Fig. 6: Comparison of regional flood probability distributions of flood generation processes aggregated for two different periods.

a–d Regional kernel density estimates of flood magnitudes (Qmax scaled by mean maximum annual flood for each study catchment) aggregated for each region and flood generation process. Insets provide a zoom into the right tails of corresponding distributions. The right tails of probability density functions are only shown up to the value of 4 (four times the mean maximum annual floods). The significance of differences of flood magnitude distributions of flood generation processes aggregated for two different periods (1960-1985 (white background) and 1985-2010 (grey background)) is tested with the pairwise two-sided Kolmogorov-Smirnov test (α = 0.01) applying a false discovery rate correction. Significant cases are displayed as thick lines. Differences between different flood generation processes within the same period are significant for all displayed cases (α = 0.01) according to the pairwise two-sided Kolmogorov-Smirnov test with false discovery rate correction (see Supplementary Table 3). n is the number of flood events in each sample. Only flood generation processes that showed considerable changes in their regional frequency (Fig. 2) and have considerable deviations in their frequency during opposing modes of flood anomalies (Fig. 4b) are displayed. The remaining processes are displayed in Supplementary Fig. 10.