Forecasting Tropical Cyclone Motion Using a Discriminant Analysis Procedure

View More View Less
  • 1 Bureau of Meteorology, Melbourne 3001, Australia
© Get Permissions
Full access

Abstract

A general method of objective tropical cyclone guidance is proposed in which the forecast variables are categories of tropical cyclone motion. Using this approach, information on more than one scenario is presented, a potentially important diagnostic consideration in uncertain forecast situations, e.g., when conflicts exist between aids or when a major track change is possible. As a first step to this type of forecasting, motion categories were defined using tercile ranges of zonal and meridional Australian region storm motion. Discriminant functions developed using both past track and current synoptic data then provided probability forecasts of each category. Forecasts for each zonal and meridional category were obtained 12, 24, 36 and 48 h in advance. With this simple approach, the accuracy of zonal (meridional) classification on the dependent data ranged from 76 (69)% at 12 h to 59 (57)% at 48 h. A no-skill category assignment would yield a figure of 33%. Classification accuracy was generally best in the below- and above-average groups where extreme storm motion and the largest forecast errors occurred.

Position forecasts were also derived for each category of motion using multiple linear regression equations. Thus, in addition to the probability information aimed at a general description of the storm's behavior, position forecasts specifically developed for each category of motion were also available. Objective use of these position forecasts resulted in lower errors than obtained with the best aid available to tropical cyclone forecasters in the Australian region, especially for situations in which some big forecast errors occurred, i.e., large movement cases.

Abstract

A general method of objective tropical cyclone guidance is proposed in which the forecast variables are categories of tropical cyclone motion. Using this approach, information on more than one scenario is presented, a potentially important diagnostic consideration in uncertain forecast situations, e.g., when conflicts exist between aids or when a major track change is possible. As a first step to this type of forecasting, motion categories were defined using tercile ranges of zonal and meridional Australian region storm motion. Discriminant functions developed using both past track and current synoptic data then provided probability forecasts of each category. Forecasts for each zonal and meridional category were obtained 12, 24, 36 and 48 h in advance. With this simple approach, the accuracy of zonal (meridional) classification on the dependent data ranged from 76 (69)% at 12 h to 59 (57)% at 48 h. A no-skill category assignment would yield a figure of 33%. Classification accuracy was generally best in the below- and above-average groups where extreme storm motion and the largest forecast errors occurred.

Position forecasts were also derived for each category of motion using multiple linear regression equations. Thus, in addition to the probability information aimed at a general description of the storm's behavior, position forecasts specifically developed for each category of motion were also available. Objective use of these position forecasts resulted in lower errors than obtained with the best aid available to tropical cyclone forecasters in the Australian region, especially for situations in which some big forecast errors occurred, i.e., large movement cases.

Save