China's "coal-to-gas" policy had large impact on PM1.0 distribution during 2016–2019.

Particulate matter with an aerodynamic diameter of less than 1 μm (PM 1.0) can be extremely hazardous to human health, so it is imperative to accurately estimate the spatial and temporal distribution of PM 1.0 and analyze the impact of related policies on it. In this study, a stacking generalization model was trained based on aerosol optical depth (AOD) data from satellite observations, combined with related data affecting aerosol concentration such as meteorological data and geographic data. Using this model, the PM 1.0 concentration distribution in China during 2016–2019 was estimated, and verified by comparison with ground-based stations. The coefficient of determination (R2) of the model is 0.94, and the root-mean-square error (RMSE) is 8.49 μg/m3, mean absolute error (MAE) is 4.10 μg/m3, proving that the model has a very high performance. Based on the model, this study analyzed the PM 1.0 concentration changes during the heating period (November and December) in the regions where the "coal-to-gas" policy was implemented in China, and found that the proposed "coal-to-gas" policy did reduce the PM 1.0 concentration in the implemented regions. However, the lack of natural gas due to the unreasonable deployment of the policy in the early stage caused the increase of PM 1.0 concentration. This study can provide a reference for the next step of urban air pollution policy development. • Use stacking machine learning algorithm and satellite data to reconstruct PM 1.0. • "Coal-to-gas" policy led to PM 1.0 concentration changes, notably in 2017 and 2019. • Provides valuable insights for mitigate air pollution in Chinese cities. [ABSTRACT FROM AUTHOR]