Incorporating Dynamic Drainage Supervision into Deep Learning for Accurate Real-Time Flood Simulation in Urban Areas

Abstract

Urban flooding has become a prevalent issue in cities worldwide. Urban flood dynamics differ significantly from those in natural watersheds, primarily because of the intricate drainage systems and the high spatial heterogeneity of urban surfaces, which pose considerable challenges for accurate and rapid flood simulation. In this study, an urban drainage-supervised flood model (UDFM) for urban flood simulation is proposed. The urban flood process is decoupled into drainage routing and surface flood inundation. On the basis of physical and deep learning drainage models, a hybrid module combining deep learning and dimensionality reduction algorithm is adopted to convert the 1D drainage overflow process into a high-resolution, spatiotemporal 2D pluvial flooding process. Compared with existing state-of-the-art surrogate models for rapid flood simulation, the UDFM more comprehensively and accurately represents the role of drainage systems in urban flood dynamics, providing high-resolution predictions of flood depth and velocity. When applied to a highly urbanized district in Shenzhen, UDFM-deep learning demonstrated real-time predictive capabilities and high accuracy, particularly in simulating flow velocity, with average Nash efficiency coefficients improved by 0.112 and 0.251 compared with those of a response surface model (RSM) and a low-fidelity model (LFM), respectively. These findings underscore the critical importance of drainage system overflow in urban surface flood simulations. The UDFM enhances accuracy, flexibility, interpretability, and extensibility without requiring additional physical model construction. This research introduces a novel hierarchical surrogate model structure for urban flood simulation, offering valuable insights for rapid flood warning and risk management in urban environments.