Bias correction and consensus have been applied to MOS temperature forecasts in effort to increase accuracy. Temperature forecasts obtained for KOKC and KPIT from 1 May 2002 to 30 January 2005 were classified according to maximum and minimum, projection, initialization, and season to examine model behavior between these divisions. All forecasts were verified using RMSE.
Compared to the uncorrected individual model forecasts, a seasonal bias correction showed a slight increase in RMSE values. A lagged bias correction decreased RMSE by approximately 0.1°F to 0.5°F compared to the RMSE for the uncorrected forecasts. An equally weighted consensus decreased RMSE by about 0.5°F to 1.0°F and 0.1°F to 0.5°F for maximum and minimum temperatures, respectively, over the uncorrected individual model forecasts. This method improved upon the lagged bias correction of individual model forecasts by several tenths of a degree. A linear regression consensus performed slightly worse than the equally weighted consensus. An unequally weighted consensus method based on lagged variance was the most accurate of all forecast enhancement methods, decreasing RMSE values by approximately 0.5°F to 1.5°F compared to uncorrected individual model forecasts and by several tenths of a degree over the lagged bias corrected individual model forecasts. Thus, based on the methods examined in this study, it is shown that a model consensus using a lagged correction based on past performance will provide the most significant MOS temperature forecast improvement.