[CESM2.1.4] Model fails due to negative snow/water values — attempted fix in SnowHydrologyMod.F90 unsuccessful

[Jeongeun]

CESM version: CESM2.1.4
Component: CLM (SnowHydrologyMod.F90)
Machine: Olaf (SLURM-based HPC)


Hello,

I encountered an error during model execution where the simulation fails immediately due to negative mass values, specifically in SnowHydrologyMod.F90, as shown below.


To investigate, I first modified the part labeled Adjust water content inside SnowHydrologyMod.F90 to prevent negative values, as shown below.
Original:

After the change:

And then, the error no longer occurred.

To validate the changes, I ran a test simulation (initial 1955~1958) under the same conditions on Aleph (other server system) and compared CESM2 outputs.
After the simulations, I selected variables related to water and snow for comparison.
Some variables (SNOWDP, H2OSNO, QSNOMELT, SALT, SOILWATER, TWS) showed very similar values between them, but the others (SNOW, SSH) are different (I also checked the significant).


So, I was thinking this change might not be correct.
I then found an old CTSM post (Need to increase min_snow_to_keep · Issue #1425 · ESCOMP/CTSM) suggesting to change the value of min_snow_to_keep in SnowHydrologyMod.F90.
So I reverted back to the original SnowHydrologyMod.F90, and only changed the min_snow_to_keep parameter like below.

Unfortunately, the model still fails with the same negative value error.

Given this, I would like to ask:
- Is it acceptable to proceed with CESM2 simulations on Olaf using revised SnowHydrologyMod.F90 below?

- Are there any additional settings or variables we should review to ensure the results are reliable?
- What is the recommended way to handle this negative value error?

I attached config_compilers, config_batch, config_machines files.

 

[slevis]

I do not know details about the snow-capping procedure, but my gut response to your attempt that worked is that it seems fine. Common practice in numerical modeling is to document your changes to the model when you go to publish your work.

I do wonder why you get the error in the first place, while other users do not. I'm guessing that you have changed something in the code or input data. If so, it would help to know about that.

Also, in your validation test, did you compare identical simulations on olaf and aleph? If so, then the diffs are due to running on different machines, right? If not, then what else is different?

 

[Jeongeun]

Thank you for your response and helpful feedback.

To clarify, I have not modified any parts of the code within SnowHydrologyMod.F90 except for the temporary change made to avoid negative values (which I have since reverted). Other than that, the only adjustments made were to the config_compilers.xml, config_machines.xml, and config_batch.xml files to suit the Olaf system specifications.

Regarding the validation test, I used identical simulation settings (initial year: 1955–1958, compset, resolution, namelists, etc.) on both Aleph and Olaf. The only difference between the runs is the machine environment. So yes, I assume that the minor differences in some output variables are likely due to system/compiler-related effects.