MALI calving dt convergence tests #386

matthewhoffman · 2022-05-05T17:58:30Z

This merge adds a new set of tests for MALI that explore convergence with timestep of the calving physics. This is added in a new test group that runs a configuration multiple times with the adaptive timestepper restricted by different fractions of the calculated calving CFL limit. The results are then analyzed and a summary plot is produced. This test can be applied to a number of different calving laws and meshes and using either data velocity or prognostic velocity. These set of tests are not intended to be run routinely as part of integration, but instead are a tool for evaluating development of calving physics. A test suite is also included that runs a number of these tests that have been evaluated as being useful. There are some known ways this test group could be improved, but given its limited intended usage, they are not worth developing further at this time.

A MISMIP+ smoke test is also added as part of this PR. The calving tests were first developed to use the MISMIP+ configuration, which was not yet in COMPASS. Later, things were reorganized to make the calving tests more general and support multiple meshes. Having a MISMIP+ smoke test is generally useful, so that test was retained in this branch after reorganizing things.

Note that this PR includes an update to the MALI-Dev submodule to include the merged MALI PR here: MALI-Dev/E3SM#33.

pep8speaks · 2022-05-05T17:58:34Z

Hello @matthewhoffman! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

In the file compass/landice/tests/calving_dt_convergence/dt_convergence_test/__init__.py:

Line 28:80: E501 line too long (86 > 79 characters)

Comment last updated at 2022-05-11 22:38:47 UTC

matthewhoffman · 2022-05-05T22:25:09Z

@trhille , this is ready for review from you. I haven't added the docs yet, but the code changes are done and I thought you might want to start looking at this before I have the docs complete. To test this, you'll need to be using the accompanying MALI PR: MALI-Dev/E3SM#33. I'll update the submodule in this PR after we merge that. For testing, I recommend running the new calving_dt_convergence test suite. Note that the last two tests in it use the FO solver and take a few hours each. But the non-FO tests collectively should finish in about an hour, so I think it would be acceptable to just test those (e.g. if your job times out before the FO tests finish). When the tests are complete, just have a look at the png image that gets generated in each test directory. The results should look similar to the images I've compiled here: https://docs.google.com/presentation/d/1YohEQBxoV84qVwrUJVM-WoB51oOv4eqTblqGx6ML3SE/edit?usp=sharing

(I'll wait to have Xylar review until I have the docs done.)

trhille

@matthewhoffman, this is really impressive! I've finished going through the code (only found minor copy-paste errors and had a few questions), but I'm still running the tests.

compass/landice/tests/calving_dt_convergence/__init__.py

compass/landice/tests/calving_dt_convergence/albany_input.yaml

compass/landice/tests/calving_dt_convergence/dt_convergence_test/__init__.py

compass/landice/tests/calving_dt_convergence/namelist.landice

compass/landice/tests/calving_dt_convergence/run_model.py

compass/landice/tests/mismipplus/albany_input.yaml

trhille · 2022-05-09T22:37:24Z

compass/landice/tests/mismipplus/namelist.landice

+config_start_time = '0000-01-01_00:00:00'
+config_run_duration = '0005-00-00_00:00:00'
+config_adaptive_timestep = .true.
+config_adaptive_timestep_CFL_fraction = 0.9


Why is this different from the CFL fraction of 0.95 in https://github.com/MPAS-Dev/compass/pull/386/files#diff-acd02e538bc2c87b44f42baa553a3630dd419f20c83568eac48f8b2a9fc497b7 ?

(that link is to the namelist.landice file in dt_convergence_test)

trhille · 2022-05-10T16:19:28Z

I ran the calving_dt_convergence test suite on Badger using an exe compiled at the head of matthewhoffman/MALI/calving_CFL (commit 258eadc1c90a616a26675538088dad383d726ed3).

humboldt.specified_calving_velocity.none (looks the same as in Matt's google slides):

humboldt.von_Mises_stress.FO (looks a little different from Matt's Google slides, maybe just because one run is longer?)

humboldt.von_Mises_stress.none (looks the same as google slides)

mismip+.specified_calving_velocity.none (looks the same as google slides)

mismip+.von_Mises_stress.FO (looks the same as google slides)

mismip+.von_Mises_stress.none (looks the same as google slides)

thwaites.specified_calving_velocity.none (looks the same as google slides)

thwaites.von_Mises_stress.FO (looks a little different from google slides: orange curve in third panel)

thwaites.von_Mises_stress.none (looks the same as google slides)

trhille

One more change I thought of while looking through output figures.

compass/landice/tests/calving_dt_convergence/dt_convergence_test/__init__.py

matthewhoffman · 2022-05-10T23:49:05Z

@trhille , thanks for reviewing this and checking all the output images. I've pushed a commit addressing your review comments. I'm going to work on the docs next, but feel free to look it over now or wait until I have the docs ready.

matthewhoffman · 2022-05-11T17:30:05Z

@xylar , this landice PR is finished and just awaiting CI to complete and a final re-run of our integration suite. If you want to review it before Trevor's final re-review and merge, let us know.

matthewhoffman · 2022-05-11T18:44:40Z

Update: CI and test suites all pass/give expected behavior

xylar · 2022-05-11T19:33:41Z

Nope, go ahead. It looks like there are some conflicts. As you know, my preference would be to rebase to handle them but you could either merge master to this branch or fix them in a merge commit if you prefer.

This uses the a spunup 2km MISMIP+ domain from Anvil. It runs with data velocity and specified calving velocity for calving. It runs with a series of values of config_adaptive_timestep_calvingCFL_fraction so that the results can be compared. A .png file is created showing the results.

May want to make options to set up either, but for now, just modifying the test, to avoid test proliferation.

Make use of new features added in MALI-Dev/E3SM#33

This is a configuration for investigating convergence of calving behavior with timestep for calving. Rather than duplicating the test in a different test groups, I've instead implemented the test as a test group, and it can be applied to three different meshes: MISMIP+, Humboldt 3km, and Thwaites 4km. It can also be applied to three different calving laws: specified calving velocity, von Mises stress, and eigencalving. It can be applied using either a data velocity field or the FO solver. This implementation replaces the previous commits that applied this same concept to just the MISMIP+ geometry. In each test, a series of runs are conducted that repeat the same simulation using different values of config_adaptive_timestep_calvingCFL_fraction. Then summary plots are generated comparing the behavior for the different fraction values. The calving law parameters are modified for the different configurations as needed to achieve steady retreat at a rate that ensures the calving CFL condition is more restrictive than the advective CFL, even for large values of config_adaptive_timestep_calvingCFL_fraction. The configurations with no velocity solver typically take a few minutes each. The ones with FO velocity take much longer, in some cases a few hours, even after setting those configurations to run for a shorter duration. As coded up now, each simulation with different fractions are run in serial. Setting this up to run them in parallel could be a strategy for getting faster results, if desired. Another limitation is that as coded up now, the analysis/plotting script is run in the validation step and therefore cannot be run manually. Therefore it will only run if all of the individual runs run to completion successfully. This could be improved by moving this operation to a separate step.

The MISMIP+ test group was added in this branch and has only a dt_convergence_test case. Now that I have created a calving_dt_convergence test group, there is no need to have a MISMIP+ test case for that purpose. This commit deletes the dt_convergence_test module and test from the MISMIP+ test group. It then adds a smoke test so that the MISMIP+ test group can actually be used. The smoke test just uses the pre-built initial condition file and runs for 5 years with the FO solver. It takes about a minute and a half on Badger. This test can be modified or replaced with something as needs require.

This brings in the calving CFL PR for MALI that is required to run the new tests: MALI-Dev/E3SM#33

trhille · 2022-05-11T21:51:06Z

All tests passed upon re-running the calving_dt_convergence test suite on Badger with an exe compiled on the head of MALI-Dev/develop (commit 9d3dcd9b0a1a15e20458a7e3394b545517ae45b5)

Test Runtimes:
02:40 ESC[92mPASSESC[0m landice_calving_dt_convergence_mismip+.specified_calving_velocity.none
02:08 ESC[92mPASSESC[0m landice_calving_dt_convergence_mismip+.von_Mises_stress.none
01:56 ESC[92mPASSESC[0m landice_calving_dt_convergence_humboldt.specified_calving_velocity.none
01:39 ESC[92mPASSESC[0m landice_calving_dt_convergence_humboldt.von_Mises_stress.none
02:49 ESC[92mPASSESC[0m landice_calving_dt_convergence_thwaites.specified_calving_velocity.none
03:16 ESC[92mPASSESC[0m landice_calving_dt_convergence_thwaites.von_Mises_stress.none
72:48 ESC[92mPASSESC[0m landice_calving_dt_convergence_mismip+.von_Mises_stress.FO
08:35 ESC[92mPASSESC[0m landice_calving_dt_convergence_humboldt.von_Mises_stress.FO
21:20 ESC[92mPASSESC[0m landice_calving_dt_convergence_thwaites.von_Mises_stress.FO
Total runtime 117:12
PASS: All passed successfully!

trhille · 2022-05-11T22:03:19Z

Figures look very similar to previous. One (shown below) is subtly but noticeably different than before. This could be due to the small changes in namelist options introduced after my last review, or maybe to the small cleanup from the review of MALI-Dev PR 33 (less likely). Unfortunately I forgot to uncomment the variable comparison lines with I ran, so the baseline comparison did not work. Just comparing these visually:
humboldt_specified_calving_velocity_none (top few curves in third panel):

The other test cases are not noticeably different than before.

trhille

@matthewhoffman, great work on this! I've noted a few totally nit-picky things, but am also approving it since these are trivial and do not affect functionality.

docs/developers_guide/landice/test_groups/mismipplus.rst

docs/users_guide/landice/suites.rst

matthewhoffman added land ice in progress This PR is not ready for review or merging MALI-Dev PR required labels May 5, 2022

matthewhoffman requested a review from trhille May 5, 2022 22:12

matthewhoffman mentioned this pull request May 5, 2022

Create a calving CFL limit and apply to adaptive timestepper MALI-Dev/E3SM#33

Merged

trhille requested changes May 9, 2022

View reviewed changes

trhille requested changes May 10, 2022

View reviewed changes

compass/landice/tests/calving_dt_convergence/dt_convergence_test/__init__.py Outdated Show resolved Hide resolved

matthewhoffman force-pushed the landice/mismipplus_calvingdt_test branch from 21a6ecf to 8e4909f Compare May 11, 2022 17:27

matthewhoffman removed in progress This PR is not ready for review or merging MALI-Dev PR required labels May 11, 2022

xylar assigned trhille May 11, 2022

matthewhoffman force-pushed the landice/mismipplus_calvingdt_test branch from 8e4909f to 625468c Compare May 11, 2022 19:47

matthewhoffman added 10 commits May 11, 2022 14:12

Switch to using vM calving law

cffa81b

May want to make options to set up either, but for now, just modifying the test, to avoid test proliferation.

Additions to test for new MALI options

a84c202

Make use of new features added in MALI-Dev/E3SM#33

Create a test suite for the calving dt tests

63720a5

Minor adjustments to configurations

db5cbeb

Fix PEP8 errors

fe0274d

Add fraction of dt with warnings to summary plot

43e407d

Changes from code review

ec92e8f

matthewhoffman added 4 commits May 11, 2022 14:12

Add docs

353083c

Update MALI-Dev submodule

a3e221e

This brings in the calving CFL PR for MALI that is required to run the new tests: MALI-Dev/E3SM#33

Add MISMIP+ docs

07e2685

Fix error in Koge Bugt docs

651bd0d

matthewhoffman force-pushed the landice/mismipplus_calvingdt_test branch from 625468c to 651bd0d Compare May 11, 2022 20:13

trhille approved these changes May 11, 2022

View reviewed changes

Docs updates after re-review

1b36835

matthewhoffman merged commit 3346e17 into MPAS-Dev:master May 11, 2022

matthewhoffman deleted the landice/mismipplus_calvingdt_test branch May 11, 2022 22:58

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

MALI calving dt convergence tests #386

MALI calving dt convergence tests #386

matthewhoffman commented May 5, 2022 •

edited

Loading

pep8speaks commented May 5, 2022 •

edited

Loading

matthewhoffman commented May 5, 2022 •

edited

Loading

trhille left a comment

trhille May 9, 2022

trhille May 9, 2022

trhille commented May 10, 2022 •

edited

Loading

trhille left a comment

matthewhoffman commented May 10, 2022

matthewhoffman commented May 11, 2022

matthewhoffman commented May 11, 2022

xylar commented May 11, 2022

trhille commented May 11, 2022

trhille commented May 11, 2022 •

edited

Loading

trhille left a comment

MALI calving dt convergence tests #386

MALI calving dt convergence tests #386

Conversation

matthewhoffman commented May 5, 2022 • edited Loading

pep8speaks commented May 5, 2022 • edited Loading

Comment last updated at 2022-05-11 22:38:47 UTC

matthewhoffman commented May 5, 2022 • edited Loading

trhille left a comment

Choose a reason for hiding this comment

trhille May 9, 2022

Choose a reason for hiding this comment

trhille May 9, 2022

Choose a reason for hiding this comment

trhille commented May 10, 2022 • edited Loading

trhille left a comment

Choose a reason for hiding this comment

matthewhoffman commented May 10, 2022

matthewhoffman commented May 11, 2022

matthewhoffman commented May 11, 2022

xylar commented May 11, 2022

trhille commented May 11, 2022

trhille commented May 11, 2022 • edited Loading

trhille left a comment

Choose a reason for hiding this comment

matthewhoffman commented May 5, 2022 •

edited

Loading

pep8speaks commented May 5, 2022 •

edited

Loading

matthewhoffman commented May 5, 2022 •

edited

Loading

trhille commented May 10, 2022 •

edited

Loading

trhille commented May 11, 2022 •

edited

Loading