Embracing Large-sample Data to Characterize Streamflow Extremes at a Global-scale

The recurrent and destructive nature of floods causes enormous economic damage and loss of human lives, leaving people in flood-prone areas fearful and insecure. It is essential to have a thorough knowledge of the factors that contribute to it. However, most catchment characterization studies are limited to case studies or regional domains. A detailed global characterization is currently unavailable due to the limitation in the holistic dataset that it demands. This study aims to fill this gap by utilizing multiple global datasets describing physiographic explanatory variables to characterize streamflow extremes. The role of catchment features such as landcover, geomorphology, climatology, lithology, etc., on spatial patterns and temporal changes of high streamflow extremes, was investigated in detail. Moreover, the multidimensional correlations between streamflow extremes and catchment features were modeled using a Random Forest approach and integrated with an interpretable machine learning framework to find the most dominating elements in different climate classes. The interpretation reveals that climatological variables are the most influential across all climates. However, the variables and their influences fluctuate between climates. Furthermore, distinct geomorphological variables dominate throughout climatic classes (such as basin relief in warm temperate and drainage texture in arid climates). Overall, the insights of this study would play a vital role in estimating the unit peak discharge at ungauged stations based on known watershed features. In addition, these findings can also help assess the nature of extremes in future climate scenarios, consequently implicating risk management methods.