关键词: 机载激光雷达, LiDAR变量, 林分蓄积量, 断面积, 模型

Abstract: Objective Point cloud density is a critical factor affecting the cost of airborne LiDAR data acquisition and pre-processing. Therefore, exploring the influence of point cloud density on the estimation accuracy of forest inventory attributes can provide a reference for optimizing technical schemes for airborne LiDAR-based large-area forest inventory and monitoring. Method In this study, we used airborne LiDAR data and field plot data collected in a subtropical mountainous and hilly region in Guangxi, China. Firstly, the original point clouds with a density of 4.35 points·m⁻² were reduced to 4.0, 3.5, 3.0, 2.5, 2.0, 1.5, 1.0, 0.5, 0.2, and 0.1 points·m⁻² using a systematic thinning method, respectively, resulting in 11 plot-level point cloud datasets, including one full-density point cloud dataset and ten reduced-density point cloud datasets. Secondly, a paired sample t-test was used to analyze the differences in 12 LiDAR-derived metrics between reduced-density point clouds and full-density point clouds in four forest types (Chinese fir, pine, eucalyptus, and broad-leaved). Thirdly, using a multiplicative power model formulation with fixed variables and stable structure, the stand volume (VOL) and basal area (BA) were estimated using various density datasets of point clouds, respectively, and their goodness-of-fit statistics, including coefficient of determination (R²), relative root square error (rRMSE), and mean prediction error (MPE), were compared. Finally, a t-test was used to analyze the differences in the means of the estimates between the reduced-density point clouds and full-density point clouds. Result 1) When the point cloud density was low, the means of the 25^th, 50^th, and 75^th height percentiles (ph25, ph50, and ph75) of the reduced-density point clouds showed statistically significant differences from those of the corresponding variables of the full-density point clouds. However, when statistically significant differences were found for different variables in various forest types, the point cloud densities differed. There were no statistically significant differences in the means of mean point cloud height (H_mean) and coefficient of variation of point cloud height distribution (H_cv) between the reduced-density point clouds and full-density point clouds in all forest types, but there were statistically significant differences in the means of maximum height (H_max) of point clouds between the reduced-density point clouds and full-density point clouds for all forest types. 2) The means of canopy cover (CC) and 25^th density percentile (dh25) of the reduced-density point clouds were not statistically significantly different from those of the corresponding variables of the full-density point clouds for all forest types (except dh25 for broadleaf forests), but statistically significant differences existed for the 50^th and 75^th density percentiles (dh50 and dh75). 3) The means of the mean leaf area density (LAD_mean) of reduced-density point clouds were statistically significantly different from those of the LAD_mean of full-density point clouds in all forest types, and while the means of the coefficient of variation of leaf area density (LAD_cv) of reduced-density point clouds were significantly different from those of the LAD_cv of full-density point clouds when point cloud density was low. 4) The differences in the estimates of VOL and BA for different density point clouds were small among the forest types, and none of the estimates were statistically significantly different from each other. However, as the density of point clouds decreased, the R² of the estimation models for VOL and BA for fir, pine, and eucalyptus forests slowly decreased, and the rRMSE and MPE slowly increased, indicating that the estimation accuracy of forest inventory attributes gradually decreased. The R², rRMSE, and MPE of the estimation models for VOL and BA for the broad-leaved forests were not obviously affected by the change in point cloud density. Conclusion The decrease in the density of point clouds leads to an increase in the standard deviation of the LiDAR-derived metrics, which is the main reason for the decrease in the estimation accuracy of forest inventory attributes. In the operational forest resources investigation and monitoring, the airborne LiDAR point cloud density should be greater than 0.5 points·m⁻².

Key words: airborne LiDAR, LiDAR-derived metrics, stand volume, basal area, model