气溶胶指悬浮在空气中除云以外的颗粒物，来源于自然和人类活动。细模态气溶胶大多来自于人为排放，因此也常常被称作是人为源气溶胶，监测其时空变化对大气污染防治和人类健康都具有重要意义。我们能够通过细模态气溶胶光学厚度（fine-mode Aerosol Optical Depth, fAOD）来推断大气中PM_2.5的含量。一般情况下通过地面观测手段可以较为准确的获得近地面细模态气溶胶的质量浓度信息，但是难以获取大尺度空间范围内的细模态气溶胶的分布信息，而卫星遥感手段能够弥补这个不足，可以为地面的大气污染观测研究提供更多的参考性信息，因此迅速成为了大气污染立体观测研究的一种重要手段。

目前，搭载在Terra和Aqua卫星上的中分辨率成像光谱仪（MODIS）提供的气溶胶光学厚度（Aerosol Optical Depth, AOD）数据产品已被国内外研究学者大量应用于全球和区域气溶胶的时空分布和变化研究中，但是基于卫星遥感手段获取到的陆地细模态气溶胶产品一直存在很大的不确定性，不能准确反映近地面细颗粒物的状况。因此长期以来包括NASA在内都没有提供公认可靠的细模态气溶胶卫星产品。

本文首先将本研究组研发的细模态气溶胶卫星遥感反演算法——查找表-光谱退卷积算法(Look Up Table-Spectrum Deconvolution Algorithm,  LUT-SDA)应用于Terra卫星的MODIS C6.1版本Level 3日平均气溶胶光学厚度产品（MOD08_D3）和波长指数（?ngstr?m Exponent, AE）产品，并对该fAOD反演算法进行了改进和测试，之后利用改进后的LUT-SDA算法获得了一套更精确的全球陆地气溶胶细模态比例(Fine Mode Fraction, FMF)产品——LUT-SDA FMF产品，其精度相比MODIS官方发布的C5.2版本的FMF产品有了很大的提高，而且该产品在全球的覆盖范围相较MODIS产品也有了明显的改进。通过与地面AERONET FMF观测产品的对比验证可知，LUT-SDA FMF产品的整体RMSE值很小，约为0.21，在期望误差（Expected Error, EE）范围内的比例约为54.88%，与地面观测结果有良好的一致性。另外，LUT-SDA FMF产品显示出两个密度中心，一个中心值在0.65附近，另一个在0.9附近，分别对应了混合型气溶胶与细模态气溶胶的模态。从空间分布来看，FMF的高值主要集中在北美洲北部、墨西哥大部、南非、中欧、中国华南地区以及俄罗斯大部分地区，说明在这些地区大气气溶胶主要以细模态气溶胶为主；低值区位于北非的萨赫勒和苏丹地区以及中东部分地区和澳大利亚的北部地区，说明这些地区以粗模态气溶胶为主导。

在此基础上，我们利用MODIS FMF产品和LUT-SDA FMF产品，结合MODIS遥感C6.1版本的DT数据集、DB数据集和融合算法DTDB数据集，分别获得了基于MODIS FMF和基于LUT-SDA FMF的两套全球陆地细模态气溶胶光学厚度产品。将基于MODIS FMF的fAOD产品与基于LUT-SDA FMF的fAOD产品分别与地面AERONET观测的fAOD进行比对验证，结果显示，基于MODIS FMF的fAOD产品的验证精度明显低于LUT-SDA fAOD产品。MODIS fAOD产品不仅存在大量零值情况，而且在整体数值上存在严重低估，在中高纬地区还存在数据缺失现象，从空间分布来看还存在许多不合理的异常高值区。基于LUT-SDA FMF的fAOD产品不仅精度较高，几乎不存在零值，而且在空间覆盖率上也有了很大的改进，其空间分布特征也较为合理。但是LUT-SDA fAOD产品在亚洲的整体精度都比较低，而亚洲正是全球细模态气溶胶光学厚度最高的地区，是细模态气溶胶的主要分布区。因此，准确反映该地区的fAOD情况对本研究来说至关重要。

于是，基于LUT-SDA fAOD的DT、DB和DTDB产品，本论文采用广义相加模型(Generalized Additive Model, GAM)算法对这3套数据集进行改进融合，获得了一套改进的全球陆地细模态气溶胶光学厚度产品——Ensemble fAOD。将Ensemble fAOD产品与地面AERONET观测值进行对比验证发现，Ensemble fAOD产品不仅在整体精度上有了很大改进，其与地基AERONET测得的fAOD之间的均方根误差可减小至0.185，平均绝对误差可减小至0.104，相关系数可增加至0.80，并且有63.3%的fAOD处于期望误差范围之内。在空间精度上，融合了基于LUT-SDA FMF fAOD三套产品数据集的优点，使得其精度在亚洲地区得到了显著提高。与MODIS官方fAOD产品相比，本研究提出的Ensemble fAOD的产品质量和完整性都有明显提高。

最后，本论文根据Ensemble fAOD产品，分析了全球陆地细模态气溶胶光学厚度的空间分布特征、季节分布特征和变化规律。结果显示，全球fAOD的高值区主要位于中国东部、印度北部和非洲中部，次高值区是在亚马逊地区、俄罗斯和印度尼西亚等地。fAOD的低值区主要分布在美国西部、南美洲南部及东部沿海地区、澳大利亚、青藏高原地区及蒙古高原地区。全球fAOD的年变化特征：从2008-2012年全球fAOD波动上升，到2012年达到最高，随后逐年下降，但2015年出现了次极大值，到2016年再次下降。全球fAOD的季节变化特征较为明显，从各个季节的平均值来看，fAOD的最高值出现在夏季，春季和秋季次之，冬季的fAOD季节平均值最低。

Aerosols refer to particles other than clouds suspended in the air, which are derived from natural and human activities. Most fine particulate aerosols come from anthropogenic emissions, so they are often referred to as anthropogenic aerosols. Monitoring its temporal and spatial changes is of great significance to the prevention and control of air pollution and human health. Therefore, the study of fine mode aerosol particles is of great significance. We can estimate the concentration of PM_2.5 from the fine-mode aerosol optical depth (fAOD). Fine mode aerosol particles are mostly from man-made emissions, so they are often referred to as anthropogenic aerosols. By means of ground observation, the mass concentration information of fine modal aerosol particles near the ground can be obtained more accurately, but it is difficult to obtain the spatial distribution information of fine modal aerosol particles in a larger area. However, this deficiency can be made up by using satellite remote sensing. By means of satellite remote sensing, more reference information can be provided for the observation and studies of air pollution on the ground. Therefore, it quickly becomes an important means of three-dimensional observation of air pollution.

Currently, the aerosol optical depth (AOD) data provided by the Moderate Resolution Imaging Spectroradiometer (MODIS) on Terra and Aqua satellites has been widely used by domestic and international researchers. However, satellite-based fAOD products such as those from the MODIS are highly uncertain over land, which cannot accurately reflect the concentration of fine particles near the ground.

Therefore, this paper firstly tested and improved an improved fAOD inversion algorithm, which called Look Up Table-Spectrum Deconvolution Algorithm (LUT-SDA), by using the MODIS Collection 6.1 version Level 3 daily average aerosol optical thickness product (MOD08_D3) and the Angstrom Exponent (AE) product. And the improved LUT-SDA algorithm is used to obtain a set of fine mode fraction (FMF) products, whose accuracy is greatly improved compared with the official MODIS Collection 5.2 FMF products. Moreover, the global coverageof LUT-SDA FMF is also significantly improved compared to MODIS FMF. By compared with ground AERONET FMF observation products, the overall RMSE value of LUT-SDA FMF products is 0.21 and the proportion in the range of expected error (EE) is about 54.88%, which is in good consistency with the ground observation results. LUT-SDA FMF has two density centers, one around 0.65 and the other around 0.9, which correspond to the modes of mixed aerosol and fine-mode aerosol respectively. In terms of spatial distribution, the high value areas of FMF are mainly concentrated in the east and north of North America, Mexico, Europe, northeast Asia, South Africa and south China, indicating that the atmospheric aerosols in these areas are mainly fine mode aerosols. The low values are located in the Sahel and Sudan regions of north Africa and parts of the Middle East and northern Australia, indicating that these regions are dominated by coarse modal aerosols.

On this basis, this study used MODIS FMF product and LUT-SDA FMF product to obtain two sets of global fine mode aerosol optical thickness products overland based on MODIS FMF and LUT-SDA FMF by combining the MODIS C6.1 DT data set, DB data set and DTDB data set. By comparing and verifying the MODIS FMF-based fAOD products and LUT-SDA FMF-based fAODs with the fAOD ground-based observation from AERONET sites, the results show that the correlation between the MODIS FMF-based fAOD products and AERONET fAOD is not as good as that between LUT-SDA FMF-based fAOD products and AERONET fAODs. MODIS fAOD products not only have many zero values, but also its overall value is relatively low, and there are many unreasonable abnormal high value areas in its spatial distribution. LUT-SDA FMF-based fAOD products not only have high precision and no zero value, but also have great improvement in space coverage and reasonable spatial distribution characteristics. However, the overall accuracy of LUT-SDA fAOD products is relatively low in Asia, which is the region with the highest fine mode aerosol optical thickness in the world. Therefore, it is crucial to accurately reflect the fAOD value in this region.

Therefore, based on the DT, DB and DTDB datasets of LUT-SDA fAOD, this study adopted the synthetic generalized additive model algorithm to obtain a set of synthetic fine mode aerosol optical thickness product -- Ensemble fAOD. Compared with AERONET fAOD observations, Ensemble fAOD products achieved great improvement. Its root mean square error(RMSE) is 0.185, and the average absolute error(MAE) is 0.104, and the correlation coefficient increases to 0.80, with 63.3% of fAOD in estimation error envelope. Moreover, in terms of spatial accuracy, Ensemble fAOD improve the spatial accuracy in Asia by combination of the LUT-SDA fAODs’ advantages. Especially, when comparing with the MODIS FMF-based fAOD products in the accuracy, Ensemble fAODproducts were overwhelming superior.

Finally, based on Ensemble fAOD product, we analyzed the spatial distribution characteristics, seasonal distribution characteristics and temporal variation characteristics of the global fine mode aerosol optical thickness over land. The results show that the high value area of global fAOD is mainly located in eastern China, northern India and central Africa, and the next high value area is in the amazon region, Russia and Indonesia. The low value areas of fAOD are mainly distributed in the western United States, the southern and eastern coastal areas of South America, Australia, the Tibet plateau and the Mongolian plateau. As for the annual change characteristics, Ensemble fAOD increased from 2008 to 2012, reached the highest level in 2012, and then decreased year by year. The sub-maximum value appeared again in 2015, and decreased again in 2016. The seasonal variation of global fAOD is relatively obvious. The seasonal average fAOD value is the highest in summer, followed by spring and autumn, and the lowest in winter.