Spatio-temporal joint species distribution modelling: a community-level basis function approach

Title: Spatio-temporal joint species distribution modelling: a community-level basis function approach
Speaker: Dr Francis K.C. Hui, ANU
Date/Time: Thursday 21 October 2021, 12:00 noon
Venue: La Trobe University
Zoom Link: latrobe.zoom.us/j/98357628534

You are welcome to attend the following Statistics and Stochastic colloquium (part of the Colloquium Series of the Department of Mathematics and Statistics) at La Trobe University.

Abstract:  The last decade in ecology has seen the development and rising popularity of joint species distribution-modelling approaches for studying species assemblages, with by far the most common approach being based around generalised linear latent variable models (LVMs). However, while methodological and computational advances continue to be made with LVMs, their application to spatio-temporal multivariate abundance data, i.e. observations of multiple species recorded across space and/or time, remains computationally challenging and not necessarily scalable when it comes to fitting and inference.

In this talk, we propose an alternative approach to spatio-temporal joint species distribution modelling which breaks away from the LVM framework. Inspired by the concept of fixed rank kriging, we employ a set of fixed, community-level spatial and/or temporal basis functions, with corresponding species-specific random slopes to account for spatio-temporal correlations both within and between species. The resulting community-level basis function model (CBFM) can be used for the same array of purposes as LVMs, but is designed to be computationally much more efficient given they can be set up and thus fitted using the same machinery as for generalized additive models.

Simulations and an application to a demersal fish dataset collected off the Northeast US continental shelf illustrate the potential of CBFMs for scalable spatio-temporal joint species distribution modelling.