With global warming and urbanization accelerating, urban flooding is becoming more severe. Real-time forecasting plays a key role in disaster mitigation, but traditional hydrodynamic models are too resource-intensive for timely prediction. Machine learning models offer high efficiency but often lack accuracy in simulating spatiotemporal flood dynamics. This study proposes a new data-driven model, which performs well in a flood-prone area of Macao.

With the rapid development of cities in China, urban flooding becomes a growing problem, which not only affect the living quality of the public, but also threatens urban safety. Recently, urban flooding is attached more attention. By analyzing and summarizing urban flooding data from recent decades, which includes frequency, scope of influence, and losses, the causes and status quo of urban flooding problems in China is generalized.

A 2D hydrodynamic model was established based on MIKE Flood, which could simulate the flooding submergence from a burst levee. The flood risks in the urban area were quantitatively evaluated for several scenarios including burst levee, proactive flood diversion and their combinations. The results could provide scientific supports for urban flood risk analysis, flood dispatching and management. The methodology has been applied in Qiqihar, which is one of the main cities in northeast China.