Global land use regression and Bayesian Maximum Entropy spatiotemporal estimation of PM(2.5) yearly average concentrations across the United States

Knowledge of PM2.5 concentrations across the United States is limited due to sparse monitoring across space and time. This work incorporates a land use regression (LUR) mean trend into the Bayesian Maximum Entropy (BME) framework along with Gaussian-truncated soft data that accounts for sampling incompleteness to provide estimations in the contiguous United States from 1999 to 2009. The LUR model was optimized to explain the most variability as possible given variable hyperparameters. Variables in the final model included elevation, average car miles driven, average traffic through-put, population density, SO2 point source emissions, and NH3 point source emissions. Compared to a kriging method with a constant mean trend this method showed a mean squared error reduction of over 35%. This is one of the few works to successfully develop a LUR model on a domain of this magnitude across space and time and incorporate the BME estimation methodology.