Estimation of 1-Degree-Resolution PM2.5 Concentrations Across China Over 1980–2019 Based on a Machine Learning Approach

Abstract:

The harmful health impacts and lack of long-term observations of PM_2.5 have emphasized the demands to estimate historical PM_2.5 concentrations in China. Many previous studies have predicted PM_2.5 mainly using aerosol optical depth (AOD) retrieved by satellites. However, they were also limited by the short temporal coverage of satellites and the partial representation of surface PM_2.5. In this study, we create the near-surface PM_2.5 concentration data across China over 1980–2019­ using the space-time random forest (STRF) model with atmospheric visibility and other auxiliary data. The modeled daily PM_2.5 concentrations were in excellent agreement with ground observations, with the coefficient of determination (R²) of 0.95, mean absolute error (MAE) of 4.84 μg/m³, root mean square error (RMSE) of 8.61 μg/m³ and mean relative error (MRE) of 12%. From 1980 to 2014, the predicted PM_2.5 concentrations had increased constantly with the maximum growth rate of 10–15 μg/m³/decade over eastern China. Due to the clean air actions, air quality has greatly improved. PM_2.5 concentrations have decreased effectively at a rate over 50 μg/m³/decade in the North China Plain and 20–50 μg/m³/decade over other regions of eastern China during 2014–2019. The newly generated 1-degree gridded PM_2.5 concentrations over 1980–2019 across China provide a useful tool for investigating environmental and regional climate impacts related to aerosols.