【摘要】 为获取城市尺度空间连续准确的PM_2.5浓度,该文以北京市为研究区,充分考虑气溶胶光学厚度和气象因素的影响,利用2017年MODIS 3km AOD产品和ECMWF-ERA5气象再分析资料,结合空气质量监测站点的PM_2.5数据,分别基于随机森林、多元线性回归、支持向量机和神经网络方法反演近地面的PM_2.5浓度,结果表明:随机森林的反演精度最高、效果最好,其平均绝对误差（MAE）为12.08μg/m³、均方根误差（RMSE）为18.42μg/m³;对自变量进行重要性分析,气溶胶光学厚度对PM_2.5反演模型的影响最大,其次为相对湿度、边界层高度和2m气温。利用随机森林模型反演非采暖期（4-11月）的PM_2.5浓度,并进行时空分析,结果表明:PM_2.5月均浓度差异显著;空间分布呈现出中部、南部和东部较高,北部、东北部和西南部较低的特点;变化幅度呈现出中部和南部较大,西北部和西南部较小的分布特点。更多还原

【Abstract】 In order to obtain continuous and accurate spatial PM_2.5concentration,this paper takes Beijing as the study area to retrieve PM_2.5 concentration based on a variety of methods including the random forest（RF）,the multiple linear regression（MLR）,the support vector machine（SVM）and the neural network（NN）,using MODIS 3-km AOD products,ECMWF-ERA5meteorological reanalysis data and PM_2.5data at air quality monitoring stations.The results show that RF method has the highest inversion accuracy,whose mean absolute error（MAE）and root mean square error（RMSE）are 12.08μg/m³ and 18.42μg/m³,respectively.Based on the analysis of the significance of the variables,aerosol optical depth has the greatest impact on the model,followed by the relative humidity,the boundary layer height and the air temperature.The RF model was used to retrieve the PM_2.5concentration during the non-heating period（April-November）in 2017,and the spatial-temporal analysis shows that the monthly average concentrations of PM_2.5are significantly different.The spatial distribution shows higher PM_2.5concentration in central,southern and eastern areas,and lower in north,northeast and southwest areas.In addition,the change magnitude of PM_2.5concentration is larger in central and southern areas,smaller in northwest and southwest areas.更多还原