Research On High Spatial Resolution AOD Inversion And PM_2.5 Estimation

In recent years,air pollution has become an important factor affecting people's health and regional sustainable development.Particalate Matter 2.5(PM_2.5)is an important indicator for measuring the quality of the atmospheric environment.Continuous spatial data of PM_2.5 concentration is the data basis for realizing refined air quality analysis and applications.Satellite remote sensing technology has the advantages of wide space coverage and low acquisition cost.Using Aerosol Optical Depth?AOD?remote sensing observation data for large-scale PM_2.5 concentration monitoring has become a research hotspot.However,current AOD products often have low spatial resolution,which is difficult to meet the demand for fine air quality monitoring in high heterogeneous urban environments.In this study,Beijing is selected as the research area,and Landsat8 remote sensing data is used to carry out high spatial resolution AOD inversion and PM_2.5 estimation,the specific research content is as follows:?1?High-resolution AOD inversion algorithm for Landsat dataAiming at the problem that the spatial resolution of current AOD products is too low,Simplified Aerosol Retrieval Algorithm?SARA?was applied to Landsat data to carry out AOD inversion research.The paper solved two key problems in AOD inversion: abnormal pixel completion and surface reflectance dataset construction.In order to reduce the data redundancy caused by too high resolution,the local spatial variance method was used to determine the optimal AOD spatial resolution is 60 m in Beijing.Results indicated that the retrieved AODs with 60 meters resolution have good spatial consistency with MODIS aerosol products,and have higher spatial coverage as well as resolution.The retrieved AODs showed a high consistency with ground-based AOD measurements,with average correlation coefficient?R?= 0.9682,root mean square error?RMSE?= 0.0517,mean absolute error?MAE?= 0.0342 and expected error?EE?= 92.8%,which is better than MOD04 products?R = 0.9814,RMSE = 0.0961,MAE=0.0858,EE=35.14%?.?2?PM_2.5 estimation model based on physical mechanisms and machine learning algorithmsConsidering the widely used multiple linear regression model cannot fit the sub-linear relationship between PM_2.5 and AOD,this paper builds models which combine the heighthumidity correction with machine learning algorithms represented by Random Forest and Extreme Gradient Boosting.Cross-validation and multiple indicators?R2?RMSE?MAE?were used to evaluate the models.The results showed that the height-humidity corrected Random Forest model has the best fitting and predictive ability,and can be used as a reasonable PM_2.5 estimation model.?3?Analysis of distribution characteristics and influence factors of PM_2.5 concentration in BeijingBased on high resolution PM_2.5 concentration estimated by height-humidity corrected Random Forest model,the spatial distribution and autocorrelation characteristics of PM_2.5 in Beijing were analyzed.Besides,this paper identifies and analyzes PM_2.5 pollution hotspots,and uses density analysis,spatial statistics,buffer analysis and other methods to study the effects of land cover type,road density and factory distribution on PM_2.5 concentration.This study improves the accuracy,spatial resolution,and coverage of PM_2.5 concentration,and explores the spatial distribution and influence factors of PM_2.5 concentration in Beijing.The outcomes of this study can promote small-scale PM_2.5 concentration monitoring and management,and is useful in many related fields,such as land use planning,transportation planning,industrial site planning,etc.