Using a computationally inexpensive statistical technique based on optimum interpolation in space and time, analyses and forecasts for 12 and 24 hours from 0000 and 1200 Q.IT were produced at 250 mb in the Australian region over a five degree latitude-longitude grid (12.5 to 47.5°S, 102.5 to 157.5°E). The forecasts were compared with numerical forecasts based on the filtered baroclinic model currently used by the National Meteorological Analysis Centre (NMAC), Melbourne, and with the manually modified numerical forecasts produced operationally at about 0800 and 2000 GMT. Over a sample of 597 forecasts from November 1975 to September 1976 the statistical forecasts, which were not manually modified, were superior to the numerical forecasts and slightly better than the operational fore-casts in the tropics. Outside the tropics, the statistical forecasts were worse than both the numerical and operational forecasts. The inclusion of aircraft winds and conventional asynoptic data up to 24 hours old produced a small improvement in a sub-sample of the statistical forecasts.
R. S. Seaman (Wed,) studied this question.