Inversion-Based Multiwell Petrophysical Interpretation of Well Logs and Core Data via Adaptive Rock Physics Models

Formation evaluation, and specifically hydrocarbon volume estimations, are tightly dependent on the rock physics model (RPM) used for the interpretation of well logs and core data. The latter models are known to exhibit small but significant variations throughout multiple wells located in the same hydrocarbon field. To improve the accuracy and reliability of the interpretation, the RPMs are typically adjusted ad-hoc. We automate the multiwell interpretation process by relying on local petrophysical inversion of well logs and core data. A spatial correlation function is used to implement the RPMs, both vertically and laterally. In addition to improving formation evaluation in each well, our inversion-based method mitigates layer-boundary, geometrical, and instrument-related effects on well logs and identifies data outliers and measurement imbalances where further quality control might be needed. First, we invert each available well log into an equivalent physical property represented by a layer-by-layer blocky log with an associated uncertainty (earth model: piecewise constant layers with discontinuities at layer boundaries). This mitigates any tool, shoulder-bed, or borehole-condition dependency. Then, we use the extra measurements (well logs and core data) from a key well to determine an initial RPM (e.g., Juhasz parameters and density of minerals), as well as probabilistic prior distributions for all properties, e.g., porosity and water saturation. Next, we propagate the RPM and prior distributions throughout the field using Bayesian petrophysical/compositional joint inversion (PJI) for all petrophysical properties in every well, concomitantly propagating uncertainties to petrophysical/compositional properties. With each non-key well having a full set of physical (from well logs) and petrophysical/compositional properties, we generate new priors and RPMs for each well by minimizing the PJI misfit. These new priors and RPMs are used to further refine priors, and RPMs on neighboring wells. We enforce consistency via spatial variograms for RPMs. The process is repeated iteratively while tightening the variogram until no further improvement is possible. This method guarantees that the variation of RPMs is consistent across spatial correlations. The accuracy of the method is improved as more field data are available to corroborate and refine local RPMs and prior distributions. By using adaptive RPMs over tool and borehole-condition-mitigated layer properties, we were able to match core data constituted by porosity, fluid saturations, and mineral composition. Our results replicated 87% of the core data within the 95% confidence interval; in contrast, using a universal RPM replicates a lower 80% of the core data within the 95% confidence interval. Traditional interpretation methods cannot capture confidence intervals and yield significantly poorer matches in all properties; when comparing specifically hydrocarbon pore volume, our method shows an average 5% accuracy improvement. We generalized a logging tool and borehole-condition-independent Bayesian inference petrophysical estimation method to a multiwell framework. By considering the entire hydrocarbon field as a single petrophysical joint inversion of well logs and core data, we increased the accuracy of formation evaluation and/or identified outliers or data imbalances that signaled poor or biased data that required further quality control.