Shawn M. Len1, Wayne D. Pennington1
(1) Michigan Technological University, Houghton, MI
ABSTRACT: Neural Network Analyses of Seismic Attributes and Facies at the Stratton Field, South Texas
Seismic "attributes" often consist of single-valued properties extracted from seismic data. Using combinations of attributes can increase the likelihood of finding a correlation with reservoir properties. Going one step further, seismic "facies" derived from the waveshape implicitly incorporate a number of attributes, such as amplitude and frequency. In thin-bedded reservoirs, correlation with a specific seismic features, such as peaks or a troughs, is not always practical; instead, thin stratigraphic features affect waveform shapes in subtle and, perhaps, non-intuitive ways. In such instances, seismic facies classification based on waveshape (implemented through neural networks) may provide an optimal approach to seismic reservoir identification and characterization.
The public-domain data set at the Stratton Field provides an excellent test case for examining the physical basis of attribute analysis and facies classifications in thin-bed reservoirs. Here, reservoir beds are thin, and difficult to identify from typical seismic interpretation techniques. We have used neural network programs to provide relationships between seismic attributes and reservoir properties, and to provide seismic classifications based on waveshape. Although each approach leads to results that appear, at first glance, to be geologically meaningful, there are a number of pitfalls that may be encountered. We have attempted to identify the results that appear to be spurious, and to provide a physical basis for those results that seem to be meaningful. To prevent erroneous conclusions, any study making use of arithmetically derived correlations should also include a study of the physical basis behind the correlations found.
AAPG Search and Discovery Article #90906©2001 AAPG Annual Convention, Denver, Colorado