Abstract: The earthquake of Nepal in 2015 and its aftershocks caused many landslides with its enormous destruction posing huge potential threats to residential lives and properties in the affected regions. Rapid and accurate detection of post-earthquake landslide is in urgent demand. Traditional pixel-based change detection methods, however, delivered a large amount of over-recognized objects. In view of this, a principal component analysis (PCA) based change detection method was proposed to recognize post-earthquake landslides. Katmandu, the capital and largest city of Nepal, was selected as the study area. First of all, to remove noise and abundant information, an orthogonal transformation was applied to before-earthquake and post-earthquake Landsat-8 images of Katmandu respectively. In subsequence, converted set of features, as the first principal component (PC1), was used for change detection. Last but not the least, non-landslides were eliminated by NDVI, PC3 and slope feature from previous results. Validation of the detected results with high-resolution images from Google Earth shows that the proposed method is able to identify landslides with relatively high accuracy (93.0%). And it also proves the applicability of Landsat-8 satellite imagery for landslide mapping with its multispectral information. The post-earthquake landslides are generally found in areas of large surface slopes (between 20° and 50° ) of the Sun Koshi Valley, which is in the Northeast of the study area. The research findings suggest that the proposed method is effective in identifying post-earthquake landslides, thus assisting post-earthquake rescue and reconstruction.

Key words: landslide mapping, remote sensing, principal component transformation, change detection, feature selection, Nepal earthquake