Exploring Unprecedented Flood Events Using Counterfactual and Stochastic Approaches

While in many regions worldwide climate change and socio-economic developments are increasing the likelihood of unprecedented extreme events, current risk management practices are often not prepared for such events, resulting in catastrophic impacts. This lack of preparedness is partly driven by the reluctance of both lay people and decision-makers to consider and plan for events that exceed those observed in the historical record. There is thus a need for approaches that generate plausible scenarios of unprecedented events that are both scientifically sound and intuitively understandable. Here we present several methods for constructing such scenarios for river flooding in Germany. These include spatial counterfactuals, in which the precipitation fields of historical floods are spatially shifted, and a perfect-storm approach, in which precipitation from historical events is combined with historical wet catchment conditions. In addition, we apply a stochastic simulation framework in which a large-scale weather generator drives a hydrological model. All three approaches produce events that are substantially more severe than those observed in Germany over the last 70 years (1951-2021). For example, even moderate deviations in the trajectory of the precipitation field of past floods, which were among the most expensive and catastrophic events in Germany, could have led to substantially higher severity across Germany. While all methods are able to provide unprecedented flood events, the choice of method depends on the intended application, such as stress-testing infrastructure or supporting risk communication.