Tree Height Measurements in Degraded Tropical Forests Based on UAV-LiDAR Data of Different Point Cloud Densities: A Case Study on Dacrydium pierrei in China

Tropical forest degradation is a major contributor to greenhouse gas emissions. Tree height can be used as an important predictor of forest growth, and yield models can provide basic data for forest degradation assessments. As an important parameter of unmanned aerial vehicle-light detection and ranging (UAV-LiDAR), it is not clear how the point cloud density affects the extraction accuracy of tree height in degraded tropical rain forests. To solve this problem, we collected UAV-LiDAR data at a flight altitude of 150 m, and then resampled the UAV-LiDAR data obtained according to the point cloud density percentage resampling method and obtained UAV-LiDAR data for five different point cloud densities, namely, 12, 17, 28, 64, and 108 points/m2. On the basis of the resampled LiDAR data, we generated a canopy height model (CHM) to extract the height of Dacrydium pierrei (D. pierrei). The results show that (1) With the increase in the point cloud density, the accuracy of tree height extraction gradually increased, with a maximum accuracy at 108 points/m2 (root mean squared error (RMSE)% = 22.78%, bias% = 14.86%). The accuracy (RMSE%) increased by 6.92% as the point cloud density increased from 12 points/m2 to 17 points/m2, but only increased by 0.99% as the point cloud density increased from 17 points/m2 to 108 points/m2, indicating that 17 points/m2 is a critical point for tree height extraction of D. pierrei. (2) Compared with the results from broad-leaved forests, the accuracy of D. pierrei height extraction from coniferous forest was higher. With the increase in point cloud density, the difference in the accuracy of D. pierrei height between two stands gradually increased. When the point cloud density was 108 points/m2, the differences in RMSE% and bas% were 3.55% and 6.22%, respectively. When the point cloud density was 12 points/m2, the differences in RMSE% and bias% were 2.71% and 4.69%, respectively. Our research identified the lowest LiDAR data point cloud density required to ensure a certain accuracy in tree height extraction, which will help scholars formulate UAV-LiDAR forest resource survey plans.