Abstract
The errors in a barotropic filtered model of the Australian region at 500 mb are shown to be correlated with the numerical prognoses. Using the latter as predictors, an optimal linear prediction of the errors is found to remove about one-third of the error variance. The prognoses have been efficiently represented by the amplitudes of a few Empirical Orthogonal Functions. Tests on independent data sets show that the predictors are reliable. The prediction shows day-to-day utility: it achieves or betters the sample mean-error reduction on two-thirds of the days in the sample.