Covariance Matrix for pestpp-ies

[DStrom1987]

Hi,

I have been trying to make use of a geospatial covariance matrix for prior monte carlo following some examples in the gmdsi tutorial notebooks.

I notice however that in the examples there is one cov-file supplied whereas the plproc workflow or if I try to generate a cov matrix form a .tpl-file generate one cov Matrix per parameter group.

Is there an example workflow on how to create a single covariance Matrix for multiple parameter groups or how to do a pest-ies run with noptmax set to -1 while supplying multiple covariance matrices for each involved parameter group?

also, having pilot point supplied with a dense pattern and manual points in various locations gives me covariance matrices that look rather strange when plotted. Nothing like the very symetrical and smooth examples in the tutorials.

Is the connection between the pilot points from my pst file neccesary? What purpose does it have? Or can you simply generate a gemeric covariance matrix specifying the distance and relationship parameters should have in space?

sincerely
David

[jtwhite79]

you can pass a pest-style .unc (uncertainty) file to the parcov argument if you want to have a cov matrix for each par group. the pattern in the cov matrix may not look like the ones in the tutorials bc its a function of the pilot point spacing and the geostruct. However unless you are using geostatistical hyperparameters, I think the cov matrix should still symmetric about the diagonal...

[DStrom1987]

Thanks Jeremy,

Sounds like that should solve my problem.
Perhaps my description was a bit off target.. my matrices are indeed symetrical around the diagonal. Its just that they have a more, in lack off better words, heterogenious look on either side of the diagonal than in the examples. Might still be in order i suppose given my pilot point pattern.

sincerely
David