Ensemble based probabilistic tropical cyclone forecasts

Ensemble variance can be used as a measure of forecast uncertainty, including forecasts of tropical cyclone track. Ensemble variance can be calibrated by comparing historical errors of ensemble mean forecasts, the ensemble mean forecast error variance, to the variance of ensemble members about the mean, the ensemble variance. Comparing Global Forecast System (GFS) ensemble mean forecast error variance to GFS ensemble variance from 2001, 2002, and 2003, a rescaling factor for GFS ensemble variance is calculated. Using raw and rescaled GFS ensemble variance as input, probabilistic forecasts of tropical cyclone track were made for the 2001, 2002, and 2003 seasons. Forecasts were made for lead times of 60 and 72 hours for storms in the Atlantic basin.

Probabilistic forecasts generated from rescaled, rather than raw, GFS ensemble variance showed a clear improvement in the estimate of ensemble mean forecast error variance, the reliability of 50% probability forecasts, and Brier scores. In addition, probabilistic forecasts generated using the rescaled variance had better Brier scores than climatology based probabilistic forecasts.