Model Output Statistics (MOS) forecasting techniques, which involve finding statistical relationships between predictands and predictors taken from numerical models and/or observations, have been in use for over 30 years. Forecasts generated by MOS systems assist meteorologists on a routine basis by providing guidance in preparing their own forecasts, and in some cases are output directly to the general public. Since MOS is so commonplace, any improvement in forecasts made by MOS systems would be useful.

In an effort to reduce the error associated with Nested Grid Model (NGM) MOS surface temperature forecasts, the techniques of multiple linear regression and Kalman filtering will be applied to the errors. NGM MOS forecast errors from 1994-2000 and the corresponding surface observations will be used as the training data. The regression equations and the Kalman filtering technique will be tested using NGM MOS forecasts from the year 2001 to correspond with an earlier MOS verification study. Regression will be carried out using the error in the MOS forecast as the predictand. The predictors will come from surface observations at the site for which forecasts are being made as well as from four surrounding sites. Bias, mean absolute error, and error relative frequencies by category calculated over the year 2001 will be presented for selected sites and lead times. Regression will be carried out for lead times of 12 and 24 hours, while longer lead times may be considered with the Kalman filter. The calculated error statistics for the unmodified NGM MOS forecasts, the forecasts adjusted using the new regression equations, and the forecasts adjusted using the Kalman filter will be compared with one another and analyzed.

Future results will be posted to the following web site: http://weather.ou.edu/~aataylor/research