본문영역 바로가기 메인메뉴 바로가기 하단링크 바로가기

Reports

Development and Application of a Deep Learning-Based Real-Time Urban Flood Monitoring System
  • 조회수48
  • 등록일2025.07.25
  • Topic Climate Change/ Environment , Digital Transformation/ Technology
  • AuthorSung Eun Kim, Jongrak Beak

We are facing climate crises due to the accelerating impacts of climate change, which are increasing the frequency and intensity of natural disasters. Among these disasters, flooding caused by heavy rains and typhoons has reached a severe level. However, disaster prevention systems are lagging behind the rapid pace of climate change, highlighting a need for new flood response strategies. 

In response, the Seoul City has announced a plan to establish an 'AI-Based Integrated Flood Management System' by 2030, aimed at rapid flood response and securing the golden time for evacuation. This system will leverage Internet of Things (IoT) and artificial intelligence technologies to detect and predict flooding situations, communicating risks in real-time through messaging. In this context, the Seoul Institute (2023) has developed a technology that employs AI deep learning-based image processing, enabling the assessment of both the occurrence and depth of flooding, rather than relying solely on traditional sensor-based monitoring methods. In this technology, AI model was applied to detect vehicles (e.g., passenger cars, buses) in road flooding videos or images, classifying the degree of flooding into five levels based on the vehicles' tire specifications. And, It achieved approximately 96% accuracy in analyzing the flooding status of roads from CCTV footage and images in near real-time. 
This study developed a real-time urban flood monitoring system that can be utilized in Seoul's AI-based Integrated Flood Management System, through the enhancement and validation of the flood depth analysis technology developed by the Seoul Institute in 2023. To improve the generalization accuracy of the flood depth analysis model, additional training images were acquired through various methods, and the data preprocessing was enhanced to increase the diversity of the training images. Furthermore, the existing flood depth analysis model was upgraded from the YOLOv8 framework to the latest YOLOv10 framework, improving the analysis performance of the flood depth algorithm and significantly reducing video lag during real-time CCTV analysis. To make the advanced flood depth analysis technology applicable in monitoring operations, a flood depth monitoring platform was established. In collaboration with the Gangnam District Office, two locations, including the Daechi Intersection and the entrance to Seonjeongneung, which experienced significant flooding damage during the heavy rains of 2022, were selected for pilot operation. The pilot operation utilized the rainy season from July 9 to August 9, monitoring 24/7 to verify the accuracy of the flood depth analysis model and address any errors that arose during platform operation.

Based on the advanced flood depth analysis technology, a real-time urban flood monitoring system prototype was developed by integrating a model for visualizing flood depth monitoring results and issuing warnings, alongside an automated flood mapping model. This system was designed to link real-time flood depth monitoring results from CCTV with a two-dimensional flow analysis model, utilizing the urban disaster solution (KUDS) developed and serviced by the Korea Institute of Science and Technology Information (KISTI). The deep learning-based urban flood monitoring system is currently undergoing tests based on various scenarios to verify its effectiveness in real disaster situations. It is expected to be utilized as an AI-based Integrated Flood Management System of Seoul, facilitating rapid flood response and securing golden time for evacuation.