LAUSR

Abstract Injection of CO2 into brine-saturated reservoir rocks reduces their elastic moduli and thus changes the seismic response. However, estimation of the stiffness reduction and 3D plume morphology have large uncertainty for regular Signal-to-Noise Ratio (SNR) and amount of prior geological information. This paper examines achievable accuracy of the time-lapse seismic inversion based on Stage 2C of the CO2CRC Otway Project, which was specifically designed to test the sensitivity of seismic monitoring. Thanks to rich geological dataset, we could build a set of adequate subsurface models to optimise the inversion workflow and assess its capability. Firstly, 1D stochastic simulations and analytical models are used to estimate the effects of imperfect repeatability and limited bandwidth of the seismic (SNR ≈ 4). Then, we test a prototype inversion workflow on a virtual seismic survey - full-scale time-lapse synthetic seismic produced by 3D simulations in a detailed full-earth model. This test shows that the errors associated with approximate nature of the seismic inversion algorithm reduces SNR to 2.1. Furthermore, detectable thickness of the plume reduces to 10 m when the plume is laterally finite. The key findings of the synthetic study help design the inversion workflow for the Stage 2C field data. Inverted parameters of the CO2 plume agree well with independent measurements, including repeat pulsed-neutron logging, in- and above-zone pressure monitoring and borehole seismic. To extract the plume body from the noisy inversion output, we use a Neyman-Pearson detector augmented by a spatial connectivity constraint. The proposed extraction criterion is then employed to compare history-matched reservoir with the monitoring data. For the simulated CO2 distribution, the proposed workflow would detect 70% of the plume areal footprint. The missed samples correspond to thin parts of the plume with low saturation, and hence their effect on the estimated total mass of CO2 in the reservoir is minimal.