Figure 4.

Neural Network predictions for alpha, beta, and gamma diversity of vascular plants. The neural network models were trained separately on alpha, beta, or gamma diversity estimates, which we compiled from available vegetation plot data (Figure 1). The alpha diversity maps (left column) show the number of vascular plant species expected to be found in a 500-m² plot (most common plot-size found in the vegetation plot data; Supplementary Figure S2). The beta diversity maps (center column) quantifies the spatial turnover and differences in species compositions (Sørensen dissimilarity index, relative to the total diversity) between such 500 m² plots within each grid cell (10 × 10 km). The gamma diversity maps (right columns) show the total species richness within each grid cell. The top row shows the predictions averaged across an ensemble of 50 independently trained models, using different starting seeds. The center row shows the coefficient of variation for each grid cell, as a measure of prediction uncertainty. High values (dark grey/black) correspond to grid cells with less consistent diversity predictions. The bottom row shows the average diversity predictions for only those grid cells with the most consistent diversity predictions (coefficient of variation smaller than median across all grid cells), while high-uncertainty grid cells are marked in grey.