积雪在地球表面的覆盖面积极大，地球表面约有三分之一的地区有季节性的积雪覆盖，积雪在地球的辐射平衡中起着至关重要的作用。作为地球第三极，青藏高原拥有丰富的积雪资源，积雪覆盖与青藏高原及其周边地区的生物和水文过程密切相关，影响着全球的能量和水循环，研究青藏高原的积雪覆盖情况有着十分重要的意义。学者们针对青藏高原的积雪特征发展了多个积雪覆盖度算法，并提供了该区域的积雪覆盖度产品。在对不同积雪覆盖度产品进行精度评价时，不同研究人员采用的“地面真值”数据不同，造成不同算法与产品精度评价结果的差异。由于Sentinel-2与Landsat-8组合可在中高纬地区达到3天的高时间分辨率。因此，本文利用高时空分辨率的Sentinel-2，结合高空间分辨率的Landsat-8数据，开展高时空分辨率积雪覆盖度验证数据集构建的算法研究，并利用此数据集对国际上现有的四种公里级的积雪覆盖度产品进行精度评价。

本论文主要从三个方面开展数据集构建的算法研究，一是高时空分辨率积雪覆盖度反演，开展将二值化积雪判识算法和基于自动端元选取的光谱混合分析算法扩展至Sentinel-2数据的研究。二是高时空分辨率积雪覆盖数据精度评价，利用GF-2数据对Landsat-8和Sentinel-2积雪覆盖数据进行精度分析。三是高时空分辨率积雪覆盖数据在不同空间尺度验证结果的差异分析，探究使用相对较低分辨率数据进行验证的可行性，从而提升计算效率。本研究的主要内容及结果如下：

1.开展了将二值化积雪判识算法和基于自动端元选取的光谱混合分析算法扩展至Sentinel-2数据的研究。本研究使用的二值化积雪判识算法包括监督分类方法，以及SNOMAP算法。监督分类得到10 m的Sentinel-2积雪覆盖数据，SNOMAP算法得到20 m以及30 m的二值积雪数据，基于自动端元选取的光谱混合分析算法得到了20 m以及30 m的积雪覆盖度数据。

2.开展了GF-2评价高时空分辨率积雪覆盖数据精度的研究。为了定量评价Sentinel-2二值积雪、积雪覆盖度与Landsat-8积雪覆盖度的精度，使用3.2 m空间分辨率的GF-2卫星数据分别对三者进行验证。由于用于精度评价的二值积雪评价指标和积雪覆盖度评价指标不同，无法对比Sentinel-2二值积雪、积雪覆盖度，因此构建指标表示GF-2与Sentinel-2二值积雪、积雪覆盖度面积差的绝对值的比，对两者积雪判识的精度进行比较评价。结论表明，Sentinel-2和Landsat-8积雪覆盖度精度较高，可以作为“地面真值”数据；Sentinel-2积雪覆盖度相较于二值积雪结果，精度更高。

3.开展了Sentinel-2二值积雪、积雪覆盖度与Landsat-8积雪覆盖度在不同尺度下验证结果差异分析的研究。将不同算法得到的五种Sentinel-2积雪覆盖和Landsat-8积雪覆盖度在不同空间尺度验证MODAGE积雪覆盖度，探究验证结果的差异。结果表明，30 m的积雪覆盖可以替代20 m的积雪覆盖结果对中低分辨率的积雪产品进行验证；监督分类算法得到的10 m二值结果可以替代MESMA-AGE 20 m/30 m的结果对中低分辨率的积雪产品进行验证；SNOMAP算法得到的20 m/30 m的二值积雪结果不能代替MESMA-AGE算法得到的结果参与验证；30 m空间分辨率的Sentinel-2、Landsat-8积雪覆盖度可以互为替代地验证中低分辨率的积雪产品。

4.基于Sentinel-2和Landsat-8数据构建积雪覆盖度验证数据集并评估国际上四种公里级尺度的积雪覆盖度产品的精度。数据集的时间范围为2018-2020年，包括426景Landsat-8积雪覆盖度影像与98景Sentinel-2积雪覆盖度影像，空间分辨率为30 m。利用数据集对比验证了GlobSnow SE日积雪覆盖度与MODAGE积雪覆盖度，并分别分析两者在青藏高原地区的适用性与精度，结果认为MODAGE积雪覆盖度精度优于GlobSnow SE产品的精度，由于GlobSnow SE日积雪覆盖度产品受云的影响严重以及存在时间尺度上的不连续，因此GlobSnow SE日积雪覆盖度产品，目前不适用于大范围的青藏高原积雪覆盖度研究，仅适用于帕米尔高原以及北纬40°的局部小区域。对于长时间尺度的积雪变化研究，则可以使用GlobSnow SE的周积雪覆盖度产品以及月积雪覆盖度产品。同时，对比验证了MOD10A1积雪覆盖度与高亚洲逐日去云的积雪覆盖度，发现与去云前的MODIS积雪覆盖度数据相比，该产品的均方根误差和平均绝对误差不变或有所减小，表明该产品的去云效果较好，然而在地表覆盖类型为草地时，产品精度低于其他地表覆盖类型。

Snow covers a huge area on the earth's surface. About one-third of the earth's surface is covered by seasonal snow, snow plays a vital role in the earth’s radiation balance. As the third pole of the earth, the Qinghai-Tibet Plateau is rich in snow resources, and snow cover is related to the living things of the Qinghai-Tibet Plateau and its surrounding areas. The snow cover of the Qinghai-Tibet Plateau also affects the global energy and water cycle. It is of great significance to study the snow cover of the Qinghai-Tibet Plateau.Scholars have developed several algorithms of snow cover on the Qinghai-Tibet Plateau, and have provided snow cover products in this region. When evaluating the accuracy of snow cover products, the "ground truth" data used are often different, leads to the difference between different algorithms and product accuracy. Sentinel-2 and Landsat-8 can achieve a 3-day temporal resolution in the middle and high latitudes. This article mainly uses Sentinel-2 and Landsat-8 data to do the research on constructing a validation dataset. And use this dataset to evaluate the accuracy of the four snow cover products.

This paper mainly studies the algorithm of dataset construction from three aspects. The first is the retrieval of snow cover. This study mainly carried out the study of extending the binary snow algorithm and spectral unmixing analysis algorithm based on automatic endmember selection to Sentinel-2 data. The second is the accuracy evaluation of snow cover data. GF-2 data was used to evaluate its accuracy. The third is the difference analysis of the validation results of snow cover data at different spatial scales. The main contents and results of this research are as follows:

(1) This study extend the binary snow algorithm and spectral unmixing analysis algorithm based on automatic endmember selection to Sentinel-2 data.The binary snow algorithm used in this study include supervised classification method and SNOMAP algorithm.The Sentinel-2 snow cover data of 10 m is obtained by supervised classification, the binary snow cover data of 20 m and 30 m is obtained by SNOMAP algorithm, and the snow cover data of 20 m and 30 m is obtained by spectral unmixing analysis algorithm based on automatic endmember selection.

(2) GF-2 was used to evaluate the accuracy of high spatial and temporal resolution snow cover data. GF-2 data with 3.2 m spatial resolution were used to verify the snow cover data respectively. And construct the index represents the ratio of the absolute area difference between GF-2 and Sentinel-2 binary snow cover and snow cover. The results show that the accuracy of snow cover between Sentinel-2 and Landsat-8 is high, and there is no significant difference. The accuracy of Sentinel-2 snow cover is higher than that of binary snow cover.

(3) The difference between Sentinel-2 and Landsat-8 snow cover is analyzed. The Sentinel-2 and Landsat-8 snow cover obtained by different algorithms were used to validate the MODAGE at different spatial scales to explore the differences in the validation results. It is concluded that the snow cover of 30 m can replace the snow cover of 20 m to validate the medium and low resolution snow products. The 10 m binary results obtained by the supervised classification algorithm can replace the MESMA-AGE 20 m/30 m results to validate the medium and low resolution snow products.The 20 m/30 m binary snow cover results obtained by SNOMAP algorithm cannot replace the results obtained by MESMA-AGE algorithm. The Sentinel-2 and Landsat-8 snow cover with spatial resolution of 30 m can be used alternatively to validate the medium and low resolution snow products.

(4) The Sentinel-2 and Landsat-8 data were used to construct snow cover validation dataset and evaluate the accuracy of four snow cover products. The dataset is composed of 426 Landsat-8 and 98 Sentinel-2 snow cover images from 2018 to 2020. The spatial resolution of the dataset was 30 m. We use the dataset validate the GlobSnow SE and MODAGE, the accuracy of the two are analyzed respectively. The result shows that the accuracy of MODAGE is better than GlobSnow SE product. GlobSnow SE’s daily snow cover products are not suitable for large-scale snow cover on the Tibetan Plateau. The research is only applicable to the Pamirs and a small area at 40°N latitude. Then validate the MOD10A1 and High Mountain Asia snow cover product. Compared with the MODIS snow cover data before the cloud removal, the error of the product remains unchanged or reduced. The cloud removal effect is good, but when the surface cover type is grass, the accuracy of this product is not as good as that of other surface cover types.