Detection of Flood Areas Using Sentinel-1 Satellite Imagery and Evaluation of Its Applicability

Abstract

The flood detection technique based on synthetic aperture radar (SAR) imagery was assessed. This technique was applied to a flood event in Busan in 2017. Sentinel-1 imagery was acquired, and then different multitemporal change detection methods were applied for SAR data analysis. Observed data were collected from newspapers, social media, and the Korean Fire Protection Association (KFPA). Specifically, flood and non-flood data were collected and used for analysis. As a result of receiver operating characteristic (ROC) curve analysis, the area under the curve (AUC) value was 0.67, which indicates a generally efficient technique, but it showed lower performance than the random model according to the threshold value. Four of 15 flood sites and a threshold value of 0.02 were selected to compare the flood area between the observed data and the flood maps derived from SAR analysis. As a result, all four flood maps included the observed flood areas, with some overestimation of flood area. A criterion for selecting the optimal threshold was suggested using ROC curve analysis, and the flood map results from this threshold were validated, but follow-up study is required to assess the accuracy numerically.