多源亮温观测支持下的被动微波雪深与雪盖参数协同反演研究

Snow is the most active natural element in the winter. In recent years, snow decreased significantly. Because of the characteristics of the long measurement history and the short revisit time of the passive microwave observations, it is of critical importance to develop the passive microwave snow retrieval algorithms to support the study of the response of the snow change to the climate. In the snow and soil system, the grain size and the background soil emission beneath the snow influence the accuracy of the snow depth retrieval results. The snow depth uses the snow density to convert to snow water equivalent. Therefore, referring to the band and the incidence angle configurations of the current AMSR2, SMAP and SMOS satellite sensors, using a massive number of passive microwave brightness temperature observations to study the feasibility of simultaneously estimating four parameters, which include the snow depth, snow grain size, soil moisture content and snow density, is the main topic of this proposal. The study will first build a synthetic retrieval test database using the brightness temperature simulated by the bi-continuous snow and the AIEM (Advanced Integral Equation Model) soil microwave emission models. Then, using the iterative Markov Chain Monte Carlo algorithm based on the forward model and the machine learning algorithms, the feasibility of the retrieval will be studied based on the assumptions of an adequate number of observations and an insufficient number of observations supported by the prior knowledge. Both the simulations and the retrievals will consider the vertical inhomogeneity of snow. This is the key to match the brightness temperature at different frequencies using the snow cover parameters. The focus of this study will be put on improving the calculation efficiency and using the real observations from the ground-based radiometer observations and the satellite radiometry to validate the conclusions from the synthetic tests. It will promote the application of the algorithm for remote sensing products.

积雪是冬季最活跃的自然要素。近年来冬季积雪日益减少，发展被动微波积雪反演算法，利用其历史数据长、重访周期短的特征，支持积雪的气候变化响应研究很有必要。在积雪-土壤系统中，雪粒径和土壤背景辐射影响着雪深反演精度，为计算雪当量又需雪深需准确估计的雪密度。因此，本课题拟参考现有AMSR2、SMAP和SMOS卫星传感器的波段和角度配置，在大量亮温观测支持下，研究同时反演雪深、雪粒径、土壤含水量和雪密度四个参数的可行性。课题将以高精度积雪和土壤辐射模型构建的亮温库为基础，结合先验知识，采用基于正向迭代的马尔科夫链-蒙特卡洛算法以及机器学习等算法，进行充分和非充分观测条件下的反演实验并评估其效果。反演和模拟将考虑积雪的垂直分层异质性，这是雪、土参数匹配多频亮温观测的关键。课题还将重点研究正向辐射模型和反演算法的效率提升，以实现基于地基辐射计和卫星亮温数据开展反演，推进算法向遥感产品转化。

积雪是冬季最活跃的陆表自然要素之一。当前被动微波雪深、雪水当量产品存在精度不足的问题。为提高反演精度，强化在雪深反演过程中综合考虑雪粒径、雪密度时空动态变化和积雪分层异质性，课题提出利用当前AMSR2（AMSR-E）、SMOS等已发射卫星和传感器的多源、多频率、多角度配置，开展雪深、雪粒径、雪密度、冻土含水量等多个雪-土参数的联合反演研究。研究内容包括积雪模型验证、可行性分析和基于实测数据的反演实验三部分；当欠定问题突出时，允许使用先验知识。课题首先在我国青藏、新疆地区对积雪过程模型和辐射传输模型模拟进行了验证。随后，根据AMSR2和SMOS配置，构建了亮温模拟数据库，证明了当同时使用10.65到89 GHz双极化四频率亮温和L波段双极化多角度亮温时，可以不使用先验知识获得反演雪深、雪粒径、雪密度和冻土未冻水含量的能力，在多层雪假设下的反演均方根误差分别为10.8 cm (0~200 cm范围)、0.18 mm (0~2.5 mm范围)、32.6 kg/m3 (0~500 kg/m3)和1.34% (0~50%范围)。然而，事实表明，由于被动微波粗分辨率、实际地表复杂性和辐射传输模型不完善之处，基于模拟数据库建立的反演算法不能直接运用到实测卫星数据。因此，课题引入先验知识，开展了三项基于实测卫星数据的反演工作。其一，对全球地面雪深观测做台站代表性分析，以实测雪深、过程模型模拟先验知识和AMSR2实测亮温构建雪深反演算法，然后根据反演雪深重新率定降雪量获取其它参数估计。其二，结合积雪过程模型和辐射传输模型构建不同气象条件下的雪-土物理参数及亮温样本，以AMSR2亮温和MODIS积雪覆盖率平差估计样本权重，应用权重获取能量水量平衡的雪-土系统多参数估计。其三，在加拿大魁北克地区开展了基于SMOS的雪密度反演研究。课题为扩展积雪被动微波遥感应用提供了新实例、新思路。