Monday, 29 January 2024: 2:15 PM
318/319 (The Baltimore Convention Center)
Reliable urban flood forecasting is highly demanded in emergency response, risk management and urban planning. We propose a hybrid physical and statistical approach for urban flood forecasting to meet the needs of multi-agencies of response at various temporal and spatial scales. A 12-year (2010-2021) urban flood inventory over China’s mainland is carefully compiled based on the mainstream social media (i.e., Baidu and Sina Weibo) which likely represents our best knowledge of the county-level urban flood history of the study domain. A support vector machine (SVM) method is optimized in reproducing the occurrence of the recorded urban flood events before it is used for event-level forecasting. Furthermore, detailed street-level inundation is numerically simulated by the DRIVE-Urban model which couples river basin-urban hydrological and hydraulic modeling with explicit delineation of the mass-energy interactions between urban drainage system (e.g., pipes and dikes) and river networks (Chen et al., 2022, https://doi.org/10.1029/2021WR031709). Urban flood simulations at multi-scale (from river-basin to street) and multi-spatial resolution (1km-90m-10m/5m) are conducted within the real-time Global to Local (Glocal) Hydrometeorological Solution on Floods (GHS-F) system (Wu et al., 2020, https://doi.org/10.1007/s00376-020-0260-y). GHS-F shows promising capability in predicting both county-level and street-level urban flood. A mean probability of detection (POD) of 0.61 with a false alarm rate (FAR) of 0.29 and a critical success index (CSI) of 0.45 are derived for 685 regions at the county-level, while the DRIVE-Urban model provides a POD of ~70% and a CSI exceeding 0.5 for street-level inundation prediction at multiple cities where the model has been in operational use.

