Estimating the daily PM2.5 concentration in the Beijing-Tianjin-Hebei region using a random forest model with a 0.01° × 0.01° spatial resolution

Abstract

High spatiotemporal resolution fine particulate matter (PM2.5) simulations can provide important exposure data for the assessment of long-term and short-term health effects. Satellite-based aerosol optical depth (AOD) data, meteorological data, and topographic data have become key variables for PM2.5 estimation. In this study, a random forest model was developed and used to estimate the highest resolution (0.01° × 0.01°) daily PM2.5 concentrations in the Beijing-Tianjin-Hebei region. Our model had a suitable performance (cv-R2 = 0.83 and test-R2 = 0.86). The regional test-R2 value in southern Beijing-Tianjin-Hebei was higher than that in northern Beijing-Tianjin-Hebei. The model performance was excellent at medium to high PM2.5 concentrations. Our study considered meteorological lag effects and found that the boundary layer height of the one-day lag had the most important contribution to the model. AOD and elevation factors were also important factors in the modeling process. High spatiotemporal resolution PM2.5 concentrations in 2010–2016 were estimated using a random forest model, which was based on PM2.5 measurements from 2013 to 2016. Keywords: PM2.5 estimation, Human exposure, High spatiotemporal resolution, Machine learning