WIP: Merge of GZ-COBALT and COBALT-DVM into eco-COBALT

[jessluo]

[WIP] Merge GZ-COBALT into eco-cobalt: gelatinous zooplankton with diel vertical migration

Merges the two generations of COBALT development into a single "ecologically heavy" model: COBALTv3 + diel vertical migration (COBALT-DVM, now in dev/eco-cobalt) gains the two tunicate groups, food-dependent assimilation efficiency, and fast-sinking iron detritus from GZ-COBALT, with the large tunicates both migrating vertically and producing fast-sinking fecal pellets. This work was done with assistance from Claude Opus 5.

Base: dev/eco-cobalt · Diff: +2,868 / −1,230 across 4 files · 4 commits

dev/eco-cobalt is currently 1 commit ahead (15e766d, the jmetabo_n diagnostic fix) — see Follow-ups.

---

Background

Two lines of COBALT development needed reconciling:

• dev/eco-cobalt — COBALTv3 (4 phytoplankton, 3 zooplankton, fast-sinking N and P detritus) plus Mathieu Poupon's diel vertical migration work, which added two vertically migrating zooplankton groups (vmmdz, vmlgz) carrying gut and metabolite pools.
• GZ-COBALT — still using COBALTv2. Adds small tunicates (appendicularians) and large tunicates (salps), a food-dependent assimilation efficiency, and fast-sinking iron detritus.

The merge target is 7 zooplankton groups in which large tunicates migrate and short-circuit the biological pump via dense fecal pellets and salp falls.

---

Design decisions

Decision	Choice	Rationale
Zooplankton IDs	Append-only: SMZ=1, MDZ=2, LGZ=3, VMMDZ=4, VMLGZ=5, SMT=6, LGT=7	Prey slots 1–10 keep their meaning; only detritus moves 11 → 13. Avoids renumbering ~200 references.
Migrator detection	zoo%does_dvm logical	Not an index-ordering test, so future groups can be appended without silently breaking DVM.
Assimilation efficiency	Unified across all 7 groups	phi_* become fractions of egestion summing to 1.0, not fractions of total ingestion.
Silica partitioning	Unchanged (phi_det_si, no AE factor)	Opal is never assimilated, so its split is independent of AE.
GZ features carried over	Salp falls (LGT aggregation), phi_aresp	phi_nh4/phi_po4 and fast-sinking Si were left out as out of scope.
LGT diet	GZ's prey list plus COBALTv3's medium phytoplankton	Leaving a v3 phytoplankton class invisible to salps is hard to defend mechanistically.
mesozoo_200 scope	Crustacean only	Targets the COPEPOD database, whose net sampling does not retain gelatinous zooplankton. Tunicates get their own tunicate_200.

The AE ↔ gge_max equivalence

Worth recording, because it governs how the legacy groups behave. GZ's metabolism reduces exactly to
COBALTv3's when AE is constant:

AE·I − bresp − phi_aresp·I  =  (AE − phi_aresp)·I − bresp     so   gge = AE − phi_aresp

The defaults AE = 0.7, phi_aresp = 0.3 therefore reproduce the previous gge_max = 0.4 for
smz/mdz/lgz. The two formulations still differ in one respect — see Answer changes.

---

What changed

cobalt_types.F90

• NUM_ZOO 5 → 7, NUM_PREY 11 → 13.
• Named IDs: SMZ…LGT and PREY_DIAZO…PREY_DET, replacing bare numeric indices throughout.
• Zooplankton type gains does_dvm, agg, frac_fast_det, ipa_smt, ipa_lgt, plus
  jaggloss_n/p and jprod_{n,p,fe}det_fast arrays and their diagnostic ids.
• generic_COBALT_type gains the fedet_fast family, tunicate tendency/pointer members, and
  jprod_tunicate_200 / f_tunicate_200.

Ecology (generic_COBALT.F90)

• Ingestion — all seven blocks regenerated in the COBALTv3 food1/food2 switching idiom with
  named prey indices. Tunicates enter the carnivory side of switching for mdz/lgz/vmmdz/vmlgz and are
  added to higher-predator switching (smt with the medium size class, lgt with the large).
• Egestion and production — unified on the dynamic AE for all seven groups, with frac_fast_det
  splitting detritus between the standard and fast-sinking pools. Migrating groups still route
  egestion through gut clearance rather than directly from ingestion.
• Salp falls — LGT aggregation (agg, quadratic, no stress dependence) routed entirely to the
  fast-sinking pools. Required adding zoo%agg and zoo%jaggloss_n/p to the v3 zooplankton type,
  which had no zooplankton aggregation at all.
• DVM — generalized from two hardcoded groups to per-group arrays with a dvm_prey_wgt mask, so
  LGT migrates. Gated on does_dvm throughout.
• Fast-sinking iron — fedet_fast end to end: production from zooplankton and higher-predator
  egestion, jfe_ads_fast scavenging (with the half-the-available-iron limiter shared across both
  detrital pools so they compete for the same dissolved pool), jremin_fedet_fast on the fast N
  pool's own specific rate, source/sink assembly, and bottom flux.
• Mass balance — tunicate pools added to the N/P/C sums; fedet_fast and the LGT gut pools to Fe
  and Si.

Diagnostics

85 new registered fields: full SMT and LGT sets mirroring vmLgz, per-group aggregation and
fast-detritus production (registered in a loop rather than group by group), the fast-iron fluxes, and
jprod_tunicate_200 / tunicate_200 ported from GZ-COBALT as the JeDI-comparable counterparts to the
COPEPOD-comparable mesozoo pair.

---

Pre-existing bugs found and fixed

These were latent on dev/eco-cobalt and are unrelated to the tunicates, but had to be resolved to
make the merge coherent.

1. zoo%assim_eff was computed but never used. It was calculated each timestep and registered as
a diagnostic, but no flux consumed it — the maths still used a static
1 − phi_det − phi_ldon − phi_sldon − phi_srdon. This is why the phi_* rescaling and the egestion
rewrite had to land together rather than as separable steps.

2. Seven zooplankton parameters were declared and read but never initialized: k_clear_gut,
k_temp_gut, k_clear_met, assim_eff_max, assim_eff_min, kae, phi_aresp. The first three
drive the live gut/metabolite path, so vmmdz and vmlgz were integrating on uninitialized memory.
All seven are now registered. Gut evacuation constants use the values from the COBALTv2-DVM code these
routines descend from (Poupon et al. 2024): k_clear_gut = 8.0 day⁻¹,
k_temp_gut = 4.32 day⁻¹ °C⁻¹, k_clear_met = 1.0 day⁻¹.

3. Sixty unguarded accesses to migrator-only arrays. does_dvm guards the allocation of the 21
gut and metabolite arrays, but accesses in three clusters — tracer fetches, vmove setters, and
tendency assembly — sat outside any guard. Benign on dev/eco-cobalt, where migration was not
switchable; fatal once it became a parameter. Now guarded per group.

Adaptations required by the convention change

Calcite and aragonite production. jprod_cadet_arag / jprod_cadet_calc used zoo%phi_det
directly as "the proportion of consumed material that ends up as detritus" — correct while phi_det
was a fraction of ingestion. Under the new convention it needed (1 − AE)·phi_det; without that
change calcite production would have roughly tripled.

---

Answer changes

Expected, and not a regression: the legacy groups will not reproduce dev/eco-cobalt
bit-for-bit even with tunicates switched off. The unified scheme applies N/P colimitation as
min(I_N, I_P/q) up front, whereas v3 applied a phosphorus cap after computing production. Small
drift is expected; large drift means something else is wrong.

jprod_mesozoo_200 and f_mesozoo_200 now include the migrating crustacean groups (from
dev/eco-cobalt), so both rise relative to pre-merge runs.

Gut residence times shorten in warm water relative to the WIP commit's initial Dam & Peterson
estimate: the two forms cross near 4.5 °C, and at 20 °C the adopted values give ~94 vs ~64 day⁻¹
(≈15 vs ≈22 min), which weakens active transport in the tropics and subtropics.

---

New model interface

9 prognostic tracers: nsmt, nlgt, nlgt_gut, plgt_gut, felgt_gut, silgt_gut,
nlgt_met, fedet_fast, fedet_fast_btf.

146 new namelist parameters, chiefly: full trait sets for smt and lgt; assim_eff_max,
assim_eff_min, kae, phi_aresp, agg, frac_fast_det, does_dvm, and the three gut/metabolite
kinetic constants for all seven groups; 26 new innate prey availabilities.

Restart compatibility: the new tracers are prognostic and TRACERS_MAY_REINIT defaults to
.false., so restarting from a pre-merge restart file is a fatal error. Cold-starting from
GENERIC_TRACER_IC_FILE is the supported path. Note that nsmt and nlgt need non-zero values
in that file: zero biomass is an absorbing state, since ingestion and basal respiration both carry
zoo%f_n as a multiplicative factor, so they would remain identically zero for the whole run. The
gut, metabolite and fedet_fast pools are fine starting at zero.

---

Configuration notes

does_dvm may only be enabled for vmmdz, vmlgz and lgt — the three groups that own gut and
metabolite tracers. Enabling it elsewhere is now rejected at initialization with an explicit message
rather than dying later with an opaque nitrogen imbalance.

Switching it off is fully supported and enables two useful sensitivity experiments:

Configuration	Isolates
swim_max_lgt = 0.0	Salps keep gut-transit physiology but never move vertically → the vertical transport alone
does_dvm_lgt = .false. and swim_max_lgt = 0.0	Salps egest where they feed, as in GZ-COBALT → transport and gut-lag physiology together

Both settings are needed for the second: the non-migrating branch still derives a swimming speed from
swim_max, and nlgt carries move_vertical, so LGT would otherwise still oscillate vertically with
none of the gut machinery.

Tracer registration is deliberately not conditional on does_dvm, so the tracer list — and hence
restart file contents — stays identical between the migrating and non-migrating configurations, which
is what makes them directly comparable.

---

Testing status

Done: all four modified files pass gfortran -fsyntax-only, and generic_COBALT.F90 compiles at
-O1 -Wall, against throwaway stub modules for the twelve FMS/MOM6 dependencies. This validates block
structure, every variable and derived-type component name, and array ranks across the whole diff,
including the programmatically generated diagnostic and ingestion blocks. A warning comparison against
dev/eco-cobalt under identical stubs shows no new warnings. The does_dvm guarding commit was
additionally verified to be pure gating: a line-by-line comparison ignoring comments and whitespace
confirms zero code lines removed or reordered, so the default configuration is unchanged.

Not done — this has never been built against real FMS/MOM6, and has never been run. The stub check
cannot catch argument disagreement with real FMS interfaces.

Suggested sequence:

1. Build against FMS/MOM6.
2. Short run with tunicates off (imax_smt = imax_lgt = 0) and do_fastsinking = .true.. Since
   the merge altered the egestion and production maths for all seven groups, this separates "did the
   refactor break the existing model?" from "do the new groups behave?" The pre/post N, P, C, Fe and Si
   conservation checks in generic_COBALT_update_from_source are the primary test — they call
   mpp_error(FATAL, …) past imbalance_tolerance, so a run that survives them exercises nearly all
   the new flux routing.
3. Enable tunicates; confirm plausible biomass, non-zero jprod_ndet_fast_Lgt, and a diel cycle in
   vmove_Lgt matching vmmdz/vmlgz.
4. Confirm jfe_ads_fast and jremin_fedet_fast are non-zero and the Fe budget still closes.

---

Known limitations and follow-ups

• Tuning. Tunicate trait defaults are GZ-COBALT's COBALTv2-tuned values carried over unchanged
  into a different food web. They are a starting point, not a calibration.
• Tunicate gut kinetics inherit copepod values. k_clear_gut / k_temp_gut are the crustacean
  numbers. Salp gut passage differs, and since LGT both migrates and produces the fast-sinking
  pellets, these directly control how much salp flux reaches depth. Flagged in-code as retuning
  candidates.
• bresp ×0.9 for migrators. Poupon's COBALTv2-DVM scaled migrator basal respiration to 90% of
  the resident equivalent, presumably because the DVM path adds a swimming cost on top. eco-cobalt
  carries the unscaled values. Unresolved: port bug or deliberate?
• Diagnostic registration coverage. Which DVM-era diagnostics are registered for which group is
  inconsistent in places. Harmless — every array is now allocated for every group, so a mismatch costs
  a missing-value field rather than a crash — but worth a tidy-up pass.
• gge_max is still registered but no longer used for zooplankton. Left in place with a comment
  rather than removed, to avoid breaking existing parameter files.
• dev/eco-cobalt is 1 commit ahead (15e766d, jmetabo_n). Should merge cleanly; the send it
  adds is guarded on numeric group indices and should become does_dvm here so LGT is included.

---

Commits

Commit	Contents
9b8811f	The GZ-COBALT merge itself: 7 groups, named IDs, tunicate tracers and ingestion, unified dynamic AE, fedet_fast, salp falls, generalized DVM, mass balance. ~2,700 lines.
946acbb	Merge dev/eco-cobalt. Resolves the mesozoo_200 and deallocation-guard conflicts; fixes a 100 m allocated-vs-registered mismatch introduced by 9b8811f; ports the tunicate 200 m diagnostics.
7cf40b6	Replace the Dam & Peterson gut evacuation estimate with the COBALTv2-DVM values.
e934fc1	Make does_dvm a valid off-switch: guard the three access clusters, reject unsupported configurations, correct a misleading comment.