Repository navigation
Description
This issue is created on the basis of a discussion with @kgoergen
Overview
During the evaluation of the EUR-12 domain, unrealistically high negative evapotranspiration values were identified over the African continent, specifically in regions with sandy soil. These anomalies significantly disrupt the surface energy + mass balance and lead to biased ground temperatures.
Key Observations:
- Magnitude: Values remain excessively high during both day and night. Order of 100-1000Wm-2
- Location: (Dry) sandy soil regions in Africa.
- Dependency: The issue persists only in coupled simulations with Parflow (eCLM-ParFlow and ICON-eCLM-ParFlow). It is absent in standalone runs or ICON-coupled simulations (eCLM or ICON-eCLM).
Problem Description
Initial investigations have ruled out the ParFlow namelist, indicator files, and eCLM surface data (surfdata) as the primary causes. The excessive evaporation appears to be triggered by how eCLM processes hydraulic state variables received from ParFlow. In sandy soils, where hydraulic conductivity and pressure gradients can be extreme, the coupling interface seems to communicate a state that eCLM interprets as "unlimited water availability," regardless of actual physical constraints, and the integrated checks of eCLM about the water and energy balance are deactivated when coupled with ParFlow. All of these suggest that the error originates in the coupling logic rather than the configuration.
When coupling with ICON, the excessive ET values also have an effect beyond the African continent. Although the excessive ET values are only observed in dry sandy soils, the underlying issue may affect the entire domain.
Figures of analysis
I took the following EUR-12 setup with ICON-CORDEX forcing. The simulation time shown is January of 2018, even though the effect is present at the whole year (not shown). The spinup was created by a coupled eCLM-ParFlow simulation from 1960-2018 using the same atmospheric forcing, there was also a spinup created with eCLM standalone for the same simulation period and atmospheric focing.
All of the following eCLM quantities shown are remapped to the ParFlow grid using cdo to easily view it with the mean of ncview.
Figure 1: shows the ET at the beginning of the simulation (2018-01-01 00UTC). The left column is eCLM standalone with eCLM standalone spinup, the center column is eCLM with eCLM-ParFlow spinup, the right column is eCLM-ParFlow with eCLM-ParFlow spinup.
Figure 2: same as Figure 1 but for 2018-01-31 23UTC (end of one-month simulation).
Figure 3: shows the ground temperature at the beginning of the simulation (2018-01-01 00UTC). The left column is eCLM standalone with eCLM standalone spinup, the center column is eCLM with eCLM-ParFlow spinup, the right column is eCLM-ParFlow with eCLM-ParFlow spinup.
Figure 4: Pressure head of ParFlow in coupled eCLM-ParFlow simulation at January 2018 started with -3m (left: uppermost layer, center left: 2nd layer from top) and with eCLM-ParFlow spinup (right: uppermost layer, center right 2nd layer from top).
Background
When coupled with ParFlow, eCLM exhibits a significantly stronger evapotranspiration (ET) response compared to standalone simulations, as illustrated in the contrast between the right and left columns of Figure 1. Even when the standalone eCLM simulation is initialized with identical conditions to the coupled run (Figure 1, center column), it does not reproduce these extreme ET values, suggesting the issue is rooted in the active coupling interface rather than the initial state. This discrepancy intensifies toward the end of the simulation (Figure 2) and fundamentally disrupts the surface energy balance, leading to biased ground temperatures (Figure 3). The physical driver appears to be the pressure head in the uppermost soil layer, which reaches extremely high negative values (Figure 4) consistent with the soil retention curve for dry sand, while the layer immediately below remains at approximately 10m. These pressure head values also show up in TSMP1 simulation (not shown), but are not taken into account in the coupling, as shown below).
The difference between TSMP1 and TSMP2 is that TSMP1 couples saturation, whereas TSMP2 couples liquid water content; in both cases, pressure head is also coupled (as can be seen in the code).
High negative pressure head values translate to high water demand, which is taken from the atmosphere.
Frozen Soil Decoupling:
In TSMP1, the moisture update was explicitly decoupled for frozen soil. This safeguard appears to be missing or handled differently in the current eCLM coupling. As the phase change of water in soil is a broader subject, this will be discussed in another issue.
Summary
Based on the current findings, I think that the current coupling interface between eCLM and ParFlow may require a functional revision on how to apply the pressure head in eCLM. Further details needs to be discussed.
Activity
s-poll commentedon Apr 1, 2026
Update:
I created the branch dev-high-et-debug to make some of the involved quantities globally defined and allow for more output variables. Here are some further insights, even though due to this investigation the role of the pressure head in eCLM became unclear to me.
As expected, the ground evaporation is the part which is leading to the high ET values:



The soil matric potential receives the pressure fields correctly from ParFlow's pressure head (in this branch with the same - no effect of limiting). The top-layer is shown.
Here are the components used to calculate the ground evaporation
eCLM/src/clm5/biogeophys/BareGroundFluxesMod.F90
Lines 334 to 372 in 98054f5
The moisture at ground seems to be very low.
(
forc_q,qg)(



raw,raiw,soilresis):Modifying the pressure head
When limiting the pressure head to just
-10instead of-10e8, the results are unexpected at least to me:Shown



ET,SMP(top-layer),RAIWSaturation comparison eCLM/ParFlow
The saturation is taken from eCLM output by using
saturation=H2OSOI_VOL/WATSATand the output of ParFlow. Left column: eCLM; Right column: ParFlow of the uppermost soil layer.These are my findings:
kvrigor commentedon Apr 1, 2026
FYI @s-poll and @kgoergen: sending a constant field from ParFlow to eCLM works as expected. More precisely, hardcoded pressure and h2osoi values from ParFlow remain exactly the same after being received in eCLM:
Note that
PFL_PSI_GRCandPFL_SOILLIQ_GRCare eCLM history fields which correspond to the pressure and h2osoi fields sent from ParFlow. To replicate this test, simply modify these two lines in ParFlow source code:Also, to log the received ParFlow fields in eCLM history file, just add
hist_fincl1 = 'PFL_PSI_GRC', 'PFL_SOILLIQ_GRC'inlnd_innamelist.kgoergen commentedon Apr 1, 2026
Thanks @kvrigor. This is helpful. According to the original idea, as I understood it, could you also set the receiving variable in eCLM as a constant, like so:
This should lead to either a compile-time of run-time warning or error once the constant field shall be in way modified.
s-poll commentedon Apr 1, 2026
@kvrigor Thanks for the update. This also matches the findings in #103 and the one in the previous post. However, could you please check if the saturation is equal in both model components, e.g. by using 'h2osoi_vol'+'watsat'and the saturation provided by ParFlow. This would clarify that there is no unexpected modifications in the processing.
kvrigor commentedon Apr 1, 2026
To @kgoergen: A received variable can never be made constant because
oasis_put/oasis_getAPIs require variables to be of apointertype, notparameter;constant_soil_liq = pointer_soil_liqwould not compile, since by definition a constant is not allowed to change its value during runtime.pointervariable into a "constant array" requires the array contents and dimensions to be pre-defined before the start of the program. A constant can only be set during compile time, not during runtime.I get that the idea is to check if the received eCLM variables are truly "read-only"; however forcing these variables to be constant will not work (with reasons mentioned above). I can find other methods to do this check.
kgoergen commentedon Apr 2, 2026
piece of information by @DCaviedesV https://link.springer.com/article/10.1023/A:1001786023282
kvrigor commentedon Apr 3, 2026
@s-poll could you check the
SMPvalues for the North African region in the eCLM history file? eCLM initializessmp_lto-1000.0 mmon all gridcells:eCLM/src/clm5/biogeophys/SoilStateType.F90
Lines 344 to 347 in 98054f5
If
SMPdoes lie within the-1000range in sandy soils, this could suggest that eCLM-ParFlow is skipping some importantsmpupdate that only happens in a standalone eCLM run.s-poll commentedon Apr 3, 2026
I believe the SMP values (-1m over the whole soil column) are only used during the cold start initialization phase. My understanding is that they are subsequently overwritten by ParFlow during the simulation, which is also what I observe. However, the experiments ParFlow initializing with a hydrostatic patch at 0m and -50m both for the same simulation period as above reveal that the ET values behave in a physical range, when the SMP is still at the order of -(O)**0-3m.
The simulation with hydrostatic patch of -50m (please note that the pressure head in the following is in m, whereas in other post it was mm) shows that at the beginning of the simulation the ET values are in a fine range (figure left). At this point in time of the simulation, the pressure head is in the order of
-10**3m(figure middle-left). At the end of the simulation month already slight excessive ET values can be observed (figure middle-right). Pressure head of uppermost layer is order of-10**4m(figure right).A side remark: I observe point-wise low saturation values in the lowermost model level of ParFlow in the restarted simulation (see figure below). This is probably not connected to this issue, but should be investigated in future.

One of the main questions I still have is why we don’t see a significant impact of SMP on ET when SMP is limited.
s-poll commentedon Apr 7, 2026
Just as an addon: The saturation in eCLM (WATSAT) keeps konstant during the year of simulation. However, when comparing the ParFlow porosity (From indicator file and ParFlow namelist, figure left) with the eclm watsat output (figure center, figure right with adjusted range), one can observe difference between the component models.
kvrigor commentedon Apr 7, 2026
Connection from latent heat flux (
eflx_lh_tot) to pressure (pfl_psi) in source code:eflx_lh_tot)eCLM/src/clm5/biogeophys/SoilFluxesMod.F90
Line 313 in 98054f5
qflx_evap_soi)eCLM/src/clm5/biogeophys/BareGroundFluxesMod.F90
Lines 371 to 372 in 98054f5
dqh)eCLM/src/clm5/biogeophys/BareGroundFluxesMod.F90
Line 261 in 98054f5
qg,qg_soil)eCLM/src/clm5/biogeophys/CanopyTemperatureMod.F90
Lines 314 to 352 in 98054f5
soilalpha,qred,hr,psit,pfl_psi)eCLM/src/clm5/biogeophys/CanopyTemperatureMod.F90
Lines 255 to 273 in 98054f5
s-poll commentedon Apr 7, 2026
Limiting the psit to
-1e5reduces the ET dramatically (see figure below). The limit ofsmpmin(-1e8) shows no effect at all as stated above. The run above with the limitation of10was unfortunately only limitingsmpand still had the limiter of smpmin forpsit. When also settingpsitto -10, the model is crashing with the error message of a nan to the cpl 7.I also did a simulation initializing ParFlow with a hydrostatic patch of
-100000mas discussed in the meeting. It shows higher ET values for the whole domain, but the excessive ET values over the sandy soils at the African continent.kvrigor commentedon Apr 8, 2026
The
smpminsetting in CLM5/eCLM appears to be only valid for the Clapp&Hornberger SWRC. Revisiting the soil alpha formula in the CLM5 Tech Guide,We see that the
1e-8lower limit applies tosmpmultiplied by a C&H-related factor (and notsmpby itself). From the code persepective, @s-poll already discovered the bug inCanopyTemperatureMod.F90:eCLM/src/clm5/biogeophys/CanopyTemperatureMod.F90
Lines 261 to 270 in 98054f5
The fix is to always apply
max(smpmin(c), psit)just beforeexp(...)gets called.The default
smpminsetting also needs to be fixed for eCLM-ParFlow which assumes a Van Genuchten SWRC.-1e5seems to work as per Stefan's experiment above.eCLM/src/clm5/biogeophys/SoilStateInitTimeConstMod.F90
Lines 189 to 191 in 98054f5
To summarize, the usage of
smpmin = -1e8in CLM5/eCLM implicitly assumes a Clapp&Hornberger SWRC.s-poll commentedon Apr 8, 2026
Introducing a threshold/limiter lead to the questions, where to set it.
I plot the soil water retention curve from ParFlow based on the ParFlow namelist
One can see that the asymptotic behavior is already reached at a pressure head of about 10**2/3m. Should we set it so this value?
I am currently not sure about possible interactions with plant hydraulics.
I also plot the soil water retention curves based on eCLM Clopp and Hornberg. Thereby I ignored organic matter, and used the pedotransferfunction values of Cosby 1984 table 5. The pressure head is calculated with Clopp and Hornberg. It needs to be verified, if this is the function which is actually used.
s-poll commentedon Apr 9, 2026
Here are the ET values between the simulations. I suggest limiting smp to
10**3 m, consistent with the approach used in this commit. Is everyone okay with applying this limit, or are there any concerns?left:
smpmin=10**2m, center:smpmin=10**3m, right: differenceleft:
smpmin=10**3m, center:smpmin=10**4m, right: differenceIn the plant hydraulics there is also the pressure head of the soil involved. This is currently calculated in a standalone and parflow coupled simulation based on the soilliq water content and based on Clopp and Hornberger (see code below). Thus, the limiter does not seem to affect the plant hydraulics, however, there is a inconsistency between the soil pressure values within the code!
eCLM/src/clm5/biogeophys/HydrologyNoDrainageMod.F90
Lines 460 to 468 in 98054f5
[-]Excessive ET in eCLM–ParFlow simulation over sandy soil[/-][+]eCLM-ParFlow: Excessively negative ET over sandy soil[/+]kvrigor commentedon May 27, 2026
As long as this SMP limit is guaranteed to apply only in eCLM-ParFlow, I'm ok with it. A more precise value could be argued upon or derived via further experiments. But so far as the negative ET issue is concerned, finding a suitable soil matric potential limit that is consistent with eCLM-ParFlow assumptions is the fix.
DCaviedesV commentedon May 27, 2026
Would it make sense to expose this setting in the namelist?
kgoergen commentedon May 27, 2026
Maybe add it as a tuning constant? Documentation in the eCLM and TSMP2 manuals alike then?
kvrigor commentedon May 27, 2026
I think no, as making the limiter setting flexible could potentially lead to the problem of subtly different TSMP2 simulations, caused by accidentally setting the wrong limiter value. Perhaps its better to prevent these future headaches by committing to a single value :).
kvrigor commentedon May 27, 2026
Posting a summary of the issue here as suggested by @jjokella . Slides are from DETECT JF last 21-May.
DCaviedesV commentedon May 28, 2026
I understand this risk, sure. I would argue nevertheless that this risk exists for absolutely all parameters exposed in the namelists, which, I would argue can be properly handled by version tracking experiments (e.g., workflow engine et al).
On the other hand, not exposing this means this is a silent knob that has been turned and locked without users being aware (yes, it is documented here, but a user will never look in here if they don't know they need to). Moreover, I would assume that the optimal super-best value for this parameter might still change, or might have some cases in which a different setting would be preferred. That is hard to track, especially if a user touches this constant in-code, runs an experiment, and, e.g., never ever commits that - it becomes non-traceable and non-reproducible. This is a real potential pitfall.
I suppose a good way to deal with it would be that you could use the namelist setting also with eCLM standalone, but that in that case the setting should be ignored (as the issue only arises when coupling).
I think the opinion of those who curate experiments (@kgoergen , @s-poll, others?) is most relevant here, as, at least how I see it, it is a mater of traceability and reproducibility.
kvrigor commentedon May 28, 2026
I'm assuming that the SMP limiter is like a "physical constant" that can be derived analytically from a given SWRC. Looking at Stefan's plots here, the functions are smooth and has a closed form, and thus it's possible to compute the 2nd derivative of the pressure head wrt to the relative saturation (2nd derivative since we want to capture the abrupt change of slope in a given SWRC). If the 2nd derivative is "too high" at a given relative saturation, both of which could be precisely defined, then that should be the SMP limit.
In general, I'm in for flexibility, traceability, and reproducibility; what I'm guarding against is pushing unnecessary empirical burden to users especially when a correct value can be known in advance. I'm not familiar with SWRC and its typical assumptions so my analysis can be flawed. We do need to have the experimentalists weigh in on this.
Fix excessively negative ET over sandy soil in eCLM-ParFlow coupling (#…