WIND>Main/Events>WebHome (31 Mar 2023, operto)EditAttach

| 2020 |
| 2022 |

2023 events

  • January 15th: WIND group submitted the following abstracts to 2023 EAGE annual meeting.
  • H. S. Aghamiry, A. Gholami , S. Operto, and M. Freitag, Efficient seismic redatuming using least-squares time-reversal, 84st Annual EAGE Meeting (Vienna). H. Aghamiry proposes a model-based redatuming technique. The idea is inspired from the wavefield reconstruction inversion or extended-source FWI: Find an approximation of the true wavefields from the recorded data and the background model and extract the values of the data-assimilated wavefield at the receiver datum level. Then, used reciprocity and use the same approach to redatum the sources. This approach can be implemented in the frequency domain or in the time domain. In the second case, first estimate for each source the scattering source of the Lippmann-Schwinger equation by solving the scattered-data fitting problem with a data-domain Hessian, then solve the wave equation with an extended source (physical source + the scattering source). In the frequency domain, use this implementation or solve the normal system built by gathering the wave equation and the observation equation.
  • K. Aghazade, A. Gholami, and H. S. Aghamiry, Elastic full waveform inversion with physical constraints: alternating direction method of multipliers' strategy, 84st Annual EAGE Meeting (Vienna). K. Aghazade describes how to implement prior constraints in elastic FWI with ADMM. Here, the constraints rely on empirical relationship between Vp and Vs following the origonal work of Data and Sava (2016). He compares three methods to implement constraints: ADMM, POCS and Dykstra.
  • K.Aghazade, A. Gholami, and H. S Aghamiry, Full waveform inversion by adaptive Tikhonov-Total variation regularization, 84st Annual EAGE Meeting (Vienna). K. Aghazade describes how to tune the relative weight between Tikhonov and Total-variation regularization with robust statistics when they are combined during extended-space FWI such as the wavefield reconstruction inversion method.
  • G. Guo, S. Operto, H. Aghamiry, and A. Gholami, Weighted time-domain extended-source full waveform inversion with layer stripping, 84st Annual EAGE Meeting (Vienna). G. Guo describes how to extend the linear regime of extended-source FWI with layer-stripping stragegies. He shows that this heps to converge toward accurate velocity models with the data are complicated by surface multiples.
  • G. Guo, S. Operto, and H. Aghamiry Time domain full waveform inversion with decomposed Gauss-Newton Hessian, 84st Annual EAGE Meeting (Vienna). In FWI, the adjoint wavefield is the adjoint approximation of the least-squares problem ailing at fitting the measured scattered data (i.e., the data residuals). Why ot using instead the least-squares solution?
  • S. Operto, et al. 3D frequency-domain FWI of full-azimuth/long-offset OBN data - The Gorgon-data FWI case study, 84st Annual EAGE Meeting (Vienna). S. Operto presents an application of 3D frequency-domain FWI based on the sparse multifrontal solver MUMPS on the OBN data from the Gorgon field.
  • Y. Wu, H. Aghamiry, S. Operto and J. Ma, Frequency-domain wave simulation using physics-informed neural networks (PINNs) with free surface boundary condition, 84st Annual EAGE Meeting (Vienna). Y. Wu describes how to implement the free-surface boundary condition in seismic modeling performed by PINN. Two possible options: implement it as a soft constraint in the loss function or as a hard constraint in the NN.

  • January 22nd: Gaoshan Guo submitted the manuscript entitled "A practical implementation of data-space Hessian in the time-domain extended-source full-waveform inversion" by G. Guo, S. Operto, A. Gholami and H. Aghamiry.
This paper discusses several approximation of the data-domain Hessian associated with the scattering-source estimation problem in time-domain extended-source FWI. In the 2022 paper of Ali Gholami et al., a scaled diagonal approximation of the Hessian was used (namely, the scattering-source estimation problem is solved with a steepest-descent method). In this paper, Gaoshan Guo proposes several approximations based on 1D Wiener matching filters, 1D Gabor matching filters, 2D Gabor matching filters. He also estimate the data-domain Hessian with truncated Gauss-Newon method by solving the normal system with the conjugate gradient method. The effects of the data-domain Hessian on the convergence of the inversion and on the accuracy of the velocity model reconstruction are illustrated with the Marmousi II model and the 2004 BP salt model. Gaosha Guo also shows the role of TV regularization to mitigate the sensitivity of ES-FWI to the accuracy of the data-domainHessian, the sensitivity of ES-FWI to the accuracy of the starting model (in terms of computational cost and accuracy of the velocity model reconstruction), the impact of the free-surface multiples on the convergence of the method. He also proposes a practical workflow that combines ES-FWI and FWI to mitigate the computational burden of extended approaches while converging to accurate velocity models.

  • March 1st: The paper "Is 3D frequency-domain FWI of full-azimuth/long-offset OBN data feasible? The Gorgon-data FWI case study" is published in the March issue of The Leading Edge. The paper is available on the intranet pages of WIND.
  • March 7th: FFWI code is available on the sponsor page. The documentation of the code is available here .
Topic attachments
I Attachment Action Size Date Who Comment
IMG_4968.JPGJPG IMG_4968.JPG manage 858.7 K 23 Jun 2022 - 20:59 Main.UnknownUser  
IMG_4970.JPGJPG IMG_4970.JPG manage 863.5 K 23 Jun 2022 - 20:59 Main.UnknownUser