Evaluating different satellite-based Aerosol Optical Depth (AOD) in predicting inland daytime PM _2.5 using machine learning-based regression approach

Abstract. Aerosols play a critical role in the development of the boundary layer and build-up of air pollution in urban environments. Their presence in the atmosphere is quantified by Aerosol Optical Depth (AOD). Satellite sensors observe and retrieve AOD at varied spatial and temporal resolutions. In air quality monitoring, satellite-based AOD products are typically utilized to predict ground concentrations of fine particulate matter (PM2.5) through various modelling approaches. This study evaluates AOD products observed by Moderate Resolution Imaging Spectroradiometer (MODIS), Visible Infrared Imaging Radiometer Suite (VIIRS) and the Advanced Himawari Imager (AHI) of Himawari-8 in predicting inland daytime PM2.5 concentrations for test sites in Japan and South Korea. Prediction models were constructed using eXtreme Gradient Boosting (XGBoost) regression with input variables from observation datasets matched on ground PM2.5 station locations. In addition to AOD, twelve (12) predictor variables representing topographic and meteorological parameters were considered. Prediction results were evaluated using Kruskal-Wallis test and effect size analysis to compare the absolute error distributions across AOD products. Statistical results indicate that while models utilizing MODIS AOD generalize better to new data, the overall difference in prediction accuracy is statistically negligible. These findings suggest that no single AOD product is significantly superior in predicting ground-level PM2.5 concentrations, highlighting the potential of integrating AOD values from various sources. This work is essential for improving the accuracy of PM2.5 estimates and for supporting more effective mapping of urban air pollution.