Funding Proposal under the BMBF Announcement “Flexible, Resilient, and Efficient Machine-Learning Models” from August 31, 2023

Motivation: Heavy rainfall and flooding are among the most significant natural hazards, with severe consequences for people, the environment, and infrastructure. Small and medium-sized river basins, which are common in Central Europe, are particularly susceptible, as they respond rapidly to extreme weather conditions. This not only drastically shortens the warning time but also amplifies forecast uncertainties. Current hydrological models reach their limits because they fail to adequately capture the complexity and accuracy required for weather prediction and runoff generation. The research project “KI-HopE-De” addresses this gap and aims to make these models more efficient, robust, and flexible.

Objectives and Approach: Modern artificial intelligence (AI) methods will be explored, developed, and applied to enable a unified, nationwide prediction system for small catchment areas (< 500 km²) in Germany. The project aims to improve forecast accuracy, particularly for extreme events, in these areas. To achieve this, a comprehensive hydro-meteorological dataset will be created, incorporating both measurement and forecast data. This dataset will serve as the foundation for future model training and validation. The focus will be on short-term forecasts (< 48 hours), with several flood forecasting centers directly involved in the project.

Innovations and Perspectives: “KI-HopE-De” will make a significant contribution to public safety and flood protection. The project will deliver an innovative, prototype platform that could potentially be adopted by all flood forecasting centers nationwide. This would effectively reduce the current dependency on and technological advantage of global tech companies. By integrating AI into large-scale natural science simulation models, the project will advance novel use cases, data integration techniques, and adaptive learning methods.