Tradeoffs and limitations in determining tree characteristics using 3D pointclouds from terrestrial laser scanning: A comparison of reconstruction algorithms on European bech (Fagus sylvatica L.) trees

A Case Study with European Beech (Fagus sylvatica)

Abstract

Terrestrial laser scanning (TLS) has quickly gained momentum in forestry as a fast and nondestructive alternative to determine tree shape and volume. Determining tree shape and volume is fundamental for a wide range of forestry applications, including the estimation of carbon stock and development of volume and biomass allometric models. However, tree shape and volume are often determined from TLS data based on different available algorithms, with direct implications on the measured tree feature. In this study we compared several algorithms for tree reconstruction from TLS data, with respect to their capacity to accurately determine tree characteristics such as diameter at breast height (DBH), tree height (H), stem volume (Vst) and total aboveground tree volume (Vtot). The following algorithms were compared using Bland-Altman limits of agreement (LoA): (i) TreeQSM, (ii) 3D Forest, (iii) RANSAC and (iv) Poisson. The data used for the comparison was collected from 10 sample plots, totalizing 40 European beech (Fagus sylvatica L.) trees, covering a DBH range from 6.2 cm to 76.0 cm and H range from 9.5 m to 36.2 m. The results showed that the algorithm used to analyse the TLS data affected notably the tree characteristics. The LoA were up to 3.65 m􀀀 for Vtot, up to 7.5 cm for DBH and up to 1.1 m for H, suggesting a rather weak agreement between algorithms. From our comparison, TreeQSM emerged as the most reliable algorithm for comprehensive trait reconstruction, while Poisson was well suited for stem volume estimation. Moreover, determining H seems to be less affected by the algorithm selection compared to DBH. Our findings raise awareness about algorithm selection in TLS data processing and highlight the importance of selecting an appropriate algorithm to meet the specific objective when using TLS to determine tree shape and volume.