Correct v14 BCs settings for GWD

[mathomp4]

Currently in v14 testing (see #767), a v14 run will have in AGCM.rc:

# TRB/GWD Parameterization
# ------------------------
NCAR_NRDG: 16

That is set to 16 if a gwd_internal_rst is found. And that is the only qualification.

However, in a discussion about v14 testing with @biljanaorescanin she said:

Here’s the background on what changed and why the fields differ so we can make decision:

The run logs (slurm-*.out) show that we properly load gwd_internal_rst with v14 BCs option and bootstrap EFFRDG/KWVRDG, which activates the NCAR ridge orographic GWD scheme.
Both the old and new 1 day experiments ran with:

NCAR_NRDG: 16   
NCAR_EFFGWORO: 1.0   (default)
NCAR_ORO_TNDMAX:400.000000, (default value)

so the configuration didn’t change but the ridge content in in gwd_internal_rst did!!!

The new restart contains a richer ridge spectrum (more populated bands and larger (HWDTH × CLNGT)/GBXAR_km² footprints), which increases the aggregate orographic stress:

EFFRDGn= NCAR_EFFGWORO * (HWDTH *CLNGT) / GBXAR_km²

With all 16 bands active, the summed forcing is stronger than in the older ridge file (where effectively only ~6 bands carried energy). This leads to larger 1-day differences in surface stress, winds, and fluxes but the timing (Run: ~2.6–3.5 s) is unchanged, so the effect is physical magnitude, not computational instability.

It’s not a code path change; it’s a forcing content change that simply needs a small retune of the orographic efficiency.

Right now both experiments use an unrealistic “maximum” configuration (EFFGWORO = 1.0 for 16 bands).

Recommended options:

1. Reduce efficiency:
   NCAR_EFFGWORO = 0.40
   or keep 16 bands and set NCAR_EFFGWORO = 0.35 with a soft cap
   sum(EFFRDG) ≤ 0.30 per grid cell.
2. Alternative: limit the active ridge count
   NCAR_NRDG = 8 and keep EFFGWORO = 1.0.

Either approach should bring the 1-day flux and wind differences back into a reasonable range while preserving the ridge-scheme benefits.

One option is we could run few short tuning options like 1 days runs with different parameters and after retuning we can check: global mean TOA net (SWTNET – OLR) and surface net (SWGNET + FLNS – SFCEM) drifts—target ≤ 1–2 W m⁻². But we also need to map TAUGWX/TAUGWY, USTAR, and CM to confirm the changes localize over orography.

She also provided an example test configuration as a possible v14 starting point:

# TRB/GWD Parameterization
# ------------------------
NCAR_NRDG: 16
NCAR_EFFGWORO: 0.35         # <— new: reduce orographic efficiency
NCAR_ORO_TNDMAX: 250.0      # <— optional: soften tendency cap (was 400 day⁻¹ before /86400)

[biljanaorescanin]

Validation summary — ridge GWD retune

Configs compared (10-day AMIP short tests):

RunID	Topography in gwd_internal_rst	NCAR_NRDG	NCAR_EFFGWORO	NCAR_ORO_TNDMAX	Notes
BASE	Old 6-band	16	1.0	400	current production
EXP1	New 16-band	16	0.35	250	recommended retune
EXP2	New 16-band	16	0.30	250	softer drag sensitivity
EXP3	New 16-band	16	1.0	400	no retune (isolates topo jump)

Summary of the issue with orographic GW source update:
The GWD restart (gwd_internal_rst) was upgraded from legacy 6-band orography to a 16-band ridge spectrum (new topography). In the NCAR ridge scheme the injected orographic stress scales roughly with

$\displaystyle \sum_{n} \frac{HWDTH_n \times CLNGT_n}{GBXAR}$

With 16 bands this sum increases over complex terrain, so default NCAR orographic efficiency = 1.0 makes near surface GW drag too strong vs. the old system. To restore budgets and surface stress to pre change levels, we reduced the orographic efficiency and slightly softened the orographic tendency cap in retune.

Global energy (EXP3 – BASE)
Mean and rms of daily domain means over each 10-day pair:

Res	ΔTOA_net [W m⁻²] (rms)	ΔSFC_net [W m⁻²] (rms)
C90	+0.01 (0.12)	+0.27 (0.65)
C180	−0.18 (0.22)	−0.09 (0.14)

Global energy (EXP1 – BASE)
Mean and rms of daily domain means over each 10-day pair:

Res	ΔTOA_net [W m⁻²] (rms)	ΔSFC_net [W m⁻²] (rms)
C48	−0.34 (0.52)	−0.17 (0.29)
C90	−0.04 (0.13)	+0.15 (0.37)
C180	−0.15 (0.18)	−0.04 (0.16)
C360	−0.13 (0.17)	−0.12 (0.25)

=> The unretuned 16-band configuration (EXP3) disturbs the short term energy balance more strongly than the retuned case (EXP1). At C90 it produces warming at the surface and TOA imbalance, while at C180 it produces a cooler response — both consistent with excessive or poorly distributed orographic drag.

=> across resolutions, the retune leaves the global radiative balance essentially unchanged (|mean| ≲ 0.4 W m⁻²; rms O(0.2–0.5) W m⁻² for 10-day variability).

Surface/BL metrics (land means; EXP1 – BASE)
Mean and rms of daily domain means over each 10-day pair:

Res	ΔUSTAR [m s⁻¹]	ΔCM [–]	ΔPBLH [m]
C48	+1.19×10⁻³	−6.9×10⁻⁵	+2.46
C90	−3.27×10⁻⁴	−2.5×10⁻⁴	−4.14
C180	+5.85×10⁻³	+6.0×10⁻⁴	+8.54
C360	+4.37×10⁻³	+4.6×10⁻⁴	+2.20

Flux context (land means):
SHFX: changes ~ −0.10 to −0.31 W m⁻² (rms O(0.2–0.6)).
LHFX: changes ~ −0.40 to +0.12 W m⁻² (rms O(0.2–0.6)).

=> surface stress/drag and BL structure respond modestly and consistently; land mean shifts are small (USTAR ~10⁻³–10⁻² m s⁻¹; PBLH few meters to ~9 m), with compensating SH/LH adjustments well below 1 W m⁻² in the mean.

Orographic focus (mountain mask; EXP1 – BASE)

USTAR_mtn and CM_mtn: O(10⁻³–10⁻⁴) changes.
PBLH_mtn: a few meters to O(10 m) depending on flow regime/resolution.
Max-abs TAU* over land/mountains nudges by O(10⁻2–10⁻1 N m⁻²), not systematic in sign.

=> responses are localized over terrain and remain small in magnitude.

Softer option (EXP2 – EXP1, C180)

ΔTOA_net = −0.03 W m⁻²; ΔSFC_net = −0.02 W m⁻².

Land means are near zero (e.g., ΔUSTAR ≈ +3.5×10⁻5 m s⁻¹, ΔPBLH ≈ −0.64 m).

=> NCAR_EFFGWORO=0.30 is an even more conservative retune! ; we can use it if we decide to further minimize perturbations.

NOTE: Diagnostic issue noticed in code: in GEOS_GwdGridComp.F90
TAUGWX and TAUGWY were added to the GWD History collection for these tests. They appear as exports in the code, but are zero in the current builds due to incomplete History wiring.
This is purely a diagnostic issue — it does not affect model physics or the GWD computation, since the stresses are already applied internally through TAUXO/TAUY* and DUDT/DVDT/DTDT.
A small History export fix will allow future runs to record nonzero TAUGWX/TAUGWY values for validation, but the physical behavior of these runs is correct.

I was adding TAUGWX and TAUGWY to HISCTORY.rc geosgcm_gwd collection but that is really 3-D collection... I added because I've seen other 2D vars there. But all 3 of those should be moved to surf which is 2-D.

geosgcm_gwd.fields:
'TAUOROX;TAUOROY' , 'GWD' ,
'TAUBKGX;TAUBKGY' , 'GWD' ,
'TAUGWX;TAUGWY' , 'GWD' ,

Conclusion

The 16-band ridge scheme + retune is numerically stable and physically balanced across C48, C90, C180, C360 on 10-day tests.
Global energy is stable; surface/BL impacts are small and localized; no signs of drift.
Either EXP1 (NCAR_EFFGWORO=0.35, NCAR_ORO_TNDMAX=250) or the softer EXP2 (0.30) are viable; recommend EXP1 as default and keep EXP2 as a conservative fallback.

Based on @mathomp4 suggestion, I suppose we want to add this as an commented option to AGCM.rc and @michellefrazer can just uncomment those in runs? @sdrabenh what do you think?

Effect of Retune vs. No-Retune

In the C180 case it is more visible, need to retune, the unretuned case drifts negative (excess outgoing energy) while the retuned one again stays close to balance.

Retuned case just gives more stable curve across all daily deltas over different test resolutions.

[mathomp4]

I've added the commented settings recommended by @biljanaorescanin for testing in 08527bc:

# The below settings are only for v14 BCs currently being tested
#v14 NCAR_EFFGWORO: 0.35
#v14 NCAR_ORO_TNDMAX: 250

So @michellefrazer or @sdrabenh can just remove the #v14 for now.