A hybrid approach for enhanced tree volume estimation of complex trees using terrestrial LiDAR

Accurate forest biomass estimation is crucial for ecological research and policymaking, yet traditional methods relying on allometric equations often struggle with structural variability in diverse forest ecosystems. Quantitative Structure Models (QSMs) using terrestrial LiDAR have improved tree volume estimation, but existing models face challenges in accurately representing complex tree geometries, particularly buttressed trees with non-cylindrical trunks. This study evaluates four widely used QSMs – TreeQSM, AdQSM, AdTree, and SimpleForest – using terrestrial LiDAR point clouds from 191 trees. Results indicate that TreeQSM achieves the lowest Mean Absolute Percentage Deviation (MAPD) of 18.16%, followed by AdQSM (25.23%), while AdTree (31.22%) and SimpleForest (56.86%) exhibit higher errors. The consistent residuals and an R2 value of 0.94 across a diverse dataset underscore the robustness of TreeQSM. Additionally, TreeQSM’s built-in noise filtering algorithm enhances its adaptability to noisy data, achieving a significant reduction in MAPD from 51.7% to 12.05% in datasets affected by noise. This study also investigates the challenges of modeling noisy data and the consequential loss of small branches by the noise filtering algorithm. Delving into the contribution and relevance of higher-order branches, the study reveals their marginal impact – accounting for approximately 5.2% of the total tree volume (including both stems and branches) – while also highlighting the increased uncertainty in volume estimations caused by their inclusion. These findings advocate for a focused modeling approach that emphasizes primary tree structures, potentially omitting detailed rendering of higher-order branches (i.e. those beyond the 2nd order) to maintain accuracy. In addressing the inadequacies of traditional QSM models in handling trees with complex geometries, this study introduces a novel hybrid approach, integrating Surface Reconstruction and QSM techniques. This approach significantly improves volume estimation for trees with complex geometries, such as buttresses. This method is fully automated, achieving a detection accuracy of 99% for buttressed trees and reducing the MAPD to 14.68%, a substantial improvement over the 21.35% MAPD observed with traditional TreeQSM applications. By advancing the understanding of the potential and limitations of different QSM models and introducing a hybrid modeling technique, this research contributes to enhanced forest carbon stock assessments. It offers a nuanced, precise, and adaptable strategy for volume estimation, thereby significantly aiding forestry management and ecological conservation efforts.