A Probabilistic Method for Characterising Fold Geometry from Structural Data

Lachlan Grose and Laurent Gautier and Laurent Ailleres. ( 2017 )
in: 2017 Ring Meeting, pages 1--12, ASGA

Abstract

Folds add additional complexity to the process of structural modeling. Recent developments in tech- niques used for 3D modeling and in the integration of different types of structural data have resulted in the ability constrain fold geometry in a data driven workflow. However, these methods do not address the un- certainty in the possible fold geometry. Previous studies have simply perturbed the input data to analyse the sensitivity of the interpolator to the dataset. In this work we propose using Bayesian inference to determine the posterior distribution of the fold parameters for a given structural data set. The posterior distribution is sampled using an Markov Chain Monte Carlo sampler and allows for the structural uncertainty to be sam- pled. Our approach characterises the error in the structural observations using a hyperparameter as part of the Bayesian inference. This means that the uncertainty is derived from the model rather than subjectively defined by the geologist. Our approach has been integrated into a new fold modeling workflow and we show that the variability between model iterations is typically associated with the locations in the model where data does not exist. This is where the most uncertainty should exist in models.

Download / Links

BibTeX Reference

@INPROCEEDINGS{Grose2017,
    author = { Grose, Lachlan and Gautier, Laurent and Ailleres, Laurent },
     title = { A Probabilistic Method for Characterising Fold Geometry from Structural Data },
 booktitle = { 2017 Ring Meeting },
    number = { 2014 },
      year = { 2017 },
     pages = { 1--12 },
 publisher = { ASGA },
  abstract = { Folds add additional complexity to the process of structural modeling. Recent developments in tech- niques used for 3D modeling and in the integration of different types of structural data have resulted in the ability constrain fold geometry in a data driven workflow. However, these methods do not address the un- certainty in the possible fold geometry. Previous studies have simply perturbed the input data to analyse the sensitivity of the interpolator to the dataset. In this work we propose using Bayesian inference to determine the posterior distribution of the fold parameters for a given structural data set. The posterior distribution is sampled using an Markov Chain Monte Carlo sampler and allows for the structural uncertainty to be sam- pled. Our approach characterises the error in the structural observations using a hyperparameter as part of the Bayesian inference. This means that the uncertainty is derived from the model rather than subjectively defined by the geologist. Our approach has been integrated into a new fold modeling workflow and we show that the variability between model iterations is typically associated with the locations in the model where data does not exist. This is where the most uncertainty should exist in models. }
}