MERRA-2 PM_2.5 mass concentration reconstruction in China mainland based on LightGBM machine learning

MERRA-2 developed by the National Aeronautics and Space Administration (NASA) provides the long-term record of surface PM_2.5 mass concentration since 1980s, but needs great improvement over mainland China according to recent studies. In this study, a newly developed light gradient boosting machine (LGBM) model is introduced to correct the MERRA-2 PM_2.5 record over mainland China by incorporating the meteorological reanalysis and satellite AOD retrievals. A 40-year surface PM_2.5 record covering mainland China is reconstructed from 1980 to 2019, providing a new dataset for exploring the interactions between climate variability and air pollution. The new record exhibits not only much better magnitude but also more excellent variabilities of surface PM_2.5 loading compared to original MERRA-2 products. The correlation coefficient, the root-mean-square error and the mean error between the observed and reconstructed records are 0.8, less than 28.5 μg·m^-3, and 0.33 μg·m^-3, respectively, which are much better than those of 0.27, 45.8 μg·m^-3, and 1.64 μg·m^-3 between the observed and MERRA-2 PM_2.5 records. The PM_2.5 record with longer term and higher accuracy developed in this study provides a better base for the research on the climate change variability and air pollution in mainland China. However, limitations of the reconstructed record still exist, especially in the Tibetan Plateau and marine regions with very sparse surface measurements, which need further correction in the future studies.