Changes in cloud/radiation interaction in GSD physics suite that uses

[tanyasmirnova]

Thompson MP, MYNN pbl and GF convection:

1. Switch the order of calls, first MYNNrad_pre (or SGSCloud_RadPre),
   then rrtmg_pre. This will add sub-grid clouds from MYNN PBL (or MYNN PBL and GF)
   to QC and QI, and these updated hydrometeors will be used to compute cloud paths
   and effective radii.
2. In rrtmg_pre with the use of THompson MP:
   • use Thompson's subroutines make_IceNumber and make_DropletNumber
     to compute number concentrations for subgrid clouds.
     • use calc_effectRad to compute effective radii for QC and QI
       with sub-grid clouds.
     • added option (clduni) to use the same subroutine to compute
       water paths as with the GFDL MP. For this input.nl should set
       effr_in=.true.
     • the progcld5 is used mostly to compute Xu-Randall cloud
       fraction.
3. Added SGSCloud_ modules to replace MYNNrad to add all subgrid
   clouds to QC and QI (from MYNN PBL and GF conv).
4. Added convective clouds qci_conv to GF scheme and SGSCloud_RadPre.
5. Computation of total cloud fraction in progcld5 is change not to
   depend on shallow/deep convection. Not needed in the current version of
   GSD suite.

[tanyasmirnova]

@hannahcbarnes
Thank you for taking a look at the modified code and also for checking with @haiqinli.
I can reply with "yes" to all four questions you asked in your comment. Therefore, it means that we can proceed with merging the changes into the gsd/develop branch.

[tanyasmirnova]

@climbfuji
Dom, one question: the sgscloud_radpre updates QC and QI with sub-grid clouds. Do we need to add qc, qi and also nwfa to the argument list in GFS_rrtmg_pre, or the updated values are already in the Statein%qgrs(:,k2,j) which will be sent to tracer1 in GFS_rrtmg_pre:

do j = 2, NTRAC
do k = 1, LM
k1 = k + kd
k2 = k + lsk
tracer1(:,k1,j) = max(0.0, Statein%qgrs(:,k2,j))
enddo
enddo

I wonder if the following loop is redundant:

  if ((Model%do_mynnedmf.or. (imfdeepcnv == imfdeepcnv_gf)) .and. Model%kdt > 1) then
  ! for MYNN PBL and GF convective include subgrid clouds into tracer1 
    do k = 1, LM
      k1 = k + kd
      k2 = k + lsk
      tracer1(:,k1,ntcw) = max(0.0, qc(:,k2))
      tracer1(:,k1,ntiw) = max(0.0, qi(:,k2))
    enddo
  endif

[climbfuji]

If the SGScloud use the correct variables (standard names) that correspond to the qgrs entries, then there is no need to do anything else. I don't know about your question whether the loop is redundant or not.

[DomHeinzeller]

I am working on this code and there are several problems with it. Generally, the changes are extensive and it will be difficult to propose them to EMC. I am trying to encapsulate the changes in easier-to-recognize blocks, but I am wary that they will accept them (unless we show significantly better results, of course). Other issues:

• The code is faulty: if Thompson without aerosols is used, it will segfault because it tries to address arrays that are not allocated.
• The code changes the answer for all microphysics using progcld5. This includes WSM6 (probably not an issue, because not in CCPP and won't be ported) and Ferrier-Aligo (imp_physics = 15). I do not think we can make the changes in radiation_clouds.f without going through a proper testing of the Ferrier-Aligo microphysics and getting the developer's buy-in.
• Coding styles should be followed: 2-space indents, vertically aligned blocks, lowercase Fortran statements (if, do, subroutine, ...).
• Several variables are declared as intent(out) but in reality they are intent(inout), because there is an option to return from the subroutine right at the beginning (for example if it is the first time step), some other intents are incorrect/there is a mismatch between the Fortran source code and the metadata.

I am importing the branches for ccpp-physics and fv3atm into the current version of the code, and I am fixing all of those as I go through them, except the change to radiation_clouds.f. The question is if we need to create progcld6 to use instead of progcld5 if certain conditions are met (for example, if SGScloud_{pre,post} is used? or every time Thompson MP is used?), or if the original progcld5 works for us.

[tanyasmirnova]

Dom,
Thank you very much for fixing the issues with my code changes. Yes, I forgot about the non-aerosol Thompson.
Considering the progcld5...
With the current version, I switched to use progclduni with the Thompson MP. The option to use progcld5 remains if clduni=.false.
The main change is progcld5 is to remove the dependence of total cloud cover computation on shallow and deep convection, which, as we discussed with John Brown, is the left-over from Thompson MP implementation in the old GFS. I was thinking to make a separate subroutine that will only compute Xu-Randall cloud cover fraction and skip the rest of progcld5, but did not get to this yet, and use progcld5 for now. If we make cldfrac subroutine inside the radiation_clouds.f then changes to progcld5 could be removed, and the original version reinstated.
I would be glad to work with you on the clean-up changes if necessary.
Thanks,
Tanya