Real-Time Short-Term Forecasting of Precipitation at an Australian Tropical Station

Abstract

The results of a major real-time trial of techniques for the short-term (12 h ahead) prediction of precipitation for the Australian tropical city of Darwin are described. The trial compared current operational manual forecasting procedures with a range of alternative techniques including statistical methods, numerical weather prediction (NWP), and model output statistics (MOS) which were developed specifically for the trial. The only technique of those tested which exhibited skill, i.e., consistent superiority over climatology and/or persistence, was a Markov chain model used either individually or in linear combination with other methods. Of particular significance was the relatively poor showing in the tropics of the numerical weather prediction model products, which were worse than both persistence and climatology. The results of this real-time trial should be treated with care because of the small sample size involved.