Representativeness of the US EPA PM monitoring site locations to the US population: implications for air pollution prediction modeling

Air pollution prediction modeling establishes relationships between measurements and geographical and meteorological characteristics to infer concentrations at locations without measurements. Since air pollution monitors are limited in number, predictions may be generated for locations different than those used to train the model. The epidemiologic impacts of this potential mismatch hinge on whether the population resides in areas well-represented by monitoring sites. Here we quantify the fraction of the population with geographical characteristics not reflected by the 2000, 2010, and 2020 EPA PM2.5 and PM10 regulatory sites. We evaluated this measure nationwide, regionally, and by race. Nationally, the networks were very representative of the population experience; however, there was less overlap regionally and in regions stratified by race. This suggests that sub-national exposure modeling should carefully consider the representativeness of monitors for their populations. It also highlights that exposure models often borrow information from distal places to predict full population exposure.