Unlocking the Potential of AI in Revolutionizing the Seismic-Driven Subsurface Workflows

Abstract Artificial intelligence (AI), especially deep learning (DL) methods, is revolutionizing the seismic-driven subsurface workflows, enabling more accurate understanding of subsurface structures and properties. Due to the sheer size of seismic data, applying DL methods to seismic data at scale has proven to be nontrivial. In this work, with DL applications ranging from preprocessing, imaging, to enrichment, we demonstrate how AI running on high performance computation devices adds value throughout the life cycle of seismic-driven workflows. The first solution is first arrival picking that substantially reduces the total turnaround time, by minimizing the QC process that usually requires significant user interaction. The second solution is automatic salt body interpretation for seismic imaging, which uses innovative DL models and latest computation resources to eliminate the costly horizon-based manual salt interpretation. Finally, is a 2D-to-3D seismic image conversion solution, crucial for initial reconnaissance of new frontier and CCUS site screening and monitoring. Such conversion is only feasible with advanced DL workflow and GPU devices with large on-device memory. We present all three solutions with field data examples, while also discussing the importance of high-performance cloud computing infrastructure as the key enabler of such AI solutions in the subsurface domain.