Interpretation of well logs and core data via Bayesian inversion

Abstract

Estimating in situ petrophysical and compositional properties of rocks (e.g., porosity, mineralogy, and fluid saturation) from well logs and core measurements is critical for the evaluation of subsurface fluid resources. Traditional multimineral analysis of well logs is susceptible to abnormal borehole and geometric conditions, such as shoulder-bed effects and tool/borehole-related biases. Moreover, traditional multimineral analysis rarely includes uncertainty quantification, much less assessments of the impact of measurement noise, borehole conditions, and/or inaccurate rock-physics models (RPMs) on estimated properties. Probabilistic formation evaluation with credible-interval estimations can improve the outcome of mineral and fluid evaluations from well logs by explicitly incorporating the effects of measurement biases and a priori compositional and petrophysical constraints in the estimation. A probabilistic workflow is developed to estimate petrophysical/compositional properties from well logs and available core data. The workflow consists of two sequential steps: borehole measurements (e.g., density, resistivity, and gamma ray) are first converted into a layer-by-layer earth model, which consists of piecewise constant property values, with associated uncertainty estimations. The estimated earth model is tool/borehole independent, allowing for a common baseline to effectively compare properties among neighboring wells. Next, the estimated earth-model physical properties are used to derive petrophysical properties (e.g., volumetric concentration of fluid and solid rock constituents) via petrophysical joint inversion. In each step, Bayesian inversion is implemented with Markov chain Monte Carlo sampling in contrast to deterministic algorithms used by traditional multimineral methods. The developed method has the following advantages over traditional approaches: (1) it propagates uncertainty from well logs to petrophysical properties by assimilating measurement noise, tool/borehole-related biases, and RPM errors and (2) it enables the implementation of a priori knowledge per rock class to constrain the petrophysical estimation, effectively integrating field-specific core/regional geologic data. Examples of thinly laminated and organic shale formations successfully verify the reliability, efficiency, and accuracy of our estimation method.