Research On MAIAC AOD Filling And Interpolation In Beijing-Tianjin-Hebei Region

With the acceleration of China’s urbanization and industrialization,heavy chemical industry and other high energy consuming industries as well as the consumption of daily living resources inevitably emit a large number of polluting gases,among which haze pollution caused by aerosol particles such as PM_2.5（fine particulate matter with an aerodynamic diameter of 2.5μm or less）has become the most important atmospheric environmental problem in Beijing-Tianjin-Hebei.it is of urgent and important significance to accurately estimate the long-term and short-term exposure of PM_2.5in heavily polluted areas such as Beijing-Tianjin-Hebei.Although China’s PM_2.5monitoring network has been operational since 2013,these are limited by sparsity and uneven distribution.Aerosol optical depth（AOD）data observed by moderate resolution imaging spectrometer（MODIS）and other satellite instruments are widely used to estimate surface PM_2.5concentration because of their wide spatial coverage and repeated observations of the earth’s surface and atmosphere.However,due to the cloud/snow/water cover,high surface reflectivity and extremely high aerosol load,satellite remote sensing AOD data also have the disadvantages of non-random missing.The results show that the non-random loss of aerosol optical thickness may lead to the deviation of PM_2.5exposure assessment.Therefore,it is necessary to explore corresponding methods to fill the lack of AOD data in order to improve the coverage and accuracy of PM_2.5prediction.Based on the MAIAC AOD data of Beijing-Tianjin-Hebei region from 2016 to 2018 and the PM_2.5mass concentration monitoring data of ground monitoring network,Spatiotemporal linear mixed effects（STLME）model and a nested model with different time scales are established to fill the AOD data of the grid where the missing situs are located.On the basis of AOD data fusion,10%of the sample points are randomly selected,and the optimal spatiotemporal variation function model is selected for spatiotemporal kriging interpolation to get a Beijing-Tianjin-Hebei fully covered AOD data set.The temporal and spatial coverage before and after interpolation and the spatiotemporal distribution of AOD after spatiotemporal kriging interpolation are analyzed.The main results are as follows:（1）The estimation accuracy of STLME model is higher than that of linear mixed effects（LME）model（model fitting R²increased by 0.057,0.062 and 0.074 respectively,and decreased by 0.034,0.039 and 0.042 respectively）.Compared with linear regression and multiple filling methods,STLME model has better filling performance,the CV R²is 0.860,0.888,0.881,respectively.CV RMSE and CV RPE do not change much.After filling,the spatial valid value ratio of monitoring stations in the study area is increased from 41.19%～57.36%to 81.15%～89.04%.（2）The estimation accuracy of the nested model（Spatiotemporal nested linear mixed effect model,STNLME）is similar to that of the non-nested model（Spatiotemporal linear mixed effects model,STLME）on the date with AOD-PM_2.5matchups.When there is no match between PM_2.5and AOD throughout the day,the prediction accuracy is low（Day-of-Year based CV is 0.361,0.315 and 0.376,respectively）.STNLME takes into account the weekly random effect and monthly random effect of the whole modeling,which can significantly improve the performance of the model without PM_2.5and AOD matchups（Day-of-Year based CV is 0.685,0.710,0.690,respectively）.After replenishing the value,the average annual AOD spatial valid value ratio of the site in the study area is increased to 95.11%～99.82%,and the temporal valid value ratio is increased from 85.29%～90.18 to 100%.（3）The spatiotemporal kriging interpolation method has better performance than the ordinary kriging method.The R²after spatiotemporal kriging cross-validation is higher than the ordinary kriging method,the MAE is lower by 0.0629 percentage points,and the RPE is0.2517 percentage points lower than the ordinary kriging method.After ordinary kriging interpolation,the average spatial coverage is 87.13%,and the spatiotemporal kriging is increased to 99.79%.And ensure the availability of AOD data for 366 days in 2016.（4）The AOD has obvious spatiotemporal variation after spatiotemporal kriging interpolation.In terms of temporal variation,AOD is the highest in summer,followed by autumn,and lower AOD is in spring or winter.In terms of spatial change,it shows the distribution pattern of high AOD value in the southeast of Beijing-Tianjin-Hebei and AOD value in the northwest.The inland plain in the south of the study area,including Shijiazhuang,Xingtai,Handan,Cangzhou and Hengshui,has a higher AOD,while the northwest of Beijing and Qinhuangdao,the mountains of Chengde and Zhangjiakou,and the northern part of Qinhuangdao have lower AOD.（5）After the filling of the STLME model and spatiotemporal kriging interpolation,the average annual AOD of Beijing-Tianjin-Hebei has been increased,indicating that a large part of the high-value AOD has been filled.This can effectively reduce the problem of high underestimation when PM_2.5is predicted by AOD,and is of great significance to reduce the deviation in PM_2.5health impact assessment.