Development of an Integrated Urban Flood Information System Using Weather Radar

Climate change–induced increases in precipitation and extreme weather events are causing severe urban flooding problems worldwide. In particular, in South Korea, urban flooding has been occurring frequently due to extreme meteorological phenomena such as localized heavy rainfall. To minimize damage caused by urban flooding, it is essential to provide flood risk information that can be practically utilized at disaster sites.

Therefore, this study developed an integrated urban flood information system using radar-based meteorological data. The system consists of five stages: data collection, rainfall prediction, flood prediction, flood risk assessment, and disaster response support. In the data collection stage, real-time meteorological data, CCTV imagery, and urban infrastructure data are gathered. For rainfall prediction, point-based observations and radar data are first applied to a Random Forest–based regression model to perform quantitative precipitation estimation (QPE) bias correction. The generated QPE is then used as input to a deep learning model, NowcastNet, to produce quantitative precipitation forecasts (QPF) at 10-minute intervals for up to 3 hours ahead. For flood prediction, the predicted QPF is applied to a hybrid U-ConvLSTM model that combines U-Net and ConvLSTM architectures to estimate urban flood occurrence patterns and inundation depth. Flood risk is assessed by integrating QPF and flood prediction results with urban flood impact factors, including pedestrian areas, transportation facilities, agricultural and livestock facilities, industrial facilities, infrastructure, and public amenities, to calculate spatially distributed flood risk. Based on the estimated risk, flood risk levels (Attention, Caution, Warning, and Severe) are determined for each administrative district. For disaster response support, a Large Language Model (LLM) is employed to generate response protocols corresponding to each flood risk level. By providing AI-generated response guidelines to users, the system enables rapid and effective disaster management.

All collected and generated information is delivered to users through tables, graphs, and visual outputs within the system interface. The developed system will be further validated through real-world application testing in a Living Lab environment to identify potential issues and enable continuous improvement. Through ongoing advancement, the system is expected to significantly contribute to enhancing urban resilience against extreme weather events and strengthening disaster response capabilities.

 

Acknowledgements

This work was supported by the Technology Innovation Program (RS202400398858, Development of AI-based urban flood damage risk prediction and evaluation technology for practical use) funded By the Ministry of the Interior and Safety (MOIS, Korea)