An ongoing challenge to seismic interpreters is to identify and extract heterogeneous seismic facies on data volumes that are continually increasing in size. Geometric, geomechanical, and spectral attributes help to extract key features but add to the number of data volumes to be examined. Common interactive analysis tools include crossplotting, interactive animation, and 3D corendering where we examine more than one attribute at a time.

When there are more than three attributes, principal component analysis, independent component analysis, self-organizing maps, and generative topographic mapping mathematically reduce the dimensionality of the data to a more manageable subset.

Corendering different components and mapping against a 2D colorbar provides a means of user defined clustering, similar to that provided by k-means. The result of such unsupervised clustering is the identification of color-coded voxels that have similar expressions. Although data reduction and clustering techniques extract important patterns across attribute volumes, the interpretation of these patterns is like traditional interactive interpretation, where the interpreters integrate their geologic understanding of the depositional environment and tectonic deformation with well control to map areas that are more prospective or pose drilling hazards.

In contrast, supervised classification such as Bayesian classification, probabilistic neural networks, and convolutional neural networks provides a statistical estimate of how likely any given voxel corresponds to one or more interpreter provided “labels”. Labels may include interpreter-painted seismic facies, hand-picked faults, or attribute vectors extracted about different facies, drilling problems, or fluid flow encountered by well bores.

For both supervised and unsupervised machine learning analysis the interpreter can improve the results by using their understanding of the geology to provide a judicious choice of inputs and training data.

A novel part of the course is a hands-on component to compute attributes and seismic facies using software developed by the Attribute Assisted Seismic Processing and Interpretation (AASPI) consortium. Supplied data volumes are limited to those that are publicly available. However, participants will be able use a copy of the software for an additional three months at home or in their workplace to allow them to continue learning and identify which workflows provide useful results for their own non-publicly available data.