Two dimensional seismic fault network interpretation using marked point processes

Fabrice Taty-Moukati and Radu S Stoica and Francois Bonneau and Xinming Wu and Guillaume Caumon. ( 2022 )
in: 2022 {RING} {Meeting}, pages 26, ASGA

Abstract

Seismic fault interpretation is an important input to subsurface models. Since in seismic images the dominant features are reflection events corresponding to horizons, fault interpretation can be achieved by computing a fault probability image. Such an image highlights fault presence while suppressing reflection events. Many methods like Machine Learning approaches have been proposed to produce probability images with good resolution. However, current approaches extract the ‘best’ fault network from the fault probability image (e.g., by thinning). The goal of this paper is to quantify the uncertainties related to the number and connectivity of faults honoring a given probability image, as all the possible fault networks can yield different outcomes in terms of subsurface behavior (e.g., reservoir flow). In this paper, we propose a rigorous approach to assess seismic fault network uncertainty. We use a mathematical framework that originates from stochastic geometry modeling, called marked point processes. Fault networks can be seen as realizations of such processes. To define the marked point process, we choose a probability density. This energy function is composed of two terms including a data term, necessary for fault localization, and an interaction energy for fault relationships. We use a simulated annealing framework based on Metropolis-Hastings algorithms, which makes it possible to find the global maximum of the probability density, built in the form of a Gibbs density. We apply the proposed approach to a 2-D seismic cross-section extracted from the Volve seismic cube provided by Equinor, and show some preliminary results of fault networks interpreted as random segments that connect and align in two dimensional space.

Download / Links

BibTeX Reference

@inproceedings{taty-moukati_two_2022,
 abstract = { Seismic fault interpretation is an important input to subsurface models. Since in seismic images the dominant features are reflection events corresponding to horizons, fault interpretation can be achieved by computing a fault probability image. Such an image highlights fault presence while suppressing reflection events. Many methods like Machine Learning approaches have been proposed to produce probability images with good resolution. However, current approaches extract the ‘best’ fault network from the fault probability image (e.g., by thinning). The goal of this paper is to quantify the uncertainties related to the number and connectivity of faults honoring a given probability image, as all the possible fault networks can yield different outcomes in terms of subsurface behavior (e.g., reservoir flow). In this paper, we propose a rigorous approach to assess seismic fault network uncertainty. We use a mathematical framework that originates from stochastic geometry modeling, called marked point processes. Fault networks can be seen as realizations of such processes. To define the marked point process, we choose a probability density. This energy function is composed of two terms including a data term, necessary for fault localization, and an interaction energy for fault relationships. We use a simulated annealing framework based on Metropolis-Hastings algorithms, which makes it possible to find the global maximum of the probability density, built in the form of a Gibbs density. We apply the proposed approach to a 2-D seismic cross-section extracted from the Volve seismic cube provided by Equinor, and show some preliminary results of fault networks interpreted as random segments that connect and align in two dimensional space. },
 author = { Taty-Moukati, Fabrice AND Stoica, Radu S AND Bonneau, Francois AND Wu, Xinming AND Caumon, Guillaume },
 booktitle = { 2022 {RING} {Meeting} },
 pages = { 26 },
 publisher = { ASGA },
 title = { Two dimensional seismic fault network interpretation using marked point processes },
 year = { 2022 }
}