Fix mEVP/aEVP: zero stress tensor before thin-ice cycle to prevent stale stress application

[patrickscholz]

Summary

Affects both mEVP and aEVP rheology variants. When an element's ice mass falls below the skip threshold, the stress tensor update is skipped but the old stress values are left in memory. A mismatched skip condition in stress2rhs_m then applies those stale stresses to the momentum equation, spuriously accelerating nearly-vanished ice. The fix explicitly zeros the stress components before cycle-ing.

---

The mismatch

The stress update and stress application use different skip conditions:

! stress_tensor_m / stress_tensor_a — stress UPDATE (skips on low MASS)
msum = sum(m_ice(elnodes)) * val3
if (msum <= 0.01) cycle        ! sigma11/22/12 keep old values

! stress2rhs_m — stress APPLICATION (skips on low CONCENTRATION)
if (sum(a_ice(elnodes)) < 0.01) cycle

Ice mass and concentration do not go to zero simultaneously. During summer melt, an element can have:

• Very low mass (msum <= 0.01 kg/m²) — stress update is skipped
• Moderate concentration (sum(a_ice) > 0.01) — stress application is not skipped

The result: stress2rhs_m picks up sigma11, sigma22, sigma12 from a previous timestep when the ice was thicker, and applies them to the momentum RHS. The nearly-vanished ice is driven by stresses that no longer reflect its physical state.

---

Impact by EVP variant

mEVP (stress_tensor_m): alpha_evp is fixed, so stale stresses decay geometrically over sub-cycling iterations (det1 * sigma each step). The bug is present but self-damps over the EVP sub-cycle. Impact is localised and most visible during rapid summer decay.

aEVP (stress_tensor_a): alpha is recomputed each sub-cycle from the current pressure, which itself depends on ice conditions. As ice thins, alpha changes, making the stale stresses increasingly inconsistent with the current state. The bug is more persistent in aEVP because the relaxation rate is not fixed.

Both variants are most affected during summer ice decay — thin, patchy ice where the mass and concentration thresholds diverge most strongly.

---

The fix

Zero sigma11, sigma22, sigma12 explicitly before cycle-ing in both subroutines (src/ice_maEVP.F90):

! Before (wrong — stale stresses survive the skip)
if (msum <= 0.01_WP) cycle

! After (correct — stress zeroed before skipping)
if (msum <= 0.01_WP) then
    sigma11(elem) = 0.0_WP
    sigma22(elem) = 0.0_WP
    sigma12(elem) = 0.0_WP
    cycle
end if

Applied identically to stress_tensor_m (mEVP, line 141) and stress_tensor_a (aEVP, line 1064). stress2rhs_m then contributes zero momentum tendency for those elements regardless of which concentration-based skip it uses — physically correct for nearly-vanished ice.

Scope

• Affects all runs using mEVP or aEVP rheology (i.e. essentially all standard FESOM2 configurations)
• No namelist changes required
• Effect is most visible in summer Arctic/Antarctic marginal ice zones during decay season

[patrickscholz]

i will make a test run to this !

[patrickscholz]

Done for aEVP:

• Difference in Summer/Winter NH/SH sea ice concentration:

• maximum NH/SH MLD2

• temperature/salinity biaS

Fazit:
this bux fix has very minor effect !!!