Observational data from the Australian Monsoon Experiment (AMEX) have been used to extensively test and verify short-term forecasts from the BMRC tropical prediction system. The main purpose of the paper is to provide a benchmark prediction study based on this unique observational data set. We document the ability of the system to predict a wide variety of tropical weather events, and identify systematic errors and problem areas in tropical prediction. These aims are achieved by examining observed and forecast time mean tropical wind field structures and by comparison between observed and predicted temporal changes in the large-scale tropical flow. Particular attention is paid to the prediction of monsoon events (onset, active and break phases) and to tropical cyclone behaviour (motion and strength). Short-term predictive skill on the synoptic scale is clearly demonstrated for all of the above phenomena. However, there is variability in skill on a day-to-day basis which we speculate is related to the quality of the initial conditions, and the sensitivity of prediction to both model resolution for certain phenomena (e.g. tropical cyclones) and parametrisation of physical processes. For AMEX tropical cyclone events, predictive skill is promising. Root mean square errors for track predictions at 24 and 48 hours are 220 and 315 km. However, even with the enhanced observational network, initial position and structure errors are still sometimes evident. Model spin-up on both the monsoon scale and the weather system scale is also suggested as a likely cause of systematic errors in the prediction system. Future enhancements to the system to cope with some of the problems identified here are discussed.
Davidson et al. (Tue,) studied this question.