Continuous time resource selection analysis for moving animals

1.Resource selection analysis (RSA) seeks to understand how spatial abundance covaries with environmental features. By combining RSA with movement, step selection analysis (SSA) has helped uncover the mechanisms behind animal relocations, thereby giving insight into the movement decisions underlying spatial patterns. However, SSA typically assumes that at each observed location, an animal makes a 'selection' of the next observed location. This conflates observation with behavioural mechanism and does not account for decisions occurring at any other time along the animal's path.

2.To address this, we introduce a continuous time framework for resource selection. It is based on a switching Ornstein‐Uhlenbeck (OU) model, parameterised by Bayesian Monte Carlo techniques. Such OU models have been used successfully to identify switches in movement behaviour, but hitherto not combined with resource selection. We test our inference procedure on simulated paths, representing both migratory movement (where landscape quality varies according to season) and foraging with depletion and renewal of resources (where the variation is due to past locations of the animals). We apply our framework to location data of migrating mule deer (Odocoileus hemionus) to shed light on the drivers of migratory decisions.

3.In a wide variety of simulated situations, our inference procedure returns reliable estimations of the parameter values, including the extent to which animals trade‐off resource quality and travel distance (within 95% posterior intervals for the vast majority of cases). When applied to the mule deer data, our model reveals some individual variation in parameter values. Nevertheless, the migratory decisions of most individuals are well‐described by a model that accounts for the cost of moving and the difference between instantaneous change of vegetation quality at source and target patches.

4.We have introduced a technique for inferring the resource‐driven decisions behind animal movement that accounts for the fact that these decisions may take place at any point along a path, not just when the animal's location is known. This removes an oft‐acknowledged but hitherto little‐addressed shortcoming of stepwise movement models. Our work is of key importance in understanding how environmental features drive movement decisions and, as a consequence, space use patterns.