Integration of Near-Proximal and Proximal Lidar Sensing for Fine-Resolution Forest Inventory

Near-proximal and proximal light detection and ranging (lidar) systems are increasingly used for high-resolution forest inventory. Near-proximal lidar systems, such as those onboard uncrewed aerial vehicles (UAVs), offer high absolute positional accuracy due to continuous global navigation satellite system (GNSS) signal accessibility. Proximal lidar systems, such as backpack-based mobile mapping, excel at capturing detailed under-canopy information, including forest floor, tree trunks, and debris. However, each system has limitations when used alone. UAV lidar encounters challenges in under-canopy data collection due to occlusions and longer sensor-to-object distance, while proximal lidar faces GNSS signal outages and incomplete top canopy scanning in dense forest areas. This study proposes an approach to integrate near-proximal (UAV) and proximal (backpack) lidar data for fine-resolution forest inventory. Specifically, a framework is presented for integrating both data sources to improve the trajectory of the backpack system and establish an inventory pipeline for individual tree detection/localization, height, and diameter at breast height (DBH) estimation. The experimental results show great potential in generating high-quality, georeferenced point clouds across northern red oak plantation and coniferous forest scenarios. For northern red oak plantation inventory, the proposed pipeline achieved a 100% F1 score in tree detection and stem mapping using integrated lidar data, with root mean square errors (RMSE s) of 3 cm and 2 m for DBH and tree height estimation, respectively, when compared to field measurements. For the mixed coniferous forest inventory, due to the lack of field reference, the integrated UAV and backpack data set have been used to manually count the trees and establish their heights. The pipeline results are compared with those derived manually. The results show that UAV data have the lowest tree-detection accuracy, with an 81.06% F1 score, while proximal and integrated data excel in stem mapping, with a 91.92% F1 score. For the estimated height results, the near-proximal and integrated data have an RMSE value of almost 1.8 m, while the backpack has an RMSE value of 2.25 m. These results demonstrate the advantage of near-proximal/proximal data integration for best evaluation of tree detection, localization, and DBH/height estimation.