Regional classification using gradients of marine species assemblages: a data-driven approach to modelling marine ecosystems

Marine management and conservation efforts often rely on predictive modelling of species observations, the output of such models can be influenced by their regional extent. This study proposes a data-driven classification of marine regions by clustering modelled gradients of species assemblages. Two clustering methods are considered, the CLARA algorithm and mean-shift segmentation, and compared with depth and geographically stratified regions. Regional classification was applied to models using three methods: Regional indices as categorical predictors, regional ensemble models, and a pre-calibration regional data-filter. Regional influence was measured in changes of MSE and R2 values. Large changes in model output were restricted to a small number of anomalous species models. Mean-shift clustered regions produced moderately improved MSE and R2 values compared to the other methods. Regional influence in the species distribution models were shown to be species dependent, necessitating an assessment of relevant species included in regional classification.