Author(s): Fernando R. L. Contreras; Sergio Martinez-Aranda; Javier Fernandez-Pato; Pilar Garcia-Navarro
Linked Author(s): Pilar García-Navarro
Keywords: Flood modeling; Neural networks; Shallow water equations; Richards equation; Surface-subsurface coupling; Infiltration
Abstract: This work presents a coupled model for flood simulation that integrates two-dimensional shallow water equations for surface flow with a subsurface component that uses a mass balance equation driven by infiltration and percolation fluxes estimated by Artificial Neural Networks (ANNs. The ANNs have been trained on extensive numerical solutions of the Richards equation, enabling accurate and efficient prediction of vertical water movement without solving the full partial differential equation. The coupling occurs through the infiltration flux at the land surface interface, ensuring mass conservation between surface and subsurface domains. The model excludes root water uptake to focus specifically on flood-related processes. Preliminary results demonstrate the model's capability to simulate complex flood events, including infiltration-driven runoff reduction and surface ponding dynamics, while maintaining computational efficiency suitable for large-scale applications.
Year: 2026