Remote Sensing Of The Regional Composite Model Of PM2.5 In The Pearl River Delta

PM2.5 pollution has attracted public attention in recent years.Remote sensing technology is an emerging technology with rapid development.The information network formed by remote sensing has a wide coverage and high resolution,and it can continuously provide people with a large amount of scientific data and dynamic information.Using satellite remote sensing to estimate the concentration of PM2.5 is extremely dependent on monitoring data.Therefore,most studies of PM2.5 remote sensing estimation models are at the site scale,and it is difficult to reflect the mutual coupling laws between PM2.5 and AOD on the regional scale,which is not conducive to the model.In view of the accuracy and space-time applicability,we selected the MODIS aerosol products and the meteorological data from December 2014 to November 2015 in the PRD region(boundary layer height,relative humidity,average temperature,and wind speed,atmospheric pressure,precipitation,and radiation intensity of sunlight).Based on the verification of MODIS AOD inversion accuracy and analysis of the relationship among PM2.5,AOD,and meteorological data,we used statistical regression method,single factor optimization method,and optimal subset method to build up the composite model of PM2.5 sites in the PRD region.Based on the composite model of the site,a regional composite model was constructed.The simulation accuracy of the model was evaluated.The conclusions are following:(1)Verification of the inversion accuracy of MODIS AOD data.Using linear regression analysis to perform the relationship between AERONET AOD and MODIS AOD data.The coefficient of determination reached 0.870,and 80%of the inversion data was within the error range.The MODIS AOD inversion results are highly reliable and can be used for modeling research.(2)Verification of the correlation among PM2.5,AOD and meteorological data on annual and seasonal scales.PM2.5 and AOD have similar distribution characteristics in temporal and spatial distribution.The correlation coefficient of the PM2.5-AOD dataset is broader than the full-scale scale in the seasonal scale,and the standard deviation is higher.It is more suitable for the study of revealing the interrelationships between PM2.5 and AOD.There is a linear correlation between meteorological factors and PM2.5,which can be used for the construction of multi-factor regression model.(3)A satellite composite site model for estimating PM2.5 in the Pearl River Delta was constructed.Using R2,Pearson correlation coefficient,RMSE,and MAPE parameter accuracy indices,compare the original regression model(OM)constructed by PM2.5-AOD,boundary layer height,and/or humidity factor corrected AOD and PM2.5 modified regression models(TM)and the fitting accuracy of the multi-factor regression model(MVM)built with meteorological factors,resulting in a site composite model.Through two stages of screening,the average fitting adjustment R2 of the site composite model CM in spring,summer,autumn,and winter was 59.804%,62.078%,43.043%,and 46.377%,which compared with the original model OM was an increase of 37.708%,39.087%,25.289%,22.903%.Compared with the optimal TM model increased by 36.528%,26.573%,28.571%,33.004%.Using meteorological data to modify the model and select the optimal site model can effectively improve the accuracy of the site model.(4)A satellite composite remote sensing model for estimating PM2.5 in the Pearl River Delta region was constructed.Based on the site optimization model,Kriging space interpolation technique,spatial clustering method,and unsupervised classification method were used in turn to successfully construct the regional compound model.On the seasonal scale,the Pearl River Delta region was divided into 12(Spring),5(Summer),15(Autumn),and 15(Winter)sub-regions based on the differences in the types of PM2.5 regional composite models.The mean values of the precision parameters for spring,summer,autumn,and winter were:R2:0.619,0.602,0.498,0.623,respectively,MAPE was 27.310%?20.307%?28.144%?19.730%,RMSE was 15.759[ig/m3?12.030?g/m3?15.795?g/m3?13.132?g/m3 have excellent fitting accuracy.Using reserve samples to evaluate the prediction effect of regional composite models.The PM2.5 predicted value and the real value showed a good consistency at each season scale.In terms of correlation,the correlation coefficients ranged from 0.421 to 0.708 in each season.In terms of model stability,the range of MAPE was 10.286%-43.780%and RMSE was 4.670-21.260?g/m3 in each season.In summary,the composite model of the Pearl River Delta region has excellent prediction results and high practical value.