CMIP7 data request → CESM3 history variables

How the variable lists were made

The CESM3 source (tag cesm3_0_alpha09e) was parsed statically, not run: every history-registration call (CAM, CTSM and FATES, CICE, MOM6 and MARBL, MOSART, CISM) gives a field’s name, dimensions, units and file:line. Names built in loops are expanded from the source’s own lists, and registrations the baseline configuration (BHISTE_MTt4s) switches off are dropped. Run logs add the “verified” status.

Downloads

Each component row has three downloads, all limited to the lines that pass the filters (requests whose spreadsheet row gives no CESM variable name are never exported). The .zip archive (top right) bundles every component’s namelist.

• namelist (user_nl_cam, user_nl_clm, …; diag_table.unresolved for the ocean): the history configuration to paste into a case, headed with the date, the experiment and how much of it rests on source evidence alone. Not yet run in a real case.
• raw_text: the plain list behind it, one VARIABLE FREQUENCY per line (:FLAG for CAM and CTSM, the time method as a comment elsewhere), CESM3 names only.
• csv: the audit trail, one row per exported line: status, component source, CMIP7 compound name(s), UID, spreadsheet line and cell, priority groups, log components, source certainty and file:line, and the estimated GB/yr, so a list can be checked against the request offline.

* How the data volume is estimated

The figure marked * (top right, and the per-component and per-line sizes) is the raw, uncompressed volume of the selected history output, in GB (10⁹ bytes) per simulated year. It is a rough planning number, not a prediction of file sizes on disk.

• One line = one variable at one frequency. Its size per time sample is 4 bytes (CESM history files are single precision by default) × the number of horizontal grid cells × the sizes of the field’s other dimensions (levels, soil layers, categories, …). The dimensions come from the field’s CESM3 source record; the cell counts and dimension sizes for the chosen grid are in data/grids.yaml and data/vertical.yaml. Land fields are priced as gridded output (one value per grid cell), as the exported namelist writes them.
• Per year = size per sample × samples per simulated year on a 365-day calendar: yearly 1, monthly 12, daily 365, 6-hourly 1,460, 3-hourly 2,920, hourly 8,760, decadal 0.1. The total is the sum over the lines that pass the filters above.
• No compression is applied. If your case writes compressed (deflated) netCDF the files will be smaller, by an amount that depends on the data. Nothing here models chunking, packing or file-system block overheads.
• Only the data variables are counted: not coordinate or bounds variables, time_bnds, file metadata, restart files, or default output that the exported files leave switched on: CICE and MOSART add to their defaults, so those components write more than the sizes shown.
• It is a lower bound. Lines with no CESM3 source record, an unsized dimension, or a frequency without a per-year count (fx, subhr) are shown as – and left out. The note under the total says how many lines were priced.