Leveraging Multi-sensor Data Fusion To Generate High Resolution AOD And PM_2.5 Concentration Maps

As an aggregate of suspended particulate matter in the air,atmospheric aerosols can not only change the atmospheric radiation balance,but also affect the regional climate.It silently damages human health through long-term accumulation,especially fine aerosols,such as PM_2.5.Aerosol Optical Depth(AOD),as an important indicator of aerosol loading,has attracted much attention for many years.With the help of satellite remote sensing technology to retrieve AOD on a global or regional scale,accurate estimation of PM_2.5 concentration has become an important task to quantify the spatiotemporal distribution of AOD and PM_2.5.However,due to the limitations of satellite platforms,sensors and inversion algorithms,the spatiotemporal resolution of current major AOD products is still relatively low.Meanwhile,for the impact of cloud,the AOD products often have a serious data gap problem,which also objectively limits the spatiotemporal coverage of predicted PM_2.5 concentration.Constrained by the deficiencies of above,it is hard to apply the existing air quality monitoring data for PM_2.5 exposure assessment in small scale areas such as cities.Therefore,how to effectively improve the spatiotemporal resolution and coverage of AOD and PM_2.5concentration under the premise of confirmed data accuracy is still a grand challenge.To answer the questions mentioned above,this study carries out scientific research on satellite AOD inversion,AOD missing information reconstruction based on multi-source data fusion technology,PM_2.5 concentration modeling and estimation based on deep learning technology.The main results and conclusions are as follows:(1)In order to effectively improve the spatiotemporal resolution and coverage of AOD,as well as generate reflectance data products with high spatial and temporal resolution,this study presents a synergistic integration of the data merging and fusion algorithms of MQQA and BME in dealing with reflectance data at both the top of atmosphere(TOA)and land surface for a comparative study.Emphasis has been placed on the distinctive performance between BME and MQQA-BME algorithms in the spatial domain and the MQQA-BME and STARFM in the temporal domain at both TOA and land surface levels.The results indicate that the BME and MQQA-BME outperform the MQQA in terms of the spatial coverage.Moreover,the MQQA-BME algorithm shows a higher prediction accuracy than STARFM at the blue band over the temporal domain.The results of this comparison will greatly empower the MQQA-BME to be used for finer aerosol optical depth predictions.(2)Based on MQQA-BME algorithm,reflectance values with different spatiotemporal resolutions at the top of atmosphere(TOA)and the land surface level were merged and fused.In order to further learn whether fused reflectance data can be applied to AOD inversion,this paper artificially adds noise to surface reflectance and apparent reflectance,and carries out sensitivity analysis.The results showed that the estimated coefficient of determination(R²)and the root?mean?square error of the reconstructed surface reflectance at blue band are 0.63 and 0.024;while,the apparent reflectance accuracy is 0.54,and 0.021.Compared with AERONET AOD,the R² of inversion AOD is 0.196,RMSE is 0.108.Results show that the error of the reconstructed reflectivity data by MQQA-BME is higher than the required accuracy of AOD inversion,which results in the high noise of high-resolution AOD.(3)This paper then uses MQQA-BME to carry out multi-source and multi-temporal AOD products fusion.It is expected to produce high spatiotemporal resolution AOD data by the complementarity of multi-source and multi-temporal AOD data.The proposed data fusion scheme successfully fuses the multi-sources AOD data from MERRA-2,GOES-16,and MAIAC,which are then further calibrated using AERONET data(ground truth).The results show that the integrated AOD product has a high accuracy,and the R² of cross-validation with ground-based observation data is0.69,RMSE?0.07,which is significantly higher than the accuracy of any single AOD product.(4)The fused high spatial-temporal resolution AOD data in the previous study can be used to estimate the ground PM_2.5 concentration for urban pollution monitoring through machine learning algorithms,Deep Belief Network(DBN).The PM_2.5 data have spatiotemporal autocorrelation in geostatistics and follow the Gaussian kernel distribution.Hence,we modified the autocorrelation model with Gaussian kernel function and integrate with DBN algorithm to estimate PM_2.5 concentration with high spatiotemporal resolution.The cross-validation results show that the R²=0.86 and RMSE=6.84?g m^-3 better than the original R²=0.67 and RMSE=10.46?g m^-3In this study,the data fusion method can provide algorithmic support for other remote sensing data products.The results in this research will greatly empower the data fusion technology in producing high spatiotemporal resolution data.The final high quality PM_2.5 concentration data can be used for urban air quality monitoring and related PM_2.5 exposure risk assessment in the future.