-
Notifications
You must be signed in to change notification settings - Fork 10
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Issue with global mean weight for multiscale ensemble runs for ig > 1 plus changes to make code work for multiscale (bugs introduced in trunk). #570
Conversation
…nal FV3 DA (FV3LAMDA)
I compared the following four tests of RRFS using Scale- and Variable-Dependent Localization (SDL: nsclgrp=2 and VDL: ngvarloc=2).
As results, 2 was stopped by segmentation fault but 3 is completed. The result of 3 was slightly different from 1 because of the changes by PR#527. However, 4 was completely the same as 3, which indicates this PR does not affect the regional SDL/VDL except for the bug fix. All of these results are as we expected. I approve this PR. |
Thank you and Ting.Sent from my Verizon, Samsung Galaxy smartphone
-------- Original message --------From: shoyokota ***@***.***> Date: 5/25/23 2:52 PM (GMT-05:00) To: NOAA-EMC/GSI ***@***.***> Cc: jderber-NOAA ***@***.***>, Mention ***@***.***> Subject: Re: [NOAA-EMC/GSI] Issue with global mean weight for multiscale ensemble runs for ig > 1 plus changes to make code work for multiscale (bugs introduced in trunk). (PR #570)
@shoyokota approved this pull request.
—Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you were mentioned.Message ID: ***@***.***>
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
One comment and one question.
@TingLei-NOAA and @shoyokota , thank you for reviewing the changes in this PR. I asked John to remove a commented out line of code. He did so. Your reviews are now stale. I requested new reviews from each of you. Thank you for your prompt attention to this. We want to get John's bug fixes into FYI, when you review code and leave comments, please remember to resolve your comments after the developer responds and you are satisfied with the response. I resolved your comments this morning to move this PR forward. I shouldn't do this since I don't know if John's response is satisfactory to you. |
@RussTreadon-NOAA I had given my approval of this PR. About "resolving "the conversation, I think, as we had discussed/verificed before, we have no "access/permission " to do this except for persons like you who have writing access to the trunk can do it. Is this github policy changed? I still failed to see that "resolving " button" . |
@TingLei-NOAA , if a reviewer does not have a button to click resolve, the reviewer should enter text in the comment window stating that the developer's response is satisfactory. This allows others to close a comment with certainty. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Two peer review approvals given. Ctests run and documented.
@jderber-NOAA , I clicked all the checkboxes except the first. Would you please confirm that you can check I think you can do so as the PR author. |
Box checked. |
oops. |
Cross check with GSI Handling Review team. This PR is ready to merge! |
Thanks for this clarification! |
Description
While looking to optimize the use of multiscale ensembles, it was noted that the global mean wgt was not changed from the default (1.0) for all values of ig. The sum of the weights over all ig should be equal to 1 for the case being used. It was found that for all ig =1 the value of the weight should remain 1., but for all ig > 1 the value should be equal to zero. The original OU code had this in it. Somehow when being put into the GSI, the code doing this was removed.
Tests using the multiscale ensemble are starting and this needs to be fixed.
The only routine changed was init_mult_spc_wgts in apply_scaledepwgts.
Also, some cleanup of the routine in which the weights are set was included (e.g., 0 used for real zero, unused arrays removed, etc.)
When change was updated to the head of the trunk, two additional issues were found. The multiscale ensemble no longer worked and the regression test for the netcdf_fv3_regional no longer scaled between the low processor count and the high processor count. Both of these errors are present in the head of the trunk, not just in this branch.
The source of these two additional issues were found by running in debug mode and adding a lot of print statements. The solution to make it run the multiscale ensemble required a change to state_vectors.f90. Note a few small changes were also found to get the code to run completely in debug mode. To remove the processor scaling issue, a code section was removed from a later loop where it was put by the previous change (by me, sorry) and put in it's previous location before the cloud liquid water calculation in setuprad.f90.
After the previous two changes, the multiscale ensemble and the regression tests worked properly.
Fixes #569
Type of change
How Has This Been Tested?
Weights produced by this routine were printed and only the global mean values for ig >1 were changed.
Regression tests were run. No differences were found (but I don't think any regression tests change the one routine changed by this update).
Results were compared to optimization tests (which also had this change, but a lot more) to ensure the results were similar (expected round off differences).
All tests were performed on Hera
Checklist
DUE DATE for this PR is 6/16/2023. If this PR is not merged into
develop
by this date, the PR will be closed and returned to the developer.