Industrial development and increased resource demand as a result of human population growth are drastically altering the environment, disrupting the balance of ecosystems relied upon for key services and increasing instances of human-wildlife conflict, risks of zoonotic infection, and instability of food source populations. Models of animal movement, space-use, and habitat selection provide insight which can be used to safeguard against these risks, by informing the conservation and management of wild and domesticated species. Historically, different approaches to habitat selection modelling have been developed independently in relation to different systems and data structures. Recently, demonstrations of the equivalence of point process methods to various approaches used in the analysis of species distribution and movement data has posited spatial and spatio-temporal point processes as a unifying framework for habitat selection modelling. The history of point process literature has been largely theoretical, due to a lack of available computationally efficient methods for fitting these models to large and complex datasets. However, this has changed with the recent development of the Integrated Nested Laplace Approximation (INLA) method for inference and its associated software packages R-INLA and R-inlabru. Data integration has also become a topic of current research interest in species distribution and movement ecology, promoting the emergence of joint likelihood models as a key framework for ongoing development in this area. Consequently, there is a demand in the ecological literature for demonstrations of the applications of joint likelihood spatio-temporal point process models to large and complex ecological datasets. This forms the underlying motivation for this thesis, which provides an exploration of the use of these methods in modelling animal movement, space use, and habitat selection, with applications in different areas of ecology. The work included in this thesis is presented in the form of three case studies, which each demonstrate a different methodological framework and ecological application for joint likelihood spatio-temporal point process modelling of habitat selection data. Chapter 2 demonstrates a marked point process approach to modelling the spread of a reintroduced population of Eurasian crane (Grus grus). Chapter 3 compares between habitat selection models at different organisational scales to analyse cattle (Bos taurus) tracking data, with applications in livestock management. Chapter 4 introduces the novel implementation of a joint likelihood framework for integrating survey and telemetry data in R-inlabru and demonstrates the advantages of this approach using simulated data. Finally, the approach developed in Chapter 4 is applied to real data in Chapter 5, in which it is used to understand habitat selection in a semi-domesticated reindeer (Rangifer tarandus tarandus) population in an area of land-use conflict. Overall, this thesis provides an exploration of different approaches and applications for joint-likelihood spatio-temporal point process modelling of animal movement, space-use, and habitat selection. This includes novel methodological contributions, such as extensions to the unique integration scheme used in the GF-iSSA movement model; and the implementation of the first individual-level continuous-time habitat selection model in R-inlabru. Key themes of accounting for availability; the impacts of spatial, temporal, and organisational scale on inference; and the balance between model complexity, interpretability, and computational efficiency are investigated throughout. Main results provide insights into the relationship between model complexity and performance; and the relevance of the spatial scales of heterogeneity in covariate structure, and of representations of availability, on habitat selection inference.
Megan Morton (Thu,) studied this question.