Bayesian estimation of multiple ecological abundances

Ecological processes driving the spatial and spatio-temporal distribution of marine species are complex to assess.Infact in several ecological studies, counts, abundances or biomass of interacting species are collected from different sites resulting in sparse datasets that include highly correlated responses. The analysis of relationships among such responses requires a suitable statistical framework to globally study the ecosystem, including relevant variables and combining in a single step environmental and community information. Inspired by Joint species distribution models, we propose a Bayesian model-based approach to deal with the zero-inflation issue, common to semi-continuous data, and with the spatial (and spatio-temporal) structure of abundance monitoring data.The proposal takes its cue from a case study concerning marine litter data collected by fishery surveys in the central Mediterranean. To jointly infer different litter categories, a multiple response Hurdle-model is proposed. This model allows to combine both information on occurrence and conditional-to-presence abundance of litter categories and the effects of environmental potential drivers. Shared spatial effects that link abundances and probabilities of occurrences, together with temporal effects, are efficiently implemented using the SPDE-INLA approach. Results support the possibility of better understanding the spatio-temporal dynamics of marine litter in the study area.