Bias corrections for speciated and source-resolved PM2.5 chemical transport model simulations using a geographically weighted regression

Abstract

The ability to provide speciated and source-resolved PM2.5 estimates make chemical transport models a potentially valuable tool for exposure assessments. However, epidemiological studies often require unbiased estimates, which can be challenging for chemical transport models. We use geographically weighted regression to predict and correct the bias in source-resolved PM2.5 species (elemental carbon, organic aerosol, ammonium, nitrate, and sulfate) across the continental U.S. for 2001 and 2010. The regression models are trained using speciated ground-level monitors from the CSN and IMPROVE networks. A 10-fold cross-validation shows minimal bias across all simulated PM2.5 species (0 – 3%) and improved agreement with ground-level monitors (R2 = 0.53 – 0.97). Corrections also improve the agreement between simulated and observed species mixtures on a fractional basis. The source-resolved exposure estimates developed in this study are suitable for use in health analyses of PM2.5 toxicity.