Integrating Pre-Harvest UAV Scans to Enhance Harvester Tree Localization Accuracy

Abstract

Accurate geolocation of individual trees during forest harvesting operations is crucial for effective decision-making, yet traditional cut-to-length (CTL) harvesters often experience significant positional errors (0.5–10 m) due to unreliable GNSS performance under dense forest canopies. This uncertainty hampers the precise integration of harvester-generated data into operational forest management systems. To address this problem, we investigated the integration of high-resolution pre-harvest UAV LiDAR data with harvester-collected positional information. UAV laser scanning (DJI Matrice equipped with Zenmuse L2 LiDAR) was conducted over a dense, mixed-species boreal forest stand scheduled for its first thinning operation. Following harvesting, stump positions were precisely recorded using centimeter-grade GNSS as ground truth. Harvester-recorded tree positions were matched to tree crowns delineated from UAV LiDAR point clouds using Canopy Height Model (CHM) segmentation. For each crown, structural (height, crown size) and spectral (RGB statistics) features were extracted, and tree species (spruce, pine, birch) were classified using Random Forest (RF) and XGBoost models. Comparative positional error analysis revealed that mean harvester GNSS errors were 1.52 m, whereas UAV-derived tree positions showed significantly lower mean errors of 0.63 m. Integrating UAV data with harvester positions successfully reduced the mean positional error to 0.76 m. Species classification accuracy exceeded 91% overall for both RF and XGBoost models, with coniferous species (pine, spruce) classified at approximately 94% accuracy and deciduous birch slightly lower at around 71%. These results highlight the potential of integrating pre-harvest UAV scans to substantially enhance tree-level geolocation accuracy, enabling precise digital twins and improved real-time operational decision-making during harvesting. The study addresses a critical research gap by developing a practical workflow for combining UAV and harvester data, thereby facilitating precision forestry applications such as targeted tree selection, automated navigation, and enforcing environmental safeguards.