Fig. 2: Decoupled spatial drivers of species occurrence for Alburnoides bipunctatus in the Aare-Rhine catchment of Switzerland.

a shows the predicted occurrence from a random forest model fitted to presence-absence data, the map inset places our focal catchment region (red) in the wider Rhine River watershed (black). b shows variable importance calculated as the average of absolute SHAP values in the calibration dataset, indicating the average variable contribution to the overall prediction in the calibration dataset. c shows the distribution of SHAP values per variable, indicating whether contributions are positive or negative. Red intensity indicates variable importance in (b–c). d shows the pairwise Spearman’s rank correlation between each variable’s SHAP values, indicating the extent to which variable contributions to predictions are spatially correlated. Blue indicates positive correlations, and red indicates negative correlations. e–l shows the spatial pattern of SHAP values and, therefore, indicates the contribution of a variable to a prediction in a given sub-catchment. We do not show the maximum temperature to simplify the figure because it was a highly similar pattern to the minimum temperature. Positive (blue) and negative (red) contributions to the environmental suitability prediction show distinct spatial patterns to each threat, as also indicated in (d). Insets show the environmental values on the x-axis and SHAP values on the y-axis and, therefore, indicate the shape of the species response curve (see Supplementary Fig. 18 for response curves across all species). Note the change in scales between panels indicated by the y-axis of the insets. Data required to reproduce this Figure is available in Supplementary Data 1 of our Figshare repository⁸⁶.