Automatic Individual Shrub Delineation from Terrestrial Laser Scanning (TLS) Data

Abstract:

Species classification, gap analysis, biomass estimates, and biodiversity assessments in semiarid and arid regions can be improved by identifying individual shrubs using ground and airborne LiDAR (Light Detection and Ranging). In this study, we develop a new 3-dimensional shrub delineation method based on a moving window and region growing segmentation approach using ground based LiDAR (terrestrial laser scanning, TLS). The method leverages the variation and distribution of point densities that characterize a shrub and its boundary. The algorithm uses a combination of a voxel-based window (5-cm cubes) and segmentation. The segmentation approach begins with a seed point of the highest vegetation point from the point cloud, from the tallest to the shortest shrub. The automatic delineation results from this new method are compared to existing neighborhood point statistics and object-oriented segmentation methods, as well as manual delineation of individual shrubs. The results demonstrate that our new method closely approximates the number and boundaries of shrubs in two plots of different shrub densities, in comparison to manual delineation. These results provide an improvement over existing methods which over-delineate the branches and sub-crowns within shrubs. Our automatic delineation method will be used to improve shrub species classification and biomass estimates within TLS data, and potentially adapted for future use with airborne LiDAR.