An evaluation of post-processing methods to correct surface forecast biases in the NAM model over the western U.S

Despite improvements in numerical weather prediction models, model biases are still unavoidable due to imperfect model physics, initialization, and boundary conditions. However, recent studies suggest that statistical post-processing of model forecasts may help to reduce model biases. We evaluated the performance of three types of post-processing methods: i) the Kalman filter (KF), ii) a seven-day running mean bias removal, and iii) Model Output Statistics (MOS), for the Eta/NAM model surface forecast of 2-m temperature, 2-m dewpoint, and 10-m wind in the summer of 2004 and the winter of 2004-2005 for 145 stations in the Western United States. In this study, all post-processing forecasts performed better in the summer than in the winter. MOS performed the best amongst all the post-processing methods in temperature, dewpoint, and wind direction. However, MOS did not perform better than the other post-processing methods in wind speed. KF and seven-day running mean bias removal had similar performance, and they performed the best when the synoptic environment is quasi-steady. However, KF was able to adapt to a change in the weather regime more quickly than seven-day running mean bias removal. Despite the shortcomings of the KF, it may still be useful in providing point-specific forecasts in many regions where traditional MOS are unavailable, especially upper-elevation mountain and mountain valley locations that are far removed from airports. This paper will also discuss a method to improve upon the existing KF bias correction to target its inherent weakness.