LAUSR

Abstract Over the past decade, sequential laser scanning has become increasingly routine for geomorphic monitoring in steep mountainous environments. Laser scans represent a permanent record of the terrain topology at a given point in time. Comparison of multitemporal datasets facilitates the identification of changes occurring in the terrain over time. Vegetation points in laser scans can inhibit the alignment of sequential scans. Therefore, vegetation points are commonly removed from the point cloud, although this can be a time-consuming process when done manually. In this study, an approach is presented and validated which makes use of multi-scale geometric operators to classify vegetation points in point clouds derived from terrestrial and aerial laser scanning. Using two study sites within the Thompson-Fraser rail corridor in British Columbia, Canada, the classification accuracies of vegetation points when compared to manual mapping was over 90%. The classified vegetation points are then segmented into individual trees. The approach is compared with outputs from the Find Individual Trees (FINT) program. The delineated trees from all approaches are incorporated into the RockyFor3D rockfall modelling program to show the implications of forest inventory construction from remote sensing methods on rockfall propagation. Over-classification of trees is shown to result in a higher predicted protective capacity of a forest when assessing a rockfall hazard.