Deep learning based Extreme Rainfall and flood warnIngs through Seamless foreCasting

The extreme precipitation event of July 2021 and the ensuing floods caused 41 deaths and over 2 billion euro in damages in Belgium alone. In order to improve preparedness and to reduce the societal and economic impacts of such extreme precipitation events, we propose to leverage deep learning (DL) techniques to create a calibrated seamless prediction system, designed to drive both early warning systems and detailed near-real time impact models. This will allow stakeholders such as local authorities, emergency services and industry to quantify and mitigate the risk of both riverine and urban flooding, and to make better-informed, timely decisions.

Early warning systems require forecasts for weeks ahead, as provided by global numerical weather prediction (NWP) models. However, such models generally fail to capture precipitation extremes, in part due to their limited spatial resolution. Kilometric-scale NWP models, integrated in seamless observation-driven short-term prediction systems such as RMI's Project IMA, better represent these extremes, making them more suitable for highresolution (urban) flood models. However, their time horizon is limited to one or a few days, making them unsuitable for early warnings and proper management of extreme events.

This project therefore aims to meet the need for a consistent and calibrated forecasting system by seamlessly combining models at different time horizons. We will correct systematic model and representativity errors by downscaling and calibrating the proposed seamless forecasting system, using cutting-edge DL methodologies such as Generative Adversarial Networks (GANs) and residual learning. The DL-based multimodal QPE (Quantitative Precipitation Estimate) developed by RMI and VUB will be used for training and nowcasting (Moraux et al., 2021). Using state-of-the-art hydrological models of KU Leuven, an early warning system will be set up and validated. To allow for effective risk and uncertainty assessment, the resulting forecasting system will be fully probabilistic.