A Comparison of Simple Methods for the Postprocessing of Wind Speed

This study examines the statistical characteristics of several commonly applied methods for the statistical adjustment of wind speed forecasts to account for model biases. The techniques to be examined will include autoregressive (AR) bias correction, linear regression (Model Output Statistics), and quantile mapping (QM). Data are examined at a set of more than 5000 stations worldwide using commonly available forecasts such as the ECMWF and NOAA global ensembles and the Global Forecast System (GFS). Preliminary results available at the time of abstract submission indicate that linear regression provides a notable improvement in the RMSE of forecasts with respect to the AR. QM increases the RMSE somewhat. An advantage of the QM methodology relative to the other techniques is the lack of conditional biases; whereas MOS resembles the observation mean with a loss of predictability, the QM retains sharpness at the expense of the increased RMSE. Following Hamill (2021, here) we will examine whether combining the results from various post-processing methods provides an improvement relative to any one method on its own.