Estimating hourly surface PM2.5 concentrations with full spatiotemporal coverage in China using Himawari-8/9 AOD and a two-stage model

PM 2.5 (fine particulate matter with an aerodynamic diameter of less than 2 . 5 μ m ) is a significant air pollutant , posing serious risks to both the atmospheric environment and human health. Satellite remote sensing Aerosol Optical Depth (AOD) data are often used to estimate surface PM 2.5 concentrations. However, satellite-derived AOD data are often affected by large-scale data gaps due to cloud contamination and high surface albedo , leading to discontinuities and incompleteness in surface PM 2.5 estimations based on AOD. PM 2.5 is influenced by natural and human activities, both of which show strong diurnal variations. Many previous studies have used AOD data from sun-synchronous orbiting satellites, whose coarser temporal resolution makes it difficult to capture these diurnal PM 2.5 variations. In this study, AOD products from the new generation of geostationary meteorological satellites , Himawari-8/9, are employed to estimate spatiotemporally continuous hourly seamless PM 2.5 grid data using a two-stage Random Forest (RF) model. This model integrates meteorological, surface, and demographic-economic factors. In the first stage, the RF model was used to fill the gaps in the satellite AOD data, achieving a good fit ( R 2 = 0 . 95 ), with a root mean square error (RMSE) and mean absolute error (MAE) of 0.05 and 0.03, respectively. In the second stage, the model estimates surface PM 2.5 grid data (5 km × 5 km) at hourly intervals during the daytime, based on the gap-filled AOD data, actual PM 2.5 measurements from ground stations, and auxiliary data. The final hourly PM 2.5 estimates were well-fitted to the ground station measurements ( R 2 = 0 . 92 ), with RMSE and MAE values of 7.14 and 4 . 90 μ g /m 3 , respectively. This study provides a valuable approach for estimating complete, hourly-level spatial and temporal distributions of PM 2.5 from incomplete satellite remote sensing AOD data, which is crucial for air quality management and assessing short-term exposure risks. • Himawari-8/9 AOD data are effectively gap-filled using auxiliary and spatiotemporal information. • Estimated PM 2.5 aligns well with observations ( R 2 = 0 . 92 , RMSE = 7 . 14 μ g /m 3 ). • The two-stage model enables seamless hourly gridded PM 2.5 concentration in space and time.