Update effective radius bin ranges for new MODIS ice-cloud CER-CWP joint histograms

[brandonduran]

We have adjusted cosp_config.F90 to add a new effective radius bin for the MODIS ice-cloud CER-CWP joint histogram. External testing has shown that some GCMs simulate ice-clouds with CER > 60 microns, which are not accounted for given the current bin limits.

Changes:
(1) A new CER bin has been added (60 to 10000 microns).

The proposed range of 60-10000 microns is intended to capture all of the ice clouds simulated by GCMs that have CER > 60 microns. We modified this range from our initial proposal of 60-90 microns as @klein21 correctly noted that there was no justification for imposing an upper limit on particle size.

[RobertPincus]

@brandonduran Thanks for this suggestion. We should solicit feedback from the CFMIP folks, especially those working on the data request. @mzelinka? @alejandrobodas ? Who else?

[alejandrobodas]

@brandonduran many thanks for this. I agree with @klein21 that the histogram should record all GCM clouds somewhere, with the upper limit of the last bin being a nominal value and not a hard limit. Do the the calculations of the MODIS histograms use the last bin edge as a hard limit? I'm asking because I think it's important that we have consistency in the definition of the axes across all the MODIS histograms. There are 2 MODIS histograms in the CMIP6 data request: jpdftaureicemodis and jpdftaureliqmodis. These are the bin edges as defined in the data request:
Ice Reff: 0.0, 10.0, 20.0, 30.0, 40.0, 60.0, 90.0
Liquid Reff: 0.0, 8.0, 10.0, 13.0, 15.0, 20.0, 30.0
COT: 0.0, 0.3, 1.3, 3.6, 9.4, 23.0, 60.0, 100000.0

Also, please note that the continuous integration tests have failed.

[brandonduran]

Hi @alejandrobodas thanks for your comment.

As far as I can tell, the MODIS histograms use the upper and lower bin edges as hard limits. I think this is the biggest factor that supports including a very large final ice cloud CER bin edge (10000 microns) since this should capture all clouds simulated by GCMs. This also begets the question of whether additional lower bins should be added for liquid (0-4 microns) and ice (0-5 microns) CER along the same theme of capturing all GCM simulated clouds. We (@caseywall7926 @qinyia) have not done any testing on whether including extra lower bins increases the total cloud fraction in the joint histograms, so the justification for changes is less clear.

Re the CMIP data request, I think the COT request is fine. The current MODIS joint histogram bin edges in COSP are:

Ice Reff: 5.0, 10.0, 20.0, 30.0, 40.0, 60.0 ( 90.0 or 10000 to be added?)
Liquid Reff: 4.0, 8.0, 10.0, 12.5, 15.0, 20.0, 30.0

Are the bin edges in the data request already finalized? If so, this might require us to think differently about our changes.

I will check on the test failures and put some commits through to address them.

[caseywall7926]

Hi all, we hoped to make the new histogram bins comparable to the observational data of Pincus et al. (2023). See attached figure for the CER-COT histograms from that observational dataset: essd-15-2483-2023.pdf

It looks like the CER histogram bin edges that @alejandrobodas listed differ from the Pincus et al. (2023) bin edges in several ways.

Bin edges from @alejandrobodas (copied from above)
Ice CER: 0.0, 10.0, 20.0, 30.0, 40.0, 60.0, 90.0
Liquid CER: 0.0, 8.0, 10.0, 13.0, 15.0, 20.0, 30.0

Bin edges from Pincus et al. (2023):
Ice CER: 5.0, 10.0, 20.0, 30.0, 40.0, 50.0, 60.0
Liquid CER: 4.0, 8.0, 10.0, 12.5, 15.0, 20.0, 30.0

Bold values are differences that deserve special attention. As a compromise between all of the comments that have been mentioned, how about the following values for histogram bin edges?

Ice CER: 0.0, 10.0, 20.0, 30.0, 40.0, 50.0, 60.0, 10000
Liquid CER: 0.0, 8.0, 10.0, 12.5, 15.0, 20.0, 30.0, 10000

[brandonduran]

The CI tests are now failing when comparing with KGO, due to array mismatches. This is not surprising, as we are adding a new bin to the CER dimension of the MODIS joint histograms, leading to dimensional conflicts. If these changes get merged, I will need to compute new md5 sums for this portion of the testing. I am noticing that the necessary .tgz files (as mentioned in #86 ) that will be needed to do this are not being uploaded with the tests, though.

[alejandrobodas]

I think there is a bug in the logic of the CI test that prevents the creation of the tgz files because the steps that produces the plots is failing before the tgz file is created. If this is indeed the case then I should be able fix it relatively quickly. Once that fix is in place we'll be able to make progress with this PR.

[alejandrobodas]

@brandonduran , #112 fixes the failure due to the different sizes in the histogram array. If you rebase your branch the CI tests will still fail but they will produce the plots and outputs that you can inspect.

[brandonduran]

Thanks for fixing the CI tests in #112 @alejandrobodas . I will inspect the outputs when the tests run and produce the .tgz files.

[alejandrobodas]

Thanks @brandonduran , I've submitted the CI tests and the outputs are now available.

[brandonduran]

Hi @alejandrobodas thanks. I have attached plots of the MODIS COT x ReICE joint histograms which have changed as a result of this pull request. They look as expected given the changes proposed. I should note that Figure 3, which shows the output from the original COSP code, is mislabeled in the current version of the cod. Currently, the largest 3 bin edges are 40, 50, and 60 microns (see cosp_config.F90), but plot_test_outputs.py lists them as 40, 60, and 90 microns. We'd expect the new code changes to show the same values for the smallest 6 REice bins, since we aren't changing those ranges.

The other variable that changed is "REICE_MODIS". These are just the bin midpoints for particle size for the histograms, so I have not plotted them, as their changes are straighforward.

I have updated plot_test_outputs.py in my latest commit to reflect these histogram changes. Please let me know what the next steps are!

COSP Figures.pdf

[alejandrobodas]

Hi @brandonduran , thanks for the update. I've been looking back at the comments and we still need to address @caseywall7926 comment about inconsistency between the bin boundaries in the obs, COSP and CMIP data request. My view is that COSP should have bin boundaries that align with the obs, and all clouds should be reported in the histogram. Therefore, my proposal is for these bin boundaries:

• Ice: 0, 5, 10, 20, 30, 40, 50, 60, 10000
• Liquid: 0, 4, 8, 10, 12.5, 15, 20, 30, 10000

Once we've agreed the COSP edges, we'll make a request for a change in the definition of the axes in the CMIP7 data request. @RobertPincus , do you agree with this proposal?

[brandonduran]

Thanks for the reminder @alejandrobodas . I agree with your proposal on the bin edges, as it meets both criteria that you outline.

Once others weigh in (@RobertPincus , @caseywall7926 ) and we agree, I'll commit the bin edge code changes.

[caseywall7926]

The bin edges proposed by @alejandrobodas sound great to me too. Thanks, all!

[alejandrobodas]

Since we haven't received objections to the proposal for the new bin edges, I believe we can move ahead with the changes. @brandonduran please could you implement the changes?

[brandonduran]

Hi @alejandrobodas - changes are committed and awaiting CI tests. Includes updated bin edges in both cosp_config.F90 and tweaks to the python plotting script for creating test output.

[brandonduran]

@alejandrobodas CI tests failing as expected given the dimension changes to both of the MODIS joint histograms. I believe all that remains is updating the known good outputs, and then the CI tests should (hopefully) pass? Please advise on next steps.

[alejandrobodas]

Hi @brandonduran , I've updated the md5 sums for the new Known Good Outputs in #113. I've also uploaded the new KGOs to Google drive. Basically, what's left to do is for this PR to point to the new KGOs. You can do that by implementing these changes (only the changes in the .yml files, not the .md5):
6c3b9f4
ff0cd8d

There are some conflicts due to the changes in the md5 files. You should be able to fix those by rebasing your branch. If I haven't made any mistake, then your changes should pass all the tests. Fingers crossed!

[brandonduran]

@alejandrobodas Hope I didn't mess that one up on my end either! Yml files updated.

[alejandrobodas]

All looks good, approving review.

[alejandrobodas]

Changes merged. @brandonduran thanks for your contribution!