Excessive ET in eCLM–ParFlow simulation over sandy soil #105
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.