We propose a novel method for velocity estimation that leverages the newest advances in Deep Learning (DL) technology. This method is fully automatic and maps seismic shot-domain data to corresponding depth-domain velocity fields via two neural networks.

Our new method is conceptually different from conventional methods such as seismic tomography or Full Waveform Inversion (FWI) that minimize a fixed objective function. Here, a system of neural networks automatically and continuously learns an objective function while training the seismic-to-velocity mapping.

The newly introduced method avoids many of the drawbacks of conventional velocity estimation techniques, such as dependence on initial models or cycle-skipping. It uses the full seismic wavefield and avoids picking of first-arrival traveltimes. The system needs to be trained with hundreds or thousands of examples of seismic data paired with their corresponding velocity models relevant for the current project. Training the system is the main computationally demanding step and produces a mapping function that contains the seismic “know-how” for the presumed geologic setting. The computational cost of the subsequent estimation of velocity from new seismic data is negligent.

Our first tests on complex two-dimensional synthetic data produce impressive results, underlining the potential of DL for velocity analysis.


Article metrics loading...

Loading full text...

Full text loading...


  1. Aminzadeh, F., Jean, B., & Kunz, T.
    (1997). 3-D salt and overthrust models. Society of Exploration Geophysicists.
    [Google Scholar]
  2. Araya-Polo, M., Jennings, J., Adler, A., & Dahlke, T.
    (2018). Deep-learning tomography. The Leading Edge, 37(1), 58–66. https://doi.org/10.1190/tle37010058.1
    [Google Scholar]
  3. Huang, Z., Shimeld, J., Williamson, M., & Katsube, J.
    (1996). Permeability prediction with artificial neural network modeling in the Venture gas field, offshore eastern Canada. Geophysics, 61(2), 422–436. DOI: https://doi.org/10.1190/1.1443970
    [Google Scholar]
  4. Isola, P., Zhu, J. Y., Zhou, T., & Efros, A. A.
    (2016). Image-to-Image Translation with Conditional Adversarial Networks. arXiv preprint arXiv:1611.07004.
    [Google Scholar]
  5. Michelsanti, D., & Tan, Z. H.
    (2017). Conditional generative adversarial networks for speech enhancement and noise-robust speaker verification. arXiv preprint arXiv:1709.01703.
    [Google Scholar]
  6. Murat, M. E., & Rudman, A. J.
    (1992). Automated first arrival picking: A neural network approach. Geophysical Prospecting, 40(6), 587–604. DOI: https://doi.org/10.1111/j.1365–2478.1992.tb00543.x
    [Google Scholar]
  7. Röth, G., & Tarantola, A.
    (1994). Neural networks and inversion of seismic data. Journal of Geophysical Research: Solid Earth, 99(B4), 6753–6768. DOI: https://doi.org/10.1029/93JB01563
    [Google Scholar]

Data & Media loading...

This is a required field
Please enter a valid email address
Approval was a Success
Invalid data
An Error Occurred
Approval was partially successful, following selected items could not be processed due to error