user controlled history density

rgknox commented 1 year ago

Description:

1) This allows FATES output to be controlled by setting density levels in the ELM/CLM namelist. Right now, the namelist setting is: "fates_hist_dense_level". The input is a list of two comma delimited integers. The first digit controls fates output at model timesteps (hi-frequency) and the second digit controls output at the dynamics (daily) timestep.

2) A cross-referencing routine has been added. This will loop through the list of variables that are in the master list from the HLM, and see if that variable has been allocated on the FATES side. It is now possible there could be a discrepancy. There will be a graceful and explanatory fail if the variable is not allocated.

3) This test will definitely require changes to the testing. For our "allvars" type tests, we should allocate all and use the highest output density levels (2,2). Perhaps we should have more allvars, and then run more tests at decreased level (1,1). At least one test should be 0,0.

4) Timing tests forthcoming

synchronized with: https://github.com/ESCOMP/CTSM/pull/2339

user documentation updates here: https://github.com/NGEET/fates-users-guide/pull/59

Here is a description of the namelist control from clm_varctl.F90:

  ! FATES history density level
  ! fates can produce history at either the daily timescale (dynamics)
  ! and the model step timescale. It can also generate output on the extra dimension
  ! Performing this output can be expensive, so we allow different history density
  ! levels.
  ! The first index is output at the model timescale
  ! The second index is output at the dynamics (daily) timescale      
  ! 0 - no output
  ! 1 - include only column level means
  ! 2 - include only output with only 1 additional dimension
  ! 3 - include all multiplexed dimensions

  integer, dimension(2), public   :: fates_hist_dense_level = (/1,1/)

CTSM-side changes: https://github.com/rgknox/ctsm/tree/fates-hist-density

Notes: This set of changes is built on-top of the two-stream pr. This set of changes will undboubtedly have conflicts with the LU PRs, but @ckoven and I have identified the differences and they won't be hard to incorporate.

Collaborators:

@ckoven

Expectation of Answer Changes:

Ideally, this should be a b4b set of changes, although it is possible with all the loop chopping being done, some things might have very small roundoff diffs.

Checklist

If this is your first time contributing, please read the CONTRIBUTING document.

All checklist items must be checked to enable merging this pull request:

NOTE: THIS UPDATE SHOULD BE ADDED TO THE USERS GUIDE, AND HAVE APPROPRIATE PR WHEN INTEGRATING THIS ONE

Contributor

[ ] The in-code documentation has been updated with descriptive comments
[ ] The documentation has been assessed to determine if updates are necessary

Integrator

[ ] FATES PASS/FAIL regression tests were run
[ ] Evaluation of test results for answer changes was performed and results provided

Documentation

Technical Note update:
User's Guide update:

Test Results:

CTSM (or) E3SM (specify which) test hash-tag:

CTSM (or) E3SM (specify which) baseline hash-tag:

FATES baseline hash-tag:

Test Output:

glemieux commented 11 months ago

Most of my comments are questions asking for clarification. The few requested changes are fairly minor, so I leave those at your discretion. Note I only skimmed the unit testing code (which I loved to see though!).

I have no idea how this got added to this PR. It's a match to the approval I wrote for 2 stream. Must be a weird github glitch :shrug: ? I did reference is in a review comment, and some of the code from here appears in 2 stream, so maybe that's why?

rgknox commented 10 months ago

Preliminary tests are passing on derecho. I compared one test with base (SMS_1x1_brazil) and that showed differences within expected tolerances. This PR has no math changes, but it does have order of operations changes and plenty of shuffling around, so I do not expect B4B. The diffs that i saw were typically smaller than abolute and relative e-10. In light of this, I'm removing the draft status and looking for reviewers.

rgknox commented 10 months ago

Also: I currently only have things split out to level "2". ie, I don't differentiate multiplexed output in that fourth (last) dimension, from output that is not multiplexed in the fourth dimension. My plan is make sure all test are passing before I do that split.

samsrabin commented 9 months ago

Concern raised by @ekluzek that "density" can be confusing. E.g., does this refer to single vs. double precision?

rgknox commented 8 months ago

@samsrabin I updated language to use "dimension" instead of density

rgknox commented 8 months ago

all tests (fates and aux_clm) are ok on derecho and izumi. Small diffs attributed to order of operations, other diffs related to targetted fixes in a small and known set of variables. Some diffs associated with two-stream tests, which is a separate issue. /glade/derecho/scratch/rgknox/tests_0312-112335de /scratch/cluster/rgknox/tests_0312-211553iz

NGEET / fates