Use of Fine-Resolution AOD-Derived Long-Term PM2.5 Concentrations to Characterize Adverse Health Outcomes

Abstract

Remote sensing data from satellites, which can be combined with land use, meteorology, and chemical transport simulations, and trained at monitoring locations, have some distinct advantages over monitoring data in assessing health effects of air pollution. Compared to monitoring data, they can provide fine scale spatial resolution and access to populations not living near monitors; compared to land use regression they can provide fine scale temporal resolution; and compared to inexpensive monitors they can provide better accuracy and historical estimates going back over a decade. I discuss several such models, their performance, and demonstrate how they perform in different studies compared to monitors. In the VIVA birth cohort, an IQR increase in modeled PM2.5 had a stronger association between Impaired Glucose Tolerance (OR=1.64 (1.11, 2.42)) compared to monitored PM2.5 (OR=1.34 (0.98, 1.84)). In the Medicare Mortality cohort they provide somewhat stronger associations between mortality and PM2.5, but much stronger associations with ozone HR= 1.011 (1.010–1.012) Vs 1.001 (1.000–1.002) for 10 ppb. In the New Jersey mortality data they allow differences in differences approach within census tracts that would not be feasible with monitors.