Data-Driven, Physically-Based Characterization of Floods Accounting for Sub-Basin Precipitation Variability

Floods account for roughly one third of all global geophysical hazards. The ability to predict and characterize floods is increasingly important, and specifically the need to account for rainfall spatial variability is paramount in order to achieve effective flash flood characterization. Flood characteristics such as location, timing, flashiness, duration, are primarily driven by the causative rainfall and basin geomorphology. Disentangling the complex interactions between precipitation forcing and hydrological processes is a challenge because of the lack of observation datasets capturing diverse conditions.

We propose here a broad characterization based on a database comprised of morphological, climatological, bioclimatic, streamflow and precipitation data from over 21,000 flood-related rainfall events that occurred over 900+ different basins over the Conterminous U.S. during 2002-2011. Basin parameters are complemented with indices derived from radar-based precipitation and representing sub-basin scale rainfall spatial variability. We specifically focus on the impact of rainfall spatial variability on basin lag time and flood stage threshold exceedance, and introduce data-driven supervised machine learning approaches for characterizing flooding conditions. Both classification and regression models were constructed, and variable importance analysis was performed in order to determine the relevant factors reflecting hydrometeorological processes. Results show consistent model performances and precipitation moments demonstrate considerable explanatory power relating to flood stage threshold exceedance as well as lag time characterization.