In recent years, the accuracy of operational tropical cyclone prediction has improved considerably. High spatial, temporal and spectral resolution satellite and radar observations, recent advances in numerical weather prediction and data assimilation, and the continuing increase in computer power, together have generated significant improvements in the prediction of tropical cyclones. In the Australian Region, average operational mean forecast position errors were reduced below 200 km at 48 hours for the first time, in the 1999 - 2000 tropical cyclone season. However, there still remain many tropical cyclones for which large errors occur. An entire season can have a large mean error as a result of just one or more large errors in track predictions. Here, a large number (30) of such cyclones have been examined, many of them chosen because of difficulties in the real time prediction of their tracks. These tropical cyclones, drawn from various basins around the world, have been modelled using the generalised inverse of a high resolution limited area primitive equation forecast model and the ingestion of near continuous high resolution satellite data. The combination of a high resolution model and its inverse, together with the near continuous data, significantly reduced many large forecast track errors, thereby resulting in errors well below the seasonal mean.

In over 90% of cases (27 out of 30 tropical cyclones) the 48 hr and 72 hr position errors were reduced. In the three cases that were degraded, the track error increases were all less than 20% at 48 and 72 hours. However, in 20 of the cases where the error was reduced, the reduction in forecast track error was over 25%, with 5 cases exceeding a 50% reduction. With the large amount of data used, it was not necessary to employ a bogus vortex in the initial field. Tropical cyclone errors often grow rapidly as a result of poor simulation of the far field. However, the high resolution data and model ensured that the far field evolved more accurately than in operations, and contributed especially to the largest error reductions. Several case studies will be presented in detail to illustrate the various aspects of the results obtained in this study.

Finally, the impact on intensity forecasting is described, and some initial results also are presented that show improved rainfall and structure forecasts obtained from the 4D-VAR assimilation of radar data.