DETECTION OF FLOOD AREA CAUSED BY TYPHOON HAGIBIS USING LEARNING-BASED METHOD WITH SENTINEL-1 DATA

Typhoon Hagibis hit Koriyama City in Fukushima Prefecture, Japan, on October 13, 2019. The outflow of the rivers, such as Abukuma, Sasahara, Yata, and Ouse, caused flood damage to built-up areas in the city center. In addition, the flood caused damage to rice production because flooding in rice paddies happened just before harvest time. Our research objective is to clarify the Sentinel-1 backscattering change in inundated built-up areas and rice paddy fields in Koriyama City caused by Typhoon Hagibis. Backscattering analysis was applied to Sentinel-1 data acquired before, during, and after flooding. The objective is to find the difference in the backscattering coefficient in non-flood and flooded areas. The backscattering coefficient increases in the built-up area during flooding, which was significant in a residential area. On the other hand, the backscattering coefficient decreases in rice paddy fields. We also applied a learning-based method to detect flooding in built-up areas and rice paddy fields by using the results of this study. We evaluated the flood detection accuracy using backscattering change with the support vector machine (SVM) as a classifier. Furthermore, adding texture information improved the detection accuracy by about 0.15 for the Kappa coefficient.