【摘要】 针对近地面PM_2.5浓度空间分布估算问题，融合卫星气溶胶遥感数据以及气象、地形、植被等方面的辅助数据，利用随机森林模型实现了对2018—2020年中国东部地区（33°N～43°N,110°E～120°E）近地面PM_2.5质量浓度空间分布特征的分析。分析结果显示：（1）在10 km空间分辨率下，PM_2.5浓度的模型预估结果与地面站点实际观测结果吻合度高，其中决定系数（R²）为0.84,均方根误差（RMSE）为16.35μg/m³,平均绝对误差（MAE）为3.17μg/m³。（2）模型月度预估数据的整体拟合度优于日度预估数据，其中R²为0.87,RMSE为10.17μg/m³,MAE为2.54μg/m³。（3）研究区PM_2.5浓度分布存在明显的季节性空间差异，不同季节预估结果的R²排序为秋季（0.82）=冬季（0.82）>春季（0.80）>夏季（0.66）,其中冬季空气质量最差(PM_2.5浓度预估均值为70.50μg/m³),夏季空气质量最好(PM_2.5浓度预估均值为25.86μg/m³)。（4）2018—2020年，研究区PM_2.5浓度逐年下降，各年度预估均值分别为46.72、43.14、37.91μg/m³。研究结果表明，基于随机森林的PM_2.5浓度估算及时空分布研究方法可以有效融合卫星气溶胶遥感数据以及辅助数据，实现对不同时间尺度近地面PM_2.5浓度的高效预估，为近地面PM_2.5浓度时空分布特征准确识别提供新的手段。更多还原

【Abstract】 A study on the spatial distribution estimation of near-surface PM_2.5 concentration was conducted by integrating satellite aerosol remote sensing data, meteorological data, topography data and other auxiliary data.The random forest algorithm was used to construct the spatial distribution characteristics of near-surface PM_2.5 concentration in eastern China（33°N-43°N,110°E-120°E） from 2018 to 2020.The results showed that:（1） The estimated PM_2.5 concentration at a spatial resolution of 10 kilometers had high degree of coincidence with the actual observation results of ground stations with R² of 0.84,RMSE of 16.35 μg/m³ and MAE was 3.17 μg/m³.（2）The overall fitting degree of the monthly average data with R² of 0.87,RMSE of 10.17 μg/m³ and MAE of 2.54 μg/m³ was better than that of the daily average data.（3） PM_2.5 had obvious seasonal variation characteristics.The R² ranking of the estimated results for different seasons was autumn（0.82） = winter（0.82） > spring（0.8） > summer（0.66）.The PM_2.5 concentration was relatively low in summer（25.86 μg/m³） and reached an annual peak in winter（70.50 μg/m³）.（4） PM_2.5 concentration decreased year by year, from 46.72 μg/m³ in 2018 to 43.14 μg/m³ in 2019 to 37.91 μg/m³ in 2020.The results showed that by integrating satellite aerosol remote sensing data and auxiliary data, the random forest algorithm can be used to predict the PM_2.5 at different time scales, providing a new method to accurately identify the spatio-temporal distribution characteristics of near-surface PM_2.5 concentration.更多还原