Using Simulation Models to Predict Uncertainty (Just Because We Can Run Millions of Cells Doesn't Mean We Should)

Larry W. Lake
The University of Texas at Austin

Risk assessment and management have always been at the core of hydrocarbon producing endeavors. Yet to date nearly all risk assessment has been qualitative; so-called quantitative risk assessment remains a new and potential fertile area of application. 

Much current practice in estimating hydrocarbon recovery is based on numerical simulation of flow through reservoirs. It has become axiomatic that incorporating more data into such models leads to greater confidence in its predictions. This statement has not been proven to be true, and, indeed, cannot be true in all cases. Furthermore, incorporating more data into a simulation model is not easy because "data" comes in all forms, covers a disparity of volumes of investigations, and many times becomes available only while the resource is being produced. This presentation deals with that portion of risk assessment known as uncertainty estimation. 

The purpose of this presentation is threefold; to

1. Review the history and benefits or large-scale flow modeling with emphasis on the benefits of history matching.
2. Enumerate the basic features that we have learned from several years of history matching.
3. Review to statistical concepts of bias and uncertainty.
4. Illustrate the benefits of inclusion of seismic data into a reservoir description, particularly with regards to how this type of data (low resolution, but exhaustive) affects bias and precision.

The presentation concludes with speculations about how much complexity is needed to model reservoir performance given uncertainties in data used for input.