Assessing fault segmentation uncertainties using Reversible Jump Monte Carlo Markov chain

Lucile Chauveau and Gabriel Godefroy and Gautier Laurent and Guillaume Caumon. ( 2018 )
in: 2018 Ring Meeting, ASGA

Abstract

Faults are composed of one or several segments which have overlapped during their growth. This segmentation potentially affects the overall fault displacement profile and the hydrodynamic behavior in faulted reservoirs. In this paper, we study the uncertainties related to fault segmentation using a Bayesian sampling probabilistic approach. The aim is to sample a posterior probability of a model combining the prior probability and its likelihood to the data. Our model is a parametrized elliptic profile of the displacement along the fault strike. This profile is the sum of the profiles of several segments. We use the Reversible Jump Monte Carlo Markov Chain (RJ-MCMC) algorithm to sample models with different number of segments. The algorithm samples possible models with different number of segments which look consistent with the displacement data. The simplest models with less parameters are sampled preferentially, which confirms that RJ-MCMC tends to generate parsimonious models.

Download / Links

BibTeX Reference

@inproceedings{RUNKJRM76,
 abstract = { Faults are composed of one or several segments which have overlapped during their growth.
This segmentation potentially affects the overall fault displacement profile and the hydrodynamic behavior in faulted reservoirs. In this paper, we study the uncertainties related to fault segmentation using a Bayesian sampling probabilistic approach. The aim is to sample a posterior probability of a model combining the prior probability and its likelihood to the data. Our model is a parametrized elliptic profile of the displacement along the fault strike. This profile is the sum of the profiles of several segments. We use the Reversible Jump Monte Carlo Markov Chain (RJ-MCMC) algorithm to sample models with different number of segments. The algorithm samples possible models with different number of segments which look consistent with the displacement data. The simplest models with less parameters are sampled preferentially, which confirms that RJ-MCMC tends to generate parsimonious models. },
 author = { Chauveau, Lucile AND Godefroy, Gabriel AND Laurent, Gautier AND Caumon, Guillaume },
 booktitle = { 2018 Ring Meeting },
 publisher = { ASGA },
 title = { Assessing fault segmentation uncertainties using Reversible Jump Monte Carlo Markov chain },
 year = { 2018 }
}