A convenient height-filtering-based strategy for multi-source forest LiDAR point cloud registration

Reconstructing the three-dimensional structure of forests in complex and heterogeneous environments is a persistent challenge in forest remote sensing, particularly when integrating multi-source LiDAR datasets with varying acquisition geometries. Traditional registration methods such as the Iterative Closest Point (ICP) algorithm have shown promise but often suffer from accuracy degradation in dense canopies and understorey clutter, limiting their applicability in operational forest monitoring. To address these limitations, this study systematically evaluates the robustness of classical ICP (point-to-point and point-to-plane) in forests with different stand densities and introduces a novel segmented registration strategy that integrates canopy height stratification.Three representative forest types were analyzed—Type I (700–1100 trees/ha), Type II (1100–1400 trees/ha), and Type III (1400–1800 trees/ha)—by jointly utilizing UAV Laser Scanning (ULS) and Backpack Laser Scanning (BLS) data. Registration performance was examined across five stratified scenarios (Type 0–0 m, 0–2 m, 0–4 m, 0–6 m, and 0–8 m), reflecting varying thresholds of aboveground points. Results demonstrate that while classical ICP consistently achieved sub-meter accuracy (RMSE < 0.50 m), the point-to-plane variant induced rigid-body displacement, leading to artifacts such as ground-negative values and stem noise. In contrast, the proposed segmented strategy, which incorporates height-based filtering of ULS data, effectively suppresses interference from understory vegetation and low-canopy structures, yielding substantial gains in registration stability and precision.Notably, Type II plots achieved the highest overall registration accuracy (average RMSE: 0.27 m), while Type I plots exhibited the most pronounced relative improvement, underscoring the method’s adaptability across stand densities. Importantly, this strategy ensures consistent canopy alignment while significantly mitigating stem-related distortions and ground-level errors.By integrating multi-source LiDAR fusion with stratified height-domain constraints, this work advances beyond conventional ICP implementations, providing a scalable framework for precise forest 3D reconstruction. The findings not only refine registration methodologies for heterogeneous forest environments but also lay the groundwork for enhanced forest inventory, biomass estimation, and ecosystem monitoring at scale.