Model with all layers treated as confined

[Philip Margarit]

Hi All,

I am currently working with a MODFLOW 6 model to do some capture fraction mapping. The model was built by another organization with each layer including the surficial aquifer represented as a confined layer rather than convertible. Likewise, I noticed some of the head-dependent boundary conditions have their heads and bottoms (drn and RIV) below the cell bottom of the first layer, however the model was able to run when the layers were all confined.

I wanted to ask whether anyone has experience working with models that represent all layers as confined rather than convertible, and whether the resulting constant-transmissivity assumption has significantly affected stress-response predictions. My concern is that this assumption could potentially bias capture-fraction estimates, particularly transient capture fractions. If a water-table aquifer is represented as confined, transmissivity remains fixed rather than decreasing as saturated thickness declines. That could overstate hydraulic diffusivity and allow the response to pumping to propagate more efficiently than it would in a convertible/unconfined formulation which may overestimate capture fractions or change their timings. 

Has anyone ran into this in their work? Thanks all! 

[Christian Langevin]

There is a paper published in Groundwater on this topic.  Would be worth a look if you haven't seen it.

Simulation of Water-Table Aquifers Using Specified Saturated Thickness

by , , , , 

First published: 06 February 2014

 

https://doi.org/10.1111/gwat.12164Digital Object Identifier (DOI)

[Philip Margarit]

Hi Chris,

Thanks so much for sending me this paper. The guidance I have seen is that it is still useful to test the impact of the confined-layer assumption on the prediction of interest.

The part I am still trying to think through is how to interpret that test. If a model was calibrated with layers treated as confined, then the calibrated K field was estimated under the assumption of constant saturated thickness and, therefore, constant transmissivity for each layer. If that same calibrated model is then changed so the layers are convertible, that assumption no longer holds. Depending on the layer thickness and the degree to which simulated heads are below the cell tops, the baseline transmissivity field could change substantially.

So my concern is that differences between the confined and convertible simulations may reflect not only the effect of the confined vs convertible assumption, but also the fact that the convertible model is now using a different effective transmissivity field than the one estimated during calibration.

What I am still wondering is if there were two calibrated versions of the model that both adequately reproduced historical system behavior, one using the confined assumption and one using convertible layers, would the resulting capture fractions be significantly different?

So my main question is this: given the general guidance to test the impact of the confined-layer assumption, is it valid to simply change the layers from confined to convertible and compare predictions, or is that comparison only fully meaningful if both model versions are calibrated or otherwise conditioned under their respective assumptions?

--
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 visit https://groups.google.com/d/msgid/modflow/16436b5e-60bf-44ec-b1d8-ea53d22fae73n%40googlegroups.com.

--

Philip Margarit

Water Resources Science PhD Candidate

College of Food, Agriculture, and Natural Resources Science 

University of Minnesota, Twin Cities

651-262-3561

[Jeremy White]

Hey Philip - 

Short of calibrating both versions of the model in the same way (with only the laycon changed as you suggest) and then calculating the capture fractions for both models, I dont think its possible to say definitively what the impact of laycon would be for a specific case.   A few things to think about (at least what I was thinking about):

- What observations were used to calibrate the model?  Were there lots of observations in the region between the capture fraction wells and the simulated SW system?  Were the observations well assimilated/matched to the point that any subtle laycon-induced signal in the residuals was transferred to the parameters?

- What inputs were parameterized and how were they parameterized?  If pilot points or other geostatistical parameterization was used (and/or conductance of the SW representation), then there might be a chance the the observations (if located in the area of interest) to mis-inform the those parameters as a result of the model error signal coming from the laycon choice, yielding parameter compensation bias (for me, the SW conductance parameters are already kinda slop factors to soak up lots of missing/misspecified processes and scale in the immediate vicinity of the exchange).

- Given the capture fraction mapping is essentially a difference prediction, then maybe some of this hypothetical parameter compensation bias might cancel out?  

So to me, this seems like a very problem-specific issue and one that will be difficult to confidently understand without repeating the whole process with the layers as convertible...presumably, the layers were treated as confined for model stability reasons...so it gets more complicated...

To view this discussion visit https://groups.google.com/d/msgid/modflow/CAO%3DMhd1D-tQ0vykWGAq0o-2%2Bdh_MKN%2BU9EmaHGnc9tYDe5ankQ%40mail.gmail.com.

[Philip Margarit]

Hi Jeremy,

Thanks so much for your response and opinion here. I did some testing myself and found there was a significant difference but to your point, it would not be possible for me to say that the impact was because of laycon without recalibrating the model first. To make it even harder for me, the model has some RIV bot and DRN bot elevations below the cell bottom elevations in parts of the model. The difference was big enough that I would most likely need to change the layer bottom elevations to get the convertible model to run which would make a new conceptual model compared to the original. 

Because of that, I am not sure if I can test this out on my current project. However, I think that's an interesting research question for someone, if you had two calibrated models with different laycons, would your capture fraction map conclusions be the same? We are using them to create a boundary zone for a river reach, and the idea is to use the most conservative results to create that boundary. We initially defined the most conservative results as most conservative under uncertainty but maybe that definition could be wider for this specific problem. My understanding is that if the layers are set to confined, at worst, they would overestimate capture. Philosophically, I imagine that would line up with the "most conservative results" goal for this boundary. 

By the way Jeremy, I plan on using your new method, MODFLOW 6 ADJ, on this project. I did some initial testing of MODFLOW 6 ADJ against the direct perturbation method from Stan Leake and we have gotten some pretty great results that lined up well to the traditional method. Awesome work you guys are doing and I am excited to see where you go with that method in the future. 

To view this discussion visit https://groups.google.com/d/msgid/modflow/CAPMrFXncazAfmtNVJ%3Di2yFhXB1zm5RppF6E6CgPHiwLpW7t%3DXw%40mail.gmail.com.