Growth only

The reference column represents ice and snow above ocean water. It starts with 1.3 m ice, 0.1 m snow, concentration 0.9 (the fraction of each cell covered by ice), 253 K air and 80 W/m2 downward longwave radiation (heat radiated from the atmosphere). It executes 150 daily Growth steps, calculating freezing and melting without ice motion or transport between cells:

python -m veris.setups.run_growth --backend cpu --output-snapshot output/growth.nc
python -m veris.setups.run_growth --backend gpu --output-snapshot output/growth-gpu.nc

The CLI uses Experiment.builtin("growth", ...) and Simulation. Python experiments use the same interface for initial conditions, forcing (prescribed inputs that are constant or vary with time), output and complete serial restart. The built-in setup defaults to Grid(2, 2). You can also choose global dimensions and explicit device meshes. See Define, run and resume a sea-ice experiment and Simulation lifecycle.

Returned Qnet/Qsw feed the next step exactly as in the original experiment. They are not reset to initial forcing. The shared allocator requires at least two cells per interior axis, so the single-column experiment uses a uniform 2 by 2 interior with two halos per edge. Halos are extra cells for boundary or neighboring data. The equations at each grid point are identical to those of a single column. Output always removes those storage halos. initialize_experiment(Experiment.builtin("growth", grid=Grid(2, 2))) creates the initial column. Use public advance(..., physics="thermodynamics") and step to advance it with pure functions and automatic differentiation (AD). Pure functions do not change outside resources; AD calculates how results change with inputs.

The setup uses the jax_halo_exchange notebook’s initial conditions and daily steps with the shared Growth routine. With the default inputs, ice decreases from 1.3 m and disappears at the end of day 140. The notebook plots values before each step (days 0–149). Scheduled instantaneous output includes the initial value and values after completed steps (days 0–150).

To record and plot the evolving ice thickness, request a history stream:

python -m veris.setups.run_growth --backend cpu \
    --output-snapshot output/growth-history-final.nc --output-history output/growth-history.nc \
    --io-variables hIceMean

Then read the netCDF time series without needing an interactive environment:

import h5netcdf
import matplotlib.pyplot as plt

with h5netcdf.File("output/growth-history.nc", "r") as result:
    history = result.groups["instantaneous"]
    days = history.variables["time"][:] / 86400
    ice = history.variables["hIceMean"][:, 0, 0]
    plt.plot(days, ice)
plt.xlabel("days")
plt.ylabel("ice thickness / m")
plt.show()

Reference adaptations

The column starts fields that were not prescribed at zero, exactly as initialize_growth.py does. Prescribed fields and physical constants keep the reference values. Experiment-specific inputs live in the scenario registry below; physical constants stay in PhysicalConstants.

Grid dimensions and run duration belong to Grid and Experiment.steps. GrowthSettings and GROWTH_SETTINGS contain only the physical column scenario. The field provider and run profiles keep the 2 by 2 grid and 150-step defaults. Scenario and physical overrides use the same grouped definition as other presets:

import jax
from veris import Experiment, Grid, initialize_experiment

jax.config.update("jax_enable_x64", True)
experiment = Experiment.builtin(
    "growth", grid=Grid(2, 2), scenario={"hIceMean": 1.3},
    constants={"rhoIce": 900.0},
)
state, settings, constants = initialize_experiment(experiment)

Tests check initial fields and shared settings/constants against the standalone reference jax_halo_exchange commit 2571998f8feb6a5943b214361c122ca78b44c3a8. The column evolves with the same thermodynamic calculations as other Veris experiments. These include heat and snow budgets that account for cells covered only partly by ice.

Derivative validation

Tests check the column’s sensitivity to longwave radiation with finite differences, which compare results after small input changes. See Automatic differentiation at model boundaries for derivative choices when ice is absent and at thresholds where the model selects a different physical branch.

Scenario controls

Growth only scenario controls

Setting	Default	Units	Description
hIceMean	1.3	m	Initial grid-cell mean ice thickness
hSnowMean	0.1	m	Initial grid-cell mean snow thickness
Area	0.9	1	Initial ice concentration
TSurf	273.0	K	Initial ice surface temperature
wSpeed	2.0	m s-1	Prescribed wind speed
ocSalt	29.0	g kg-1	Prescribed ocean salinity
oceanTemperatureC	-1.66	degC	Prescribed ocean temperature
Qnet	173.03212617345582	W m-2	Initial net upward heat flux
LWdown	80.0	W m-2	Prescribed downward longwave flux
ATemp	253.0	K	Prescribed air temperature