Incorporating Uncertainties in Emission Inventories Into Air Quality Modeling

Abstract

In modeling ambient ozone concentrations, NOx emissions estimated based on emission inventories are used as input to air quality models. A concern regarding the quality of ozone predictions is the uncertainty inherent in emission inventories. This work aims at developing new methodologies for quantifying uncertainty in emission inventories and propagating the uncertainty through a photochemical grid air quality model.

Time series techniques were used to develop new methodologies for developing probabilistic emission inventories. These methodologies were applied to a case study for NOx emissions for each of 32 units of 9 coal-fired power plants in the Charlotte domain, North Carolina. Probabilistic inventories for a near-term future episode based on the data of the year 1995 and for a distant future episode in 2007 based on the data of the year 1998 were developed. In order to investigate how much of an effect does correlation between emissions from different units has on the developed probabilistic inventory, two different approaches were used. Univariate time series techniques were applied in the first where each unit is assumed to be dispatched independently of all other units. Multivariate time series techniques were applied in the second in order to account for the inter-unit dependence in developing the inventory. A methodology for accounting for intra-unit dependence between variables used in estimating the inventory was also developed. For the first approach, the 1995 case showed that the 95% confidence interval for the daily inventory lied between 562 t/d and 698 t/d. This represents approximately ±10% uncertainty range from the average value which is 639 t/d. The daily inventory for the 2007 case showed an uncertainty range of ±8% of the average value which is 192 t/d. The second approach showed that the total daily inventory for the year 1995 had a 95% confidence interval of 548 to 778 t/d, corresponding to an uncertainty range of -15% to +22% of the average value while the 2007 case showed an uncertainty range of -8% to +15% of the average value. Comparison of the simulated results of the two approaches with observations showed that the dependent approach produced a distribution for uncertainty that more accurately represents the observed data.

Both inventories were used as input to an air quality model to investigate the propagation of uncertainties in emission inputs through the model. Uncertainties in the maximum 1-hour and 8-hour ozone predictions were estimated at different locations in the modeling domain. Forty-three grid cells were estimated to have a probability greater than 0.9 that the maximum hourly ozone concentration exceeds the 120 ppb 1-hour ozone standard. A similar analysis was conducted for the 8-hour ozone standard where 1654 grid cells showed a probability greater than 0.9 of exceeding the 80 ppb standard. The results of the case study demonstrate that the range of hourly variability in power plant emissions is sufficient large to justify a quantitative analysis of uncertainty, and that the range of uncertainty in air quality predictions is large enough to imply ambiguity regarding development of control strategies. The developed methodologies are very useful tools for decision making. These methodologies can be employed to develop control strategies to achieve attainment with an acceptable degree of confidence, such as 90 or 95 percent.