Encountered a pop up during PEST calibration run saying Exception EOleSysError

[Ishita Bhatnagar]

Dear All, I got an error pop up during pest calibration run, when the run was at optimization iteration 5. Could you tell what does this error mean because the pest is still running further. I had put the pest on run yesterday. Error message is attached. 

Thanks and Regards
Ishita Bhatnagar
Research Scholar (Water Resources )
Department of Civil Engineering
Indian Institute of Technology Delhi

[Jakab Andras - Gmail]

This is a system error not directly related to PEST or Visual MODFLOW. BPL files are Borland Package libraries. My feeling is that there might be some kind of incompatibility issue here that is due to an unsupported operating system.

Please remember that the Visual MODFLOW Classic interface is no longer supported, especially if you run it under newer operating systems. Such error messages might be encountered, therefore. Nevertheless, the GUI is fully functional under most circumstances.

On another note, the PEST version implemented in VMOD Classic is version 4 if I'm not mistaking. That is a very old version of PEST that doesn't use pilot point parameterisation and regularization. Therefore it relies on manual regularization using zones with piecewise homogeneity. In most of the cases,.however, this kind of regularization results in a non-unique inversion process and unsuccessful PEST runs. I strongly recommend reverting to the Visual MODFLOW Flex interface which implements the latest features of PEST.

--
This group was created in 2004 by Mr. C. P. Kumar, Former Scientist 'G', National Institute of Hydrology, Roorkee. Please visit his webpage at https://www.angelfire.com/nh/cpkumar/
---
You received this message because you are subscribed to the Google Groups "MODFLOW Users Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to modflow+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/modflow/CAF1FOW0KLU8f_bw3HYtz%2BTfYAoaznSTvZQc_3PM8vzQLW1Je0w%40mail.gmail.com.

[Ishita Bhatnagar]

Dear Jakab 

Thank you for a detailed explanation of the issue. The error, in my case, was that of unrealistically high upper and lower bounds, driving PEST crazy. I do take note of the PEST version, you are correct. I have had to do intense manual adjustments to ensure PEST does not perform recklessly. The issue is that my model grid size is 2km by 3km and I have limited observation points (just 500-600 points) while the model area is approximately 0.4 million sqkm, thus it is difficult to provide distributed parametrization and opt for pilot point calibration. I am targeting at minimizing the average head residuals for each zone. However this too is very challenging. For instance for my steady state model, I could boil down to an overall mean residual of 2.991m after extensive manual and steady-state PEST calibration, but the standard error is 24.03m and variance of residuals is 577.6. Are these figures acceptable if I am only looking to achieve a zone-wise well-calibrated model? 

Thanks and Regards

ISHITA BHATNAGAR
Research Scholar (Water Resources Engineering)
Department of Civil Engineering
Indian Institute of Technology (IIT)  Delhi
Alternate Email: cez1...@civil.iitd.ac.in
Ph no.: +91 8394834109

To view this discussion on the web visit https://groups.google.com/d/msgid/modflow/CAKmr6MsQJZgYpy08TKVtUczJbApXCDm-2kL%2BZ40dB1XTJot_gA%40mail.gmail.com.

[Jakab Andras - Gmail]

I don't see any reason why not use pilot points with the number of observations you mention. You probably will have to place pilot points denser in the vicinity of the observations and more sparsely farther away from them. Add regularization and covariance matrices to the regularization observations and give it a go; create the covariance matrices taking into account pilot point separation distances.

There are of course many things you have to pay attention to when using PEST. The most important probably is that your model should have integrity, i.e., make sure that conceptually is acceptable and manually calibrated to a reasonable extent, i.e., your starting parameters are approximately at the center of their prior probability distributions (i.e., you trust them being close to their "real" values).

To view this discussion on the web visit https://groups.google.com/d/msgid/modflow/CAF1FOW0v%2BQzqN5HQyweq2PPg3LHYH1d%2BJh5U8jpQcw6U8L6%2BiA%40mail.gmail.com.

[Ishita Bhatnagar]

Dear Jakab

As always, thanks for the detailed suggestion. I will definitely look into it. I have one doubt, for pilot point calibration I need to have measured data points for say conductivity (if I am calibrating K), but the problem is that I do not have any such information for my basin. It's a blind fold. Whatever district or state level lumped information I had I used it to generate initial conductivity zones. So that's why I had dropped the idea of pilot points. Let me know if I understood correctly. 

Yes I am performing extensive manual calibration and directing PEST. Thanks. 

Thanks and Regards
Ishita Bhatnagar
Research Scholar (Water Resources )
Department of Civil Engineering
Indian Institute of Technology Delhi

To view this discussion on the web visit https://groups.google.com/d/msgid/modflow/CAKmr6MuaOOYcM5rb0DQnzjOYZdvAJR38XYu0sBcWvLOHHrOCfg%40mail.gmail.com.

[Jakab Andras - Gmail]

Ishita,

You don't necessarily need to have measured point conductivities. They are good of course if you have them, but not necessary. I.e., if you performed manual calibration, you already have an idea of the magnitude of the K-values (or any other parameter values) for your layers and possibly their rough horizontal variance. You may then use these manually calibrated values as initial values for the pilot points if you trust them. Set some reasonable bounds for these and, unless it's absolutely necessary, also dispense with the horizontal zonation and let PEST look for heterogeneity. 

I would like to remind less experienced modelers that PEST is not meant to "calibrate a model" that is structurally not correct. Only use PEST if you are confident that the model has integrity. Of course, one may use it also to detect structural problems with the model if that's the goal (i.e., if the outcomes of the inversion process bear no physical meaning at all), but I'm afraid in many cases it is applied with the expectation to "fix" a model that is conceptually wrong.

András

To view this discussion on the web visit https://groups.google.com/d/msgid/modflow/CAF1FOW36HWuLdOabFsASvq7VCRVr1C4TfCD9dXAfVum%3D3YCJ1A%40mail.gmail.com.