The incorporation of reactivity into the Chemical Mass Balance (CMB) receptor model to improve source attribution of NMHC in Houston Texas

摘要

The ability of the CMB receptor model to resolve sources of reactive pollutants was assessed using simulated ambient concentrations of non-methane hydrocarbons (NMHC) typical of Southeast Texas. Ambient concentrations were simulated using a multistage unsteady state photochemical box model which used simplified reaction, convection, and dispersion schemes to describe the decay, transport and dispersion of emissions en route to the Galleria site in Houston Texas. Emission rates into the model were based on those reported in the Texas official gridded inventory. This paper reports on a subset of the 1792 cases performed to explore the impact of reasonable uncertainties in four key assumptions implicit in the CMB formulation. These assumptions are: 1) all major sources are known; and for each of these major sources, 2) a single fitting species exists, which is both unique to the source and inert; 3) a source profile is available that reasonably describes the distribution of pollutants in the fresh emission; and 4) this source profile adequately represents the source even after photochemical reaction in the ambient air has non-uniformly distorted it. The first three assumptions were investigated by omitting or adding key sources, by varying the fitting species made available to CMB, or by using source profiles generated in one area to describe sources in another. To test the last assumption, both fresh emissions profiles and photochemically aged source profiles were made available to CMB. CMB model performance was assessed by comparing its source contribution estimates (SCEs or Sj) to the actual emissions input to the photochemical box model. When typical uncertainties were introduced into the model inputs,CMB was not able to consistently allocate sources of reactive compounds. When reactivity was incorporated into the model using aged source profiles, some improvement in model performance was observed.Although the model was less likely to achieve a solution (i.e., cases for which model reached a solution decreased from 87% to 58%), the model predictions were more accurate when it did (i.e., cases which were inaccurate decreased from 12% to 4%).