Using machine learning to quantify drivers of aerosol pollution trend in China from 2015 to 2022

Atmospheric aerosol pollution, such as fine particulate matter (PM2.5) and inhalable particulate matter (PM10), is one of the most important environmental problems in China. To mitigate particulate air pollution, the Chinese government has been implementing a series of clean air actions. Here, we conducted a comprehensive analysis on surface air pollutants data obtained from the China air quality observation network using a random forest (RF) approach, to evaluate the impact of clean air actions on aerosol pollution from 2015 to 2022. Overall, the observed PM2.5 and PM10 showed evidently decreasing trends during 2015-2022 in each season, with the largest trends in winter. Based on the RF model analysis, we further quantified the seasonal-dependent trends in PM2.5 and PM10 concentrations driven by anthropogenic emissions over China, which were approximately -3.84 and -6.14, -2.82 and -4.71, -2.58 and -4.45, and -2.77 and -4.06 mu g m- 3 yr- 1 for winter, spring, summer, and autumn, respectively. Furthermore, anthropogenic emissions were estimated to contribute 34.44 and 54.98, 24.10 and 37.22, 23.48 and 36.20, and 21.04 and 30.69 mu g m- 3 to declines in annual mean PM2.5 and PM10 concentrations during the eight years in megacity clusters of eastern China for winter, spring, summer, and autumn, respectively. These seasonal-dependent trend analyses reveal the largest reduction in ambient aerosol pollution due to anthropogenic emission abatements during cold season and further demonstrate substantial effectiveness of the clean air actions for improvement in particulate air quality in China. Correlation analysis on PM2.5 with the major meteorological parameters and the RF model built-in feature importance suggested that a combination of high relative humidity, shallow boundary layer heights, and low wind speeds could promote wintertime aerosol pollution in short-time scales over northern China (e.g., the North China plain).