Application of Artificial Intelligence in Rainfall Nowcasting and Flash Floods Forecast in an urban catchment: First results from Research Project KIWaSuS in Germany

Overview & Synopsis

The increase in the frequency and intensity of extreme weather events is already a reality due to global warming. In order to better inform the disaster management teams about the upcoming flooding situation in an urban catchment, the research project KIWaSuS (AI-based warning system for heavy rain and urban flash floods) is going to develop three models with artificial intelligence (AI) for precipitation nowcasting, forecasting of sewer overflow and forecasting of inundation areas. In comparison to a usual rainfall nowcasting model for a specific heavy rainfall event, the first developed AI precipitation prediction model can deliver better results for forecast horizons between 20 and 60 minutes. It can also maintain its performance constant up to a lead time of 35 minutes. In the first investigations for the flood prediction model, it was analysed whether the inclusion of other inputs additionally to precipitation information could improve the performance of the AI model. It is demonstrated that the addition of sewer overflow information improves the model performance, which corroborates the planned coupling of these three AI models to achieve the goals of the research project KIWaSuS.