Sampling method with geological prior models for solving inverse problems

Christoph Jaggli and Julien Straubhaar and Philippe Renard. ( 2017 )
in: 2017 Ring Meeting, pages 1--6, ASGA

Abstract

When facing inverse problems in the hydrogeological framework, one is often confronted with significant structures of sharp petrophysical contrasts such as karstic aquifers, cliffs, faults, lenses, etc. Extensive exploration of such complex model spaces by Markov chain Monte Carlo (McMC) methods often results in considerable computational efforts. Most optimization methods, on the other hand, are limited to the approximation of the posterior measure by a Gaussian distribution function. PoPEx is a recently introduced method that tries to overcome these issues. On the basis of a synthetic inverse problem, the method showed very promising performance. However, it has also been highlighted that using the method for predicting expectations of random variables may suffer from a slight bias. In the following work we use a technique inspired from importance sampling for considerably improving such predictions without significantly increasing the computational costs. Introduction

Download / Links

BibTeX Reference

@inproceedings{Jaggli2017,
 abstract = { When facing inverse problems in the hydrogeological framework, one is often confronted with significant structures of sharp petrophysical contrasts such as karstic aquifers, cliffs, faults, lenses, etc. Extensive exploration of such complex model spaces by Markov chain Monte Carlo (McMC) methods often results in considerable computational efforts. Most optimization methods, on the other hand, are limited to the approximation of the posterior measure by a Gaussian distribution function. PoPEx is a recently introduced method that tries to overcome these issues. On the basis of a synthetic inverse problem, the method showed very promising performance. However, it has also been highlighted that using the method for predicting expectations of random variables may suffer from a slight bias. In the following work we use a technique inspired from importance sampling for considerably improving such predictions without significantly increasing the computational costs. Introduction },
 author = { Jaggli, Christoph AND Straubhaar, Julien AND Renard, Philippe },
 booktitle = { 2017 Ring Meeting },
 number = { 2011 },
 pages = { 1--6 },
 publisher = { ASGA },
 title = { Sampling method with geological prior models for solving inverse problems },
 year = { 2017 }
}