Standardize regridding to `conservative_normed` #940

tomvothecoder · 2025-02-14T23:22:20Z

Description

Closes [Bug]: Investigate differences in xESMF and ESMF regridding that lead to "Model - Observation" plot diffs #938
e3sm_diags_driver/utils/regrid.py:
- Update _align_grid_to_lower_res() to update "conservative" regrid method to "conservative_normed" for proper alignment with CDAT
- Update _apply_land_sea_mask() to add a mask variable to the dataset for regridding, which helps prevent missing values from bleeding into the regridding.
- Remove unused Notes comments referring to old CDAT-based functions that have been removed
e3sm_diags/driver/default_diags/lat_lon_model_vs_model.cfg and `e3sm_diags/driver/default_diags/lat_lon_model_vs_obs.cfg
- Remove regrid_method = "bilinear" to default to "conservative" (which becomes "conservative_normed")

Checklist

My code follows the style guidelines of this project
I have performed a self-review of my own code
My changes generate no new warnings
Any dependent changes have been merged and published in downstream modules

If applicable:

New and existing unit tests pass with my changes (locally and CI/CD build)
I have added tests that prove my fix is effective or that my feature works
I have commented my code, particularly in hard-to-understand areas
I have made corresponding changes to the documentation
I have noted that this is a breaking change for a major release (fix or feature that would cause existing functionality to not work as expected)

tomvothecoder · 2025-02-18T20:29:19Z

e3sm_diags/driver/utils/regrid.py

+    # TODO: Remove this conditional once "conservative" references are updated
+    # to "conservative_normed" throughout the codebase.
+    # NOTE: this is equivalent to "conservative" in cdms2 ESMF. If
+    # "conservative" is chosen, it is updated to "conservative_normed". This
+    # logic can be removed once the CoreParameter.regrid_method default
+    # value is updated to "conservative_normed" and all sets have been
+    # refactored to use this function.
+    if method == "conservative":
+        method = "conservative_normed"


Related to #938 (comment)

I think I forgot to add this conditional in _align_grids_to_lower_res(), which resulted in the abnormal latitude edge in the polar plot here. This fixes that issue and hopefully more diffs between "Model - Observation" plots.

@chengzhuzhang I just did a run with the specific variables from #930/#931 that have differences due to regridding. This change doesn't seem to affect the plot differences for these cases. The regression testing notebooks show the same differences between *_diff.nc files and the plot diffs look similar.

I think this is the new, expected result as far as I can tell (did not try the bounds dimension name alignment fix from #938).

This fix does improve the polar plots though, as you mentioned.

@tomvothecoder see comment here: #938 (comment). It looks like changing to convervative_normed has little impact, I'm interested to look at the plots and tables. Could you point me to the urls of both runs?

Could you point me to the urls of both runs?

v2.12.1: https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-04-main-zppy-diffs/viewer/

Latest main with conservative: https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-04-branch-930-zppy-diffs/viewer/index.html

This branch with conservative_normed: https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-18-branch-940-xesmf-diffs/viewer/index.html

This branch with conservative_normed for all variables (no bilinear): https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-18-branch-940-xesmf-diffs-no-bilinear/viewer/index.html

Thanks for the links! I think we should use conservative_normed to replace bilinear. Now there is a caveat: here, look at the diff plot, there is a gap along the coastal line, in conservative_normed. And I realize that, part of the reason is that we are using a very coarse resolution file for land-sea mask. I'm creating a new land-sea mask to see if it helps.

e3sm_diags/driver/utils/regrid.py

- Convert `"conservative"` regridding method to `"conservative_normed"` to align with CDAT codebase

chengzhuzhang · 2025-02-19T18:17:49Z

@tomvothecoder Following:https://xesmf.readthedocs.io/en/latest/notebooks/Masking.html, I applied the fix (93bf5e3) by adding a mask variable to the dataset before regridding, it actually resolved the issue that mask bleeding into data:
After:

Before:

chengzhuzhang · 2025-02-19T18:20:00Z

Maybe it is hopeful to wrap up this PR with some more testing?

tomvothecoder · 2025-02-19T20:58:55Z

Maybe it is hopeful to wrap up this PR with some more testing?

Here is the latest run with the mask fix: https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-19-branch-940-xesmf-diffs-mask-fix/viewer/index.html

chengzhuzhang · 2025-02-19T22:04:25Z

@tomvothecoder thank you. I'm putting together all the urls and inspecting the metrics results change:
v2.12.1: https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-04-main-zppy-diffs/viewer/
Latest main with conservative: https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-04-branch-930-zppy-diffs/viewer/index.html
This branch with conservative_normed: https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-18-branch-940-xesmf-diffs/viewer/index.html
This branch with conservative_normed for all variables (no bilinear): https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-18-branch-940-xesmf-diffs-no-bilinear/viewer/index.html
Here is the latest run with the mask fix: https://web.lcrc.anl.gov/public/e3sm/cdat-migration-fy24/25-02-19-branch-940-xesmf-diffs-mask-fix/viewer/index.html

I noticed large difference in RMSE in SST plots to use conservative_normed compare to bilinear.
Conservative_normed:

Variables Unit Test_mean Ref._mean Mean_Bias Test_STD Ref._STD RMSE Correlation
SST global HadISST_CL degC 19.909 18.616 1.293 7.995 9.587 1.771 0.979
SST global HadISST_PD degC 19.909 18.73 1.178 7.995 9.589 1.839 0.977
SST global HadISST_PI degC 19.909 18.897 1.011 7.995 8.988 1.802 0.977

Variables Unit Test_mean Ref._mean Mean_Bias Test_STD Ref._STD RMSE Correlation
Bilinear:
SST global HadISST_CL degC 19.909 18.698 1.211 7.995 9.536 1.477 0.985
SST global HadISST_PD degC 19.909 18.807 1.102 7.995 9.543 1.551 0.984
SST global HadISST_PI degC 19.909 18.978 0.931 7.995 8.933 1.478 0.984

I'm going to plot out the regrided maps to examine and compare.

tomvothecoder commented Feb 18, 2025

View reviewed changes

e3sm_diags/driver/utils/regrid.py Outdated Show resolved Hide resolved

tomvothecoder force-pushed the bug/939-xesmf-diffs branch from 879dc66 to d5776a4 Compare February 18, 2025 20:50

tomvothecoder mentioned this pull request Feb 18, 2025

[Bug]: Investigate differences in xESMF and ESMF regridding that lead to "Model - Observation" plot diffs #938

Open

tomvothecoder added 9 commits February 19, 2025 11:46

Add initial debugging script and code changes to fix xESMF diffs

c84cab7

Update align_grids_to_lower_res()

88583d8

- Convert `"conservative"` regridding method to `"conservative_normed"` to align with CDAT codebase

Update dir name for scripts

5e47e7a

Update e3sm_diags/driver/utils/regrid.py

cc2d2e4

Update scripts

f8eb18b

Remove "bilinear" regridding workaround for specific variables

551706d

Fix unit tests

11bcc2e

Update scripts

1ba9308

Add scripts for debugging trefht-land coastlines

37c39ef

tomvothecoder force-pushed the bug/939-xesmf-diffs branch from c36e239 to 37c39ef Compare February 19, 2025 17:46

add mask variable to fix xesmf regridding

93bf5e3

tomvothecoder added 2 commits February 19, 2025 14:59

Update regression testing script for mask fix

80797fc

Fix unit tests and add comment about mask var

bda65d8

tomvothecoder self-assigned this Feb 19, 2025

tomvothecoder added this to the FY25 Q2 (01/01/25 - 03/31/25) milestone Feb 19, 2025

tomvothecoder changed the title ~~Address "Model - Observation" plot differences due to xESMF~~ Convert xESMF regridding from bilinear to conservative_normed Feb 19, 2025

tomvothecoder marked this pull request as ready for review February 19, 2025 21:22

tomvothecoder added bug Bug fix (will increment patch version) enhancement cdat-migration-fy24 CDAT Migration FY24 Task labels Feb 19, 2025

tomvothecoder changed the title ~~Convert xESMF regridding from bilinear to conservative_normed~~ Standardize regridding to conservative_normed Feb 19, 2025

Update debug_diffs_min.py

c0e5e29

Fix unit tests and add comment about mask var

fe3f2aa

tomvothecoder force-pushed the bug/939-xesmf-diffs branch from 7aea4ed to fe3f2aa Compare February 19, 2025 22:06

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Standardize regridding to `conservative_normed` #940

Standardize regridding to `conservative_normed` #940

tomvothecoder commented Feb 14, 2025 •

edited

Loading

tomvothecoder Feb 18, 2025

tomvothecoder Feb 18, 2025 •

edited

Loading

tomvothecoder Feb 18, 2025 •

edited

Loading

chengzhuzhang Feb 18, 2025

tomvothecoder Feb 18, 2025 •

edited

Loading

chengzhuzhang Feb 18, 2025

chengzhuzhang commented Feb 19, 2025 •

edited

Loading

chengzhuzhang commented Feb 19, 2025

tomvothecoder commented Feb 19, 2025

chengzhuzhang commented Feb 19, 2025

Standardize regridding to conservative_normed #940

Are you sure you want to change the base?

Standardize regridding to conservative_normed #940

Conversation

tomvothecoder commented Feb 14, 2025 • edited Loading

Description

Checklist

tomvothecoder Feb 18, 2025

Choose a reason for hiding this comment

tomvothecoder Feb 18, 2025 • edited Loading

Choose a reason for hiding this comment

tomvothecoder Feb 18, 2025 • edited Loading

Choose a reason for hiding this comment

chengzhuzhang Feb 18, 2025

Choose a reason for hiding this comment

tomvothecoder Feb 18, 2025 • edited Loading

Choose a reason for hiding this comment

chengzhuzhang Feb 18, 2025

Choose a reason for hiding this comment

chengzhuzhang commented Feb 19, 2025 • edited Loading

chengzhuzhang commented Feb 19, 2025

tomvothecoder commented Feb 19, 2025

chengzhuzhang commented Feb 19, 2025

Standardize regridding to `conservative_normed` #940

Standardize regridding to `conservative_normed` #940

tomvothecoder commented Feb 14, 2025 •

edited

Loading

tomvothecoder Feb 18, 2025 •

edited

Loading

tomvothecoder Feb 18, 2025 •

edited

Loading

tomvothecoder Feb 18, 2025 •

edited

Loading

chengzhuzhang commented Feb 19, 2025 •

edited

Loading