Posted Thu, 17 Jul 2014 17:37:25 GMT by
Im working both Fepest 6.1 and 6.2  .  However,I encountered that the 6.1(default setup) is more flexible for highly heterogeneous  calibration than newest 6.2. I think the problem is with Regularization setup change in new  version, might be Correlation length or co variance matrix.Please help me to make changes in regularization setup, similar to 6.1 setup. Thanks
Posted Thu, 17 Jul 2014 18:06:06 GMT by Carlos Andres Rivera Villarreyes Global Product Specialist - FEFLOW
Dear Joachim,

Please notice that FePEST 6.1 was a quasi beta version compared to FePEST 6.2, which is fully developed. I strongly recommend you to use 6.2 version.
What exactly would you like to do? Could you be more especific in what kind of regularization setings you need to chance?


Posted Fri, 18 Jul 2014 20:03:52 GMT by
Hi Carlo,
Thanks for the respond,
Regarding my work,I'm working with highly heterogeneous, fractured, about 5000sqkm area,  6 layer model.I calibrated the model several times to understand the influence of changing boundary conditions because of the the groundwater dynamics of the area is not well understood.

I used Feflow 6.1 (PEST- pilot point) earlier and I could calibrate the model accurately (perfectly with groundwater heads) although It took very long time for inversion. After releasing the Feflow 6.2 last December,I started to use FePEST. I have access Server and with the aid of BEOPEST parallel processing, the calibration is very fast. But,objective function start to not to fall down and calibration stops without achieving good fit.However ,Fepest 6.1 gives best mach for the same model and I'm fully satisfied with the results.Actually,I have followed the calibration procedure more than 500 times with different boundary conditions.

I do not need any calibration regularization method at this moment,I use only initial parameter,Upper-lower levels only for the Hydraulic conductivity calibration.Logically,As I think,If I use manual calibration,  fepest 6.1 ,or fepest  6.2  to get the parameters and the parameters fits to manage observed heads with modelled heads,It does not have any effect on final results.

Earlier,I thought new FePEST uses regularization method in default setup to adjust and have smooth distribution without bumpy appearance of conductivity in space.

Please let me know the  suggestions.

Posted Fri, 25 Jul 2014 15:34:11 GMT by
who has similar experiences as mine? please share....
Posted Fri, 01 Aug 2014 09:29:40 GMT by
I tested the calibration with some simple models as well. It shows same problem again and again. FePEST 6.2 does not give the good results and objestive function does not drop after some value. please help me with this manner.
Posted Fri, 01 Aug 2014 15:33:46 GMT by Carlos Andres Rivera Villarreyes Global Product Specialist - FEFLOW
Dear Joachim,

Indeed there are many changes in the FePEST version 6.2, especially the regularization settings and interpolation techniques (e.g. Kriging) are different. This is because [i][b]they need to be adapted to specific case studies[/b][/i]. For example, a "typical" misunderstanding is that variogram parameters are applicable for all the models and all parameter types. You may take a look on Geostatistical books to review the different methods for empirical variogram fitting, you will notice that these parameters depends a lot on sampling design, parameter types (conductivity, transmissivity, porosity, etc.). Therefore, FePEST cannot suggest values on which the modeller has the responsibility.

You may take a look in the following settings in FePEST version 6.2:

1) Kriging configuration (variogram type, range, nugget, etc.).

2) Tikhonov regularization settings (Objective function, weight factor and weight adjustment). These options were not available in FePEST Beta version. Especially, objective function values can control strongly the balance between observation over-fitting and/or adequate (or geologically-realistic) parameter distributions.

3) SVD regularization: Adjust number super parameters (Default value is zero).



You must be signed in to post in this forum.