| Title |
Utilizing News and Social Media Data for Performance Evaluation of Disaster Early Warning Systems: A Case Study on Flash Flood Prediction |
| Authors |
송윤섭(Song, Yunseop);황석환(Hwang, Seokhwan);윤정수(Yoon, Jungsoo);강나래(Kang, Narae);나우영(Na, Wooyoung) |
| DOI |
https://doi.org/10.12652/Ksce.2025.45.5.0577 |
| Keywords |
돌발홍수;예측 성능 평가;비정형 데이터;뉴스 미디어;SNS Flash flood;Predictive performance evaluation;Unstructured data;New media;Social network services |
| Abstract |
Conventional approaches to evaluating Early Warning Systems (EWS) have primarily relied on real-time hydrological data?such as precipitation, water level, and runoff?to quantitatively assess the accuracy and lead time of alerts. However, the recent proliferation and real-time dissemination of unstructured data from news media and Social Network Services (SNS) have opened new avenues for performance evaluation. This study proposes a novel methodology that utilizes unstructured media data to assess the predictive performance of a flash flood early warning system and examines the validity of this approach. A database was constructed by collecting nationwide media reports on flood and inundation events that occurred in South Korea from 2020 to 2024. Posts from platforms with high user engagement and accessible search functions?among major domestic news outlets and SNS?were treated as proxy observations for actual flood occurrences. Based on this media-derived event record, the study evaluated the spatiotemporal accuracy of the EWS predictions. Although some variations in evaluation outcomes were observed depending on the spatial and temporal granularity of the media data, the results generally indicated that the early warning system effectively captured real flood and inundation events. Furthermore, comparative analyses showed that the predictive performance was not significantly different between evaluations based solely on news articles and those incorporating SNS data. This suggests that media-based validation?regardless of data type?can serve as a viable tool for assessing the reliability of flood early warning systems. |