Abstract
Emission factors and emission inventories are subject to both variability and uncertainty. Variability refers to observed differences attributable to true heterogeneity or diversity in emissions. Uncertainty refers to lack of knowledge regarding the true value of emissions. Variability in emissions can be attributed to variations over time, space or across different populations. Uncertainty in emissions typically arises due to limited sample size, lack of accuracy, non-representativeness of data, measurement errors, use of surrogate data, and human errors. This work successfully demonstrated new applications of quantitative methods for characterizing variability and uncertainty in emission estimates. The methods were demonstrated with respect to cases studies on nitrogen oxides (NOx) and volatile organic compound (VOC) emissions from natural gas-fueled internal combustion engines, and VOC emissions from consumer/commercial product use, gasoline terminal loading, cutback asphalt paving, architectural coatings and wood furniture coatings.
Emission data must be nonnegative, typically are positively skewed and have limited sample size. The restrictive assumption of normality used in analytical methods can lead to biased uncertainty estimates. Therefore, in this work, variability was characterized by fitting parametric distributions and uncertainty due to random sampling errors was quantified based upon numerical bootstrap simulation. Uncertainty in mean emission factors was found as much as minus 90 percent to plus 180 percent in a relative basis. Key methodological issues, including separation of intra- and inter-facility/engine variability, and methods for fitting parametric distributions to unequally weighted data, were addressed. Recommendations include extending these efforts to more emission source categories and for EPA and others to routinely report well-documented emission data to facilitate uncertainty analysis.
|
