Abstract

A pairwise dissimilarity index is a measure used to quantify changes in species composition within a spatial region. The index is often calculated using species presence-absence data and, typically, statistical models are fitted to indices in order to identify key environmental drivers of species-diversity change. However, such a modelling approach does not make a distinction between the noisy, incomplete species data and the underlying ecological process. Consequently, dependence in the process' pairwise dissimilarities is not properly accounted for. Here, we use a hierarchical statistical framework that incorporates a data model and a process model. This framework takes into consideration data-sampling errors and enables inference on the latent pairwise dissimilarities directly from species data. Moreover, we model the spatial dependence in dissimilarities using a generalised chi-squared process, which encourages a stronger correlation between pairs of dissimilarities that are close in space. We study the model analytically and through simulation studies, and we apply our methodology to an ecological data set. This is joint work with Andrew Zammit-Mangion and Noel Cressie from the University of Wollongong.