Random Forest Model for PM2.5 Concentration in China Using Himawari-8 Hourly AOD Product

Based on Himawari-8 hourly AOD product, meteorological variables, auxiliary data and PM2.5 observations, this study established a random forest (RF) model to estimate the PM2.5 concentration in China in 2018. The results showed that the estimated PM2.5 has high consistency with the ground observations. Ten-fold cross-validation (CV) result showed that the model has high accuracy, with R2 of 0.801, RMSE of $15.677\mu\mathrm{g}\cdot\mathrm{m}^{-3}$ In areas with sparse sites, there are fewer ground observation data, resulting in fewer training samples and poor model accuracy. During the day, the RF model performed best around noon due to the meteorological observation conditions. Seasonally, the RF model has the highest fit goodness in autumn and the lowest in summer. In addition, the model has the lowest RMSE in summer. Result of the model prediction showed that the spatial distribution of PM2.5 prediction and PM2.5 observation are relatively consistent. The high values in PM2.5 are mainly concentrated in Jiangsu, Shandong, Henan and Anhui, southeast of Beijing-Tianjin-Hebei and east of Sichuan Province.