| In recent years,the regional composite air pollution problem become particularly outstanding in China,especially in the Beijing–Tianjin–Hebei region.The influence of fine particulate matter(PM2.5)plays a prominent part of the problem,and has attracted wide attention of scholars and public in the world.Due to the limited PM2.5 ground-level monitors that mainly concentrate in urban area,estimation of ground-level PM2.5 concentrations using satellite aerosol products become an effective method to solve the spatial imbalance of PM2.5 observations for lack of monitors in suburb and rural area.On the one hand,it can extend spatial coverage and fill in the gaps of the sparse ground-level monitors especially in suburb and rural area,on the other hand,it will correct the bias resulting from PM2.5 ground-level observation when this observation data is applied to represent the average pollution level of the region including urban,suburb and rural area.Meanwhile,how the aerosol products,statistical model construction effect the relationship between aerosol optical depth?AOD?and PM2.5 is an issue that urgently needs to further research.In this work,we focus on Beijing–Tianjin–Hebei region.Firstly,using ground-based aerosol observations,air quality monitoring data,satellite remote data,ERA-interim reanalysis data and land used information,the difference between MERSI AOD and MODIS AOD retrieved by Dark Target?DT?algorithm are compared.And the regional and seasonal characters of the MODIS C6AOD products retrieved by each algorithm or each resolution are also discussed.Secondly,the performance of AOD-PM2.5 linear correlation and the improvement after planetary boundary layer height?PBLH?and relative humidity?RH?adjustments are systematically investigated.Thirdly,the regional correction scheme for aerosol products is proposed by combining the advantages of all AOD products,and the seasonal mixed effects model is presented to improve the applicability of estimating PM2.5.5 spatial distribution within the study domain,by calibrating day-specific AOD-PM2.5 relationship,incorporating PBLH and RH adjustments,meteorological variables and land use information.As a result,the ability to estimating fine particle concentration in the study area is enhanced.Lastly,the bias is explored when the monitoring value of PM2.5 is used to present that average pollution level in the whole district.We expect that results will help to assess the regional pollution level more reasonably,and to control pollution more effectively.The main conclusion are described as follows:?1?The regional and seasonal applicability of AOD productsThe characteristic of AOD distribution and seasonal variation is generally consistent for FY-3A/MERSI and Terra/MODIS AOD products retrieved by DT algorithm.Both products have better performance in autumn,but less availability in winter.On the whole,MODIS AOD products were more consistent with the ground-based observation than MERSI for various underlying surfaces?the correlation coefficient between MODIS and ground-based observation is 0.530.98,while it is only 00.93 for MERSI?.The percentages falling within the NASA expected errors?EE?for MODIS products is higher than that for MERSI?except Beijing City?.The error of MODIS AOD products from the ground-based AOD is mostly contributed by overestimation in urban and suburb,while it is mainly from underestimation in forest and cropland.This implies that error belongs to systematic bias for MODIS.The error of MERSI AOD is mainly contributed by underestimation in suburb and cropland,while it is both from overestimation and underestimation in the rest of area.The data dispersion degree of MERSI collocations from regression line is more obvious than MODIS.In short,MODIS AOD products have better performance than MERSI.The retrieval accuracy of MODIS DT-10km,DT-3km,Deep Blue?DB?and DB/DT“merged”?MD?products refined in C6 reprocessing show distinctive regionality and seasonality.In most cases,the percentage falling within EE for the four products is largest in autumn.And we prefer to choose DB product due to the poor collocations number of DT products in winter.The best performing product in cities vary with geographical location and aerosol sources.However,we can draw similar conclusion from all suburban sites that DB product has advantages in spring and winter,and DT-3km is better in summer and autumn.Additionally,for the forest land,the best product is DT-3km all the year except DB in winter.In a word,none of products consistently outperforms the others,although a product may shows better performance for some area or in some season.Therefore,we should choose a suitable product according to some category?such as region or season?,which will benefit to reflecting the variation of aerosol optical feature more effectively in the whole area.?2?Model performance of AOD-PM2.5 linear regression,and the impact of PBLH,RH adjustmentsThe air quality monitoring data exhibits a similar seasonal variability within the study area,but a considerable regional divergence of PM2.5 concentration and the proportion of fine particles in particulate matter.When a simple linear regression model is used to obtain the AOD-PM2.5relationship,the correlation of MODIS AOD and the ground-level PM2.5 value is poor and changes obviously with AOD products,seasons,and monitors?R2 for the highest correlation in each monitor range from 0.080.63 in spring,0.170.60 in summer,0.150.75 in autumn and 0.130.57 in winter?.In most cases,the correlation of AOD-PM2.5.5 will improve after PBLH and RH adjustments.But the improvement degree also depends on AOD products,seasons,and monitors?R2 for the highest correlation in each monitor range from 0.090.78 in spring,0.110.76 in summer,0.380.76 in autumn and 0.100.75 in winter?.These imply the relations of AOD and PM2.5 is influenced by some parameters varying temporally,such as the regional suitability of AOD products,aerosol vertical profiles,fine particle compositions,and pollution diffusion conditions in a season.So it is insufficient if we estimate PM2.5 concentration in the whole region only from a single AOD product by AOD-PM2.5 linear regression or by PBLH,RH adjustments.?3?Combination of AOD products,development and application of the seasonal mixed effects model.Based the results above,we divide the study domain into some subregions,and merge the AOD products according to seasons and subregions for obtaining more reliable data.Then a regional correction scheme that combines the advantages of all AOD products is built for PM2.5 estimation.Furthermore,a mixed effects model is used to account for the daily variability in the AOD-PM2.5relationship,and incorporating PBLH,RH adjustments,meteorological variables and land use parameters to improve the accuracy of the model.At last,mixed effects models in four season are established respectively to estimate day-to-day PM2.5 concentration.Then the regional distribution of PM2.5.5 levels for selected days is presented.Out-of-sample“ten-fold”cross validation?CV?method was employed to access the robustness of the model.The results show that the final constructed models enhance our ability to estimate PM2.5 concentration using new merged AOD products.CV R2 between the estimated and measure PM2.5 value is 0.71 in spring,0.70 in summer,0.81 in autumn,0.77 in winter respectively.Additionally,comparing with single AOD product,the regional AOD correction scheme not only extends spatial and temporal coverage,but also captures the small scale PM2.5 variation within cities,transitional zone from urban to suburb,forest land and so on.It shows similar spatial distribution trend,and has good consistence between model estimated PM2.5 mass concentrations and the actual observations during different pollution events.However,there still exists some shortage when the monitoring value of PM2.5 is used to present the average pollution level in the whole district and indicate spatiotemporal variation,since the monitors is so limited and distributes unevenly.The average regional pollution level is often overestimated in Northerm Hebei Province,such as ZJK and CD area,where the air is clean.While the difference between the monitoring value and the regional average level retrieved from satellite AOD products is related to pollution level in the rest of the study area.The average regional pollution level is always overestimated if the air in urban area is very clean(PM2.5?10?g/m3),and it will be underestimated in most case of low PM2.5 concentration(10?g/m3<PM2.5?40?g/m3).But the monitoring value of PM2.5 tend to overestimate the regional level when PM2.5 concentration in urban area further increases,especially in severe pollution events. |
PDF Full Text Request