Most weather forecasting models are deterministic in that they forecast exactly one outcome. However, some models are run over and over again with slightly different initial conditions to capture a better understanding of what the range of outcomes could be. Models are usually started with observations from weather stations, which could have small errors in how they are recording data. Ensembles try to account for slight variations in these initial observations in order to show what the range of possible outcomes could be for a forecast model. The GFS and the Euro global models are both run as ensembles with variations between the ensemble models showing how different the future weather patterns could be based on small changes in the starting conditions. The ensemble form of the GFS is the GEFS (Global Ensemble Forecast System) and the ensemble form of the Euro is the EPS (Ensemble Prediction System). The GEFS uses 21 separate ensemble members, while the EPS uses 50. These separate ensemble members give probabilistic information about what weather patterns are most likely to occur.
The GEFS and the EPS are valuable in showing the anomalies, or how much different from normal certain atmospheric variables are based on the average of the ensemble members. One common ensemble output is the surface temperature anomaly, which would show whether the majority of ensemble members show that the surface temperatures are going to be warmer or colder than normal for a certain location. These anomaly forecasts can also be useful for determining how likely large scale patterns are in the 5-10 day range by showing if there is agreement between ensemble members about how far from normal certain atmospheric conditions will be. For example, one model run of the GFS showing a strong jet stream over a particular area is not convincing as the GEFS showing agreement between the ensemble members of anomalously strong upper-level winds, indicating that strong jet stream over a particular area.