distribution

Julia

struct Distribution
Distribution(kind::AbstractString; location, scale, cov, low, high)

Distribution

A parameter’s value drawn from a probability distribution rather than fixed (esm-spec §6.3, schema Distribution). The closed set is normal, lognormal, and uniform.

The three kinds spell their parameters differently on the wire, so the struct stores them by ROLE and the wire names are restored on emit:

A distribution is univariate when its location parameter is a number and multivariate when it is an array, in which case cov gives the full covariance matrix and the parameter’s shape must agree. WHEN the value is drawn is decided by the parameter’s ParameterUpdate : with no update it is sampled once at setup (sampled_parameters); with kind: "wiener" it is resampled every step with √dt scaling (brownian_parameters).

Source

TypeScript

TypeScript does not export this type.

Python

class Distribution
Distribution(kind: Literal['normal', 'lognormal', 'uniform'], mean: float | list[float] | None = None, std: float | list[float] | None = None, mu: float | list[float] | None = None, sigma: float | list[float] | None = None, low: float | list[float] | None = None, high: float | list[float] | None = None, cov: list[list[float]] | None = None)

A parameter’s value drawn from a probability distribution (esm-spec §5.4 / §6.3). The closed set is normal / lognormal / uniform; a binding implements exactly these and rejects anything else.

Univariate when the location parameter (mean / mu / low) is a number, multivariate when it is an array — in which case the parameter’s shape must agree and cov gives the full covariance matrix. The location/spread slots are spelled per kind (mean/std for normal, mu/sigma for lognormal, low/high for uniform) and only the slots belonging to kind are ever populated or emitted.

WHEN the value is drawn is decided by the parameter’s update, and that is the whole difference between the two uses: with NO update it is sampled once at setup (the uncertainty-quantification / ensemble case); with update.kind == "wiener" it is resampled every step with sqrt(dt) scaling (the stochastic-process case that makes the model an SDE).

Source

Rust

pub enum Distribution {
    Normal { mean: DistributionParam, std: Option<DistributionParam>, cov: Option<CovarianceMatrix> },
    Lognormal { mu: DistributionParam, sigma: Option<DistributionParam>, cov: Option<CovarianceMatrix> },
    Uniform { low: DistributionParam, high: DistributionParam },
}

impl Distribution { pub fn kind(&self) -> &'static str }

impl Distribution { pub fn location(&self) -> &DistributionParam }

impl Distribution { pub fn is_multivariate(&self) -> bool }

impl Distribution { pub fn cov(&self) -> Option<&CovarianceMatrix> }

A parameter’s value drawn from a probability distribution rather than fixed (esm-spec §6.3, schema Distribution). The closed set is normal, lognormal, uniform — bindings implement exactly these and reject anything else.

Univariate when the location parameter is a number, multivariate when it is an array — in which case cov gives the full covariance matrix and the parameter’s shape must agree.

Source · Rust docs

Go

type Distribution struct {
	Kind string `json:"kind"`
	// normal
	Mean any `json:"mean,omitempty"`
	Std  any `json:"std,omitempty"`
	// lognormal
	Mu    any `json:"mu,omitempty"`
	Sigma any `json:"sigma,omitempty"`
	// uniform
	Low  any `json:"low,omitempty"`
	High any `json:"high,omitempty"`
	// Cov is the symmetric positive-semidefinite covariance matrix (row-major)
	// of a multivariate `normal` / `lognormal`. Mutually exclusive with
	// Std / Sigma.
	Cov [][]float64 `json:"cov,omitempty"`
}

Distribution is a parameter’s value drawn from a closed set of probability distributions (esm-spec 6.3): normal (Mean + Std|Cov), lognormal (Mu + Sigma|Cov), or uniform (Low + High). A distribution is univariate when its location parameter is a number and multivariate when it is an array, in which case the parameter’s shape must agree and Cov gives the full covariance matrix – which is how 1.0.0 spells correlated noise, replacing the 0.x correlation_group tag that never carried a matrix.

The scalar-or-vector fields are any (float64 or []any) because the schema admits both spellings at every one of them.

Source · Go docs