Efficient characterization of pollutant-emission response under parametric uncertainty

An essential requirement of modeling for air quality management is to accurately simulate the responses of pollutant concentrations to changes in emissions. Uncertain model input parameters such as emission rates and reaction rate constants lead to uncertainty in model responses. However, traditional methods for characterizing parametric uncertainty are exceedingly computationally intensive. This paper presents methods for using high-order sensitivity coefficients in analytical equations to efficiently represent how the responsiveness of pollutants to emission reductions in the underlying photochemical model varies with simultaneous perturbations in multiple model input parameters. Separate approaches are introduced for characterizing the parametric uncertainty of pollutant response to a fixed or a variable amount of emission reduction. The approaches are demonstrated for an air pollution episode used in recent attainment planning in Georgia. For hypothetical scenarios in which domain-wide emission rates and photolysis rates are perturbed simultaneously by 50%, the reduced form models yield highly accurate predictions of the ozone impacts due to 50% reductions in nitrogen-oxide emissions in Atlanta (normalized mean bias 6.0%, normalized mean error 9.7%, R2=0.992) and inorganic particulate responses to Atlanta sulfur-dioxide emissions (-2.9% bias, 3.7% error, R2=1.000). Similar accuracy is achieved for pollutant responses to power plant emission controls.