Spiky Ice Sheet Surface in transient flow line model

Extension of Elmer in computational glaciology
Post Reply
alden
Posts: 7
Joined: 02 Jun 2020, 20:33
Antispam: Yes

Spiky Ice Sheet Surface in transient flow line model

Post by alden »

Hi all,
I have been trying to run a thermo-mechanically coupled transient land-terminating flowline model based on an EISMINT test (i.e. no flow bottom, SMB(x), Surface Temp(y), no ice at the start). The setup is fairly basic, it is really just a modified version of the transient toy flowline example model on a larger flat bed. My issue is that as the simulation progresses, the surface of the ice sheet oscillates, and when the ice sheet approaches a steady-state profile it settles into a fixed spiky pattern. The spikiness causes errors in the velocity solution which in turn causes errors in the temperature solution. I know that increasing the mesh resolution near the terminus localizes the effects of the problem to near the terminus. This indicates that the instability stems from the terminus. The reason for not wanting to increase the mesh resolution is that because the terminus if free to move, a large portion of the domain must be run at a finer mesh resolution resulting in simulations that are prohibitively slow with my current computer setup.

My question is, is there a way to alleviate this problem so as to get more reasonable solutions while still using a relatively course mesh resolution?

Thanks,

I have attached an image from 40 thousand years into a 10km horizontal resolution simulation, as well as the sif, and the geo file to generate the mesh used in this simulation.
In the figure, the ice sheet is approximately 3km tall at the divide and 500 km long.
Attachments
spikey_ice_sheet.png
spikey_ice_sheet.png (61.53 KiB) Viewed 1130 times
EISMINT_test.sif
(12.2 KiB) Downloaded 57 times
EISMINT.geo
(511 Bytes) Downloaded 49 times
kevinarden
Posts: 653
Joined: 25 Jan 2019, 01:28
Antispam: Yes

Re: Spiky Ice Sheet Surface in transient flow line model

Post by kevinarden »

Without a mesh density increase, higher order elements is required. It will still require more memory and compute time.

see page 131 Elmer Solver manual

http://www.nic.funet.fi/pub/sci/physics ... Manual.pdf
raback
Site Admin
Posts: 3752
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Spiky Ice Sheet Surface in transient flow line model

Post by raback »

Hi

Free surface loosely coupled with N-S is only conditionally stable. I guess the only thing you can currently do is to reduce the timestep.

-Peter
alden
Posts: 7
Joined: 02 Jun 2020, 20:33
Antispam: Yes

Re: Spiky Ice Sheet Surface in transient flow line model

Post by alden »

Thanks for the timely replies.

I hadn't thought about higher order elements, and I will look into that.

I had reduced the time step by a factor of ten (to a tenth of a year) previously, and had no success in disappearance of the spiky surface. I will try to reduce it further to see if that helps as well.

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

Re: Spiky Ice Sheet Surface in transient flow line model

Post by tzwinger »

Hello
there is active work going on to get a remedy for this instability. But until then, I am afraid, the only way to avoid instability is to reduce the time-step size. In particular, the no-slip condition on the bedrock does not help. What is your minimum flow-depth (i.e., your RefZs - 0.1 ?). Increasing this might help.

Regards,
Thomas
fgillet
Posts: 41
Joined: 30 Sep 2010, 16:58

Re: Spiky Ice Sheet Surface in transient flow line model

Post by fgillet »

Hello;
I think it is also sensitive to the stabilisation method. By default the Free surafce solver uses the Bubbles; SUPG stabilisation is also possible; However in this case the stabilisation parameter is optimal in steady-state. There is a lot of litterature on this subject and possible stabilisation parameters for transient simulations and unstructured grids.
In the Thickness Solver I have implementend something from this paper: Akin and Tezduyar, Calculation of the advective limit of the SUPGstabilization parameter for linear and higher-order elements, Comput. Methods Appl. Mech. Engrg. 193 (2004).
I also saw that in the level-set solver the stab. parameter is also limited by dt/2.
I never did a clean comparison for the Thickness solver coupled with Shallow shelf; but results looks smoother than bubbles.
Might be interesting to properly look at this issue if somone has time.

Also increasing the bdf order could help.

Fabien
alden
Posts: 7
Joined: 02 Jun 2020, 20:33
Antispam: Yes

Re: Spiky Ice Sheet Surface in transient flow line model

Post by alden »

Thanks again for all of the help. It does appear that trying to enforce a no slip boundary condition was making the problem worse. The simulations seem to run better if there is some amount of sliding allowed near the terminus.
Post Reply