Abstract:Particulate matter with a diameter less than 2.5 microns (PM2.5) on the ground has a negative impact on human health and the economy. Most methods obtain PM2.5 from satellite derived aerosol optical depth (AOD) indirect products or daytime atmospheric top reflectance. The purpose of this paper is to directly use the advanced geosynchronous radiation imager (AGRI) infrared data from Fengyun-4B satellite and artificial intelligence models to retrieve PM2.5 with spatial and temporal resolution of 4 kilometers and 15 minutes in the Yangtze-Huaihe region in near-real time throughout the entire time period (including day and night). Firstly, explore the signal response of AGRI brightness temperature to different levels of PM2.5 in different seasons; Secondly, a study on AGRI brightness temperature retrieval of PM2.5 was conducted based on the random forest method in different seasons. The experimental results showed that the PM2.5 correlation coefficients obtained from the seasonal retrieval were all over 0.87; Finally, based on SHapley Additive exPlanations (SHAP), the model was interpretable and the contribution of geographic information to PM2.5 was found to be significant. And further explored the application of the products mentioned in this paper.