rdb_ocean_cavity_flux Module

Composed solve mask (0/1): cover_frac AND wet AND “the far-field sample found mass”. This is what the kernel’s cover argument is given.

Top drag coefficient for the MELT friction velocity only.

Per-cell top-layer thickness RATIO h_new/h_old left behind by the volume_compensation sink, exactly 1 everywhere the sink did not act. Every PASSIVE tracer’s top-layer load is multiplied by it (cavity_comp_scale_tracer_impl) so the removed parcel carries that cell’s own concentration and changes none of them. Allocated with the rest of the slot (full size iff enable); left at 1 when volume_comp = CAVITY_VC_NONE.

Thickness (m) the compensation sink removed from each open -ocean column this thermo step; zero when compensation is off.

Thermodynamic + turbulence constants. Flat POD, left at the ISOMIP+ protocol values (Asay-Davis et al. (2016) Table 4): rho_w, c_w, alpha_T, beta_S here are the MELT LAW’s own calibrated constants, NOT copies of the model’s Boussinesq rho_0 — the same “out of scope on purpose” category as RHO_WATER and &ocean_ice_nml rho_ocean in src/core/ocean/README.md’s reference-density table. Overriding them would break ISOMIP+ comparability, which is the reason this path exists.

&ocean_cavity_melt_nml enable, latched before init.

Cell-centred Coriolis parameter (1/s), from the same fill_coriolis_centre every other slot uses. Read by hj99 only, as |f|; the law does not exist at f = 0.

Thickness (m) below the ice base the far field is averaged over. See the module docstring: metres, not layers.

&ocean_cavity_melt_nml freshwater, parsed. CAVITY_FW_VIRTUAL (default) ⇒ ocean_cavity_mass_step is an immediate return and the path is byte-identical.

Haline exchange velocity (m/s), same convention.

Thermal exchange velocity (m/s) the column’s solve converged on — NOT par%gamma_t_coeff, which is the dimensionless coefficient. Under hj99 / yung25 it is an implicit function of the converged interface state, so it is stored rather than re-derived: a diagnostic that rebuilt it from u* alone would report the NEUTRAL value instead of the stratification-suppressed one. Zero where inactive.

Ice-conduction selector + T_ice. Flat POD.

Set by init after every allocation succeeds; cleared by destroy first. Always test this, never allocated(...).

Canonical melt mass flux (kg/m^2/s), > 0 melting.

Domain-integrated meltwater VOLUME (m^3) added this thermo step over the INTERIOR (ghosts excluded, globally combined). Diagnostic + the input the compensation sink spends.

Active columns returning NO_PHYSICAL_ROOT this step.

Running total of NO_PHYSICAL_ROOT.

Active columns returning NONFINITE_INPUT/NONFINITE_STATE this step. FATAL — the driver fails loud.

Active columns returning NOT_CONVERGED this step (zero melt applied there — the kernel’s documented safe state).

Running total, drained to the console like continuity_t%n_limited_total. int64 because a long run with one stubborn column can exceed 2e9.

Active columns returning any other non-OK status (BAD_INPUT, NO_CORIOLIS, LAW_DOMAIN, …).

Running total of every other non-OK, non-fatal status.

Columns this thermo step whose top layer could NOT give up the thickness the real-mass path asked of it without falling through H_CAVITY_FLOOR — a FREEZING column (m < 0) whose top layer is thinner than |m|*dt/rho_0, or a compensation sink deeper than the open-ocean top layer. The withdrawal is CLAMPED to what is there ABOVE the floor (so h stays strictly above the vanish marker, never on or below it) and the count is FATAL: a clamped withdrawal is a withdrawal the tracked mass source no longer matches, so continuing would print a budget that silently stopped closing. Same stance as the non-finite melt column.

Running total of n_thin_step, for the record.

Interior wet area (m^2) the ice does NOT cover, globally combined — the denominator of the uniform sink.

Exchange-law selector + coefficients. Flat POD; its f_cor member is unused here (the driver substitutes the per-column f_cor array).

Turbulent heat flux ocean → interface (W/m^2), > 0 cools the ocean.

Boussinesq reference density (kg/m^3) — THE density that converts the melt MASS flux into a thickness tendency, and the same one ocean_surface_flux_t%rho0 divides the surface tracer sources by. Assigned at configure from the single rho_0 of record; the literal is only the pre-configure type default. The two MUST agree or the salt undo is not the exact negation of what the apply kernel stamped, which is asserted at configure.

Interface salinity (g/kg).

Far-field salinity (g/kg), same average.

Ice salinity (g/kg).

CAVITY_MELT_* per column; CAVITY_MELT_OK where inactive.

Interface temperature (degC), on the liquidus.

Far-field temperature (degC), thickness-weighted over far_field_depth.

Far-field x velocity at the cell CENTRE (m/s), same average.

RMS tidal velocity (m/s) in the melt u*.

Melt friction velocity (m/s).

Friction-velocity floor (m/s).

Far-field y velocity at the cell CENTRE (m/s), same average.

&ocean_cavity_melt_nml volume_compensation, parsed. Requires freshwater = CAVITY_FW_MASS (validate_config).

Meltwater volume this step (m^3); either sign.

Uncovered wet area (m^2), >= 0.

n_thin_step.

n_nonfinite_step from cavity_status_counts_impl.

First dimension (ghosts included).

Second dimension.

Layer count; only k = nz is touched.

Thickness (m) to remove from each open-ocean column.

Static wet (1) / land (0) mask.

Ice-cover fraction (v1 binary).

Layer thickness (m).

h*S.

h*T.

ms%salt_budget_surface.

ms%heat_budget_surface.

h_new/h_old where the sink acted, exactly 1 elsewhere.

Columns whose withdrawal had to be clamped.

First dimension.

Second dimension.

Layer count.

h_new/h_old from cavity_comp_apply_impl.

The tracer’s h*C.

First dimension (ghosts included).

Second dimension.

Layer count; k = nz is the surface / ice-base layer.

Sampling thickness (m), > 0.

Ice-cover fraction (v1 binary).

Static wet (1) / land (0) mask.

Layer thickness (m).

h*T (degC m).

h*S ((g/kg) m).

Zonal face velocity (m/s).

Meridional face velocity (m/s).

Composed solve mask, 0 or 1.

Sampled temperature (degC); 0 where inactive.

Sampled salinity (g/kg); 0 where inactive.

Sampled centred x velocity (m/s); 0 where inactive.

Sampled centred y velocity (m/s); 0 where inactive.

First dimension (ghosts included).

Second dimension.

Ice salinity (g/kg).

Composed solve mask.

Melt mass flux (kg/m^2/s), > 0 melting.

Ocean → interface heat flux (W/m^2).

Far-field salinity the solve used (g/kg).

Heat component (W/m^2), positive down into the ocean.

Virtual salt component, positive salinifies.

First dimension (ghosts included).

Second dimension.

Layer count; k = nz is the surface / ice-base layer.

dt/rho_0 (m^3 s / kg) — multiplies melt to give dh.

The same dt/rho_0, used to undo the surface-flux stamp.

Ice salinity (g/kg).

Composed solve mask; 1 only on covered, wet, sampled columns.

Melt mass flux (kg/m^2/s), > 0 melting.

Far-field salinity (g/kg) the solve used. The VIRTUAL salt component -melt*(s_far - s_ice) is REBUILT from it here, with the same expression cavity_flux_fill_impl uses, rather than read back off sf: on a covered column the assembler’s open-water factor is exactly zero, so Q_salt IS that component bit for bit, and rebuilding it keeps this kernel free of the surface-flux slot.

Interface temperature (degC): the temperature the meltwater joins the column at.

Layer thickness (m); only k = k_top(i,j) is touched.

h*S ((g/kg) m).

h*T (degC m).

ms%salt_budget_surface.

ms%heat_budget_surface.

ms%k_top — the first LIVE layer counting down from the top, nz wherever nothing vanishes against the top. Under a quasi-geopotential coordinate beneath the shelf k = nz is an inert filler on every covered column: growing IT would put the meltwater where the remap drain deletes it, and shrinking it would trip the fatal thin-withdrawal clamp on every column at once (a filler is already AT the marker).

Columns whose withdrawal had to be clamped.

First dimension.

Second dimension.

Layer count.

dt/rho_0.

The same dt/rho_0 (see the sibling’s docstring).

Ice salinity (g/kg).

Composed solve mask.

Melt mass flux (kg/m^2/s).

Far-field salinity (g/kg); the virtual component being undone is rebuilt from it, exactly as in the sibling.

Pseudo-salt h*C.

The SAME first-live-layer index the salt branch used — the mirror’s whole contract is that its increment is bit-identical to salinity’s, which includes landing on the same row.

First dimension (ghosts included).

Second dimension.

Halo width to exclude on every side.

dt/rho_0.

Composed solve mask.

Static wet (1) / land (0) mask.

Ice-cover fraction (v1 binary).

Melt mass flux (kg/m^2/s).

Cell area (m^2).

Meltwater volume this step (m^3).

Uncovered wet area (m^2).

First dimension.

Second dimension.

Composed solve mask.

CAVITY_MELT_* per column.

Count of NONFINITE_INPUT + NONFINITE_STATE.

Count of NOT_CONVERGED.

Count of NO_PHYSICAL_ROOT.

Count of every other non-OK status.

Horizontal grid (for nx_total/ny_total).

The cavity-melt slot.

Reads cover_frac only.

Reads h_layer, the S/T tracer loads, the face velocities, p_top (THE interface pressure) and wet_mask.

Shared EOS handle — the liquidus. Flat POD, by value.

Writes heat_cavity/salt_cavity and latches has_heat/has_salt host-side.

Thermo-cadence gate. Present-and-false ⇒ early return; absent ⇒ run (the ocean_surface_flux_assemble convention).

Horizontal grid (nx_total/ny_total/nghost).

Reads cover_frac and areaT.

The cavity-melt slot; reads melt/active/t_b, writes comp_scale and the thin counters.

Writes h_layer(:,:,nz), the S/T/passive tracer loads, the two surface budget contributors and mass_src.

Thermo timestep (s) — the same therm_dt the surface-flux apply was given.

Per-stage weight for the mass_src accumulator (0.5 per SSP-RK2 stage; 0 / 1 for the pred_corr predictor / corrector), matching ocean_accumulate_mass_out.

Thermo-cadence gate. Present-and-false ⇒ early return; absent ⇒ run.