Rapid simulation of flood routing using deep learning and hydrodynamic model

The rapid simulation and early warning of flood disasters are crucial for flood prevention and mitigation. However, the current simulation efficiency of urban flood models based on physical mechanisms remains low. In this study, a deep learning model based on convolutional neural network(CNN) is constructed by combining flood inundation data generated by hydrodynamic model and deep learning techniques to rapidly simulate urban flood routing. The results show that the developed CNN model can effectively simulate the flood inundation, with a peak water depth prediction error within 8%, and a good performance in simulating the inundation extent. The CNN model demonstrates a significantly higher efficiency in flood inundation simulation, achieving approximately 400 times faster computation while maintaining comparable accuracy to hydrodynamic models. This study can provide valuable insights for rapid simulation of urban flood inundation, early warning and forecasting of flood disasters, and the development of digital twin basins.