L3SI: A Novel LiDAR 3-D Structural Index Framework for Forest Attribute Retrievals—A Case Study on Stand Volume Estimation

Abstract

Airborne light detection and ranging (LiDAR) provides detailed 3-D structural information for forest attribute retrieval. However, existing parametric models for stand volume estimation often exhibit limited transferability and robustness across heterogeneous forest conditions. This study introduces a LiDAR 3-D structural index (L3SI) framework that flexibly combines LiDAR-derived canopy height and canopy density metrics within a unified allometrically grounded model structure for stand volume estimation. In this framework, the height component represents vertical canopy development, whereas the density component represents canopy occupation or closure, allowing canopy density to modulate the contribution of canopy height to stand volume. Using airborne LiDAR and field plot data from three subtropical forest regions in Guangxi, China, we developed log-linear allometric models with L3SI as the sole predictor and evaluated their performance against representative single-metric and multimetric parametric baseline models developed from the same datasets and assessed under identical data partitions. Across the three regions, L3SI formulations combining mean height or the 95th height percentile with canopy density metrics showed the strongest correlations with the natural logarithm of stand volume. The L3SI-based models achieved coefficients of determination (R2) of 0.65–0.72 and reduced root-mean-square error (RMSE) by 11.20%, 26.20%, and 18.93% in the Eastern, Nanning, and Western regions, respectively, relative to the best-performing single-metric baseline models. They also showed consistently higher accuracy than the tested multimetric parametric models while retaining a single predictor structure. Cross-site validation indicated that L3SI-based models maintained stable performance when transferred among the three subtropical regions, and calibration sample sensitivity analysis further showed their robustness under reduced field plot availability. These results demonstrated that L3SI provides a structurally interpretable, density-modulated height descriptor that can improve LiDAR-based stand volume estimation while maintaining model parsimony and transferability within subtropical forest conditions. Its applicability to other forest attributes, sensor configurations, and biomes should be further evaluated in future studies.