Ensemble variance analysis - parameter worth?

[cnicol-gwlogic]

Hi, this is more than likely a daft question, but is there a pyemu method, or would it be an ok-ish idea to fudge using pyemu.EnDS.get_posterior_prediction_moments(), to assess the conditioning effect of (a) parameter(s) (eg if we fixed them) on a given "prediction's" uncertainty?

The fudge I am thinking of is to expand the obs ensemble with obs of parameters, and then remove selected pars from the ensemble that way. But I am guessing this will wreak havoc on the internals of EnDS() (??).

Any ideas? Thanks a lot.

[rhugman]

Hey Chris - we've applied EVA to determine "parameter importance". I confess when I did so I ripped out the pertinent pieces of code in pyemu and made a standalone fnx. I cannot quite recall why I did that instead of using EnDS...as you mention, iirc the way the fnxs are setup in pyemu are not streamlined for this use case...

[cnicol-gwlogic]

Ok thanks Rui, appreciate it.