Application of a Markov technique to the operational, short-term forecasting of rainfall

The short-term prediction of precipitation remains a major forecasting problem. Most Australian Regional Forecasting Centres (RFC) still use basically subjective techniques and exhibit modest levels of skill. In this study it is shown that the Markov chain technique has sufficient skill to form an objective basis for the short-term operational prediction of the probability of precipitation. This is demonstrated by the performance of the technique in two comparative trials. The first trial was a real-time trial for Melbourne, carried out over a period of more than two years. In this trial the performance of the Markov chain model, as judged by the standard measurement of skill (half-Brier scores), was clearly superior to probability of precipitation forecasts issued by the Victorian Regional Office. The second trial involved comparing the Markov chain technique with persistence and climatology forecasts for all eight Australian RFCs, using 14 years of data for verification. Again it was found that the Markov chain technique showed such a high level of skill relative to persistence and climatology that it can be considered for use as an operational technique for short-term rainfall prediction. Finally, we discuss how the skill level of the Markov chain technique can be enhanced by combining it in a linear manner with other independent forecasting techniques, such as numerical weather prediction, as part of an integrated system.