Disclaimer: this issue is for internal discussion only.
Last Updated: 2026-09-24 12:00
Goal for the next version
All test cases run with TSMP2 WFE (EUR-12 domain and beyond).
- ICON-eCLM: resolve the winter cold bias.
- eCLM-ParFlow: resolve the soil-moisture difference and the reduced LHF bias.
- eCLM sta: clarify summer wet bias
1. ICON-eCLM — winter cold bias
After all 1.2 code changes, the winter cold bias seems to be solved.
1.1 Where the code is
Build tree / work directories:
- Source + build:
/p/scratch/cslts/poll1/sim/testcases/TSMP2_flxcpl/models/{icon,eCLM}
- WFE:
/p/scratch/cslts/poll1/sim/paper/wfe_eur-11_revsetup_icon-eclm_excoef
1.2 Code changes going into the new version
- Exchange albedo and downward shortwave radiation for both radiation bands (closed a 26–28 % surface shortwave inconsistency to −0.7 %; this was the main driver of the winter cold bias).
- Send topography (
ICOTOPOG) as an additional coupling field to eCLM (topographic downscaling of the atmospheric forcing).
- Convert momentum flux to use eCLM's drag:
tvm derived from the coupled TAUX/TAUY, no new coupling fields. Momentum is now dynamical: ~+14 % near-surface wind, decaying to zero by mid-troposphere.
- ( to be decided:) couple the exchange coefficient for heat into
TKE_ke1. tvh already reaches the TKE through the circulation term (+46 %/+65 % surface TKE by day 7 in calm conditions).
- Add a new exchange-coefficient-based coupling scheme, selectable via a new namelist switch (
loascplscheme_exchcoef)
- Robustness fixes:
- only overwrite
oas_rcv_field on a successful oasis_get
- default of oasis_mask false
- preserve oasis-coupled surface fluxes through
turbtran (incl. the shfl_s retrieval fix and the missing j-loop over the oasis temperature in mo_nwp_sfc_interface.f90);
- correct radiation fluxes for cold starts. The OASIS init exchange preceded the only radiation call; cold-start-only, restart path unchanged.
- Modularisation: the coupled surface block moved out of
mo_nwp_sfc_interface.f90 into oasis_apply_surface in mo_oasis_land_coupling.f90; invalid oasis values reported via oasis_report_invalid.
- DEBUG build:
cmake/BuildICON.cmake DEBUG branch was fixed to actually set -O0 and enable bounds/FP/NaN checks.
- Remove ncoup from turbtrans, as it disconnects the coupling of heat fluxes from ocean to atmosphere and now is not needed anymore due to lcoup
1.4 Configuration
- SST is using lowest model level instead of dedicated LBC
- use SPICE v2.4 config
1.5 Long-term tests to run
| # |
Code |
Setup |
| 1 |
old code |
old setup |
| 2 |
new code |
old setup |
| 3 |
new code (without TKE) |
old setup |
| 4 |
new code |
SST correction |
| 5 |
new code (without TKE) |
SST correction |
- SST correction seems to worsen RMSE
- Only small difference between new code and new code (without TKE) for T2M
- New code improve RMSE about 2K. Still winter cold bias visible.
Results
- in the old setup the SST is taken from initialization (Jan)
- with TKE coupling overall boundary layer state seems to be improved, t2m/q2m RSME increased
1.6 Long-term tests II
| # |
Code |
Setup |
| 0 |
ICON sta |
recommended CLMcom |
| 1 |
new code oce |
recommended CLMcom |
| 2 |
new code oce (resistance coupling) |
recommended CLMcom |
| 3 |
new code oce (without TKE) |
recommended CLMcom |
1 ICON-eCLM
Massive improvements compared to cplv2.01. Surface energy fluxes are better against GLEAM. 2m temperature is about 50% better compared to ICON sta (in ME most improvement). 2m humidity decrease in skill compared to eobs.
Without upper-boundary nudging
The land-scheme signature does not depend on upper-boundary nudging. With spectral nudging switched off in both ICON sta and ICON-eCLM, every mean ICON - ICON-eCLM gap stays within ~20 % of its nudged value. The eCLM Bowen ratio is unchanged in SON and only slightly lower in DJF. Only TERRA seems to be more sensitive in summer time.
2. eCLM-ParFlow: soil-moisture difference, reduced LHF bias
Porosity synchronization is already included.
| Configuration |
Repository |
Branch |
| WFE |
https://github.com/HPSCTerrSys/TSMP2_workflow-engine/ |
master |
| Forcing |
/p/scratch/cslts/poll1/sim/paper/wfe_eur-11_revsetup_eclm-pfl_v2/dta/forcing |
|
| Geo |
/p/data1/slts/poll1/sim/paper/wfe_eur-11_revsetup_icon-eclm-pfl/dta/geo |
|
| NML |
/p/scratch/cslts/poll1/sim/testcases/wfe_eur-11_revsetup_eclm-pfl_swrc/nml |
|
2.1 Where the code is
Work directory: ``/p/scratch/cslts/poll1/sim/testcases/wfe_eur-11_revsetup_eclm-pfl_swrc`
2.2 State of the diagnosis
| # |
Effect |
Size |
Status |
| 1 |
Soil ice never debited when ParFlow overwrites the liquid leads to unbounded water creation |
dominant; ~1.6 mm/day domain-wide, up to 619 mm in a column over 7 days |
fixed, verified |
| 2 |
eCLM and ParFlow are on genuinely different grids; every exchange is a 4-neighbour IDW interpolation |
RMSE ≈ 0.03 in S; irreducible without regridding |
inherent to the setup |
| 3 |
WATSAT (eCLM) ≠ ParFlow porosity at the same point, because of (2) |
bias −0.0044 in θ, RMSE 0.029 at depth; 2.4 % of points off by >0.05 |
mitigate in post-processing |
2.3 Open issues (eCLM-ParFlow)
- soil moisture do not match between eCLM and ParFlow
- reduced LHF bias (fixed by eCLM PR#129)
2.4 Configuration
2.5 Bugfixes
2.6 Feature Implementation
2.7 Backlog
- runoff / overland flow treatment
- Bedrock depth treatment
- Coupling depth
nlevgrnd (25) vs nlevsoi (20): below layer 20 the eCLM and ParFlow layer thicknesses differ, SPo proposal: couple nlevsoi only.
- Rosetta n values +1
- SWRC Clapp and Hornberger vs van Genuchten. prototype in branch
dev-vangenuchten-swrc
3. eCLM standalone
The bias is probably connected to a faulty eCLM-ParFlow spinup.
4. Fully coupled ICON-eCLM-ParFlow
long term test of fully coupled including bugfixes 2.5 and new code change 1.2. with a 5a eCLM-ParFlow spinup.
First results with 5a spin-up
Good Thing: the dry bias improves (q2m) with coupling ParFlow, which also feeds back positively to cloud and shortwave radiation
Bad Thing: The transpiration is still unnatural low, whereas the evaporation is unnatural high. The reason for the high evaporation seem to be large areas of ponding water in the domain.The surface energy fluxes (ET/H) looses in skill compared to the observations.
Summary: ParFlow improves moisture state, but for the wrong reasons.
Initial state or ParFlow? (ICON-eCLM from the 5a eCLM-ParFlow restart)
ICON-eCLM was started from the identical eCLM restart as the fully coupled run.
- The initial state has no effect on the atmosphere: 4 % of the RH improvement, ~0 % of LE, cloud and SWdown.
- eCLM (free drainage) empties the top metre within ~1 month. The deep store (1–8.6 m) stays wet all year but never reaches the roots.
- Every ParFlow signal seems to come from ParFlow running, not from the spin-up.
Why the ground evaporation is high: two causes
- Ponding / shallow water table (hydrology): The top layer is near saturation (> 0.8) on 24 % of samples vs 3 % without ParFlow. At the same point, hour and top-layer wetness above ~0.25 saturation, both runs evaporate the same amount (within ~10 %).
- Dry-soil evaporation (code). Check
smpmin = -1e6 mm clamps the soil suction seen by the ground humidity and move to eCLM defaults -1e8 mm? (add experiment)
30a spin-up (running)
Nearly have of spin-up done. The surface converged by ~1993: ~39 % of columns are ponded, the median water table is at 0.8 m, and both did not change over the past ~5 a. The deep store is still draining.
5. Finished Tasks
6. Current Tasks
Goal for the next version
All test cases run with TSMP2 WFE (EUR-12 domain and beyond).
1. ICON-eCLM — winter cold bias
After all 1.2 code changes, the winter cold bias seems to be solved.
1.1 Where the code is
oas-coup-dev-raddev-iconeclm-exchcoefBuild tree / work directories:
/p/scratch/cslts/poll1/sim/testcases/TSMP2_flxcpl/models/{icon,eCLM}/p/scratch/cslts/poll1/sim/paper/wfe_eur-11_revsetup_icon-eclm_excoef1.2 Code changes going into the new version
ICOTOPOG) as an additional coupling field to eCLM (topographic downscaling of the atmospheric forcing).tvmderived from the coupledTAUX/TAUY, no new coupling fields. Momentum is now dynamical: ~+14 % near-surface wind, decaying to zero by mid-troposphere.TKE_ke1.tvhalready reaches the TKE through the circulation term (+46 %/+65 % surface TKE by day 7 in calm conditions).loascplscheme_exchcoef)oas_rcv_fieldon a successfuloasis_getturbtran(incl. theshfl_sretrieval fix and the missingj-loop over the oasis temperature inmo_nwp_sfc_interface.f90);mo_nwp_sfc_interface.f90intooasis_apply_surfaceinmo_oasis_land_coupling.f90; invalid oasis values reported viaoasis_report_invalid.cmake/BuildICON.cmakeDEBUG branch was fixed to actually set-O0and enable bounds/FP/NaN checks.1.4 Configuration
1.5 Long-term tests to run
Results
1.6 Long-term tests II
1 ICON-eCLM
Massive improvements compared to cplv2.01. Surface energy fluxes are better against GLEAM. 2m temperature is about 50% better compared to ICON sta (in ME most improvement). 2m humidity decrease in skill compared to eobs.
Without upper-boundary nudging
The land-scheme signature does not depend on upper-boundary nudging. With spectral nudging switched off in both ICON sta and ICON-eCLM, every mean ICON - ICON-eCLM gap stays within ~20 % of its nudged value. The eCLM Bowen ratio is unchanged in SON and only slightly lower in DJF. Only TERRA seems to be more sensitive in summer time.
2. eCLM-ParFlow: soil-moisture difference, reduced LHF bias
Porosity synchronization is already included.
2.1 Where the code is
dev-eclm-parflow-smWork directory: ``/p/scratch/cslts/poll1/sim/testcases/wfe_eur-11_revsetup_eclm-pfl_swrc`
2.2 State of the diagnosis
WATSAT(eCLM) ≠ ParFlow porosity at the same point, because of (2)2.3 Open issues (eCLM-ParFlow)
2.4 Configuration
2.5 Bugfixes
2.6 Feature Implementation
2.7 Backlog
nlevgrnd(25) vsnlevsoi(20): below layer 20 the eCLM and ParFlow layer thicknesses differ, SPo proposal: couplenlevsoionly.dev-vangenuchten-swrc3. eCLM standalone
The bias is probably connected to a faulty eCLM-ParFlow spinup.
4. Fully coupled ICON-eCLM-ParFlow
long term test of fully coupled including bugfixes 2.5 and new code change 1.2. with a 5a eCLM-ParFlow spinup.
First results with 5a spin-up
Good Thing: the dry bias improves (q2m) with coupling ParFlow, which also feeds back positively to cloud and shortwave radiation
Bad Thing: The transpiration is still unnatural low, whereas the evaporation is unnatural high. The reason for the high evaporation seem to be large areas of ponding water in the domain.The surface energy fluxes (ET/H) looses in skill compared to the observations.
Summary: ParFlow improves moisture state, but for the wrong reasons.
Initial state or ParFlow? (ICON-eCLM from the 5a eCLM-ParFlow restart)
ICON-eCLM was started from the identical eCLM restart as the fully coupled run.
Why the ground evaporation is high: two causes
smpmin = -1e6 mmclamps the soil suction seen by the ground humidity and move to eCLM defaults-1e8 mm? (add experiment)30a spin-up (running)
Nearly have of spin-up done. The surface converged by ~1993: ~39 % of columns are ponded, the median water table is at 0.8 m, and both did not change over the past ~5 a. The deep store is still draining.
5. Finished Tasks
6. Current Tasks