Study on the application of sentinel-2 optical imagery to inventory landslides using random forest classification model

Abstract

    In the study of landslides in Vietnam, the inventories of landslide has still been insufficient due to the difficulty in measuring and detecting location and time of landslide sites. With the development of the Earth - Observing Science and Computer Science, remote sensing technology is considered a solution to
this problem. This study utilised optical imagery Sentinel 2 for landslide detection, analysed by SNAP and QGIS software. The pre-event and post-event Sentinel 2 images acquired at the same study area were selected for the analysis. Location of landslide points is determined based on the change of NDVI index, using Random Forest (RF) classification model and overlay mapping technique. The validation results showed that this model has performed well with the accuracy and kappa values are 98.2% and 0.95 respectively. In addition, the test results at 2 actual landslide locations have shown the applicability of this method