Question about host-dependencies for CCPP schemes

[grantfirl]

During the process of adding the Tiedtke prognostic cloud scheme from GFDL's AM4 code to the CCPP, it is apparent that there is significant dependency on FMS, at least for the scheme's initialization. In the short-term, adding this scheme to be used in the UFS is the goal, so perhaps as a "prototype" leaving the dependency on FMS is OK for now. In the long-term, though, if we want this scheme to be host-agnostic, we may need to replicate the parts of FMS that the scheme uses within the CCPP.

@climbfuji @ligiabernardet @joeolson42 Does anyone have any thoughts on the best way forward?

[grantfirl]

@climbfuji I'm looking in the gfdl_fv_sat_adj, and it looks at least some of what the Tiedtke prognostic cloud scheme needs from FMS has already been put in the CCPP via the code in fv_sat_adj_init. I think that you worked on this originally. For making the fast part of GFDL microphysics CCPP-compliant, did you grab the saturation table initialization stuff from FMS code when you did this, or was it already in the GFDL MP codebase when you started work on it?

[climbfuji]

I am pretty sure that this was in the original GFDL MP codebase ...

[climbfuji]

I looked at the code again. Originally all this code was in module_gfdl_cloud_microphys.F90 in IPD, which got transferred to CCPP. It's still there, because you need to do the init in the gfdl cloud microphys init step if the fast physics are not used (there is logic that makes sure the init runs only once, i.e. to skip the table inits in the slow physics when the fast physics already did the init).

[ligiabernardet]

It at all possible, make it host-agnostic, so that we can run the code in the single-column model and other hosts. Can you explore what would be involved?

[grantfirl]

I brought this up with Joe Olson, Wayne Angevine, and Dave Turner, and they mentioned that it would probably be a good idea to use the same saturation vapor pressure curve/algorithm as the rest of the suite where the scheme will be called from, specifically to be consistent with what the Thompson scheme is using. It looks like it doesn't require lookup tables.

Even if we didn't try to be consistent with Thompson MP, after looking at the AM4 code some more (where the scheme is coming from), for the Tiedtke prognostic cloud scheme, it looks like they are initializing tables that don't even get used, so, either way, at least for this, it looks like we can avoid a dependency on FMS.

Thanks for the responses!