Full waveform inversion for CO₂ monitoring: a feasibility study at the CMC Newell County facility

Ninoska Amundaray

Carbon capture and storage stands out as a method to aid in mitigating the effects associated with the anthropogenic emissions of carbon dioxide (CO₂). However, to adopt this technology, an understanding of the monitoring methods that aid to ensure the accuracy, conformance, and containment of the injecting fluid in the subsurface must be refined. The Carbon Management Canada Newell County facility in southern Alberta simulates a controlled leakage of CO2 from a deeper injection target as an alternative to assess geophysical and geochemical methods in their capacity to monitor the development of a CO₂ plume. Seismic monitoring of CO₂ encompasses stages of acquisition, processing, and inversion. This thesis deals with the last part through full waveform inversion (FWI) by examining an acoustic and elastic wave propagation for three stages of CO₂ injection using vertical seismic profiles (VSP). VSP are among the seismic acquisition surveys assessed at the study site. These are attractive for seismic inversion due to their high signal-to-noise ratio, preservation of frequency content, and fixed acquisition. The simulated acoustic experiments examined in this thesis set expectations about the resolution of velocity structures. While, the elastic experiments simulate complexities that are closer to those expected in the subsurface, providing a better view of the challenges of multi-parameter FWI for the study site. To mitigate the latter, I suggest implementing a sequential approach for multi-parameter cases and conditioning the inversion utilizing prior information. This constraint was assessed through the model-fitting and the data-fitting term of the objective function for FWI. Overall, inverted results demonstrate reliable P-wave and S-wave velocity structures but generally poor density models. Although the geometry of the CO₂ effects was imaged with relative accuracy in the neighbouring areas to the injection well in most examples, with increasing offset, this became more challenging, especially for S-wave velocity and density. Approximately 50% to 70% of the expected variation of P-wave velocity was reconstructed between stages, and about 50% of the S-wave variation was also estimated in the best-case scenarios.