DestinE by ECMWF

How can waterboards and municipalities be better prepared for flood?

Weather Impact has partnered with Hydrologic and the Royal Netherlands Meteorological Institute (KNMI) to develop a flood inundation forecasting system for the European Centre for Medium-Range Weather Forecasts (ECMWF). This system combines the latest advances in weather forecasting, seamless precipitation prediction and hydrological modeling, to improve flood preparedness of water boards and municipalities.

At the heart of the project is the Destination Earth Extremes Digital Twin (DestinE Extremes DT), a new weather model developed by ECMWF. This model can make higher resolution weather forecasts for Europe than ever done before, with a resolution higher than radars can measure. It is designed to capture extreme rainfall events with much greater accuracy than before.

AI-driven nowcasting

To make these forecasts useful for a wide range of users, Weather Impact improved upon pysteps, an open-source tool for rainfall nowcasting and blending. Normally, pysteps combines short-term radar observations with longer-term weather forecasts to create a seamless rainfall prediction. Our improvement is to connect pysteps with machine learning based nowcasts, especially the Deep Generative Model of Rainfall (DGMR). DGMR improves the representation of small-scale rainfall patterns, especially during fast-developing thunderstorms, making forecasts more accurate and reliable.
In addition, we apply machine learning optimization to improve how forecasts are blended. Instead of fixed rules, the system uses data-driven methods to dynamically determine the best mix of nowcasts and model outputs, adapting to changing weather situations.

Accurate, timely and impact-based precipitation and flood forecasts

The result is a seamless, high-quality precipitation forecast that feeds into Hydrologic’s flood inundation model, delivering more accurate, timely, and impact-based flood forecasts. These forecasts will be visualised in KNMI’s GeoWeb platform, making them accessible and user-friendly for decision makers. To ensure real-world applicability, the models and visualisations will be beta-tested by water boards, municipalities, and safety regions in the Netherlands, as well as by partners in South Africa and New Zealand. This international collaboration not only strengthens the system’s reliability under diverse climate and hydrological conditions, but also supports the protection of communities, infrastructure, and ecosystems in a time of increasing climate extremes.

By combining DestinE forecasts, AI-driven nowcasting (DGMR), and machine learning–enhanced blending, this project sets a new standard for seamless precipitation forecasts, helping to enhance flood risk prediction across Europe.