National NO2 exposure models for measuring its impact on vulnerable people in the US metropolitan areas.

Epidemiological research requires accurate prediction of the concentrations of air pollutants. In this study, satellite-based estimates (OMI NO2), distance-weighted models (DWMs), and universal kriging (UK) are applied to land use regression (LUR) to predict annually and monthly averaged NO2 concentrations in the continental United States. In addition, to assess environmental risk, the relationship between NO2 concentrations and people potentially exposed to NO2 within urban areas is explored in 377 metropolitan statistical areas (MSAs). The results of this study show that the application of a combination of OMI NO2, UK, and DWMs to LUR yielded the highest cross-validated (CV) R2 values and the lowest root mean square error of prediction (RMSEP): 82.9% and 0.392 on a square root scale of ppb in the annual model and 70.4-83.5% and 0.408-0.518 on square root scale of ppb in the monthly models, respectively. Moreover, the model presented a spatially unbiased distribution of CV error terms. Models based on LUR provided more accurate NO2 predictions with lower RMSEP in urban areas than in rural areas. In addition, this study finds that the people living in the urban areas of MSAs, with larger populations and a higher percentage of children under 18 years of age, are likely to be exposed to higher NO2 concentrations. By contrast, people living in the urban areas of MSAs with a higher percentage of the elderly over 65 years of age are likely to be exposed to lower NO2 concentrations.