Evaluating the sensitivity of Sentinel-2 reflectance to vertical canopy profiles using a voxel-based point density proxy framework

Optical satellite data are widely used for forest monitoring, yet their potential for retrieving vertical canopy structure remains underexplored. To address this, we employed the Voxel-Based Point Density Proxy (VPDP) framework to simulate the attenuation of solar irradiance within the canopy and its effect on Sentinel-2 near-infrared reflectance using UAV-LiDAR point clouds from eight broadleaved forest plots. Plots in which canopy occlusion influenced Sentinel-2 reflectance patterns were identified based on the correlation between simulated attenuated irradiance and observed reflectance (r > 0.5). These selected plots were then used for sensitivity analysis, focusing on cases where canopy-induced shading effects were evident and the simulation approach showed reasonable validity. Gaussian functions were fitted to the vertical canopy distributions, and their peak positions (dominant canopy height, μ) and widths (vertical diffuseness, σ) were systematically varied. The results showed that reflectance was moderately sensitive to the Gaussian peak position (representing the dominant canopy layer height). Correlations generally decreased when the peak shifted upward under narrower vertical distributions (σ ≤ 1.0), whereas broader distributions (σ = 1.5) mitigated this decrease. Narrower distributions (σ = 0.5) generally produced lower correlations, implying that reflectance sensitivity to canopy height is highly dependent on the vertical diffuseness of the canopy elements. Although this study was limited to a small number of broadleaved forest plots, the findings provide fundamental evidence that satellite-observed reflectance encodes information on dominant canopy height and vertical diffuseness. This work offers a conceptual basis for developing inversion frameworks to infer vertical canopy structure from optical satellite imagery, contributing to future large-scale mapping of forest height distributions.