Statistical Method for One-to Three-Day Tropical Cyclone Track Prediction

摘要

This newly developed method of forecasting one- to three-day cyclone motion is an improvement over existing forecast schemes. A screening--multiple regression technique--is used with the 'perfect-prog' concept to introduce numerically forecast synoptic data in the development of the regression coefficients. Actual forecast fields are used to test the skill of the new model to be referred to as CSU84. Comparison with operational forecast techniques on a nearly homogeneous sample using a persistance-climatology model as a normalizer to indicate relative skill of the models as compared to a no-skill forecast indicate that CSU84 is six times as skillful as the best operational model in the Atlantic and at least four times as skillful as the best operational model in the northwest Pacific. These very encouraging verification results suggest the potential for significant reduction in forecast position errors and should hasten the testing of the model in an operational environment.