In order to retrieve the large-scale ground PM_2.5 concentration distribution in eastern China, a model was built using machine learning. The model used the top-of-atmosphere reflectance data of the Himawari-8 geostationary satellite in 2019 and the meteorological data of the European Center as the input data, and the ground PM_2.5 concentration was the output data. Validation results showed that the model had high accuracy on different time scales in eastern China. The ten-fold cross-validation of the model had a correlation coefficient of 0.82 and a root-mean-square error of 20.11 μg/m³ for hourly PM_2.5 concentration inversion. The hourly satellite-meteorological grid data throughout the year of 2019 were input to the model, and the corresponding PM_2.5 grid data for the eastern China obtained. Good results were achieved for the PM_2.5 grid data after analyzing the seasonal variation and spatial distribution of PM_2.5 concentration over eastern China.

为了得到中国东部地区大范围的地面PM_2.5浓度分布, 用机器学习方法建立一个模型, 用2019年葵花8号静止卫星大气顶层反射率数据和欧洲中心气象资料作为输入数据, 地面PM_2.5浓度作为输出数据。验证结果表明, 在不同时间尺度下, 该模型对中国东部地区均有较高的精度。对于小时PM_2.5的浓度反演, 该模型的十折交叉验证的相关系数为0.82, 均方根误差为20.11 μg/m³。将2019全年卫星?气象格点数据放入模型, 得到中国东部地区全年逐小时的PM_2.5格点数据, 利用该格点数据分析中国东部地区PM_2.5浓度的季节变化和空间分布, 取得良好的效果。