How the Choice of Spatial Resolution Affects Freshwater Fish Species Distribution Models

Species distribution models (SDMs) are widely used to understand and predict how species respond to changes in their environment. Hence, it is critical to understand how SDMs are influenced by methodological aspects and choices. The choice of spatial resolution is an understudied subject, especially for freshwater fish SDMs. Here, we aimed to analyse how the choice of spatial resolution affects the predictive accuracy, predictor variable importance and spatial predictions of SDMs of freshwater fish species in Europe. We fitted SDMs for 49 freshwater fish species based on point occurrence records combined with environmental and anthropogenic predictor variables aggregated at five spatial resolutions as represented by nested hydrological basins (level 8 to 12 in the HydroSHEDS database). Following an ensemble modelling approach, we employed nine algorithms to establish the SDMs at each of the resolutions, using temperature, topography, streamflow, land use, human population and dam density as predictor variables. We analysed the differences in predictive performance by testing the models on a geographically independent subset of the data. Predictive performance and variable importance were highly similar across the resolutions, with a median TSS between 0.37 and 0.39 and temperature and topography being the most important variables. In contrast, predicted range size decreased towards higher resolutions, while the difference between predicted and observed range size increased. Our results indicate that the choice of spatial resolution has a small effect on the performance and predictor importance of continental freshwater fish SDMs, while significantly influencing predicted range size. The latter may have important consequences for conservation and extinction risk assessments, which often rely on estimates of range size. From a precautionary principle, establishing SDM at the highest resolution possible may help to prevent the risk of overestimating range size hence underestimating extinction risk.