Estimating CO2 Saturation Maps from Seismic Data using Deep Convolutional Neural Networks: A Case Study at Frio-II Injection Site

Current methods to monitor sequestered CO2 involve extensive petrophysical and geophysical processing steps which are tedious, expensive, and subjected to human error, all of which impede accurate CO2 estimation in the subsurface. Monitoring underground sequestered CO2 plume is crucial to prevent unwanted environmental issues from occurring (e.g., CO2 plume subsurface contamination). To date, the direct estimation of CO2 saturation from seismic gathers is not well-studied. Neural networks (NNs) have seen burgeoning applications in highly non-linear inversion problems pertaining to geophysics such as in seismic inversion. We investigate the use of supervised NN learning to estimate CO2 saturation maps directly from seismic data. Our work is divided into two parts, Part I and II. Part I explores the use of NNs to invert for CO2 maps under synthetic settings. In Part II, we explore the NN inversion method on field crosswell seismic data acquired at Frio-II CO2 injection site in Texas. We compare the inverted field results with conventional physics-based inversion results from literature review. Our results indicate fair data matching at the first arrivals along with reasonable CO2 plume estimates.