Title: Flexible inference for continuous-time models of wildlife movement

Abstract:

The majority of statistical models of animal movement are formulated in discrete time, modelling separately each `step’ from one location (e.g. GPS fix) to the next. This can make it difficult to deal with missing or unequally-spaced observations, to compare studies with different time scales, or to interpret results biologically. In reality, animals exist and move in continuous time, and I will describe some switching diffusion models that try to capture some of the complexities of real behaviour in continuous time. Computational cost is an increasingly important issue in fitting movement models, and I will talk about some algorithms that allow exact inference for such models, even in the presence of spatial heterogeneity, borrowing ideas from Hidden Markov Models and Kalman Filtering. An increasingly important area of application is collective movement, where we model the locations of simultaneously-tracked animals as they interact; I will discuss some recent developments in modelling and computation for this situation.

Some of this work is joint with Mu Niu (University of Plymouth) and a number of recent or current research students at the University of Sheffield.