融合无人机与地面激光雷达点云的杉木单木参数提取：林龄依赖的精度制约与机理解释

Individual tree parameter extraction of Chinese fir via fusion of UAV and terrestrial LiDAR point clouds: age-dependent accuracy constraints and mechanistic explanation

摘要

【目的】森林结构随林龄的动态演变会导致激光雷达（LiDAR）参数提取精度产生异质性。当前单一机载（ALS）或地基（TLS）激光雷达在获取林分垂直三维信息时均存在物理视角局限，而多源点云融合虽能提升精度，但其增益效果如何受林分生长过程的系统性制约，内在物理机制尚不明确。为此，本研究以不同林龄阶段的杉木人工林为对象，系统评估ALS与TLS融合点云对单木参数（树高、胸径）提取精度的林龄制约规律，并定量解析其底层物理驱动机制，旨在为构建适用于全生长周期的高精度、自适应森林监测体系提供理论与方法支撑。【方法】在9块不同林龄（幼、中、成熟林）的杉木标准地，同步获取ALS、TLS点云及人工实测单木数据。采用无标靶策略对多源点云进行高精度配准与去噪、归一化等预处理后，利用比较最短路径算法对融合点云进行单木分割及树高、胸径提取。通过与实测数据及单一平台结果对比，采用均方根误差（RMSE）、决定系数（ R²）等指标综合评价提取精度。在此基础上，进一步利用广义加性模型定量检验点云核心物理特征与参数提取误差之间的非线性关联，以解析误差演变的底层物理机制。【结果】（1）融合点云显著提升了单木几何表征精度，其单木分割 F₁分数（≥0.961）与高密度TLS相当，远优于单一ALS（≥0.726）。（2）树高提取方面，融合点云在各林龄组均表现最优，相较于单一ALS和TLS，其 R²分别平均提升了0.11和0.10；RMSE较TLS降低了0.197 m，有效修正了地基扫描易低估树高的系统性偏差。（3）胸径提取精度的提升具有显著的林龄依赖性：在幼、中龄林中，多源融合能有效校正底层的几何误差（RMSE分别降至0.69和1.73 cm）；但在高郁闭度的成熟林中，因林下有效信号缺失，融合对胸径的精度补偿效应基本消失。（4）误差机理分析表明，树高提取残差与点云最大反射强度在中、成熟林中呈极显著的非线性相关（ p ≤ 0.001），证实高郁闭度引发的光束散射与物理退化是导致系统性测量偏差的主要原因；而胸径提取偏差则未表现出与宏观结构特征的显著相关性，趋于随机分布。【结论】融合ALS与TLS激光雷达能有效整合冠层与林下探测优势，显著提升杉木单木参数提取的整体精度。然而，这种精度提升受林分结构演变的严格制约，不区分郁闭度差异的简单点云融合在面对茂密林分时，会因激光穿透受限而难以发挥数据的协同优势。林分发育驱动的物理结构演变构成了参数提取系统误差的底层机制，未来研究需探索“林龄/密度自适应”的动态融合策略，并引入基于点云强度的信号衰减补偿模型，以实现在复杂森林生境下的精准反演。

Abstract

[Objective] The dynamic evolution of forest structure with stand age leads to heterogeneity in the accuracy of LiDAR parameter extraction. Currently, both airborne (ALS) and ground-based (TLS) LiDAR systems face physical viewing angle limitations when acquiring vertical 3D information of forest stands. While multi-source point cloud fusion can improve accuracy, the extent to which this enhances accuracy is systematically constrained by the forest stand growth process, and the underlying physical mechanisms remain unclear. Therefore, this study investigated Chinese fir plantations at different age stages to systematically evaluate the age-dependent patterns of accuracy in extracting individual tree parameters (tree height and diameter at breast height) from fused ALS and TLS point clouds, and to quantitatively analyze the underlying physical driving mechanisms. The aim is to provide theoretical and methodological support for the development of a high-precision, adaptive forest monitoring system applicable to the entire growth cycle. [Method] In nine standard plots of Chinese fir of different ages (young, middle-aged, and mature), ALS and TLS point clouds as well as manually measured individual tree data were collected simultaneously. After performing high-precision registration, denoising, and normalization of the multi-source point clouds using a target-free strategy, the fused point cloud was segmented into individual trees, and tree height and diameter at breast height (DBH) were extracted using the comparative shortest path algorithm. Extraction accuracy was comprehensively evaluated using metrics such as root mean square error (RMSE) and coefficient of determination (R ²) through comparison with field measurements and results from a single-platform approach. Building on this, a generalized additive model was further employed to quantitatively examine the nonlinear relationship between the core physical characteristics of the point clouds and extraction errors, thereby elucidating the underlying physical mechanisms driving error evolution. [Result] (1) The fused point cloud significantly improved the accuracy of individual tree geometric characterization; its F₁ score for individual tree segmentation (≥0.961) was comparable to that of high-density TLS and far superior to that of ALS alone (≥0.726). (2) Regarding tree height estimation, the fused point cloud performed best across all age classes. Compared to standalone ALS and TLS, its R ² values increased by an average of 0.11 and 0.10, respectively, while the RMSE was reduced by 0.197 m compared to TLS, effectively correcting the systematic bias of ground-based scanning that tends to underestimate tree height. (3) The improvement in DBH extraction accuracy exhibits significant age-dependence: in young and middle-aged forests, multi-source fusion effectively corrects underlying geometric errors (RMSE reduced to 0.69 cm and 1.73 cm, respectively); however, in mature forests with high canopy closure, the accuracy-compensating effect of fusion largely disappears due to the lack of effective signals beneath the canopy. (4) Analysis of error mechanisms indicates that tree height extraction residuals exhibit a highly significant nonlinear correlation (p ≤ 0.001) with maximum point cloud reflectance in middle-aged and mature forests, confirming that beam scattering and physical degradation caused by high canopy closure are the primary sources of systematic measurement bias; in contrast, diameter at breast height (DBH) extraction errors do not show a significant correlation with macrostructural features and tend to follow a random distribution. [Conclusion] The integration of airborne and terrestrial LiDAR effectively combines the advantages of canopy and understory detection, significantly improving the overall accuracy of individual tree parameter extraction for Chinese fir. However, this improvement in accuracy is strictly constrained by stand structure evolution; simple point cloud fusion that does not distinguish between differences in canopy closure struggles to leverage the synergistic advantages of the data in dense stands due to limited laser penetration. The physical structural evolution driven by stand development constitutes the underlying mechanism of systematic errors in parameter extraction. Future research should explore “age/density-adaptive” dynamic fusion strategies and introduce signal attenuation compensation models based on point cloud intensity to achieve accurate inversion in complex forest habitats.

关键词

无人机激光雷达 / 地基激光雷达 / 点云融合 / 杉木人工林 / 林龄效应 / 广义加性模型 / 精度评估

Key words

UAV-LiDAR / terrestrial laser scanning / point cloud fusion / Chinese fir plantation / stand age effect / generalized additive model / accuracy assessment

参考文献

[1] Liu B, Liu Q, Zhu C, et al. Seed rain and soil seed bank in Chinese fir plantations and an adjacent natural forest in southern China: implications for the regeneration of native species[J].Ecology and Evolution, 2022, 12: e8539.

[2] 吕梓晴, 段爱国. 不同产区杉木生物量与碳储量模型[J].林业科学, 2024, 60(2): 1-11.

[3] Lü Z Q, Duan A G. Biomass and carbon storage model of Cunninghamia lanceolata in different production areas[J].Scientia Silvae Sinicae, 2024, 60(2): 1-11.

[4] Spadavecchia C, Belcore E, Piras M, et al. An automatic individual tree 3D change detection method for allometric parameters estimation in mixed uneven-aged forest stands from ALS data[J].Remote Sensing, 2022, 14(18): 4666.

[5] Zhang H, Ye Z, Qin X, et al. Improving regional forest biomass estimation with a UAV-driven hierarchical framework: a case study in Chinese fir plantations[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2026, 19: 5758-5773.

[6] 丁志文, 邢艳秋, 尹伯卿, 等. 融合无人机和地基激光雷达点云数据估测单木结构参数[J].森林工程, 2024, 40(1): 142-151.

[7] Ding Z W, Xing Y Q, Yin B Q, et al. Fusion of UAV and TLS LiDAR point cloud data for estimating individual tree structure parameters[J].Forest Engineering, 2024, 40(1): 142-151.

[8] Panagiotidis D, Abdollahnejad A, Slavík M. 3D point cloud fusion from UAV and TLS to assess temperate managed forest structures[J].International Journal of Applied Earth Observation and Geoinformation, 2022, 112: 102917.

[9] 朱俊峰, 刘清旺, 崔希民, 等. 地基与无人机激光雷达结合提取单木参数[J].农业工程学报, 2022, 38(14): 51-58.

[10] Zhu J F, Liu Q W, Cui X M, et al. Extraction of individual tree parameters by combining terrestrial and UAV LiDAR[J].Transactions of the Chinese Society of Agricultural Engineering, 2022, 38(14): 51-58.

[11] 吴树萍, 曹小玉, 闫文德, 等. 不同龄组杉木林空间结构和土壤养分特征及其关联度[J].中南林业科技大学学报, 2025, 45(7): 115-122,173.

[12] Wu S P, Cao X Y, Yan W D, et al. Spatial structure and soil nutrient characteristics of Cunninghamia lanceolata forest of different age groups and their correlations[J].Journal of Central South University of Forestry & Technology, 2025, 45(7): 115-122,173.

[13] Zhang X, Wang H, Chhin S, et al. Effects of competition, age and climate on tree slenderness of Chinese fir plantations in southern China[J].Forest Ecology and Management, 2020, 458: 117815.

[14] Feng T, Chen S, Feng Z, et al. Effects of canopy and multi-epoch observations on single-point positioning errors of a GNSS in coniferous and broadleaved forests[J].Remote Sensing, 2021, 13(12): 2325.

[15] 林承达, 谢良毅, 韩晶, 等. 基于激光点云的农田玉米种植株数数目识别[J].激光技术, 2022, 46(2): 220.

[16] Lin C D, Xie L Y, Han J, et al. Recognition of the number of corn plants in farmland based on laser point cloud[J].Laser Technology, 2022, 46(2): 220.

[17] Besl P J, McKay N D. Method for registration of 3-D shapes[J].Sensor Fusion IV: Control Paradigms and Data Structures, 1992, 1611: 586-606.

[18] Chen J, Innes J L. The implications of new forest tenure reforms and forestry property markets for sustainable forest management and forest certification in China[J].Journal of Environmental Management, 2013, 129: 206-215.

[19] Ye Z, Guo Q, Wei J, et al. Recognition of terminal buds of densely-planted Chinese fir seedlings using improved YOLOv5 by integrating attention mechanism[J].Frontiers in Plant Science, 2022, 13: 991929.

[20] Li W, Guo Q, Jakubowski M K, et al. A new method for segmenting individual trees from the lidar point cloud[J].Photogrammetric Engineering & Remote Sensing, 2012, 78(1): 75-84.

[21] Tao S, Wu F, Guo Q, et al. Segmenting tree crowns from terrestrial and mobile LiDAR data by exploring ecological theories[J].ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 110: 66-76.

[22] Xiang W, Li L, Ouyang S, et al. Effects of stand age on tree biomass partitioning and allometric equations in Chinese fir (Cunninghamia lanceolata) plantations[J].European Journal of Forest Research, 2021, 140(2): 317-332.

[23] Eysn L, Hollaus M, Lindberg E, et al. A benchmark of lidar-based single tree detection methods using heterogeneous forest data from the Alpine space[J].Forests, 2015, 6(5): 1721-1747.

[24] Gupta S, Koch B, Weinacker H. Tree species detection using full waveform lidar data in a complex forest[C]//Wagner W, Székely B.Symposium: 100 Years ISPRS-Advancing Remote Sensing Science. Vienna: IAPRS, 2010: 249-254.

[25] Zhang H, Cai J, Cai M, et al. An upscaling framework for estimating regional-scale fractional cover of Chinese fir (FCCF) by integrating UAV and satellite images[J].International Journal of Digital Earth, 2025, 18(1): 2510570.

[26] 宋子戈, 肖海, 张泰, 等. 机载 LiDAR 森林蓄积量非参数模型构建与残差分析[J].中南林业科技大学学报, 2025, 45(10): 86-95,123.

[27] Song Z G, Xiao H, Zhang T, et al. Nonparametric model for forest stock volume estimation based on airborne LiDAR point cloud and residual analysis[J].Journal of Central South University of Forestry & Technology, 2025, 45(10): 86-95,123.

[28] Wang Y, Lehtomäki M, Liang X, et al. Is field-measured tree height as reliable as believed: a comparison study of tree height estimates from field measurement, airborne laser scanning and terrestrial laser scanning in a boreal forest[J].ISPRS Journal of Photogrammetry and Remote Sensing, 2019, 147: 132-145.

[29] Kukko A, Kaijaluoto R, Kaartinen H, et al. Graph SLAM correction for single scanner MLS forest data under boreal forest canopy[J].ISPRS Journal of Photogrammetry and Remote Sensing, 2017, 132: 199-209.

[30] 许珊珊, 李常春, 张超. 结合背负式激光雷达和无人机机载激光雷达的云南松单木胸径和树高提取[J].浙江农林大学学报, 2024, 41(5): 939-948.

[31] Xu S S, Li C C, Zhang C. Extraction of DBH and height of Pinus yunnanensis individual trees by combining backpack laser scanning and unmanned aerial vehicle laser scanning[J].Journal of Zhejiang A&F University, 2024, 41(5): 939-948.

[32] Paris C, Kelbe D, van Aardt J, et al. A novel automatic method for the fusion of ALS and TLS LiDAR data for robust assessment of tree crown structure[J].IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(7): 3679-3693.

[33] Stereńczak K, Mielcarek M, Wertz B, et al. Factors influencing the accuracy of ground-based tree-height measurements for major European tree species[J].Journal of Environmental Management, 2019, 231: 1284-1292.

[34] 曹林, 代劲松, 徐建新, 等. 基于机载小光斑 LiDAR 技术的亚热带森林参数信息优化提取[J].北京林业大学学报, 2014, 36(5): 13-21.

[35] Cao L, Dai J S, Xu J X, et al. Optimized extraction of forest parameters in subtropical forests based on airborne small footprint LiDAR technology[J].Journal of Beijing Forestry University, 2014, 36(5): 13-21.

[36] Yang X, Wang C, Xi X, et al. Assessment of multiple scattering in LiDAR canopy waveform[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2024, 17: 14394-14407.

[37] 王睿宇, 景会成, 包启龙. 基于特征提取和改进 ICP 的点云配准方法[J].应用光学, 2025, 46(4): 805-812.

[38] Wang R Y, Jing H C, Bao Q L. Point cloud registration algorithm based on feature extraction and improved ICP[J].Journal of Applied Optics, 2025, 46(4): 805-812.

[39] 尤号田, 邢艳秋, 冉慧, 等. 基于 LiDAR 点云能量信息的樟子松郁闭度反演方法[J].北京林业大学学报, 2014, 36(6): 30-35.

[40] You H T, Xing Y Q, Ran H, et al. Inversion method for the crown density of Mongolian scotch pine from point cloud data of small-footprint LiDAR[J].Journal of Beijing Forestry University, 2014, 36(6): 30-35.

[41] Lu W, Wan G, Zhou Y, et al. DeepVCP: An end-to-end deep neural network for point cloud registration[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Seoul: IEEE, 2019: 12-21.

[42] Dai W, Yang B, Liang X, et al. Automated fusion of forest airborne and terrestrial point clouds through canopy density analysis[J].ISPRS Journal of Photogrammetry and Remote Sensing, 2019, 156: 94-107.

基金资助

国家自然科学基金项目(31901298)

十四五国家重点研发计划(2023YFF130440502)

福建省自然科学基金项目(2021J01059)

福建农林大学学生科研创新训练项目(202510389184)