Statistical movement ecology has grown rapidly in recent years due to advances in animal tracking technologies and increasing concern for the natural environment. This has initiated the development of models to capture the complex behaviours of animals using this detailed data. Since data is gathered at discrete time intervals, many models are formulated in discrete time; however, these suffer from problems when data is irregularly spaced or observed with error. Continuous-time models are able to overcome this, but often at a high computational cost, and uptake is low due to parameter interpretability.
In this thesis, we introduce the continuous-time velocity-jump model, which links movement behaviour to space use by parametrising the stationary distribution as an animal’s utilisation distribution. Spatial covariates can be included, allowing the integration of movement and environmental influences. As a case study, we demonstrate how the velocity-jump model can generate realistic simulations of seabird movement, showing how multiple behavioural states, prey distributions and wind farms can be incorporated.
We develop novel inference methods for the velocity-jump model by approximating it as a time-inhomogeneous hidden Markov model (HMM), in which the state indicates whether a velocity “refreshment” event occurred. Two inference approaches are presented: the first ignores the previous state, assigning states based on velocity correlation; the second conditions on the previous interval, modelling event velocities as a weighted average of neighbouring velocities and integrating out the event time. This second method is further extended to incorporate a utilisation distribution defined by environmental covariates. Finally, we illustrate its application to studying foraging behaviour of three seabird species during the breeding season.
