Accounting for terrain in neighborhood competition: fused UAV and backpack LiDAR for fine-scale volume estimation in Chinese fir (Cunninghamia lanceolata) plantations

Abstract

Chinese fir ( Cunninghamia lanceolata ) is an important fast-growing timber species in southern China, and accurate estimation of individual-tree volume is important for stand management and carbon accounting. In plantations located on complex terrain, neighborhood competition among trees is difficult to describe using planar distance alone because terrain-induced elevation differences may alter inter-tree spatial relationships. This study analyzed 12 pure Chinese fir plots in Jingdezhen, Jiangxi Province, China, representing low-, medium-, and high-density stand conditions. Field growth survey data were integrated with fused unmanned aerial vehicle (UAV) and backpack light detection and ranging (LiDAR) points, and three-dimensional competition indices were constructed from digital elevation model (DEM)-constrained Voronoi–Delaunay neighborhoods. Machine-learning models combined with Shapley additive explanations (SHAP) were used to evaluate the predictive contributions of focal-tree attributes, neighbor attributes, and competition variables to equation-derived individual-tree volume. Random Forest showed the most consistent performance gains when conventional two-dimensional competition indices were replaced by DEM-constrained three-dimensional competition indices, and was therefore selected for detailed comparison. Using individual-tree volume calculated from field-measured diameter at breast height and tree height as the reference response variable, we found that, under the competition-pair configuration, replacing the conventional two-dimensional competition indices with their DEM-constrained three-dimensional counterparts increased R 2 from 0.74 to 0.85 for the overall dataset and from 0.81 to 0.90 for high-density stands, while RMSE decreased by 0.0088 m 3 tree −1 (20.3%) and 0.0199 m 3 tree −1 (42.8%), respectively. Furthermore, SHAP analysis indicated that focal-tree size attributes remained dominant in model prediction, whereas DEM-constrained three-dimensional competition indices provided clearer complementary predictive information compared to their corresponding two-dimensional indices in medium- and high-density stands.