David Handwerger

Effective Core Sampling for Improved Calibration of Logs and Seismic Data

Abstract Core data are commonly used to calibrate reservoir parameters with wireline logs, so that the log data can be used to propagate the reservoir properties over the zone of interest or into three-dimensional reservoir models. However, such a calibration is often suboptimal, either because it is built without incorporating basic tenets of statistical modeling or built with biased samples. We discuss pitfalls in calibration between data from different sources (e.g., core, log, and seismic) using linear regression because it is the most commonly used modeling method. Many of the problems with fitting a regression model between core analyses, wireline logs, and/or seismic data can often be mitigated in the sample selection process. It is advantageous to choose samples with limited bias, and in a manner that accounts for the heterogeneity inherent in most unconventional reservoirs. By acknowledging both the assumptions inherent in statistical modeling and the formation heterogeneity, one can collect the data necessary for construction of a calibration model efficiently without necessarily requiring more samples. To achieve this, it is useful to deconstruct the log response into zones of similar bulk material properties, and then sample each zone in a relevant manner.

Using Multivariate Data Classification on Frontier Exploration Basins to Enhance the Information Value of Suboptimal 2D Seismic Surveys for Unconventional Reservoir Characterization

We developed a workflow that allows integrating legacy 2D seismic surveys with modern log and core data, validating their consistency, classifying them into rock classes with consistent properties, propagating material properties across each of these rock classes, and using this information to improve reservoir characterization and the assessment of their hydrocarbon resource potential. As proof of concept, we analyzed two intersecting 2D seismic lines shot in 2001 in a frontier basin in Canada to determine the distribution of reservoir quality. Each of these had been separately prestack inverted, but have modern core and log data (as well as legacy log data) which were integrated with the inverted attributes. Results identify a most prospective class for reservoir quality within the zone of interest, and show that it increases in thickness to the south in the seismic section.