Leveraging simulated source contributions to improve understanding of atmospheric ozone

Tropospheric ozone causes a variety of health outcomes ranging from mild breathing discomfort to mortality, and is atmospherically produced via precursor emissions from natural, intra-continental anthropogenic, and inter-continental anthropogenic. We evaluate estimates ozone source contributions from two nested version continental United States simulations (Zhang et al. 2011; Emery et al. 2012) at CASTNET, AQS and NOAA monitoring sites for 2006. Each model estimates ozone contributions from zero-out simulations. The predicted ozone and contributions share many features, but the differences offer an opportunity to evaluate uncertainty in emissions sources and meteorology. Both models are evaluated as a function of season and region, model performance and discrepancies are leveraged to evaluate underlying processes. Most apparent model-bias and inter-model differences occur at the ends of the concentration distribution (e.g., for <10th and >90th percentiles). For example, both models show limited ability to capture the timing and strength of episodic sources such as stratospheric intrusions. The models show better agreement, however, with observations when they are paired by quantile rather than time/space, i.e., if episodic events are treated as stochastic. Even when quantile-paired, some model biases and inter-model differences persist. Some persistent features are attributed to spatial resolution as expected. Others, however, can be attributed to the models' representation of chemistry, their emissions from wildfires, and their production of NOx from lightning. The spatial patterns of model biases suggest significant differences in the efficiency of ozone transport and local ozone production. For example, non-North American sources contribute substantially more in the Northeast and Midwest. Model processes will be isolated through a series of sensitivity simulations. The component processes will then be correlated with model bias. The correlation between component processes and model bias will help to dis-aggregate model bias and suggest areas for improvements. Preliminary results indicate that spatial resolution is not the only factor to be considered when attempting to simulate or attribute regional air quality. Rather, difference in models' treatments of atmospheric chemistry and physics must be considered. These preliminary results also confirm earlier findings that agreement between models and measurements is improved as the averaging time of the simulation and measurements are increased. It is also apparent that when analyzing time series over longer time periods (e.g., months), special care should be taken to examine temporal trends in bias as this will improve understanding of the processes in the model. This study evaluates model bias and ozone contributions using time/space categories that account spatial/temporal trends, treating the simulations as representative days rather than traditional time/space paired evaluation. Ultimately, we will attribute model bias to component process and make recommendations for future modeling efforts.