Bayesian generalized dissimilarity model for marine biodiversity analysis

Abstract

Marine biodiversity is crucial for ocean ecosystems and global ecological services. The spatial changes in the biodiversity can be assessed by modeling the beta diversity indices using the Generalized Dissimilarity Model (GDM) which captures nonlinear species-environment relationships through I-splines but the method lacks interval estimates. The Bayesian Bootstrap GDM (BBGDM) also provides confidence intervals but does not incorporate the knowledge of ecological priors. Therefore, this study aimed to propose a Bayesian Generalized Dissimilarity Model (BGDM) that integrated ecological priors such as non-negative regression coefficients into a fully Bayesian framework. Hamiltonian Monte Carlo (HMC) was used for efficient posterior sampling. The results showed that BGDM improved both uncertainty quantification and model interpretability. It was further applied to analyze the marine biodiversity patterns in the Lesser Sunda Islands to show more robust responses to environmental gradients compared to GDM and BBGDM.