The decision-oriented world: effective management of uncertainty in geomodelling workflows

Abstract Given a realistic set of sub-surface data, the number of plausible geological models that can be constructed can run into the millions or tens of millions. It is therefore impractical for a geologist to examine the entire space of possible models and thereby correctly characterize uncertainty. Commonly applied approaches to this problem use simplifications such as experimental design, parameter sensitivity analysis or Monte Carlo simulation to attempt to reasonably account for input uncertainty while balancing computational expenditure and time. Broadly, this is undertaken by assessing which model inputs (for example porosity or permeability) have the most significant impact on response variables of choice, typically volume based or commercial metrics such as net present value (NPV) or estimated ultimate recovery (EUR). Specific parameters are then perturbed to produce an assumed representative subset (in many cases a low, mid and high case) of the range of possible models (see for example Pyrcz and Deutsch, 2014).