Short-range prediction of tropical cyclone tracks has improved greatly in the decade since the Tropical Cyclone Motion observational program (TCM-90) was carried out in 1990. The reductions in mean track errors have come about from improved models, more sophisticated data assimilation and initialization techniques, and dramatically increased computer power that allow corresponding increases in model resolution.

The TCM-90 data set has been used to "re-visit" three tropical cyclones that occurred in the 1990 season. These tropical cyclones were Yancy, Flo and Dot. All of the tropical cyclones, except Flo, made landfall. Moreover, some of the storms proved very difficult to predict at the time and even subsequently in research mode.

In the current program, the HIRES model at The University of New South Wales was used for the predictions, which were out to 72 hours ahead. All three tropical cyclone tracks were predicted accurately, with mean errors at 48 hours averaging just over 200km. Both tropical cyclones that made landfall were predicted to do so, close to the observed locations and times.

Sensitivity experiments have been carried out to assess why the forecasts are as accurate as they are. The major factors contributing to the excellent tracks have been identified by degrading the initial conditions for a range of variables such as domain size, SST distributions and orography. Ensemble prediction techniques have been applied to the tropical cyclones to gain further insight into the evolution of the errors present in the initial state.

Finally, the study is being extended to include all of the tropical cyclones available in the TCM-90 dataset.