A supermodel is an ensemble of different models in which the models interact with one another in run time.
It has been shown that a supermodel can out-perform both the individual models and any average of the model
outputs. I will review the supermodeling concept, as presented recently by Frank Selten, with stress on the relationship
between supermodeling and data assimilation, as well as the conditions under which supermodeling is better
than ex post facto averaging. Finally, I will explain why supermodeling is even more appropriate for short-range weather forecasting
than it is for climate projection.