Stratigraphical Consistent Seismic Profile for Geologically Informed Machine Learning Interpretation

David J. Emery, Daniel O. Trad

Machine Learning solutions have become increasingly popular and should be a natural tool for seismic stratigraphic and seismic facies analysis. As a detailed stratigraphic analysis is time-consuming, it tends to be done post-structural interpretation. These methods are based on pattern recognition, meaning an experienced interpreter could use these machine-learning observations to prioritize their initial structural interpretation. Moving this time-consuming and multi-dimensional analysis earlier in a workflow and flagging potential explanations without personal bias should significantly improve sub-surface analysis.Published geophysical machine learning solutions have generally focused on extending an interpretation to adjacent uninterpreted lines. The learning data is, therefore, from the volume being interpreted, and analysis becomes a style of auto-tracking. The alternative is to create synthetic stratigraphically consistent seismic profiles for training and use these models to provide a first-pass analysis before starting an interpretation. This report focuses on creating a stratigraphically consistent profile using geological principles. The intent is to modify the existing 2D code into a 3D for publication for consortium members next year.