Probabilistic Geological Information: Elicitation, Cognitive Bias and Herding

Conceptual models (structural, stratigraphic, diagenetic) of the origin of observed reservoir geology are important for predicting reservoir heterogeneity or behaviour beyond the limits of our ability to measure such information directly. However, such models are often qualitative, and based on the prior experience of particular geologists. Even within each conceptual framework it is difficult to quantify the relative likelihood (probability) of occurrence of different geological processes, and of the different values of controlling parameters. This has a significant impact on our ability to assess the risk in an oilfield asset.

By contrast, methods used to interrogate a reservoir directly are predominantly geophysical in nature, and are usually based on quantitative data. Geophysical data alone is never, however, able to constrain a reservoir model uniquely, and the injection of additional information is always necessary. An improved workflow that creates a direct interface between geologists' qualitative, process-based knowledge, and geophysicists' quantitative information would therefore be of considerable utility.

This talk will introduce novel methods in Bayesian uncertainty analysis applied to oilfield geological models, and in the field of elicitation theory (which incorporate statistics, cognitive psychology and computer science) to motivate new workflows that quantify geological knowledge explicitly and probabilistically. These enable geological information to be used quantitatively by integrating such information within a joint geological-geophysical workflow so as to reduce uncertainty in 2D/3D/4D geophysical models directly. The result is a more robust geological and geophysical interpretation of reservoir models, and ultimately more accurate prediction of reservoir behaviour during production.