Improving species distribution models for climate change studies: ecological plausibility and performance metrics

• Ecological plausibility of species-distribution models is often neglected • Traditional optimal performance metrics often yield implausible models • Temperature response should comply with the ecological niche concept • Performance metrics based on decadal changes are better for climate studies • Temperature bell-shape responses provide better models for climate projections Species Distribution Models (SDMs) are widely used tools for studying potential climate-induced shifts in species distribution to support future marine spatial planning. However, the ecological plausibility of selected models is often neglected, particularly in marine ecosystems, resulting in potentially misleading outcomes, especially when climate effects are of concern. In this study we modeled the distribution of 11 commercial fish species in the North Sea using 57 years of observations and 60 SDMs with various degrees of freedom and aimed to improve the ecological plausibility of SDMs by evaluating common model selection techniques. Model performance was evaluated using deviances obtained with three cross-validation designs, Akaike Information Criterion, median absolute deviation and percentage of local mismatch. We identified top performing models based on consistent good scores of those metrics and assessed the ecological plausibility of all models. Specifically, we tested whether the modeled temperature response curve aligned with the ecological niche concept, i.e. having a bell shape within the plausible temperature range for each species, where the highest habitat suitability should relate to optimal conditions. The tested performance metrics often yielded conflicting outcomes and selected models with implausible temperature response curves that had poor extrapolation skills in temperature space and, thus, may result in unreliable predictions under climate change. Building on our findings, we provide recommendations for future SDM applications to improve their accuracy, ecological plausibility and predictive skills in climate-related studies.