Will DISU be part of imod-python?

[phaehnel]

Dear developers,

I just wanted to ask if there are any plans to also include completely unstructured grids for MODFLOW 6 (DISU) in imod-python? Especially, the option to represent vertically staggered horizontal connections (so horizontal connections between cells in different layers) seems quite promising for, e.g., strongly tilted geological layers or glacial tunnel valleys with steep sidewalls.

There is some really interesting, recently published work regarding the discretization options in MODFLOW 6 and how they affect simulation results:

• Provost et al. (2025), https://doi.org/10.1111/gwat.13459
• Bardot et al. (2025), https://doi.org/10.1111/gwat.13273

The key point basically is that the vertical staggering with DISU is extremly helpful for strongly ditching aquifers. I also did some synthetic tests with a steep channel structure producing a connection between two otherewise vertically separated aquifers. Here, the vertical staggering with DISU also improved modeling results significantly.

They also developed a nice Python tool to convert a DISV package in flopy to a DISU package with vertical staggering for use with flopy (disv2disu in https://github.com/kerrybardot/loopflopy).

Since this can be very helpful in reducing structural errors in the simulation results for certain aquifer geometries, I wanted to ask if something like including DISU and vertical staggering options is planned in general for imod-python?

Cheers,
Patrick

[phaehnel]

Well I should have gotten to the bottom of things first:
https://deltares.github.io/imod-python/index.html#not-all-modflow-6-features-are-supported
I found my answer here.

[Huite]

Hi @phaehnel,

For what it's worth, I definitely intend to support DISU on the long term. What has been holding us back is a rich data structure that can represent the 3D unstructured topology in a way that xarray does for DIS grids, and xugrid does (reasonably) for DISV grids. The natural place for this is a Ugrid3d topology in xugrid, next to Ugrid1d (think SFR networks) and Ugrid2d (think DISV): https://github.com/Deltares/xugrid

However, the challenge with fully 3D unstructured grids is that geometric operations such the intersection algorithms (required for cross-sections, but also for translating between different topologies) are rather more complex than 2D.
Fortunately, full 3D isn't required for MODFLOW 6 DISU. Because the cells are prismatic, the 2D algorithms can be applied and the z-axis can be treated separately (this has performance and efficiency benefits as well). My concrete plans are to add a PrismaticGrid3d topology to xugrid. Once that exists, DISU support is relatively straightforward.

The staggered grids are relevant for our national model too, and for variable density modeling. In the first case, we have areas which consist clearly of one contiguous aquifer. However, hypothetical interfaces must be constructed to guarantee connectivity. This rather complicates the act of deriving the geohydrological schematization from the geological layers. Concretely, it means that our national models has 8 aquifers everywhere; with DISU we can adjust this easily. IDOMAIN = -1 in DIS and DISV helps, but is insuffcient. For variable density modeling, the benefit is clear too: we don't need additional vertical discretization in areas that are fully fresh or fully saline.

Unfortunately, I am rather busy with other things at the moment, so it seems unlikely that we will add DISU support within the next two years. Five years, probably yes. (We can quibble whether five years counts as "foreseeable future" like the docs page mentions.)

[phaehnel]

Thank you for the detailed insights and good to know that DISU is also relevant for the NHI. The infos help us a lot in planning how to build our model. I've been using xugrid intensively in the past months and it's a great and easy-to-use tool to work with once you've understood the xarray logic. Really appreciate it, that Deltares is putting so much effort into open-source software.

And I'm looking forward to the foreseeable future, whenever it may be.