Physical Initialization and Hurricane Ensemble Forecasts

Abstract

Ensemble forecasting of hurricane tracks is an emerging area in numerical weather prediction. In this paper, the spread of the ensemble of forecast tracks from a family of different First Global GARP (Global Atmospheric Research Program) Experiment analyses is illustrated. All forecasts start at the same date and use the same global prediction model. The authors have examined ensemble forecasts for three different hurricanes/typhoons of the year 1979. The authors have used eight different initial analyses to examine the spread of ensemble forecasts through 6 days from the initial state. A total of 16 forecasts were made, of which 8 of them invoked physical initialization. Physical initialization is a procedure for improving the initial rainfall rates consistent with satellite/rain gauge based measures of rainfall. The main results of this study are that useful track forecasts are obtained from physical initialization, which is shown to suppress the spread of the ensemble of track forecasts. The spread of the tracks is quite large if the rain rates are not initialized. The major issue here is how one could make use of this information on ensemble forecasts for providing guidance. Toward that end, a statistical framework that makes use of the spread of forecast tracks to provide such guidance is presented.