Tropical Cyclone Intensity Prediction Based on Empirical Orthogonal Function Representation of Wind and Shear Fields

摘要

An objective techniques for predicting 24, 48 and 72 h tropical cyclone intensity is investigated using 1216 cases in the western North Pacific from 1979 to 1983. Potential predictors include conventional storm-related parameters, such as date, intensity, motion and position. Additional potential predictors include empirical orthogonal function (EOF) coefficients of the zonal and meridonal components of the environmental wind (250, 400 and 700 mb) and vertical wind-shear (250-400, 400-700, and 250-700 mb) fields. These coefficients represent the synoptic forcing in the vicinity of the storm. The intensity change information is filtered to eliminate data for storms affected by landfall from the sample. The regression equations are verified against a homogeneous sample of Joint Typhoon Warning Center (JTWC) official forecasts, which are also demonstrated to be significantly better (95% confidence) than persistence at all forecast intervals. Regression equations developed using EOF coefficient predictors along with conventional predictors are comparable to the JTWC official forecast, even at 48 and 72 h. The regression equations based on the complete set of predictors have slightly more skill than those based only on conventional predictors. If the regression equations are derived from a smaller sample to allow for an independent test, the results appear to be better in the dependent set, but are degraded in the independent sample. Nevertheless, these independent sample results are comparable in skill to the JTWC forecasts at all intervals. Regression equations generated from three subsets stratified by 12 h old intensity are significantly better than the 48 and 72 h JTWC official forecast. (Author)