News

Helmholtz AI project call showcase: Identifying causal flood drivers

Floods are among the most damaging natural hazards, causing thousands of deaths and millions of euros in damages every year. Efficient monitoring and prediction tools are key in the mitigation of flood’s adverse effects, for both short-term forecasting and long-term projections. Developing these tools requires a deep understanding of floods’ causes and outcomes, which can be accomplished using artificial intelligence.

 

How can AI methods help in understanding the drivers behind floods and improve their risk management? Read in this week’s Helmholtz AI project showcase how researchers at the Helmholtz Centre for Environmental Research (UFZ) and the German Aerospace Center (DLR) are working on this challenge.

Could you introduce yourself, giving your affiliation, area of work, and of course, the project title?

My name is Jakob Zscheischler, I work for the Helmholtz Centre for Environmental Research (UFZ). I am coordinating the 2020 Helmholtz AI project “Identifying causal flood drivers” (CausalFlood) along with Jakob Runge from the Institute of Data Science of the German Aerospace Center (DLR). The project aims to combine the expertise on compound weather and climate events (UFZ) with novel causal inference techniques (DLR).

In simple words, what specifically is your project about? And, how and why do you think it is a high risk, high gain endeavour?

The goal of CausalFlood is to identify how and at which time scale variables such as precipitation, soil moisture and temperature combine to cause floods. 

Figure: Extreme events such as the July 2021 flood in Germany (left image) often are a result of compound extremes: The combination of two or more variables in an extreme state, for example soil dryness and precipitation, aggravates flood extremes (right image).

We will rely on observational data and use novel statistical tools from the field of causal inference. Causal inference consists in studying a succession of events that seem to predict others. It is a relatively new research field in AI, still being thoroughly tested and further developed. Runoff observations, which form the basis of identifying floods in our case, have many statistical properties that make the straightforward application of causal inference difficult. A successful identification of causal flood drivers via causal inference tools would substantially improve our understanding of these high-impact events and help create better flood forecasting tools.

How important has the Helmholtz AI funding and platform been to carry out this project?

While contacts between both groups have existed for some years, the Helmholtz AI funding helped to make this collaboration more concrete. In particular, it allowed the hiring of two PhD students, one at DLR and one at UFZ. Both work closely interlinked on complementary topics, namely developing the causal inference methodology (DLR) and applying it to runoff data (UFZ).