Spatiotemporal Continuous Estimates of PM2.5 Concentrations in China, 2000-2016: A Machine Learning Method with Inputs from Satellites, Chemical Transport Model, and Ground Observations

Abstract

Background: Exposure to ambient PM2.5 is known to harm public health in China. Satellite aerosol optical depth (AOD) were statistically associated with in-situ observations after 2013 to predict PM2.5, while the lack of monitoring data before 2013 created difficulties in historical estimates. Hindcast approaches using chemical transport models (CTM) can overcome this limitation, but still suffer incomplete coverage due to missing AOD or limited accuracy due to uncertainties of CTM.Objects: We attempted to produce historical PM2.5 estimates with complete spatiotemporal coverage and improved accuracy for exposure assessments in short- and long-term.Methods: First, we designed a machine learning (ML) model, which linked in-situ PM2.5 with high-dimensional expansion of numerous predictors (AOD, CTM outputs and etc.). To interpolate the missing predictions due to incomplete AOD, we incorporated another generalized additive model in next stage.Results: Cross-validations show that ML estimates were highly correlated with in-situ PM2.5, with R2 of 0.61, 0.68, and 0.75 for daily, monthly and annual averages, respectively. The two-stage estimates sacrificed accuracy on daily timescale (R2=0.55), but achieved complete spatiotemporal coverage and improved the accuracy of monthly (R2=0.71) and annual (R2=0.77) averages. The model was used to predict daily PM2.5 across China during 2000-2016 and estimate long-term trends for the period. The population-weighted PM2.5 significantly increased, by 2.10 (95% confidence interval [CI]: 1.74, 2.46) μg/m3/year during 2000–2007, and rapidly decreased by 4.51 (3.12, 5.90) μg/m3/year during 2013–2016.Conclusions: The data products could support large-scale epidemiological studies and risk assessments of PM2.5 in China.