This is the authors' abstract. We don't add key points for this paper.
Accurate stem volume estimation is fundamental to forest resource inventory and carbon stock assessment. Traditional methods rely on destructive sampling, whereas terrestrial laser scanning (TLS) offers a non-destructive alternative. However, the accuracy of individual tree segmentation in structurally complex subtropical natural forests is constrained by crown overlap, species mixing, and vertical stratification. In this study, we developed a semantic voxel-based framework for individual tree segmentation and robust stem volume estimation from plot-level TLS point clouds. The method integrates 3D-CNN-based voxel semantic classification with bottom-up tree growth segmentation, followed by parameter extraction, taper equation fitting, and volume estimation using the sectional measurement method. Evaluation across 18 plots (1451 trees) in Guangxi, Southern China, demonstrated that the proposed method achieved an F-score of 0.881 for individual tree segmentation, significantly outperforming conventional CHM-based (0.533) and geometric voxel-based (0.794) approaches. Optimal taper equations were established for Chinese fir (Zeng Weisheng model, validation R2 = 0.943) and Eucalyptus (Yan Ruohai model, validation R2 = 0.987). TLS-based volume estimates yielded R2 values of 0.94–0.97 and RMSE of 0.022–0.037 m3 per tree, with negligible systematic bias. These findings demonstrate that the proposed semantic voxel framework enables accurate and non-destructive stem volume estimation in complex subtropical forests, providing a practical technological pathway for modernizing forest inventory practices.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: