Dynamic forecasting of urban waterlogging process based on real-time calibration of the SWMM model

ABSTRACT The graphical abstract presents a three-stage workflow for dynamic urban waterlogging forecasting using real-time SWMM model calibration. Step 1: Multi-source Data Fusion integrates rainfall forecasts, urban drainage monitoring data, and crowdsourced waterlogging information from social media platforms such as Weibo.Step 2: Real-time Dynamic Forecasting Framework runs the SWMM model, judges forecast errors (triggering calibration when the water depth error exceeds 5 cm), identifies sensitive parameters, performs real-time calibration, and updates forecasts iteratively. Step 3: The results show two key improvements: (1) Accuracy improvement: average Nash-Sutcliffe Efficiency (NSE) increased by 0.29, with all nodes reaching Class B accuracy; (2) Timeliness superiority: the timeliness coefficient (TC) is ? 0.9 (Class A), with each node’s calibration completed in 40 seconds to 1 minute. The framework balances accuracy and timeliness for real-time urban waterlogging early warning. Under the climate change paradigm, urban waterlogging triggered by extreme rainfall is intensifying. Precise, dynamic waterlogging forecasting is vital for urban management and residents to cope with disasters. Current studies often use near-real-time rainfall forecasts to drive hydro-hydraulic models for short-term urban waterlogging prediction, with static inputs and parameters, termed static forecasting. Absent a correction mechanism, static forecasts' accuracy often falls short for effective waterlogging control. This study takes the SWMM model as an example, introduces a real-time calibration mechanism for model parameters, and proposes a method for dynamic forecasting of urban waterlogging processes. A case application was carried out in Zhengzhou, China. The results show that compared with static forecasting, dynamic forecasting can improve the forecasting accuracy in stages during the process, with the average Nash–Sutcliffe Efficiency (NSE) coefficient increased by 0.29, and all nodes achieving a forecasting level above Class B. This method balances accuracy and timeliness, and can be effectively applied to urban waterlogging emergency management.