Risky rivers: physioclimatic controls of basins' penchant for extreme floods

Reliable appraisal of the occurrence of extreme flood events, their magnitude and probability are crucial for a multitude of societal and economic activities. Extreme floods are however elusive by nature and difficult to estimate through current methods, which heavily rely on limited flood records unable to characterize processes that might be more variable than suggested by available observations.

In this work, we first capitalized on advances in the mechanistic-stochastic descriptions of precipitation, soil moisture and runoff dynamics in river basins to derive a new tool: the Physically-based Extreme Value (PHEV) distribution of river flows. PHEV provides a theoretical underpinning to the study of manifold flood-related issues, such as the emergence of heavy tails in streamflow and flood distributions, flood rich and poor periods, and the watershed features leading to the occurrence of extreme flood events.

We then utilized PHEV to aid investigation of the latter phenomenon in a large set of catchments in the USA and Germany. We positively verified its capability to identify rivers which exhibit a sharp increase of the magnitude of the rarer floods, and the flood value for which this marked increment of magnitude occurs, which we label flood divide. We then leveraged the mechanistic nature of PHEV and a dimensional analysis tool to identify key hydroclimatic and geomorphologic determinants of the occurrence of a flood divide. The hydrograph recession exponent, which embodies the branching pattern of the stream network, and the river flow regime characterized by its streamflow variability play a pivotal role in this regard, as confirmed by observations from the large dataset analyzed in this study.

Finally, we applied binary logistic regression using hydrograph recession exponent and streamflow variability as explanatory variables to predict in what basins the occurrence of flood divides and extreme floods shall be expected. Analyses for independent realizations of subsets of data indicate high prediction accuracy. This remarkable result emphasizes the feasibility of inferring river basins which are prone to generating extreme floods by means of two simple hydrologically-meaningful and readily available indices, thus raising awareness of the intrinsic peril of floods in these cases.