Highly Accurate Tree Models Derived from Terrestrial Laser Scan Data: A Method Description

Abstract

This paper presents a method for fitting cylinders into a point cloud, derived from a terrestrial laser-scanned tree. Utilizing high scan quality data as the input, the resulting models describe the branching structure of the tree, capable of detecting branches with a diameter smaller than a centimeter. The cylinders are stored as a hierarchical tree-like data structure encapsulating parent-child neighbor relations and incorporating the tree's direction of growth. This structure enables the efficient extraction of tree components, such as the stem or a single branch. The method was validated both by applying a comparison of the resulting cylinder models with ground truth data and by an analysis between the input point clouds and the models. Tree models were accomplished representing more than 99% of the input point cloud, with an average distance from the cylinder model to the point cloud within sub-millimeter accuracy. After validation, the method was applied to build two allometric models based on 24 tree point clouds as an example of the application. Computation terminated successfully within less than 30 min. For the model predicting the total above ground volume, the coefficient of determination was 0.965, showing the high potential of terrestrial laser-scanning for forest inventories.

Does Tree Architectural Complexity Influence the Accuracy of Wood Volume Estimates of Single Young Trees by Terrestrial Laser Scanning?

“…When working with imperfect and unorganized raw point cloud data (without topological information on the trunk and branches), a filter algorithm is needed which prevents branch structures with low point densities from being removed. Although subjective and individual quality control in point cloud filtering is beneficial in preserving branch structures from being removed [17,28], processing hundreds of trees requires automated filtering processes. Skeletonization of raw point clouds and resultant topological tree models could preserve structural information and prevent low point densities from being removed or could even add structural information by closing gaps between disconnected branch segments [22].…”

“…For example, Hopkinson et al [10] estimated DBH and TH based on TLS data, but still relied on standard allometric equations for wood volume estimates. Moreover, several authors [14][15][16][17] have developed and deployed methods to estimate wood volume based solely on TLS data.…”

“…Alternatively, other approaches depend on geometric feature recognition within segments of the TLS points. Potentially representing a more natural model of the curved surface of the trunk and branches, cylinders are fitted within segments [21] and their volumes are summed for volume estimations [15,17,[22][23][24][25].…”

A high‐resolution approach for the spatiotemporal analysis of forest canopy space using terrestrial laser scanning data

“…Even geographical coordinates can be added to model specific light absorption. Furthermore, a differentiation of branch order levels can be performed using both voxel‐ and cylinder‐based tree modeling approaches (Gorte & Winterhalder, 2004; Hackenberg et al., 2014; Raumonen et al., 2013). The estimation of leaf area densities of trees scanned under leaf‐on conditions (in particular using full‐waveform scanners) can also be used to further describe and differentiate the voxel's potential degree of shadowing (Hosoi & Omasa, 2006).…”

“…Terrestrial laser scanning (TLS) has been established as a complementary approach for the time‐efficient and nondestructive measurement of three‐dimensional (3D) structural elements of trees (Liang et al., 2016). TLS has been applied to analyze standard tree and stand dendrometrics, such as stem diameter and tree height (Li, Hess, von Wehrden, Härdtle, & von Oheimb, 2014; Liang et al., 2016), and more recently to analyze 3D tree topology, canopy structure, and wood volume of branches (Bienert, Hess, Maas, & von Oheimb, 2014; Calders et al., 2015; Hackenberg, Morhart, Sheppard, Spiecker, & Disney, 2014; Hess, Bienert, Härdtle, & von Oheimb, 2015; Kunz et al., 2017; Raumonen et al., 2013). …”