Balance errors with custom atmospheric / shortwave radiation input for single point ensemble setup

[lstrebel]

Hello,

I am running an ensemble of single point simulations and try to force them with local observations from the FLUXNET dataset with CLM5 (starting as I2000Clm50BgcCropGs but modified the atm. forcing variables).
To generate the ensemble I perturb sand and clay fraction in the surface files and air temperature, precipitation, short- and long-wave radiation in the forcings.
When I use this perturbation method with reanalysis data (e.g. COSMO REA6) as forcing, I can run an ensemble of 100 members without issues.
However, when I start using the local FLUXNET data, some members run at different times into different errors.
The errors I have noticed the most are:
- "Negative conc. in ch4tran."
- "ERROR: Methane demands exceed methane available. ERROR in ch4mod.F90 at line 3408"
- "ERROR: ERROR in CNBalanceCheckMod.F90 at line 219"

Through a process of elimination I have determined that the errors only occur once the local shortwave radiation is included in the forcings.
All the other variables do not cause errors.
However, I can not find a reasonable explanation why the shortwave radiation inputs would cause these errors for some ensemble members but not others.
I checked for missing / incorrect values or for extremes at the time the error occurs.
For example in the following plot, I marked the date, where the error occurs (for the blue line) with a pink arrow and clearly the difference to the ensemble member where no error occurs (red line) is much larger for other dates where no error occurs.



So, now I wanted to ask, if anyone here has experience with custom forcings, specifically with custom shortwave radiation input and knows what exactly in these inputs causes these kinds of balance errors.
Is there a pre-processing step I can perform that can detect / correct the shortwave radiation data that causes these errors?
Or do I have to manually perturb the specific input files for each ensemble member with an error until they also run through without balance errors?

Thanks for any advice,

Best regards,
Lukas

 

[oleson]

Based on the plot you've shown I agree that it is unlikely that it is solar radiation alone that is causing the problem. It's more likely something that is causing some unrealistic soil temperature or moisture states. You could look for any correlation with the sand and clay perturbations you are using, and examine the soil temperature and moisture profiles leading up to the error. Are your input percent sand and percent clay vertical profiles reasonable?

 

[NIU]

I had the same problem. In my case, I used the observation of FLUXNET data. The simulation results will crash after setting the spin-up to 100 years. I am also trying to solve the problem.

 

[lstrebel]

From the suggestions by Keith, I reduced the magnitude of sand and clay perturbations and it seems to work now. So in this case it looks like it was a combination of the solar radiation and the sand and clay percent vertical profiles. I had assumed / hoped that limiting the sand and clay percent to 0-100 and their sum (including organic matter as well) also to 0-100 and in combination with the lower and upper limits in the pedotransferfunction for each parameter would be enough to avoid these kind of issues.

@NIU If your simulations crashes after 100 years of spin-up this sounds like a different issue. I assume in the 100 years you are able to cycle through the forcings a few times before the crash while my issue appeared within the first few years.

 

[NIU]

From the suggestions by Keith, I reduced the magnitude of sand and clay perturbations and it seems to work now. So in this case it looks like it was a combination of the solar radiation and the sand and clay percent vertical profiles. I had assumed / hoped that limiting the sand and clay percent to 0-100 and their sum (including organic matter as well) also to 0-100 and in combination with the lower and upper limits in the pedotransferfunction for each parameter would be enough to avoid these kind of issues.

@NIU If your simulations crashes after 100 years of spin-up this sounds like a different issue. I assume in the 100 years you are able to cycle through the forcings a few times before the crash while my issue appeared within the first few years.

Thank you for your reply. Now I am doing a short simulation and the problem will not appear again.