Extracting individual trees from lidar point clouds using treeseg

Abstract

Recent studies have demonstrated the potential of lidar-derived methods in plant ecology and forestry. One limitation to these methods is accessing the information content of point clouds, from which tree-scale metrics can be retrieved. This is currently undertaken through laborious and time-consuming manual segmentation of tree-level point clouds from larger-area point clouds, an effort that is impracticable across thousands of stems. Here, we present treeseg, an open-source software to automate this task. This method utilises generic point cloud processing techniques including Euclidean clustering, principal component analysis, region-based segmentation, shape fitting and connectivity testing. This data-driven approach uses few a priori assumptions of tree architecture, and transferability across lidar instruments is constrained only by data quality requirements. We demonstrate the treeseg algorithm here on data acquired from both a structurally simple open forest and a complex tropical forest. Across these data, we successfully automatically extract 96% and 70% of trees, respectively, with the remainder requiring some straightforward manual segmentation. treeseg allows ready and quick access to tree-scale information contained in lidar point clouds. treeseg should help contribute to more wide-scale uptake of lidar-derived methods to applications ranging from the estimation of carbon stocks through to descriptions of plant form and function.