Quantification of Variability and Uncertainty in Emission Factors

The quality of emission factors is typically described using data quality ratings. Such ratings are qualitative and provide no indication of the precision of the emission factor for an average emission source, nor of the variability in emissions from one source to another within a category. Advances in methodology and computing power enable the application of a quantitative approach to characterizing both variability and uncertainty in emission factors. Variability refers to actual differences in emissions from one source to another due to differences in feedstock composition, design, maintenance, and operation. Uncertainty refers to lack of knowledge regarding the true emissions because of measurement errors (both random and systematic), limited sample sizes (statistical random sampling error), and non-representativeness (which can introduce additional errors, including systematic errors). The set of numerical methods generically known as “bootstrap simulation” are a powerful tool for characterization of both variability and random sampling error. In this paper, we demonstrate the use of bootstrap simulation and related techniques for the quantification of variability and uncertainty for selected examples, including NOx emissions from coal-fired power plants and CO, NOx, and hydrocarbon emissions for light duty gasoline vehicles. While our examples are focused upon emission factors for selected criteria pollutants or their precursors, the same methodology can be applied to other pollutants (e.g., hazardous air pollutants, greenhouse gases). The work described here was conducted at North Carolina State University in a project sponsored by the U.S. Environmental Protection Agency. INTRODUCTION Emission Inventories (EIs) are a vital component of environmental decision making. For example, emission inventories are used for: (a) characterization of temporal emission trends; (b) emissions budgeting for regulatory and compliance purposes; and (c) prediction of ambient pollutant concentrations using air quality models. If random errors and biases in the EIs are not quantified, they can lead to erroneous conclusions regarding trends in emissions, source apportionment, compliance, and the relationship between emissions and ambient air quality.