A Minimal Model for the Short-Term Prediction of Rainfall in the Tropics

Abstract

A ?minimal? model is proposed here for the short-term prediction (up to 12 h ahead) of precipitation occurrence in the tropics. The model is purely statistical, consisting of an optimally weighted linear combination of a Markov chain and persistence. It is minimal in the sense that only surface data are needed, and the computing requirements are almost nil. In this study the skill of the minimal model, i.e., accuracy relative to climatology and/or persistence, is demonstrated in theory and practice. The model was tested in real time during the 1986/87 Australian monsoon season at the tropical city of Darwin. Results of the real-time experiment reveal that the minimal model was the only model of those available to the Australian Bureau of Meteorology (including manual forecasts, a regional NWP model, and a model output statistics (MOS) scheme) that exhibited forecast accuracy greater than that of both climatology and persistence.