Rising Thermal Bubble

In this example, we demonstrate the usage of the ClimateMachine AtmosModel machinery to solve the fluid dynamics of a thermal perturbation in a neutrally stratified background state defined by its uniform potential temperature. We solve a flow in a box configuration - this is representative of a large-eddy simulation. Several versions of the problem setup may be found in literature, but the general idea is to examine the vertical ascent of a thermal bubble (we can interpret these as simple representation of convective updrafts).

Description of experiment

  1. Dry Rising Bubble (circular potential temperature perturbation)
  2. Boundaries Top and Bottom boundaries:
    • Impenetrable(FreeSlip()) - Top and bottom: no momentum flux, no mass flux through walls.
    • Impermeable() - non-porous walls, i.e. no diffusive fluxes through walls.
    Lateral boundaries
    • Laterally periodic
  3. Domain - 2500m (horizontal) x 2500m (horizontal) x 2500m (vertical)
  4. Resolution - 50m effective resolution
  5. Total simulation time - 1000s
  6. Mesh Aspect Ratio (Effective resolution) 1:1
  7. Overrides defaults for
    • CPU Initialisation
    • Time integrator
    • Sources
    • Smagorinsky Coefficient

This experiment setup assumes that you have installed the ClimateMachine according to the instructions on the landing page. We assume the users' familiarity with the conservative form of the equations of motion for a compressible fluid (see the AtmosModel page).

The following topics are covered in this example

  • Package requirements
  • Defining a model subtype for the set of conservation equations
  • Defining the initial conditions
  • Applying source terms
  • Choosing a turbulence model
  • Adding tracers to the model
  • Choosing a time-integrator
  • Choosing diagnostics (output) configurations

The following topics are not covered in this example

  • Defining new boundary conditions
  • Defining new turbulence models
  • Building new time-integrators
  • Adding diagnostic variables (beyond a standard pre-defined list of variables)

Loading code

Before setting up our experiment, we recognize that we need to import some pre-defined functions from other packages. Julia allows us to use existing modules (variable workspaces), or write our own to do so. Complete documentation for the Julia module system can be found here.

We need to use the ClimateMachine module! This imports all functions specific to atmospheric and ocean flow modeling.

using ClimateMachine
using ClimateMachine.Atmos
using ClimateMachine.Orientations
using ClimateMachine.ConfigTypes
using ClimateMachine.Diagnostics
using ClimateMachine.GenericCallbacks
using ClimateMachine.ODESolvers
using Thermodynamics.TemperatureProfiles
using Thermodynamics
using ClimateMachine.TurbulenceClosures
using ClimateMachine.VariableTemplates
[1636503129.036492] [hpc-92-37:14383:0]       ib_verbs.h:84   UCX  ERROR ibv_exp_query_device(mlx5_0) returned 38: No space left on device

In ClimateMachine we use StaticArrays for our variable arrays. We also use the Test package to help with unit tests and continuous integration systems to design sensible tests for our experiment to ensure new / modified blocks of code don't damage the fidelity of the physics. The test defined within this experiment is not a unit test for a specific subcomponent, but ensures time-integration of the defined problem conditions within a reasonable tolerance. Immediately useful macros and functions from this include @test and @testset which will allow us to define the testing parameter sets.

using StaticArrays
using Test
using CLIMAParameters
using CLIMAParameters.Atmos.SubgridScale: C_smag
using CLIMAParameters.Planet: R_d, cp_d, cv_d, MSLP, grav
struct EarthParameterSet <: AbstractEarthParameterSet end
const param_set = EarthParameterSet();

Initial Conditions

This example demonstrates the use of functions defined in the Thermodynamics package to generate the appropriate initial state for our problem.


The following variables are assigned in the initial condition

  • state.ρ = Scalar quantity for initial density profile
  • state.ρu= 3-component vector for initial momentum profile
  • state.energy.ρe= Scalar quantity for initial total-energy profile humidity
  • state.tracers.ρχ = Vector of four tracers (here, for demonstration only; we can interpret these as dye injections for visualization purposes)
function init_risingbubble!(problem, bl, state, aux, localgeo, t)
    (x, y, z) = localgeo.coord

    # Problem float-type
    FT = eltype(state)
    param_set = parameter_set(bl)

    # Unpack constant parameters
    R_gas::FT = R_d(param_set)
    c_p::FT = cp_d(param_set)
    c_v::FT = cv_d(param_set)
    p0::FT = MSLP(param_set)
    _grav::FT = grav(param_set)
    γ::FT = c_p / c_v

    # Define bubble center and background potential temperature
    xc::FT = 5000
    yc::FT = 1000
    zc::FT = 2000
    r = sqrt((x - xc)^2 + (z - zc)^2)
    rc::FT = 2000
    θamplitude::FT = 2

    # This is configured in the reference hydrostatic state
    ref_state = reference_state(bl)
    θ_ref::FT = ref_state.virtual_temperature_profile.T_surface

    # Add the thermal perturbation:
    Δθ::FT = 0
    if r <= rc
        Δθ = θamplitude * (1.0 - r / rc)

    # Compute perturbed thermodynamic state:
    θ = θ_ref + Δθ                                      ## potential temperature
    π_exner = FT(1) - _grav / (c_p * θ) * z             ## exner pressure
    ρ = p0 / (R_gas * θ) * (π_exner)^(c_v / R_gas)      ## density
    T = θ * π_exner
    e_int = internal_energy(param_set, T)
    ts = PhaseDry(param_set, e_int, ρ)
    ρu = SVector(FT(0), FT(0), FT(0))                   ## momentum
    # State (prognostic) variable assignment
    e_kin = FT(0)                                       ## kinetic energy
    e_pot = gravitational_potential(bl, aux)            ## potential energy
    ρe_tot = ρ * total_energy(e_kin, e_pot, ts)         ## total energy

    ρχ = FT(0)                                          ## tracer

    # We inject tracers at the initial condition at some specified z coordinates
    if 500 < z <= 550
        ρχ += FT(0.05)

    # We want 4 tracers
    ntracers = 4

    # Define 4 tracers, (arbitrary scaling for this demo problem)
    ρχ = SVector{ntracers, FT}(ρχ, ρχ / 2, ρχ / 3, ρχ / 4)

    # Assign State Variables
    state.ρ = ρ
    state.ρu = ρu
    state.energy.ρe = ρe_tot
    state.tracers.ρχ = ρχ
init_risingbubble! (generic function with 1 method)

Model Configuration

We define a configuration function to assist in prescribing the physical model. The purpose of this is to populate the ClimateMachine.AtmosLESConfiguration with arguments appropriate to the problem being considered.

function config_risingbubble(
) where {FT}
    # Since we want four tracers, we specify this and include the appropriate
    # diffusivity scaling coefficients (normally these would be physically
    # informed but for this demonstration we use integers corresponding to the
    # tracer index identifier)
    ntracers = 4
    δ_χ = SVector{ntracers, FT}(1, 2, 3, 4)
    # To assemble `AtmosModel` with no tracers, set `tracers = NoTracers()`.

    # The model coefficient for the turbulence closure is defined via the
    # [CLIMAParameters
    # package](https://CliMA.github.io/CLIMAParameters.jl/latest/) A reference
    # state for the linearisation step is also defined.
    T_surface = FT(300)
    T_min_ref = FT(0)
    T_profile = DryAdiabaticProfile{FT}(param_set, T_surface, T_min_ref)
    ref_state = HydrostaticState(T_profile)

    # Here we assemble the `AtmosModel`.
    _C_smag = FT(C_smag(param_set))
    physics = AtmosPhysics{FT}(
        param_set;                                     ## Parameter set corresponding to earth parameters
        ref_state = ref_state,                         ## Reference state
        turbulence = SmagorinskyLilly(_C_smag),        ## Turbulence closure model
        moisture = DryModel(),                         ## Exclude moisture variables
        tracers = NTracers{ntracers, FT}(δ_χ),         ## Tracer model with diffusivity coefficients

    model = AtmosModel{FT}(
        AtmosLESConfigType,                            ## Flow in a box, requires the AtmosLESConfigType
        physics;                                       ## Atmos physics
        init_state_prognostic = init_risingbubble!,    ## Apply the initial condition
        source = (Gravity(),),                         ## Gravity is the only source term here

    # Finally, we pass a `Problem Name` string, the mesh information, and the
    # model type to  the [`AtmosLESConfiguration`] object.
    config = ClimateMachine.AtmosLESConfiguration(
        "DryRisingBubble",       ## Problem title [String]
        N,                       ## Polynomial order [Int]
        resolution,              ## (Δx, Δy, Δz) effective resolution [m]
        xmax,                    ## Domain maximum size [m]
        ymax,                    ## Domain maximum size [m]
        zmax,                    ## Domain maximum size [m]
        param_set,               ## Parameter set.
        init_risingbubble!,      ## Function specifying initial condition
        model = model,           ## Model type
    return config
config_risingbubble (generic function with 1 method)

Keywords are used to specify some arguments (see appropriate source files).


Here we define the diagnostic configuration specific to this problem.

function config_diagnostics(driver_config)
    interval = "10000steps"
    dgngrp = setup_atmos_default_diagnostics(
    return ClimateMachine.DiagnosticsConfiguration([dgngrp])

function main()
    # These are essentially arguments passed to the
    # [`config_risingbubble`](@ref config-helper) function.  For type
    # consistency we explicitly define the problem floating-precision.
    FT = Float64
    # We need to specify the polynomial order for the DG discretization,
    # effective resolution, simulation end-time, the domain bounds, and the
    # courant-number for the time-integrator. Note how the time-integration
    # components `solver_config` are distinct from the spatial / model
    # components in `driver_config`. `init_on_cpu` is a helper keyword argument
    # that forces problem initialization on CPU (thereby allowing the use of
    # random seeds, spline interpolants and other special functions at the
    # initialization step.)
    N = 4
    Δh = FT(125)
    Δv = FT(125)
    resolution = (Δh, Δh, Δv)
    xmax = FT(10000)
    ymax = FT(500)
    zmax = FT(10000)
    t0 = FT(0)
    timeend = FT(100)
    # For full simulation set `timeend = 1000`

    # Use up to 1.7 if ode_solver is the single rate LSRK144.
    CFL = FT(1.7)

    # Assign configurations so they can be passed to the `invoke!` function
    driver_config = config_risingbubble(FT, N, resolution, xmax, ymax, zmax)

    # Choose an Explicit Single-rate Solver from the existing [`ODESolvers`](@ref ClimateMachine.ODESolvers) options.
    # Apply the outer constructor to define the `ode_solver`.
    # The 1D-IMEX method is less appropriate for the problem given the current
    # mesh aspect ratio (1:1).
    ode_solver_type = ClimateMachine.ExplicitSolverType(
        solver_method = LSRK144NiegemannDiehlBusch,
    # If the user prefers a multi-rate explicit time integrator,
    # the ode_solver above can be replaced with
    # `ode_solver = ClimateMachine.MultirateSolverType(
    #    fast_model = AtmosAcousticGravityLinearModel,
    #    slow_method = LSRK144NiegemannDiehlBusch,
    #    fast_method = LSRK144NiegemannDiehlBusch,
    #    timestep_ratio = 10,
    # )`
    # See [ODESolvers](@ref ODESolvers-docs) for all of the available solvers.

    solver_config = ClimateMachine.SolverConfiguration(
        ode_solver_type = ode_solver_type,
        init_on_cpu = true,
        Courant_number = CFL,
    dgn_config = config_diagnostics(driver_config)

    # Invoke solver (calls `solve!` function for time-integrator), pass the driver,
    # solver and diagnostic config information.
    result = ClimateMachine.invoke!(
        diagnostics_config = dgn_config,
        user_callbacks = (),
        check_euclidean_distance = true,

    # Check that the solution norm is reasonable.
    @test isapprox(result, FT(1); atol = 1.5e-3)
main (generic function with 1 method)

The experiment definition is now complete. Time to run it.

Running the file

julia --project tutorials/Atmos/risingbubble.jl will run the experiment from the main ClimateMachine.jl directory, with diagnostics output at the intervals specified in config_diagnostics. You can also prescribe command line arguments for simulation update and output specifications. For rapid turnaround, we recommend that you run this experiment on a GPU.

VTK output can be controlled via command line by setting parse_clargs=true in the ClimateMachine.init arguments, and then using --vtk=<interval>.

Output Visualisation

See the ClimateMachine API interface documentation for generating output.

are two commonly used programs for .vtu files.

For NetCDF or JLD2 diagnostics you may use any of the following tools: Julia's NCDatasets and JLD2 packages with a suitable

or the known and quick NCDF visualization tool: ncview plotting program.

Test Passed

This page was generated using Literate.jl.