Bayesian data assimilation for stochastic multiscale models of transport in porous media.

Youssef Marzouk, Bart van Bloemen Waanders, Matthew Parno, Jaideep Ray, Sophia Lefantzi, Luke Salazar, Sean Andrew McKenna, Katherine A. Klise2011 · DOI

TL;DR

This paper introduces novel Bayesian techniques for reconstructing complex subsurface fields (like permeability) from sparse data, effectively bridging the gap between grid-resolved structures and unresolvable subgrid features. By developing new subgrid models and leveraging multiscale finite elements, the authors enable more accurate and robust inference of these challenging multiscale systems.

What Problem Does This Paper Address?

Reconstructing subsurface properties like permeability from limited field data is a critical but challenging task in fields ranging from hydrology to petroleum engineering. Traditional methods often struggle with fields that have a wide spectrum of length scales, meaning structures exist at scales both larger and much smaller than the resolution of the computational grid. Estimating only the grid-resolved structures leaves out crucial subgrid information, while trying to resolve everything at the finest scale becomes computationally prohibitive. This paper tackles the problem of inferring not just grid-resolved features but also statistical summaries of these subgrid structures, which are essential for accurate modeling of transport in porous media.

Methods

The authors employ a Bayesian framework for inverse problems, focusing on quantifying uncertainty. They introduce two primary approaches for handling multiscale structures:

  1. Subgrid Model: A physical (phenomenological) parameterized subgrid model is developed to capture the impact of unresolved scales on the coarse grid. This model leverages properties of truncated Gaussian fields and point-process theory to estimate effective conductivity without explicitly resolving the fine-scale binary field.
  2. Multiscale Finite Elements (MsFEM): This approach uses specially designed operators to link information between coarse and fine meshes, effectively upscaling the problem without requiring an explicit subgrid model.

Dimensionality reduction is achieved using the Karhunen-Loève expansion of a multiGaussian field. For inference, they utilize adaptive Markov Chain Monte Carlo (MCMC) methods and scalable Ensemble Kalman Filters (EnKF), extending EnKF to handle both static and dynamic data.

Datasets

The paper demonstrates its techniques on several synthetic datasets, including:

  • Binary Medium Reconstruction: A synthetic binary medium with inclusions too small to be grid-resolved is used to test the subgrid model and MCMC reconstruction.
  • Fractured Medium Reconstruction: A permeability field with long, narrow fractures is reconstructed using EnKF.
  • High Plains Aquifer Model: A real-world application involving the Ogallala aquifer in Kansas, using a MODFLOW model and data from 122 monitoring wells and 7 streamflow gages.

Key Insights and Ideas

  • Bridging Scales: The core innovation lies in effectively bridging the gap between grid-resolved and subgrid structures, allowing for a more complete representation of complex subsurface heterogeneity.
  • Subgrid Model Effectiveness: The developed subgrid model, based on truncated Gaussian fields and distance-based upscaling, proves effective in summarizing subgrid impacts without requiring fine-scale resolution.
  • EnKF for Scalability: Ensemble Kalman Filters are shown to be a scalable approach for multiscale inference, particularly for large-scale problems where MCMC becomes computationally intractable.
  • Data Fusion: Combining static (permeability) and dynamic (breakthrough times) data significantly improves the accuracy and robustness of the reconstructions.
  • Uncertainty Quantification: The Bayesian framework inherently provides uncertainty estimates, crucial for understanding the reliability of the inferred parameters.

Key Results

  • Accurate Reconstruction: The proposed methods successfully reconstruct permeability fields with both grid-resolved and subgrid structures, outperforming traditional approaches in certain scenarios.
  • Scalability Demonstrated: The EnKF approach, particularly when hybridized with MCMC, shows promise for tackling large-scale, computationally intensive problems like the High Plains aquifer model.
  • Improved Predictive Skill: Multiscale reconstructions, especially those using combined static and dynamic data, lead to more accurate predictions of transport phenomena (e.g., breakthrough times).
  • Parameter Sensitivity: The study highlights the sensitivity of the inference to the number of sensors, ensemble size, and the choice of KL modes, providing guidance for practical applications.

Limitations and Future Work

  • Computational Intensity: While multiscale methods aim to reduce computational cost, some approaches, like multiscale finite elements, remain computationally intensive.
  • Nonlinearity Challenges: Highly nonlinear dynamics, as seen in the High Plains aquifer model, still pose challenges for EnKF, sometimes leading to unstable estimates.
  • Subgrid Model Dependence: The effectiveness of the subgrid model approach relies on the ability to construct a suitable model for the unresolved structures.
  • Future Directions: The authors suggest extending the framework to higher dimensions, incorporating time-dependent parabolic equations, and further developing EnKF for nonlinear systems. They also propose exploring the use of KDE-based error models for improved inference and investigating the impact of structural errors.

Summary updated

Abstract

We investigate Bayesian techniques that can be used to reconstruct field variables from partial observations. In particular, we target fields that exhibit spatial structures with a large spectrum of lengthscales. Contemporary methods typically describe the field on a grid and estimate structures which can be resolved by it. In contrast, we address the reconstruction of grid-resolved structures as well as estimation of statistical summaries of subgrid structures, which are smaller than the grid resolution. We perform this in two different ways (a) via a physical (phenomenological), parameterized subgrid model that summarizes the impact of the unresolved scales at the coarse level and (b) via multiscale finite elements, where specially designed prolongation and restriction operators establish the interscale link between the same problem defined on a coarse and fine mesh. The estimation problem is posed as a Bayesian inverse problem. Dimensionality reduction is performed by projecting the field to be inferred on a suitable orthogonal basis set, viz. the Karhunen-Loeve expansion of a multiGaussian. We first demonstrate our techniques on the reconstruction of a binary medium consisting of a matrix with embedded inclusions, which are too small to be grid-resolved. The reconstruction is performed using an adaptive Markov chain Monte Carlo method. We find that the posterior distributions of the inferred parameters are approximately Gaussian. We exploit this finding to reconstruct a permeability field with long, but narrow embedded fractures (which are too fine to be grid-resolved) using scalable ensemble Kalman filters; this also allows us to address larger grids. Ensemble Kalman filtering is then used to estimate the values of hydraulic conductivity and specific yield in a model of the High Plains Aquifer in Kansas. Strong conditioning of the spatial structure of the parameters and the non-linear aspects of the water table aquifer create difficulty for the ensemble Kalman filter. We conclude with a demonstration of the use of multiscale stochastic finite elements to reconstruct permeability fields. This method, though computationally intensive, is general and can be used for multiscale inference in cases where a subgrid model cannot be constructed.
BibTeX
@misc{marzouk2011bayesian,
  author = {Youssef Marzouk and Bart van Bloemen Waanders and Matthew Parno and Jaideep Ray and Sophia Lefantzi and Luke Salazar and Sean Andrew McKenna and Katherine A. Klise},
  title = {{Bayesian data assimilation for stochastic multiscale models of transport in porous media.}},
  year = {2011},
  doi = {10.2172/1030232},
}
Figure