“The major pitfall of machine learning of any kind is to be overly confident in the results. We run the risk of garbage in gospel out.”

This discussion offers a rare chance to go a little deeper into a Leading Edge article and hear directly from the authors about the thinking behind their workflow. Satinder Chopra and Kurt Marfurt walk through how unsupervised machine learning, careful attribute selection, and simple preprocessing steps can reveal subtle channel features in a deepwater New Zealand example. It feels less like a theory lesson and more like practical guidance on using machine learning as a helpful partner in everyday seismic interpretation.

Key Takeaways

• Small workflow choices have big impact. Clean input data, thoughtful attribute selection, and simple normalization steps often determine whether machine learning highlights geology or just amplifies noise.
• The value is in the combination of tools and judgment. Unsupervised methods quickly expose patterns, but interpreters still need to compare results with seismic sections, wells, and regional context to confirm what is real.
• PCA and SOM make complex attribute sets easier to explore. By reducing dozens of attributes into clearer clusters, they help interpreters see channel shapes and reservoir variability that might otherwise be overlooked.

Links

• Read the December 2025 special section
• Seismic characterization with unsupervised machine learning applications for facies classification by Satinder Chopra and Kurt Marfurt

Download this episode

Please visit this page to download this episode as an .MP3 file.

Show Credits

Andrew Geary at TreasureMint hosted, edited, and produced this episode. The SEG podcast team comprises Robin Dupre, Kathy Gamble, and Ally McGinnis. 

If you have episode ideas or feedback for the show or want to sponsor a future episode, email the show at [email protected].