[Review of Machine Learning in Air Pollution Research].

Air pollution is one of the most serious global challenges at present， and it has great practical importance to study and improve air quality. Traditional research methods mostly rely on multi-source observations and numerical models constructed based on atmospheric physics and chemistry theories， although these methods are limited in terms of accuracy， spatial and temporal coverage， and computational resources. As a powerful data processing and information mining tool， machine learning has begun to be applied by researchers in the field of air pollution research， aiming to reveal the changing patterns and influencing factors of air pollution through analyzing large amounts of data and predict future trends in air quality. This study reviews the typical applications of machine learning in air pollution research in recent years， mainly involving the following four aspects： inverting and estimation， monitoring， and prediction of atmospheric composition based on satellite remote sensing； improvement of air quality simulation and forecast accuracy； analysis of air pollution causes； and fusion of multi-source data. In addition， the scientific problems and technical difficulties in the current research are further discussed. Future research should focus on how to combine machine learning with traditional numerical models， such as developing intelligent parameterization schemes and learning model parameters. The application of machine learning in pollution source analysis， air quality health impact assessment， and multi-source data fusion techniques should also be explored to achieve more accurate air quality management and policy making.