Angular Normalization of Airborne Spectral Imagery in Forested Areas by Improved Pseudo Multiangle Observation

Airborne spectral imagery is significantly affected by angular effects due to the push-broom sensor geometry and the anisotropic reflectance of forest canopies, introducing uncertainties in vegetation variable retrieval. Conventional angular normalization methods rely on multi-angle datasets and kernel-driven BRDF models, however, the limited availability and inherent uncertainty of true multi-angle observations constrain their effectiveness. This study integrates structural and spectral similarity metrics (including canopy cover, point density profile, and NDVI distribution) derived from co-registered airborne LiDAR and spectral imagery. These metrics enable the construction of pseudo multi-angle data by identifying spatially distinct forest patches with highly similar structural and spectral properties but different viewing geometries.Validation across multiple forest types using simulated datasets demonstrates that the extracted forest patches exhibit consistent structural and spectral characteristics, and that the constructed pseudo multi-angle data effectively capture angular variations in bi-direction reflectance factor (BRF). Kernel-driven BRDF models applied to these data show strong agreement between predicted and observed BRFs, particularly along the principal plane. Correspondingly, the angular-normalized BRFs closely align with the real BRF obtained in the nadir direction for all three representative spectral bands where slight underestimation was observed in the visible bands. The performance of the normalization approach was also found to be influenced by canopy structural complexity, underscoring the importance of incorporating canopy structure into angular correction processes. Given that airborne remote sensing platform usually can simultaneously obtain spectral and LiDAR data, the strategy proposed in this study offers better potential for improving angular normalization.