Satellite-driven prediction of fine particulate matter (PM 2.5 ) concentrations: machine learning and explainable artificial intelligence

This study provides a systematic review of recent trends in forecasting fine particulate matter (PM2.5) concentrations, which are important for managing air quality in regions where there is limited ground monitoring. This study reviews literature from the years 2015 to 2025 and examines the combination of satellite data with machine learning (ML) and deep learning (DL) techniques. The study reviews the use of various satellites, including low Earth orbit and geostationary satellites, for obtaining aerosol optical depth (AOD) and top-of-atmosphere reflectance and Ångström exponent. The review highlights the importance of meteorological, land use, demographic, and infrastructure datasets in reducing prediction bias. Significant improvements to model stability are attributed to advancements

in data fusion and preprocessing, especially with respect to data normalization and missing values imputation for AOD. Performance comparisons among models show that traditional ML models such as Random Forest provide strong baseline performance while modern DL models such as hybrid DL models and transformer-based models provide better performance by capturing complex spatiotemporal relationships. This study also examines the growing importance of explainable artificial intelligence (XAI) in promoting the acceptability of PM2.5 prediction models.

Message: Hi FE, I need help with “Satellite-driven prediction of fine particulate matter (PM 2.5 ) concentrations: machine learning and explainable artificial intelligence” in “Environmental Engineering”