Time step dependency for Stokes grounding line

Extension of Elmer in computational glaciology
Post Reply
gagliar
Posts: 69
Joined: 04 Sep 2009, 16:34
Location: LGGE - Grenoble
Contact:

Time step dependency for Stokes grounding line

Post by gagliar » 29 Jan 2020, 11:08

Hi all,

You may have seen this paper in TCD ( https://www.the-cryosphere-discuss.net/tc-2019-315/) by Brandon Berg and Jeremy Bassis regarding the "time step dependence (and fixes) in Stokes simulations of calving ice shelves". As they are solving the contact problem in the same way as us in Elmer/Ice, it might be that we encounter the same issues but also that the same fixes would work.

I have two questions here to the Elmer/Ice community:
- who has encounter similar problems? (or who is trying to solve GL dynamics with large changes or large hydrostatic unbalance?)
- who would have test cases to test the fixes proposed by Brandon Berg and Jeremy Bassis?

Regards,
Olivier

tzwinger
Site Admin
Posts: 94
Joined: 24 Aug 2009, 12:20
Contact:

Re: Time step dependency for Stokes grounding line

Post by tzwinger » 29 Jan 2020, 11:17

I think every initialization run of any real-world geometry with marine ice sheets involved (e.g., Antartica from BedMap2 or perhaps even BedMachine) faces that issue, as shelf-geometries usually are inconsistent with floatation.

martinw
Posts: 1
Joined: 29 Jan 2020, 13:11
Antispam: Yes

Re: Time step dependency for Stokes grounding line

Post by martinw » 29 Jan 2020, 13:22

I have been using ELMER/Ice to investigate the evolution of basal melt channels on ice shelves. I have encountered some problems with oscillating surface heights in my results when I reduced my time steps. To investigate this I ran a simulation of a freely floating ice shelf with uniform basal melting and uniform surface accumulation. When melting and accumulation were in balance there were no oscillations. However, when surface accumulation exceeded basal melting the position of the ice-shelf surface would oscillate about an upward trend rather than increasing monotonically as expected. I have yet to vary the size of the time step to determine whether the effect this has on the size of the oscillations.

Having briefly read the Berg and Bassis paper, it appears that accounting for the acceleration term in the conservation of momentum may help to solve this. But I have not looked into this yet.

cmosbeux
Posts: 2
Joined: 31 Oct 2017, 00:36
Antispam: Yes

Re: Time step dependency for Stokes grounding line

Post by cmosbeux » 30 Jan 2020, 01:23

Hi all,

To answer to the two questions :

1) Recently, I have been using the Stokes solver in a similar context than the one mentioned in Berg and Bassis paper: super-buoyancy of an underwater foot at the front of an ice shelf triggering an uplift of the ice front (called a rampart-moat in Scambos and others (2005)). I have been comparing the response of the Stokes solver, in terms of deformation and maximal stress position, to the response of the linear elastic solver in the same condition.

I could see a small time-step dependence in the early deformation of the ice. The strain-rates seem indeed overestimated when using very small timesteps and the shelf probably deforms too fast. However, I noticed that changing the timestep, as far as we still have a convergence of the model at early-stage, does not really affect the response of the ice shelf when we get closer to the hydrostatic equilibrium (later in the simulation) but only the time needed to get to this equilibrium (which is what I really cared about in my case).

2) Since Berg and Bassis use a similar 2D-flowline geometry than the one I used, I can send you my files if needed (pretty standard setup). I have some python/bash scripts to automatically create the ice shelf geometries (with a configuration file containing the parameters such as the ice thickness or the size of the underwater foot and the size of the elements) and call Gmsh and Elmergrid to create the mesh in Elmer format. Then, I have the .sif I used for the simulations and the bash scripts to run everything automatically on my machine.

Here after you will find an example of the EPS I can compute once the uplift at the front has occurred and for an underwater foot of 100m and initially located 10m below the water surface (mesh resolution in this area is ~2.5m).

Best,
Cyrille
Viscous_EPS_L1000m_l100m.png
(73.43 KiB) Not downloaded yet

gagliar
Posts: 69
Joined: 04 Sep 2009, 16:34
Location: LGGE - Grenoble
Contact:

Re: Time step dependency for Stokes grounding line

Post by gagliar » 31 Jan 2020, 14:56

Hi Cyrille and Martin,

What would be great is that you test the fixes proposed by Brandon Berg and Jeremy Bassis on your setups to see if it changes the dependence to time step of your simulations?
My understanding of this fix is that you simply need to reintroduce the acceleration term in the force balance, so to switch from Stokes to Navier-Stokes.

Regards,
Olivier

joeatodd
Posts: 36
Joined: 02 Feb 2012, 18:49
Antispam: Yes

Re: Time step dependency for Stokes grounding line

Post by joeatodd » 03 Feb 2020, 14:50

Hi all,

I had previously done some investigations with a basal channel mesh. I have been rather busy recently and haven't had time to continue investigations into this, but Brandon provided me with the mesh they use in their investigations (subaerial calving at the right margin). I can provide this (and a SIF) if someone wants to investigate.

Joe

Post Reply