Author(s): Jared Ortiz-Angulo; Paula Camus; Beatriz Perez-Diaz; Laura Cagigal; Sonia Castanedo; Fernando Mendez

Linked Author(s):

Keywords: Machine learning; Metamodel; Compound flooding; Flooding map prediction

Abstract: This work presents a hybrid statistical–dynamical approach for developing an operational Early Warning System for estuarine flooding, addressing the complexity of compound coastal–fluvial interactions driven by tides, surge, waves, wind, and river discharge. Instead of relying on fully dynamical hydrodynamic simulations—which are too computationally expensive for real-time forecasting—the method characterizes historical flood events, parametrizes their key drivers, and builds a hydrodynamic metamodel trained on a catalogue of pre-computed simulations. The metamodel can be used to convert forecasted atmospheric conditions into high-resolution flood maps within seconds, preserving physical consistency while reducing computational costs, thus supporting real-time decision-making and enhanced estuarine risk management. The methodology is applied and validated in Santoña Estuary, northern Spain, demonstrating high skill in replicating flood extents from historical events.

DOI:

Year: 2026