Global-scale characterization of streamflow extremes

• A global characterization of streamflow extremes was performed based on a representative dataset augmented with geo-climatic variables at catchment scale and combined random forest an interpretable machine learning framework (SHAP). High unit peak discharges are characterized by meteorological such as annual precipitation, precipitation wettest month quarter, geomorphologic basin magnitude, first-order streams length, perimeter, drainage area. Envelope curves over the exhibit maximum magnitude floods higher than those developed continental United States Europe. It is observed that discharge values extreme hydrologic events mountainous terrain along oceans high across world. The increasing risk globe needs focused attention because extensive damage to human lives economy. comprehensive understanding its causative factors vital importance. Yet studies generally limited case or regional domains. currently unavailable, which requires collecting collating large number datasets vast areas. This study embraces large-sample data-driven science new paradigm characterize utilizing physiographic explanatory could explain various facets streamflows. Along spatial temporal variations extremes, their correlation characteristics geomorphology, meteorology, climatology, landcover, lithology, etc. were examined. multidimensional relationships between modeled using Random Forest approach identify most dominant in varying climate classes. Interpretation SHAP (SHapley Additive exPlanations) reveals influential climatic However, influences change among different Moreover, geomorphological come into dominance classes (such relief warm temperate texture arid climates). Overall, insights from play crucial role predicting ungauged stations known characteristics. these findings can also formulating management strategy.