AI-driven rapid simulation and forecasting techniques for flooding processes and their practical application

To address the high demand for timely flood forecasting, this paper introduces AI technology in conjunction with physics-based hydrodynamic models. By training on typical heavy rainfall and flood processes, a rapid AI-based method for flood prediction is developed. Initially, a hydrodynamic numerical model for studying the flood processes in the research area is established. Subsequently, this model is used to simulate and compute flood processes under different rainfall scenarios, forming a database of outcomes. Different AI learning methods are then employed to the machine learning for the correlation between key rainfall features and flood processes, validating the reliability of this learning method. This leads to the creation of a rapid simulation and forecasting machine learning model specific to the studied area's flood processes. Finally, inputting forecasted rainfall values allows the application of the machine learning model for fast flood prediction. This paper delineates methodologies and showcases applications for two flood types as urban inundation and watershed flooding.Results demonstrate that the developed AI model can achieve over 300 ~ 400 times acceleration compared to physics-based models with a similar level of accuracy.