[FEATURE REQUEST] Investigate saving REAL*8 fields to the restart file

[yantosca]

Overview

We are looking into removing differences between GEOS-Chem runs that execute over a long period, and runs that are restarted. One of the reasons for these differences might be insufficient precision in the GEOS-Chem restart files. We can experiment with saving REAL*8 data to the SpeciesRst collection and see if this helps to improve the situation.

[yantosca]

While the History diagnostics had derived types that would allow for carrying diagnostics as REAL*8, the I/O code (in History/history_netcdf_mod.F90 would cast everything to REAL*4 and then write it to netCDF. This situation has been corrected in commit 652bb23. Several modifications were made to the following modules:

• Headers/registry_mod.F90
• Headers/state_diag_mod.F90
• History/history_mod.F90
• History/history_netcdf_mod.F90

Specifically, we now have included a variable within derived-type objects that indicates the KIND value (i..e. REAL*8 or REAL*4) for the netCDF output. Prior to this, it was always assumed that netCDF variables would be written to disk with the same KIND value with which they were defined.

NOTE: Similar modifications need to be made in HEMCO in order to read in REAL*8 data. Right now HEMCO can only read in REAL*4 data from netCDF files.

[msulprizio]

This will likely go into an early GEOS-Chem 13 version

[yantosca]

The GCC History can now save output as REAL*4 or REAL*8 (cf. 6e004b7 merge).

However, at present, HEMCO can still only read and carry REAL*4 data in its data structures. This needs to be resolved first before this issue is closed.

[yantosca]

The remaining work needs to be done on the HEMCO side, as the GEOS-Chem "Classic" History component can now write REAL*8 fields to disk.

Therefore, I am closing out this issue, and have opened a new feature request (geoschem/HEMCO#69) on the HEMCO repository issue tracer,