Quantifying Uncertainty in Subsurface Systems (QUSS), aka Geophysical Monograph 236, by Céline Scheidt, Lewis Li and Jef Caers (all from Stanford University) is a co-publication of the American Geophysical Union and John Wiley and Sons.
QUSS is a textbook, almost a work of literature, that provides broad coverage of current subsurface assessment in the fields of oil and gas, water, geothermal, storage and more. QUSS is bang up to date with for instance, the use of data science in North American shale development. QUSS covers much more than it indicated by its title, discussing the limits of a fundamental scientific approach in the face of complex geological and other phenomena. The authors compare predictive (forward) modeling from first principle with data-derived models. And that is before we get into quantifying uncertainty (QU) in all of the above.
QUSS takes quite a while to get around to the topic of the title. QUSS per se is discussed in a 50 page chapter starting on page 217! As the authors explain in the introduction, “the primordial question is not necessarily the quantification of uncertainty of all the [...] variables but [rather] of a decision-making process involving any of the target variables in question.” Such decision making might include whether to acquire more seismic data over an oil field, introducing the topic of value of information.
A chapter on decision making under uncertainty introduces the ‘science’ of decision analysis. This section is an extensive, well-illustrated overview of data science including dimension reduction, principle component analysis, regression (including boosted trees), kriging and kernel processes and cluster analysis. A further chapter covers Monte Carlo-based and its use in model simplification with “Sobol” regionalized sensitivity analysis.
A whole chapter is devoted to the philosophy behind Bayesian methods with a historical context explaining why Bayesianism is now a ‘leading paradigm’ for quantifying uncertainty. This leads-on to a presentation of the role of the prior distribution in both deterministic and stochastic inversion, geological heterogeneity and geostatistics. The philosophy of science approach is leavened with some interesting asides, including the attempts to model Saudi Arabia’s Ghawar field, where flowmeter measurement led to ‘ridiculous’ permeability values (200 Darcys!) and ad-hoc ‘fixes’ to previous models. There is a tendency in the modeling community to ‘shy away from bold hypotheses certainly if one wants to obtain government funding’ and the fact that modelers tend to ‘gravitate toward consensus under the banner of being good at team-work.’ The chapter concludes with a discussion of the nature of geological priors and their relationship to inter alia, model building and flume tank sedimentological experiments.
Chapter 7, billed as ‘the most novel technical contribution’ of QUSS, introduces a collection of methods called ‘Bayesian evidential learning’ (BEL). These address the problem of matching large, realistic geological models with limited computing resources. BEL leverages Monte Carlo methods to generate a training set of data and prediction variables that can ‘allow for predictions based on data without complex model inversions.’
QUSS is a fantastic compendium of terminology and methods addressing a wide range of subsurface problems. Some practitioners may find the comprehensive approach rather bewildering which is probably in the nature of the subject. But even the amateur decision-maker will find much to intrigue and challenge. QUSS undoubtedly merits a more leisurely read that this reviewer could afford. In our rapid run-through we spotted an amusing typo on page 23 where a geological model of a “buried valet” is discussed. Poor chap!
Quantifying Uncertainty in Subsurface Systems (QUSS) by Céline Scheidt, Lewis Li and Jef Caers all from Stanford University. Wiley ISBN: 978-1-119-32583-3.
© Oil IT Journal - all rights reserved.