Direct sensitivity analysis in air quality models

The theory and applications of sensitivity analysis in air quality modeling is investigated. Direct sensitivity analysis is applied for three-dimensional reactivity assessment of volatile organic compounds (VOCs). The effect of different domains, episodes, and emission inventories are investigated. Various relative reactivity metrics for about 40 (explicit and lumped) compounds are developed in a total of 5 episodes in California and the Eastern US. Most reactivity metrics are usually consistent with box model scales and among themselves. However, 3-D reactivity assessment also reveals important differences with box model scales, notably box model reactivities for most carbonyls are high. High levels of consistency among different domains, episodes and inventories are observed. Overall, relative reactivity scales prove to be a robust and reliable measure of ozone formation potential.; First-order sensitivity analysis cannot accurately characterize the nonlinear behavior in atmospheric response. To alleviate this limitation, the decoupled direct method (DDM) is extended to calculate higher-order sensitivity coefficients. High-order DDM (HDDM) is very efficient and calculates sensitivity coefficients of different orders at approximately the same computational cost as the first-order calculations. By applying higher-order analysis, a more accurate prediction of the response is achieved when nonlinearity is present. Some of the most nonlinear behaviors (largest magnitudes of high-order sensitivity coefficients) are seen in the high ozone concentration areas downwind of urban or industrial emissions. Applying high-order analysis in developing control strategies for such cases will result in markedly improved accuracy when assessing the response to large changes in emissions. Also, by using HHDM in calculation of the cross-derivatives (sensitivity to two or more parameters), the time-dependent and location-specific ozone isopleths are developed for a photochemical episode. Finally, by applying HDDM to reactivity simulations of central California, local uncertainty analysis of 3-D organic reactivities is carried out for six species and 27 uncertain inputs and parameters. As expected, the relative reactivities are found to have significantly lower uncertainty than the absolute values. Uncertainty contributions from different input parameters show a great deal of spatial variation and largely depend on the prevailng chemical environment. In general, uncertainty levels in relative reactivities are low-to-moderate (10–50%) and reaffirm the conclusion that the relative reactivity scales can be used reliably for VOC control regulations.