On the Orientation of Grid Systems for the Statistical Prediction of Tropical Cyclone Motion

Abstract

Statistical models for the prediction of tropical cyclone motion traditionally have been formulated in a coordinate system oriented with respect to zonal and meridional directions. An investigation is made here into the forecast error reducing potential of a grid system reoriented with respect to initial storm heading. The developmental data comprise Atlantic forecast situations from 1965 through 1980 on all storms initially north of about 25°N. Reorientation of the coordinate system reduces the total variance in 24 h storm motion by 40%, projects most of the motion onto one (along-track) component of displacement, and makes the components nearly independent of each other. For 48 and 72 h displacements, however, these advantageous effects are substantially diminished or eliminated. Synoptic predictors derived from current deep-layer mean heights on a grid of 1700 km radius are used to forecast storm displacements. For the developmental data, grid reorientation lowers the 24 h forecast error by 13%, and reduces the slow speed bias by a factor of 2/4. For 24 h forecasts the skill in the prediction of cross-track motion is small. Empirical Orthogonal Function and Principal Estimator Patterns provide insight into the role of reorientation in the reduction of forecast error, and the position of grid-point height predictors selected by a screening technique.