There are plans on their way about how to do this. The big issue is, to which degree it is possible to combine models in a systematic way and not just do "combinations of opportunity". However, there are not many completed simulations yet. It would be quite easy to do mixed ensembles; each simulation is independent, so it does make sense to mix and sea.
Actually, there is significant dependencies through similar model schemes - most models are not independent. Is there an attempt planned to look towards an objective way to combine the ensemble into a subset that still maintains most of the 'information' on spread and variability?
Many attempts have been tried so far in the scientific papers (filling the matrix with statistical methods, sub-sampling the ensemble keeping « independent » runs). Not sure that the community has reached an agreement on the best way to do it. The answer may depend on the user application
