We present a method for choosing a member in a global circulation model (GCM) ensemble forecasting. We show that the r2 correlation between members and independent data is related to the flux of information entropy contained into the forecasts. This allows to classify the members in terms of goodness-of-fit. To validate the method we use several ECMWF-ensemble forecast from the DEMETER project, and the ERA-40 database. We compare ERA-40 actual estimates with our entropy-based choice of the DEMETER members in each of the 180 days, obtaining consistent results. Our hypothesis is based on energy-balance, maximum entropy production considerations so that the method might only be applicable to global simulations.