GitHub

ENA experiment definitions

ENA identifies the Eastern North Atlantic site. A case additionally specifies a date, forcing dataset, grid, physics, and experiment protocol. BreezeLab keeps these choices separate from its shared model assembly. Changing an option produces a recorded sensitivity experiment; it does not silently change the protocol's definition.

Covert et al. (2022)

Sources: paper, public bin inputs, and public bulk inputs. The fetch script pins each repository revision and records SHA-256 checksums.

The supported covert_public_bin protocol follows the runnable public bin input configuration: 18 July 2017, 06–12 UTC, 256×256×192, 35 m horizontal spacing (8.96 km square). It uses time-dependent prescribed sensible/latent heat and stress, simple longwave radiation, full-field large-scale vertical advection, and no mean-wind nudging. The default Breeze microphysics is P3 with N75 initialization.

The paper describes 864×864×192 (30.24 km square), 06–15 UTC experiments. The public configuration is therefore a development benchmark, not the complete paper setup. It omits the original vertical-grid file: BreezeLab reconstructs 192 cells, with 10 m spacing to 1.5 km and stretching to 20 km. Recovering and checking the actual paper grid and experiment variants is required before claiming paper reproduction. The bulk repository is also downloaded for comparison, but is not a separately validated protocol in this package.

LASSO-ENA

See the ENA-LASSO audit for the adopted member, the bundle validation rules and the SAM-vs-Breeze physics differences recorded against the SAM source.

Sources: ARM bundle browser, LASSO-ENA dataset, and LASSO SAM source (lasso_ena_noice, reference revision 12d02446a2147388dc89d828e6e0553106abea0f).

Obtain a samin bundle through an ARM account and stage it with julia data_wrangling/stage_lasso_bundle.jl ARCHIVE.tar, which extracts snd, lsf, sfc, prm and grd into data/lasso/<run ID>/ and freezes the archive checksum, member tokens, DOI and SAM revision in bundle.toml. The adopted member is 20170718era5d25x100_sbmwrm-aer2-flxsst (the plan's …era5s1n0… spectral-bin candidate does not exist at ARM; see the audit). Do not substitute the public Covert inputs.

The current adapter targets warm-cloud, ocean, SST-flux, wind-nudged LASSO members. inspect_lasso_bundle reads the grid from grd, horizontal spacing and timestep from prm, duration from nstop*dt, radiation cadence from nrad*dt, the nudging timescale, perturb_type and compute_reffc, and rejects every unsupported namelist setting with its reason. The member run ID, the UTC epoch (SAM's day0 does not identify the year) and, when the namelist carries neither nx_gl,ny_gl,nz_gl nor a numeric Nx×Ny×Nz case ID, the dimensions must be given explicitly; they are recorded in provenance and checked against the bundle. Use Nx, Ny, and z_faces overrides for grid sensitivities.

With the staged bundle, the readable script runs the member end to end:

julia --project cases/ena_lasso.jl        # member from ENA_LASSO_MEMBER, bundle from ENA_LASSO_BUNDLE

or, from the CLI, after confirming the bundle's date, inputs and grid:

julia --project cases/cli/run_case.jl \
  --protocol lasso_ena_official \
  --member 20170718era5d25x100_sbmwrm-aer2-flxsst \
  --data data/lasso/20170718era5d25x100_sbmwrm-aer2-flxsst \
  --epoch 2017-07-18T00:00:00 --dimensions 256,256,260 \
  --arch gpu --float Float32 --output output/ena_lasso

(the epoch above is the day-of-year of day0 in the member's year; the adapter rejects a mismatch). The API equivalent is:

using BreezeLab, Oceananigans, Dates
case = ena_lasso(; member="20170718era5d25x100_sbmwrm-aer2-flxsst",
                   epoch=DateTime(2017, 7, 18, 0), arch=GPU())   # dimensions=:documented → 256×256×260
write_provenance("provenance.toml", case)
run!(case.simulation)

The adapter defaults to SST-based bulk surface exchange, RRTMGP longwave/shortwave, mean-wind nudging on the tauls timescale, full-field vertical advection, P3-aer2, and diagnostic CCN replenishment. These defaults describe a Breeze counterpart of the LASSO forcing protocol; the symbol lasso_ena_official identifies the input protocol, not a certification of equivalent results.

Shared forcing formulation

For a transported variable φ, the imposed large-scale vertical-advection contribution is

\[(∂φ/∂t)_{LS,v} = -w_{LS}(z,t) ∂φ/∂z.\]

It acts on the resolved field. Mean-wind nudging contributes a horizontally uniform acceleration, independently for each horizontal component:

\[(∂u/∂t)_{nudge} = [u_{LS}(z,t) - ⟨u⟩(z,t)] / τ_{LS}.\]

SAM moisture inputs are mixing ratios r per dry-air mass. In clear air the Breeze specific-humidity conversion is q = r/(1+r), with tendency dq/dt = (dr/dt)/(1+r)^2. In cloudy air, holding condensate mass fraction qᶜ fixed, the vapor tendency is (qᵈ)^2/(1-qᶜ) * drᵛ/dt, where qᵈ is dry-air mass fraction. Thermodynamic forcing is converted to the chosen Breeze prognostic variable while preserving the specified physical temperature tendency. See large_scale_forcings.jl and its tests for the implemented conversions.

Known differences to resolve or quantify

  • Breeze P3/one-moment microphysics differs from SAM's HUJI bin microphysics.
  • The default subgrid closure/advection are Smagorinsky–Lilly/WENO; they differ from the SAM reference's TKE closure and transport.
  • RRTMGP columns currently stop at the LES top and use prescribed effective radii. Matching the reference requires checking the atmospheric column above the LES, radiation state, optical assumptions, and heating/flux profiles.
  • Bulk surface exchange differs from SAM's iterative stability/gustiness treatment. The current bulk-flux path uses a zero translation velocity to retain ground-relative surface winds; the SAM translation is recorded in the case label.
  • The diagnostic CCN reservoir projection is applied once per model timestep, rather than at every microphysics call as in the reference algorithm.
  • Converting SAM scalar levels to Breeze finite-volume cells needs comparison with the selected bundle's staggered grid, especially near the surface and model top.

LASSO reproduction acceptance checklist

  1. Freeze one ensemble member. Archive the bundle ID, DOI, input archive checksum, SAM revision, and reference outputs. Confirm date, spin-up/evaluation windows, day0, dimensions, scalar/face heights, aerosols, SST, nudging, and radiation flags.
  2. Match imposed tendencies. Compare interpolated sounding/forcing and surface time series at selected times. Check units/signs, wind frame, pressure/height conversion, column mass/water budgets, and horizontal/vertical advection separately.
  3. Check each physics component. Compare surface H/LE/stress and radiation fluxes/heating against SAM before the full coupled run. Resolve the differences above or label them as controlled model differences with sensitivity tests.
  4. Run a short GPU test, then the complete member. Inspect finite values, timestep stability, negative water/number concentrations, budgets, and output completeness. Preserve both Float32 and Float64 short-run checks where practical.
  5. Evaluate statistically. Compare LWP, cloud fraction and boundaries, precipitation, mean thermodynamic/wind profiles, and surface fluxes over common sampling windows against SAM and ARM observations. Agree on tolerances and uncertainty estimates before declaring reproduction; matching chaotic snapshots is not the criterion.

The automated protocol test uses a synthetic SAM-style fixture and a tiny sensitivity with one-moment microphysics, radiation disabled, and no aerosol replenishment. The public-input CPU smoke also uses a reduced grid/duration. Neither establishes full LASSO or Covert scientific fidelity, and no official ARM bundle is checked into this repository.