Accurate estimation of individual-tree aboveground biomass is a fundamental challenge in ecological informatics, as it is crucial for forest carbon accounting and ecosystem monitoring. Although unmanned aerial vehicle (UAV)-based hyperspectral imaging offers high spatial and spectral resolution, it remains sensitive to illumination and viewing geometry, leading to temporal spectral inconsistencies compromising retrieval accuracy and model transferability. To address this challenge, we developed a physics-informed informatics framework: the Hapke-based bi-temporal UAV hyperspectral index (HBUHI). Canopy single scattering albedo ( ω ) was first retrieved using the Hapke radiative transfer model from single-day bi-temporal observations to mitigate directional effects, followed by an exhaustive search to identify the optimal band combination at 774 nm (red edge), 822 nm (near-infrared NIR plateau), and 914 nm (longer wavelength NIR). In two homogeneous Mongolian pine ( Pinus sylvestris var. mongolica ) plantations, five-fold cross-validation revealed that HBUHI achieved an R 2 of 0.4763 (RMSE = 19.8220 kg; MAE = 16.3648 kg), outperforming seven widely used vegetation indices and high-dimensional full-spectrum PLSR, LASSO, and Random Forest models. Comparisons using only sunlit canopy pixels or raw reflectance substantially reduced accuracy, highlighting the necessity of physics-based directional correction over simple image masking techniques. Although the performance of HBUHI in structurally complex natural forests and its sensitivity to wind-induced canopy deformations remain to be evaluated, the proposed framework demonstrates a promising, lightweight, and physically interpretable alternative to “black-box” machine learning. It provides a robust pathway for reducing illumination-induced spectral variability in automated UAV-based estimation of individual-tree biomass. • A Hapke-based bi-temporal UAV hyperspectral index (HBUHI) estimates tree biomass. • Retrieving canopy single scattering albedo ( ω ) effectively mitigates directional effects. • HBUHI outperforms standard vegetation indices and machine learning models like Random Forest. • Optimal bands at 774, 822, and 914 nm provide robust physics-based biomass relationships.
Zhao et al. (2026) studied this question.