Integration of data to improve uncertain geological models through non-parametrical empirical Bayesian networks.

Miguel Varga and Alexander Schaaf and Florian Wellmann. ( 2016 )
in: 2016 RING Meeting, ASGA

Abstract

Several techniques exist to construct structural geological models on the basis of different as- sumptions and varying types of geological observations. However, two problems are prevalent when constructing these models: (i) observations and assumptions, and therefore also the constructed model, are subject to uncertainties, and (ii) additional information is often available, but cannot be considered directly in the geological modelling step, although it may be suitable to reduce model uncertainties. A range of methods have recently been developed to analyse and quantify the first aspect, for example on the basis of Monte Carlo error propagation from uncertain input data to model results. We extend these previous approaches here to include information to the model in form of observation likelihoods in a non-parametric empirical Bayesian inference framework, and describe the implementation in a probabilistic programming framework. To illustrate our developments, we apply the method to several model studies with increasing complexity, and highlight the potential of the method to consider additional geophysical data (grav- ity measurements), as well as geological information, integrated in the form of suitable likelihood functions. First results are promising, and we expect to gain more insights into uncertainties in structural models through this flexible implementation with the consideration of more complex geological likelihood functions in future work.

Download / Links

BibTeX Reference

@INPROCEEDINGS{,
    author = { Varga de la, Miguel and Schaaf, Alexander and Wellmann, Florian },
     title = { Integration of data to improve uncertain geological models through non-parametrical empirical Bayesian networks. },
 booktitle = { 2016 RING Meeting },
      year = { 2016 },
 publisher = { ASGA },
  abstract = { Several techniques exist to construct structural geological models on the basis of different as-
sumptions and varying types of geological observations. However, two problems are prevalent when
constructing these models: (i) observations and assumptions, and therefore also the constructed
model, are subject to uncertainties, and (ii) additional information is often available, but cannot be
considered directly in the geological modelling step, although it may be suitable to reduce model
uncertainties. A range of methods have recently been developed to analyse and quantify the first
aspect, for example on the basis of Monte Carlo error propagation from uncertain input data to
model results. We extend these previous approaches here to include information to the model in
form of observation likelihoods in a non-parametric empirical Bayesian inference framework, and
describe the implementation in a probabilistic programming framework.
To illustrate our developments, we apply the method to several model studies with increasing
complexity, and highlight the potential of the method to consider additional geophysical data (grav-
ity measurements), as well as geological information, integrated in the form of suitable likelihood
functions. First results are promising, and we expect to gain more insights into uncertainties in
structural models through this flexible implementation with the consideration of more complex
geological likelihood functions in future work. }
}