Flood Prediction with Physics-Informed Neural Networks: Challenges and Future Directions

Physics-Informed Neural Networks (PINNs) bridge the gap between slow, physics-based models and fast, less reliable artificial intelligence (AI) approaches for flood modelling. This paper reviews their landscape, finding that PINNs effectively create fast surrogate models, infer river characteristics, and reconstruct flow fields from sparse data. A key challenge, however, is the field's heavy reliance on simulated data for validation. To become trustworthy for flood forecasting, future research must prioritize validation with real-world observations and establish robust methods for quantifying prediction uncertainty. This paper was presented at the 21st Computing and Control in the Water Industry Conference (CCWI 2025) at the University of Sheffield (1st - 3rd September 2025).