Efficient characterization of uncertainty in control strategy impact predictions.
Antara Digar and Daniel Cohan, Rice University

Air quality models often estimate pollutant responsiveness to control strategies inaccurately due to uncertain input parameters, hampering the formulation of effective abatement plans. While these parametric uncertainties could be characterized by conducting a large number of Monte Carlo simulations of the photochemical model, to do so would often be impractical, especially for considering numerous control measures at fine resolution over lengthy episodes. We propose an alternative, computationally efficient approach that uses high-order sensitivity coefficients to develop analytical representations (i.e., surrogate models) of how both pollutant concentrations and their responsiveness to emission reductions vary with input parameter uncertainties. Probability distribution functions for these pollutant concentration and responsiveness estimates are then characterized by conducting Monte Carlo simulations of the analytical surrogate model. This computationally efficient method can provide a powerful tool for model evaluation and help air quality planners explore the range of possible impacts of potential control strategies. We demonstrate the methodology by its application to a case study for ozone and fine particulate matter attainment plan development in Georgia. We quantify the uncertainty in the impacts of emission controls (both regional and point-source) arising due to uncertainty in the photochemical model inputs (emission rates, boundary conditions, and reaction rates).