Robust facies classification from 3D seismic data plays a key role in successful reservoir identification and characterization. Machine learning, particularly supervised convolutional neural network (CNN), has been extensively implemented for both improved efficiency and accuracy of seismic facies classification in the past years. In most seismic applications, however, the amount of available expert annotations is small, which raises the risk of overfitting a CNN particularly when only seismic amplitudes are used. In such a case, the trained CNN would have poor generalization capability, causing the facies interpretation of obvious artifacts, limited lateral consistency, and thus restricted application to subsequent geological modeling workflows. This study proposes to address such an issue by using relative geologic time (RGT), which explicitly preserves the large-scale continuity of seismic patterns, to constrain a seismic facies identification CNN. Such constrained learning is enforced in two ways: (1) from the perspective of input, the RGT is used as an additional feature channel besides seismic amplitudes; and more innovatively (2) the CNN has two output branches, with one for matching the target seismic facies and the other for reconstructing the RGT. The performance of such an RGT-constrained CNN is validated by interpreting six facies in the Parihaka dataset. Compared to facies predictions purely from seismic amplitudes, those from using the proposed RGT constraint demonstrate significantly reduced artifacts and improved consistency throughout the entire seismic survey.

This content is only available via PDF.
You can access this article if you purchase or spend a download.