Using spatio-temporal models to assist ecosystem-based fisheries management: a New Zealand case study

Tools to understand or predict ecosystem changes (e.g., fish community analyses) and the impacts of management measures on ecosystem components (e.g., ecosystem simulation models) are instrumental to the success of ecosystem-based fisheries management (EBFM). We developed a workflow using vector autoregressive spatio-temporal (VAST) models to assist EBFM in the Chatham Rise middle depth (CHAT MD) region, located in New Zealand waters. Our workflow employs research survey biomass catch rate data collected within the CHAT MD region and informs EBFM indirectly, by delivering inputs to ecosystem simulation models, and directly, by supporting spatial and temporal fish community analyses. While many previous studies worldwide used only joint species spatio-temporal models (JSSTMs) to inform EBFM, our results highlight that good practice is to develop and compare single-species models and JSSTMs for all species and for phylogenetically related species, to ultimately retain for each individual species the type of spatio-temporal models that is most appropriate to address the study objectives. We then show how the predictions of the most appropriate spatio-temporal models can be used to generate maps of species density, annual biomasses, and spatial overlap indices for parameterisation or fitting of ecosystem simulation models. Next, we showcase spatial fish community analyses employing spatio-temporal model densities, including biogeography and beta diversity analyses, which can contribute information to marine spatial planning efforts targeting multiple fish species. Finally, we demonstrate temporal fish community analyses (principal component analyses, chronological clustering, and heatmaps) relying on spatio-temporal model biomasses, which can, for example, help resource managers evaluate fish community health relative to historical changes to reflect on management measures for fisheries catching multiple species (as targets or bycatch). Our workflow could easily be adapted to other statistical spatio-temporal modelling platforms (e.g., tinyVAST) and other regions worldwide.