地表反照率（Albedo）表征了地表对短波辐射能量的收支能力，是中长期天气预报和全球气候变化研究中的重要参数之一。地球系统科学及气候变化等研究对多种尺度、长时间序列、时空连续的高质量地表反照率产品有着迫切的应用需求。目前对地观测累积了大量卫星数据，可以支持二十世纪八十年代年至今的长时间序列遥感反照率产品生成。此外，近年来在轨运行的遥感卫星日益增多，对地观测信息空前丰富，为基于多源卫星数据的高时空分辨率反照率产品生成提供了条件。但是，现有基于多源卫星的反照率产品中还面临反演模型不统一、产品时间分辨率低、部分区域和时段存在产品缺失或产品质量低、不同产品的时间序列一致性差等诸多挑战。 时间序列多源卫星数据的反照率产品生成有两种方式：一是针对同时期的多源卫星数据构建联合模型直接联合反演，该方法可以获得高精度的二向反射（Bi-directional Reflectance Distribution Function, BRDF）参数和地表反照率，但对数据量和质量有较高要求，无法应用到早期卫星稀少的年代；另一种方式是对长时间序列非同期多源卫星数据使用同一系列的模型和算法分别反演，先分别生成反照率产品，再讨论它们之间的差异和一致性。为了提高长时间序列遥感反照率产品的质量，论文在反演算法、产品生产及精度评价分析方面开展工作，并取得了一定成果： （1）发展了多传感器联合反演模型（Multi-sensor Combined BRDF Inversion model, MCBI），可以支持高时空分辨率的BRDF/反照率反演，并且联合MODIS（Moderate Resolution Imaging Spectroradiometer）和VIIRS（Visible Infrared Imaging Radiometer Suite）传感器数据反演了5天和每日时间分辨率的BRDF/反照率产品。验证和分析结果表明：1）虽然MCBI产品的合成周期（10天）比国际上最常用的MCD43产品合成周期（16天）短，但是反演状态相近，且前者具有更多的完全反演；2）MCBI反演的BRDF参数相比MCD43的BRDF参数精度更高；3）站点验证的MCBI反照率相比MCD43反照率精度大致相当，在MCD43产品为备用算法但MCBI完全反演时，MCBI反照率具有更高的精度。MCBI反演模型已成为MuSyQ（Multi-source data Synergized Quantitative remote sensing production system）系统中反照率产品生成的主算法。 （2）引入机器学习方法对长时间序列GLASS（Global LAnd Surface Satellite Products）反照率产品生产流程中的直接估算算法和全球反照率背景场的生成进行了改进，支持完成了GLASS反照率产品的版本升级，并对算法和产品开展了全面的精度验证和评价。评价结果表明：1）机器学习方法的估算精度优于原直接估算算法中的多元线性回归方法；2）基于机器学习方法和雪水当量等观测数据对全球反照率背景场增强后，补偿了原背景场对北半球积雪地表反照率的低估；3）第4版GLASS反照率产品相比第3版本在整体精度上有所提升，且基于MODIS数据和基于AVHRR（Advanced Very High Resolution Radiometer）数据生产的两种GLASS子产品具有良好的一致性。 （3）利用长时间序列的GLASS反照率产品分析了中国地区反照率的空间分布特征和年际变化，并讨论了大尺度上反照率变化与雪水当量、植被指数和土地覆盖变化的相关性。考虑到大尺度上平均反照率变化为弱信号，为了消除时序分析中干扰因素的影响，研究探讨了反照率的相对修正方法，修正后的反照率时间序列与雪水当量等参数的相关更为明显。分析结果表明：1）我国北部地区反照率与雪水当量数据有较强相关，因此降雪是引起大尺度上反照率变化的主导因素；2）我国平均反照率在35年来有缓慢上升的趋势；3）基于GLASS-AVHRR反照率产品35年时间序列的分析可以得出其它较短时间序列产品所不能揭示的现象和结论。

Albedo indicates land surface energy budget, and it is also one of the key parameters of middle/long-term weather prediction and global climate change. Long-term, temporal spatial continuous, and high quality albedo products at various scales are required by earth system science and climate change researches. With the development of earth observation system, time series satellite observations from 1980s can support the production of long-term satellite albedo. Recent years, multiple satellites are available at the same time and provide diverse observations with more information. This can support the production of high quality, high temporal resolution and spatial resolution albedo. However, there are still many challenges, such as: no unified model, poor temporal resolution, gaps or low quality product, and inconsistence among different products. The study aims at the production of long-term, high quality albedo and is conducted in two way. On one hand, a union model is built for synchronously use multi-satellite observations and used to retrieve high quality albedo. This model can be applied to satellites at different times and retrieve consistent albedo product in long time series. The inversion of this model is based on the accumulation of effective and high quality observations in a period. However, early satellites provide sparse observations, which cannot support high quality inversion with this model. Thus, this method is recommended for the estimation of high quality albedo in recent years. On the other hand, the study retrieves albedo separately from different satellites but with similar model and assesses the consistence among them. To improve the long-term albedo quality, the study is conducted at three aspects: inversion method, product produce, and quality assessment. The main conclusions of this study is as follows: (1) The study develops a union model (Multi-sensor Combined BRDF Inversion model, MCBI) for synchronously use multiple satellites’ observations and produces a 5-day albedo and a daily albedo with MODIS (Moderate Resolution Imaging Spectroradiometer) and VIIRS (Visible Infrared Imaging Radiometer Suite) observations. The results show: 1) The inversion status of MCBI product (10-day accumulation period) is equivalent to that of MCD43 product (16-day accumulation period); 2) The accuracy of MCBI BRDF is higher than MCD43 BRDF; 3) The albedo accuracy of MCBI product is equivalent to that of MCD43 product, while MCBI albedo shows better accuracy than MCD43A1 albedo when MCBI under full inversion and MCD43A1 under magnitude inversion. MCBI is the main algorithm of MuSyQ (Multi-source data Synergized Quantitative remote sensing production system) albedo production. (2) For the produce of long-term GLASS albedo product, a machine learning method is applied to estimate GLASS albedo and enhance the global albedo background. Then the study assesses the quality and characteristic of GLASS albedo. The results show: 1) The accuracy of the machine learning method is higher than that of multiple linear regression method; 2) The enhanced background remedies the under-estimation of snow area at the North Hemisphere; 3) The version 4 products show better overall accuracy than the version 3 products, and they are high consistent. (3) The study analysis the distribution and time series albedo change of China, and discusses the relation between albedo and snow water equivalent (SWE), vegetation index, and land cover change separately. Furthermore, to remove the fake signal in albedo trend, the study explores a relative correction method of long-term albedo. The corrected albedo are more consistent with SWE. The results show: 1) The albedo of North China is of high correlation with SWE, thus snow is the major factor which influences albedo; 2) China averaged albedo increases slightly over 35 years; 3) The analysis based on 35 years GLASS-AVHRR albedo can reveal more and different conclusions compared to shorter time series products.