CCPP Framework Meeting Minutes 2019 02 28

Discussion with NRL (Alex Reinecke, John Michalakes)

• upcoming changes to metadata format and new parser (two-step change)
  • new features, benefits, challenges
  • time constraints do not allow transitioning FV3 before the training at EMC
• psychedelic music from google meeting
• diagnostics plan and restarts: coming after the transitioning to the new metadata and the new parser
• build options: dynamic build, static build: notion that NRL would also like the option of multiple suites compiled into one executable for the static build

NUOPC/ESMF layer on top of CCPP: use cases

• chemistry
• radiation
• lsm
• anything that can be sent to somewhere else (GPUs, different nodes, ...)

Avoid redundancy of namelist options and suite definition file

• having to specify consistent options in namelist (e.g. choice of microphysics) and in SDF (which MP scheme to run) is error-prone and should be avoided
• scheme-dependent namelist options (e.g. tuning parameters): easier to deal with, because the model has to accept how a scheme calls its namelist options
• model-dependent namelist options (e.g. choice of microphysics imp_physics in FV3) for which the scheme does not have to know what "its corresponding option" is:
  • have a registry that translates scheme names into namelist options
  • where should the namelist read/override sit, and when should it be done
  • may require breaking up some of the logic of setting up DDTs, initialization etc in FV3
  • discuss this on a github issue?

Feedback from NRL

• possibility to maintain own ccpp-physics library:
  • a common framework should be able to draw physics from multiple libraries
• when can we guarantee backward compatibility (physics, host)
  • v2 metadata will be backward compatible
  • host model interface to CCPP may see some changes, for example wrt. error handling
• NRL will need multiple instances of physics data types in Neptune, i.e. can't use module variables inside a CCPP_data module as it is done now in FV3 (and in the first shot we took with Neptune); most of this will be resolved by moving to the new cap_gen utility