A new segmentation method for rotational landslide detection using outlier detection

Abstract

A new segmentation method for detection of rotational landslides from two epochs of Light Detection and Ranging System data (LiDAR) was developed. The developed method is based on outlier detection using Z scores of standard normal distribution. In this method, outlying height differences were obtained from two epochs of LiDAR data. The height differences were determined for each ith point uniquely by gridding the area along the steepest slope direction. Then, each profile containing the outlying height differences was tested for conformity to a possible rotational landslide and grouped into horizontal sets. Finally, the procedure was applied to horizontally grouped sets in the direction perpendicular to the steepest slope to determine each rotational landslide. To test the developed method, the area containing five different sizes of rotational landslides was scanned by the developed method. The results showed that the developed method segmented the area successfully identifying each rotational landslide.