Predicting changes in intensity of tropical cyclones using a Markov chain technique

At present no skilful, quantitative technique exists for the operational prediction of changes in tropical cyclone intensity. Here we describe the application of a Markov chain technique to this difficult problem. The method predicts transitional probabilities between five discrete categories of intensity and these probabilities are applied to short-period (24 hours) forecasting of tropical cyclones in the Australian region. When tested on independent data, the Markov technique out-performed a standard 'zero-skill' climatology-persistence (CLIPER) intensity forecast technique by up to 22 per cent, out to 24 hours. The greatest gains were found in the 6 to 18-hour range. In addition, the Markov predictions show a degree of improvement of between 25 and 40 per cent over the Dvorak method, which is the current operational objective technique used by the Australian Bureau of Meteorology. The effectiveness of the method under operational conditions is illustrated by applying it to four tropical cyclones that caused operational forecast difficulties. We suggest that this approach complements the traditional methods where position and intensity are forecast directly, largely based on information provided by persistence, climatology, satellite data and synoptic interpretation.