Monday November 2nd, 4pm, Mann Hall 323
Abstract: This work can best be seen as the second step in a long term project, the ultimate goal of which is to develop and evaluate the ability of a process-based space-time statistical model of ozone to address multiple regulatory and research needs. Those needs include decomposing ozone into parts attributable to background, local creation, and regional transport; retrospective prediction of ozone at unmonitored space-time locations for use in health and eco-system impact studies; assessment of the efficacy of past emission control programs such as the NOx SIP Call; assessment of the potential for proposed emission controls to bring future ozone concentrations into attainment; and quantification of the uncertainty of model parameters and ozone predictions. The first version of this model expressed ozone as a function of meteorology, observed NOx, a latent VOC field, and a transport term to meaningfully account for space-time correlation. The latent VOC field was a function of VOC emissions and meteorology. Both the ozone and VOC fields had a different set of spatial covariance parameters for each of four time periods. The original model and all modifications permit two predictors: one for retrospective prediction that uses spatial correlation of deviations from the mean trend, and another for forecasting that does not. The performance of the forecasting predictor also reflects the ability of the model to be used for decomposition. We present a variety of modifications to the original model, each designed to make the model more true to the atmospheric processes that create, destroy, and transport ozone. All models are evaluated using a withheld dataset and are compared to CMAQ, a deterministic model used by EPA.