Research progress, challenges, and prospects of PM2.5 concentration estimation using satellite data

Abstract

Satellite data are vital for understanding the large-scale spatial distribution of particulate matter (PM2.5) due to their low cost, wide coverage, and all-weather capability. Estimation of PM2.5 using satellite aerosol optical depth (AOD) products is a popular method. In this paper, we review the PM2.5 estimation process based on satellite AOD data in terms of data sources (i.e., inversion algorithms, data sets, and interpolation methods), estimation models (i.e., statistical regression, chemical transport models, machine learning, and combinatorial analysis), and modeling validation (i.e., four types of cross-validation (CV) methods). We found that the accuracy of time-based CV is lower than others. We found significant differences in modeling accuracy between different seasons ( p < 0.01) and different spatial resolutions ( p < 0.01). We explain these phenomena in this article. Finally, we summarize the research process, present challenges, and future directions in this field. We opine that low-cost mobile devices combined with transfer learning or hybrid modeling offer research opportunities in areas with limited PM2.5 monitoring stations and historical PM2.5 estimation. These methods can be a good choice for air pollution estimation in developing countries. The purpose of this study is to provide a basic framework for future researchers to conduct relevant research, enabling them to understand current research progress and future research directions.