基于MCD19-A2数据和GWR模型的2011-2020年中国大气PM_(2.5)质量浓度反演

Inversion of atmospheric PM_(2.5) mass concentration in China from 2011 to 2020 using MCD19-A2 data and GWR model

为探究MODIS高时空分辨率气溶胶产品数据在长时间序列下对于反演中国陆地PM_(2.5)质量浓度的适用性和准确性。该研究基于MCD19-A2数据研究2011-2020年中国陆地气溶胶光学厚度(aerosol optical depth,AOD)的时空分布特征,以降水、风速等8个气象要素为辅助变量建立反演PM_(2.5)的地理加权回归模型(geographically weighted regression,GWR)并分析中国陆地PM_(2.5)的空间分布。结果表明:1)2011-2020年中国陆地气溶胶时空分布基本符合“西低东高、逐年下降”的规律且10 a间AOD值存在较大季节差异,春季(0.294)>夏季(0.262)>冬季(0.223)>秋季(0.194)。2)利用方差膨胀系数(variance expansion coefficient,VIF)对变量进行多重共线性检验,建立并分析2011-2020年GWR模型,发现建模集决定系数均大于0.760,验证集决定系数均大于0.740,且均方根误差均小于7.070μg/m3,模型拟合效果良好。3)将GWR模型预测的PM_(2.5)浓度值分别通过样条函数插值法、反距离加权插值法、克里金插值法和自然邻近插值法进行空间插值,发现4种插值方法决定系数均大于0.910,均方根误差均小于7.030μg/m3,插值结果十分可靠,并与国家地球系统科学数据中心所提供的“1km高质量中国PM_(2.5)分布”数据一致。该研究表明结合MCD19-A2数据与GWR模型反演PM_(2.5)浓度具有较好的适用性。

Temporal and spatial transformation characteristics of China’s land-based atmospheric aerosol were obtained from the high spatial resolution MODIS MCD19-A2 data from 2011 to 2020 using remote sensing and geographically weighted regression model(GWR).This article was adopted:1)Vertical-humidity revision:The vertical revision of aerosol optical depth(AOD)was carried out under the different spatial distributions of satellite and PM 2.5 concentration data.The AOD value was obtained from the MODIS,and then divided by the atmospheric boundary layer height data,in order to obtain the AOD value close to the ground level.The humidity was revised to remove the influence of relative humidity on the AOD value.The vertically revised“dry”AOD value was divided to obtain a vertical-humidity revised“wet”AOD by the humidity impact factor,according to the calculated humidity impact factor in each city.There was an increase from 0.365 to 0.779 in the correlation coefficient between the“wet”AOD value and PM 2.5 concentration data.2)Multiple colinear tests:multiple colinear tests were carried out to verify the GWR fitting using the eight variables of the model.Variance inflation factor(VIF)was selected to test the multiple colinear all over the eight variables for the better fitting of the GWR model.3)GWR model was established to inspect the accuracy:The processed PM_(2.5) and auxiliary variables data of Chinese cities were selected to establish 10 annual GWR models and the fitting of each model using the Geographically Weighted Regression function in Modeling Spatial Relationships tool of ArcGIS.SPSS software was used to verify the accuracy of the fitting in each model,with the adjusted coefficient of determination(R2)and root mean square error(RMSE).Conclusions were drawn as follows:1)The spatial and temporal distribution of aerosol was basically conformed to the“low in the west and high in the east,decreasing year by year”.There was an outstanding seasonal difference in the AOD value,where the highest value was 0.294 in spring,followed by 0.262 in summer,and the lower values were 0.194 and 0.223 in autumn and winter,respectively.2)The VIF demonstrated that the strongest multicollinearity was observed in 2018,indicating the monthly scale GWR model of each city.The VIF variables were close to 1,which fully met the requirements of the GWR model.3)The better fitting of the model was achieved,where the best year was 2013(R2=0.933),and the worst year was 2018(R2=0.761).There was a basically consistent trend in the spatiotemporal distribution of PM_(2.5) concentration and AOD in each year,indicating the high applicability of MCD19-A2 data in the GWR model.4)In terms of PM 2.5 visualization,the distribution of PM 2.5 was inverted by the Spline,Inverse Distance Weighted,Kriging and Natural Neighbor interpolation in 2013.The spatial distribution and concentration range were basically consistent,compared with the“Distribution of 2013 mean PM_(2.5) concentrations in China”data provided by National Earth System Science Data Center.IDW and Natural Neighbor more accurately described the high-value areas of PM_(2.5) concentration around the Tarim Basin in Xinjiang and the Beijing-Tianjin-Hebei region.