Immutable Variables

[zmenzo]

Hello,

I plan to run RCP (4 & 8) simulations with MPAS and therefore need to allow the model to spin up to the imposed SST. I plan on running the model for a year as Spin-Up. Do you think that is long enough/too long for the model to reach equilibrium?

Also, during the Spin-Up, I plan to remove as many variables as possible to save on cost. Are there any immutable variables in the streamlists that are necessary for the model to run (i.e., xCell, zVertex, etc.)? Or can I run the model for a year without producing any variables?

Thanks as always,

Zach

 

[mgduda]

Regarding time to adjust to imposed SSTs, that's probably a science question that I'm not qualified to answer. Hopefully some past studies with the CESM may be able to provide some guidance here. Or, if some other, kind forum users would like to respond, that would be most welcome, too!

When it comes to model output, it is certainly possible to run MPAS-Atmosphere without writing any NetCDF output (you'll always get text log files, of course); if you'd like to do this, all that's needed is to set output_interval="none" for each of the output streams in the streams.atmosphere file. If you'll be running for long periods of wallclock time, though, you may want to periodically write restart files in case, e.g., you run out of wallclock time on batch nodes, or in case the model crashes and you'd like to restart with some modifications to your namelist.

 

[zmenzo]

Gotta love when the solution is so simple.

Thanks as always,
Zach

 

[zmenzo]

Hello,

I was able to run the model with the output set to 'none.' However, I wanted to understand better the purpose of variables such as zCell, indexToCellID, lonEdge, etc. I'm just concerned that I would run a long simulation without them, only to find that they are required for something like mpas_convert or other crucial MPAS-related operations.

Thanks as always,
Zach

 

[mgduda]

Variables without a time dimension can probably be omitted from your model output streams without any real consequence -- as long as you have one copy of these variables, e.g., in your model initial conditions ("init.nc") file, that should be sufficient. Most of the post-processing tools that we support should be capable of reading time-independent variables from one file (again, probably your "init.nc" file) while reading the time-varying fields to be processed from a different file (perhaps a model history or diag file).

For example, the latitudes and longitudes of cells, edges, and vertices don't change in time, so there's no need to write xCell, yCell, zCell, latCell, lonCell, xEdge, yEdge, zEdge, latEdge, lonEdge, xVertex, yVertex, zVertex, latVertex, or lonVertex to every history file. Similarly, the geometric heights of model vertical coordinate surfaces don't change in time, so the zgrid field also doesn't need to be written to every history file. I think we had included many of these fields in the default history stream to simplify post-processing in the past, and as our tools evolved to not require time-invariant and time-varying fields to be in the same file, we simply never updated the default history stream.