version: "1.0.0" name: meridional-transport description: "Meridional heat, volume, freshwater transport: section masks, overturning and gyre decomposition, RAPID comparison at 26.5N." user-invocable: false

meridional-transport

Background expertise for the METHOD of computing meridional transports natively and framing them honestly. Dataset-specific facts (basin scope of a given product, expected values, the RAPID-comparison discipline) are not restated here: they live in the knowledge bundle (the ecco- gotchas, recipes, and attested computations the ecco skill lists) and are consulted per analysis by the transport-analysis workflow and the ecco-scout agent, as the core consult-knowledge skill sets out. Grid mechanics live in ocean-grids and the ecco skill's grid reference; expected values and spreads live in the recipe and computation concepts.

Sections are masks, not index rows

On any curvilinear grid, a latitude circle crosses cells irregularly; a section is a set of cell FACES with signs, not a j = const slice. ecco_v4_py.get_section_line_masks (and get_available_sections for named straits) produces the west- and south-face masks; transports sum face fluxes times face geometry over those masks. Grid-native face directions make rotation unnecessary in the sum (the ecco skill's grid reference, step 8).

Volume, heat, freshwater

• Volume (Sv): sum of UVELMASS * dyG * drF and

VVELMASS * dxG * drF over the masked faces; the MASS variables already carry hFac and bolus is separate (below).

• Heat (PW): prefer the model's advective heat-flux diagnostics

(ADVx_TH, ADVy_TH for ECCO) summed over the section, times rho0 * c_p on the temperature flux: these capture exactly what the model transported, including sub-monthly covariance that reconstructed v-times-theta from monthly means misses. A heat transport is physically well defined only across a section with zero net mass transport; where net mass transport is nonzero, the number is a temperature transport relative to a stated reference, and the reference must be stated.

• Freshwater (Sv): same structure with salt fluxes and a stated

reference salinity; the reference-dependence caveat applies identically.

Total means Eulerian plus bolus

In eddy-parameterized models the tracer transport includes the bolus (GM) component; ECCO ships it as its own product (UVELSTAR family). Eulerian-only transport labeled as total is a systematic bias; the flux diagnostics (ADV*_TH) already include everything the model did, which is another reason to prefer them for heat.

Overturning and gyre decomposition

Zonally integrate meridional transport in depth space and cumulate vertically for the overturning streamfunction (AMOC strength is its maximum at the section, in Sv). Decompose section transport into overturning (zonal-mean velocity times zonal-mean property) and gyre (deviation) components to attribute mechanisms; on depth surfaces this decomposition is convention-laden near western boundaries, so name the convention used.

The RAPID comparison at 26.5N

The RAPID-MOCHA array at 26.5N is the standard ground truth for Atlantic sections. The METHOD discipline: compute the model section, then consult the section's recipe and gotcha concepts in the knowledge bundle for what to compare against (expected range, spread, the basin scope the observation uses, validation provenance) and cite them; this skill carries no expected ranges, spreads, or product-specific scope rules of its own. Time-sampling matters generally: a single year's mean sits inside a wider envelope than a multi-year mean, and a model window overlaps an observational array only where their periods intersect.

Must NOT

• Never hardcode an expected transport range or spread; read the

recipe concept and cite it.

• Never compute transports on regridded fields (ocean-grids rule;

native faces only).

• Never slice j = const as a section on a curvilinear grid.
• Never quote a heat transport across a mass-unbalanced section

without stating the reference temperature.

• Never label Eulerian-only transport as total in an

eddy-parameterized model.

• Never compare a model mean against RAPID without matching the

overlap period or saying the periods differ.