Review of satellite-driven statistical models [formula omitted] concentration estimation with comprehensive information

Particulate matter pollution is increasingly serious due to the acceleration of urbanization, which has an adverse effect on human health. In order to offset that the nonuniform spatial distribution of PM2.5 and the short of long-term observational data from ground monitoring, optical products based on satellite retrieval, such as aerosol optical depth (AOD), had been the main data to estimate PM2.5 concentration. This paper summaries the principle and progress of statistical models for PM2.5 concentration estimation in recent decades. Specifically, we first reviews the characteristics and gap-filling methods of satellite-driven AOD products. Secondly, the auxiliary variables used to enhance the explanatory of PM2.5 changes are introduced and their impact on fine particles is analyzed. Third and most important, we summarizes the statistical models of PM2.5 concentration estimation, and discusses the model performance of each subtype separately (i.e., regression-based, machine learning-based and hybrid model). According to the summary and discussion of the above work, there are still problems and challenges that need to be paid attention and further improved. Finally, this paper also provides some feasible improvements about quality control of PM2.5 measurements, maximize the role of PM2.5 measurements in the model and the trend analysis of fine time scale PM2.5.