Accurate estimation of individual tree above-ground biomass (AGB) and its component-wise allocation is crucial for advancing ecological research and forest management. However, current biomass estimation methods, such as destructive sampling and allometric equation–based approaches, face limitations in both operational efficiency and cost-effectiveness, and only destructive sampling can provide component-wise biomass measurements, which is impractical for large-scale studies or repeated measurements.

In this study, we present a terrestrial laser scanning (TLS)-based workflow integrating wood–leaf separation, voxel-based foliage estimation and detailed 3D reconstruction of tree architecture to achieve accurate estimation of individual tree AGB and its component-wise allocation. A total of 68 trees were scanned to obtain high-resolution TLS data and subsequently destructively harvested to acquire field reference measurements for validation.

The results demonstrate that the workflow achieved high accuracy in predicting AGB at the individual tree level (coefficient of determination/R2 = 0.88, root mean squared error/RMSE = 16.83 kg, mean absolute error/MAE = 12.18 kg), significantly outperforming estimates derived from locally calibrated allometric equations (R2 = 0.61, RMSE = 29.86 kg, MAE = 24.52 kg). Furthermore, this study provides evidence of the strong capability of TLS in estimating branch-level biomass, with high accuracy achieved across branch orders (R2 ranging from 0.66 to 0.91, RMSE from 3.55 to 380 g and MAE from 2.97 to 290 g).

By providing precise, non-destructive estimates of biomass distribution across branches and leaves, this workflow demonstrates strong potential for improving the accuracy of tree biomass quantification, supporting investigations of resource allocation strategies, and enhancing forest carbon monitoring.