Rapid drying of prescribed soil moisture in high latitude regions

[sukim]

Hi,

I'm running prescribed soil moisture experiments and want the model to stick to the prescribed values rather than actively evolve over time. We conducted two experiments on Derecho prescribing monthly and daily soil moisture data, where the daily values were interpolated from the monthly data. The model was run from April to June.

However, both experiments share a similar issue of rapid drying of soil moisture in the high latitude regions.
I expected the simulated soil moisture values to closely match the prescribed values. But the result concerning as there is a significant difference in the high-latitude regions. Is this because CLM allows soil moisture to evolve through hydrological processes? Still, the deviation seems too large, and surface soil moisture in high-latitude regions is depleting too quickly.

Could someone help me understand what I might be missing?


Differences in surface soil moisture (H2OSOI, levsoi=0) averaged over the first 30 days.


Globally averaged differences in soil moisture.

​

How can we enforce the prescribed soil moisture and prevent the model from diverging?
Would modifying the source code—such as repeatedly calling PrescribedSoilMoistureInterp after any subroutine alters soil moisture—help achieve this?

Any insights would be greatly appreciated!



Below is the information about my simulation setup and changes I've made.

CESM version: CESM2.2.2
Compset: FHIST (HIST_CAM60_CLM50%SP_CICE%PRES_DOCN%DOM_MOSART_SGLC_SWAV_SIAC_SESP)

user_nl_clm
use_soil_moisture_streams = .true.
stream_fldfilename_soilm = ‘/path/to/soil_moisture_data/LFMIP-pdLC-SST.H2OSOI.0.9x1.25.20levsoi.natveg.1980-2014.MONS_climo.c190716.nc’
soilm_ignore_data_if_missing = .true.

env_run.xml
CLM_BLDNML_OPTS: -bgc sp -ignore_warnings

soil moisture data: LFMIP-pdLC-SST.H2OSOI.0.9x1.25.20levsoi.natveg.1980-2014.MONS_climo.c190716.nc

BalanceCheckMode.F90
blocked calling endrun at line 470
! call endrun(decomp_index=indexc, clmlevel=namec, msg=errmsg(sourcefile, __LINE__))

 

[slevis]

First let me confirm that I understand.

You tried two simulations prescribing soil moisture from the file stream_fldfilename_soilm:
1) Prescribe monthly soil moisture
2) Prescribe daily soil moisture (the monthly data interpolated to daily)

And you hoped to get a smaller difference between "simulated" and "climatology", right?

Questions and ideas:
1) You wrote that both simulations show the underestimation, but I would expect the monthly to be MORE underestimated. Is this true?
2) Maybe also try a simulation where you do not prescribe the soil moisture or change anything else about the model (let's call it CONTROL simulation). Is this even drier than the prescribed cases? If so, how much improvement do you get by prescribing the soil moisture?
3) You may find it helpful and insightful to look at other variables that likely change in the three simulations, for example:
- drainage and runoff related variables
- evapotranspiration and its components
- partitioning of H2OSOI into liquid and solid soil moisture

 

[sukim]

Thanks for your suggestions, and you got the points right.
And just a heads-up. I'm about to spam lots of figures below.

1) You wrote that both simulations show the underestimation, but I would expect the monthly to be MORE underestimated. Is this true?
2) Maybe also try a simulation where you do not prescribe the soil moisture or change anything else about the model (let's call it CONTROL simulation). Is this even drier than the prescribed cases? If so, how much improvement do you get by prescribing the soil moisture?

You're right. The monthly simulation more underestimates soil moisture, especially over Eurasia. The panels below show the average H2OSOI or H2OSOI_PRESCRIBED_GRC from the first layer in April for each simulation and the climatology.



And when compared to the control simulation, where I neither prescribed soil moisture nor modified the source code, it's producing a totally different field of soil moisture (H2OSOI).
View attachment 6637

I assume that this discrepancy between the monthly and daily simulation occurs due to time interpolation. I didn't change soilm_tintalgo, so it's set to 'linear'. However, H2OSOI_ PRESCRIBED_GRC shows a 15-day delay. This explains some of the differences but doesn't explain the drying of high latitudes. Could it be that the prescribed soil moisture values are just too wet by default in our simulation setup?

3) You may find it helpful and insightful to look at other variables that likely change in the three simulations, for example:
- drainage and runoff related variables
- evapotranspiration and its components
- partitioning of H2OSOI into liquid and solid soil moisture

I found that there's no issue with drainage and runoff related variables, but I suspect that changes in vegetation and ice are contributing to changes in radiation and temperature or vice versa.






So, the only explanation I can think of for the rapid drying is that the default soil moisture values prescribed into the model are too wet for our simulation set up. In other words, the difference between "simulated" and "climatology" likely stems from the climatology that might be not aligned with the simulation.

 

[sukim]

Actually, I have a follow-up question regarding the prescribed soil moisture experiment. It's also related to the strange and unexpected values we're seeing in high latitudes.

The goal of these experiments was to see the response of the atmosphere to dry soil moisture over North America, expecting the strongest anomalous temperature to appear over that region. The control simulation (cntl) was run using daily soil moisture climatology, while the perturbed simulation (ptbd) was prescribed with drier soil moisture only over a part of North America. The only difference between simulations was the soil moisture over the targeted region in North America.

Although we were able to force the model to use the prescribed values, the liquid and ice component of soil moisture began to diverge as shown below. In line with this, strong temperature anomalies emerged in high latitude regions.

We forced the model to use prescribed soil moisture whenever subroutines that modifies water_inst%waterstatebulk_inst. As a result, while h2osoi_vol remains consistent, both h2osoi_ice and h2osoi_liq values start to diverge. This suggests that the partitioning between liquid and ice differs, potentially due to evolving temperature fields and vice versa.

So here's my question:
Is there a way to isolate and obtain a temperature response that is more directly tied to the prescribed soil. moisture anomalies, without introducing unexpected changes elsewhere in the model?



Difference between simulated and prescribed soil moisture at the first layer


Difference in SOILLIQ and SOILICE at the first layer between two simulations


Difference in TBOT and EVAP at the first layer between two simulations