search

E3SM-Project / ACME-ECP

E3SM MMF for DoE ECP project
Other
9 stars 1 forks source link

Fix Physgrid #109

Closed whannah1 closed 4 years ago

whannah1 commented 4 years ago

This includes some fixes to restore physgrid capability. Most important, the 2-way tendency mapping that was originally added when physgrid was merged to ECP is removed, including several restart file variables. This PR also includes a fix for a restart bug when running aquaplanet cases with the physgrid and support for the ne120pg grid. There is also partial support added for some other girds (ne4pg3, ne4pg4, ne30pg3, ne30pg4) which are not planned for serious use, but are useful for testing and may likely be removed later.

whannah1 commented 4 years ago

There's a lot of changes to go over here, and a lot of them probably won't make any sense, but anything that involving removal of variables of the form "*_in" relates to removal of the 2-way physgrid tendency mapping that was deemed unviable because of drift issues. For the other stuff I'm happy to clarify.

whannah1 commented 4 years ago

Note: I originally intended to add ne120pg2 support in E3SM first and have it percolate down to ECP. However, there's an issue running with this grid where the domain check in CIME complains that the domain file coordinates are very slightly off from the model coordinates. The error message I typical saw was:

(seq_domain_check_grid)  maximum           difference for lon  0.147792889038101E-11
(seq_domain_check_grid)  maximum allowable difference for lon  0.100000000000000E-11

I tried various things to remove this discrepancy without success, but this can easily be avoided by setting EPS_AGRID=1e-11 in env_run.xml. There does not seem to be any way to have this set automatically for this specific grid, so I decided to hold off adding support for a grid that I knew would not work in E3SM. But in the interim, the ECP project is small enough that anyone who runs into this problem will probably ask around and quickly realize what needs to be set.