Hillslope roughly doubles the median NSE (0.106 to 0.219) and raises median KGE from 0.251 to 0.348 and median correlation from 0.642 to 0.712. Details: Results (Configuration sensitivities).
Case card¶
| Item | Value |
|---|---|
| Baseline | see the contrast table in Results (Configuration sensitivities) |
| Configuration | Common setup (Configuration summary, and the PE layout row of the shared setup table) |
| Setup scripts | setup_hillslope_spinup.sh and setup_hillslope_eval.sh, see Code and data (Scripts) |
| Spinup case | hillslope_on |
| Spinup start | 1980-10-01 |
| Spinup length | 24 yr in 6-yr chunks (STOP_N=6, REST_N=2) |
| Spinup final restart | 2004-10-01 |
| Evaluation run | hillslope_eval |
| RUN_REFCASE / RUN_REFDATE | hillslope_on / 2004-10-01 |
| CTSM finidat | hillslope_on.clm2.r.2004-10-01 |
| mizuRoute fname_state_in | hillslope_on.mizuroute.r.2004-10-01-00000.nc |
| Extra output | QLATFLOWOUT |
Spinup¶
Adequacy checks
Hillslope, last 10 years: TWS, TOTSOILLIQ and TOTSOILICE drift at 1.0-1.3 times their interannual noise; snow and water table depth converged. The residual soil-moisture drift was accepted.
Restart boundaries: a TOTSOILICE jump at the 1998-10 hillslope segment boundary (+0.56) is far below the median month-to-month change (13.7); it is the freeze-thaw cycle, not a restart artifact.
Evaluation run¶
hillslope_eval: a separate case from hillslope_on, with the restart set copied in, so the spinup case stays untouched; run as STOP_N=6, REST_N=6, RESUBMIT=6, with NTASKS_ROF=-14 (1,792 tasks) (see Issues and fixes).
Issues and fixes¶
Note: I forgot to change NTASKS_ROF to -5 for hillslope case. After spinup run, I tried to change it but as mizuRoute restarts record the task count, so hillslope_eval failed to read the hillslope_on restart with NTASKS_ROF=-5 and was rerun with -14 , used the same mizuRoute task count as its spinup.
The hillslope spinup lost ~6-7 simulated years because REST_N was left unset and defaulted to STOP_N (see Lessons, caveats and future work).
Hillslope with use_init_interp triggers a build-namelist warning that surfaces as a Perl error; add -ignore_warnings to CLM_BLDNML_OPTS.
QLATFLOWOUT exists only with use_hillslope=.true.. Requesting it in other configurations fails.