overlap 2008  12  0.91 (0.74–1.08)  1.06 (0.95–1.17)  6  1.15 (1.02–1.28)  0.17 (−0.00–0.35)  12  1.08 (1.01–1.15)  0.39 (0.08–0.71)  Home range overlap 2009  12  1.63 (1.26–2.01)  0.95 (0.93–0.98)  10  1.06 (0.98–1.14)  0.82 (0.54–1.11)  18  1.16 (1.05–1.26)  0.69 (0.50–0.88)    Rainy  Early flood  Late flood  Location  Number  Open  Cluttered  Number  Open  Cluttered  Number  Open  Cluttered  Fission  7  1.12 (−2.29–2.53)  0.96 (0.49–1.43)  5  0.67 (0.34–1.01)  2.17 (−0.30–4.64)  8  0.63 (0.35–0.91)  2.26 (1.13–3.39)  Fusion  7  1.69 (−0.25–3.63)  0.76 (0.30–1.21)  5  0.85 (0.55–1.14)  1.56 (0.15–2.96)  8  0.87 (0.62–1.11)  1.44 (0.47–2.41)  Home range overlap 2008  12  0.91 (0.74–1.08)  1.06 (0.95–1.17)  6  1.15 (1.02–1.28)  0.17 (−0.00–0.35)  12  1.08 (1.01–1.15)  0.39 (0.08–0.71)  Home range overlap 2009  12  1.63 (1.26–2.01)  0.95 (0.93–0.98)  10  1.06 (0.98–1.14)  0.82 (0.54–1.11)  18  1.16 (1.05–1.26)  0.69 (0.50–0.88)  Significant results are in bold. View Large The larger sample sizes for dyads with overlapping HRs allowed us to split the data by year as well as season. There was no difference between habitat composition in the HRs and HROs during the 2008 rainy season and the 2009 early flood season, but there was more open and less cluttered habitat in HROs than HRs during the other seasons (Table 3). Water availability The relatively small sample size for dyads displaying fusion events did not allow the data to be split between years, so we grouped seasonal data to compare water availability at fission and fusion locations. During the rainy season, mean ± SD pan density was 6.8 ± 6.5 and 3.6 ± 4.1 pans/0.28 km2 at fission and fusion locations, respectively. The most parsimonious model included the effect of pan density (AICc = 99.8, AICω = 0.89), indicating that pan density was higher at fission than fusion event locations during the rainy season. During the early flood season, mean ± SD distance to permanent water was 302.7 ± 348.5 m and 223.7 ± 335.2 m at fusion and fission event locations, respectively. The most parsimonious model was the null model (AICc = 62.5), indicating that there was no effect of water availability on fission/fusion during the early flood season. During the late flood season, mean ± SD distance to permanent water was 1275.4 ± 4177.3 m and 1639.2 ± 5264.1 m at fusion and fission locations, respectively. The most parsimonious model was the null model (AICc = 123.4), indicating that there was no effect of water availability on fission/fusion during the late flood season. The larger sample sizes for dyads with overlapping HRs allowed us to split the data by year as well as season. During the 2008 rainy season, mean ± SD pan density was 6.0 ± 8.5 and 6.6 ± 9.0 pans/0.28 km2 in HRs and HROs, respectively. The most parsimonious model included the effect of pan density (AICc = 2783.36, AICω = 1.00), indicating that pan density was higher in HROs than in HRs during the 2008 rainy season. During the 2009 rainy season, mean ± SD pan density was 6.4 ± 9.7 and 5.0 ± 6.8 pans/0.28 km2 in HRs and HROs, respectively. The most parsimonious model included the effect of pan density (AICc = 32942.47, AICω = 1.00), indicating that pan density was higher in HRs than in HROs during the 2009 rainy season. During the 2008 early flood season, mean ± SD distance to permanent water was 1526.5 ± 5438.9 m and 94.0 ± 151.8 m in the HRs and HROs, respectively. The most parsimonious model included the effect of distance to permanent water (AIC = 15480.17, AICω = 1.00), indicating that water availability was lower in HRs than in HROs during the 2008 early flood season. During the 2009 early flood season, mean ± SD distance to permanent water was 1526.6 ± 4364.9 m and 394.6 ± 577.5 m in the HRs and HROs, respectively. The most parsimonious model included the effect of distance to permanent water (AICc = 26834.63, AICω = 1.00), indicating that water availability was lower in HRs than in HROs during the 2009 early flood season. During the 2008 late flood season, mean ± SD distance to permanent water was 778.6 ± 3792.8 m and 77.8 ± 186.6 m in the HRs and HROs, respectively. The most parsimonious model included the effect of distance to permanent water (AICc = 37576.9, AICω = 1.00), indicating that water availability was lower in HRs than in HROs during the 2008 late flood season. During the 2009 late flood season, mean ± SD distance to permanent water was 1585.9 ± 3400.1 m and 1046.7 ± 1447.1 m in the HRs and HROs, respectively. The most parsimonious model included the effect of distance to permanent water (AICc = 62891.4, AICω = 1.00), indicating that water availability was lower in HRs than in HROs during the 2009 late flood season. DISCUSSION Understanding animal social systems provides in-depth knowledge of possible routes for disease transmission (Body et al. 2015), information sharing, and gene flow (Biosa et al. 2015), and allows the identification of key processes and pathways for population connectivity and health (Morales et al. 2010). Previous studies of African buffalo in different parts of their range have recorded a variety of social systems, from steady herds with permanent membership (Melletti et al. 2007) to more fluid groups occupying identifiable home ranges (Ryan et al. 2006). Our results show that, in the Okavango Delta, UDOI could be used to identify subpopulation membership for individual collared animals and fusion between individuals in different subpopulations was rare. Number of fusion events and duration of fusion periods did not vary seasonally, total fusion time varied seasonally and annually, and UDOI varied with year and subpopulation. Fission events were more likely in cluttered habitats, but only in the late flood season, and HROs contained more open than cluttered habitats compared to HRs in most seasons. Pan density was higher at fission than fusion events sites during the rainy season, but water availability had no effect on fission/fusion in other seasons. Pan density in rainy season HROs was lower and higher than in HRs in 2008 and 2009, respectively, and in all flood seasons, HRO areas were closer to permanent water than HRs. Buffalo were not collared concurrently, so we could have deployed a collar in a group that had previously contained a collared animal. However, the high levels of social fluidity and environmental variation suggest that HR locations may have varied between years, which would also have led to variation in HRO locations and levels of interaction with neighbors, as indicated by the different rainy season fusion duration between 2008 and 2009, when data from 4 of the same individuals were used. Hourly GPS fixes provide large datasets, but they are still relatively coarse and cannot be used to identify precise moments when buffalo groups became aware of each other and decided whether to approach and fuse. Equally, the moment of fission is very difficult to identify because consecutive Euclidean distances between collared buffalo could shift from < 300 m to ≥ 1000 m over a 1 h period, and it was not possible to identify exactly when or where fission occurred. Indeed, when 3 collared animals were in the same group, our methods detected concurrent fusion between 2 dyads, but did not identify a fusion pattern for the 3rd. Some fusion events may therefore have gone undetected based on our criteria, perhaps for very large groups in which collared animals could have been >300 m apart, but raising the threshold distance for defining fusion would have been arbitrary and could have led to the inclusion of false fusion events when collared individuals were close to each other without being in the same group. Also, some fission–fusion events could have occurred without being recorded because they involved noncollared animals, but it is unlikely that the behavior of collared animals was substantially different from that of other buffalo within the study area. There were no impermeable barriers between the subpopulations: buffalo could cross channels and resident herds could follow the migration route taken by the migratory subpopulation (Bennitt et al. 2016). The resident subpopulation may have migrated in a different direction prior to the erection of a veterinary fence in 1982 (Mbaiwa and Mbaiwa 2006), and they may not have the spatial memory or cognitive capacity to adjust to changing conditions caused by the erection of a hard boundary. Resident buffalo frequently moved along the fence during the rainy season, suggesting that they would have pushed further if the fence had not been there. Habitats on either side of the fence were similarly dominated by mopane woodlands and grasslands, possibly forming a corridor that buffalo could have followed ancestrally. Competition between the subpopulations could have led to the observed spatial distribution, but the migratory subpopulation left their flood season HR during the rainy season, leaving almost entirely unoccupied habitat that the resident subpopulation could have exploited during the rainy season but did not. The very low levels of HRO between subpopulations indicate that their encounter levels were too low for competition because they rarely converged on shared resources. Competition within subpopulations would have been possible in HRO areas, but our results suggested that groups congregated and fused in areas with key resources, rather than competitively excluding other groups. The subpopulations displayed similar reproductive productivity and body condition, despite smaller resident HRs and group sizes (Bennitt et al. 2016), suggesting that resource availability in the subpopulation HRs was comparable and that there was no strong driver for interaction between subpopulations. This suggests that there may have been social reasons for the disparate migratory strategies, possibly related to spatial information passed down through interaction with other subpopulation members. We hypothesized that the abundance and wide distribution of water and productive forage across the landscape during the rainy season would lead to higher levels of fusion, but the frequency of fusion events and duration of fusion periods did not appear to be affected by environmental factors such as seasonal changes in resources. Group sizes may have increased with males joining breeding groups to exploit mating opportunities during the rainy season, but fusion behavior between collared females was not affected. Buffalo occupied more cluttered habitats during the rainy season (Bennitt et al. 2014), so predation pressure could have reduced optimal group size, leading to fission events (Fortin et al. 2009). When the flood waters recede during the limiting late flood season, nutrient-poor floodplain grasses offer the only fresh forage growth (Murray-Hudson et al. 2006), and this could drive buffalo to congregate in areas of limited resources, thereby leading to group fusion (Bercovitch and Berry 2010). Total time spent in fusion varied with season and year, being lowest during the 2009 rainy season, when there appears to have been a drop in environmental pressures that would have caused fusion. Rainfall in the 2009 rainy season started early (Okavango Research Institute weather monitoring website, http://www.okavangodata.ub.bw/ori/monitoring/rainfall/), possibly leading to rapid growth of productive vegetation that would also have allowed buffalo to be in a state of fission. UDOI varied with subpopulation and, to some extent, with year: UDOI was higher in the resident subpopulation, especially in 2009. The resident subpopulation utilized a smaller area than the migratory subpopulation, which could have caused the higher levels of HRO, but there was also one dyad, B9B10, collared in the resident subpopulation in 2009, which had much higher UDOI than any other dyad, and this could have affected these results. UDOI did not have an effect on the number of fusion events between dyads, even though encounters between dyads with extensive HRO would probably have been more frequent because population density in HRO areas should have been higher (Kolbe and Weckerly 2015). Indeed, elk (Cervus elaphus) encounter rates increased with HRO, although overall population density also had a strong effect (Vander Wal et al. 2014). Some species with high levels of HRO partition the overlap temporally by utilizing shared resources at different times (Markham et al. 2015), but our results suggest that buffalo were utilizing shared resources at the same time, potentially leading to opportunistic fusion events (Bauder et al. 2016). In contrast to expectations, fusion was not more common in open habitats, and there was only 1 season when fission was more frequent in cluttered habitats. However, in 4 of the 6 seasons, the proportion of open habitat was higher in HROs than in HRs. Open habitats included grassland and floodplains, which are all favored grazing habitats (Bennitt et al. 2015), and the latter were identified as sources of permanent water. Therefore these habitats provide key resources and would form vital components of HRs, possibly leading to overlapping patterns of use by 2 or more individuals. The only time of year when there were differences in water availability between fusion and fission sites was during the rainy season, when pan density was higher in the latter. Pan density was also higher in HRs than in HROs during the 2009 rainy season, despite being slightly lower in HRs than in HROs in the 2008 rainy season. Pans are more common in poorly-drained soils where water can accumulate, such as in mopane woodland, but these are often not the most productive soil types in terms of herbaceous resources (Bennitt et al. 2016). Buffalo HRs may therefore have overlapped in areas with high quality or abundant forage, which may also have been areas with lower water availability, as seen in semi-free-ranging cattle (Bos taurus; Zengeya and Murwira 2016). The wide distribution of abundant pans across the landscape would have removed the limiting factor of water availability present during the flood seasons. In contrast, during the early and late flood seasons, HROs were closer to permanent water than the rest of the seasonal HRs. As obligate drinkers, buffalo need access to water every day, and their movements are therefore restricted by water availability outside the rainy season. During the late flood season, forage is senescent in dry habitats, but fresh grasses grow on the floodplains as the floodwaters recede (Murray-Hudson et al. 2006), providing an additional attractant for buffalo. Seasonal variation in forage and water availability in a flood-pulsed ecosystem such as the Okavango Delta may have exerted different pressures on buffalo that required similar levels of spatial and social interaction, such as the availability of abundant but spatially restricted resources. Cape buffalo in the Okavango Delta therefore behaved differently to other populations in eastern (Sinclair 1977) and southern Africa (Taolo 2003), maintaining similar levels of group cohesion throughout the year. We were unable to record group sizes accurately during the rainy season because of low visibility, so groups could have been larger at that time, but there were no differences in group size between the early and late flood seasons, when median group size was 100–200 individuals (Bennitt et al. 2016). CONCLUSIONS Our results show that Cape buffalo in the Okavango Delta form fluid groups within relatively distinct subpopulations that occupy separate HRs and show very little social interaction, and that fusion events within subpopulations occur in HROs located in areas with key environmental resources. Buffalo social systems vary in different parts of their global range, indicating a disparity in social complexity that may be linked to resource availability. Previous studies of buffalo social systems in eastern and southern Africa have described several large herds that subdivide (Prins 1996; Cross et al. 2005), typical of species displaying fission–fusion social dynamics (Sueur and Maire 2014). Our results suggest that buffalo in the Okavango Delta form numerous small groups that occasionally fuse into larger aggregations when they congregate on key resources, which appear to relate to forage during the rainy season and water during the early and late flood seasons. Our results therefore call into question the definition of a “herd”, particularly in reference to animals that are difficult to identify individually and occur in large aggregations, such as Cape buffalo. We suggest that the term “herd” should be used with caution, and that groups of animals with undetermined levels of cohesion and membership permanence should be called “groups”. Previous research has shown a higher level of sociality between dyads with high levels of spatial overlap (Chaverri et al. 2007), but it is difficult to distinguish cause and effect: animals may be attracted to each other, leading them to share space, or they may be attracted to resources, leading to higher encounter rates (Best et al. 2014). Our results suggest that buffalo HRs overlapped in areas with key seasonal resources, which led to fusion events followed by fusion periods when all animals in a herd made the same collective foraging decisions (Ramos-Fernández et al. 2006). Such spatial causes of higher levels of group cohesion and social interaction could be more important than kinship and relatedness (Best et al. 2014), and it is possible that, once fusion had occurred, herds moved together for several hours, allowing social interaction between herd members. Fission could ensue when key individuals differed in their decisions about the next foraging location, and were joined by regular followers, leading to a group split (Ramos-Fernández and Morales 2014). From aerial observations, the resident and migratory subpopulations in our study area were estimated at 5000 and 8000 buffalo, respectively (E.B., Personal observation, 2007–2010), group sizes that are too large to be sustained by clumped resources. Indeed, total seasonal fusion duration showed that collared buffalo were in a fission state for the majority of their time, with only 1 dyad spending almost 50% of 1 season in a fusion state. The lack of interfaces between subpopulations suggests that they are relatively distinct in terms of spatial patterns and social interactions, which could have consequences for gene flow and the transmission of adaptive behavior. These 2 subpopulations, despite occupying HRs with common boundaries, could therefore constitute a meta-population, with minimal contact between subpopulations experiencing different environmental pressures, and further research should be conducted to determine the level of gene flow between subpopulations. Populations displaying fission–fusion social dynamics can be more vulnerable to anthropogenic impacts, as well as being more difficult to count reliably (Haulsee et al. 2016). Dynamic levels of social cohesion could mean that some meta-populations are more at risk, whereas others are safer, decreasing their vulnerability to the effects of stochastic environmental change through increased variability (Holmes et al. 2016). Disruptions from anthropogenic influences or stochastic climatic variation could have large impacts on buffalo in the Okavango Delta through a reduction in access to seasonal ranges (Bennitt et al. 2014), particularly for buffalo on the periphery of the ecosystem. Peripheral subpopulations are also more vulnerable to human-wildlife conflict (Bennitt et al. 2014) and could encounter ecological traps, whereby incomplete knowledge of risks lead herds to utilize resources in areas with higher levels of human activity and hence increased mortality (Sigaud et al. 2017). Cape buffalo in the Okavango Delta should therefore be managed as meta-populations exposed to different ecological pressures and threats rather than 1 continuous population with homogeneous mixing of genetic material (Titcomb et al. 2015). FUNDING This work was supported by Jenny and Martin Bennitt; the Dulverton Trust; Harry Ferguson; Ian Fuhr; Rodney Fuhr; Dane Hawk; Idea Wild; the North of England Zoological Society; the Roberts Fund; and Wilderness Safaris Wildlife Trust. The authors thank the Botswana Ministry of Environment, Wildlife and Tourism for permission to conduct this study; Dane Hawk, Rob Jackson, and Larry Patterson for veterinary support; Laura Atkinson, Janette Baarman, Roz Balen, Shavaughn Davies, and Jennifer Gilbert for assistance in the field; and Guy Lobjoit and Duncan Rowles for logistical support. Data accessibility: Analyses reported in this article can be reproduced using the data provided by Bennitt et al. (2017). REFERENCES Akaike H. 1974. A new look at the statistical model identification. IEEE T Automat Cont . 19: 716– 723. Google Scholar CrossRef Search ADS   Alexander RD. 1974. The evolution of social behavior. Annu Rev Ecol Syst . 5: 325– 383. Google Scholar CrossRef Search ADS   Amici F, Aureli F, Call J. 2008. Fission-fusion dynamics, behavioral flexibility, and inhibitory control in primates. Curr Biol . 18: 1415– 1419. Google Scholar CrossRef Search ADS PubMed  Association for the Study of Animal Behaviour. 2012. Guidelines for the treatment of animals in behavioural research and teaching. Anim Behav . 83: 301– 309. CrossRef Search ADS   Aureli F, Schaffner CM, Boesch C, Bearder SK, Call J, Chapman CA, Connor R, Di Fiore A, Dunbar RIM, Henzi SPet al.   2008. Fission-fusion dynamics: new research frameworks. Curr Anthropol . 49: 627– 654. Bates D. 2010. lme4: mixed-effects modeling with R. www.R-project.org/. Accessed 23 March 2017. Bauder JM, Breininger DR, Bolt MR, Legare ML, Jenkins CL, Rothermel BB, McGarigal K. 2016. The influence of sex and season on conspecific spatial overlap in a large, actively-foraging colubrid snake. PLoS ONE . 11: e0160033. Google Scholar CrossRef Search ADS PubMed  Benhamou S, Cornélis D. 2010. Incorporating movement behavior and barriers to improve kernel home range space use estimates. J Wild Manage . 74: 1353– 1360. Google Scholar CrossRef Search ADS   Benhamou S. 2011. Dynamic approach to space and habitat use based on biased random bridges. PLoS ONE . 6: e14592. Google Scholar CrossRef Search ADS PubMed  Bennitt E. 2012. The ecology of African buffalo (Syncerus caffer) in the Okavango Delta, PhD Thesis , Botswana: University of Bristol. Bennitt E, Bonyongo MC, Harris S. 2014. Habitat selection by African buffalo (Syncerus caffer) in response to landscape-level fluctuations in water availability on two temporal scales. PLoS ONE . 9: e101346. Google Scholar CrossRef Search ADS PubMed  Bennitt E, Bonyongo MC, Harris S. 2015. Behaviour-related scalar habitat use by cape buffalo (Syncerus caffer caffer). PLoS ONE . 10: e0145145. Google Scholar CrossRef Search ADS PubMed  Bennitt E, Bonyongo MC, Harris S. 2016. Effects of divergent migratory strategies on access to resources for cape buffalo (Syncerus caffer caffer). J Mammal . 97: 1682– 1698. Google Scholar CrossRef Search ADS   Bennitt E, Bonyongo C, Harris S. Data from: cape buffalo (Syncerus caffer caffer) social dynamics in a flood-pulsed environment. Dryad Digital Repository . http://dx.doi.org/10.5061/dryad.d8j34 Bercovitch FB, Berry PSM. 2010. Ecological determinants of herd size in the thornicroft’s giraffe of zambia. Afr J Ecol . 48: 962– 971. Google Scholar CrossRef Search ADS   Best EC, Dwyer RG, Seddon JM, Goldizen AW. 2014. Associations are more strongly correlated with space use than kinship in female eastern grey kangaroos. Anim Behav . 89: 1– 10. Google Scholar CrossRef Search ADS   Biosa D, Grignolio S, Sica N, Pagon N, Scandura M, Apollonio M. 2015. Do relatives like to stay closer? spatial organization and genetic relatedness in a mountain roe deer population. J Zool . 296: 30– 37. Google Scholar CrossRef Search ADS   Body G, Weladji RB, Holand Ø, Nieminen M. 2015. Measuring variation in the frequency of group fission and fusion from continuous monitoring of group sizes. J Mammal . 96: 791– 799. Google Scholar CrossRef Search ADS   Calenge C. 2006. The package “adehabitat” for the R software: a tool for the analysis of space and habitat use by animals. Ecol Model . 197: 516– 519. Google Scholar CrossRef Search ADS   Chaverri G, Gambia-Rios M, Kunz TH. 2007. Range overlap and association patterns in the tent-making bat Artibeus watsoni. Anim Behav . 73: 157– 164. Google Scholar CrossRef Search ADS   Conradt L, Roper TJ. 2000. Activity synchrony and social cohesion: a fission-fusion model. Proc Roy Soc B . 267: 2213– 2218. Google Scholar CrossRef Search ADS   Cornélis D, Benhamou S, Janeau G, Morellet N, Ouedraogo M, de Visscher M-N. 2011. Spatiotemporal dynamics of forage and water resources shape space use of West African savanna buffaloes. J Mammal . 92: 1287– 1297. Google Scholar CrossRef Search ADS   Couzin ID. 2006. Behavioral ecology: social organization in fission-fusion societies. Curr Biol . 16: R169– R171. Google Scholar CrossRef Search ADS PubMed  Couzin ID, Laidre ME. 2009. Fission-fusion populations. Curr Biol . 19: R633– R635. Google Scholar CrossRef Search ADS PubMed  Cross PC, Lloyd-Smith JO, Getz WM. 2005. Disentangling association patterns in fission-fusion societies using African buffalo as an example. Anim Behav . 69: 499– 506. Google Scholar CrossRef Search ADS   Farine DR, Whitehead H. 2015. Constructing, conducting and interpreting animal social network analysis. J Anim Ecol . 84: 1144– 1163. Google Scholar CrossRef Search ADS PubMed  Fieberg J, Kochanny CO. 2005. Quantifying home-range overlap: the importance of the utilization distribution. J Wild Manage . 69: 1346– 1359. Google Scholar CrossRef Search ADS   Fortin D, Fortin M-E, Beyer HL, Duchesne T, Courant S, Dancose K. 2009. Group-size-mediated habitat selection and group fusion-fission dynamics of bison under predation risk. Ecology . 90: 2480– 2490. Google Scholar CrossRef Search ADS PubMed  Fushing H, Wang H, VanderWaal K, McCowan B, Koehl P. 2013. Multi-scale clustering by building a robust and self correcting ultrametric topology on data points. PLoS ONE . 8: e56259. Google Scholar CrossRef Search ADS PubMed  Fynn RWS, Murray-Hudson M, Dhliwayo M, Scholte P. 2015. African wetlands and their seasonal use by wild and domestic herbivores. Wetlands Ecol Manage . 23: 559– 581. Google Scholar CrossRef Search ADS   Haulsee DE, Fox DA, Breece MW, Brown LM, Kneebone J, Skomal GB, Oliver MJ. 2016. Social network analysis reveals potential fission-fusion behavior in a shark. Sci Rep . 6: 34087. Heinl M, Neuenschwander A, Silva J, Vanderpost C. 2006. Interactions between fire and flooding in a southern African floodplain system (Okavango Delta, Botswana). Landscape Ecol . 21: 699– 709. Google Scholar CrossRef Search ADS   Holmes SM, Gordon AD, Louis EE, Johnson SE. 2016. Fission-fusion dynamics in black-and-white ruffed lemurs may facilitate both feeding strategies and communal care of infants in a spatially and temporally variable environment. Behav Ecol Sociobiol . 70: 1949– 1960. Google Scholar CrossRef Search ADS   Isvaran K. 2007. Intraspecific variation in group size in the blackbuck antelope: the roles of habitat structure and forage at different spatial scales. Oecologia . 154: 435– 444. Google Scholar CrossRef Search ADS PubMed  Jarman PJ. 1974. The social organisation of antelope in relation to their ecology. Behaviour . 48: 215– 267. Google Scholar CrossRef Search ADS   Kelley JL, Morrell LJ, Inskip C, Krause J, Croft DP. 2011. Predation risk shapes social networks in fission-fusion populations. PLoS ONE . 6: e24280. Google Scholar CrossRef Search ADS PubMed  Kie JG. 1999. Optimal foraging and risk of predation: effects on behavior and social structure in ungulates. J Mammal . 80: 1114– 1129. Google Scholar CrossRef Search ADS   Kolbe NR, Weckerly FW. 2015. Home-range overlap of roosevelt elk herds in the bald hills of redwood national park. Calif Fish Game . 101: 208– 217. Korte LM. 2008. Variation of group size among African buffalo herds in a forest-savanna mosaic landscape. J Zool . 275: 229– 236. Google Scholar CrossRef Search ADS   Manly BFJ, McDonald LL, Thomas DL, McDonald TL, Erickson WP. 2002. Resource selection by animals: statistical design and analysis for field studies . Dordrecht: Kluwer Academic Publishers. Markham AC, Gesquiere LR, Alberts SC, Altmann J. 2015. Optimal group size in a highly social mammal. PNAS . 112: 14882– 14887. Google Scholar CrossRef Search ADS PubMed  Mbaiwa JE, Mbaiwa OI. 2006. The effects of veterinary fences on wildlife populations in the Okavango Delta, Botswana. Int J Wilderness . 12: 17– 41. McCarthy TS, Ellery WN. 1998. The Okavango Delta. Trans R Soc S Afr . 53: 157– 182. Google Scholar CrossRef Search ADS   Melletti M, Penteriani V, Mirabile M, Boitani L. 2007. Some behavioral aspects of forest buffalo (Syncerus caffer nanus): from herd to individual. J Mammal . 88: 1312– 1318. Google Scholar CrossRef Search ADS   Merkle JA, Sigaud M, Fortin D. 2015. To follow or not? how animals in fusion-fission societies handle conflicting information during group decision-making. Ecol Lett . 18: 799– 806. Google Scholar CrossRef Search ADS PubMed  Morales JM, Moorcroft PR, Matthiopoulos J, Frair JL, Kie JG, Powell RA, Merrill EH, Haydon DT. 2010. Building the bridge between animal movement and population dynamcs. Phil Trans R Soc B . 365: 2289– 2301. Google Scholar CrossRef Search ADS PubMed  Mourier J, Vercelloni J, Planes S. 2012. Evidence of social communities in a spatially structured network of a free-ranging shark species. Anim Behav . 83: 389– 401. Google Scholar CrossRef Search ADS   Murray-Hudson M, Wolski P, Ringrose S. 2006. Scenarios of the impact of local and upstream changes in climate and water use on hydro-ecology in the Okavango Delta, Botswana. J Hydrology . 331: 73– 84. Google Scholar CrossRef Search ADS   Owen-Smith N, Fryxell JM, Merrill EH. 2010. Foraging theory upscaled: the behavioural ecology of herbivore movement. Phil Trans R Soc B . 365: 2267– 2278. Google Scholar CrossRef Search ADS PubMed  Prins HHT. 1996. Ecology and behaviour of the African buffalo: social inequality and decision making . London: Chapman & Hall. Google Scholar CrossRef Search ADS   Ramos A, Petit O, Longour P, Pasquaretta C, Sueur C. 2015. Collective decision making during group movements in European bison, Bison bonasus. Anim Behav . 109: 149– 160. Google Scholar CrossRef Search ADS   Ramos-Fernández G, Boyer D, Gómez VP. 2006. A complex social structure with fission-fusion properties can emerge from a simple foraging model. Behav Ecol Sociobiol . 60: 536– 549. Google Scholar CrossRef Search ADS   Ramos-Fernández G, Morales JM. 2014. Unraveling fission-fusion dynamics: how subgroup properties and dyadic interactions influence individual decisions. Behav Ecol Sociobiol . 68: 1225– 1235. Google Scholar CrossRef Search ADS   Ryan SJ, Knechtel CU, Getz WM. 2006. Range and habitat selection of African buffalo in South Africa. J Wildl Manage . 70: 764– 776. Google Scholar CrossRef Search ADS   Sigaud M, Merkle JA, Cherry SG, Fryxell JM, Berdahl A, Fortin D. 2017. Collective decision-making promotes fitness loss in a fusion-fission society. Ecol Lett . 20: 33– 40. Google Scholar CrossRef Search ADS PubMed  Sikes RS; Animal Care and Use Committee of the American Society of Mammalogists. 2016. 2016 guidelines of the american society of Mammalogists for the use of wild mammals in research and education. J Mammal . 97: 663– 688. Google Scholar CrossRef Search ADS   Sinclair ARE. 1977. The African buffalo: a study of resource limitation of populations . Chicago: University of Chicago Press. Sueur C, King AJ, Conradt L, Kerth G, Lusseau D, Mettke-Hofmann C, Schaffner CM, Williams L, Zinner D, Aureli F. 2011. Collective decision-making and fission-fusion dynamics: a conceptual framework. Oikos . 120: 1608– 1617. Google Scholar CrossRef Search ADS   Sueur C, Maire A. 2014. Modelling animal group fission using social network dynamics. PLoS ONE . 9: e97813. Google Scholar CrossRef Search ADS PubMed  Taolo C. 2003. Population ecology, seasonal movement and habitat use of the African buffalo (Syncerus caffer) in Chobe National Park, Botswana . PhD Thesis, Trondheim: Norwegian University of Science and Technology. Titcomb EM, O’Corry-Crowe G, Hartel EF, Mazzoil MS. 2015. Social communities and spatiotemporal dynamics of association patterns in estuarine bottlenose dolphins. Mar Mamm Sci . 31: 1314– 1337. Google Scholar CrossRef Search ADS   Vander Wal E, Laforge MP, McLoughlin PD. 2014. Density dependence in social behaviour: home range overlap and density interacts to affect conspecific encounter rates in a gregarious ungulate. Behav Ecol Sociobiol . 68: 383– 390. Google Scholar CrossRef Search ADS   VanderWaal KL, Wang H, McCowan B, Fushing H, Isbell LA. 2014. Multilevel social organization and space use in reticulated giraffe (Giraffa camelopardalis). Behav Ecol . 25: 17– 26. Google Scholar CrossRef Search ADS   Wittemyer G, Douglas-Hamilton, I, Getz WM. 2005. The socioecology of elephants: analysis of the processes creating multitiered social structures. Anim Behav . 69: 1357– 1371. Google Scholar CrossRef Search ADS   Wolf M, Frair J, Merrill E, Turchin P. 2009. The attraction of the known: the importance of spatial familiarity in habitat selection in wapiti Cervus elaphus. Ecography . 32: 401– 410. Google Scholar CrossRef Search ADS   Zengeya FM, Murwira A. 2016. Intraspecific variations in home range overlaps of a semi-free range herbivore are explained by remotely sensed productivity. Int J Geogr Inf Sci . 30: 5– 19. Google Scholar CrossRef Search ADS   © The Author(s) 2017. Published by Oxford University Press on behalf of the International Society for Behavioral Ecology. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com

Journal

Behavioral EcologyOxford University Press

Published: Jan 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 12 million articles from more than
10,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Unlimited reading

Read as many articles as you need. Full articles with original layout, charts and figures. Read online, from anywhere.

Stay up to date

Keep up with your field with Personalized Recommendations and Follow Journals to get automatic updates.

Organize your research

It’s easy to organize your research with our built-in tools.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve Freelancer

DeepDyve Pro

Price
FREE
$49/month

$360/year
Save searches from Google Scholar, PubMed
Create lists to organize your research
Export lists, citations
Access to DeepDyve database
Abstract access only
Unlimited access to over
18 million full-text articles
Print
20 pages/month
PDF Discount
20% off