LAI Estimation using Simulated Full-Waveform Lidar Data in a Complex Forest Scene

Leaf Area Index (LAI) is a vital parameter for understanding vegetation structure , canopy complexity, and ecosystem productivity. Full-waveform LiDAR system allows the collection of continuous backscattered energy profiles that enables a more detailed characterization of canopy structure and related variations compared to traditional discrete-return LiDAR systems. By utilizing waveform-derived metrics, LAI can be estimated with improved accuracy, specially in a dense or complex vegetation. This study presents a novel approach to LAI estimation by combining multiple full-waveform-derived metrics derived from simulated NASA's Land, Vegetation, and Ice Sensor (LVIS) footprints over the Harvard Forest scene. The truth data for modeling LAI is derived through voxelization of the forest scene, leveraging the known geometry of leaves (where individual leaves are modeled as smaller facets of known dimensions) and their arrangements within the tree to determine column-based projected leaf area onto the ground, which, by definition, represents LAI. Out of 30 candidate met-rics, the optimal combination-includes (1) Gini Coefficient (GC), (2) Centroid (intensity-weighted mean), and (3) Total Return Energy (RWE). Ordinary Least Squares (OLS) regression using these metrics on raw waveform data produced an Adjusted R 2 of 69.4%, indicating a strong predictive relationship. Among the selected predictors, RWE (β = 2.0976, p < 0.001) had the highest positive contribution to LAI estimation, followed by Centroid (β = −2.9073, p < 0.001), while GC (β = −4.8867, p < 0.001) also showed a significant negative effect. Moreover, we investigated the impact of removing the system contribution effect using a deconvolution-based approach. This preprocessing step improved model performance, increasing the Adjusted R 2 to 72.3%, suggesting that deconvo-lution improves the predictive power of waveform-derived metrics. Noticeably, after deconvolution, GC (β = −0.3026, p = 0.743) became statistically insignificant , while RWE (β = 1.5856, p < 0.001) and Centroid (β = −2.2318, p < 0.001) remained strong predictors. Overall, this study corroborates the ability of full-waveform LiDAR combined with signal processing techniques for accurate and robust LAI estimation in complex forest environments and suggest its broader applicability for ecosystem function analysis.