Interpretable Physics-Informed Graph Neural Networks for Flood Forecasting

This paper introduces HydroGraphNet, a physics-informed graph neural network for flood forecasting that combines mass-conservation-aware training, autoregressive stability, and interpretable KAN-based feature transformations. On a real-world flood case study, the method delivers markedly better predictive accuracy than a baseline GNN while maintaining near-zero mass balance error and stronger detection of major flood events.

Recommended citation: Taghizadeh, M., Zandsalimi, Z., Nabian, M. A., Shafiee-Jood, M., & Alemazkoor, N. (2025). "Interpretable Physics-Informed Graph Neural Networks for Flood Forecasting." Computer-Aided Civil and Infrastructure Engineering.