A Bayesian joint model for exploring zero-inflated bivariate marine litter data

Acknowledging the spatial and spatio-temporal behavior of natural processes is crucial for management purposes. Semi-continuous datasets are common in Ecology: combining information on occurrence and conditional-to-presence abundance of species allows to improve environment effects estimates. Based on a marine litter case study, this paper proposes a two-parts model to handle 1) the zero-inflation problem and 2) the spatial correlation characterizing abundance monitoring data. In the spirit of multi-species distribution models, we propose to jointly infer different litter categories in a Hurdle-model framework. Shared spatial effects that link abundances and probabilities of occurrences of litter categories, are implemented via the SPDE approach in the computationally efficient INLA context.