add stream field specific unit conversions in stream_defintion_MODEL.xml

[mvertens]

Summary

Allow a unit conversion factor on stream variables

Description of changes

Adds an optional third token to a <var> line in a stream definition, giving a factor applied to that field as it is read. Uses it to fix nitrogen deposition, where the conversion was previously inferred from field names and got cplhist forcing wrong by a factor of 1000.

datm_pres_ndep_mod decided whether to convert nitrogen deposition from gN/m2/s to kgN/m2/s by looking at which stream fields were present: four fields meant no conversion, two fields meant divide by 1000.

A <var> line may now carry a conversion factor:

<var>drynhx  Faxa_ndep_nhx_dry</var>
<var>NDEP_NHx_month  Faxa_ndep_nhx  1.0e-3</var>

• The factor is applied in dshr_strdata_mod as the field is read, so the data enters the system already in the units the model expects and no component carries conversion logic of its own.
• Existing two-token entries are unaffected
• The factor sits on the variable rather than the stream because units are a property of the field. A single stream may carry some fields needing conversion and others not, which a stream-wide factor could not express.
• datm_pres_ndep_mod now only sums the fields it is given.

Specific notes

Contributors other than yourself, if any: claude

CDEPS Issues Fixed: #437

Are there dependencies on other component PRs:

Are changes expected to change answers: different to roundoff

• Since we are now multiply by 1.0e-3 rather than dividing by 1000 - the answers are round off different
• Testing was done in NorESM for the following compset/resolution:
  COMPSET: 1850_DATM%GSWP3v1_CLM50%SP_SICE_SOCN_MOSART_CISM2%GRIS-EVOLVE_SWAV_SESP
  GRID: a%1.9x2.5_l%1.9x2.5_oi%null_r%r05_w%null_z%null_g%gris4_m%tnx1v4
  And the following was obtained for the diffs:

atmImp_Faxa_ndep1   (atmImp_nx,atmImp_ny,time)  t_index =      1     1
      6630    13824  (    43,    55,     1) (    23,     6,     1) (    43,    56,     1) (   123,    15,     1)
              13824   3.338589824444479E-11   3.220082898594232E-16 6.5E-27  2.973562882264960E-11 8.1E-17  2.278254265933821E-13
              13824   3.338589824444479E-11   3.220082898594232E-16          2.973562882264959E-11          2.278254265933820E-13
              13824  (    43,    55,     1) (    23,     6,     1)
         avg abs field values:    6.873940049853111E-13    rms diff: 1.9E-28   avg rel diff(npos):  8.1E-17
                                  6.873940049853111E-13                        avg decimal digits(ndif): 15.8 worst: 15.4
RMS atmImp_Faxa_ndep1                1.8978E-28            NORMALIZED  2.7609E-16

atmImp_Faxa_ndep2   (atmImp_nx,atmImp_ny,time)  t_index =      1     1
      6683    13824  (     5,    50,     1) (    50,     8,     1) (    13,    50,     1) (   117,    28,     1)
              13824   6.316566797644321E-11   9.343714994332430E-15 1.3E-26  4.566923188557707E-11 8.1E-17  9.105195619604070E-13
              13824   6.316566797644321E-11   9.343714994332428E-15          4.566923188557706E-11          9.105195619604066E-13
              13824  (     5,    50,     1) (    50,     8,     1)
         avg abs field values:    6.168908270901164E-13    rms diff: 3.2E-28   avg rel diff(npos):  8.1E-17
                                  6.168908270901162E-13                        avg decimal digits(ndif): 15.8 worst: 15.4
RMS atmImp_Faxa_ndep2                3.1973E-28            NORMALIZED  5.1829E-16

Any User Interface Changes (namelist or namelist defaults changes):

Testing performed (e.g. aux_cdeps, CESM prealpha, etc):

Hashes used for testing:

[mvertens]

@ekluzek @billsacks - the problem I ran into when I originally thought of doing this in the xml is that I was trying to put the unit conversion on a stream rather than on a field - and that of course will not work if the different stream fields have different units. With this PR now the unit conversion(s) are on stream field(s).

• I have not tested this in ESCOMP - but I have tested this in NorESM successfully for the cmip6 forcing - which is where you wanted the conversion. I did not test the cplhist yet - but it should work.
• This conversion can also be applied in other parts of CDEPS if units are expected to differ between what is read in and what is passed to the mediator. However, I only implemented the change for ndep.

[billsacks]

This looks great - thanks a lot for implementing this! I just have a couple of requests about comments:

[ekluzek]

There's a couple things I ask for. The one that's important is to upgrade the constants in the stream definition file from single to double precision. I think that's important enough to do. But, it still not make answers identical because of the conversion from a divide to a multiply, but certainly a good thing to have in place.

[mvertens]

@ekluzek @billsacks - thanks for your review. I think I've addressed all of your comments.

[ekluzek]

Awesome. Looks good.

[billsacks]

Looks good - thank you!

[billsacks]

@mvertens - is it quick for you to redo your baseline test with the latest code (where you have changed the e to a d specifier)? I'm curious if you still see roundoff level diffs. I expect you probably will, but it would be good to know. If it would be at all of a pain to redo the testing, don't worry about it.