Evaluation of High-Resolution Rapid Refresh (HRRR) model performance for use in air dispersion modeling

The accuracy, spatial resolution and run time of meteorological forecasting models has been steadily improving. The number of models available through the National Weather Service through Model Analysis and Guidance is increasing, too. Some of the models available include GFS, NAM, NAM-HIRES, HRRR, HRW-NMMB, HRW-ARW, and RAP.

Dispersion models can use meteorological data from these models to drive transport and dispersion of pollutants. The meteorological model that has been most commonly used for the past decade, both in dispersion and photochemical modeling, is the Weather Research and Forecasting model (WRF). WRF is used both in operational forecasts and for a variety of atmospheric research projects, including many air dispersion research studies and regulatory applications.

To prepare meteorological data inputs on an air dispersion modeling project, WRF model would be run for one or more years with the highest resolution nest covering the project area of interest. The WRF output would then be checked against observations to evaluate the model performance, using a typical set of statistical measures.

In this study we used the same metrics that are commonly used when evaluating performance of WRF for use in dispersion modeling but apply them to HRRR, one of the highest resolution operationally available models. HRRR uses the WRF-ARW platform to produce hourly short-range forecasts which include extensive observational data assimilation. Hour zero analysis fields are available operationally in real time and are captured in several archive databases. As an operational model HRRR has been extensively evaluated for performance in weather forecasting, but few studies have focused on its application to air dispersion modeling. Here, we evaluate the model output for a 3-year period, for a domain covering New England. Statistics employed include established metrics and benchmarks associated with Index of Agreement (IOA), Mean Bias, Root Mean Square Error (RMSE), and Gross Error for wind, temperature and specific humidity.