Tuesday, April 29, 2008

Null Space Monte Carlo - Almost There!

The latest version of Groundwater Vistas (5.17 Build 2) now has the Null Space Monte Carlo for PEST Version 11.5 hooked up. If you have the advanced version, you will see it light up now on the Pest menu. I must caution you that it has not been fully tested yet. I am working on that today so there will likely be additional updates by tomorrow as I break and fix things. I am preparing a tutorial on it for my upcoming course in late May. Hopefully everything will work.

I also fixed a couple of bugs related to importing existing models. The first was that GV was not importing the GMG package input file. The second involved not properly recording the layer type codes when importing the HUF package. There was also an odd problem with importing arrays from older vesions of GMS.

Monday, April 21, 2008

New Stuff in Version 5.16 Build 10 to Build 16

It's been a busy 2 weeks but I have a free minute now to update you on the latest changes in Groundwater Vistas. These are all enhancements (no bugs to report):

- On Model/Pest/Parameters we added several new buttons to make it easier to set up parameters. There is an automated parameter setup, a button to clear all parameters so you can start over, and a button to make all parameters use log transform. For models with lots of parameters, these buttons make it much less tedious to enter all data in the spreadsheet.

- Added the SUPCALC utility to the Model/Pest/SVD Assist menu. This program computes the recommended minimum and maximum number of super parameters when using SVDAPREP.

- Added a restart option on the Pest menu. If you stop a Pest run, this command restarts it either from the last full iteration or from the last model run.

- Added a restart and stop option on the SVD Assist menu for this style of Pest run.

- We have had a few problems with people typing in the path to the working directory under Model/Paths to Models. Some users have entered a trailing "\" character which then causes problems in writing files and running some programs. GV will now look for this and delete it if it is found.

- We added an option under BCs/Delete/Reach that asks if you want to delete the reach in all layers. Before it would only delete in the current layer.

- When importing stream boundary conditions from a text file, we did not have the option to read inflow rates and tributary numbers. You can now import inflow rates and up to 3 tributaries.

You will notice a significant number of changes related to Pest in the coming months, particularly on the SVD Assist and the Null Space Monte Carlo menus. The latter is not yet fit for human consumption but should be working by late May. I plan on covering this topic in my next seminar in Tampa. Those two aspects of Pest are extraordinary and set Pest apart from any other inverse model.

Monday, April 07, 2008

MODFLOW & Vertical Leakance Coefficient

I seem to get a lot of questions about vertical leakance vs. vertical hydraulic conductivity in MODFLOW. In the original versions of MODFLOW (88 and 96), the Block Centered Flow (BCF) Package required you to enter a value for vertical leakance or VCONT. That value was based on the half-thickness of layers and the vertical K of the layers. All preprocessors that support MODFLOW88/96 compute this for the user, although there are slight differences in how they compute VCONT. Like most preprocessors, Groundwater Vistas computes VCONT based on the saturated thickness of the layer. So as the starting heads change during the evolution of a model, the VCONT may change as well for unconfined layers. However, during any particular simulation, VCONT is constant, even if the saturated thickness of the layer changes.

In MODFLOW2000, the BCF Package is still available with the same assumptions. A new package was introduced that can be used in place of BCF. The new package is the Layer Property Flow Package (LPF). Instead of a vertical leakance, LPF takes as input the vertical hydraulic conductivity directly and computes a vertical conductance internally. The main difference is that it updates vertical leakance at each solver iteration for partically saturated convertible layers. This was a big change and caused many models developed in MODFLOW96 to give different results in MODFLOW2000.

MODFLOW2005 is essentially the same as MODFLOW2000 with one exception. A new option was added that makes LPF more BCF-like. You can choose to have the vertical leakance computed once at the beginning of the simulation and then it remains constant as in BCF.

Which is correct? I think that theoretically the LPF approach is the better one. It is hard to say, though, that all of our older MODFLOW96 models are incorrect. Perhaps with all of the other uncertainties in a model it really does not matter.