Modeling regional haze in the BRAVO study using CMAQ‐MADRID: 1. Model evaluation

Abstract

The Big Bend Regional Aerosol and Visibility Observational (BRAVO) study was a multiyear monitoring and assessment study of the causes of regional haze in Big Bend National Park (BBNP), Texas. The Community Multiscale Air Quality model augmented with the Model of Aerosol Dynamics, Reaction, Ionization and Dissolution (CMAQ‐MADRID) was used to simulate tracers and regional haze in Big Bend National Park for the full BRAVO period of July‐October 1999. The BRAVO monitoring network provided an opportunity to conduct a comprehensive evaluation of CMAQ‐MADRID over a 4‐month period. Tracer simulations revealed uncertainties in the model representation of advection and diffusion processes and the effects of uncertainties in meteorological fields on transport simulations. Results improved with the implementation of a more diffusive horizontal diffusion scheme. The 12‐km resolution provided better results than the 4‐ and 36‐km resolution simulations for 15–25 August 1999 and the 36‐km resolution provided better results for 5–15 October. Model performance for tracers suggests that as currently formulated, grid‐based Eulerian models are not well suited to simulate the impacts of long‐range transport of individual point source emissions at specific receptors. Nonetheless, they are suitable for resolving the contributions of source regions that may contain multiple area and point sources. Using a 36‐km resolution for a 4‐month simulation, the model performance was good in comparison with contemporary models for sulfate (the major PM2.5 component) in the region of interest (i.e., BBNP), with a low bias and coefficient of determination better than 0.5. However, the model overestimated sulfate and total sulfur significantly in other parts of the modeling domain. For organic particulate matter (OM, the second most prevalent PM2.5 component at BBNP), the model correctly reproduced the dominance of secondary organic aerosols and explained most of the variance in the OM concentrations; however, it underestimated OM concentrations consistently. Model performance was poor for the less prevalent components of PM2.5 (i.e., nitrate and black carbon) at BBNP. Diagnostic analyses suggest that the discrepancies between model simulation results and observations are due not only to limitations in the model formulation but also to uncertainties in the model inputs, including emissions, meteorology, and boundary conditions.