Integrated post-stack seismic inversion, multi-attribute analysis, and machine learning for carbon storage

The Gulf of Mexico known for its prolific hydrocarbon potential and extensive oil and gas infrastructure, makes it a promising candidate for carbon dioxide storage. The significant carbon storage potential of depleted hydrocarbon reservoirs and saline aquifers could significantly contribute to addressing climate change mitigation. Within this region, Mississippi Canyon Block 118 (MC-118) presents both opportunities and challenges for CO2 storage implementation. While MC-118 has abundant 3-D seismic data, it faces two main limitations: a scarcity of direct subsurface measurements from well logs, and structural deformation from salt tectonics that have caused faulting and reduced data quality at various depths. This research evaluates MC-118's carbon storage potential through advanced seismic analysis techniques, including post-stack inversion and machine learning methods such as K-means clustering and self-organizing maps. The study identified storage capacity in several phases. First in chapter two, using the Department of Energy's (DOE) methodology and the Offshore CO2 Saline Storage (OCSS) Calculator alongside post-stack seismic inversion, a potential storage capacity of 10-20 megatons of CO2 was estimated in reservoirs with existing well control. In the third chapter, self-organizing maps were employed to identify additional storage reservoirs in areas lacking well data. This analysis relied on various seismic attributes, with Aberrancy, Similarity, Peak Frequency, Envelope, and Curvature proving most effective for identifying key geological features including gas hydrates, high-quality sands, salt structures, and faults. Finally, in chapter three, by comparing self-organizing maps with K-means clustering and post-stack inversion results, the research identified an additional 2 megatons of potential storage capacity in areas without well control. This study also examined the potential for these unsupervised machine learning methods for fault identification, which is crucial for understanding both seal integrity and reservoir compartmentalization.