Experiments using Model Output Statistics to predict precipitation at a tropical location

Probabilistic forecasts of short-period precipitation for a location in the Australian tropics were developed using Model Output Statistics (MOS). During equation development the selected expressions were applied to historical data withheld for independent testing. The resulting predictions showed skill relative to climatology, despite a number of adverse factors: the lead time of the forecasts was approximately 24 hours; the prognosis model from which many of the MOS predictors were generated was primarily designed for use in mid-latitudes; and the rarity of some of the events being predicted was such that predictions based on climatology were not easily improved on. The levels of skill achieved in the developmental phase were not maintained during an experiment conducted in the summer of 1986-87. The prognosis model in use during that period differed in many respects from that which provided the archive from which the prediction equations were developed, and the statistical characteristics of many of the MOS predictors were changed thereby. The resultant forecasts were affected adversely as a consequence.