Add 5-bin sea salt aerosol tracer by ziglaser · Pull Request #4713 · CliMA/ClimaAtmos.jl

ziglaser · 2026-07-19T08:57:17Z

PR Dependency

Ocean fraction wiring requires ClimaCoupler.jl #2049, AMIP runs or outdated Coupler runs currently default to all 100% ocean coverage

New feature

Sea salt aerosol emitted via Gong 2003 wind-speed dependent scheme (with toggled sea surface temperature dependence as per Jaegle 2011) in proportion to ocean_fraction of cells
Emission fluxes are precomputed per bin in set_sea_salt_emission_flux! (aerosols/sea_salt.jl) during set_explicit_precomputed_quantities!, using a 10 m wind extrapolated from the lowest model level via MOST, and applied as a bottom boundary flux by sea_salt_emission_tendency! in surface_flux_tendency!
Aerosols are treated as $\rho \chi$ standard passive trated into horizontal advection viahorizontal_tracer_advection_tendency! and vertical transport via vertical_transport in explicit_vertical_advection_tendency! (both in prognostic_equations/advection.jl)
Sea salt aerosol is removed through a tendency enacting a 0.55 day half life for all bins and vertical levels (sea_salt_deposition_tendency!,
called from additional_tendency!), a placeholder until weional
Created diagnostics in tracer_diagnostics.jl for individual aerosol size bins (mmrsslt01–mmrsslt05, and analogous mmrdst0X for dust); new diagnostics for SSA emissions: total flux emiss and per-bin emisslt01–emisslt05

Config changes

New optional interactive_aerosols config option, takes list of (currently sea salt aerosol) i.e. ["SSLT01", ...], defaults empty. Stored on AtmosModel as a Val of names so tracer sets are inferrable at compile time.

Wiring changes

Ocean fraction is passed from ClimaCoupler.jl via Intcluded in AtmosSimulation` struct
Prescribed and interactive aerosols stored in tracer_cache in tracer_cache.jl
Interactive aerosol bins are added to the PrognosticEDMFX updraft state as specific tracers (interactive_aerosol_sgs_variables in prognostic_variables.jl), zero-initialized following wiring of ClimaAtmos #4629
radiation/update_inputs.jl: update_aerosol_concentrations! split into update_prescribed_aerosol_concentrations! (evaluates MERRA-2 time-varying inputs) and update_radiation_aerosol_concentrations! (packages per-layer column mass densities for RRTMGP). For each species, the interactive (prognostic) mass is used when active, falling back to the prescribed climatology otherwise, so interactive sea salt feeds aerosol radiation directly. The RRTMGP↔MERRA-2 name mapping is hoisted into a shared RRTMGP_AEROSOL_NAMES_PAIR constant.
radiation/radiation.jl: sea-salt bin radii passed to RRTMGP now come from ClimaParams (MERRA2_seasalt_aerosol_bin0X_radius). Aerosol check in radiation_model_cache is derived from RRTMGP_AEROSOL_NAMES_PAIR rather than a duplicated name list, and now accepts interactive as well as prescribed aerosol names.

Renames

Uses of "aerosol_[VAR]" renamed to the following scheme: "prescribed_aerosol_[VAR]" when only applicable to MERRA-2 prescribed aerosols, "interactive_aerosol_[VAR]" only applicable to interactive aerosols, "aerosol_[VAR]" when applicable to both. This includes the AtmosSimulation kwarg aerosol_names -> prescribed_aerosol_names.

Future Work

More thoughtful treatment of 10m wind estimate with MOST — current formulation assumes bottom level center wind value is point value, not layer averaged.
Replace Gong 2003 parameterization with more physically supported parameterization (likely Deike 2022)
Replace half-life decay with wet and dry deposition tendencies
Implementation of hygroscopic growth: wet/dry deposition and optics are dependent on RH-influenced particle size
Interactions with cloud microphysics

I have read and checked the items on the review checklist.

github-actions · 2026-07-19T09:20:22Z

📖 Docs preview for this PR: https://clima.github.io/ClimaAtmos.jl/previews/PR4713/

szy21 · 2026-07-20T23:02:23Z

+            center_dust1_radius = 0.73,
            center_dust2_radius = 1.4,
            center_dust3_radius = 2.4,
            center_dust4_radius = 4.5,
            center_dust5_radius = 8,
-            center_ss1_radius = 0.55,
-            center_ss2_radius = 1.4,
-            center_ss3_radius = 2.4,
-            center_ss4_radius = 4.5,
-            center_ss5_radius = 8,
+            center_ss1_radius = aero_params.SSLT01_radius * FT(1e6),
+            center_ss2_radius = aero_params.SSLT02_radius * FT(1e6),
+            center_ss3_radius = aero_params.SSLT03_radius * FT(1e6),
+            center_ss4_radius = aero_params.SSLT04_radius * FT(1e6),
+            center_ss5_radius = aero_params.SSLT05_radius * FT(1e6),


Let's not change the radius here. It depends on how RRTMGP creates the lookup table for aerosol optics, and I need to double check it. But I think what we have before is correct.

trontrytel · 2026-07-20T23:24:06Z

    precomputed = precomputed_quantities(Y, atmos)
+    tracers = tracer_cache(
+        Y,
+        prescribed_aerosol_names,


Do we want to support prescribed and prognostic aerosol at the same time? To allow having prognostic sea salt but prescribed everything else? This is appealing for testing but may result in complicated code. The alternative would be to allow for either fully prescribed or fully interactive.

trontrytel · 2026-07-20T23:26:28Z

+        prescribed_aerosol_names,
+        time_varying_trace_gas_names,
+        start_date,
+        _interactive_aerosol_names(atmos.interactive_aerosols),


I would add it right after prescribed aerosol names, drop the underscore and name in prognostic aerosol names. "Interactive" is a new way to say prognostic and will be confusing if we start to use it interchangeably in the model.

trontrytel · 2026-07-20T23:26:45Z

 prescribed_aerosols:
  help: "Which aerosols to add. List of keys from the data file (e.g., CB1, CB2)."
  value: []
+interactive_aerosols:


prognostic_aerosols

trontrytel · 2026-07-20T23:28:04Z

    ORGW,
    RAD,
    TRAC,
+    OFRAC,


Not for this PR, but we should re-think this pattern of adding new fields to this struct whenever we add new functionality to the model

trontrytel · 2026-07-20T23:28:31Z

+        start_date,
+        _interactive_aerosol_names(atmos.interactive_aerosols),
+    )
+    ocean_fraction = ones(axes(Fields.level(Y.f, Fields.half))) # Populated via Interfacer.update_field! before first timestep


What to do in purely Atmos runs?

Also, do we really have to cache it?

trontrytel · 2026-07-20T23:30:08Z


    set_precomputed_quantities!(Y, precomputing_arguments, FT(0))

+    radiation_aerosol_names = (


Is this used anywhere else? Maybe we can add both categories directly to radiation_args without specifying another entity

trontrytel · 2026-07-20T23:31:18Z

    non_orographic_gravity_wave = non_orographic_gravity_wave_cache(Y, atmos)
    orographic_gravity_wave = orographic_gravity_wave_cache(Y, atmos)
    radiation = radiation_model_cache(Y, atmos, radiation_args...)
-    tracers = tracer_cache(Y, aerosol_names, time_varying_trace_gas_names, start_date)


We don't need the trace_gas_names here anyomre?

trontrytel · 2026-07-20T23:31:58Z

        SurfaceConditions.update_surface_conditions!(Y, p, FT(t))
    end

+    set_sea_salt_emission_flux!(Y, p)


Do we want to always apply it, or only under some user defined option?

trontrytel · 2026-07-20T23:33:27Z

            aod_sw_scattering = NaN,
            # assuming fixed aerosol radius
-            center_dust1_radius = 0.55,
+            center_dust1_radius = 0.73,


Why this needs to be changed?

trontrytel · 2026-07-20T23:34:07Z

+            center_ss2_radius = aero_params.SSLT02_radius * FT(1e6),
+            center_ss3_radius = aero_params.SSLT03_radius * FT(1e6),
+            center_ss4_radius = aero_params.SSLT04_radius * FT(1e6),
+            center_ss5_radius = aero_params.SSLT05_radius * FT(1e6),


We should not change those numbers by accident. They are consistent with RRTMGP assumptions, I think

trontrytel · 2026-07-20T23:56:59Z

    turbconv_model::PrognosticEDMFX,
    microphysics_model,
    chemistry_model,
+    interactive_aerosols,


aerosol_model ?

trontrytel · 2026-07-20T23:58:24Z

    # tracer_diagnostics.jl
    "loadss", "mmrbcpi", "mmrbcpo", "mmrdust", "mmrocpi", "mmrocpo",
    "mmrso4", "mmrss", "o3",
+    "mmrdst01", "mmrdst02", "mmrdst03", "mmrdst04", "mmrdst05",


Do we document anywhere what thise acronyms stand for?

trontrytel · 2026-07-20T23:59:04Z

    # at the cost of water vapor.
    tracer_nonnegativity_vapor_tendency!(Yₜ, Y, p, t, microphysics_model)

+    sea_salt_deposition_tendency!(Yₜ, Y, p, t)


Should we dispatch on it based on the aerosol_model type?

trontrytel · 2026-07-20T23:59:37Z

        end
    end
+
+    sea_salt_emission_tendency!(Yₜ, Y, p, t)


Consider dispatching based on aerosol model type?

trontrytel · 2026-07-21T00:00:20Z

            # assuming one updraft
-            ᶜχʲₜ = MatrixFields.get_field(Yₜ.c, get_χʲ_name_from_ρχ_name(ρχ_name))
-            @. ᶜχʲₜ -= specific(btt, p.precomputed.ᶜρʲs.:(1))
+            ᶜχʲ_name = get_χʲ_name_from_ρχ_name(ρχ_name)


Could you explain why this needs the additional if now?

trontrytel · 2026-07-21T00:00:47Z

 function get_tracers(parsed_args)
-    aerosol_names = Tuple(parsed_args["prescribed_aerosols"])
+    prescribed_aerosol_names = Tuple(parsed_args["prescribed_aerosols"])
+    interactive_aerosol_names = Tuple(parsed_args["interactive_aerosols"])


prognostic_aerosol_names ?

trontrytel · 2026-07-21T00:13:34Z

+species is interactive; otherwise the prescribed (climatology) concentration
+is used.
+"""
+function update_radiation_aerosol_concentrations!((; u, p, t)::I) where {I}


I think the naming is a bit confusing here. We have the update_prescribed_aerosol_concentrations! followed by update_radiation_aerosol_concentrations! To me this reads as prescribed aerosol do not affect radiation. There is a separate group of aerosol that do affect radiation that gets updated later.

trontrytel · 2026-07-21T00:16:53Z

+
+# BIN DEFNITIONS
+const SEA_SALT_BIN_BOUNDS = (
+    (0.03, 0.1),   # μm, rep size 0.079 μm, SSLT01,


Are we ok with hard coding those, or could this be a parameter?

trontrytel · 2026-07-21T00:17:59Z

+        UF.dimensionless_profile(uf_params, z_target, ζ_target, z₀, UF.MomentumTransport())
+    F_anchor =
+        UF.dimensionless_profile(uf_params, z_anchor, ζ_anchor, z₀, UF.MomentumTransport())
+    return max(u_anchor * F_target / F_anchor, FT(0))


Is it possible for F_anchor to be zero?

trontrytel · 2026-07-21T00:18:26Z

+Precomputable radius-dependent part of the Gong (2003) integrand
+"""
+function _gong2003_r_integrand(r, theta)
+    A = 4.7 * (1 + theta * r)^(-0.017 * r^(-1.44))


Those should not be hardcoded. We should put them in ClimaParams

trontrytel · 2026-07-21T00:19:01Z

+    return 1.373 * r^(-A) * (1 + 0.057 * r^3.45) * 10^(1.607 * exp(-B^2))
+end
+
+function _gong_bin_integral(r_lo, r_hi, theta = 30.0, N = 512)


What is theta and N?

trontrytel · 2026-07-21T00:19:59Z

+function sea_salt_emission_flux(u_10, T_sfc, bin_index; SST_adj = false)
+    FT = typeof(u_10)
+    bin_integral = FT(SEA_SALT_BIN_R_INTEGRALS[bin_index])
+    number_flux = bin_integral * abs(u_10)^FT(3.41)


Those parameters should be part of Clima Params

trontrytel · 2026-07-21T00:20:18Z

+)
+
+"""
+    sea_salt_emission_flux(u_10, T_sfc, bin_index)


Could you add a short entry to ClimaAtmos documentation?

trontrytel · 2026-07-21T00:20:47Z

+
+const SEA_SALT_BIN_R_INTEGRALS = ntuple(
+    i -> _gong_bin_integral(SEA_SALT_BIN_BOUNDS[i]...),
+    Val(5),


Do we ever want to support more than 5 bins?

trontrytel · 2026-07-21T00:21:53Z

+
+Called during `set_explicit_precomputed_quantities!'. `sea_salt_emission_tendency!` reads the cached values rather than recomputing.
+"""
+function set_sea_salt_emission_flux!(Y, p)


You could dispatch based on the aerosol_model and avoid the isempty. I think its a cleaner pattern. Thats what we do in most other tendencies

trontrytel · 2026-07-21T00:22:58Z

+    surface_fluxes_params = CAP.surface_fluxes_params(p.params)
+    uf_params = SFP.uf_params(surface_fluxes_params)
+    κ = SFP.von_karman_const(surface_fluxes_params)
+    ustar_p = parent(sfc_conditions.ustar)


Why do we need the parent here?

trontrytel · 2026-07-21T00:23:42Z

+    roughness_spec = SF.COARE3RoughnessParams{FT}()
+
+    z0_eff = p.scratch.temp_field_level
+    z₀_p = parent(z0_eff)


I don't think the parent is needed here?

trontrytel · 2026-07-21T00:24:01Z

+
+    z0_eff = p.scratch.temp_field_level
+    z₀_p = parent(z0_eff)
+    z₀_p .= SF.momentum_roughness.(roughness_spec, ustar_p, surface_fluxes_params, nothing)


Maybe

Suggested change

z₀_p .= SF.momentum_roughness.(roughness_spec, ustar_p, surface_fluxes_params, nothing)

@. z₀_p = SF.momentum_roughness(roughness_spec, ustar_p, surface_fluxes_params, nothing)

trontrytel · 2026-07-21T00:24:30Z

+    z₀_p = parent(z0_eff)
+    z₀_p .= SF.momentum_roughness.(roughness_spec, ustar_p, surface_fluxes_params, nothing)
+
+    z_c1_p = parent(Fields.level(Fields.coordinate_field(axes(Y.c)).z, 1))


Same as above

trontrytel · 2026-07-21T00:25:25Z

+    )
+    mass_per_particle = ntuple(
+        i -> FT(4 / 3 * π * bin_radii[i]^3 * aero_params.seasalt_density),
+        Val(5),


I think its safer to make the number of bins a parameter. If we ever change it it'll be hard to trace all the places it is being hardcoded

trontrytel · 2026-07-21T00:25:42Z

+        aero_params.SSLT05_radius,
+    )
+    mass_per_particle = ntuple(
+        i -> FT(4 / 3 * π * bin_radii[i]^3 * aero_params.seasalt_density),


I suspect we don't need FT here

Add 5-bin sea salt aerosol tracer

528e1b8

ziglaser requested review from tapios and trontrytel July 19, 2026 08:57

ziglaser mentioned this pull request Jul 19, 2026

SSA configs and adding ocean_fraction to Interfacer update CliMA/ClimaCoupler.jl#2049

Draft

szy21 reviewed Jul 20, 2026

View reviewed changes

trontrytel reviewed Jul 20, 2026

View reviewed changes

trontrytel reviewed Jul 21, 2026

View reviewed changes


		set_precomputed_quantities!(Y, precomputing_arguments, FT(0))

		radiation_aerosol_names = (

	z₀_p .= SF.momentum_roughness.(roughness_spec, ustar_p, surface_fluxes_params, nothing)
	@. z₀_p = SF.momentum_roughness(roughness_spec, ustar_p, surface_fluxes_params, nothing)

Uh oh!

Conversation

ziglaser commented Jul 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Dependency

New feature

Config changes

Wiring changes

Renames

Future Work

Uh oh!

github-actions Bot commented Jul 19, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

trontrytel Jul 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

ziglaser commented Jul 19, 2026 •

edited

Loading

trontrytel Jul 21, 2026 •

edited

Loading