Assessing space use and landscape navigation
To determine the individual home ranges of the seven collared hyenas, we
used the ‘adehabitatHR’ package in R (R Core Team, 2018) to assess 50%
and 95% kernel utilization distributions (KUD) and calculated overlap
among home ranges.
To determine landscape feature selection by spotted hyenas, we derived
resource selection functions (RSFs) using the ‘lme4’ package in R, using
only the six collared hyenas (3 male and 3 female) whose collars
remained active throughout the study. Random points generated were equal
to the number of GPS points used, and we found no evidence of
collinearity among our covariates (Dellinger et al. 2013). We
assessed resource selection using generalized linear mixed-effects
models with a logit link. Using the raster package and base R, we scaled
the following covariates for use in the RSFs: normalized difference
vegetation index (NDVI; 30m, Landsat 8), elevation, slope, distance to
rivers, distance to lakes, distance to roads, and distance to protected
area boundaries. NDVI in this study site can serve as a proxy for land
cover, because areas of higher NDVI are generally brush or forest,
whereas lower NDVI areas are typically grasslands. After initial data
exploration, we included an interaction term—individual and distance
to boundary—as a random effect in our candidate model to control for
variation in individual boldness. We assessed a global model (for all
hyenas) and compared global seasonal (rainy and dry) models, as well as
models for hyenas whose dens were in LNNP (n=3), and for hyenas whose
dens were in Soysambu Conservancy (n=3). We also assessed and compared
models with the following combinations of variables: ecological
variables only, ecological and infrastructure, infrastructure only,
infrastructure and human perception/experience, and human
perception/experience only. Models were ranked based on their Akaike
Information Criterion (AIC; Burnham & Anderson 2002), and models within
Δ AIC ≤2 were retained to use when assessing coefficient values.