WP2 - Unit boundary detection

Interpreting geological structures from seismic images is a time-consuming task for big projects, that requires distinct experience and domain knowledge. Additionally, multiple domain experts will interpret the data differently based on different backgrounds (sedimentology, structural geology, etc.) implying that there is a subjective uncertainty which is not accounted for numerically at present. This work package aims to address all these points, where the solution could also conceivably enable new possibilities for working with geophysical data.

Recent advances in deep learning have demonstrated the vast potential of deep neural networks for a range of inputs and tasks. In this work we will draw inspiration from the field of computer vision, where the current paradigm is to use “foundation models” which are trained in a self-supervised way which allows for training on vast datasets without the need for annotated data.

Sheng et al. (2025a) used these techniques to train a foundation model based on high resolution 3D seismic data, commonly used in petroleum exploration. The authors managed to acquire a dataset of size comparable to well-known benchmark datasets in computer vision. As a contribution to further research the authors also made the data, code and model open-source through their GitHub repository (Sheng et al., 2025b). The plan for this work package is to build on this work and apply the model for ultra-high resolution seismic (UHRS) data.

The aim is to apply this foundation model for the downstream task of discerning changes in layering within a framework which allows for sampling as a means of uncertainty quantification. We aim to perform benchmark tests against real projects to test and quantify the efficiency of working with the proposed method. Furthermore, the deep learning approach will be tested against other seismic software and evaluated for the end purpose.

WP2 will further investigate possibilities and limitations of alternative software for interpreting seismic data including uncertainty quantification, focusing on what is important for ground modelling in offshore wind.

References:

Sheng, H., Wu, X., Si, X., Li, J., Zhang, S., & Duan, X. (2025a). Seismic foundation model: A next generation deep-learning model in geophysics. GEOPHYSICS, 90(2), IM59–IM79. https://doi.org/10.1190/geo2024-0262.1

Sheng, H., Si, X., & Li, J. (2025b). Shenghanlin/SeismicFoundationModel [Python]. https://github.com/shenghanlin/SeismicFoundationModel (Original work published 2023)