Approximating the Hessian matrix in seismic full waveform inversion

Abdrahamane Berete and Giusi Ruggiero and Paul Cupillard. ( 2024 )
in: Proc. 2024 RING Meeting, pages 11, ASGA

Abstract

Seismic Full Waveform Inversion (FWI) is a technique for imaging the Earth's interior offering high-resolution models of physical properties of the subsurface, widely used for various applications from exploration to geohazard assessment. This study investigates the performance of two different initial Hessian approximations for Gauss-Newton optimization algorithms in the context of 2D seismic FWI. In particular, experiments are carried out employing the BFGS and L-BFGS algorithms and results are compared in terms of accuracy of the inverted model and computational time. The study reveals the importance of initial Hessian approximations in guiding optimization processes and highlights FWI's potential to enhance our understanding of subsurface structures.

Download / Links

BibTeX Reference

@inproceedings{berete_approximating_RM2024,
 abstract = {Seismic Full Waveform Inversion (FWI) is a technique for imaging the Earth's interior offering high-resolution models of physical properties of the subsurface, widely used for various applications from exploration to geohazard assessment. This study investigates the performance of two different initial Hessian approximations for Gauss-Newton optimization algorithms in the context of 2D seismic FWI. In particular, experiments are carried out employing the BFGS and L-BFGS algorithms and results are compared in terms of accuracy of the inverted model and computational time. The study reveals the importance of initial Hessian approximations in guiding optimization processes and highlights FWI's potential to enhance our understanding of subsurface structures.},
 author = {Berete, Abdrahamane and Ruggiero, Giusi and  Cupillard, Paul},
 booktitle = {Proc. 2024 RING Meeting},
 language = {en},
 pages = {11},
 publisher = {ASGA},
 title = {Approximating the Hessian matrix in seismic full waveform inversion},
 year = {2024}
}