Design, Application, and Recommendations for Including Inventory Uncertaintiesin Emission Inventory Preparation for Modeling

The air quality modeling community widely acknowledges that there are considerable uncertainties in the emission inventories used for modeling. Nonetheless, these uncertainties are typically ignored in air quality modeling applications. Three reasons for ignoring these uncertainties are (1) uncertainties in model inputs typically are not well quantified, (2) emissions and air quality modeling codes are not equipped to support uncertain inputs, and (3) methodologies are lacking that address the previous two issues. In this paper, we describe a methodology for integrating some types of emissions uncertainties into emissions modeling. The methodology is applied to an air quality modeling application, allowing the effects of emissions uncertainties on air quality simulations to be characterized. To facilitate this application, the Sparse Matrix Operator Kernel Emissions (SMOKE) modeling system has been modified to accept statistical or empirical distributions describing emissions factors as well as multiple realizations of hour-specific emissions data. SMOKE was used with these data to simulate many alternative realizations of the emissions inventory. These realizations were then modeled with the Multiscale Air Quality Simulation Platform (MAQSIP) and the uncertainty in the air quality model predictions was characterized. Computational intensity is addressed by distributing the model runs over a cluster of inexpensive Linux computers. The paper concludes with recommendations for improving the characterization of uncertainty in emissions inventories, with the goal of making probabilistic assessments more practical.