Automated Delineation of Individual Tree Crowns from Lidar Data by Multi-Scale Analysis and Segmentation

A canopy height model (CHM) derived from lidar data can be segmented to obtain individual tree crowns. However, branches, tree crowns, and tree clusters usually have similar shapes and overlapping sizes. This causes current individual tree crown delineation methods for CHMS to work less effectively on closed canopy deciduous or mixed wood forests consisting of various-sized tree crowns. Based on mult-scale analysis and segmentation, an innovative tree crown delineation method was developed in this study. In this method, the scale levels of target tree crowns are first morphologically determined; the CHM is filtered at the multiple scale levels; and local maxima within each filtered CHM are taken as markers to segment the original CHM using the marker-controlled watershed method. After tree crown segments are selected from the multiple resulting segmentation maps and integrated together, a complete tree crown map is generated. In an experiment on natural forests in Ontario, Canada, the proposed method yielded crown maps having a good consistency with manual and visual interpretation. For instance, when compared to a manually delineated forest map, the automated method correctly delineated about 69 percent, 65 percent, and 73 percent of the tree crowns from plots of closed canopy coniferous, deciduous, and mixed wood forests, respectively.