are usually directed toward maximizing the repeatability of the final image. Given the amount of effort devoted to realizing this goal, it is reasonable to ask if we ever reach the point where we have all the repeatability we need. More precisely, does repeatability ever become sufficiently good that further improvement is not worth the additional effort? This article addresses this question using model-based value of information analysis (VOI).Although there has been much discussion of how repeatability might be improved, there has been less discussion of the benefits of improved repeatability. Part of the reason for this might be that, when time-lapse seismic was in its early stages of development, repeatability tended to be poor and the need for improvement was beyond question. Recently, however, surveys with average nrms differences of less than 20% have become common, and cases have been reported where nrms values below 10% have been achieved; see, for example, Smit et al. (2005). Improving on these levels of repeatability is likely to require permanently installed receivers or costly, special purpose acquisition and processing methods.These higher acquisition and processing costs might be justified if the improved repeatability produces an offsetting improvement in the outcomes of reservoir management decisions. VOI analysis is an approach that quantifies the economic value of a data set based on its expected impact on business decisions. This article uses VOI to estimate the economic value of improvements in image repeatability. After a brief description of how a model-based implementation of VOI can be applied to time-lapse seismic, I will show two examples using VOI to quantify the effect of different levels of repeatability. Both examples involve selecting locations for infill drilling based on pore fluid predictions made from time-lapse data sets. Nonseismic information is an important contributor to drilling decisions, so the analysis includes variations in both seismic and nonseismic uncertainty.Model-based VOI. The premise behind VOI is that information has economic value to the extent that it reduces the uncertainty associated with a decision that involves money. Because of this emphasis on decisions, a decision-tree approach has traditionally been used for VOI analysis. For complex situations involving multiple dependent variables, however, decision trees become too large to be practical and Bayesian decision networks (BDN) are more appropriate. Bhattacharjya and Mukerji (2006) demonstrate the construction of a BDN, also known as an influence diagram, for the general problem of reservoir management using timelapse seismic. Figure 1 is a BDN representation of the decision problem addressed in this paper. The network in Figure 1 is much simpler than that developed by Bhattacharjya and Mukerji because it includes only the aspects of the problem that I have chosen to model. Figure 1 depicts the factors involved in deciding to drill an infill well at a particular loca-tion. The rounded rectangles indic...

No abstract

No abstract

Lithologic interpretations of amplitude variation with offset (AVO) information are ambiguous both because different lithologies occupy overlapping ranges of elastic properties, and because angle‐dependent reflection coefficients estimated from seismic data are uncertain. This paper presents a method for quantifying and combining these two components of uncertainty to get a full characterization of the uncertainty associated with an AVO‐based lithologic interpretation. The result of this approach is a compilation of all the lithologies that are consistent with the observed AVO behavior, along with a probability of occurrence for each lithology. A 2‐D line from the North Sea illustrates how the method might be applied in practice. For any data set, the interaction between the geologic and measurement components of uncertainty may significantly affect the overall uncertainty in a lithologic interpretation.

No abstract