Coupling Artificial Neural Networks and high-resolution hydrodynamic models for an Early-Warning System for coastal flooding

Coastal regions are increasingly vulnerable to flooding due to extreme events and sea-level rise driven by climate change. Effective Early Warning Systems (EWSs) are essential for enhancing resilience and enabling proactive disaster management. Traditional EWSs often rely on physically-based numerical models to simulate flood scenarios. However, the high computational demand of high-resolution, downscaled models makes them incompatible with real-time predictions and limits their integration into EWSs. To address this limitation, we integrate numerical modeling with machine learning, specifically employing an Artificial Neural Network (ANN) trained on numerical model outputs to enable rapid and accurate flood predictions. Our approach allows the rapid transformation of offshore conditions forecasts into zone-scale predictions of coastal flood metrics. A wide range of wave and sea level scenarios were considered as inputs, in order to ensure a comprehensive training dataset. This range was based on historical wave data alongside future climate projections, capturing a comprehensive set of potential flooding events. The Maximum Dissimilarity Algorithm (MDA) was employed to select a smaller subset of conditions, in order to reduce computational time while ensuring the representativeness of the dataset. These selected offshore parameters were then used in hydrodynamic simulations conducted with a numerical chain, consisting in a nested SWAN+XBeach model. The model accounts for vegetation’s dissipative effects by incorporating habitat maps that define its characteristics and spatial distribution. Our results demonstrate that the trained ANN effectively predicts the zone-scale inundation metrics obtained from the hydrodynamic simulations. Its performance was evaluated against satellite-derived observations of total inundated area for a historical event. Building on this capability, the ANN was then embedded into an EWS framework for real-time forecasting and alerting. The system automatically classifies flood risk levels across coastal zones to easily and fastly disseminate timely warnings to authorities and stakeholders. This approach offers a robust, scalable, and easily replicable tool for proactive flood risk management and operational deployment under climate change scenarios.