Inversion functions

calculate_chi2(
    s::AbstractSolver
) -> Dict{AbstractSpectralWindow, Float64}

Calculates the reduced $χ^2$ statistic for the spectral fit within solver.

Details

The reduced $χ^2$ statistic is calculated as

\[χ^2 = \frac{1}{N_{\text{spec}} - N_{\text{sv}} - 1} \sum_i^{N_{\text{spec}}} \frac{M_i - O_i}{\varepsilon_i}\]

calculate_scale_factor_AK(
    buf::AbstractAtmosphereBuffer,
    s::AbstractSolver,
    gas::GasAbsorber,
    isrf::Dict{<:AbstractSpectralWindow, <:AbstractISRF};
    G
) -> Any

Calculates the averaging kernel profile for a gas absorber gas if that gas was retrieved through a scaling factor (GasLevelScalingFactorSVE) rather than a full profile retrieval. This function assumes that the partial derivatives of optical depths with respect to volume mixing ratios have been calculated - which should be the case if the general set-up through buffer helper functions (e.g. EarthAtmosphereBuffer(...)).

Details

(TODO: add derivation)

check_solver_validity(s::AbstractSolver) -> Bool

Checks if the solver object s is valid, meaning that the mapping indices are not empty, the model radiances do not have any NaNs and the model Jacobians do not have any NaNs. Returns true if all of those checks pass, otherwise returns false.

This function can be used in inversions to prevent later parts of the code from failing due to NaNs being propagated into matrices etc.

compute_finite_difference_jacobian(
    s::AbstractSolver,
    sve::AbstractStateVectorElement,
    Δx
) -> Tuple{Any, Any, Any}

Computes the finite-difference Jacobian for the forward model embedded in solver s. Users must supply a perturbation value Δx (must be unit-compatible with sve), such that (F(x + Δx) - F(x)) / Δx can be computed. Δx must have units. Returns F(x), F(x + Δx), [F(x + Δx) - F(x)]/Δx.

create_K_from_solver(s::AbstractSolver) -> Any

Turns the Jacobians from the solver object into a freshly allocated matrix, so they can be used in the inversion part of the algorithm.

create_Se_from_solver(
    s::AbstractSolver;
    return_inverse
) -> Any

Creates and returns the instrument noise covariance matrix. If the optional keyword return_inverse is set to true, the inverse is returned. For now, this function only supports diagonal noise covariance matrices.

get_iteration_count(s::AbstractSolver) -> Any

Returns the number of iterations performed in an AbstractSolver object s, and also checks if all state vector elements have the same count (as they should).

get_jacobian(
    s::AbstractSolver,
    SVE::AbstractStateVectorElement,
    swin::AbstractSpectralWindow;
    view
) -> Any

Returns the partial derivative (Jacobian) with respect to state vector element SVE, as embedded in solver s for a spectral window swin. If the optional keyword view is true, then a view on the Jacobian is returned, otherwise a copy is created and returned.

get_jacobian(
    s::AbstractSolver,
    SVE::AbstractStateVectorElement
) -> Any

Returns the partial derivative (Jacobian) with respect to state vector element SVE, as embedded in the solver s. This function does not calculate the Jacobian itself, so in order for this function to return a meaningful quantity, users must call a forward model beforehand, which implements the correct calculations and stores them in s.jacobians. This function returs a newly created vector.

See also create_K_from_solver and get_jacobian(s, SVE, swin).

get_measured!(measured, s)

Returns the measured radiances as of an AbstractSolver s, as seen by the full solver, meaning that all spectral windows are inserted into an existing array measured. This function does not check if measured is big enough.

get_measured(
    s::AbstractSolver,
    swin::AbstractSpectralWindow;
    view
) -> Any

Returns the measured radiances of an AbstractSolver s, belonging to an AbstractSpectralWindow swin. The optional argument view determines whether to return a view onto the array (true), or a copy (false).

get_measured(s::AbstractSolver) -> Any

Returns the measured radiances as of an AbstractSolver s, as seen by the full solver, meaning that all spectral windows are concatenated into one newly allocated array. This reads the contents of s.measured but mapped via the solver indices.

get_modeled!(modeled::AbstractVector, s::AbstractSolver)

Returns the modeled radiance field of an AbstractSolver s for all spectral windows, and stores them into modeled, according to the solver's mapping indices. Does not check whether modeled is large enough to hold all values.

get_modeled(
    s::AbstractSolver,
    swin::AbstractSpectralWindow;
    view
) -> Any

Returns the modeled radiance currently stored in the radiance field of solver s, corresponding to the spectral window swin. The optional argument view determines whether to return a view onto the array (view=true), or a copy (view=false).

get_modeled(s::AbstractSolver) -> Any

Returns the modeled radiance currently stored in the radiance field of AbstractSolver s, for all spectral windows, with the ordering given by s.indices. Allocates a new vector.

get_noise!(noise::AbstractVector, s::AbstractSolver)

Returns the noise-equivalent radiances in AbstractSolver s, where the values for all spectral windows inside s are concatenated into the vector noise. Does not check whether noise is large enough to hold all values.

get_noise(
    s::AbstractSolver,
    swin::AbstractSpectralWindow;
    view
) -> Any

Returns the noise-equivalent radiances in AbstractSolver s, belonging to spectral window swin. If view is true, a view to the vector inside s is returned, otherwise a new vector is allocated and returned.

get_noise(s::AbstractSolver) -> Any

Returns the noise-equivalent radiances in AbstractSolver s, where the values for all spectral windows inside s are concatenated into a new vector and returned.

get_wavelength(
    s::AbstractSolver,
    swin::AbstractSpectralWindow;
    view
) -> Any

Returns the wavelengths in AbstractSolver s, belonging to spectral window swin, as pointed to by the dispersion object. If view is true, a view to the vector inside s is returned, otherwise a new vector is allocated and returned. See also get_wavenumber.

get_wavelength(s::AbstractSolver) -> Any

Returns the wavelenth vector in AbstractSolver s, where the values for all spectral windows inside s are concatenated into a new vector and returned. See also get_wavenumber.

print_posterior(s::AbstractSolver)

Print a convenient summary of the current state vector, including the associated uncertainties. Note that this function does not check for convergence.

print_posterior(q::OEQuantities)

Print a convenient summary of the current state vector, including the associated uncertainties. Note that this function does not check for convergence.