Bayesian inference of petrophysical properties with generative spectral induced polarization models

Abstract

Mechanistic induced polarization (IP) models describe the relationships between the physical properties of geomaterials and their frequency-dependent complex conductivity. However, practitioners rarely use mechanistic models to interpret the IP data because the uncertainties associated with estimating petrophysical properties from complex conductivity spectra are still poorly understood. We propose a framework for critically assessing any IP model’s sensitivity and parameter estimation limitations. The framework consists of a conditional variational autoencoder (CVAE), an unsupervised Bayesian neural network specializing in data dimension reduction and generative modeling. We apply the framework in a case study of the “perfectly polarized interfacial polarization” model by training the CVAE on the IP signatures of synthetic mixtures of metallic mineral inclusions hosted in electrolyte-filled geomaterials. First, the CVAE’s Jacobian reveals the relative importance of each petrophysical property for generating the spectral IP data. The most critical parameters are the conductivity of the host, the volume fraction of the inclusions, the characteristic length of the inclusions, and the permittivity of the host. Contrastingly, the inclusions’ diffusion coefficient, permittivity, and conductivity, as well as the host’s diffusion coefficient, have marginal importance. A parameter estimation experiment using various model constraints yields the standardized accuracy of petrophysical properties and corroborates the sensitivity analysis results. Finally, we visualize the effects of data transformations and model constraints on the petrophysical parameter space. We conclude that a common logarithm data transformation yields optimal parameter estimation results and that constraining the electrochemical properties of a geomaterial improves the estimates of the size of its metallic inclusions and vice versa.