业务化MODIS BRDF模型对冰雪BRDF/反照率的反演能力评估

Snow/ice surface albedo is an important climate parameter because it governs the surface energy budget on the earth-atmosphere system. MODIS snow/ice albedo products, which are generated on the basis of a kernel-driven Rossthick-LisparseR (RTLSR) model, are extensively used in quantitative remote sensing community. However, the model is developed on the basis of the physics of soil-vegetation system. In theory, the scattering properties between the snow/ice and soil/vegetation systems are distinctly different, thereby causing potential problems in characterizing a Bidirectional Reflectance Distribution Function (BRDF) and albedo for snow/ice cover type using this operational model. In this study, we assess the ability of the operational RTLSR model in characterizing the BRDF and retrieving the surface albedo of snow/ice using Polarization and Directionality of the Earth's Reflectances (POLDER) BRDF data. To verify and compare BRDF/albedo for snow and ice using high-quality collected POLDER observations, we introduce an extensively used asymptotic radiative transfer (ART) model to estimate the BRDF/albedo surface with POLDER observations as a reference given the lack of in situ albedo observations. Results show that (1) the RTLSR model fits the POLDER observations with higher accuracy than the result derived using the ART model. The root mean square error (RMSE) reaches 0.0498 in the 1020 nm band, which increases by 53.70% in comparison with the result using the ART model. (2) The resultant albedos are significantly influenced by the inversion of the RTLSR model for the snow/ice cover type in comparison with the result using the ART model. The RMSE and bias between the two models reach 0.0333 and 0.0274, correspondingly, thus indicating a significant underestimation of surface albedos using the RTLSR model in comparison with the ART model. However, determination coefficient (R²) reaches 0.529, which indicates a high-level correlation between the RTLSR and ART models. Our findings show that the current operational RTLSR model cannot characterize the directional reflectance of snow/ice, thereby affecting the model's capability in estimating the surface albedo for snow/ice, and thus, this model must be modified. Such development can improve the inversion accuracy of surface albedos when this model is used as an operational algorithm for snow/ice cover type in the future.