Questions about parameter tuning and spin-up

[Jeline]

Dear all,
I’ve run into difficulties regarding parameter calibration and spin-up procedures.

(1) Parameter calibration:
Our goal is to calibrate PFT-related parameters in CLM using an I compset to improve the accuracy of simulations, and then run the model with a B compset.

• Does CESM provide any official tools for parameter calibration?
• If not, is it feasible to perform calibration using other approaches?
• Are there any recommended or widely used methods?

I’m concerned because running CESM is computationally expensive, making it impractical to perform repeated runs with algorithms like genetic optimization. Also, I found very few discussions about parameter calibration on the forum, which makes me a bit uncertain about the best approach.

(2) Spin-up question:
I plan to run a BHIST_BPRP simulation (1980–2020) to study the climatic effects of global afforestation. As I understand it, there seem to be three possible spin-up options:

1. Run B1850 with the default forcing data for 1850–1980.
2. Repeatedly cycle the 1980–1990 forcing data to spin up B1850 for 1900–1980 (80-year spin-up).
3. Run B1850 with the default forcing data for 1900–1980 (80-year spin-up).

Which of these three approaches would be considered correct or most appropriate?
Sorry if some of my questions seem basic. I would be truly grateful for your advice and any guidance you could share.
Many thanks in advance!

 

[slevis]

I am not aware of official Parameter Calibration tools. I will let others recommend methods. You may wish to look in the literature for insights into this type of work. One paper I can think of is Kennedy et al. (2025).

Model spin-up is part of your scientific method, so you and your collaborators should agree on a method. Again, reading how others spun up their simulations in the literature can offer insight. Also, since you're talking about B-cases, I will move this thread to a more appropriate forum.

 

[slevis]

I will add that, if one runs with active biogeochemistry (bgc), then it's beneficial to do the long bgc spin-up in land-only mode before going on to the coupled mode. The CLM User's Guide discusses land-only spin-ups. Other threads in these forums have discussed the topic, as well. And I'm repeating myself when I say that you will also find information in the scientific literature.

 

[Jeline]

I am not aware of official Parameter Calibration tools. I will let others recommend methods. You may wish to look in the literature for insights into this type of work. One paper I can think of is Kennedy et al. (2025).

Model spin-up is part of your scientific method, so you and your collaborators should agree on a method. Again, reading how others spun up their simulations in the literature can offer insight. Also, since you're talking about B-cases, I will move this thread to a more appropriate forum.

Thank you very much for your reply! I will carefully review these references and reconsider the spin-up scheme. Now I have some new questions:

1. I understand that BPRP and BDRD differ in how atmospheric CO₂ concentration is treated when calculating CO₂ surface fluxes—whether it is prescribed (constant) or interactively coupled between the atmosphere, land, and ocean. However, I’m not sure how to decide which one to use. I’m working on feedback from afforestation on the global carbon and water cycles during 1980–2100, using customized surface data and landuse timeseries data. My understanding is that BPRP would be more realistic when focusing on carbon uptake, yet some studies related to carbon sequestration still used BDRD, which really confuses me.

2. I know that CMIP6 historical forcing data end in 2014, and CESM historical forcing is consistent with that. If I want to run BHIST up to 2022, is that feasible, or would it cause inconsistencies in the forcing setup?

 

[slevis]

1. As you said, you can publish a study in the scientific literature as long you are able to explain and justify your choices and assumptions. As a scientist, you will decide what makes the most sense for your research. And you may try one thing and find that it doesn't do what you had hoped. Then you may decide to repeat the work with a different choice or assumption. These things are worth discussing with more experienced scientists, so I will leave it open to others in the community to offer feedback.

2. Running BHIST to 2022 seems equivalent to running BHIST to 2014 and then continuing with one of the BSSP cases, so it is feasible and I think you just need to pick the SSP you want and the rest will fall into place.