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

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Ambient sulfur dioxide (SO2) pollution significantly impacts air quality, human health, and ecosystems.
• Existing ground-level monitoring data in China is limited before 2014, hindering the analysis of long-term SO2 pollution trends and responses to human activities.

Observation:

• A high-quality random forest (RF) model was developed to simulate SO2 concentrations across China from 1973 to 2014.
• The model utilized meteorological factors, SO2 emission intensity, and land use types as explanatory variables.
• The RF model achieved a 10-fold cross-validation R² of 0.64, outperforming other models like BPNN and GRNN.

Findings:

• Temperature (T), wind speed (WS), precipitation (Prec), SO2 emission intensity, and relative humidity (RH) were key predictors.
• Estimated SO2 concentrations showed significant spatial variation, with higher levels in the North China Plain (NCP) and Inner Mongolia.
• Overall annual mean SO2 increased from 29.46 ± 9.79 μg/m³ in 1973 to 31.44 ± 8.77 μg/m³ in 2014.

Implications:

• SO2 pollution in China is a long-standing issue, dating back to at least 1973.
• SO2 levels in NCP increased until 2002, then decreased due to emission controls, while western regions saw increases post-2003.
• The simulated long-term SO2 data provides essential information for epidemiological studies and air pollution prevention strategies.