Parameter Constraints for Sensitivity Analysis in CLM5

[krivers]

Hi all,

I'm currently conducting a sensitivity analysis for a few grid cells using CLM5. As part of this, I've modified values in the ctsm60_params.c241119.nc file to assess the influence of specific parameters.

I'm looking for guidance on whether any of these parameters are internally constrained by others within CLM5. For example, in the NOAH-MP model, field capacity is constrained by the soil's maximum saturation—any violation of this relationship prevents the model from running. I'm wondering if similar dependencies exist in CLM5 that I should account for when creating parameter samples.

I’ve reviewed several relevant studies, including:

• Kennedy et al. (2024): One-at-a-time parameter perturbation ensemble of the Community Land Model Version 5.1
• Dagon et al. (2020): Machine learning to assess Earth system model predictability
• Massoud et al. (2019): Ensemble modeling of land surface processes
• Gao et al. (2021): Exploring parameter interactions in CLM

While these are incredible helpful, they don’t go into detail about parameter interdependencies or constraints.

If anyone has experience with parameter interactions in CLM5 or knows of relevant documentation or best practices, I’d really appreciate your insights. Understanding these constraints is key to conducting a robust and physically consistent sensitivity analysis.

Thanks in advance!


What version of the code are you using?
------------------------------------------------------------------------
git describe:
release-cesm2.1.5-0-g7a6c5b0
------------------------------------------------------------------------
release-clm5.0.37

Have you made any changes to files in the source tree?
No changes.

 

[kdagon]

This is a great question, thanks for inquiring. In our work generating and simulating perturbed parameter ensembles (PPEs) across different versions of CLM, we have encountered parameters where it's made sense to group them together in the perturbation experiments. One example of this is briefly discussed in the Kennedy et al. preprint you cited above: the case of nitrogen fixation costs (labeled there as "KCN") which actually represents 6 individual parameters that are perturbed independently and together. There are a couple other cases we've identified but haven't yet documented them in a formal way. For a comprehensive summary of CLM5 parameters, you can take a look at this spreadsheet, in particular the "flag for parameter dependencies" column H. If there is a common entry like KCN that is identifying a specific parameter grouping. For a smaller subset of parameters we are focusing on with the upcoming CLM6 codebase, you can take a look at this spreadsheet from Linnia Hawkins, also note the "flag" column E which identifies a few other groupings. Whether or not violating these would prevent the model from running is somewhat case by case, but these are the ones we think make sense to group (and there may be others!).

Soil hydrology presents similar challenges as you alluded to with Noah-MP. We have experimented with perturbing soil hydraulic properties directly, and also by perturbing soil texture, in order to include this part of the model in our PPEs. It's still a bit open ended as to how best to approach this.

Depending on what you are looking for, we have existing PPE simulations we have run with CLM5 and CLM6 on a sparse grid, both with one-at-a-time and multifactor parameter perturbations. You may also want to consider which parameter file you are perturbing relative to the codebase version you are using.

In the long term, we'd also like to better document and provide tools for generating parameter files and analyzing PPEs to support the user community. We have some public github repos set up (here and here), though they are in progress. You are welcome to interact with us there, and/or respond further here to continue the discussion.