Geostatistical data fusion estimation methods of ambient PM2.5 and polycyclic aromatic hydrocarbons

摘要

Fine Particulate Matter (PM2.5) is a complex air pollutant associated with a host of adverse health effects. In epidemiologic studies there is a need to accurately predict exposures to reduce misclassification. Recently there has been a surge in data fusion methods which combine observed data with gridded modeled data like the regulatory Community Multiscale Air Quality (CMAQ) model. Substantial resources are allocated to the evaluation of CMAQ. However, this model has inherent error and uncertainty. Currently, CMAQ can only be operationally evaluated at locations where observed data exist, leaving potentially large spatial and temporal gaps in a given modeling domain. This study develops a framework for evaluating gridded air quality modeled data that can then be corrected for systematic error and combined with observed data in a geostatistical framework. First, this dissertation develops the novel Regionalized Air quality Model Performance (RAMP) method that performs a non-homogenous, non-linear, non-homoscedastic model evaluation at each CMAQ grid for a well-documented 2001 regulatory episode across the continental United States. The RAMP method comparatively outperforms other model evaluation methods with a 22.1% reduction in Mean Square Error (MSE). Secondly, the RAMP corrected CMAQ modeled data are combined with observed data in the modern Bayesian Maximum Entropy (BME) geostatistical framework which combines the accuracy of observed data with the spatial and temporal coverage of gridded modeled data. RAMP BME resulted in a 6-7 times increase in spatial refinement compared to using kriging alone. Lastly, the data rich PM2.5 environment is contrasted with the data poor environment of Polycyclic Aromatic Hydrocarbons (PAHs). The Mass Fraction (MF) BME method is developed through a relatively small number of paired PM2.5 and PAH values and is applied to PM2.5 observed locations where PAH have not been observed to create the first detailed spatial maps of PAH across North Carolina in 2005. The MF BME method reduces MSE by over 39% compared with using kriging alone. Accurate assessment of ambient air pollutants is essential in public health to explore and elucidate true underlying relationships between pollutants and health endpoints.