Preparing Dispersion Model Surface Meteorological Inputs Using High-Resolution Rapid Refresh (HRRR) Data

Accurate meteorological inputs are essential for air pollution dispersion modeling. Traditionally, dispersion models rely on observational meteorological data collected from weather stations at fixed locations. However, the sparse distribution of weather stations limits the ability to capture fine-scale meteorological variability, particularly in areas far from weather stations. In this study, we developed a novel framework for generating American Meteorological Society/Environmental Protection Agency (EPA) Regulatory Model (AERMOD) compatible surface meteorology data (.sfc) using the High-Resolution Rapid Refresh (HRRR) dataset, which provides predicted meteorological variables at a 3-km spatial resolution and an hourly temporal resolution. We followed the AERMOD Model Formulation document to create three scenarios of surface meteorology data, including two exploratory scenarios by setting HRRR meteorology parameter ranges and recalculating key parameters based on whether an hour filled in convective or stable planetary boundary layer status. We then applied these HRRR-derived meteorology data in the Research LINE source (R-LINE) dispersion model to predict traffic-related nitrogen dioxide (NO₂) concentrations at 443 Air Quality System monitoring sites across the United States (U.S.) in the year 2019. For comparison, we also ran R-LINE using observational-based surface meteorology data preprocessed with AERMET. NO₂ concentrations predicted by R-LINE were compared against NO₂ measurement data by the meteorology inputs (three HRRR scenarios versus weather station data) using simple linear regression coefficient of determination (R²) and Index of Agreement (IOA). The three HRRR scenarios yielded a higher R² on average (0.26) than did the observational data (R²=0.16), suggesting over 60% increase in explained variance. In simple linear regression analyses stratified by distance between NO₂ sites and weather stations as well as by traffic magnitude around NO₂ sites, HRRR data generally outperformed observational data. Site-specific IOA analyses further showed that, compared to observational meteorology data, HRRR inputs performed better across most of the continental U.S. but not as well in urban areas. Overall, our findings demonstrate that HRRR data has the potential to be utilized in air pollution dispersion modeling and that it has superior predictive ability in locations that lack nearby weather stations.