Por Theo Michelot (University of St Andrews).
Abstract: Large telemetry data sets are collected on wild animals, and they can be used to learn about their behaviour, and their response to environmental features. Hidden Markov models (HMMs) provide a flexible and rigorous statistical framework to describe animal movement as the result of behavioural states. From an HMM fitted to telemetry data, we can infer the sequence of unobserved behavioural states over the duration of the study, and movement parameters for each behaviour. HMMs are relatively fast to fit, usually taking only a few minutes for tens of thousands of data points. The effects of environmental covariates can be modelled, to learn about the drivers of behavioural switches. I will describe the general formulation of HMMs, and then illustrate the models with various data sets from marine and terrestrial species.
