Use of high-order sensitivity analysis and reduced-form modeling to quantify uncertainty in particulate matter simulations in the presence of uncertain emissions rates: A case study in Houston

Abstract

Abstract Regional air quality models are widely used to evaluate control strategy effectiveness. As such, it is important to understand the accuracy of model simulations to establish confidence in model performance and to guide further model development. Particulate matter with aerodynamic diameter less than 2.5 μm (PM2.5) is regulated as one of the criteria pollutants by the National Ambient Air Quality Standards (NAAQS), and PM2.5 concentrations have a complex dependence on the emissions of a number of precursors, including SO2, NOx, NH3, VOCs, and primary particulate matter (PM). This study quantifies how the emission-associated uncertainties affect modeled PM2.5 concentrations and sensitivities using a reduced-form approach. This approach is computationally efficient compared to the traditional Monte Carlo simulation. The reduced-form model represents the concentration-emission response and is constructed using first- and second-order sensitivities obtained from a single CMAQ/HDDM-PM simulation. A case study is conducted in the Houston-Galveston-Brazoria (HGB) area. The uncertainty of modeled, daily average PM2.5 concentrations due to uncertain emissions is estimated to fall between 42% and 52% for different simulated concentration levels, and the uncertainty is evenly distributed in the modeling domain. Emission-associated uncertainty can account for much of the difference between simulation and ground measurements as 60% of observed PM2.5 concentrations fall within the range of one standard deviation of corresponding simulated PM2.5 concentrations. Uncertainties in meteorological fields as well as the model representation of secondary organic aerosol formation are the other two key contributors to the uncertainty of modeled PM2.5. This study also investigates the uncertainties of the simulated first-order sensitivities, and found that the larger the first-order sensitivity, the lower its uncertainty associated with emissions. Sensitivity of PM2.5 to primary PM has the lowest uncertainty while sensitivity of PM2.5 to VOC has the highest uncertainty associated with emission inputs.