Estimating historical SO2 level across the whole China during 1973-2014 using random forest model

Abstract

Ambient SO₂ pollution poses a great threat on air quality, human health, and ecosystem safety. The ground-level SO₂ monitoring sites over China have been established during the past years, while the long-term SO₂ data was still missing before 2014, which cannot reveal the evolution trend of SO₂ pollution and assess its response to the anthropogenic activity. In this work, we developed a high-quality random forest (RF) model to simulate the long-term SO₂ concentration across the entire China from 1973 to 2014, based on substantial explanatory variables (e.g., meteorological factors, SO₂ emission intensity, land use types). The 10-fold cross-validation R² value and root mean square error (RMSE) over China reached 0.64 and 17.06 μg/m³, respectively, both of which were significantly higher than those of other models such as back propagation neural network (BPNN) and generalized regression neutral network (GRNN). Among all of the predictors, T displayed the highest relative importance value, followed by WS, Prec, SO₂ emission intensity, RH, DOY, elevation, and the lower one for land use types and P. The estimated mean SO₂ concentration during 1973-2014 displayed the remarkably spatial variation with the higher value in North China Plain (NCP) and Middle part of Inner Mongolia. This historical SO₂ level estimation suggested that air pollution was not a new environmental issue that could be dated back to 1973. Overall, the annually mean SO₂ level for each grid increased from 29.46 ± 9.79 to 31.44 ± 8.77 μg/m³ from 1973 to 2014. The annually mean SO₂ concentration in NCP showed rapid increase from 34.32 ± 3.05 to 36.97 ± 3.18 μg/m³ during 1973-2002, whereas they decreased significantly after 2003 (from 37.46 ± 3.20 to 36.13 ± 3.48 μg/m³ during 2003-2014). The gradual decrease since 2003 was benefitted from the adjustment of the energy consumption structure and the adoption of emission control technologies. However, the SO₂ levels in some western regions showed the violent increases since 2003 due to the proposal of "development of the western region". The estimated daily SO₂ concentration across the entire China could provide the essential data for epidemiological research and air pollution prevention.

Keywords: China; Emission inventory; Random forest model; SO(2) level.