Statistical Prediction of Tropical Cyclone Motion: An Analog–CLIPER Approach

Abstract

In recent years, numerical models have shown significant improvement in their ability to predict tropical cyclone motion. This has brought about a decline in the development, improvement, and use of models that treat environmental data in a statistical prediction framework. However, the very basic statistical climatology and persistence (CLIPER) models continue to be functional in both operational and research environments. In this paper, an attempt is made to improve the performance of these models by combining the CLIPER concept with the analog concept. The new model, referred to here as the Modèle Climatologiques de Cyclones par Analogues (MOCCANA), is a statistical 72-h forecast model and uses two sets of regression equations (zonal and meridional displacement) to forecast tropical cyclone motion. The predictors entering the model are dependent on an analog selection function of current tropical cyclone motion, current time of year, and location. The coefficients of the regression equations are calculated in a transform space (or eigenspace) using principal component analysis (PCA). MOCCANA was run for all cases during the 1988?97 period in seven major ocean basins: north Indian, western North Pacific, eastern North Pacific, North Atlantic, southwest Indian, southeast Indian, and southwest Pacific. A comparison with pure CLIPER models for various basins shows that, for all forecast periods and for all basins, MOCCANA exhibit a smaller track forecast error.