Iterative regularization for ensemble data assimilation in reservoir models

TL;DR

This paper introduces iterative regularization ensemble methods for Bayesian inverse problems in reservoir modeling, improving robustness and stability over traditional methods by leveraging ideas from deterministic inverse problem regularization.

Contribution

It develops two novel ensemble methods, IR-enLM and IR-ES, based on iterative regularization principles, specifically tailored for Bayesian inverse problems in reservoir models.

Findings

The proposed methods match standard approaches in linear Gaussian cases.

They provide more robust and stable posterior approximations in nonlinear scenarios.

Numerical experiments validate the effectiveness of iterative regularization in ensemble data assimilation.

Abstract

We propose the application of iterative regularization for the development of ensemble methods for solving Bayesian inverse problems. In concrete, we construct (i) a variational iterative regularizing ensemble Levenberg-Marquardt method (IR-enLM) and (ii) a derivative-free iterative ensemble Kalman smoother (IR-ES). The aim of these methods is to provide a robust ensemble approximation of the Bayesian posterior. The proposed methods are based on fundamental ideas from iterative regularization methods that have been widely used for the solution of deterministic inverse problems [21]. In this work we are interested in the application of the proposed ensemble methods for the solution of Bayesian inverse problems that arise in reservoir modeling applications. The proposed ensemble methods use key aspects of the regularizing Levenberg-Marquardt scheme developed by Hanke [16] and that we recently applied for history matching in [18]. In the case where the forward operator is linear and the prior is Gaussian, we show that the proposed IR-enLM and IR-ES coincide with standard randomized maximum likelihood (RML) and the ensemble smoother (ES) respectively. For the general nonlinear case, we develop a numerical framework to assess the performance of the proposed ensemble methods at capturing the posterior. This framework consists of using a state-of-the art MCMC method for resolving the Bayesian posterior from synthetic experiments. The resolved posterior via MCMC then provides a gold standard against to which compare the proposed IR-enLM and IR-ES. We show that for the careful selection of regularization parameters, robust approximations of the posterior can be accomplished in terms of mean and variance. Our numerical experiments showcase the advantage of using iterative regularization for obtaining more robust and stable approximation of the posterior than standard unregularized methods.