Posted Wed, 26 Sep 2018 08:54:55 GMT by ToUbY
Hey there,

at the moment I am working on my master thesis. Therefor I need to calibrate a FeFlow groundwater model with FePEST and create several fields with a Monte Carlo simulation afterwards.

So for a field with like 70 observations and 100 parameters (pilot points) the calibration reaches a phi of lets say 1.6. When I run a post calibration Monte Carlo (with auto nullspace projection and optimization of the objective function on) afterwards the phis of the MC fields have values between 8 and 100 so they are up to 2 scales bigger, which makes a comparison between those fields nearly senseless.
Adding one or two iterations of optimization runs doesn't improve the phis to much and more iterations are just too time consuming.

My next idea was to reduce observations to 20 and pilot points to 24 so there would be more "space" for the MC results to reach similar phis. But actually calibration resulted in an even lower phi of 0,0027 and those phis of the MC are around 8-100 again.

So my question is how to reduce the phis of the MC simulation to get comparable fields for uncertainty calculations. Is there any other way or settings within the calibration or post-calibration MC interface of FePEST I should change or am I just missing/misunderstanding something.

So far I am only using FePEST and haven't worked with PEST itself.

I am happy for any suggestions. If you guys need further information please tell me.

Best regards

Posted Sun, 30 Sep 2018 11:25:22 GMT by Björn Kaiser
It's difficult to give suggestions without knowing more details about your model. I suggest to contact the support:

You must be signed in to post in this forum.