Evaluation of Techniques for the Operational, Single Station, Short-Term Forecasting of Rainfall at a Midlatitude Station (Melbourne)

Abstract

Probability of precipitation (POP) for the 12 hours 0600 to 1800 local time was predicted for Melbourne each day for the three months (winter) period June-August 1986 using six different techniques. These were: a Markov chain model based on 20 years of three-hourly observations', the Australian region limited-area numerical weather prediction (NWP) model; a weighted linear combination of Markov and NWP models; a model output statistics scheme based on the NWP model; an analogue statistics procedure in which a set of the ?best? analogues of the NWP forecast were selected; and the manual ?official? Bureau of Meteorology forecast for Melbourne, issued by the duty forecaster. The six techniques were evaluated and compared in terms of Brier scores and also compared with predictions based on climatology. The results of this operational trial indicate that the skill of the combined Markov-NWP model forecasts considerably exceeds the other techniques. The Markov model was next, followed by the other methods which were close together in skill. Some cam was taken in the interpretation of thew findings as there were differences in lead times associated with the NWP model predictions and the MOS and analogue schemes which were dependent upon it, owing to the operational schedule at Melbourne.