Centre-developed seismic foundation model is now open source!

The NCS model, a seismic foundation model trained on data from the Norwegian data repository for subsurface data, is now available as an open-source model, allowing anyone to download, utilize, and further develop it.

By Petter Bjørklund, Communications Officer at SFI Visual Intelligence

Analyzing seismic data is essential for understanding the subsurface. This data is acquired onshore or offshore by sending acoustic waves into the ground to create images of the Earth's subsurface, revealing rock layers and important geological structures.

However, seismic data interpretation is often a time-consuming and resource-intensive task. Now, Visual Intelligence researchers have developed and openly published a foundation model which can provide geologists with more efficient tools for interpreting seismic data.

"A strong starting point for automated seismic interpretation"

The model, dubbed the NCS (Norwegian Continental Shelf) model, is trained on data from almost all public seismic surveys available in DISKOS: the Norwegian subsurface data repository. A goal of the model is to contribute to more efficient analyses of seismic data and subsurface structures – to support applications such as energy exploration or carbon capture and storage.

Alba Ordoñez is a Senior Research Scientist at the Norwegian Computing Center. She developed the model alongside NR colleagues Anders Ueland Waldeland and Theodor Johannes Line Forgaard, and says it provides a number of benefits for geologists.

"The NCS model provides a strong starting point for automated seismic interpretation. Because it is trained directly on seismic data, the patterns it learns are better suited to seismic tasks than those learned by foundation models trained on natural images," Ordoñez says.

Foundation models, also called general models, are AI models trained on very large datasets using self-supervised learning.

While traditional AI methods usually depend on manual annotations for a specific task, self-supervised learning allows a model to learn directly from the data itself, without needing labeled examples.

"This helps them capture the underlying structure and broader patterns in the data. As a result, foundation models can be adapted to many different applications, such as language, vision, and increasingly in scientific fields like Earth observation and medicine," Ordoñez explains.

Key for research and innovation

The NCS model can be utilized as a basis for developing new AI models for various seismic tasks. As it has taught itself to recognize seismic structures, the model can then be used for purposes such as interactive data interpretation.

Making it open source means that anyone can download, utilize, and further develop the model – which is key for research and innovation, as well as future development of new commercal products and services.

"We released the model as open source to make advanced seismic AI tools accessible to the wider geoscience community, while also enabling faster progress in the field through openness, reuse, and collaboration," Ordoñez says.

User partners Equinor and Aker BP participated in the NCS model project. Their involvement reflects a strong interest in methods that can make seismic interpretation more efficient, scalable, and useful across a wide range of subsurface applications.

The links below provide access to the paper preprint, code, and model weights:

Paper preprint

Code

Model weights