Evaluation of a seven-year air quality simulation using the Weather Research and Forecasting (WRF)/Community Multiscale Air Quality (CMAQ) models in the eastern United States

The performance of the Weather Research and Forecasting (WRF)/Community Multi-scale Air Quality (CMAQ) system in the eastern United States is analyzed based on results from a seven-year modeling study with a 4-km spatial resolution. For 2-m temperature, the monthly averaged mean bias (MB) and gross error (GE) values are generally within the recommended performance criteria, although temperature is over-predicted with MB values up to 2 K. Water vapor at 2-m is well-predicted but significant biases (>2 g kg(-1)) were observed in wintertime. Predictions for wind speed are satisfactory but biased towards over-prediction with 0 < MB <0.5 m s(-1) and root mean square error (RMSE) around 1.5 to 2 m s(-1). Wind direction, predicted without observation nudging, is not well-reproduced with GE values as large as 50 degrees in summertime. Performance in other months is better with RMSE around 20-30 degrees and MB within +/- 10 degrees. O-3 performance meets the EPA criteria of mean normalized bias (MNB) within +/- 0.15 and accuracy of unpaired peak (AUP) within 0.2. Normalized gross error (NGE) is mostly below 0.25, lower than the criteria of 035. Performance of PM10 is satisfactory with mean fractional bias (MFB) within +/- 0.6, but a large under-prediction in springtime was frequently observed. Performance of PM2.5 and its components is mostly within performance goals except for organic carbon (OC), which is universally under-predicted with MFB values as large as -0.8. The predicted frequency distribution of PM2.5 generally agrees with observations although the predictions are slightly biased towards more frequent high concentrations in most areas. Elemental carbon (EC), nitrate and sulfate concentrations are also well reproduced. The other unresolved PM2.5 components (OTHER) are significantly overestimated by more than a factor of two. No conclusive explanations can be made regarding the possible cause of this universal overestimation, which warrants a follow-up study to better understand this problem. (C) 2013 Elsevier B.V. All rights reserved.