Stochastic parameter estimation and uncertainty analysis applied to photochemical air quality models

Incremental reactivity estimates (IRs) for aromatic hydrocarbons and higher molecular weight volatile organic compounds (VOCs) from consumer products have been considered highly uncertain because of gaps in understanding their oxidation mechanisms. Parameterized mechanisms derived from chamber experiments are used in air quality models for these compounds. Uncertainties in rate constants, experimental conditions, and chamber artifacts affect the parameters so derived. Using stochastic programming, this study finds that uncertainties for the chamber-derived parameters range from 18% (1sigma relative to the mean) for the organic nitrate yield of 2-butoxy ethanol to 71% for the MGLY (alpha-dicarbonyls) yield for p-xylene in the SAPRC-97 mechanism. The aromatics parameters are most influenced by rate constants for the aromatics+OH, NO 2+OH reactions, the light intensity, parameters representing unknown chamber radical sources and initial aromatic concentrations in the experiments. Light intensity and radical source parameters are also highly influential to the 2-butoxy ethanol parameters, while PPN formation and initial m-xylene and n-butyl acetate concentrations are strongly influential for n-butyl acetate. The effects of the chamber-derived parameters on IRs are then investigated through Monte Carlo analysis with Latin hypercube sampling. The uncertainties for absolute maximum incremental reactivities (MIRs), maximum ozone incremental reactivities (MOIRs) and equal benefit incremental reactivities (EBIRs) range from 27 to 49% for most of the aromatics, 2-butoxy ethanol and n-butyl acetate. Uncertainties in IRs of 25 other VOCs without chamber-derived parameters in their mechanisms range from about 20 to 37%. The uncertainties in the corresponding relative IRs are fairly consistent for most of the VOCs, ranging from 10 to 30%. The chamber-derived aromatics parameters strongly influence the IRs of most of the VOCs studied, including non-aromatics. But the effects of the chamber-derived parameters for 2-butoxy ethanol and n-butyl acetate are relatively small.; Decoupled stochastic optimization was applied to the August 28, 1987 air pollution episode in Riverside, CA to demonstrate its use in designing ozone control strategies, accounting for modeling uncertainties including emissions, chemical mechanism parameter and meteorology. The optimal control strategies obtained by the isopleth approach for the deterministic case only have about 50% probability of meeting the target levels. The relative effectiveness of VOC and NOx controls depends on the proposed ozone concentration target. This approach appears promising in analyzing the effectiveness of VOC and NOx controls under uncertainty.