Reduction of Tropical Cyclone Position Errors Using an Optimal Combination of Independent Forecasts

View More View Less
  • 1 Bureau of Meteorology Research Centre, PO Box 1289K, Melbourne, Australia
  • | 2 Institut fur Meteorology, Freie Universitat Berlin
© Get Permissions
Full access

Abstract

It is shown that an optimal linear combination of independent forecasts of tropical cyclone tracks significantly reduces the mean forecast-position errors. In this study the independent forecasts are provided by a statistical scheme (CLIPER) and a numerical weather prediction (NWP) model operating over the Australian tropics.

A comparison is made between the optimal linear combination and four other forecast techniques, over the five Australian tropical cyclone seasons 1984/85–1987/88. The combination method gave a mean position error of 157 km at 24 h using independent “best track” data, an improvement of 15% over the next most accurate method. At 48 h, the mean position error of 312 km was 17% less than the next most accurate scheme.

The combination method was assessed further in a real-time trial on operational data during the 1988/89 Australian tropical cyclone season. The results of this trial confirmed the superiority of the combination technique over the other methods. It will be used operationally in the next Australian tropical cyclone season (1989/90) either in its present form or as part of an integrated “expert” system being developed specifically for tropical cyclone motion prediction.

Abstract

It is shown that an optimal linear combination of independent forecasts of tropical cyclone tracks significantly reduces the mean forecast-position errors. In this study the independent forecasts are provided by a statistical scheme (CLIPER) and a numerical weather prediction (NWP) model operating over the Australian tropics.

A comparison is made between the optimal linear combination and four other forecast techniques, over the five Australian tropical cyclone seasons 1984/85–1987/88. The combination method gave a mean position error of 157 km at 24 h using independent “best track” data, an improvement of 15% over the next most accurate method. At 48 h, the mean position error of 312 km was 17% less than the next most accurate scheme.

The combination method was assessed further in a real-time trial on operational data during the 1988/89 Australian tropical cyclone season. The results of this trial confirmed the superiority of the combination technique over the other methods. It will be used operationally in the next Australian tropical cyclone season (1989/90) either in its present form or as part of an integrated “expert” system being developed specifically for tropical cyclone motion prediction.

Save