Accurate assessment of forest structure and biomass is hampered by extensive field measurements that are time-consuming, costly, and inefficient. This is especially true in mangrove forests that have developed complex above-ground root structures for stability and survival in the harsh, anaerobic, and reducing conditions of water-logged sediments. These diverse structures can differ even among similar species, providing complex three dimensional structures and making them difficult to accurately assess using traditional allometric methods. Terrestrial laser scanners (TLS) have been used widely in collecting forest inventory information in recent years, mainly due to their fine-scale, detailed spatial measurements and rapid sampling. In this work we detected stems and roots in TLS data from three mangrove forests on Pohnpei Island in Micronesia using 3D classification techniques. After removing noise from the point cloud, the training set was acquired by filtering the facets of the point cloud based on angular orientation. However, many mangrove trees contain above-ground roots, which can incorrectly be classified as stems. We consequently trained a supporting classifier on the roots to detect omitted root returns (i.e., those classified as stems). Consistency was assessed by comparing TLS results to concurrent field measurements made in the same plots. The accuracy and precision for TLS stem classification was 82% and 77%, respectively. The same values for TLS root detection were 76% and 68%. Finally, we simulated the stems using alpha shapes for volume estimation. The average consistency of the TLS volume assessment was 85%. This was obtained by comparing the plot-level mean stem volume (m³/ha) between field and TLS data. Additionally, field-measured diameter-at-breast-heights (DBH) were compared to the lidar-derived DBH using the reconstructed stems, resulting in 74% average accuracy and an RMSE of 7.52 cm. This approach can be used for automatic structural evaluation, and could contribute to more accurate biomass assessment of complex mangrove forest environments as part of forest inventories or carbon stock assessments.

对森林结构和生物量的准确评估受到费时，昂贵且效率低下的广泛实地测量的影响。在已开发出复杂的地上根结构以在苛刻，厌氧和减少浸水沉积物条件下稳定和生存的红树林中，尤其如此。这些不同的结构甚至在相似的物种之间也可能有所不同，从而提供了复杂的三维结构，并使它们难以使用传统的异速测量方法进行准确评估。近年来，陆地激光扫描仪（TLS）已被广泛用于收集森林清单信息，这主要是由于其精细，详细的空间测量和快速采样。在这项工作中，我们使用3D分类技术从密克罗尼西亚Pohnpei岛上的三片红树林中检测了TLS数据的茎和根。从点云中删除噪声后，通过基于角度方向对点云的各个面进行滤波来获取训练集。但是，许多红树林树木都具有地上的根，这些根可能被错误地归类为茎。因此，我们在词根上训练了一个支持分类器，以检测遗漏的词根（即归类为词干的那些）。通过将TLS结果与在同一图中进行的并发现场测量进行比较来评估一致性。TLS词干分类的准确性和准确性分别为82％和77％。TLS根检测的相同值分别为76％和68％。最后，我们使用alpha形状对茎进行了模拟以进行体积估计。TLS容量评估的平均一致性为85％。通过比较样地水平的平均茎体积（m³ / ha）之间的字段和TLS数据。此外，使用重建后的杆，将实地测得的胸高直径（DBH）与源自激光雷达的DBH进行了比较，得出的平均精度为74％，RMSE为7.52 cm。这种方法可以用于自动结构评估，并且可以作为森林清单或碳储量评估的一部分，对复杂的红树林环境进行更精确的生物量评估。