Satellite-Based Estimates of Daily NO2 Exposure in China Using Hybrid Random Forest and Spatiotemporal Kriging Model

A novel model named random-forest-spatiotemporal-kriging (RF-STK) was developed to estimate the daily ambient NO2 concentrations across China during 2013-2016 based on the satellite retrievals and geographic covariates. The RF-STK model showed good prediction performance, with cross-validation R-2 = 0.62 (RMSE = 13.3 g/m(3)) for daily and R-2 = 0.73 (RMSE = 6.5 g/m(3)) for spatial predictions. The nationwide population-weighted multiyear average of NO2 was predicted to be 30.9 +/- 11.7 mu g/m(3) (mean +/- standard deviation), with a slowly but significantly decreasing trend at a rate of -0.88 +/- 0.38 mu g/m(3)/year. Among the main economic zones of China, the Pearl River Delta showed the fastest decreasing rate of -1.37 mu g/m(3)/year, while the Beijing-Tianjin Metro did not show a temporal trend (P = 0.32). The population-weighted NO2 was predicted to be the highest in North China (40.3 +/- 10.3 mu g/m(3)) and lowest in Southwest China (24.9 +/- 9.4 mu g/m(3)). Approximately 25% of the population lived in nonattainment areas with annual-average NO2 > 40 g/m(3). A piecewise linear function with an abrupt point around 100 people/km(2) characterized the relationship between the population density and the NO2, indicating a threshold of aggravated NO2 pollution due to urbanization. Leveraging the ground-level NO2 observations, this study fills the gap of statistically modeling nationwide NO2 in China, and provides essential data for epidemiological research and air quality management.