This study aimed to address the inefficiency of manual measurements in current forest resource surveys.

The LiDAR point cloud data of the natural forests of the Greater Khingan mountains were obtained from unmanned aerial vehicle laser scanning (ULS) and handheld laser scanning (HLS). After trajectory solution, filtering, and noise removal, coarse registration of ULS and HLS was performed using the 3D Harris algorithm for key point extraction, followed by fine registration through the Iterative Closest Point (ICP) algorithm. Based on ULS and fused point cloud data (U-HLS), individual trees were segmented using two approaches: the canopy height model (CHM)-based segmentation and a seed-based K-means hierarchical clustering method. Tree height and diameter at breast height (DBH) were extracted from the optimal segmentation results, and a two-variable additive biomass model was applied to estimate biomass and carbon storage at both individual-tree and plot levels, thereby verifying the applicability of multi-platform LiDAR data for refined forest monitoring.

1) Significant differences were observed in individual tree segmentation accuracy among different methods and data sources. The CHM-based method identified 5 790 individual trees from ULS data but exhibited a high false detection rate, with a recall of only 0.61 and an F-score of 0.43. In contrast, the seed-based K-means hierarchical clustering method, constrained by stem position, achieved notably higher completeness and accuracy, with an F-score of 0.87 and recall, precision, and F-score improved by 107.07%, 70.46%, and 87.49%, respectively; 2) Using U-HLS data and K-means hierarchical clustering method, the tree parameters achieved high accuracy with minimal systematic bias. Both DBH and tree height reached R²=0.93, with RMSE values of 1.6 cm and 1.15 m, respectively, and relative RMSEs below 16%. The inclusion of stem position constraints effectively mitigates errors caused by canopy occlusion, improving the reliability of individual-tree parameters extraction; 3) The estimation accuracy of carbon storage based on U–HLS data is high at both the individual-tree and plot levels, with R² values of 0.97. The carbon storage estimation at the individual tree level has small RMSE (7.14 kg) and bias (-0.68 kg), although underestimation and overestimation occur in large (26-36 cm) and small (<12 cm) DBH classes, respectively. At the plot scale, the R² of carbon storage estimation achieves 0.97, and the RMSE and bias were only 3.62 t C·hm^-2 and -1.27 t C·hm^-2, respectively. The rRMSE, in particular, was 6.22%, significantly smaller than that at the individual tree-level (23.32%). This suggests that at the plot level, individual tree estimation errors are further balanced, partially offsetting local overestimations and underestimations, resulting in high consistency between carbon storage estimates and observed values.

Multi-platform LiDAR data effectively enhances the accuracy of individual-tree parameter extraction and carbon storage estimation, maintaining stable and reliable performance under complex forest conditions. It can partially replace traditional field measurements at both individual-tree and plot scales, providing a valuable reference for efficient and refined forest resource surveys and carbon sink monitoring.