Prescribed ocean geometry

Use public OceanGeometry inside an Experiment for a complete standalone run with physical arrays, initial ice and prescribed ocean/atmospheric fields. Prescribed fields are inputs supplied by you, such as ocean temperature. Prescribed ocean geometry and custom fields provides a runnable example using only public imports. It supports a single device or explicit x/y device meshes and needs no external ocean package. Supply geometry on the global physical grid. Initialization adds halos (extra boundary cells) and calculates grid metrics (geometric quantities such as areas).

The low-level adapter below accepts geometry that includes halos. Creating its arrays alone does not define a complete physical experiment.

veris.grid.ocean.initialize_from_ocean creates a fresh State, Configuration and PhysicalConstants through veris.execution.initialization.initialize. State holds the model arrays. The adapter accepts OceanGeometry directly; no existing Veris state is needed. The geometry arrays include two halo cells at each horizontal boundary. Volume masks identify ocean and land through depth and use the last vertical level as the surface. The geometry determines horizontal dimensions, which take priority over settings overrides. The grid metric equations keep the original periodic four-cell corner-area mean.

Supply ocean forcing (prescribed inputs) and initial ice fields as full arrays that include halos:

from veris.grid.ocean import initialize_from_ocean

state, conf, phys = initialize_from_ocean(
    geometry,
    dtype="float32",
    settings_overrides={"deltatDyn": 600.0},
    physical_overrides={"rhoIce": 920.0},
    state_overrides={"theta": ocean_temperature, "ocSalt": ocean_salinity},
)

State fields you do not supply use the VARIABLES registry defaults. Masks, metrics and surface temperature calculated from geometry take priority over state overrides. geometrySurfaceTemperature sets the initial surface temperature. All arrays and physical coefficients use the selected precision. Geometry and your override dictionaries stay unchanged. The host (the CPU running Python) rejects invalid shapes, nonfinite geometry (NaN or infinity) and spacings or areas that are not positive.

The ocean adapter lives in veris.grid.ocean. Its input structure is veris.grid.geometry.OceanGeometry. See Scientific package hierarchy for the modules responsible for each part.

To repeat timesteps, fix the initialized model objects as arguments of the physics function:

from functools import partial
from veris import advance, step

transition = partial(
    advance, configuration=conf, constants=phys,
    forcing={"Qnet": 100.0, "Qsw": 0.0}, physics="coupled",
)
final = step(state, transition, 3, has_aux=True)

Choose and supply the ice and forcing fields needed by that function. Geometry initialization alone does not define the experiment. The example uses the coupled island physics sequence, which combines ice motion with freezing and melting. See Scan rollouts and checkpointing for the shared rules for repeating timesteps and automatic differentiation (AD), which calculates how results change with inputs.

See initialize_from_ocean() in the Python API for its signature and docstring.