TurbulenceConvectionUtils

ModelEvaluator

A structure containing the information required to perform a forward model evaluation at a given parameter vector.

Fields

• param_cons::Vector{FT} where FT<:Real: Parameter vector in constrained (i.e. physical) space

• param_names::Vector{String}: Parameter names associated with parameter vector

• param_map::CalibrateEDMF.HelperFuncs.ParameterMap: A mapping operator to define relations between parameters. See ParameterMap for details.

• ref_models::Vector{CalibrateEDMF.ReferenceModels.ReferenceModel}: Vector of reference models

• ref_stats::CalibrateEDMF.ReferenceStats.ReferenceStatistics: Reference statistics for the inverse problem

Constructors

ModelEvaluator(
;
    param_cons,
    param_names,
    param_map,
    ref_models,
    ref_stats
)

defined at /home/runner/work/CalibrateEDMF.jl/CalibrateEDMF.jl/src/TurbulenceConvectionUtils.jl:51.

ModelEvaluator(param_cons, param_names, param_map, RM, RS)

defined at /home/runner/work/CalibrateEDMF.jl/CalibrateEDMF.jl/src/TurbulenceConvectionUtils.jl:64.

run_SCM(
    u::Vector{FT},
    u_names::Vector{String},
    param_map::ParameterMap,
    RM::Vector{ReferenceModel},
    RS::ReferenceStatistics;
    error_check::Bool = false,
    failure_handler = "high_loss",
) where {FT <: Real}
run_SCM(
    ME::ModelEvaluator;
    error_check::Bool = false,
    failure_handler = "high_loss",
) where {FT <: Real}

Run the single-column model (SCM) using a set of parameters and return the value of outputs defined in y_names, possibly after normalization and projection onto lower dimensional space using PCA.

The function also outputs a boolean diagnosing whether the requested SCM simulation failed.

Inputs:

• u :: Values of parameters to be used in simulations.
• u_names :: SCM names for parameters u.
• param_map :: A mapping operator to define relations between parameters. See ParameterMap for details.
• RM :: Vector of ReferenceModels
• RS :: reference statistics for simulation
• error_check :: Returns as an additional argument whether the SCM call errored.
• failure_handler :: Method used to handle failed simulations.

Outputs:

• sim_dirs :: Vector of simulation output directories
• g_scm :: Vector of model evaluations concatenated for all flow configurations.
• g_scm_pca :: Projection of g_scm onto principal subspace spanned by eigenvectors.
• model_error :: Whether the simulation errored with the requested configuration.

run_reference_SCM(m::ReferenceModel; overwrite::Bool = false, run_single_timestep = true)

Run the single-column model (SCM) for a reference model object using default parameters, and write the output to file.

Inputs:

• m :: A ReferenceModel.
• overwrite :: if true, run TC.jl and overwrite existing simulation files.
• run_single_timestep :: if true, run only one time step.

eval_single_ref_model(
    m_index::IT,
    m::ReferenceModel,
    RS::ReferenceStatistics,
    u::Vector{FT},
    u_names::Vector{String},
    param_map::ParameterMap,
) where {FT <: Real, IT <: Int}

Runs the single-column model (SCM) under a single configuration (i.e., ReferenceModel) using a set of parameters u, and returns the forward model evaluation in both the original and the latent PCA space.

Inputs:

• m_index :: The index of the ReferenceModel within the overarching ref_models vector used to construct the ReferenceStatistics.
• m :: A ReferenceModel.
• RS :: reference statistics for simulation
• u :: Values of parameters to be used in simulations.
• u_names :: SCM names for parameters u.
• param_map :: A mapping operator to define relations between parameters. See ParameterMap for details.

Outputs:

• sim_dir :: Simulation output directory.
• g_scm :: Forward model evaluation in original output space.
• g_scm_pca :: Projection of g_scm onto principal subspace spanned by eigenvectors.
• model_error :: Whether the simulation errored with the requested configuration.

run_SCM_handler(
    m::ReferenceModel,
    tmpdir::String,
    u::Array{FT, 1},
    u_names::Array{String, 1},
    param_map::ParameterMap,
) where {FT<:AbstractFloat}

Run a case using a set of parameters u_names with values u, and return directory pointing to where data is stored for simulation run.

Inputs:

• m :: Reference model
• tmpdir :: Temporary directory to store simulation results in
• u :: Values of parameters to be used in simulations.
• u_names :: SCM names for parameters u.
• param_map :: A mapping operator to define relations between parameters. See ParameterMap for details.

Outputs:

• output_dir :: directory containing output data from the SCM run.
• model_error :: Boolean specifying whether the simulation failed.

run_SCM_handler(
    case_name::String,
    out_dir::String;
    u::Vector{FT},
    u_names::Vector{String},
    param_map::ParameterMap,
    namelist::Dict,
    uuid::String = "01",
    les::Union{NamedTuple, String} = nothing,
)

Run a TurbulenceConvection.jl case and return directory pointing to where data is stored for simulation run.

Inputs:

• case_name :: case name
• out_dir :: Directory to store simulation results in.

Optional Inputs:

• u :: Values of parameters to be used in simulations.
• u_names :: SCM names for parameters u.
• param_map :: A mapping operator to define relations between parameters. See ParameterMap for details.
• namelist :: namelist to use for simulation.
• uuid :: uuid of SCM run
• les :: path to LES stats file, or NamedTuple with keywords {forcingmodel, month, experiment, cfsitenumber} needed to specify path.

Outputs:

• output_dir :: directory containing output data from the SCM run.
• model_error :: Boolean specifying whether the simulation failed.

create_parameter_vectors(u_names, u, param_map, namelist)

Given vector of parameter names and corresponding values, combine any vector components into single parameter vectors for input into SCM.

Inputs:

• u_names :: SCM names for parameters u, which may contain vector components.
• u :: Values of parameters to be used in simulations, which may contain vector components.
• param_map :: A mapping to a reduced parameter set. See ParameterMap for details.
• namelist :: The parameter namelist for TurbulenceConvection.jl

Outputs:

• u_names_out :: SCM names for parameters u.
• u_out :: Values of parameters to be used in simulations.

generate_scm_input(model_evaluators, iteration, outdir_path = pwd(), batch_indices = nothing)

Writes to file a set of ModelEvaluator used to initialize SCM evaluations at different parameter vectors, as well as their assigned version, which is on the form i<iteration>_e<ensemble_index>.

parse_version_inds(version)

Given version = "ix_ey", return the iteration index x and ensemble index y as integers.

precondition(
    param::Vector{FT},
    priors,
    param_map::ParameterMap,
    ref_models::Vector{ReferenceModel},
    ref_stats::ReferenceStatistics;
    counter::Integer = 0,
    max_counter::Integer = 10,
) where {FT <: Real}

Substitute parameter vector param by a parameter vector drawn from the same prior, conditioned on the forward model being stable.

Inputs:

• param :: A parameter vector that may possibly result in unstable forward model evaluations (in unconstrained space).
• priors :: Priors from which the parameters were drawn.
• param_map :: A mapping to a reduced parameter set. See ParameterMap for details.
• ref_models :: Vector of ReferenceModels to check stability for.
• ref_stats :: ReferenceStatistics of the ReferenceModels.
• counter :: Accumulator tracking number of recursive calls to preconditioner.
• max_counter :: Maximum number of recursive calls to the preconditioner.

Outputs:

• new_param :: A new parameter vector drawn from the prior, conditioned on simulations being stable (in unconstrained space).

precondition(ME::ModelEvaluator, priors)

Substitute the parameter vector of a ModelEvaluator by another one drawn from the given priors, conditioned on the forward model being stable.

Inputs:

• ME :: A ModelEvaluator.
• priors :: Priors from which the parameters were drawn.

Outputs:

• A preconditioned ModelEvaluator.

get_gcm_les_uuid(cfsite_number; [forcing_model::String, month, experiment])

Generate unique and self-describing uuid given information about a GCM-driven LES simulation from [1].

Examples

julia> get_gcm_les_uuid(1; forcing_model = "HadGEM2-A", month = 7, experiment = "amip")
"1_HadGEM2-A_07_amip"

save_tc_data(cases, outdir_path, sim_dirs, version, suffix)

Save full TC.jl output in <results_folder>/timeseries.<suffix>/iter_<iteration>/Output.<case>.<case_id>_<ens_i>.

Behavior of this function is specified in the output config, i.e. config["output"]. If the flag save_tc_output is set to true, TC.jl output is saved, and if additionally a list of iterations is specified in save_tc_iterations, only these EKP iterations are saved.

Arguments:

• config :: The calibration config dictionary. To save TC.jl output, set save_tc_output to true in the output config. To only save specific EKP iterations, specify these in a vector in save_tc_iterations in the output config.
• outdir_path :: The results results_folder path
• sim_dirs :: List of (temporary) directories where raw simulation output is initially saved to
• version :: An identifier for the current iteration and ensemble index
• suffix :: Case set identifier; is either "train" or "validation".