Automatic detection of landslides in object-based environment using open source tools

Abstract

Automatic detection of landslides from very high resolution satellite images is a prerequisite for rapid damage assessment and supporting disaster management activities. In this study, a novel method using open source tools was developed for extracting landslides from bi-temporal satellite images based on Object Based Image Analysis (OBIA). The methodology employed involves image segmentation followed by elimination of non-landslide candidates using object based change detection techniques and lastly, unsupervised classification. Brightness of a post-landslide image is higher in comparison to its pre-landslide image hence a suitable threshold value for post image brightness was set to demarcate the landslide affected regions from the other land cover types. Further, landslide diagnostic parameters such as difference in Green Normalized Difference Vegetation Index (GNDVI), Digital Elevation Model (DEM, slope, Principal Component Analysis (PCA) and difference in Top of Atmosphere (ToA) values were used to eliminate challenging false candidates such as snow cover, barren land and river sediments. The objects retained after eliminating false candidates are then classified into two classes using k-means clustering algorithm. The local features associated with an image can be computed by finding the key points using a Speeded Up Robust Feature (SURF). Performance of this method was investigated using Resourcesat-2 LISS-IV multispectral (5m) bi-temporal satellite image covering parts of Uttarakhand state in India. Results show that the proposed methodology will aid rapid inventorisation of landslides.