These final verifications should be performed on the simulations using in FWT.

Here are the details about the code and how to run it is :

FWI verification code

Here is the example output of the code for Emod3D version V3.0.7

 https://wiki.canterbury.ac.nz/display/QuakeCore/Cybershake+v18.6+verification

Forward simulation: 

  1. Velocity model visualization

    Test: 3-D visualization of Vs, Vp model using a Matlab script.
    Pass criterion: The Vs, Vp models have correct dimensions and orders. The user can access the different profiles including: homogeneous, 1-D layering profiles or 3-D profiles imported from a geological software (i.e Velocity Development Programs).

    Note: The color scale should indicate the relationship between Vs and Vp. This visual examination is required before running a simulation to assure the correct of the input velocity models.

  2. Modification of source in terms of locations, source shapes, ...

    Test a): Input a Ricker wavelet as an active synthetic source at a customized location for forward emod3D simulation using a homogeneous velocity profile. Set up a list of stations for recording seismograms.
    Plot the seismograms according to a station for visual examination.

    Pass criterion: The seismogram output should have expected shape and indicate the correctness in implementation of wave equation, stability criteria, dispersion criteria, free surface and absorbing boundary conditions.

    Test b): Input a source signal recorded from a real event at a given location for forward emod3D simulation using a developed 3-D velocity profile. Import a list of stations for recording seismograms (Select some stations from a full list of existing stations for the event).
    Plot the seismograms according to a station for visual examination.

    Pass criterion: The seismogram output should have similar trend in terms of magnitude and phase to the source signature and the observed seismograms (if available).

    Note: 3-D animation of the wavefield can be implemented to access the wave simulation. (In our case, the evolution of the strain field can be used for animation).

  3. Access the output  from forward emod3d simulation for different velocity models

    Test: Input a source signal recorded from a real event at a given location for forward emod3D simulation using a developed 3-D velocity profile (true model) and an homogeneous profile (initial model). Access the seismograms from 2 forward runs (observed and estimated data) in time series and in frequency contents.
    Disable the strain tensor output for the run with true model and Enable the strain tensor output for the run with initial model. The enable strain tensor output for a forward run can be plotted according to an array of cells along a specific direction (x, y or z).

    Pass criterion: The seismograms from 2 runs should have the same format and comparable in term of magnitude. The strain field plotted along a line in one direction should indicate the consistent in term of magnitude and phase.

    Note: The strain field output has 6 components:eii3, exxp, eyyp, exyp, exzp, eyzp which are the deviatoric strain tensors. The deviatoric strain tensor also can be converted to regular train tensor.

  4. Low-pass filter of the seismogram

    Test: Plot the frequency content of the seismogram at each station and for each component (using FFT transform). Choose a low-pass filter with the high-cut frequency up to 1 Hz. Apply the filter and plot the filtered sesmogram together with the original seismogram in time serie and frequency content.

    Pass criterion: The filtered seismogram should have the frequency content in a specified bandwidth and the filter does not modify the seismogram in an unexpected way.

    Note: The low-pass filter can also be applied for the source signature and then be inherited by the wavefields during the simulation.

 

Adjoint source & Adjoint simulation: 

  1. Adjoint simulation with source signal as Ricker wavelet applying at station locations

    Test: Input the Ricker wavelet/ Time-derivative of Ricker wavelet as a body force at a random station from files with similar format to the recorded seismograms. The adjoint emod3d simulation should perform similarly to the forward emod3d with the only difference:  the source using was multi-channel source applied simultaneously at all receiver locations.
    Pass criterion: The recorded seismograms at  stations should be similar to the combination of seismograms from multi forward simulations with a single source location at each station. The seismograms can be examine in term of time series (magnitude and phase) and frequency contents.

    Note: For visual examination of the adjoint simulation, we can apply the active source at one station and record the seismograms at the other stations.

  2. Calculation of adjoint source for P-arrival components for each seismograms using GSDF method

    Test:
    Pass criterion:

    Note:

  3. Adjoint simulation using adjoint sources from GSDF

    Test:
    Pass criterion:

    Note:

Kernel calculation: 

  1. Banana - donut Kernel calculation for adjoint source with only P-wave arrival signal for one seismogram component (vx)
    Test:
    Pass criterion:

    Note:

  2. Kernel calculation for adjoint source with all 3 components (vx, vy, vz)

    Test:
    Pass criterion:

    Note:

  3. Kernel calculation for synthetic study with an 1-D true model and a homogeneous initial model:
    Pass criterion:

    Note:

 

 

Tentative items for future verification tasks:

  1. Modification of the source mechanism
  2. ...
  • No labels