Estimation of PM2.5 Concentrations in the Conterminous U.S. Using MODIS data and a Three-Stage Model

Abstract:

previous studies showed that fine particulate matter (PM_2.5, particles smaller than 2.5µm in aerodynamic diameter) is associated with various adverse health outcomes. Many efforts have been made to develop PM_2.5 prediction models using satellite-derived aerosol optical depth (AOD) to take advantage of its comprehensive spatiotemporal coverage. However, those models are generally built on regional scales. To date, attempts to develop models to predict PM_2.5 concentrations in the conterminous U.S. has not been seen in literature probably because of the difficulties of building such a model that can adapt to a great variety of meteorological conditions and land covers. In this study, we combined the Moderate Resolution Imaging Spectroradiometer (MODIS) dark target and deep blue AOD to increase the spatiotemporal coverage. A three-stage model was developed to predict spatiotemporal-resolved PM_2.5 concentrations in the conterminous U.S. using MODIS AOD as the primary predictor and meteorological fields and land use variables as secondary predictors. The first two stages, including a linear mixed effects model and geographically weighted regression, account for the spatiotemporal relationship between PM_2.5 and AOD, and the third stage generalized additive model was developed to predict PM_2.5 concentrations in areas where AOD is missing. The results show that model fitting generated R² of 0.60 and RMSPE of 4.23 μg/m³, indicating a good fit between the dependent variable and predictor variables. The spatial pattern shows that high PM_2.5 concentrations occur in big cities such as the Houston metro area, and the eastern U.S. is more polluted than western regions.