DestinE: real-time flood forecasting with AI

Introduction

Within the European initiative Destination Earth (DestinE), HydroLogic has developed an AI-driven flood forecast that supports water managers in swift and well-founded decision-making.

  • Duration: 2023–present
  • Client: ECMWF
  • Partners: KNMI and Weather Impact
  • Keywords: Destination Earth, digital twin, floods, flood forecasting, AI, climate adaptation
  • Models & Tools: machine learning models, DestinE platform, Extremes Digital Twin, Climate Digital Twin, GeoWeb, High Performance Computing

Background

Floods

Floods caused by extreme rainfall are among the most impactful natural disasters worldwide. Although rainfall forecasts are steadily improving, it remains difficult to determine quickly and reliably where and when water nuisance will occur. Traditional hydrodynamic models are often too slow for real-time applications, while water managers in particular need more lead time and clear actionable information. Within DestinE, new, faster methods for flood forecasting are therefore being explored.

Destination Earth (DestinE) is developing a digital twin of the Earth to better predict climate change and extreme events. The initiative is funded by the European Union and implemented by ECMWF, ESA and EUMETSAT. Within this European initiative, HydroLogic has worked on an AI-driven flood forecast that can be deployed operationally for any location in the world and supports water managers in swift and well-founded decision-making.

Method

AI models

HydroLogic heeft samen met het KNMI en Weather Impact binnen DestinE FloodMIND (Machine Learning Demonstrator Together with KNMI and Weather Impact, HydroLogic has developed two coupled machine learning models within DestinE FloodMIND (Machine Learning Demonstrator for rainfall-induced floods). The rainfall blending model combines short-term nowcasts and medium-term forecasts into improved rainfall forecasts. The flood forecasting model then translates this rainfall into high-resolution flood forecasts, considerably faster than traditional models.

The project uses the DestinE infrastructure, including the Extremes Digital Twin. Results are made accessible via interactive GeoWeb visualisations.

The result

Reliable information

The rainfall blending and flood forecasting models can generate real-time probabilistic flood forecasts with a high spatial resolution. Water managers, municipalities and safety regions gain insight into risks earlier, including uncertainties and scenarios. This supports timely warnings, evacuation decisions and measures to better protect infrastructure and limit damage.

What did the project deliver?

  • Real-time flood forecasts at high resolution
  • Longer forecast horizon than traditional models
  • Applicable to operational water management and crisis management
  • Successfully tested in the Netherlands, New Zealand and South Africa
“With DestinE, we show how AI, data and water expertise come together in practically applicable flood forecasts.”
Bram Schnitzler | project lead
More information