积雪是地表覆盖的重要组成部分之一，其年际和季节变化会影响区域的气候和水资源平衡，进而影响全球能量平衡和气候变化。中分辨率成像光谱仪（Moderate-resolution Imaging Spectroradiometer, MODIS）积雪覆盖产品已广泛应用于积雪覆盖研究的各个领域。然而，作为光学传感器，MODIS不可避免地受到云层的严重影响，致使其积雪产品，尤其是逐日积雪产品存在大量的数据缺失。因此，发展新的MODIS积雪产品去云方法，获取准确的高时空分辨率的积雪覆盖信息，对气候变化背景下积雪覆盖时空动态变化格局、融雪径流建模及径流量的准确估算乃至水资源管理都具有重要的理论价值和实践意义。

青藏高原（Tibetan Plateau, TP）是我国主要积雪区之一，其积雪的时空演变对亚洲季风乃至东南亚洪涝灾害具有重要的影响。鉴于此，本研究选择该区域为典型试验区，针对现有研究存在的问题，在分析积雪累积和消融过程基本原理的基础上，提出了一种针对MODIS雪盖指数（Normalized Difference Snow Index, NDSI）产品的时空插补方法，制备了2003 ~ 2018年青藏高原地区逐日无云雪盖指数（NDSI）时间序列产品；在此基础上，分析了青藏高原地区积雪覆盖的时空格局、变化趋势以及驱动因素。最后基于融雪径流模型（Snowmelt Runoff Model, SRM）模拟的径流结果分析了融雪季积雪覆盖率差异对径流模拟精度的影响。本研究的主要工作和结论如下：

（1）提出基于相似像元选取的去云算法框架，耦合了时间、空间、雪盖指数丰度和积雪变化历史规律信息，去云后全年平均云覆盖率小于2%，总体精度接近97%。

本研究提出基于相似像元选取的去云算法框架，该框架包含下午星融合（Terra and Aqua Combination, TAC）、3日时间滤波（3 Day Temporal Filter, 3DTF）和相似像元选取（Similar Pixel Selecting Algorithm, SPSA）三个步骤。实验结果表明TAC和3DTF分别可去除约8%和9%的云量。SPSA算法在不造成明显精度损失的情况下可将云量降低至 < 2%，平均绝对误差（Mean Absolute Error, MAE）为2.77%，平均绝对百分比误差（Mean Absolute Percentage Error, MAPE）为42.40%，均方根误差（Root Mean Square Error, RMSE）为3.78%，决定系数R²为0.78，平均总体精度（Overall Accuracy, OA）为96.92%，高估误差（Overestimation Error, OE）为1.10%，低估误差（Underestimation Error, UE）为1.98%。同时，SPSA算法具有很好的鲁棒性，去云精度不受云量影响，即使云覆盖比例超过60%（3DTF产品），SPSA仍保持较高精度。

（2）基于MODIS逐日无云（SPSA）与8日合成（MOD10A2）两种积雪覆盖产品，分析了2003 ~ 2018水文年间青藏高原积雪覆盖的时空分布格局、年际变化趋势以及驱动因素，并探讨了两种数据源在分析结果方面的异同。

从时空分布格局来看，青藏高原积雪覆盖空间分布具有较强空间异质性，积雪覆盖日数（SCD）高值区主要集中在青藏高原边缘以及腹地的高大山脉。受大气环流影响，青藏高原积雪覆盖时间分布表现为冬季型、冬/春型和秋季型三种特征。从年际变化趋势来看，2003 ~ 2018水文年间，青藏高原57.4%区域的SCD呈现下降趋势，平均SCD下降速率为-5.45 d?10a^-1。其中在冬季和秋季的下降速率最大，其值分别为-2.92 d?10a^-1（p < 0.1）和-2.13 d?10a^-1。SCD减少速率峰值出现在海拔5000 ~ 6000m之间。从SCD驱动因素来看，青藏高原年SCD与气温和降水均表现为负相关，其相关系数分别为-0.80（p < 0.05）和-0.76（p < 0.05）。SCD与气温在四季均呈显著负相关（p < 0.05），但在青藏高原大型水体表面，由于受“湖雪效应”影响，秋季SCD与气温呈显著正相关（p < 0.05）；SCD与降水在春季、夏季与秋季呈显著负相关（p < 0.05），而在冬季呈显著正相关（p < 0.05）。

对MODIS逐日无云（SPSA）与8日合成（MOD10A2）两种积雪覆盖产品对比分析表明：从空间格局来看，相较于SPSA产品，MOD10A2产品总体上高估了SCD，尤其在青藏高原全局尺度上，相对高估偏差达56.34%。而在青藏高原南部的雅鲁藏布江流域、萨尔温江流域、湄公河流域等流域的低海拔地区（海拔 < 3600 ~ 4000 m），以及青藏高原的高海拔地区（海拔 > 6000 m），MOD10A2产品会低估SCD。从时间格局来看，MOD10A2产品对SCD在积雪稳定期（冬季）高估程度较小，积雪过渡期（秋季和春季）高估程度较大。从SCD年际变化趋势来看，基于MOD10A2产品计算得到的SCD趋势略高于SPSA产品，这种差异在秋季最为明显，在冬季最不明显。从SCD驱动因素来看，基于SPSA与MOD10A2产品计算得到的SCD与气温和降水之间的相关性存在流域差异，具体表现为内流区一致性较差，外流区一致性较好。

（3）基于MOD10A2和SPSA积雪覆盖产品的SRM融雪径流模拟精度差异分析表明：在融雪季中期，基于SPSA产品的径流模拟精度明显优于MOD10A2产品，径流量体积差异率（D）由17.83%降低至4.15%。

分别采用基于MOD10A2产品构建积雪消融曲线插值得到和基于SPSA产品直接计算得到的逐日积雪覆盖率作为融雪径流模型（SRM）的积雪覆盖率参数，对黑河流域上游地区2008年融雪季进行融雪径流建模，进而分析了积雪覆盖率数据差异可能导致的径流模拟结果差异。结果表明，在整个融雪季内，两种产品计算得到的积雪消融曲线总体上变化趋势较为相似，但SPSA产品直接计算得到的逐日积雪覆盖率曲线能够捕捉到更多积雪覆盖的时空变化细节特征。在整个融雪季，基于两种积雪覆盖产品模拟的径流量与实测径流量之间均具有较好的相关关系，两者的纳什效率系数（Nash-Sutcliffe efficiency coefficient, NSE）系数和R²较为接近，但基于SPSA产品模拟的径流量具有更高的模拟精度，体积差异率（D）由12.75%降低至8.75%。在融雪季的不同阶段，由于积雪覆盖率差异所导致的径流模拟精度差异有所不同，在融雪季初期和后期影响小，在融雪季中期影响较大。在融雪季中期，基于SPSA产品的径流模拟精度优于MOD10A2产品，径流量体积差异率（D）由17.83%降低至4.15%，精度提升明显。

Snow cover is one of the most important components of land cover types, and its inter-annual and seasonal variation can affect the radiation and energy balance, hydrological and biogeochemical cycles, and even the global energy balance and climate change. Among other widely-utilized snow cover assessment methods, Moderate resolution Imaging Spectrometer Spectroradiometer (MODIS) products have become one of the main data sources for ice and snow research due to their global coverage, high spatial and temporal resolutions. However, as an optical sensor, cloud occlusion in MODIS snow cover products, especially in MODIS daily product data, often leads to numerous data gaps, which hinders their promotion and adoption in environmental research. Therefore, the development of a new cloud removal algorithms of MODIS snow product and obtain accurate snow cover information with high spatial and temporal resolution has important theoretical value and practical significance for the patterns of spatial and temporal dynamics of snow cover in the context of climate change, modelling of snowmelt runoff and accurate estimation of runoff volume, and even water resource management.

The Tibetan Plateau (TP) is a stable snow cover region in Asia, and the temporal and spatial evolution of snow cover has important implications for the Asian monsoon and even for flooding in Southeast Asia. In this study, the TP is chosen as a typical test area, and to address the existing research problems, we propose a novel gap-filling method based on non-local spatiotemporal similar pixels and conditional probabilities to eliminate cloud occlusion in daily MODIS snow cover product, and the time series products of daily cloud-free snow cover in the TP from 2003 to 2018 are prepared. Based on that, the temporal and spatial patterns, trends, and drivers of snow cover were assessed in the TP, and the effect of snow cover differences on runoff simulation accuracy during the snowmelt season based on simulated results of the snowmelt runoff from the Snowmelt Runoff Model (SRM) was also analyzed in this study. The main conclusions of this study are as follows:

(1) A cloud removal algorithm framework based on spatial-temporal similar pixel interpolation is proposed, with deep coupling of time, space, abundance of Normalized Difference Snow Index (NDSI) and history law information of snow cover, the annual average cloud coverage is less than 2% after de-clouding, and the overall accuracy is close to 97%.

Terra and Aqua Combination (TAC), three-Day Temporal Filter (3DTF), and Similar Pixel Selecting Algorithm (SPSA) were included in the innovative cloud removal method. The experimental results show that TAC and 3DTF method could remove approximately 8% and 9% of the cloud gaps, respectively. The SPSA method provided good stability and applicability. The annual averages were as follows: Mean Absolute Error (MAE) was 2.77%, Mean Absolute Percentage Error (MAPE) was 42.40%, Root Mean Square Error (RMSE) was 3.78%, and R² was 0.78. The annual average Overall Accuracy (AO), Overestimation Error (OE), and Underestimation Error (UE) was 96.92, 1.10, and 1.98%, respectively. Meanwhile, the SPSA algorithm has good robustness, the de-clouding accuracy is not affected by cloud volume, even if the cloud-gap fraction exceeded 60% (3DTF product), the SPSA still performed better.

(2) Based on the SPSA and MOD10A2 products, the spatial and temporal distribution patterns, change trends, drivers of snow cover on the TP were evaluated during 2003-2018 hydrological period, and the similarities and differences between SPSA and MOD10A2 were also analyzed.

In the view of the spatial and temporal distribution pattern, the spatial distribution of snow cover on the TP had strong spatial heterogeneity, and the high values regions of snow cover days (SCD) were mainly concentrated on the edges and high mountains of the hinterland of the TP. The temporal distribution of snow cover on the TP is characterized by winter, winter-spring and autumn patterns. From the inter-annual trend, 57.4% of SCD on the TP showed a declining trend at respective rates of -5.45d?10a^-1 from 1982 to 2017 hydrological period. Seasonally, the greatest rate of decline was observed in winter and autumn, with values of -2.92 d?10a^-1 (p < 0.1) and -2.13 d?10a^-1. From different altitudes, the greatest rate of decline was occurred between 5000-6000m above sea level. In terms of SCD drivers, annual SCD on the TP was negatively correlated with temperature and precipitation, and the correlation coefficients was - 0.80 (p < 0.05) and -0.76 (p < 0.05), respectively. Seasonally, SCD in the TP was associated with temperature in all seasons. However, the notable positive correlations (p < 0.05) between autumn temperature and SCD on the surface of large water bodies of the TP due to the "lake snow effect". SCD was significantly correlated with precipitation in spring, summer and autumn (p< 0.05). In contrary, a significantly negative correlation (p < 0.05) between SCD and precipitation was found in winter.

The results of the comparative analysis between SPSA and MOD10A2 were as follows. From a spatial pattern perspective, the MOD10A2 product overall overestimated the SCD, and the relative overestimation error is 56.34% in the total TP. In contrast, MOD10A2 products will underestimate SCD in the southern part of the TP (such as Brahmaputra, Salween, and Mekong river basins) and high altitudes (> 6000m) of the TP. In terms of temporal trends, MOD10A2 products have lower errors of SCD during the snow stabilization period (winter). However, the higher errors of SCD were found in the snow transition period (spring and autumn). In terms of interannual trends in SCD, the SCD trend calculated based on the MOD10A2 product was slightly higher than the SPSA product, this difference was most pronounced in the autumn and least pronounced in the winter. The watershed differences in the correlations were exited between SCD and temperature and precipitation based on the SPSA and MOD10A2 product, and showed poor consistency in the inner flow zone and good consistency in the outer flow zone.

(3) The results of snowmelt runoff simulations of SRM based on the MOD10A2 and SPSA snow cover products showed that the runoff simulation accuracy of SPSA produce was significantly higher than that of the MOD10A2 product in the middle of the snowmelt season, and the runoff volume variance (D) was reduced from 17.83% to 4.15%.

The snowmelt runoff was simulated based on the daily snow cover come from the MOD10A2 and SPSA product for the upper Black River Basin in 2008 snowmelt season, and the possible differences in runoff simulation results due to differences in snow cover data was also analyzed. Results showed that the snow melt curves calculated for the two products have generally similar trends throughout the snowmelt season, but the daily snow cover curves calculated by the SPSA products capture more detailed features of the spatial and temporal variation of snow cover. Good correlations were observed between simulated and measured runoff based on the SPSA and MOD10A2 snow cover products. The results of Nash-Sutcliffe efficiency coefficient (NSE) and R² indicated that the simulates runoff based on the SPSA product had higher simulation accuracy, and the volume difference ratio (D) was reduced from 12.75% to 8.75%. There were differences in runoff simulation accuracy due to the snow cover varied during the different stages of the snowmelt season, with small effects at the beginning and end of the snowmelt season and larger effects at the middle of the snowmelt season. The runoff simulation accuracy based on the SPSA product was significantly higher than that of the MOD10A2 product, and the runoff volume variance (D) was reduced from 17.83% to 4.15% in the middle of the snowmelt season.