Diminishing wild space and population fragmentation are key drivers of large carnivore declines worldwide. Persistence of large carnivores in fragmented landscapes often depends on the ability of individuals to move between separated subpopulations for genetic exchange and recovery from stochastic events. Where separated by anthropogenic landscapes, subpopulations’ connectivity hinges on the area's socio-ecological conditions for coexistence and dispersing individuals' behavioral choices. Using GPS-collars and resource- and step-selection functions, we explored African lion (Panthera leo) habitat selection and movement patterns to better understand lions’ behavioral adaptations in a landscape shared with pastoralists. We conducted our study in the Ngorongoro Conservation Area, Tanzania, a multiuse rangeland, that connects the small, high density lion subpopulation of the Ngorongoro Crater with the extensive Serengeti lion population. Landscape use by pastoralists and their livestock varies seasonally, driven by the availability of pasture, water, and disease avoidance. The most important factor for lion habitat selection was the amount of vegetation cover, followed by the distance to human settlements and the interaction between those two variables, with selectivity strengths varying with season and time of day. All lions were more willing to approach human settlements at night and during the dry season, selecting strongly for cover when moving closer to humans during the day. Resident females most consistently used areas close to humans, but also relied more consistently on cover than males. Connectivity of lion subpopulations, facilitated by nomadic males, does not appear to be blocked by sparse pastoralist settlements and nomadic males avoided humans on the landscape more strongly than did resident lions. These results are consistent with lions balancing risk from humans with exploitation of livestock by altering their behaviors to reduce potential conflict. Our study lends some optimism for the adaptive capacity of lions to promote coexistence with humans in shared landscapes.

Lion movements

To capture lion movements and use of habitat, we deployed satellite GPS collars (models GPS Plus or Vertex Lite by Vectronic Aerospace (www.vectronic-aerospace.com) on lions of both sexes. Between Oct. 2012- Mar. 2023, we had 22 different individuals collared: eight females (190 months total) and 14 males (252 months total). On average, individual lions were collared 19.9 months (range: 1.7 to 56.4 months) with nine lions being collared for two years or more (S1 Table). The collars had Iridium satellite data transmission (for regular data transfer), VHF-beacon (for real-time manual tracking), remote drop-off function, and were scheduled to take positions every 1-2 hours continuously, day or night. The collar batteries lasted for 2-3.5 years, after which the collar was either removed (using remote drop-off) or replaced. We targeted lions for collaring from adult individuals of either sex from outside the Ngorongoro Crater, in NCA’s multiuse area, from different groups and areas, based on our knowledge of the usual range of their pride or group. We prioritized lions in areas with heightened risk of conflict. Apart from studying the lions’ behavior through fine-scale movements, the purpose of the collars was to provide early-warning to livestock herders in the area.

Lions in NCA community lands are few and elusive, hence immobilization to deploy collars is challenging and done opportunistically, mainly at night, following observations of lions in the area. To attract lions closer to the vehicle for immobilization, we broadcasted a high-volume recording of feeding hyenas, a bleating buffalo calf, or the roars of a lion. The lions were immobilized and collared under permission from the Tanzania Wildlife Research Institute and the Ngorongoro Conservation Area Authority, whose veterinarians immobilized and supervised the collaring of all animals. Lions were immobilized using a dart-gun from a vehicle at 10–20-meter distance, and collars were fitted by an experienced field researcher. All lions were monitored until alert and deemed safe and well recovered after waking up from the immobilization.

Landscape variables

We modelled lion habitat selection using landscape variables (Table 1) representing habitat characteristics and conflict risk (i.e., humans on the landscape). Habitat covariates included the percent of shrub/forest habitat within a 50m radius (hereafter termed cover) to capture lion preference for sheltering and/or stalking under cover (Hopcraft et al., 2005; Loarie et al., 2013); a terrain ruggedness index (hereafter “TRI”) (Riley et al., 1999) to capture lions’ use of rugged, less accessible areas for shelter and/or stalking; an enhanced vegetation index (hereafter “EVI”; MODIS) to capture greenness as an indication of herbivore (lion prey) abundance; and the distance to the nearest riverbed (S1 Figure) which lions commonly use for stalking prey or resting in thick riverine vegetation (Hopcraft et al., 2005; Mosser et al., 2009). Although EVI values for the region are available monthly, to reduce computation time and model complexity, we summarized EVI for wet (Dec-May) and dry (Jun-Nov) seasons, respectively, by averaging over each year from 2012-2022. We used the month of April to represent the wet season and October to represent the dry season, typically the peak months of greenness for each season (Metzger et al., 2015). Human-related covariates included the distance to the nearest boma (hereafter “human activity; S1 Figure), the distance to the nearest water point, and the density of bomas within a defined radius (3km, 5km; see below) (hereafter, “intensity of human activity”). Since many bomas and water points shift seasonally, we created separate covariates for distance to human activity, intensity of human activity, and distance to water points to model wet and dry seasons separately.

Models

We used SSFs to quantify lions’ habitat selection at the local scale by comparing habitat covariates at locations that the lions had visited with habitat covariates at a random set of available locations (i.e. the habitats available along animal movement tracks) (Avgar et al., 2016). The available locations for SSF modelling are conditioned on the last step taken by the animal, meaning they are relative to the current location, resulting in a stratified dataset (Muff et al., 2020). Again we used the amt package (Signer et al., 2019) to estimate each animal’s movement parameters for step length and turn angle, then sampled from these parameter distributions to generate random available locations (n=50 locations) around each used location. We did not restrict this analysis to the study area, allowing inclusion of long-distance tracks around which we had targeted additional collection of covariate data. We extracted covariate vectors for each location and used the Poisson formulation of conditional logistic regression to estimate associated slope coefficients (Fieberg et al., 2021). We included random intercepts for each strata (i.e., set of used and available steps) (Muff et al., 2020) and used R package glmmTMB to run our models, fixing the variance of the random intercept for strata to 10³ (Muff et al., 2020).