Modelling latent animal movement and behaviour in statistical ecology with spatial hidden Markov models

Animal movement and behaviour are key processes that drive scientific observations made on wild populations, allowing us to infer where animals spend their time, how they interact with their environment, and how many animals there are in a population. One obstacle for statistical methods when modelling animal movement or encounters between animals and scientific observers is that both the animal’s true movement path and their changing behavioural state are unobserved.

In this talk, I will introduce a flexible modelling framework that uses spatial hidden Markov models where animal movement and behaviour are jointly modelled as a latent process and inferred from a wide variety of possible observation processes including animal telemetry data and widely used population abundance survey methods such as distance sampling and spatial capture-recapture. The movement-behaviour process is described by a system of advection-diffusion partial differential equations that can be approximated by a continuous-time Markov chain. With this approximation and further computational techniques, the marginalisation over all possible movement-behaviour histories of an animal can be efficiently implemented, allowing models with latent animal movement and behaviour to be fit by marginal maximum likelihood.

In this talk, I will show three examples of how this general statistical model can be used to improve upon existing approaches in statistical ecology including reducing bias in population size estimates for Eastern spotted dolphins in the Tropical Pacific, quantifying inter-individual interactions between jaguars in Belize from camera trap data, and inferring the effect of spatial covariates on marine animal movement whilst respecting land boundaries.