A Sparse Bayesian Estimation Framework for Conditioning Prior Geologic Models to Nonlinear Flow Measurements

We present a Bayesian framework for reconstruction of subsurface hydraulic properties from nonlinear dynamic flow data by imposing sparsity on the distribution of the solution coefficients in a compression transform domain.

💡 Research Summary

The paper introduces a novel Bayesian inversion framework designed to condition prior geological models with nonlinear dynamic flow measurements while exploiting sparsity in a transformed coefficient space. Traditional Bayesian approaches to subsurface property estimation suffer from the curse of dimensionality and the computational burden of repeatedly solving nonlinear flow equations. The authors address these challenges by (1) projecting the high‑dimensional hydraulic property field (e.g., permeability, porosity) onto a compression transform domain such as discrete cosine transform (DCT) or wavelet bases, and (2) imposing a sparsity‑promoting hierarchical prior on the transform coefficients using an Automatic Relevance Determination (ARD) scheme.

In the ARD hierarchy each coefficient (w_i) is assigned a zero‑mean Gaussian prior with its own precision (\alpha_i). Large values of (\alpha_i) shrink the corresponding coefficient toward zero, effectively pruning irrelevant basis functions and automatically determining the intrinsic dimensionality of the problem. The hyper‑parameters ({\alpha_i}) and the observation noise precision (\beta) are learned from the data, eliminating the need for manual regularization tuning.

Because the forward model – typically a set of nonlinear partial differential equations governing multiphase, anisotropic flow – is highly nonlinear, the authors employ an iterative linearization strategy. At iteration (k) the forward operator (\mathcal{F}(\mathbf{w})) is approximated by a first‑order Taylor expansion around the current estimate (\mathbf{w}^{(k)}):
\