A hydrodynamic-machine learning coupled (HMC) model of real-time urban flood in a seasonal river basin using mechanism-assisted temporal cross-correlation (MTC) for space decoupling

The intricate nature of urban waterlogging models arises from the compounding effects of human activities and dynamic alterations in the natural environment. Based on extensive data correlation and hydrodynamic process analysis, this study offers an a priori index mechanism-assisted temporal cross-correlation (MTC) for model space decoupling, which helps to reduce the computational complexity of the urban flood model. Furthermore, a hydrodynamic-machine learning (ML) coupled (HMC) model is proposed for predicting the river and drainage pipe water levels in urban seasonal river basins. The observations of the rainfalls, water levels of the river and manholes during 10 rainstorm events are gathered, with 8 of which serving as training datasets and 2 as prediction datasets. The simulation results show that MTC can provide a thorough evaluation of the impact of input parameters on predicting objects. It can also demonstrate the causality of hydrological processes behind monitoring data correlation. In addition, in comparison to ML and hydrodynamic models, the simulation stability of the HMC model is observed to be superior. Meanwhile, the NSE of the river channel and manhole is observed to be higher than 0.95 and 0.90, respectively, and the simulation uncertainty is found to be reduced by 42.2 %. The HMC model requires an average of 25 s to simulate an 8-hour rainfall event. This approach is found to be effective for space decoupling and rapid urban flood simulation.