Simultaneous estimation of leaf directional-hemispherical reflectance and transmittance from multi-angular canopy reflectance

Remote sensing retrieval of foliar biochemical traits via inversion of radiative transfer models is hampered by the uncertainties arising from the ill-posed problem of insufficient remote sensing information for the inversion. The use of leaf transmittance in addition to leaf reflectance in retrieving leaf traits could be a key step to alleviate this issue. To achieve this objective, our study introduces a novel theoretical framework for the simultaneous retrieval of leaf directional-hemispherical reflectance and transmittance, leveraging multi-angular optical observations at the canopy level. This framework partitions the canopy's spectral reflectance in the range of 400 nm to 2500 nm into two distinct components, each aligning with the contributions of 1st-order and multiple-order scatterings of radiation in the canopy. At a specific illumination-observation (IO) angle, these two scattering components are assumed to be determined by leaf hemispherical reflectance and scattering albedo (i.e., the sum of leaf reflectance and transmittance). The average leaf directional-hemispherical reflectance and transmittance can be estimated from combined canopy reflectance observed under different IO angles. The proposed framework was implemented using a single site data collected on August 19th, 2018, in Saihanba National Forest Park, Hebei, China. A comprehensive dataset was compiled, including hyperspectral images of a dense white birch canopy acquired at 8 different illumination-observation (IO) angles. This dataset was augmented with corresponding canopy structural traits, including leaf area index, diameter at breast height, crown radius, tree number, and domain size. Additionally, terrain features such as slope and aspect, along with leaf spectra and background spectra, were included in this extensive dataset. The geometrical-optical model GOST2 was utilized to establish the relationships between canopy reflectance and leaf reflectance and transmittance. Initially, the performance of GOST2 was evaluated, followed by the construction of a leaf spectral dataset generated by PROSPECT-D and input into GOST2 to construct the LUT. Subsequently, leaf reflectance and transmittance were estimated for different combinations of IO angles. The results demonstrated a close agreement between the canopy reflectance factors modelled by GOST2 and the observations obtained from UAV. Overall, leaf reflectance was more accurately retrieved than leaf transmittance across the entire spectrum of interest, primarily due to the significant influence of the 1st-order scattering in canopy reflectance. In the near-infrared region, where the magnitudes of 1st- and multiple-order scatterings were similar, the proposed framework produced satisfactory leaf transmittance with a relative error of <10% when compared to the measured reference data. Furthermore, the results indicated that the absolute error of model simulations should be less than half of the multiple-order scattering, which is particularly stringent for wavelengths characterized by strong absorption features. The proposed framework holds potential for reducing uncertainties in various fields, including global mappings of leaf biochemical traits.