Question about restart file

[xnnzka]

Hi everyone,
I use CESM 2.1.4 and have some questions about restart files:
1. When restarting an experiment—whether it's a branch run or recovering a crashed simulation—what is the exact function of the cam.rs file and cam.i file? I understand that cam.r typically contains the initial conditions, but what additional information do cam.rs and cam.i hold?
2. I am working with an ensemble of restart files from different initial conditions. My goal is to select the restart file that represents a model state most similar to observational data for SST or surface temperature for globe. My plan is to take the observational data for a specific date (e.g., 2000-01-01) and compare it directly with the model outputs from each restart file that correspond to the same calendar date (i.e., the state representing 2000-01-01 in each simulation). I would then calculate the pattern correlation and RMSE for variables like SST and Surface Temperature to identify the best match. I am not sure if it is a valid approach, or are there better-recommended practices? And I also want to know are there any standard metrics (excluding SST and TS) designed for this kind of "best analog" selection from restart files?

 

[xnnzka]

My restart exp type is branch. Sorry for forgeting writing this.

 

[peverley]

Hi,

To answer your first question:

• cam.i.* files represent the initial conditions files that can be used to start a new hybrid run. They are not used in a branch run.
• cam.rs.* files contain information that is passed in and out of the coupling layer between the surface and the atmosphere. It's in its own file because some surface fields are to/from the ocean and some are to/from the land. This file is used in any restart run.

In regards to your second, I'd like a bit more information:

• What compset and grid are you using?

Courtney

 

[xnnzka]

Hi,

To answer your first question:

• cam.i.* files represent the initial conditions files that can be used to start a new hybrid run. They are not used in a branch run.
• cam.rs.* files contain information that is passed in and out of the coupling layer between the surface and the atmosphere. It's in its own file because some surface fields are to/from the ocean and some are to/from the land. This file is used in any restart run.

In regards to your second, I'd like a bit more information:

• What compset and grid are you using?

Courtney

Hi Courtney,
Very thanks for your reply! Now I am clear about i* and rs*.
Now I use BHIST compset and f09_g17 grid. Looking forward to your reply!
xnnzka

 

[xnnzka]

And also, I can use hybrid type to set ensemble members through pertlim.

 

[mdfowler]

Just to add - I don't know that there are established ways to select the "best" restart file. Maybe one question is why you would want to start from a CAM restart file? There are methods for initialized prediction, where you could essentially start from reanalysis products like ERA5, which would align well with observations and from which you could let the model run freely to look at evolution of key fields. I'm not exactly sure on how to do this, but know that others have that knowledge!

 

[xnnzka]

Just to add - I don't know that there are established ways to select the "best" restart file. Maybe one question is why you would want to start from a CAM restart file? There are methods for initialized prediction, where you could essentially start from reanalysis products like ERA5, which would align well with observations and from which you could let the model run freely to look at evolution of key fields. I'm not exactly sure on how to do this, but know that others have that knowledge!

Hi, Meg
I am sorry that I did not describe my question clearly. Specifically, I use 8 restart files from CESM2LENS, modify the emission files and have run these cases to a specific year. I need to select a "best" restart file among 8 cases in this specific year and add the micro atmosphere temperature perturbation (in user_cam_nl, use pertlim) to generate ensemble members. Currently, I use RMSE and pattern correlation for surface temperature and SST to select. I am not sure if this is a good idea. My experiments use BHIST and BSSP370.

 

[mdfowler]

Thanks for clarifying! In that case, I think this approach is reasonable; I'm not aware of a "best" criteria at least. You might also consider some atmospheric variable above the surface (OMEGA500 maybe) to assess agreement for circulation patterns at initialization, but I'm not sure the pro/con of more vs. less criteria for selection.

 

[xnnzka]

Thanks for clarifying! In that case, I think this approach is reasonable; I'm not aware of a "best" criteria at least. You might also consider some atmospheric variable above the surface (OMEGA500 maybe) to assess agreement for circulation patterns at initialization, but I'm not sure the pro/con of more vs. less criteria for selection.

Very thanks for your reply! I will try it!