Classification of Landslide Features Using a LiDAR DEM and Back-Propagation Neural Network

Abstract

The purpose of this research was to detect landslide features using a light detection and ranging (LiDAR)-based digital elevation model (DEM) and back-propagation neural network (BPNN). The study area is in north-east of Taiwan. A high-resolution LiDAR-based DEM was used. Six training and four testing data sets were selected and manually digitized on landslide features were used as ground truth data. The relationship between landslide features and six trigger factors (slope angle, area solar radiation, roughness, profile curvature, plan curvature, and topographic wetness index) was computed from the LiDAR-based DEM. The experimental results indicated that the overall accuracy and kappa accuracy of the classification of landslide features using the BPNN algorithm were 0.950 and 0.772, respectively.