IDENTIFICATION OF VEGETATED LANDSLIDES USING ONLY A LIDAR-BASED TERRAIN MODEL AND DERIVATIVES IN AN OBJECT-ORIENTED ENVIRONMENT

Light Detection and Ranging (LiDAR) and its derivative products have become a powerful tool in landslide research, particularly for landslide identification and landslide inventory mapping. In contrast to the many studies that use expert-based analysis of LiDAR derivatives to identify landslides only few studies, all pixel-based, have attempted to develop computer-aided methods for extracting landslides from LiDAR. It has not been tested whether object-oriented analysis (OOA) could be an alternative. Therefore, this study investigates the application of OOA using 2 m resolution slope gradient, roughness, curvature, and openness maps calculated from single pulse LiDAR data, without the support of any spectral information. More specifically, the focus is on the possible use of these derivatives for segmentation and classification of slow-moving landslides in densely vegetated areas, where spectral data do not facilitate accurate landslide identification. A semi-quantitative method based on support vector machines (SVM) was developed for a test area in the Flemish Ardennes (Belgium). The procedure was then applied without further modification to a validation area in the same region. The results show that OOA using LiDAR derivatives allows recognition and characterization of profound morphologic properties of deep-seated landslides on soil-covered hillslopes such as those in the Flemish Ardennes, because approximately 70 % of the landslides of an expert-based inventory were also included in the object-oriented inventory. For mountain areas with bedrock, o