Question about GeoClaw SGN with continuously moving bathymetry

[Jialing Dai]

Dear all, 

I'm using the GeoClaw v5.14.0 SGN solver for a landslide-tsunami problem with continuously evolving dtopography, and I'd like to ask about the intended applicability of the SGN implementation to moving bathymetry.

In our case, the paired SWE and SGN runs use the same executable, forcing, grid, AMR, and numerical settings; the only effective model difference is bouss_equations=0 versus 2. The SWE solution remains moderate, whereas the SGN solution develops very large early free-surface excursions (about −127 m at 32 s and −433 m at 52 s).

We have done several checks:

• increasing the SGN linear-solver iteration limit does not remove the anomaly;
• the large depth changes occur in the FV update after the PRE-SGN momentum correction;
• using the refreshed bathymetry in PRE-SGN changes the solution, but does not consistently remove the anomaly;
• spatial bathymetry-gradient/curvature terms also have a measurable but location-dependent influence;
• in a small-amplitude flat-bed moving-bottom benchmark, the numerical generation response is close to an unfiltered coefficient (C\approx1), rather than the linear water-wave factor (C=\mathrm{sech}(kH)).

From the source, we understand the SGN operator is assembled from the instantaneous spatial bathymetry and its derivatives, while we could not identify a separate moving-bottom temporal/history contribution.

Could you please clarify whether the current SGN formulation and PRE/FV/POST splitting are intended to be used with continuously time-dependent bathymetry? In particular, is there a derivation or reference establishing the moving-bottom form implemented in GeoClaw?

Best regards,
Jialing Dai

[Randall J LeVeque]

Hi Jialing,

Sorry for the slow response due to travel and the CLaSH Annual meeting (relevant to your interests, I think, where it was nice to see lots of people are now using or planning to use D-Claw for landslide and debris flow modeling).

Regarding the problems you observe with SGN for moving topography, perhaps you can give some more details about where the Boussinesq terms are turned on and where you are observing the problem.  These terms have mostly been tested in the context of relatively small amplitude waves and I can imagine things might break down if they are used in the landslide region itself, particularly if the dtopo file you are using specifies very large perturbations of the surface that perhaps is also non-smooth and might have lots of high-frequency components.  The derivation of the depth-averaged equations are not valid in such a regime either, and so it is not clear that these terms would necessarily improve the solution over what the more robust shallow water equations do in such regions.   If this is where the problem arises, it may be worth trying to restrict the SGN terms to regions away from the landslide (where they may be more useful and stable in modeling the dispersive tsunami that is generated as it propagates away and decays).  

There is also a setrun  parameter rundata.bouss_data.bouss_tstart  that can be used to specify when to start using the Boussinesq terms, which we originally added because of problems seen at early times in other tests with large amplitude initial waves.

Regarding the manner in which the moving topography is incorporated in the GeoClaw solution, the topography grids are updated every time step during the time that dtopo is active, while the fluid depth h is left alone. This means that the water column above each grid cell is simply shifted up or down with the topography, giving an instantaneous motion of the sea surface that exactly matches the sea floor motion, as you observed in your test.

Alex Dolcimoscolo gave a talk at the recent GeoClaw Users' Workshop on implementing the Kajura filter in GeoClaw, which I think you are referring to, but this hasn't been incorporated more generally yet.

Hope this helps,

  Randy

On Sun, Sep 20, 2026 at 8:02 AM Jialing Dai <jialin...@gmail.com> wrote:

Dear all, I'm using the GeoClaw v5. 14. 0 SGN solver for a landslide-tsunami problem with continuously evolving dtopography, and I'd like to ask about the intended applicability of the SGN implementation to moving bathymetry. In our case, the

ZjQcmQRYFpfptBannerStart

This Message Is From an Untrusted Sender

You have not previously corresponded with this sender.

See https://it.uw.edu/email-tags for additional information. Please contact the UW-IT Service Center, he...@uw.edu 206.221.5000, for assistance.

 

ZjQcmQRYFpfptBannerEnd

--
You received this message because you are subscribed to the Google Groups "claw-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to claw-users+...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/claw-users/7c1d095e-4eea-45ef-a1b2-328e79721a6bn%40googlegroups.com.

[Jialing Dai]

Thanks Prof. Randy! Following your suggestions, I find some parameters related to SGN might be useful for my case: bouss_tstart,  bouss_min_level, bouss_max_level. I will do the test this week.

Jialing