Flash Flood Hazard: A Counterfactual Analysis for Germany

Abstract

In response to heavy rainfall, flash floods can arise from rapid runoff concentration in the landscape, presenting significant damage potential due to high flow velocities and minimal lead times. Flash floods are among the most destructive natural hazards. Managing their risks usually necessitates the application of extreme value statistics. However, the small temporal and spatial scale of flash floods poses a challenge, as the requisite data for statistical methods is often unavailable or incomplete.  Furthermore, the effects of climate change may compromise the robustness of extreme value statistics. To enhance our understanding of flash flood hazards in Germany, we present a novel "counterfactual" scenario analysis. This approach considers alternative ways of how events could have unfolded. To identify worst-case scenarios is particularly interesting for risk assessment. Accordingly, we assumed that historical rainfall events could have happened anywhere else in Germany: What would have happened if a particular rainfall event occurred in a different area? Would it result in a flash flood? To address these questions, we created a catalog of extreme rainfall events for the years 2001-2022 from radar rainfall estimates. Because flash flood triggering rainfall is often embedded in precipitation fields of larger spatio-temporal extent, we used the cross-scale weather extremity index (xWEI) to identify and rate the events. We then shifted the ten most extreme events systematically across Germany and modeled the peak discharge for every shifted realization (counterfactual peaks), thus creating close to a billion runoff datasets. This approach preserves the spatio-temporal event structure that significantly influences the overlapping scales of runoff processes and hence the hazard. Results are provided to users via an interactive web interface. Our results reveal that, on average, the worst case counterfactual peaks would exceed the maximum original peak by a factor 5.3. Furthermore, it shows that not every event is equally likely to trigger high runoff peaks, even when rated similarly extreme. Our study might help to expand the view on what could happen in case certain extreme events occurred elsewhere, help to identify flash flood prone areas, and thereby reduce the element of surprise in disaster risk management. The proposed method is transferable and could be a valuable asset, especially in data-scarce regions.