FuzzyQSM: A Novel Tree Structure Reconstruction Approach Using Uncertainty

Forests constitute a large proportion of vegetation biomass, and laser scanning enables the determination of various ecological metrics by reconstructing the tree geometry from a point cloud into a quantitative structure model (QSM). QSMs commonly consist of a series of cylinders fitted to the point cloud using a least-squares-based approach. Due to the finite laser beamwidth, laser scanning data is, however, inherently uncertain. Additionally, data quality is commonly poor for large parts of a tree due to occlusion and beamwidth being comparable in size to the diameter of smaller branches. Therefore, instead of using a point cloud directly, we quantify uncertainty in the point cloud to arrive at a fuzzy cloud. The QSM’s cylinders are fitted to this fuzzy cloud using the expected Mahalanobis distance in an approach called FuzzyQSM. This approach has the benefit of taking the magnitude and orientation of uncertainty into account to better reflect the underlying nature of the data; however, it also requires an initial geometry estimate. To test this approach, we derived twenty-five artificial tree models of five different species from reconstructed trees. Performance was compared to TreeQSM using simulated point cloud data. We found that FuzzyQSM achieves a stem fit comparable to TreeQSM, whilst providing a clear improvement in total and branch volume estimates for the tested artificial trees and scanning configurations. On average, TreeQSM underestimates total and branch volumes by 16.9% and 22.2%, respectively, whereas the corresponding underestimation with FuzzyQSM is only 6.3% and 2.2%. Although we evaluated it only on synthetic data, we believe the approach shows promise for tree volume estimation, albeit at a higher computational cost.