Evaluation of Three Numerical Weather Predictions Using the Weather Research and Forecasting Model

Meteorology greatly influences air quality, a complex process involving direct emissions, chemical and physical transformations, and removal of air pollutants. In order to generate accurate air quality forecasts, reliable numerical weather predictions are essential. We used model output from three variants of the Weather Research and Forecasting (WRF) model: the University of Washington’s (UW) WRF-GFS with 4-km resolution, Washington State University’s (WSU) WRF-Chem with 18-km resolution, and the National Oceanic and Atmospheric Administration’s High-Resolution Rapid Refresh (HRRR) with 3-km resolution. Each model differs in selection of atmospheric physics options and data assimilation level. WSU WRF is a coupled model, accounting for feedbacks between meteorology and chemistry, while both UW WRF and HRRR are decoupled models. We selected the 12-day evaluation period, August 29 to September 9, 2017, for model comparison because of the high wildfire activity in the Pacific Northwest during this time. MesoWest—a cooperative project led by the University of Utah, the National Weather Service, and other organizations—was our source of weather observations. Observed and predicted near-surface temperature, wind speed, wind direction, and relative humidity were used to determine the normalized mean bias (NMB), normalized mean error (NME), root mean square error (RMSE), and coefficient of determination (R²) at each MesoWest station and for all stations, within the UW WRF domain. We discuss factors that may have contributed to the biases and errors, namely resolution, coupling, and physics options. The results indicate that the WSU WRF underperforms the other models overall, suggesting that a high resolution might be paramount for numerical weather prediction. This research provides valuable information on the quality of each WRF variant and the implications for air quality forecasting.