Abstract: Primary gaseous pollutants such as carbon monoxide (CO), sulfur dioxide (SO2) and nitrogen oxides (NOx) are important targets for controlling sources of air pollution. Satellite remote sensing can achieve large-scale concentration monitoring of these pollutants, which serves as a significant supplement to ground-based station monitoring. Based on the observation data including primary gaseous pollutants, aerosol optical depth, meteorological and other auxiliary data from Shanxi Province's national control stations, Lüliang City's micro monitoring stations, Sentinel-5P/Tropomi satellite and MODIS, this work conducts a study on the estimation and mapping of the concentrations of these primary gaseous pollutants (CO, NO2, SO2) with a spatial resolution of 0.01° in Lüliang City, China. Firstly, the DINEOF method is employed to reconstruct the missing data of satellite remote sensing, and then the eXtreme Gradient Boosting (XGBoost) method is utilized for concentration estimation. The finding reveals a strong correlation between the estimated concentrations and the station observation data of primary gaseous emissions, and the inversion results effectively illustrate the variations in the distribution of primary gaseous emissions across different regions within the city. It is indicated tht this approach compensates for the limitations posed by the sparse distribution of national control stations and enhances the precision of urban air quality control.

Key words: primary pollutant emission, machine learning, satellite remote sensing, concentration inversion