ENA-LASSO: the official LASSO-ENA member in Breeze

Companion to ENA experiment definitions. This page records, against the LASSO SAM source, what the lasso_ena_official adapter does and does not do, which inputs it requires explicitly, and every physics difference that matters for comparing Breeze with the SAM member. It is an audit, not a claim of reproduction: no official bundle had been run when it was written (7 October 2026).

1. Member identity and data status

1b. The staged bundle (ARM order 284988, staged 8 October 2026 05:06 UTC)

data/lasso/20170718era5d25x100_sbmwrm-aer2-flxsst/ (stage_lasso_bundle.jl; archive SHA-256 b7594c2d…, 53 739 520 B; samstat f5cb2bed…, sam2d 4a4c6682… under data/lasso/archives/). Bundle README: SAM v6.10.3 plus LASSO modifications, model_source_git_hash f83adf58 (the public branch head audited above is 12d0244; the member was produced by the earlier revision), stagesam_ena 8404ad7. Facts read from the files:

validate_lasso_bundle reports 0 problems and 8 recorded warnings (CPU job 241; analysis/check_lasso_bundle.jl), and tiny CPU cases built on the full 260-level column step finitely.

1c. Short GPU integration (plan step 6, first part)

Slurm job 243 on A100 GPU-c764cc9c (gpu-p4de-2c-2), source fa26811, cases/cli/run_case.jl with the documented command and --hours 1 --profile_interval 600 --slice_interval 1800: 3600 steps of 1 s in 44.1 min (0.69 s per step, RRTMGP every 30 steps), finite throughout, cloud fraction 1 from t = 10 min, LWP 53 → 91 → 61 g m⁻², max |w| ≤ 2.9 m s⁻¹, max qᶜˡ 1.6 g kg⁻¹, rain 0.002 mm d⁻¹, T ∈ [252.3, 295.6] K. Aerosol check (runs/lasso_aer2_1h_job243/checks/): nᵃ + nᶜˡ equals the capped total everywhere (2.384×10⁸ kg⁻¹), nᵃ constant outside cloud, in-cloud nᶜˡ/CF ≈ 2.1–2.7×10⁸ kg⁻¹ (≈ 250–320 cm⁻³), i.e. the activation still reaches the cap (the Breeze ratchet, now bounded by the SBM's own maximum).

First-hour comparison with the SAM member (runs/lasso_aer2_1h_job243/checks/sam/, analysis/compare_sam_reference.jl; SAM samstat is a 2-min domain statistic, Breeze a 60-s domain mean): cloud water path 72.8 (SAM) vs 67.3 g m⁻² (Breeze) over the hour, both rising to a peak of ≈ 115 / 100 g m⁻² at 00:37 UTC and relaxing; shaded cloud fraction 1.00 vs 0.98; surface precipitation ≈ 0 in both; Breeze's rain water path reaches 0.06 g m⁻² against SAM's 0.002 (earlier drizzle onset in P3). The hour-1 profiles of θ, qᵛ, u and v coincide with SAM's to within the line width up to 3 km, and both models carry two cloud layers (≈ 500–800 m and ≈ 1200–1500 m) with the same shapes; Breeze additionally holds a thin layer at ≈ 250 m that SAM does not, and its lower-layer qᶜˡ peaks at 0.13 vs SAM's 0.17 g kg⁻¹ (sam_profiles.png). The member starts with near-saturated air below the stratocumulus, which is why both models report very low cloud bases in hour 1 (SAM's GCSS ZCB/ZCT are undefined/≈ 0 there). MWRRET has no samples before 03:10 UTC.

The full 24-h run (Slurm job 249, runs/lasso_aer2_24h_job249/, --time=22:00:00, ≈ 17 h expected) was queued for the next physically idle A100 at 06:25 UTC; its comparison with samstat/sam2d and the ARM observations over 03:10–24:00 UTC (analysis/compare_sam_reference.jl, analysis/compare_ena_observations.jl) is plan step 7.

2. What protocol = :lasso_ena_official does

inspect_lasso_bundle / validate_lasso_bundle (src/lasso_bundle.jl) read the five bundle files and either return the protocol defaults for build_case or reject the bundle with every offending setting named. The namelist conventions were taken from setparm.f90 (&PARAMETERS), SGS_TKE/sgs.f90 (&SGS_TKE) and params.f90/grid.f90 (defaults for absent keys).

Required explicitly (never inferred):

• the member run ID (member=; aerosol setting and SAM microphysics come from it, because the HUJI-SBM spectrum is compiled into the SAM binary, not read from prm);
• the UTC epoch (SAM's day0 has no year; it is checked against day0 and the member date);
• the domain dimensions=(Nx, Ny, Nz) unless prm carries nx_gl/ny_gl/nz_gl or a numeric NxxNyxNz case ID (SAM compiles the domain into domain.f90; the documented d25x100 domain 256×256×260 can be requested as dimensions=:documented in ena_lasso and is recorded as such).

Read from the bundle: dx, dy, dt (fixed step: Δt = max_Δt = dt), nstop (duration nstop × dt), day0, latitude0/longitude0 (f-plane), tauls (wind-nudging timescale), nrad (radiation every nrad × dt), doupperbound, dodamping, read_in_geostrophic_wind (7- or 9-column lsf), perturb_type (cases 0 and 5 implemented), compute_reffc (prescribed effective radius: 14 μm when .false., SAM's ocean value), ug/vg (translation frame, recorded, not applied), dosmagor (recorded). Scalar levels come from grd with SAM's interface rule (setgrid.f90: interfaces halfway between levels, top interface one half-spacing above the top level; missing lines extended with the last spacing).

Rejected with the reason: prescribed fluxes (SFC_FLX_FXD, SFC_TAU_FXD), land or CEM modes, per-column fluxes (UNIFORM_SFC_FLX = .false.), any T/q nudging (donudging_tq/t/q, transient or inversion-following nudging — the n1/n2 members), a wind-nudging height window, delayed forcing (timelargescale > 0), rad_simple or missing LW/SW, prescribed radiative forcing, restarts, subensembles, tracers, smoke, column mode, walls, SCAM input, perpetual or fixed-sun insolation, homogenized radiation/SST, slab ocean, scaled CO₂, removed trace gases, a Coriolis parameter detached from the latitude, a constant-dz grid, perturb_type ∉ {0, 5}, a sounding that does not reach the model top (SAM would continue with the 1976 standard atmosphere), snd records not bracketing day0, lsf starting after day0 and sfc not covering the whole run (SAM extrapolates these linearly; Breeze clamps).

Conversions: pressures hPa → Pa, moisture g kg⁻¹ → kg kg⁻¹ dry mixing ratio; the initial sounding is the snd pair bracketing day0 interpolated in time (setdata.f90), its mixing ratio interpolated to the grid before conversion to a mass fraction q = r/(1+r); heights of pressure-level records from SAM's dry hydrostatic recipe (R = 287, cp = 1004, g = 9.81); lsf columns interpolated linearly in height (height records) or pressure (pressure records) with zero tls/qls/wls and held winds above the record top (forcing.f90); qls (a dry-basis vapor source) converted to a mass-fraction rate dq/dt = qᵈ²/(1−qᶜ) qls (= qls/(1+r)² in clear air); tls converted to the energy prognostic so that dT/dt = tls; wls applied once, as full-field first-order upwind advection of every prognostic (subsidence.f90, levels 2…Nz−1), never as mean-profile subsidence in addition; winds are ground-relative (no translation frame; bulk fluxes need the absolute wind). All of this is checked by test/ena_lasso.jl on the synthetic fixture at several heights and times.

3. Physics differences that matter (audit)

4. Observations for 18 July 2017

Staged at /shared/home/greg/breezelab-runs/20261006/inputs/observations/20170718/ (checksums in its manifest.json): enamwrret2turnC1.c1 (phys_lwp g m⁻² with qc_phys_lwp, phys_qc_flag, 1-σ uncertainty; irregular ~1-min samples from 03:10 UTC), enaarsclkazrbnd1kolliasC1.c0 (4-s cloud_layer_base_height/cloud_layer_top_height for up to 10 layers, cloud_base_best_estimate), enavdisC1.b1 (1-min rain_rate mm hr⁻¹ with qc_rain_rate). src/arm_observations.jl reads them with their QC; analysis/compare_ena_observations.jl compares a run's domain-mean LWP, rain rate, cloud fraction and cloud base/top with them over the run's own UTC window and labels the run by its protocol. The LES is a 25.6 km domain mean; the observations are a point/zenith time series, so the comparison is statistical (means, quantiles, cloudy fractions) and carries the retrieval uncertainty; matching snapshots is not the criterion.

5. Status against the plan's "done" criteria

• Separate run script (cases/ena_lasso.jl), constructor (ena_lasso), staging/freeze tooling (data_wrangling/stage_lasso_bundle.jl), launcher (execution/submit_ena_lasso.sbatch), comparison tooling and this audit: done; CPU tests in Pkg.test().
• Real-bundle validation and the 1-h GPU integration: done (sections 1b, 1c). The 24-h baseline and its SAM/ARM comparison: queued/pending; see the status file. The Covert benchmark runs are not relabelled as LASSO.