Landslide detection using polarimetric ALOS-2/PALSAR-2 data: a case study of 2016 Kumamoto earthquake in Japan

Abstract

Landslide events occur annually induced by heavy rain or an earthquake in the world. Remote sensing technique is an effective for landslide mapping and monitoring. Synthetic aperture radar (SAR) has a great potential due to its all-weather day and night imaging capabilities. Therefore, the utilization of SAR data for rapid damage assessment is expected. In a previous study, we demonstrated that landslide detection derived from correlation coefficient using pre- and post-event COSMO-SkyMed HH single polarization data. On the other hand, fully-polarimetric SAR (PolSAR) data contain various information compared to single polarization SAR data. In this study, we demonstrated the applicability of polarimetric analysis from SAR images for detection of the landslide area. The 2016 Kumamoto earthquake in Japan caused landslide damage in Kumamoto prefecture, Japan. Rapid damage assessment after natural disasters is crucial to fast crisis response. Three ALOS-2/PALSAR-2 polarimetric data acquired on 3 December, 2015, 21 April, 2016 and 5 May, 2016 were used. Entropy/α angle/anisotropy were calculated from each PolSAR data. Yamaguchi four-component decomposition analysis of PolSAR was also conducted. The polarimetric coherence (γHH – V) was calculated from the correlation between HH and VV polarization from pre- and post-event PolSAR data. In this study, we deal with the detection of landslides using pre- and post-event polarimetric parameters from PolSAR data. The largest landslide area in Minami-Aso village was clearly showed the surface scattering because the landslide induced by earthquake removed forested vegetation on the ground surface. The extent of the landslides was detected using pre- and post-event PolSAR data with Random forest (RF) classifier. It is clarified that pre- and post-event alpha angle, entropy and γHH – V from PolSAR data with the RF classifier is effective for landslide detection.