Dataset used for training and testing in “Physics-informed Graph Neural Networks for Operational Flood Modeling” paper

Abstract

This dataset contains 2D hydrodynamic model simulation outputs and associated geometry files used for training and testing graph neural network (GNN) models presented in the paper “Physics-informed Graph Neural Networks for Operational Flood Modeling.” The paper has been accepted to the AI4Tech track of the IJCAI-ECAI 2026 conference, which will be held in Bremen from 15–21 August 2026. The preprint is available on arXiv (https://arxiv.org/abs/2512.23964v1), and the codebase can be accessed through GitHub (https://github.com/acostacos/dual_flood_gnn). Please refer to the README file for additional details regarding dataset structure and usage. Flood simulations were generated using the HEC-RAS hydrodynamic modelling software developed by the US Army Corps of Engineers (https://www.hec.usace.army.mil/software/hec-ras/). The Digital Elevation Model (DEM) for the Wollombi catchment was obtained from the ELVIS – Elevation and Depth – Foundation Spatial Data portal, accessible at ELVIS Portal (https://elevation.fsdf.org.au/). Synthetic forcing data used in the simulations were adapted from the paper “Interpretable physics-informed graph neural networks for flood forecasting” available at Wiley Online Library (https://onlinelibrary.wiley.com/doi/full/10.1111/mice.13484).