Imagine a machine learning model, more familiar with the seismic structures on the Norwegian Continental Shelf (NCS) than any other model – or human expertise. This is the NCS Seismic Foundation Model, trained on all publicly available post-stack seismic data (30 TB) in the DISKOS database.
At the Dig X Subsurface 2025 conference in Oslo in December, Senior Research Scientist Anders Waldeland at Norwegian Computing Center will present the first-of-its-kind model, funded by the Research Council of Norway and developed with industry partners Aker BP and Equinor.
It leverages 1,050 seismic cubes to master geological features like continuity, layering, and amplitude textures without human labels. Using a masked autoencoder approach, it reconstructs random 2.5D and 3D patches, trained on Norway’s Olivia supercomputer with 1,000 GPU hours.
The result? A model that “knows” NCS seismic structures, enabling faster, more robust fine-tuning for tasks like salt-body delineation, flatspot mapping, and geological package segmentation.
Tested against global seismic models, the NCS model excels in accuracy and generalization, using a simple classifier on extracted features to deliver high-quality predictions across multiple interpretation tasks.
Waldeland will share results, including segmented salt bodies and injectite mappings, showcasing the model’s edge for NCS exploration.
This leap toward AI-driven seismic workflows paves the way for smarter, more efficient subsurface analysis. Future applications may include AVO exploration and real-time geobody tools.
Register for the Dig X Subsurface 2025 conference at Scandic Fornebu in Oslo, December 03-04, 2025, to learn more about this groundbreaking model. The program can be found on the conference website.