Accurate flood extent mapping in suburban areas using a single SAR image: FFT-based artifact removal approach Article Swipe
YOU?
·
· 2025
· Open Access
·
· DOI: https://doi.org/10.1016/j.jag.2025.104941
· OA: W4415757595
Rapid and reliable flood inundation mapping is essential for disaster response. While multi-temporal SAR analysis is widely used, suitable pre-event imagery is often unavailable in real disaster scenarios. This study proposes a method for extracting inundation areas from single SAR data by removing strong scattering components from SLC images using FFT. The method was tested with ALOS-2 PALSAR-2 data from the 2018 Western Japan Heavy Rainfall in Mabi, Kurashiki City. Quantitative evaluation using SSIM confirmed that narrower Hann window sizes enhanced suppression of strong scatterers. Among three histogram-based binarization methods (Mean, Tsai, Otsu), the histogram mean achieved the highest Cohen’s Kappa (0.444). Accuracy assessment demonstrated that the proposed method significantly improved farmland inundation detection compared with conventional binarization, with Recall and F1 scores increasing by approximately 2.5%. In contrast, improvements for flooded buildings were limited due to inherently low baseline accuracy. Overall, the approach enables practical and reliable farmland inundation mapping using single-polarization, single-SAR imagery, even when no pre-event data are available; however, further validation across different conditions and frequency bands is needed.
