基于GF-4卫星的长三角城市群PM2.5遥感反演

基于静止卫星高分四号（GF-4）遥感数据，利用6SV辐射传输模型与暗目标算法进行高空间分辨率气溶胶光学厚度（AOD）遥感反演；在此基础上，结合地面监测站大气细颗粒物（PM_2.5）浓度、气象资料等数据，采用物理订正方法及线性混合效应模型，实现长三角城市群区域大尺度空间连续的PM_2.5浓度遥感反演；最后利用十折交叉验证法对反演精度进行验证.结果表明：GF-4反演的AOD结果分辨率较高，空间连续性好，与AERONET地基监测相关性R达到0.82；利用GF-4 AOD的PM_2.5估算模型精度较高，模型估算PM_2.5浓度与地面实测数据拟合度R²为0.74；在分春夏秋冬4个季节建模情景下，交叉验证R²依次为0.67，0.59，0.63和0.72，平均绝对误差MAE为10.40，7.42，10.10，13.34μg/m³，表明GF-4卫星适用于区域PM_2.5浓度监测.

Remote estimation of PM2.5 based on GaoFen-4 satellite data in the Yangtze River Delta urban agglomeration

This paper used 6 SV model and dark target algorithm to retrieve AOD with a high spatial resolution based on GaoFen-4(GF-4) geostationary satellite data. Afterwards, combined with the PM_2.5 concentration data of the ground air quality observation sites, meteorological factors and other data, physical correction methods and linear mixed effects(LME) model were used to monitor the large-scale and spatial continuous PM_2.5 concentration in the Yangtze River Delta urban agglomeration(YRDUA). The results showed that the retrieved GF-4 AOD has good spatial resolution and spatial continuity, and the correlation coefficient(R) with AERONET ground-based monitoring data reached 0.82. The LME model based on GF-4 AOD showed a good agreement between the estimated PM_2.5 concentration and the in situ observed values(R²=0.74). The 10-fold cross-validation R² of spring, summer,autumn and winter were 0.67, 0.59, 0.64 and 0.72, respectively; and the mean absolute error(MAE) were 10.40, 7.42, 10.10 and 13.34μg/m³, respectively, which indicates that GF-4 can be used for regional PM_2.5 concentration monitoring.