SEA STARR — driver inventory and Breeze mapping

SEA STARR (Southeast Atlantic Stratocumulus Transitions with Aerosol-Rain-Radiation interactions) is an LES/SCM intercomparison of a 66-hour Lagrangian stratocumulus-to-cumulus transition under a biomass-burning smoke layer, driven by a composite of GEOS-5/SEVIRI trajectories (Diamond et al., in prep.). Official inputs: Zenodo record 22241697 (DOI 10.5281/zenodo.22241697), pinned in cases/seastarr/inputs.toml. This page records, per plan step 1, what the setup PDF and the DEPHY driver NetCDF actually contain and how each quantity maps onto Breeze. It is an inventory of the inputs and of the implementation choices; it is not evidence that the Breeze CTRL reproduces any other model.

Protocol summary (setup PDF, `SEA_STARR_Model_Setup_Specifications.pdf`)

Item	Specification
Horizontal grid	Δx = Δy = 50 m; 384² (19.2 km) or, if prohibitive, 192² (9.6 km)
Vertical grid	10 m spacing from the surface to 2500 m, then each level 10 % thicker until 6500 m is reached
Time step	1 s
Period	2017-08-15 21:00 UTC → 2017-08-18 15:00 UTC (66 h; the first 3 h are spin-up)
Nudging	only above the domain maximum inversion height (maximum ∂θₗ/∂z) + 100 m, with a Blossey et al. (2013) ramp over the next 200 m; T, qₜ, Nₐ with τ = 1800 s; u, v with τ = 10800 s
Advection	no horizontal advective tendencies for free-tropospheric variables (the nudging accounts for them); vertical advection from the prescribed vertical velocity (`wa` or `wap`)
Geostrophic forcing	on, with the Coriolis parameter from the (single) latitude
Radiation	on, interactive aerosol with single-scatter albedo 0.85 at 550 nm
Aerosol	single lognormal accumulation mode: geometric mean diameter 185 nm, σ = 1.5, κ = 0.2; surface source 70 cm⁻² s⁻¹ (Yamaguchi et al. 2017)
Surface	ocean; time-varying skin temperature (`ts`); no prescribed moisture/wind forcing (bulk fluxes); z₀ = 10⁻⁴ m

Driver files (`SEA_STARR_{CTRL,N100,N030}_SCM_driver.nc`, DEPHY SCM format v1)

Dimensions: time = 67 hourly records from 2017-08-15 21:00 to 2017-08-18 15:00 UTC, t0 = 1, zh = 336 mid-levels stored top-down (76 255 m → 5 m; 10 m spacing for the 250 levels below 2500 m, stretched above), lat = lon = 1 (−11.649° N, −8.336° E), orog = 0. Global attributes: forc_wap = forc_wa = forc_geo = 1, adv_ta = adv_thetal = 0, radiation = "on", surface_forcing_temp = "ts", surface_forcing_moisture = "none", surface_forcing_wind = "none", surface_source_aerosol = "70 cm-2 s-1", nudging_* = 1800 s / 10800 s, zh_nudge_gap = "200 m".

The members differ only in na and na_nud (verified by element-wise comparison of every numeric variable and attribute): CTRL has 100 mg⁻¹ in the boundary layer and a smoke plume of ≈1000–1100 mg⁻¹ in the free troposphere (nudging target up to 1481 mg⁻¹ later); N100 is 100 mg⁻¹ everywhere and at all times; N030 is 30 mg⁻¹ everywhere and at all times.

Variable table

Column "role": IC = initial condition (dimension t0), F = time-varying forcing (dimension time), N = nudging target. "Breeze target" names the quantity in the BreezeLab implementation (src/sea_starr.jl).

Variable	Dims	Units	Role	Interpolation	Breeze target / treatment
`zh`	zh	m	coordinate (mid-levels, top-down)	—	reversed to ascending by the reader; the LES grid is built from the protocol (10 m / 10 % stretch), not from `zh`
`lat`, `lon`	1	deg	geometry	—	Coriolis `FPlane(latitude = -11.649)`; solar position (see departures)
`ps`	t0	Pa	IC	—	`ReferenceState(base_pressure = ps)` at z = 0
`pa`	zh, t0	Pa	IC (diagnostic)	—	not used for the reference state (Breeze integrates hydrostatically from `ps`); recorded. Note `pa(5 m) = 101 659 Pa` versus `ps = 101 984 Pa`, a 325 Pa drop over 5 m, so the driver column is not hydrostatic with `ps` near the surface
`thetal`	zh, t0	K	IC	linear in z	liquid-water potential temperature → `ReferenceState(potential_temperature)` and `set!(model; θ)` (`:LiquidIcePotentialTemperature` formulation)
`ta`	zh, t0	K	IC (diagnostic)	—	"liquid water temperature"; not used (θₗ is the prognostic basis)
`qt`	zh, t0	1 (mass fraction)	IC	linear in z	total water → `ReferenceState(vapor_mass_fraction)` and `set!(model; qᵗ)`; no mixing-ratio conversion (`moisture_basis = :mass_fraction`)
`na`	zh, t0	mg⁻¹ (per mg of air)	IC	linear in z	aerosol number mixing ratio `nᵃ = 10⁶ na` [kg⁻¹]; set as a height-varying initial field of P3's prognostic reservoir `ρnᵃ`
`ua`, `va`	zh, t0	m s⁻¹	IC	linear in z	`set!(model; u, v)`
`ts`, `z0`	t0	K, m	IC	—	initial SST; momentum roughness length for the bulk formulae
`wa`	zh, time	m s⁻¹	F	linear in z, linear in time	large-scale vertical velocity → `LargeScaleVerticalAdvection` (full-field upwind −w ∂ₓϕ) on u, v, θ, qᵛ and every microphysical species, exactly once per field. `wa` is identical at all 67 times (a single profile); ≈ −4.5 mm s⁻¹ at 2500 m
`wap`	zh, time	Pa s⁻¹	F (alternative)	—	not used (`wa` is used)
`pa_force`	zh, time	Pa	F (diagnostic)	—	not used by the anelastic model (fixed reference pressure); recorded
`ug`, `vg`	zh, time	m s⁻¹	F	linear in z and time	`TimeVaryingGeostrophicForcing` (`-f vg`, `+f ug`). `ug ≡ ua_nud`, `vg ≡ va_nud` in all three files
`ta_nud`	zh, time	K	N	—	not used (θₗ is nudged)
`thetal_nud`	zh, time	K	N	linear in z and time	target of the inversion-following mean-profile nudging of θ (τ = 1/`nudging_constant_thetal` = 1800 s)
`qt_nud`	zh, time	1	N	linear in z and time	target of the nudging of the moisture variable (P3: `qᵛ`; above the inversion there is no condensate so qₜ = qᵛ)
`na_nud`	zh, time	mg⁻¹	N	linear in z and time	target of the nudging of `nᵃ` (× 10⁶)
`ua_nud`, `va_nud`	zh, time	m s⁻¹	N	linear in z and time	targets of the wind nudging (τ = 10800 s)
`nudging_constant_*`	zh, time	s⁻¹	N	—	uniform in height and time: 1/1800 s⁻¹ (ta, thetal, qt, na), 1/10800 s⁻¹ (ua, va). The inversion mask is therefore not in the file and must be diagnosed online
`ts_force`	time	K	F	linear in time	SST → shared surface-temperature field (bulk fluxes and radiation); 293.33 → 298.06 K
`ps_force`	time	Pa	F (diagnostic)	—	not used; 101 984 → 101 523 Pa
`o3`	zh, time	mol mol⁻¹	F (radiation)	—	ozone profile for the radiation column; see departures

Does any thermodynamic tendency already include vertical transport or radiation? There are no horizontal advective tendencies in the file (adv_ta = adv_thetal = 0; no tnta_*, tnqt_* variables). The only large-scale thermodynamic forcings are (i) vertical advection by wa, which the LES must apply itself, and (ii) the free-tropospheric nudging, which the PDF states "already accounts for" advection of free-tropospheric variables. Radiation is fully interactive in the LES; none of the driver tendencies contains it.

Initial state

At t₀ the column is a well-mixed layer (θₗ = 290.0 K, qₜ = 9.0 g kg⁻¹, na = 100 mg⁻¹) up to the inversion at 1160 m (maximum ∂θₗ/∂z = 0.064 K m⁻¹, jump of ≈ +10 K), capped by the smoke layer (θₗ increasing ≈ 10 K km⁻¹, qₜ = 2.36 g kg⁻¹, na ≈ 995–1100 mg⁻¹ up to ≈ 3 km, decaying to ≈ 130 mg⁻¹ by 5 km). The boundary layer is saturated above ≈ 850 m (qₛₐₜ(1000 m) ≈ 7.7 g kg⁻¹ < 9 g kg⁻¹), i.e. the initial state contains a stratocumulus deck. The nudging targets' "inversion" (maximum gradient of thetal_nud) jumps between 600 m and 1790 m during the period, which is why the protocol restricts nudging to above the LES's own inversion.

Raw trajectories (`SEA_STARR_Raw_Trajectories.nc`)

40 GEOS-5 trajectories (ind), 67 hourly times, 72 GEOS levels, with SEVIRI cloud retrievals (Nd, CF, LWP, rₑ, cloud-top T/p/z, SZA) and GEOS state/aerosol species. The composite's mean path moves from (−16.4° N, −1.2° E) at t = 0 to (−7.6° N, −14.9° E) at t = 66 h; the driver's single (lat, lon) = (−11.65, −8.34) is the mid-period position (t ≈ 30 h). The mean TS of the trajectories equals ts_force. The SEVIRI retrievals are the observational reference for evaluation; no comparison has been made yet.

Standard variables (`SEA_STARR_Standard_Variables.xlsx`)

Three sheets: statistical output (15 min; profiles and scalars, tier 1/2), 2D output (15 min) and 3D output (1 h). Time axis "seconds since 2017-08-16 00:00:00". The tier-1 statistical set is: pa, ta, qv, qt, hur, ua, va, wa, theta, thetal, tke, hfss, hfls, ps, ts, zi, w2, ql, qr, cl, clt, pr, cwp, lwp, rwp, nt, na, nc, nr, rlh, rsh, rsdt, rsut, rlut, rsutcs, rlutcs; 2D tier 1: prw, zi, pr, cwp, lwp, rwp, optc, optr, zct, zcb, rsdt, rsut, rlut, rsutcs, rlutcs. Number concentrations are per kg (kg-1). Cloud fraction cl uses a 0.01 g kg⁻¹ liquid threshold; clt counts columns with cloud optical thickness > 2. nt is total aerosol (dry + in droplets + in rain); na is dry (unactivated) aerosol.

Implementation (`src/dephy_driver.jl`, `src/sea_starr_aerosol.jl`, `src/sea_starr_forcings.jl`, `src/sea_starr.jl`)

sea_starr(; member, arch, FT, Nx, Ny, ...) reads the driver, builds the protocol grid (sea_starr_vertical_faces: 10 m × 250 to 2500 m, then 10 % growth, last interface clipped to 6500 m, Nz = 288), the anelastic reference state (hydrostatic from ps, θₗ and qₜ of the driver), P3 microphysics with one KappaAerosolMode and a prognostic reservoir, full-field upwind subsidence (LargeScaleVerticalAdvection with the driver wa) on u, v, θ, qᵛ and every microphysical prognostic exactly once, the time-varying geostrophic wind, bulk surface fluxes from ts_force with the driver z0, RRTMGP LW+SW with the driver ozone profile, inversion-following mean-profile nudging of θ, qᵛ, nᵃ (τ = 1800 s) and u, v (τ = 10800 s), the surface aerosol source, evaporation regeneration, an equilibrium initial cloud, JLD2 output (15-min statistics, 1-min time series with the aerosol number budget columns, 15-min 2D fields, hourly 3D fields) and a 3-hourly checkpoint. cases/sea_starr.jl constructs, writes provenance, runs and plots; execution/sea_starr.sbatch is the UUID-pinned launcher; test/sea_starr.jl holds the reference-value tests.

Aerosol processes (plan step 4), and what the pinned Breeze P3 provides:

Process	Pinned Breeze P3	BreezeLab
Activation	M&G2007 erf form with the mode's constant number, only capped by the local reservoir	`aerosol_activation_rate` specialized for `KappaAerosolMode`: target `f(S) (nᶜˡ + nᵃ)` from the κ-Köhler critical supersaturation of the local reservoir; one aerosol consumed per droplet (Breeze)
Surface source	none (`FluxBoundaryCondition` on `ρnᵃ` is ignored by P3's field materialization)	`SurfaceAerosolSource`: 7×10⁵ m⁻² s⁻¹ as a bottom-cell tendency
Regeneration on droplet evaporation	none (`nᶜˡ` kept until `qᶜˡ` is exhausted, then discarded)	`EvaporationRegeneration`: nᶜˡ removed ∝ P3's own diagnosed cloud-evaporation rate and returned to nᵃ (one-step lag)
Collision scavenging	droplet/rain number lost in autoconversion, accretion, self-collection; nothing returned	unchanged (this is the coalescence sink of aerosol)
Rain removal	rain number sediments through the surface	unchanged
Regeneration on rain evaporation	P3 reduces `nʳ`; nothing returned	not implemented (nʳ ≪ nᵃ; labelled)
Interstitial/impaction scavenging	none	not implemented (labelled)
Transport	—	advection, subsidence, nudging on `nᵃ` like every other scalar

The column budget ∫(ρnᵃ + ρnᶜˡ + ρnʳ) dz is written every minute (n_total_column); it changes only through the surface source, nudging, subsidence, coalescence and rain removal.

Implementation choices and departures

The following are either not specified by the PDF/driver or not available in the pinned Breeze revision. Each is listed in case.config.departures and written to the provenance file. A run with them is an exploratory CTRL, not a protocol-complete one.

Nudging ramp. Blossey et al. (2013) could not be retrieved from this cluster (paywall). The implemented ramp is linear in height: weight 0 at zb = max(zᵢ) + 100 m, 1 at zb + 200 m (the driver's zh_nudge_gap). The inversion height is the per-column height of maximum ∂θₗ/∂z (searched below 4 km), and max(zᵢ) is the maximum over all columns ("domain maximum inversion height"). The nudging acts on the horizontal mean so that resolved free-tropospheric eddies are not damped.
Nudged thermodynamic variable. θₗ (the prognostic) is nudged to thetal_nud; ta_nud is not used. The P3 moisture variable is the vapor mass fraction qᵛ, nudged to qt_nud.
Aerosol chemistry. κ enters the activation directly (κ-Köhler critical supersaturation; identical to Morrison & Grabowski's β_act form with β = κ).
Aerosol processes. See the table above (proportional evaporation regeneration; no regeneration from rain evaporation; no interstitial scavenging). ENA's DiagnosticCCNProjection reset is not used.
Radiation. Breeze's RRTMGP has no aerosol optics (single-scatter albedo 0.85 not representable, so the smoke-layer absorption is absent), its column ends at the LES top (6.5 km; no atmosphere above, zero downwelling longwave at the top), and ApparentSolarPosition takes a fixed coordinate (the driver's single lat/lon rather than the moving composite trajectory, which spans ≈ 9° of latitude and 14° of longitude over the period).
Vertical grid top. 38 stretched levels reach 6504 m; the last interface is clipped to exactly 6500 m (Nz = 288).
Upper sponge. SAMSponge over the top 15 % of the domain (z > 5525 m), not in the protocol.
Initial cloud. Warm-phase saturation adjustment of the driver (θₗ, qₜ) gives the initial cloud; in cloudy cells the whole aerosol reservoir is initially activated (what the activation scheme would do at the large initial supersaturation), elsewhere nᵃ is the driver profile.
Surface fluxes. Breeze bulk formulae (Large & Yeager polynomials, Li et al. stability) with z₀ = 10⁻⁴ m and 0.1 m s⁻¹ gustiness; the protocol only prescribes ts and z0.
Time step. Fixed 1 s as specified; the CFL wizard may only shorten it.