Hurricane Track Prediction with a New Barotropic Model

Abstract

A new barotropic prediction system for the tracks of tropical cyclones is presented. The system (referred to as WBAR) consists of an initialization procedure, a vortex enhancement scheme, and a shallow water model, formulated in a geographical coordinate system. During model initialization, the operational wind field analyses and forecasts of a global model are postprocessed to remove unwanted features such as mislocated weak vortices. Then a synthetic vortex, constructed from the information given by operational tropical cyclone advisories, is implanted at the observed position. In the vicinity of the storm, the wind fields are modified such that the total flow at the storm center corresponds with the observed translation velocity. Geopotential height is adjusted to the wind field by solving a nonlinear divergence equation without the tendency term. The numerical model integrates the shallow water equations in a storm-relative circular domain over a period of 72 h, using the postanalyses of global model analyses and forecasts as time-dependent boundary conditions. The present research version of WBAR has been developed on the basis of 167 cases (13 storms) during the 1996 Atlantic hurricane season, using predefined operational deep layer mean global model analyses and forecasts and tropical cyclone advisories of the U.S. National Centers for Environmental Prediction. The predicted 12-, 24-, 36-, 48-, 60-, and 72-h mean position errors in 1996, verified against best-track positions, are 78, 129, 184, 235, 295, and 360 km, respectively, with corresponding standard deviations of 40, 76, 111, 133, 168, and 182 km. At all prediction times except 12 h, WBAR has a consistent 30%?40% positive skill (defined as negative relative error) relative to CLIPER, a statistical regression model consisting of climatology and persistance predictors. With that, the performance of WBAR is comparable to that of the best track prediction model available in 1996, the Geophysical Fluid Dynamics Laboratory (GFDL) model. In the 1996 Atlantic hurricane season, WBAR shows the following sensitivities: position errors are smaller than the mean position error of all 167 forecasts in the case of stronger storms, storms with smaller influence radii, storms at base date/time latitudes greater than about 15°, storms with higher translation speeds and, storms that move west- or northwestward. The seasonal performance is similar to that of VICBAR (Vic Ooyama's Barotropic Model), showing a decrease in skill later in the season. The new system has only a minor lack of skill relative to the GFDL model that results almost exclusively from three weak storms in 1996 (Gustav, Josephine, and Marco) and is possibly a consequence of inappropriate deep layer mean fields.