Unmanned aerial vehicle laser scanning (UAVLS) systems present a relatively new means of remote sensing and are increasingly applied in the field of forest ecology and management. However, one of the most essential parameters in forest inventory, tree diameter at breast height (DBH), cannot be directly extracted from aerial point cloud data due to the limitations of scanning angle and canopy obstruction. Therefore, in this study DBH-UAVLS point cloud estimation models were established using a generalized nonlinear mixed-effects (NLME) model. The experiments were conducted using Larix olgensis as the subject species, and a total of 8364 correctly delineated trees from UAVLS data within 118 plots across 11 sites were used for DBH modeling. Both tree- and plot-level metrics were obtained using light detection and ranging (LiDAR) and were used as the models’ independent predictors. The results indicated that the addition of site-level random effects significantly improved the model fitting. Compared with nonparametric modeling approaches (random forest and k-nearest neighbors) and uni- or multivariable weighted nonlinear least square regression through leave-one-site-out cross-validation, the NLME model with local calibration achieved the lowest root mean square error (RMSE) values (1.94 cm) and the most stable prediction across different sites. Using the site in a random-effects model improved the transferability of LiDAR-based DBH estimation. The best linear unbiased predictor (BLUP), used to conduct local model calibration, led to an improvement in the models’ performance as the number of field measurements increased. The research provides a baseline for unmanned aerial vehicle (UAV) small-scale forest inventories and might be a reasonable alternative for operational forestry.

Over an extended period, remote-sensing-based individual tree analysis has played a critical role in modern forest inventory and management research. The segmentation of individual trees from aerial point clouds usually depends on the characteristics of peak-like uplift on the crown surface; however, the performance inevitably decreases with increasing visibility of such features in point clouds, especially for high-density forests. Herein, we developed a novel hierarchical region-merging algorithm that first over-segmented the entire forest scene based on local density and then merged the over-segmented partitions into pairs through a stepwise optimal process to produce the final segmentation. In the region-merging method, a global merging cost was introduced to shift from local detection of crown features to use the overall compactness of forest point clouds. The experiments were conducted using unmanned aerial vehicle light detection and ranging (UAV-LiDAR) point clouds from three coniferous stands with different densities and a high-density coniferous and broad-leaved mixed stand. A total of 5510 field-measured trees in 36 plots were used to assess the accuracy of the proposed method. Our method achieved F-scores of 0.91, 0.88, 0.84, and 0.80 for low- (~700 stems/ha), medium- (~1000 stems/ha), and high-density (~2000 stems/ha) conifer stands and coniferous and broad-leaved mixed forests (~1800 stems/ha), respectively. Compared with the classical individual tree segmentation methods (marker-controlled watershed segmentation and point cloud region-growing algorithm), our method obtained comparable performance in low-density conifer stands and superior performance in the other stands. Furthermore, the region-merging algorithm could detect 10% more suppressed trees on average, which led to an apparent improvement in detection accuracy. The proposed algorithm provides a flexible segmentation framework that could be further improved by a different design that merges costs or applies multiscale segmentation with different stopping criteria.

In recent years, airborne Light Detection and Ranging (LiDAR) that provided three-dimensional forest information has been widely applied in forest inventory and has shown great potential in automatic individual tree crown delineation (ITCD). Usually, ITCD algorithms include treetop detection and crown boundary delineation procedures. In this study, we proposed a novel method called region-based hierarchical cross-section analysis (RHCSA), which combined the two procedures together based on a canopy height model (CHM) derived from airborne LiDAR data for ITCD. This method considers the CHM as a three-dimensional topological surface, simulates stereoscopic scanning from top to bottom using an iterative process, and utilizes the individual crown and vertical structure of crowns to progressively detect individual treetops and delineate crown boundaries. The proposed method was tested in natural forest stands with high canopy densities in Liangshui National Nature Reserve and Maoershan Forest Farm, Heilongjiang Province, China. Its performance was evaluated by an accuracy procedure that considered both the relative position of treetops and overlapped area of crowns. The average overall accuracy achieved was 85.12% for coniferous plots, 83.86% for deciduous plots and 86.44% for coniferous and broad-leaved mixed forest plots. The results revealed that the RHCSA method can detect and delineate individual tree crowns with little influence from forest types and crown size. It could provide technical support for individual tree crown delineation in coniferous, deciduous and mixed forests with high canopy densities.

As a core content of forest management, the height to crown base (HCB) model can provide a theoretical basis for the study of forest growth and yield. In this study, 8364 trees of Larix olgensis within 118 sample plots from 11 sites were measured to establish a two-level nonlinear mixed effect (NLME) HCB model. All predictors were derived from an unmanned aerial vehicle light detection and ranging (UAV-LiDAR) laser scanning system, which is reliable for extensive forest measurement. The effects of the different individual trees, stand factors, and their combinations on the HCB were analyzed, and the leave-one-site-out cross-validation was utilized for model validation. The results showed that the NLME model significantly improved the prediction accuracy compared to the base model, with a mean absolute error and relative mean absolute error of 0.89% and 9.71%, respectively. In addition, both site-level and plot-level sampling strategies were simulated for NLME model calibration. According to different prediction scale and accuracy requirements, selecting 15 trees randomly per site or selecting the three largest trees and three medium-size trees per plot was considered the most favorable option, especially when both investigations cost and the model’s accuracy are primarily considered. The newly established HCB model will provide valuable tools to effectively utilize the UAV-LiDAR data for facilitating decision making in larch plantations management.

Unmanned aerial vehicle (UAV) laser scanning, as an emerging form of near-ground light detection and ranging (LiDAR) remote sensing technology, is widely used for crown structure extraction due to its flexibility, convenience, and high point density. Herein, we evaluated the feasibility of using a low-cost UAV-LiDAR system to extract the fine-scale crown profile of Larix olgensis. Specifically, individual trees were isolated from LiDAR point clouds and then stratified from the point clouds of segmented individual tree crowns at 0.5 m intervals to obtain the width percentiles of each layer as profile points. Four equations (the parabola, Mitscherlich, power, and modified beta equations) were then applied to model the profiles of the entire and upper crown. The results showed that a region-based hierarchical cross-section analysis algorithm can successfully delineate 77.4% of the field-measured trees in high-density (>2400 trees/ha) forest stands. The crown profile generated with the 95th width percentile was adequate when compared with the predicted value of the existing field-based crown profile model (the Pearson correlation coefficient (ρ) was 0.864, root mean square error (RMSE) = 0.3354 m). The modified beta equation yielded slightly better results than the other equations for crown profile fitting and explained 85.9% of the variability in the crown radius for the entire crown and 87.8% of this variability for the upper crown. Compared with the cone and 3D convex hull volumes, the crown volumes predicted by our profile models had significantly smaller errors. The results revealed that the crown profile can be well described by using UAV-LiDAR, providing a novel way to obtain crown profile information without destructive sampling and showing the potential of the use of UAV-LiDAR in future forestry investigations and monitoring.

Accurate tree species information is a prerequisite for forest resource management. Combining light detection and ranging (LiDAR) and image data is one main method of tree species classification. Traditional machinelearningmethods rely on expert knowledge to calculatea large number of feature parameters.Deep learning technology can directly use the original image and pointclouddata to classify tree species. However, data with different patterns require the use of different types of deeplearningmethods. In this study, a multimodal deeplearningframework (TSCMDL) that fuses 2D and 3D features was constructed and then used to combine data from multiple sources for tree species classification. This framework uses an improved version of the PointMLP model as its backbone network and uses ResNet50 and PointMLP networks to extract the image features and pointcloudfeatures, respectively. The proposed framework was tested using UAV LiDAR data and RGB orthophotos. The results showed that the accuracy of the tree species classification using the TSCMDL framework was 98.52%, which was 4.02% higher than that based on pointcloudfeatures only. In addition, when the same hyperparameters were used for training the model, the efficiency of the model training was not significantly lower than for models based on pointcloudfeatures only. The proposed multimodal deeplearningframework extracts features directly from the original data and integrates them effectively, thus avoiding manual feature screening and achieving more accurate classification. The feature extraction network used in the TSCMDL framework can be replaced by other suitable frameworks and has strong application potential.