Paper submitted to SEG Advancing Data Analytics & Machine Learning for Exploration Geophysics Workshop, by Mark Roberts, Olga Brusova, David Brookes, Leandro Gabioli and Alejandro Valenciano (TGS)
Summary
In offshore seismic imaging, reducing turnaround time is critical for project success. While computational power has increased, the rising complexity of algorithms like deblending and deghosting has kept production timelines static. Traditionally, "fast-track" products sacrificed quality for speed. This work presents a hybrid workflow that integrates Machine Learning (ML) with physics-based foundations, extending previous hydrophone-only successes to multi-component (geophone and hydrophone) streamer data. By targeting the most parameter-sensitive stages, such as denoising, and deghosting, this approach eliminates weeks of manual effort while maintaining high-fidelity results.
The implementation of a Deep-Learning based velocity estimation allows the rapid generation of stacked seismic sections to provide an initial look at the data and facilitate the QC process.