CESM Models

CESM 1.0 CICE PORT Validation

 

In a successful port, CESM CICE sea ice model solutions are expected to be the same "to roundoff level" as solutions generated on a trusted machine, although determining what is and what is not roundoff level is somewhat difficult and subjective. Presently no automated testing tools exist for validating the successful port of the CESM CICE model to a new system.

In order to facilitate the evaluation of a successful port, we suggest that you set up and run on your local machine the two cases described below. When time permits, we plan to create scripts to automatically set up and run those cases, but until then, you will need to apply these steps manually.

The two cases ideally would be an ice-only control run. This is a D compset in the CESM. The steps to set up the cases are as follows:

1. Create a case with a D compset and T62_gx1v6 resolution on your machine.
2. Configure your case and turn on every timestep output by setting histfreq = 'm','1','x','x','x' in the cice.buildnml.csh file or via user_nl_cice. Also, make sure the variables aisnap and hisnap go to the timestep stream by setting f_aisnap = 'm1xxx' and f_hisnap = 'm1xxx'.
3. Run your case for 5 days (default out of the box).
4. Concatenate (ncrcat) the timestep history files (h1_inst) into a single file.
5. Set up a pertubation test identical to the control case. Make sure to turn on the timestep output.
6. Set the number of subcycles in the EVP dynamics to 240, ndte = 240 in the CICE namelist.
7. Run the perturbation test for five days.
8. Concatenate (ncrcat) the timestep history files (h1_inst) into a single file.
9. Run the script (below) on the two concatenated files to compare your tests.
10. Contact the PCWG to get the control and perturbation tests from bluefire to compare to your tests.

The bluefire control (D_cntle) and perturbation (D_perte) cases are available through the PCWG community liaison.

Following the successful execution of these two cases, you can run a script to determine the error growth in sea ice thickness over a five-day run. Next, you can run out the control for 10 years with monthly-mean history (turning off the timestep output). The ice diagnostics package can be run on the ten-years of monthly-mean history files to as another check to verify that the simulations are indistinguishable or not. Again, contact the PCWG liaison to access the 10-year control run from bluefire. A succesful port should have an RMS error difference of less than 0.001 m in global ice volume in the five day test and the ten-year runs should be nearly indistinguishable in the time series plots. Spatial differences in the two runs will be larger, but these should appear to be random with a fairly equal distribution of plus and minus differences. While more systematic differences may not indicate a problem, these are usually indicative of a problem with the port, particularly with larger RMS differences.

Port-Validation Plots

The following plots illustrate a successful port to the NCCS Cray XT jaguar. The plots display timeseries of RMS differences of the sea ice volume field between the jaguar solutions and those generated on a "trusted machine," the NCAR IBM bluefire.

Tools

The CICE diagnostic package is available at:

The sea ice thickness difference script is available at:

Contact the PCWG liaison to obtain the datasets or help with this procedure.