State–space models link elk movement patterns to landscape characteristics in Yellowstone National Park

Forester, J. D., Ives, A. R., Turner, M. G., Anderson, D. P., Fortin, D., Beyer, H. L., … & Boyce, M. S. (2007). State–space models link elk movement patterns to landscape characteristics in Yellowstone National Park. Ecological Monographs77(2), 285-299.

Classic analyses of animal movement paths (e.g. from GPS loggers) are based on correlated random walks, which are essentially null models where “step lengths” (usually the distance traveled between GPS fixes) are drawn from a distribution and turning angles are autocorrelated. Any departure from these random walks can be used as evidence for biological processes such as habitat selection. However, these methods usually use an “average” landscape over an area, rather than one that an animal is experiencing at any given moment, whereas in reality animals are constantly responding to stimuli, including both external (e.g. patch quality) and internal (e.g. hunger) states. Further, there are feedbacks between animals and their environments that are not considered by models that assume an “average” landscape.

In this paper, Forester et al. (2007) develop a state-space model of animal movement, which considers both the behavioral state of an animal and landscape characteristics concurrently. Their model models step length as a product of two main processes: an immediate response to the environment and an autocorrelated response to the (past) environment. By allowing this second process to vary in strength, they are able to model the importance of an animal’s behavioral state in determining movement speed. They apply this model to both simulated and real GPS and landscape data of elk in Yellowstone National Park to examine the relationships between landscape characteristics, autocorrelation of movement, and home range size.

Overall, the authors found that when the importance of autocorrelation was high (i.e. behavioral state was strongly dependent on past conditions), both simulated and real elk had larger home ranges than when the importance of autocorrelation was low. This result suggests that home range size depends not only on landscape structure, but also on the intrinsic tendency of animals to move according to their past experience. Surprisignly, they also saw large variability in the importance of any given landscape characteristic for the movements of GPS-tagged elk, pointing to a high inter-individual variation in movement strategy within the population. The authors suggest that behavioral responses may occur on multiple spatial and temporal scales, which may account for the inconsistency of parameter estimates between individuals. They conclude that the state-space models provide a flexible and generalizable modeling strategy for animal movements.