Ensembles of numerical weather forecasts can be used to produce probabilistic predictions of future tropical storm position. The spread between ensemble members can be used as a measure of forecast uncertainty. However – raw ensemble spread does not account for any errors of the models in the ensemble.

The 11 member GFS ensemble is used to make probabilistic predictions of tropical storm position. The ensemble mean position and variance about the mean (the sample variance) are computed for forecasts from the Atlantic Basin from 2002. These two quantities are compared to the actual position of the storm and the variance of the ensemble members about that true position (the population variance). Corrections to the ensemble mean and variance about the mean are then computed via linear regression, and applied to future forecasts. Results from 2003 Atlantic basin tropical storms will be presented.