Estimate objective function threshold for Null space Monte Carlo

[VirkNu]

Hi,

I have a groundwater model that needs to estimate areas contributing recharge to wells through backwards particle tracking. The parameters estimation has more parameters than observations and is therefore regularized with SVD and Tikhonov. I want to use Null space monte carlo to create many realization of the parameter sets, preferably though python and/or pyemu.

I need threshold for which realizations to evaluate if realizations are acceptable. Currently I have set an objective function threshold based on "hydro-sense".

But is there a correct way of estimating an objective function threshold to evaluate the realizations and is it hidden somewhere in pyemu?

[jtwhite79]

I think this will always be subjective (unless we somehow have a perfect model with a perfect bayesian sampler - one can dream!). I tend to use a more relative threshold like culling the X% worst realizations rather than an absolute threshold phi value. I guess once you have all the realizations gathered up, might be worth a few experiments of how the threshold choice propagates thru to the contributing areas?

[VirkNu]

I'll try it out!

Thanks for the answer.