Bayesian Hierarchical models for presence-availability data, with application to GPS locations of bears and wolves.

Resource Selection Functions allow knowing which habitat characteristics are preferred by a species. To determine them, two types of locations are considered: a set of locations in which the animal has been recorded (usually through GPS), which are called “used locations”, and a set of “available locations” where the individual has not been observed. The latter are usually sampled uniformly over the animal home range. The thesis aims to construct Bayesian Hierarchical models for use-availability data of two animals that are modeled jointly. We formalized four models and, through a simulation study, we evaluated the identifiability of the parameters and the convergence of the model with respect to the number of available locations. The models are then estimated on a real dataset of GPS locations of bears and wolves, collected in the PNALM area (National Park of Abruzzo, Lazio, and Molise). We fit a different model for each of the four time-windows where the data are recorded (spring, early summer, late summer, autumn), using environmental data as model covariates. Results obtained are interpretable and give insights on how the animals interact and choose the resources.