AI4FLOOD: adaptation and improvement of municipal flood prevention through the integration of Artificial Intelligence and citizen participation

The main objective of the project is to develop early warning systems based on artificial intelligence, aimed at strengthening the capacity of municipalities to reduce disaster risk, particularly in the face of flooding. By analysing historical data from different river basins, AI4FLOOD generates predictions about the future behaviour of rivers with sufficient advance notice to enable local authorities to identify risks more accurately and implement preventive measures in a timely manner.

To achieve this objective, the project is developing a series of AI-based tools designed to improve both the collection of hydrometeorological data and the response capacity in emergency situations. These early warnings will be integrated into three new municipal emergency plans, which will be complemented by participation and training activities aimed at organisations, businesses and citizens, thus promoting a more robust and sustainable culture of prevention.

Within this framework, the following specific objectives are proposed:

• Conduct a comprehensive initial assessment in the participating municipalities, through the hydrological characterisation of the Nive, Nivelle, Urola and Cidacos river basins. This assessment will include a review of existing plans and the identification of weaknesses and opportunities for improvement in current risk management systems.
• Design and validate an early warning tool based on artificial intelligence techniques (deep learning) that merges historical data from the basin, hydrometeorological stations, weather radars and drone observations to improve flood risk prediction in the three regions covered by the project.
• Develop two-dimensional models of the study basins, capable of virtually simulating different flood events. These scenarios will be used to feed deep learning algorithms and thus improve the accuracy of predictive flood maps.
• Modernise existing crisis management platforms by integrating new artificial intelligence tools and develop a complete demonstration system covering everything from flood modelling to operational visualisation for decision-making.
• Co-design three new emergency municipal plans adapted to the specific characteristics of each territory and strengthen the local culture of prevention through training and citizen participation activities, thus ensuring the sustainability and transferability of the results.

NAIR participates in the project with a strategic role focused on the development and research of advanced Artificial Intelligence techniques to improve the generation of early warnings for flood risk scenarios. Its contribution has taken the form of Long Short-Term Memory (LSTM) neural networks, combined with recursive imputation and feature engineering techniques based on discrete wavelet transform, in order to optimise the operational prediction of short-term flows in transboundary basins in the Pyrenees and thus support flood risk management.

The work has been applied in three representative basins in Spain and France (Urola, Nivelle and Nive), where flood forecasting in mountainous and transboundary environments is critical for the protection of the population and infrastructure. These areas are characterised by rapid hydrological responses that call into question the effectiveness of conventional modelling approaches.

The proposed methodology integrates multiscale feature extraction using discrete wavelet transform with a recursive LSTM-based strategy for handling missing data, which is particularly relevant during extreme hydrometeorological events. The model implements a hierarchical funnel architecture, composed of three stacked LSTM layers with residual connections, capable of processing 48-hour historical sequences and generating flow predictions up to 12 hours in advance.

The AI4FLOOD project has been 65% co-financed by the European Union through the Interreg VI-A Spain-France-Andorra Programme (POCTEFA 2021-2027). The objective of POCTEFA is to strengthen the economic and social integration of the Spain-France-Andorra border area.