Abstract CSelecting appropriate rock physics models in geologically complex settings with limited core or lab measurements poses a significant challenge for reliable calibration and property prediction. This is further hindered in reservoirs with multiple stacked pay zones and varying depositional environments, where separate rock physics models (RPMs) are typically required for different rock types to ensure accurate log prediction. This study presents a rock physics-guided machine learning workflow (RPML) that integrates statistical learning with physically constrained rock physics models to predict facies, elastic, and petrophysical properties across multiple wells simultaneously. The method applies an Expectation–Maximization (EM) algorithm within a maximum-likelihood framework in which facies probabilities and RPM fitting parameters are jointly updated while remaining constrained within physically realistic bounds. This enables automatic calibration of rock physics templates while simultaneously generating facies classifications and log predictions. The workflow is tested in two geologically distinct settings: the Athabasca Oil Sands in northeastern Alberta and the Barnett interval of the Midland Basin. Using common well-log inputs (gamma ray, density, neutron porosity, resistivity, and elastic logs where available), RPML calibrated rock physics models for multiple facies and generated consistent elastic and petrophysical predictions across wells. Results from training and blind wells demonstrate that the method can reconstruct missing log suites, estimate elastic and petrophysical properties from limited inputs, and generate per-facies depth trends consistent with regional geology. The workflow not only streamlines RPM calibration but supports reservoir characterization workflows such as log repair, well-to-seismic ties, and construction of low-frequency models for seismic inversion, thus enabling faster and more reliable interpretation in data-scarce environments.
Bhatnagar et al. (Thu,) studied this question.