eCLM-ParFlow: Soil moisture disagreements in real case Parflow-coupled runs

[s-poll]

Summary

TSMP2 EUR-12, ICON-CLMcom forcing, 60yr spinup, sim year 2018 (TSMP2_workflow-engine setup). Every configuration coupling ParFlow shows a large, domain-wide LHF deficit vs. ParFlow-free configs. Root cause traces to eCLM and ParFlow disagreeing on their own shared soil saturation state within the same coupled run, which was already a finding in #103, now shown to drive a real energy-balance bias.

Details

Every ParFlow-coupled config underperforms its ParFlow-free counterpart by ~30-50 W/m² at peak season. Deficit is spatially even across the whole domain; not only local artifacts. This leads to a different flux partitioning and to larger RMSE against GLEAM and has negative feedback especially in the fully coupled simulation.

Domain average (EUR CORDEX focus domain) latent heat flux.

eCLM's own state shifts when coupled to ParFlow. eCLM's relative saturation (level 3) differs between standalone eCLM/ICON-eCLM and their ParFlow-coupled counterparts, especially spring dry-down/summer. Thus, coupling changes eCLM's hydrological state, not just the exchanged flux (not shown).

A deeper analysis revealed that within one coupled eCLM-ParFlow run, the two component models disagree. Same eclm-pfl experiment, level 3: eCLM's H2OSOI/WATSAT vs. ParFlow's own Pressure-derived saturation (van Genuchten) shows.

Relative saturation from eCLM and ParFlow (level three from the top of the soil) from the eCLM-ParFlow coupled Simulation.

eCLM saturation in simulation coupled with ParFlow locally exceeds 100%, up to ~20x nominal at specific points (especially in winter time), leading to averages above 100%.

Yearly average from eCLM-ParFlow simulation.

Switching the coupled exchange variable from liquid water content to saturation directly would likely eliminate the most visible symptom reported in this issue. However, this would only mask rather than resolve the underlying problem.

Data and analysis scripts

Data: /p/scratch/cslts/poll1/sim/paper/wfe_eur-11_revsetup_eclm-pfl/dta/simres
Script: eurcordex_soilmoisture_eclm_vs_parflow.R

[s-poll]

The origin of the relative soil water saturation discrepancy between model components appears to be differing porosity values. Even though the static fields share the same database (SoilGrids), processing differs: eCLM calculates porosity based on sand/clay/organic matter content, whereas ParFlow uses Rosetta values. This leads to strong differences, especially in peatland areas.

Porosity values for ParFlow (top) and eCLM (bottom) for the uppermost model level.

Force synchronization between model components by adopting eCLM porosity values in ParFlow. While eCLM values can theoretically vary over time, organic matter remains constant in SP mode (verified in 2018 runs). I created a file processing eCLM porosity to ParFlow with the mean of claude (parflow static file generator)

Additional Points:

Deep Soil Indicator Default: ParFlow soil indicators > 13 default to a fixed porosity of 0.4. This effect all deeper soils indicator classes. Is this intended, or was the parameterization forgotten?

eCLM over-saturation: Porosity differences explain the relative saturation gap, but do not explain why eCLM values exceed 100%.

Redundant Conversion in Coupling: The liquid water content coupling seems arbitrary. ParFlow calculates liquid water content from saturation using its own porosity before sending it to eCLM, where porosity otherwise seems to be used mainly for diagnostics (volumetric soil moisture calculation). Not checked in depth for ParFlow, but coupling the saturation seems to be more intuitive for me.

[kvrigor]

The origin of the relative soil water saturation discrepancy between model components appears to be differing porosity values.

Not surprising given that ParFlow and eCLM independently define soil texture characteristics. Should we start modifying eCLM and ParFlow towards a single source of truth for soil texture specification? We can let ParFlow inherit the soil characteristics defined in the eCLM surface data file.

[s-poll]

EDIT: Update. In the previous version of this post there was a misconfiguration with eCLM restart file in forced porosity sycned simulations. This has been fixed. The overall message did not change.

The forced synchronization of porosity (with the procedure/script posted in my last post) has clearly led to some improvements in respect to the soil moisture.

There also still seems to be an issue, especially during winter, as already pointed out in my previous comment. I will open a separate issue to track this independently.

The new results look much better with respect to the soil moisture, even though there is still a discrepancy between eCLM and ParFlow's soil moisture state. The latent heat flux in the eCLM-ParFlow simulation improved with forced porosity synchronization, however, is still lower compared to eCLM, leading to higher RMSE.