Spatiotemporal patterns of PM₁₀ concentrations over China during 2005-2016: A satellite-based estimation using the...

Area of Science:

• Environmental Science
• Atmospheric Science
• Public Health

Background:

• Limited historical national-scale PM10 exposure data in China using satellite-based Aerosol Optical Depth (AOD).
• Lack of investigation into long-term PM10 trends in China.

Purpose of the Study:

• Estimate daily PM10 concentrations across China for the past 12 years (2005-2016).
• Utilize ground monitoring data, AOD, land use, and weather data with a machine learning approach.
• Investigate long-term PM10 trends and spatial variations.

Main Methods:

• Collected daily PM10 measurements from 1479 sites in China (2014-2016).
• Integrated Moderate Resolution Imaging Spectroradiometer (MODIS) AOD, land use, and meteorological data.
• Developed and compared a random forests model against traditional regression models for PM10 estimation at ~10 km resolution.

Main Results:

• Random forests model explained 78% of daily PM10 variability (RMSE=31.5 μg/m³), improving to 82% for monthly and 81% for annual averages.
• The random forests model demonstrated superior predictive ability and lower bias compared to regression models.
• Estimated PM10 levels exceeded national standards in one-third of China, with highest concentrations in Xinjiang and Beijing-Tianjin, and lowest in Tibet, Yunnan, and Hainan.
• PM10 levels peaked nationally in 2006-2007 and showed a declining trend from 2008 onwards.

Conclusions:

• First study to estimate historical PM10 pollution in China using satellite-based AOD and a random forests model.
• The developed model provides valuable data for assessing long-term health impacts of PM10 exposure in China.