Abstract

AVO crossplotting of intercept and gradient attributes has been a mainstay of AVO interpretation and analysis for more than two decades. In binary sand-shale settings, the viability of a lead or prospect can be further evaluated by identifying a background trend of wet sands and shales, and observing how removed the prospect's intercept-gradient pairs are from the background trend using small temporal or stratigraphic-consistent windows. This technique has primarily been performed visually, and, as such, contains inherent interpretation bias. While overall the visual methodology of the technique has been successful, methods to reduce such errors are valued as prospect economics become more challenging. To mitigate this interpretation bias, an unsupervised neural network methodology is presented. The unsupervised vector quantizer (neural network) assists in subgrouping (classifying) those seismic pairs that comprise the AVO background trend and those pairs that may subgroup as outliers — possibly indicative of hydrocarbon-saturated reservoirs. With this methodology, a clearer determination of what is AVO background trend and what is AVO outlier is presented using 2D seismic data in a frontier exploration province. Use of unsupervised neural networks for AVO crossplot assessment is accompanied with additional benefits that further assist in understanding geologies or seismic facies within the AVO crossplot subset.

You do not currently have access to this article.