Dataset for Seismic waveform tomography of the Central and Eastern Mediterranean upper mantle

Dataset corresponding to the Seismic waveform tomography of the Central and Eastern Mediterranean upper mantle

This dataset belongs to the seismic waveform tomography of the Central and Eastern Mediterranean by Blom, Gokhberg and Fichtner, Solid Earth (Discussions), 2019. Seismic tomography is an inverse problem where the internal elastic structure of the Earth (the upper ~500 km) is determined from seismograms (the vibrations of the Earth as a result of earthquakes, as recorded by seismometers at the Earth's surface). This inverse problem is cast as an optimisation where the misfit between observed and synthetic seismograms is minimised: waveform tomography (often referred to as full waveform inversion or FWI). Synthetic seismograms are produced by simulating the elastic wavefield of earthquakes within the Earth. The optimisation problem is solved by iterative, deterministic, gradient-based inversion. Gradients are computed using the adjoint method, which requires one forward wavefield simulation and one adjoint wavefield simulation per earthquake used in the project.

The inversion was carried out over several frequency bands, starting with the longest periods and including a progressively broader frequency band. Within each frequency band, ~10-20 iterations were carried out, totalling to a hundred iterations. Synthetic seismograms and iteration information are stored for a subset of iterations, notably those where human interaction (i.e. the selection of events / data windows) took place.

Here, we describe:

• The contents of this package
• How to set up the package such that all the data can be accessed and used, and reproduce the figures.

Contents of this package

• Data that was used for the seismic waveform inversion: raw and processed seismograms, station information, earthquake information, as well as the window selection (designating the parts of the data that were actually used at each stage in the inversion) and synthetic seismograms produced during various stages of the inversion. This information is gathered in the LASIF project "EMed_full.complete.tar".
• Models and misfit development across the iterations, as well as models relating to model testing, as carried out after the inversion. This information is gathered in the tarball "MODEL_FILES.tar". Model files are both given in the ses3d ascii format (text file drho, dvsv, dvsh, dvp and block_x, block_y,  block_z) and in bundled .vtu format. Conversion to .vtu was done using the tools in SCRIPTS. These vtu files can be viewed using Paraview.
• information on the tools and code that was used to do the inversion:
  • ses3d: a seismic wave propagation spectral element code in spherical coordinates. This will run both forward and adjoint simulations. This is available publicly through the developers on https://cos.ethz.ch/software/production/ses3d.html. See Gokhberg & Fichtner, 2016.
  • LASIF: a waveform inversion workflow managing package, where we have made small adaptations to make it suitable for our workflow. The original package is available via www.lasif.net and on github (see Krischer et al, 2015), the modified version is added to this package as 'LASIF-master.zip'.
  • LASIF_scripts: bespoke scripts in order to interact with the LASIF project and generate different types of analyses and plots that are used in the publication. This is included in the tarball 'LASIF_scripts.tar'
  • SCRIPTS: containing some modified tools that were originally written for ses3d, as well as some additional tools - notably to interact with models converted to the VTK format. This is included in the tarball 'SCRIPTS.tar'
  • A description of the conda environment named lasif_ext (which is used for all the data analysis), in the form of the yml file 'lasif_ext.yml'
• An additional LASIF project which is used just to compute sensitivity kernels for different windows within the same trace: 'EMed_window_kernels.tar'. This is used as an example in one of the manuscript figures.

How to set up the data package

1. Download the entire data package. We will assume it is located in `~/Downloads/`.
2. Get miniconda or anaconda if you don't have it.
3. Install LASIF. This can be done using the instructions from the LASIF website, but with a few adaptations, which are detailed in the lasif_ext.yml file. This amounts to the following:
   1. Add the channel conda-forge to your standard channels
   2. Name the environment "lasif_ext"
   3. Manually replace the files in the LASIF source directory with those in LASIF-master.zip.
   4. Install the specific version of pyqt=4.11.
   5. Install the additional packages jupyter, vtk=7.0.0, pandas=0.23.4 (these are the ones that work for me).
4. Extract the LASIF_scripts.tar to the site-packages directory of your conda environment:

   tar -xf ~/Downloads/LASIF_scripts.tar -C [/path/to/conda/environments]/lasif_ext/lib/python2.7/site-packages/

5. Make a project directory and extract all needed packages into it:

   # make project directory
   mkdir CEMed_project_Blometal
   cd CEMed_project_Blometal
   
   # extract data tarballs into it
   tar -xf ~/Downloads/EMed_full.complete.tar
   tar -xf ~/Downloads/EMed_window_kernels.tar
   tar -xf ~/Downloads/MODEL_FILES.tar
   
   # make scripts directory and extract scripts into it
   mkdir conda_stuff
   tar -xf ~/Downloads/SCRIPTS.tar -C conda_stuff
   
   # make data analysis directory
   mkdir data_analysis
   cd data_analysis
   
   # extract analysis tools
   tar -xf ~/Downloads/NPY_FILES.tar
   tar -xf ~/Downloads/FIGURE_SCRIPTS.tar
   tar -xf ~/Downloads/figs_png.tar

Now the project should be ready for inspection. The following things can be done, for example:

• Reproduce the figures in the manuscript. All scripts for this are located in CEMed_project_Blometal/data_analysis/FIGURE_SCRIPTS/.

  conda activate lasif_ext
  cd CEMed_project_Blometal
  jupyter notebook

  This should open up a browser tab that shows the directory structure. Navigate to data_analysis/FIGURE_scripts and click on one of the .ipynb files to open it. If you press 'Kernel' > 'Restart kernel and run all' at the top, all cells will be launched automatically. This should work out of the box.

• Interact with the lasif project. For this, refer to the LASIF website. Note that above jupyter notebooks do so extensively, using the lasif communicator.
• Build additional analysis tools, using the tools supplied in SCRIPTS and LASIF_scripts.

References:

• Blom, N., Gokhberg, A., and Fichtner, A.: Seismic waveform tomography of the Central and Eastern Mediterranean upper mantle, Solid Earth Discuss., https://doi.org/10.5194/se-2019-152, in review, 2019.

• Gokhberg, A., Fichtner, A., 2016. Full-waveform inversion on heterogeneous HPC systems. Comp. & Geosci. 89, 260-268. https://doi.org/10.1016/j.cageo.2015.12.013

• Krischer, L., Fichtner, A., Zukauskaitė, S., and Igel, H. (2015), Large‐Scale Seismic Inversion Framework, Seismological Research Letters, 86(4), 1198–1207. doi:10.1785/0220140248