Real-Time Flood Prediction Using Physics-Informed Neural Networks and Rainfall-Runoff Data

Complex and unpredictable, natural disasters, especially floods, are hard to predict and plan for. Risk management, policymaking, fatality reduction, and property damage mitigation require better flood prediction models. Flood prediction models are invaluable for handling extreme events and identifying hazards. The proposed model involves preprocessing, feature selection, and model training. Data is preprocessed by cleaning, segmenting, and converting before analysis. Multi-feature and principle component analysis (PCA) minimise dimensionality and increase model accuracy in feature selection. The Physics-Informed Neural Network (PINN) model predicted floods with 93.49% accuracy. This shows that the PINN-based model can predict floods, giving policymakers and emergency responders a significant tool. The model could reduce community harm and boost disaster readiness by improving flood predicting accuracy.