What is it? An ensemble mean is the average of multiple quantitative climate model projections for a given climate variable (i.e., temperature).
Why is it used? A model predicts a given amount of change for a specific climate variable, but it is based on a specific set of initial model conditions and known and unknown uncertainties. Individual climate model projections should not be viewed as an exact picture of future climate, especially since multiple predictions even from the same model do not agree. This variance between what models predict for future climate change creates a need for summarizing the ensemble of several projections considered. An ensemble mean is one way to summarize a collection of projections but it is not necessarily the most appropriate measure for climate change.
When should it be used? It is difficult to define when an ensemble mean should be used because each climate question being asked will have a different amount of sensitivity to averages. An ensemble mean is not an appropriate summary statistic when models show negative and positive change, because the mean would be show close to zero change when in reality there is higher uncertainty.