DestinE ML demonstrator – Machine learning for flood nowcasting
A machine learning (ML) Demonstrator will be developed to assist optimising power flow and ensuring resource adequacy across European electricity grids.
In a nutshell
- Seamless rainfall-to-flood prediction pipeline – Combines DestinE Weather-Induced Extremes Digital Twin data, national nowcasts, impact-based post-processing, hydrological models, and advanced machine learning to deliver faster, more accurate flood forecasts at short to medium lead times.
- Actionable insights for decision-makers – Supports national authorities, water managers, and civil protection with timely, risk-based flood information for early warning, evacuation planning, and proactive measures, reducing potential impacts of flooding.
- Scalable, user-driven approach – Co-developed with a Dutch community on extreme precipitation events and tested in the Netherlands. Fully integrated with the DestinE infrastructure.
Technical Overview
Challenge
Flooding is among the most frequent and destructive natural disasters, causing severe human, economic, and environmental losses. In 2024 alone, storms and floods in Europe affected over 400,000 people, claiming hundreds of lives and causing billions in damage. The challenge lies in providing timely, accurate, and actionable forecasts to enable early warnings, evacuation, and protective measures. Traditional forecasting methods are often produced at coarser resolution, while our approach delivers much finer-resolution predictions for more precise and locally relevant flood risk management.
DestinE Solution
This demonstrator harnesses the power of the DestinE Extremes DT to deliver faster, more accurate, and higher-resolution flood forecasts.
It combines the unique DestinE dataset with operational national nowcasts, impact-based precipitation post-processing, hydrological models, and advanced machine learning (ML) techniques to form a seamless rainfall-to-flood prediction pipeline. The first ML model blends DestinE forecasts with national meteorological data to produce highly detailed short- to medium-range precipitation predictions. These feed directly into a second ML model that predicts the extent and depth of flooding, enabling risk-based warnings and proactive flood management.
Co-developed with the a Dutch user community—a network of disaster risk authorities, water boards, and municipalities in the Netherlands—this approach ensures that the demonstrator meets real-world operational needs. Beyond the Netherlands, this service pilots in South Africa and New Zealand to demonstrate its adaptability to diverse hydrological and climatic conditions, paving the way for future international uptake.
Fully integrated into the DestinE infrastructure and accessible through GeoWeb – developed by KNMI and its partners-, the platform will provide actionable, user-friendly insights to civil protection and water managers, enhancing preparedness and decision-making in the face of increasingly frequent and severe flooding events.
Impact
By delivering faster, more accurate, and higher-resolution flood forecasts, this Machine Learning Demonstrator will significantly improve preparedness and response to extreme weather events. Civil protection authorities, water managers, and policymakers will gain access to actionable, location-specific insights, enabling timely evacuations and targeted protective measures. Beyond the Netherlands, the demonstrator’s scalability offers global potential, supporting communities worldwide in mitigating rainfall-induced flood risks and building resilience against the growing impacts of climate change and weather-induced extremes.
Contributions
Providers
