Aerosol configuration audit of the ENA-covert P3-aer2 runs

Runs audited (Covert-public-bin development benchmark, source commit 3330f3c, A100, 06–12 UTC 18 July 2017, 256×256×192 at 35 m): p3_aer2_dt0.5_job168 and p3_aer2_dt0.25_job166 under /shared/home/greg/breezelab-runs/docs-20261007-8dc824d/production/. Evidence: hourly-mean profiles of nᵃ, nᶜˡ, qᶜˡ, cloudy fraction and the slices, read on a CPU node by analysis/ena_aerosol_audit.jl; figures and the numeric summary in runs/ena_aerosol_audit/ (aerosol_profiles.png, slices_*.png, aerosol_audit.toml). Code paths: src/case_setup.jl (lasso_aerosol_modes, build_microphysics, the p3_initialization = :condensate_free branch, the DiagnosticCCNProjection callback), src/large_scale_forcings.jl (DiagnosticCCNProjection, AerosolReplenishment), Breeze 5264d3c PredictedParticleProperties/aerosol_activation.jl and cloud_droplet_activation_rates.jl, and the LASSO SAM MICRO_HUJISBM (12d0244).

Verdict

The aerosol fields are not corrupted (no NaN, collapse or blow-up; nᵃ + nᶜˡ is held exactly at the initial total by the diagnostic-CCN projection; the two time steps agree to three digits), but they are not physically meaningful for this case:

1. Wrong aerosol for the case. The member applies the LASSO-ENA "aer2" prescription (276 + 281 = 557 cm⁻³ at the surface) to the Covert et al. (2022) case, whose control droplet number is the observed Nc = 75 cm⁻³ ("based on airborne in situ measurements of the 18 July 2017 case"; sensitivities 50 and 100 cm⁻³). The P3-aer2 member of the Covert benchmark is therefore a LASSO-aerosol sensitivity, not the Covert case; the case-consistent P3 member is p3_n75.
2. Over-activation. In cloud, nᶜˡ ≈ 3.3–3.6×10⁸ kg⁻¹ (≈ 350–400 cm⁻³, 60–75 % of all aerosol, including part of the Aitken mode) throughout the six hours, 5× the observed 75 cm⁻³, and above the ≈ 270 cm⁻³ the SBM itself can activate from aer2 (its ss_max = 0.3 % cap). Breeze's Morrison–Grabowski activation has no supersaturation cap and is a one-way relaxation (ncnuc = max(0, N_act(S_grid) − nᶜˡ)/τ_act, τ_act = 1 s) toward the equilibrium count at the grid-cell supersaturation, so every overshoot ratchets nᶜˡ up. The largest overshoot is the initial burst: with p3_initialization = :condensate_free the cloud layer starts with ≈ 0.55 g kg⁻¹ of excess vapor (S ≈ 6 %), which activates ≈ 390×10⁶ kg⁻¹ (≈ 430 cm⁻³) in the first seconds (first-hour mean peak, both runs). Later, the cloud-base/top supersaturation spikes keep nᶜˡ near 330–360×10⁶ kg⁻¹.
3. Reservoir semantics. nᵃ is initialized everywhere (free troposphere included) at the constant mixing ratio 4.618×10⁸ kg⁻¹ (SAM does the same: FCCN0 = FCCNR·ρ(k)/ρ(1)), stays exactly constant outside cloud, and is depleted to 140–200×10⁶ kg⁻¹ in the cloud layer. Because the projection rewrites ρnᵃ ← ρ n₀ − ρnᶜˡ − ρnʳ every step, the reservoir is a pure diagnostic of the droplet number: activation's own depletion and the advection of ρnᵃ have no lasting effect (no double counting, no drift), and CCN are "regenerated" wherever droplets evaporate or coalesce, as in the SBM's diagCCN. This is the intended emulation, but it also means the reservoir cannot limit the ratchet.

So the symptom Greg saw ("aerosol fields wrong") is real and expected from the configuration: a 557 cm⁻³ LASSO aerosol applied to a 75 cm⁻³ case, activated by a scheme without the SBM's supersaturation cap from a supersaturated condensate-free start.

Evidence

runs/ena_aerosol_audit/aerosol_audit.toml (both runs; hour k = mean over hour k−1…k):

job166 (Δt = 0.25 s) differs by < 3 % in every entry. Initial total n₀ = 4.618×10⁸ kg⁻¹ = 557 cm⁻³ / 1.206 kg m⁻³ (surface reference density), recorded in provenance as n₁ + n₂. Time series: mean LWP 155 g m⁻² (job168) / 165 g m⁻² (job166), final 190 / 210 g m⁻², cloud fraction 1, rain 2–3×10⁻⁴ mm day⁻¹ (drizzle suppressed, consistent with 350–400 cm⁻³ droplets; the MWRRET 06–12 UTC mean is 168 ± 13 g m⁻² and the one-moment run gives 100 g m⁻²).

Units check: LASSO quotes cm⁻³; lasso_aerosol_modes converts with the surface reference density to kg⁻¹ (1e6 N/ρ₀), which is SAM's constant-mixing-ratio initialization, and Breeze's AerosolMode.number_mixing_ratio is per kg of air and set! writes ρnᵃ = ρ Σ modes everywhere. Modal parameters (r = 0.018 / 0.066 μm, σ = 1.53 / 1.78) match the LASSO methodology table 6. Chemistry: density 1790 kg m⁻³, M = 0.115 kg mol⁻¹, ν = 3 match micro_prm.f90 (RO_SOLUTE = 1.79, mwaero = 115, ions = 3).

Specific misconfigurations (campaign source 3330f3c)

Not a misconfiguration: CloudDroplets(number_concentration = 75e6) passed alongside AerosolActivation only sets the Liu–Daum shape parameter and the prescribed-number fallback, which the prognostic path does not use (process_rates.jl L1055–1059); AerosolReplenishment was not active (aerosol_replenishment = diagnostic_ccn).

Fix on this branch

• lasso_aerosol_modes(; mass_fraction_soluble = 1, maximum_supersaturation = nothing): explicit full solubility (fixes #1) and an optional SBM-cap emulation that scales each mode's number to the fraction activatable below maximum_supersaturation (activated_fraction, evaluated with Breeze's own activated_number): for aer2 at 0.3 % that is ≈ 8 % of mode 1 and ≈ 89 % of mode 2, i.e. ≈ 270 cm⁻³ activatable in total (fixes #2 as an explicit, recorded approximation: the lognormal shape is kept, so activation at low supersaturation is slightly underestimated). Exposed as aerosol_supersaturation_cap in build_case/ena_simulation, --aerosol_ss_cap in the CLI, recorded in provenance with the activatable fractions.
• #3 is left as the protocol default (the SBM also starts condensate-free), but with the cap the initial burst cannot exceed the SBM's activatable total; the alternative p3_initialization = :equilibrium, initial_droplet_number = 75e6 remains available.
• #4 is a labelling matter: the Covert benchmark's case-consistent P3 member is p3_n75; p3_aer2 is documented here as a LASSO-aerosol sensitivity.
• A proper fix of #2 belongs in Breeze (AerosolActivation(; maximum_supersaturation) applied inside activated_number), which would also bound the in-cloud ratchet.

Relaunch (plan: the affected configuration p3_aer2, Δt = 0.5 s, with the cap and full solubility, on a physically idle A100 under runs/): see plans/status/ENA-LASSO.md for the job ID and output path once submitted.