Spatial and temporal predictions of moose winter distribution.

Herbivores are usually distributed unevenly across the landscape often because of variation in resource availability. We used zero-inflated generalised additive models (to account for data with a high number of zeros) that include georeferences to predict winter distribution of a large herbivore (moose Alces alces). Moose distribution was analysed in relation to forage availability and distance to neighbouring sites. Our results showed that the ability to explain moose distribution indexed by pellet count data at a local scale increased when spatial information (longitude and latitude) was added to the model compared to the model only including food availability. By using the relationship between moose and forage distribution, and the spatial information, we predicted patch choice by moose reasonably well in 2 out of 4 years. However, the distribution of moose was also influenced by weather conditions, as it was most clumped in the year with most snow. In conclusion, our study lends support for a non-linear approach using georeferences for a comprehensive understanding of herbivore distribution at a small scale. This result also indicates that the use of a certain patch by moose not only depends on the selected patch itself but is also influenced by the neighbouring patch and factors at a larger spatial scale, such as moose management influencing the density above moose home range level. The relatively high proportion of unexplained variation suggests that the use of a certain patch is also influenced by other factors such as topography, predation, competition, weather conditions, and wildlife management strategies.