Search

Investigating the relationship between organisms and the environment has long been a focus of study in the conservation and ecology fields. Central to this research is the use of ecological models to explain, predict and project species distributions. Transferring ecological models into novel environments, both spatially and temporally, can prove vitally important when there is insufficient response data to create a model in the target area. However, there remain gaps in the knowledge and unanswered questions surrounding the concept and practice of transferring models. Therefore, in this thesis, I investigate how the performance and transferability of correlative SDMs are influenced by 1) the number of points a model is trained with, 2) the spatial resolution of predictor data, and 3) the choice of algorithm used. This research focuses on twenty amphibian, bird, insect, mammal, plant, and reptile species and utilises three popular correlative modelling algorithms; boosted regression trees (BRTs), generalised linear models (GLMs) and Maxent, before I investigate the transferability of a further five algorithms and an ensemble approach. Furthermore, I investigate the transferability of a simple and potentially generic mechanistic risk model for an emerging plant pathogen. In general, the correlative models, particularly the machine learning methods performed well and were transferable, though to what degree varied by the algorithm chosen and species modelled. However, in all chapters, perhaps the greatest influences on model transferability were data quality and differences in data between the area in which models were trained and transferred to. Nevertheless, this research demonstrates model transferability is achievable and can be improved through testing and selecting the most appropriate modelling approach, resolution, and complexity for both the correlative and mechanistic models.