Development of algorithms to automatically identify and spatialize all the trees in complex one-hectare 3D point cloud stands.

Abstract

The development of the mobile LiDAR technology for mapping forest ecosystems is rapidly advancing and has great potential in many ways. However, to date, the use of this technology remains marginal both in conducting inventories and in research in forest ecology. This is probably due to the difficulty of translating LiDAR point clouds into usable information. While the identification and spatialization of trees is certainly one of the most important pieces of information to provide, the algorithms for this identification remain poorly performing, especially in complex structured stands. In this study, we developed a series of algorithms adapted to complex data clouds. We then used these algorithms in 10 stands (1 ha) showing a dense shrub layer. Our results show that nearly 95% of trees over 10 cm DBH are automatically identified by our algorithms; a fast visual inspection subsequently allows us to identify missed trees. Our algorithms also allow us to identify trees less than 10 cm in DBH, but some of these may have been missed by the LiDAR due to the obstruction created by the foliage. These algorithms, along with the extremely accurate algorithm developed for estimating DBH (Nolet et al, under revision), represent an important step in deploying the mobile LiDAR technology on a larger scale.