Repository navigation
Description
LND_TUNING_MODE was added as an XML variable that helps us handle parameters that need to be tuned for different major physics versions of the model. When we added it in, the hope was that we would only add options that we have tuned the important parameters for that configuration. So options that weren't tuned wouldn't be available. This would then require the user to explicitly choose a different option that was tuned. And the idea is that we'd only add new options when we have it fully tuned.
We tuned clm4_5, clm5_0 with GSWP3, CRUv7, and CAM6.0. So there are tuning options for those setups. And we did add a clm5_1_GSWP3 option which of course is not tuned, so many of the parameters just used the clm5_0_GSWP3 version. That already goes against the idea of only adding fully tuned options. In #2117 I need to add LND_TUNING_MODE options for CAM4.0, CAM5.0 and CAM6.0 that's needed for dust, so the other settings have to come from the clm5_0_GSWP3 settings. So some of the options are now a mix of things that are truly tuned and that aren't. That's probably OK, but we do want it to be clearly documented. And it would be good if the user was aware of this. The question is how to do this?
The LND_TUNING_MODE is set by default according to the compset name. And it matches the CLM physics version with the ATM forcing. The clm physics version HAS to match, but the ATM forcing isn't required to, so that you can try different tuning options for the same clm physics.
One question I have is how untuned compset options should be handled? For example, clm5_1_CRUv7?
The way I propose is that when options are chosen that aren't fully tuned, there would be some documentation on it. And we could setup the buildnml to print a warning, for those options. I think that ALL of the LND_TUNING_MODE combinations of physics with forcing be provided, but the ones that aren't tuned give a warning, and say what they are based on.
What are other's reaction to this idea?
Definition of done:
- Add all possible combinations for physics and Met forcing to config_components.xml (clm4_5, clm5_0, clm5_1 physics, and GSWP3, CRUv7, QIA, ERA5, CAM4, CAM5, CAM6, 1PT, and NLDAS2 forcing)
- Have a reduced set in namelist_definition, and buildnml that points to the right one with a warning
- Warn the user in buildnml if a tuning set is not fully tuned and based on another one.
- Add the CAM4, CAM5 forcing options
- Add warnings to buildnml that CAM4, CAM5 are based on the CAM6 tunings
Metadata
Metadata
Assignees
Labels
Type
Projects
- StatusShow more project fieldsDone
Activity
ekluzek commentedon Sep 23, 2023
What I'm doing in #2008 is to have warnings like this in preview_namelist
There's also a warning that clm5_1_CRUv7 does not exist so that clm5_1_GSWP3 will be used. The bad thing about that is that the XML variable will be set to something different than what's used in the namelist generation. The idea is that there will be a setting for each CLM physics type and forcing combination. And that will be set for the user based on the COMPSET. The user can override it, but as setup now you can't change the CLM physics option in the tuning from the CLM physics that you are running. So it seems that the XML options should be larger than the internal namelist settings. I'll add comments regarding this in both places.
I also realized that the other forcing is Qian, which we didn't tune, so it should be added, but changed to one of the others for the user. The user can suppress the warning for options that aren't available by changing the setting of LND_TUNING_MODE to the one that is available.
The advantage for developers is that the config_component.xml file will just blindly point to the physics and forcing option that the compset gives. Inside of buildnml namelist_defaults and namelist_definition it will figure out which ones aren't tuned and choose the other option, or give a warning that it's not fully tuned and the tuning is based on which one. So this part does need to be coordinated as new tuning happens. I think this is probably fine as some tuning options won't be fully tuned for a long time, or may even remain untuned and based on another.
For users I think this chooses the best default choice, but in most cases allows the user to override the default choice. I think in general users use the default choice rather than setting it themselves. The warning isn't always seen but it will point this out. We could make the warning a fail and make the user pick a valid option, but since I don't think most users change this I think that would be more annoying than helpful.
[-]How should we handle LND_TUNING_MODE?[/-][+]Change handling of LND_TUNING_MODE so user is warned of which option is used, add more supported options[/+]ekluzek commentedon Sep 23, 2023
Besides CRUv7 this also means adding these forcing options: 1PT, NLDAS2, and QIA which will point to the GSWP3 option by default.
ekluzek commentedon Sep 25, 2023
We discussed this some in today's standup with @billsacks and @slevis-lmwg. They are supportive of the idea of adding warning messages, but suggest we call it something like "Important Note: ", and make sure the naive user would know what to do with the message. With the vast space of users for CESM, many need to use CLM, but are working in other areas (for example WACCMX users) and can not by nature be an expert in it. So we need to help them as well as we can for usability of the CESM communitiy.
ekluzek commentedon Sep 25, 2023
With my above proposal if you try to use an OCN DATM forcing mode it fails at create_newcase time with a message like this:
without this it does allow you to setup a case that way although undoubtably other issues are going to happen. Maybe the simulation would work, but answers wouldn't be useful because it's daily forcing. It's better to give a message early on that this configuration is not valid.
ekluzek commentedon Apr 20, 2024
This was completed with #2464 and in ctsm5.2.0