create new tag
view all tags

eSDO 1121: Local Helioseismology Inversion

This document can be viewed as a PDF.
Deliverable eSDO-1121: Local Helioseismology Inversion
S. Zharkov, M. Thompson
28 June 2005


The aim of this package is to provide front-end interface for specifying and launching inversions of helioseismic data, on local or remote machines; maintain queueing system and indices of data and inversions results; retrieve inversions results and allow display and local storage.


  • Travel times (mean/difference)
  • corresponding travel time sensitivity kernels
  • noise covariance matrix
  • regularisation option
  • trade-off parameter


  • Solar interior inversion, error estimates, resolution estimates

Tool Interface

  • commandline: input of AIA images, output of FITS files containing images and statistical data.
    1. AstroGrid CEA web service: this algorithm will be deployed as a CEA service hosted in the UK that users can call the web service to process datasets on the grid.
    2. SolarSoft routine: the C module will be wrapped in IDL and distributed through the MSSL SolarSoft gateway. Users will access to a SolarSoft installation can call the routine from the commandline or GUI to process locally held data.
    3. JSOC module: the C module will be installed in the JSOC pipeline. Users can access the routine through pipeline execution to operate on data local to the JSOC data centre.

Science Use Case

In general, the inverse problem in local helioseismology is ill posed and thus requires some "a priori" knowledge which is used to select the proper solution from the set of possible solutions given by the measurements. This normally consists of choosing an additional constraint which causes the chosen solution to be "smooth" in some way. The relative influence of such regularization is controlled by "regularization trade-off parameter". In general, model solutions are computed for a range of different values, and then the "best" model is chosen.

In addition, the degree to which any inversion method succeeds depends on the accuracy of the forward model, noise level in travel times, the depth and spatial scales of real variations in the Sun, the number and type of travel times that are available as input to inversion, the accuracy of the travel time covariance matrix. This package will provide the user with tools to apply different inversion methods with varying trade-off parameters to the data enabling the user to choose "best" results.

For Perturbation Map Generation and Subsurface Flow Analysis algorithms during the inversion part the choice of regularisation and trade-off parameter will be made automatically as a first approximation. User will be able to take travel-times, error covariance matrix and travel time sensitivity kernels available as additional output in the above algorithms and use those as input to further local helioseismology inversion systems to refine the inversion results.

  1. User begins by inputing travel times, time sensitivity kernels, regularisation option and noise covariance matrix to local helioseismology inversion.
  2. Trade-off parameter is chosen automatically for first approximation.
  3. Inversion is calculated and returned to the user with error estimates.
  4. Inversion can then be re-calculated with different trade-off parameters to refine inversion results.

Technical Use Case

Inversion methods considered for implementation: Multi channel deconvolution, Regularized Least Squares, Singular Value Decomposition, Subtractive Optimally Localized Averages.

Option 1: Multi Channel Deconvolution (Priority for implementation)

  1. Input: Travel times, corresponding sensitivity kernels, Solar model, noise covariance matrix, trade-off parameter, regularization operator
  2. Perform 2D Fourier transforms of the input mean travel time perturbations and sensitivity kernels
  3. Calculate weight matrices for model vector using chosen regularization and trade-off parameter.
  4. Calculate the Fourier transform of the inverse operator using weight matrices, Fourier transformed model data.
  5. Calculate the Fourier transform of the estimated soundspeed perturbation
  6. The soundspeed perturbation estimate is found by inverse Fourier transforming of 3 layer by layer.
  7. Calculate the covariance matrix of the estimated model using inverse operator and data error covariance matrix
  8. Calculate the resolution matrix using inverse operator and Fourier transformed data.
  9. output: Solar interior inversion, error estimates, resolution estimates

Option 2: Subtractive Optimally Localized Averages

  1. Input: travel times, corresponding sensitivity kernels, noise covariance matrix, trade-off parameter
  2. From sensitivity kernels, noise covariance matrix and trade-off parameter calculate the the mode kernel cross-correlation matrix and calculate its inverse.
  3. Choose Gaussian target function parameters, ie spatial location in 3D, spatial horizontal and vertical extent.
  4. For each data point calculate the cross-correlation vector of the mode kernels with the target function with an extra column used for constraint.
  5. From 2. and 4. calculate weight coefficients at each data point.
  6. Calculate the solar interior parameters using input data and weight coefficients.
  7. Calculate error bars and spatial resolution attained.

Option 3: Generalized Singular Value Decomposition

Here we are essentially minimizing the expression

where Lambda is the trade-off parameter, L is the regularization operator, and ||.|| denotes L2 norm.

  1. Input: travel times, corresponding sensitivity kernels, Solar model, noise covariance matrix, trade-off parameter
  2. Using sensitivity kernels and top hat functions as basis calculate matrix A.
  3. Solve generalized SVD for A and regularization operator L obtaining generalized singular values of (A, L), rank of A and decomposition matrices U, V, W-1.
  4. Calculate W and filter factors.
  5. Using U, V, W, filter factors and non-zero singular values calculate the inversion from input data at each data point.
  6. Calculate resolution matrix and error estimates.

Regularization operator choice

The choice of the regularization operator will be the zero-th, first or second derivative of the model.

Quicklook Products


Support Information

  1. B.H. Jacobsen, I. Moller, J.M. Jensen and F.Efferso, Multichannel Deconvolution, MCD, in Geophysics and Helioseismology, Phys. Chen. Earth(A), Vol. 24, No. 3, 215-220, 1999
  2. Pijpers, F. P., Thompson, M. J., The SOLA method for helioseismic inversion, A&A, 1994,
  3. Christensen-Dalsgaard, J., Hansen, P. C., Thompson, M. J., Generalized Singular Value Decomposition Analysis of Helioseismic Inversions, 1993
  4. Tikhonov A N and Arsenin V Ya, Solution of Ill-Posed Problems, 1977

-- ElizabethAuden - 29 Jun 2005

Edit | Attach | Watch | Print version | History: r20 < r19 < r18 < r17 < r16 | Backlinks | Raw View | More topic actions
Topic revision: r20 - 2005-09-30 - ElizabethAuden
This site is powered by the TWiki collaboration platform Powered by PerlCopyright © 2008-2019 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback