2. Running and modifying the CESM land ice component¶

This section provides an overview of some of the most important information needed when running a land ice-focused CESM case. This assumes you are familiar with the basic process of obtaining and running CESM, as described in the CESM Quickstart guide and the CIME documentation.

In CESM, land ice processes are simulated by two components: The land ice (glc) component and the land (lnd) component. The land ice component is CISM, the Community Ice Sheet Model; the land component is CLM, the Community Land Model (which is now part of CTSM, the Community Terrestrial Systems Model). CLM is responsible for computing the surface mass balance and surface temperature of ice sheets, along with snow pack evolution for all land cover types. CISM is responsible for ice sheet dynamics and other ice sheet internal processes.

2.1. Choosing a CESM configuration for land ice work¶

2.1.1. Choosing a compset¶

CESM high-level configurations are known as “compsets” (short for “component sets”, described in more detail in the CIME documentation, the CESM documentation, and the table of available compsets). At the compset level, there are three main modes for configuring CESM’s ice sheet component:

Using CISM with ice evolution turned off.

This is the standard configuration used by most CESM compsets. These compsets have CISM2%NOEVOLVE in their compset long name. CISM is built into the system and is called periodically by the CESM coupler, but it does very little. CISM serves two roles in the system in this configuration:
- Over the CISM domain (typically Greenland in CESM2), CISM dictates glacier areas and topographic elevations, overriding the values on CLM’s surface dataset. CISM also dictates the elevation of non-glacier land units in its domain, and only in this domain does CLM downscale atmospheric fields to non-glacier land units.
- CISM provides the grid onto which SMB is downscaled.
Using CISM with ice evolution turned on.

These compsets have CISM2%EVOLVE in their compset long name, and typically have the letter G somewhere near the end of their alias. CISM is fully active, receiving SMB from CLM and evolving dynamically. In addition, by default, CISM feeds information back to other climate system components — i.e., it is two-way coupled. Specifically, it sends:
- Glacier areas and elevations to CLM
- Ice and liquid runoff to the ocean
Part or all of this two-way coupling can be turned off if desired. Note that the liquid runoff sent to the ocean consists of meltwater computed in the interior or at the base of the ice; surface liquid runoff is handled separately by CLM.
Using a stub glacier component (SGLC), completely avoiding the use of CISM.

These configurations have SGLC in the compset long name, and typically have Gs somewhere near the end of their alias. This is similar to (1), and CLM still computes ice sheet surface mass balance. However:
- Glacier areas and elevations are taken entirely from CLM’s surface dataset, and CLM does not perform any downscaling to non-glacier land units.
- Although SMB is still computed in CLM, it won’t be downscaled to a high-resolution ice sheet grid.
There are a few possible reasons why you may want to run in this configuration:
- Single-point and regional runs over glacier regions, where it doesn’t make sense to run a dynamic ice sheet model.
- Runs with a Gregorian calendar (i.e., with leap years): CISM does not currently support a Gregorian calendar, so these runs need to use SGLC. This includes data assimilation and CAM specified dynamics runs.
- Other cases where you don’t want to include CISM because of the extra complexity this involves, such as when setting up a new grid (which would require additional mapping files for CISM).

After creating a case, you can switch between (1) and (2) by setting the xml variable, CISM_EVOLVE.

It is also possible to run with the older CISM1 physics, using the serial, shallow-ice- approximation dynamical core (Glide), rather than the parallel, higher-order dynamical core (Glissade). This can be selected at create_newcase time by changing CISM2 to CISM1 in the compset long name (aliases using CISM1 have G1 in place of G), or after case creation by setting the xml variable, CISM_PHYS.

2.1.2. Choosing a CISM grid¶

The CESM components can be run on a variety of grids, listed in this table. The atmosphere and land are often run on the same grid, but CESM supports running them on different grids. The ocean and sea ice components must always run on the same grid. The ice-sheet component has a grid for each active ice sheet. Unlike the other component grids, which are global, the ice-sheet grids have limited domains. The current grids are polar stereographic projections with rectangular grid cells.

Currently, CESM only supports running CISM over Greenland. When running with CISM2 (Glissade), the standard grid has a resolution of 4 km; when running with CISM1 (Glide), the standard grid has a resolution of 5 km. Both CISM1 and CISM2 also support a 20 km grid for software testing purposes. There is out-of-the-box support for running either the 4 km (CISM2) or 5 km (CISM1) grids with most or all of the commonly-used atmosphere/land and ocean grids.

The alias for a typical CESM grid lists the atmosphere/land grid followed by the ocean grid. For example, f09_g17 runs the atmosphere and land on a 0.9°x1.25° finite-volume grid and the ocean and sea ice on a displaced Greenland pole 1° grid. If you don’t specify the CISM grid explicitly in the grid alias, it will use the default grid (4 km for a compset with CISM2, 5 km for a compset with CISM1). For some common grids, you can also specify the grid explicitly in the alias, using a _gl element following the ocean grid. For example, for a compset with CISM2, f09_g17_gl4 is equivalent to f09_g17.

For the T1850G compset (described in Section 3), you should use grid f09_g17_gl4. For information on introducing new ice sheet grids, see A.1 Introducing a new ice sheet grid.

2.1.3. Special considerations for hybrid cases¶

As described in the CIME documentation, hybrid cases are commonly used to start a case from a pre-existing case. Generally, all components start up from the restart files in the reference case. However, in some situations, you may want to run a case that is a hybrid start for most components — so that they start from spun-up initial conditions — but for which CISM starts with observed initial conditions instead of a restart file. For example, you may not have a CISM restart file for this case, or the scientific purpose of the run might warrant starting with observed initial conditions rather than an existing restart file.

To do this, set the CISM_OBSERVED_IC variable in env_run.xml:

./xmlchange CISM_OBSERVED_IC=TRUE

This will force CISM to start from the same observed initial conditions that are used for a startup run.

Note that CISM_OBSERVED_IC is ignored for startup runs; for branch runs, it must be FALSE.

If you are doing a hybrid run where you have changed cisminputfile to point to a restart file from a standalone CISM case (i.e., a run done outside of CESM) or a case with different physics options, then you must also set restart = 0 in user_nl_cism. (Otherwise restart has a default value of 1 for a hybrid case, which can give incorrect behavior if you are using a restart file from a case with different physics options.)

2.2. Modifying namelist settings¶

Once the code is running, you may want to change namelist or configuration variables. Variables related to land ice are set in the files cism_in, cism.config and lnd_in. These files appear in the run directory, and also in the CaseDocs subdirectory of the case directory. User modifications can be made to these files by adding lines to user_nl_cism (for variables in cism_in or cism.config) or user_nl_clm (for variables in lnd_in); this is described in more detail below. The various user_nl_xxx files are created when you first run case.setup for your case. They can be modified any time between running case.setup and the start of the run: the model does NOT need to be rebuilt after making namelist changes in these files.

Most parameters directly relevant to ice sheet modeling are set in cism.config. This config file contains settings used by CISM — for example, grid information (which is set automatically based on the resolution set at create_newcase time) and physics parameter settings. The cism_in file contains some additional parameters controlling the CISM run. All of the available settings are given in this table (variables in namelist groups with config in their name will go into the cism.config file; others will go into the cism_in file).

The file lnd_in provides settings for CLM. CLM’s available settings are given in this table.

Changes to both cism.config and cism_in can be made by adding lines with the following format to the user_nl_cism file in your case directory:

namelist_variable = value

Note that there is no distinction in user_nl_cism between variables that will appear in cism_in vs. those that will appear in cism.config: CISM’s build-namelist utility knows where each variable belongs. For example, to set the value of cism_debug to .true. and dt to 0.05, include the following in user_nl_cism:

cism_debug = .true.
dt = 0.05

After running preview_namelists, the following will appear in cism_in:

&cism_params
...
cism_debug = .true.
...
/

and the following will appear in cism.config:

[time]
...
dt = 0.05
...

Changes to lnd_in can be made by adding similar lines to user_nl_clm. For example, to change the ice albedo (values give albedo in the visible and near-infrared), add the following line to user_nl_clm:

albice = 0.55,0.45

After changing any of the user_nl_xxx files, you can preview the generated namelists by running the preview_namelists utility in the case directory. Generated namelists will then appear in the CaseDocs subdirectory of your case as well as in the run directory.

Note: There appears to be a bug in the parsing of strings in user_nl_cism that are bound for cism.config: These appear to be handled correctly if they are single-quoted, but double-quoted strings lead to buggy behavior.

2.3. Modifying source code¶

2.3.1. Source code directory structure and repositories¶

Within a CESM checkout, after all components are obtained with the manage_externals tool, CISM source code can be found in the directory components/cism. The source code for CISM within CESM is contained in two separate git repositories:

The code for CISM itself is in the directory rooted at components/cism/source_cism. This is a git repository associated with the CISM GitHub repository.
The wrapper code for connecting CISM with CESM is in the other subdirectories of components/cism. This top-level directory is a git repository associated with the CISM-wrapper GitHub repository.

Other subdirectories of the top-level CISM-wrapper code are:

source_glc, which contains much of the logic for driving CISM via CESM
drivers, which contains the code for coupling with CESM via the MCT coupling framework
bld, which mainly consists of code and xml files needed to create the namelist and configuration files. Note that the actual build of the model is handled by files in cime_config. The most important files in bld are build-namelist and the xml files in the namelist_files subdirectory, which describe all possible namelist / configuration settings and their default values.
cime_config, which contains build scripts and configuration files used by CIME (the scripting and infrastructure of CESM).
tools, which contains tools (now somewhat obsolete) for generating land/ice-sheet grid overlap files
test, which contains code (now somewhat obsolete) for testing some parts of the source code
mpi and serial, which have appropriate versions of source code that can be used for parallel and serial runs, respectively. The serial directory is obsolete; now the mpi directory is used even when running on a single processor.
manage_externals, which is the same tool that CESM uses to obtain its components. This instance of the tool can be used to obtain a working set of code when cloning directly from the CISM-wrapper GitHub repository.

There are two ways to modify source code for a CESM case:

Changing code in-place in the source tree
Using SourceMods

These two methods are described in more detail below.

2.3.1.1. Changing code in-place in the source tree¶

Changing code in-place allows you to make a git branch of the source code and do your work leveraging the power and convenience of git. As noted above, there are two separate git repositories making up CISM within CESM. If you need to modify code both in CISM itself and in the CISM-wrapper, you will need to create a separate git branch for each (as well as a separate GitHub fork of each repository if you want to push your work up to GitHub).

This method is best for changes that have some of these characteristics:

Long-term developments
Incremental changes towards a final solution
Changes that apply to many cases
Changes that touch many files
Any change intended to eventually come back to master

2.3.1.2. Using SourceMods¶

Within a case, there is a SourceMods directory that can contain modified source code that applies to that specific case. The location for modified CISM files depends on whether the files are part of the CISM code itself or part of the CISM-wrapper code. For any source code that appears in the source_cism subdirectory of components/cism, the modified file should go in $CASEROOT/SourceMods/src.cism/source_cism. For any other source code (e.g., modifications to code from source_glc), the modified file should go directly in $CASEROOT/SourceMods/src.cism. (C++ code in the source_cism directory can NOT be modified via SourceMods; this C++ code must be modified in-place in the main source code directory.) Once the modified files are in place, the code can be rebuilt. The files in the SourceMods directories automatically replace the files of the same name in the model directories.

This method is suitable for changes that have these characteristics:

Short-term developments (merging in new changes from master is a pain and very error-prone: you don’t have the version control system to help you keep track of your changes over time)
Changes that apply to just one or two cases
Changes that touch just a few files