冰雪反照率作为区域和全球能量收支平衡的一个关键参数，它在气候变化 、水循环和碳交换过程中起着非常重要的作用。目前，基于核驱动模型理论(尤其是RossThick-LiSparseReciprocal, RTLSR)生产的全球中分辨率成像光谱仪(MODIS) BRDF/Albedo产品在定量遥感中有着广泛的应用。该模型最初是从连续和离散植被冠层的简化场景发展而来，并且已被广泛用于拟合土壤-植被系统的多角度观测和反演地表反照率。然而，冰雪的散射特性和植被的散射特性具有显著地差异，冰雪通常在前向具有较高的反射，而植被往往在后向具有较高的反射。因此，核驱动模型在表征冰雪BRDF特征和估算冰雪反照率方面需要进一步评估和发展。 针对上述问题，本研究首先基于渐进辐射传输模型(asymptotic radiative transfer，ART)和全球POLDER数据全面评估了MODIS业务化算法(即RTLSR)在表征冰雪散射特征和反演冰雪反照率的能力；然后，在核驱动模型框架下，从ART模型中推导了一个冰雪散射核，并通过多种数据源验证了该方法的可行性；在此基础上，本研究将该冰雪散射核进一步与Qu等人基于Rahman-Pinty-Verstraete (RPV)模型发展的冰雪散射核进行了详细比较(以下分别简称为ART方法和Qu方法)；最后，本研究基于多种数据源在核驱动RossThichk-Roujean(RTR)模型框架下评估了从ART模型发展的冰雪散射核反演冰雪反照率的能力(RossThichk-Snow，RTS)。因此，本研究的结论主要包括以下几个方面： (1)基于全球POLDER冰雪数据的研究表明，MODIS线性核驱动模型基本上不能很好地表征冰雪的二向反射特征，并且该模型反演的冰雪反照率相较于ART模型整体上存在明显的低估，其平均偏差可达到0.027。为了使核驱动模型能准确反演冰雪二向性反射和反照率，该模型需要针对冰雪散射特征进一步发展。 (2)本研究基于校正形式的ART模型在核驱动模型框架提出了一个冰雪散射核，将核驱动模型的适用性从土壤-植被体系扩展到了冰雪，使该模型能够更好地表征冰雪的散射特征。通过使用多种数据源、从多个尺度验证分析表明：该冰雪散射核在核驱动RTLSR模型框架下可以准确地拟合各种冰雪BRDF数据，表现出较高的反演精度(R2=~0.9)。 (3)Qu等人基于RPV模型发展的冰雪散射核仅考虑了较强的前向散射特性，难以表征在小太阳天顶角条件下BRDF形状的变化，特别是在太阳天顶角范围为35°-55°。同时，Qu方法在前向存在明显的低估，特别是在观测天顶角和太阳天顶角较大的情况下，而基于ART模型发展的冰雪散射核在各种太阳天顶角变化条件下都表现出较高的拟合精度。 (4)虽然RTR模型反演的冰雪反照率与bicontinuous photon tracking (bic-PT)模型模拟的反照率具有很高的一致性，但该模型反演的结果相较于bic-PT模拟的反照率在统计学上具有显著性差异。在太阳天顶角范围为30°-70°时，RTR模型反演的反照率相较于bic-PT模型在红和近红外波段分别低估了0.71％和0.69％。而RTS模型在太阳天顶角范围为30°-70°处反演的冰雪反照率具有可忽略的偏差，并且与这些模拟的反照率数据没有显著性差异。对于实测和卫星数据，通过RTR模型反演的冰雪反照率相较于RTS模型也具有明显的低估，RTR模型反演的POLDER短波黑白天空反照率相较于RTS模型的反演结果分别低估了1.43％和1.54％。

The snow albedo is a key parameter for regional and global energy balances, which plays a vital role in global climate change and water cycle. The global MODIS BRDF/Albedo products have been widely used in the quantitative remote sensing community, which are generated based on the kernel-driven BRDF model (e.g. RossThick-LiSparseReciprocal, RTLSR). This model is built up by simplifying scenarios of continuous and discrete vegetation canopies and has been extensively utilized in characterizing the BRDF signatures and retrieving the surface albedo of the soil-vegetation systems. However, the scattering characteristics of snow are distinctly different with the soil-vegetation systems. The snow usually has a higher reflectance in the forward direction, while the vegetation tends to have a higher reflectance in the backward direction. Therefore, the kernel-driven BRDF model needs a further evaluation and development in characterizing the snow BRDF characteristics and estimating snow albedo. In this study, we assess and develop a snow kernel in the kernel-driven BRDF model based on the asymptotic radiative transfer (ART) model regarding these problems. Firstly, we assess the ability of the kernel-driven RTLSR model in characterizing snow scattering signatures and retrieving snow albedo based on the ART model and POLDER BRDF database. Then we derive a snow kernel from the ART model in the kernel-driven RTLSR model framework (i.e. RossThick-LiSparseReciprocal-Snow, RTLSRS), and we validate the method using various BRDF data sources. Next, we further assess the performances of this snow kernel and the snow kernel proposed by Qu et al. based on the Rahman-Pinty-Verstraete (RPV) model in characterizing snow BRDF signatures (hereinafter named the ART and Qu methods, respectively). Finally, we assess the performance of the snow kernel derived from the ART model (i.e. RossThichk-Snow, RTS) in the kernel-driven RossThichk-Roujean (RTR) model framework in retrieving snow albedo. Our results show that: (1)The kernel-driven RTLSR model has difficulty in capturing snow BRDF characteristics for the POLDER data. The snow albedo retrieved by this model has a significant underestimation compared with the ART model, and the bias can reach 0.027. Therefore, the kernel-driven RTLSR model needs to be further developed for the snow scattering characteristics. (2)We propose a snow kernel in the kernel-driven BRDF model framework based on the corrected form of the ART model, which extends the applicability of the kernel-driven model from the soil-vegetation system to the snow. The snow kernel in the kernel-driven RTLSR model framework has been validated by a variety of snow BRDF data and shows to have a high accuracy in capturing the snow bidirectional signatures(R2=~0.9). (3)The snow kernel developed by the Qu et al. based on the RPV model only considers strong reflectance in the forward-scattering direction for snow cover and has difficulties capturing the variations in the BRDF shape in a range of small solar zenith angles (SZAs), especially at SZA=35o-55o. Although the Qu method only focuses on characterizing the strong reflectance of snow surfaces in the forward-scattering direction, this method still underestimates the observed reflectance of snow surface in the forward-scattering direction, especially at large SZAs. However, the snow derived from ART model shows to have a high accuracy under various variations of SZA. (4)The snow albedo retrieved by the RTR model has statistically significant difference with the albedo simulated by the bicontinuous photon tracking (bic-PT). At SZA = 30°-70°, the albedo retrieved by the RTR model underestimates 0.71% and 0.69% in the red and near-infrared bands, respectively. In contrast, the RTS model has a higher accuracy with these simulated albedo for all SZAs. The albedo retrieved by the RTS model at SZA = 30°-70° has a negligible bias and shows no significant difference. The albedo retrieved by RTR model has a significant underestimation compared with the results of the RTS model for the field-measured and POLDER data. The shortwave albedo retrieved by the RTR model is underestimated by 1.43% and 1.54% in the black-sky albedo and white-sky albedo compared with the RTS model, respectively.