Modelling resource utilisation of temporally irregular and spatially autocorrelated GPS data of red deer

The application of resource selection functions (RSF) for large datasets of GPS-positions is limited by the relatively high proportion of used grid cells compared to the available area. Resource utilisation functions (RUF) assume that the degree of use of areas without relocations follows a kernel density distribution of the relocations (Millspaugh et al. 2006). This assumption though is often violated by the fact that most relocation techniques are discrete in time and therefore display only a narrow time window of resource use. We have modified the RUF method by relating kernel density values to available resources at the site of relocations only. Contrary to grid cells, relocations are not systematically distributed in space and often not systematically sampled in time, and we accounted for temporal and spatial autocorrelation by applying the time kernel method (Katajisto and Moilanen 2006) and subsampling relocations with a minimum nearest neighbour distance defined by the relocation error and the map pixel size of the resources. We used generalized linear mixed models (GLMM) with a spatial covariance matrix and different combinations of random intercept and coefficients (Gillies et al. 2006) to describe resource use. Different resource variables were derived from vegetation maps based on Landsat images and from a terrain model.