Satellite-derived long-term estimates of full-coverage PM1 concentrations across China based on a stacking decision tree model

Abstract

Fine particles with aerodynamic diameters less than 1 μm (PM1) often exert a greater threaten on human health, and thus it is highly imperative to accurately characterize the spatiotemporal variation of PM1 concentrations and to assess the potential health risks. Our study attempted to predict the long-term full-coverage PM1 concentrations across China during 2004–2018 using a stacking decision tree model based on satellite data, meteorological variables, and other geographical covariates. The result suggested that the stacking model captured strong prediction capability with a higher cross-validation (CV) R2 value (0.64), and the lower root-mean-square error (RMSE: 18.60 μg/m3) and mean absolute error (MAE: 11.96 μg/m3) compared with the individual model. The higher PM1 concentrations were mainly concentrated on North China Plain (NCP), Yangtze River Delta (YRD), and Sichuan Basin due to intensive anthropogenic activities and poor meteorological conditions especially in winter. The annual mean PM1 concentration in China exhibited a remarkable increase during 2004–2007 by 1.34 μg/m3/year (p < 0.05), followed by a gradual decrease during 2007–2018 by −1.61 μg/m3/year (p < 0.05). After 2013, the mean PM1 concentration at the national scale experienced a dramatic decrease by −2.96 μg/m3/year (p < 0.05). The persistent increase of PM1 concentration across China during 2004–2007 was mainly caused by the rapid increases of energy consumption and inefficient emission control measures, while the dramatic decrease since 2013 was attributed to increasingly strict control measures, particularly the implementation of the Air Pollution Prevention and Control Action Plan (the Action Plan). The long-term PM1 estimates obtained here provide a key scientific basis and data support for epidemiological research and air pollution mitigation.