Beyond Historical Records: Using Counterfactual Scenarios to Improve Flood Risk Management

Floods caused by heavy precipitation events (HPEs) rank among the most damaging natural hazards. Under climate change, HPEs are projected to intensify in both spatial extent and rainfall magnitude. Yet extreme rainfall does not necessarily translate into extreme flooding because flood severity depends on the spatial coincidence of intense rainfall with catchments that have the hydrological properties to produce extreme floods. Such rare alignments may be poorly captured in historical observations, rendering conventional flood risk assessment, typically based on stream gauge records and extreme value analysis (EVA), inherently uncertain.

To address this uncertainty, counterfactual analysis - exploring alternative, hypothetical event scenarios - can help remove randomness in the spatial distribution of rainfall and reduce the element of surprise. Advances in precipitation monitoring, such as weather radar, together with increased computational capacity, now enable the systematic application of counterfactual approaches in flood risk management. This way the data basis can be artificially broadened. As a result, the method is gaining momentum in both the United States and Europe, supporting the development of more robust flood scenarios, also for ungauged catchments.

We introduce a framework to include counterfactual scenarios in conventional EVA for flood hazard assessments, with a particular focus on flash floods, and demonstrate that this approach substantially improves the anticipation of extreme floods. However, a central challenge lies in ensuring the physical plausibility of counterfactual scenarios. We therefore present and compare multiple methods for selecting counterfactual events and evaluate their influence on overall EVA-based hazard estimates. By identifying potential flood hotspots and reducing uncertainty, counterfactual thinking offers a valuable tool for disaster risk management, particularly in data-scarce regions.