On April 29, the AI4FLOOD project consortium, co-funded by the European Interreg POCTEFA programme, held a new technical meeting at the Syndicat Mixte de l’Usine de la Nive (France). The session focused on Action 4: Design and development of artificial intelligence tools for early warning systems, and highlighted significant progress that reinforces the project’s position as a technological benchmark in flood risk management.

Key findings from the French territory

During the session, the Communauté d’Agglomération Pays Basque (CAPB) presented the results of a survey sent to 60 municipalities in the French Basque Country. The results revealed a high level of local awareness, but also concrete needs such as updating emergency plans and improving communication with residents.

In addition, the installation of rain gauges in three strategic areas  was announced, aiming to improve monitoring and flood forecasting in the Nive and Nivelle river basins.

Progress in Navarre

On the Spanish side, Tesicnor reported the identification of a critical risk point at a level crossing in Tafalla, where the installation of flood detection sensors is currently being evaluated.

Furthermore, our PhD researcher Iñaki Pérez del Notario (TESICNOR) presented advances in artificial intelligence focused on:

• Nowcasting models (very short-term forecasting)

• Quantitative Precipitation Estimation (QPE)

• Semi-supervised classification of rainfall types using radar imagery

These developments are a key step toward the project’s goal of accurately anticipating severe storms and flash floods.

Flow prediction models and ultra-fast simulations

Next, the NAIR Center shared results from its LSTM models applied to river flow forecasting. These models have shown strong potential for generating early warnings with sufficient lead time for decision-making.

In parallel, CEREMA and SIXENSE presented innovations in hydraulic simulation, including:

• The use of the Lattice-Boltzmann method on GPU, enabling 48-hour simulations in just 9 minutes

• The development of AI-based reduced models that cut simulation times from 30 hours to just 3 seconds

Next steps: integration and cross-border cooperation

Finally, the consortium agreed to continue the progressive integration of these developments into real operational environments. The focus will be on automating data flows and deploying real-time predictive tools. In addition, methodological harmonisation between territories will be strengthened to improve cross-border emergency coordination.

All these topics will be explored in more detail at the next biannual project meeting, scheduled for June 2025.

More information on the project website AI4FLOOD