A TLS-Based Framework for Individual-Tree Structural Reconstruction and Improved Biomass Estimation

Accurate representation of forest structure and above-ground biomass (AGB) at the individual-tree scale is essential for improving carbon cycle assessments and enabling emerging forest Digital Twin applications. Terrestrial Laser Scanning (TLS) provides high-resolution three-dimensional observations of vegetation structure; however, transforming dense point clouds into biologically meaningful and computationally efficient tree-level models remains challenging. Here, we present a TLS-based end-to-end framework for individual-tree structural reconstruction and component-wise AGB estimation using high-density point cloud data acquired over a small forest copse within the Indian Institute of Technology Kharagpur campus using a Terrestrial Laser Scanning system. The workflow begins with multi-scan point cloud registration and preprocessing, including noise removal, ground filtering, and vegetation isolation. Multi-view acquisition is employed to mitigate occlusion, and residual data gaps are addressed through model-based structural interpolation of woody elements. Individual trees are delineated using geometric clustering-based instance segmentation (TreeIso), with segmentation quality assessed against field-mapped stem locations. Leaf and woody components are separated using a combination of graph-based structural analysis. Woody architecture, including stems and branches, is reconstructed using quantitative structure modelling (TreeQSM), where adaptive cylinder fitting is applied to derive branching topology and woody volume. Leaf biomass is estimated independently by converting classified leaf points into a voxel-based crown representation from which leaf area is derived. Leaf mass is then calculated using species-specific specific leaf area (SLA) values obtained from field sampling. Species identity and corresponding wood density values are assigned using concurrent field inventory data. Component-wise woody and foliar masses are combined to obtain tree-level AGB estimates. Each stage is implemented using alternative models and parameterisations. Selected models are regionally calibrated to improve performance under the conditions of the study area. To ensure biological realism, reconstructed tree geometry is validated against field measurements and reference allometry using stem diameter at breast height (DBH), tree height, and crown metrics, and sensitivity analyses are conducted to quantify uncertainty propagation from segmentation, classification, and QSM parameterisation into final biomass estimates. The final framework demonstrates the potential of TLS-derived point clouds to produce validated, structurally explicit tree-level models that support carbon accounting, ecosystem modelling, and calibration of airborne and satellite-based biomass products, thereby bridging in situ measurements and multi-scale Earth observation systems. Keywords : Terrestrial Laser Scanning (TLS); Individual-tree segmentation; Leaf–wood separation; TreeQSM; Above-ground biomass (AGB).