A Meta-Analysis and Systematic Review of Integrating Satellite-Derived Aerosol Optical Depth Data with Machine Learning for Estimating Fine Particulate Matter (PM2.5) Concentrations

Exposure to fine particulate matter (PM₂.₅) significantly impacts public health, particularly in regions where annual average levels of PM₂.₅ exceed the World Health Organization (WHO) guidelines. According to the literature, in Iran, elevated fine particulate matter levels contribute substantially to mortality among adults. The spatial coverage limitations and intermittent data gaps of ground PM₂.₅ monitoring stations pose challenges for effective air quality management.The products of remote sensing technologies, such as Aerosol Optical Depth (AOD) from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensors, offer a promising alternative for fine particulate matter estimation. This study reviews previous research on using machine learning algorithms to predict PM₂.₅ ground concentrations based on AOD data. A structured analysis of 127 selected studies reveals varying correlations between AOD and PM₂.₅ (with the resultant coefficient of determination, R², between ground PM₂.₅ concentrations and AOD data ranging from 48 to 99%), influenced by auxiliary variables like meteorological conditions and environmental factors.Integrating these variables enhances prediction accuracy, though it may increase complexity and potential errors in machine learning models. The hybrid machine learning models demonstrate superior performance compared to individual algorithms, leveraging their adaptability, parallel processing capabilities, and ability to handle missing data. Despite advancements, challenges persist due to data uncertainty and meteorological dynamics.In conclusion, while machine learning offers robust tools for PM₂.₅ forecasting using AOD data, ongoing research is essential to address existing limitations and optimize model performance amidst environmental variability.