Keywords: particle filters, telemetry, animal movement, Bayesian computation, tracking, Kalman filters, ecology
Particle filters are a computational Bayesian technique used to sequentially model a dynamic system, such as a moving object’s sensor measurements over time. We show an application of this technique to modeling animal movement---in our case sharks---with emphasis on statistical inference on the underlying parameters, such as speed and turning angle. Specifically, we use the parameters to infer the animal’s unobserved behavioral state (feeding vs transiting), behavioral influences between animals, and regional (i.e., ecological) differences in the animals’ behavior. Additionally, we will address the issue of modeling the animal’s behavior when the location measurements occur at irregularly-spaced time intervals; regular intervals are a feature that many animal tracking movements assume, but which often is not feasible. Because particle filtering is a topic that is not typically familiar to many statisticians, we also will present brief animations to illustrate these algorithms.