Estimating the period of probable landslide event using advanced D-InSAR technique for time-series deformation study of Kotrupi region

Abstract

Himalayan region is prone to landslides and every year many landslides occur. Some of these events are reported while, many of these landslides are not reported due to their remote location of occurrence, or nature of impact, and its date of occurrence is also unknown. This study tries to find a method to estimate the probable dates of landslide occurrences using time-series deformation analysis applying advanced DInSAR approach by adopting a short temporal baseline graph. We used 118 Sentinel 1-A SLC images to identify the nature, pattern, and trends of landslide deformation. Based on the time series cumulative deformation values, we found a ‘V’-shaped kink that matches with the landslide events. Such kinks were observed around 12 June 2015, 25 May 2016, 05 August 2016, 23 June 2017, 15 September 2019, 27 May 2018, 08 June 2018, and 15 July 2020. The majority of these kinks are visible during monsoon season which is expected. High-resolution Planet satellite data around these time ranges were used to verify the existence of new landslides. Moreover, the performance is evaluated using AUC-ROC, precision, recall; and F-Score. This approach has good prediction capabilities with the precision of 87%, AUC-ROC of 77% and F-score of 74%.