B1719
Title: Modelling species communities through space and time using opportunistic datasets
Authors: Maxime Fajgenblat - KU Leuven & UHasselt (Belgium) [presenting]
Robby Wijns - KU Leuven (Belgium)
Luc De Meester - KU Leuven - Leibniz-Institute of Freshwater Ecology and Inland Fisheries (Belgium)
Thomas Neyens - Hasselt University (Belgium)
Abstract: Understanding how species communities are shaped by local and regional processes is one of the grand aims of ecology. Community datasets are, however, typically challenging to analyse due to their high dimensionality. Recent statistical advances increasingly facilitate the joint analysis of multispecies data and have given rise to several influential joint species distribution models (JSDMs). The rising popularity of JSDMs coincides with the increasing availability of crowdsourced biodiversity data through citizen science initiatives. Most JSDMs, however, cannot deal with the challenges intrinsic to these opportunistic data sources, such as imperfect and heterogeneous detection probabilities. We developed a spatio-temporal joint species distribution model that flexibly acknowledges imperfect detection. Specifically, we combined a spatio-temporal latent factor approach with a comprehensive site-occupancy approach to model occupancy and detection patterns across the considered species. We performed Bayesian inference through the probabilistic programming language Stan and applied the developed model to large datasets on invertebrate occurrences in Belgium. By doing so, we were able to gain insights at both the species and the community level. We argue that extending joint species distribution models to flexibly accommodate imperfect detection enables the study of species communities at an unprecedented scale due to their ability to harness a wider variety of datasets.
