Objective This article takes the deciduous broad-leaved forest in the Tianma National Nature Reserve in Anhui Province as the research subject, exploring the application potential of the quantitative structure model in biomass estimation of individual tree in complex environment using unmanned aerial vehicle laser scanning data.

Method Through ground surveys and the use of unmanned aerial vehicles, plot data and LiDAR point cloud data were collected, with the ground data serving as the reference measurements. The comparative shortest-path algorithm was used for point cloud segmentation. Subsequently, tree parameters (such as diameter at breast height, trunk volume, branch volume, number of branches, canopy base height, canopy area, canopy volume, and crown width) were extracted from the segmented individual tree point clouds using the quantitative structure model. Pearson correlation coefficient and variance inflation factor were then employed for variable selection of the parameters. Finally, an individual tree biomass estimation model was constructed based on the three machine learning algorithms.

Result Among the biomass models, the one based on random forest (RF) achieved the best training performance (R² = 0.880 0, RMSE = 192.81 kg, rRMSE = 29.88%). The performance of the multilayer perceptron (MLP) model (R² = 0.820 0, RMSE = 233.62 kg, rRMSE = 36.65%) was quite similar to that of the support vector machine (SVM) model (R² = 0.810 0, RMSE = 243.67 kg, rRMSE = 37.77%).

Conclusion This article confirms that the tree parameters extracted from the quantitative structure model can be used to construct a high-precision biomass estimation model for deciduous broadleaf species, providing a new method for resource surveys in complex forest environment.