风云四号A星（FY-4A）是中国第二代静止气象卫星的首颗星，多通道扫描成像辐射计AGRI（Advanced Geosynchronous Radiation Imager）是搭载在FY-4A上的主要光学载荷之一。AGRI具有高频率观测特点（每天观测205次），在大气气溶胶的遥感高频监测方面具有良好应用潜力，但目前官方还未发布相应的气溶胶数据集。本文旨在针对AGRI数据的特点开发基于地表反射率比值库的反演算法以生产高精度的AGRI气溶胶数据集。本文首先基于再分析数据对去云后的AGRI L1级数据进行气体吸收订正；然后利用背景气溶胶光学厚度AOD（Aerosol Optical Depth）对一个月内的“次暗像元”进行大气校正以获取AGRI 0.65 μm 和 0.83 μm 通道的地表反射率，进而获取这两个通道的地表反射率的比值，完成每个月的地表反射率比值库的构建；最后便可以基于已构建的地表反射率比值库实现地气解耦，完成气溶胶的遥感反演。该算法已被应用于2019年5—10月京津冀地区的气溶胶反演，AGRI AOD反演结果与美国国家航天局发布的MODIS（Moderate-resolution Imaging Spectroradiometer）AOD数据集、日本气象厅发布的AHI（Advanced Himawari Imager）AOD数据集的对比结果显示它们具有基本一致的空间分布趋势。利用AERONET（Aerosol Robotic Network）数据验证的结果显示AGRI AOD数据集具有较高的精度，且其精度要优于AHI AOD数据集和MODIS AOD数据集。AGRI AOD数据集的平均绝对误差，均方根误差，与地基数据的相关系数和误差落在期望误差±(0.05+0.15AOD)AERONET范围内的反演结果所占的比例分别是0.09%，0.12%，0.91%和65.86%。上述验证结果表明基于地表反射率比值库的方法反演AGRI AOD具有可行性，且反演结果具有较高的精度。

FY-4A, as the latest generation of China’s geostationary meteorological satellite, has been launched on December 11, 2016. The Advanced Geosynchronous Radiation Imager (AGRI) is the primary payload onboard FY-4A, and it can image China and its surrounding areas 205 times per day. The AGRI with high-frequency observation capabilities can provide sufficient data support for aerosol monitoring, but few aerosol products are developed using AGRI. The accuracy of the available FY-4A AOD datasets is also inferior to that of the Himawari-8 aerosol product, which can also cover China. In this study, an aerosol optical depth (AOD) retrieval algorithm based on the database of ratio of surface reflectance of different channels is proposed to develop high-accuracy FY-4A AOD dataset.This algorithm involves four steps: (1) detect and remove cloud pixels in FY-4A L1 data; (2) perform gas absorption correction on FY-4A L1 data using the reanalysis data released by European Center for Medium-Range Weather Forecasts; (3) select the sub-dark pixels for each month from FY-4A data after gas absorption correction, and perform atmospheric correction on these sub-dark pixels using the background AOD (i.e., AOD at 550 nm is 0.02) to obtain the surface reflectance of VIS06 and NIR08 channels, obtain the ratio of surface reflectance of these two channels, and perform the abovementioned operations for all the pixels in the study area to complete the construction of surface reflectance ratio database of VIS06 and NIR08 channels; (4) retrieve AOD using FY-4A L1 data after gas absorption correction based on the constructed surface reflectance ratio database.The algorithm has been applied to aerosol retrieval over Beijing–Tianjin–Hebei region from May 2019 to October 2019. Comparison of FY-4A AOD retrieval results with MODIS AOD dataset released by NASA shows that the two AOD datasets have a consistent spatial distribution trend. Validation result of MODIS AOD dataset, the official Himawari-8 AOD dataset released by the Japan Meteorological Agency, and FY-4A AOD dataset against ground-based AOD data provided by Aerosol Robotic Network shows that the accuracy of FY-4A AOD dataset is better than that of Himawari-8 AOD dataset and MODIS AOD data. The root mean square error, mean absolute error, correlation coefficient with ground-based data, and percentage of retrieval results with error within ±(0.05+0.15AOD_AERONET) of FY-4A AOD dataset are 0.12, 0.09, 0.91, and 65.86%, respectively.Although the signal-to-noise ratio of FY-4A/AGRI is lower than that of Himawari-8/AHI, the absolute and relative errors of FY-4A AOD dataset are better than those of Himawari-8 AOD dataset. The statistical parameters of FY-4A AOD dataset are also slightly better than those of MODIS AOD dataset, which is one of the widely used AOD datasets with high accuracy. Therefore, the FY-4A AOD dataset developed in this study has high accuracy.