Stochastic Time Lapse Seismic Inversion for Monitoring CO2 Sequestration: CO2CRC Otway Project Modelling Study
Monitoring of a CO2 plume is an important component in any CO2 sequestration project because it can identify the occurrence of fingering, leakage through sub-seismic fractures in the cap-rock or permeability barrier not recognized on the 3D seismic data or present in the flow simulations. Early detection of such issues can result in adjustment of the project's plans and determine the volume of gas that can be injected. In this work we use the stochastic time lapse inversion algorithm Delivery 4D to invert a synthetic seismic data used in a feasibility study of CO2 injection at the Otway basin, Australia. The three main steps for the construction of the synthetic seismic datasets were: 1) Computation of synthetic seismograms for the 1D velocity models for different plume thicknesses with 5% of gas saturation. 2) Generation of 3D seismic volumes for models with gas plumes presenting infinite lateral extent and a constant thickness using the same acquisition geometry and the synthetic seismograms from the previous step 3) Construction of synthetic seismic volumes that represents the presence of gas plume with actual finite lateral extent and thickness predicted by the flow simulator, through combination of the synthetic volumes of the previous step. The synthetic seismic vintages were contaminated with coherent and random noise with different root mean square (RMS) amplitudes in order to determine the limits of contaminations at which it was possible to perform a quantitative interpretation in the time lapse seismic data. The prior model was generated with information gathered from two wells present in the area. The outputs comprises a Maximum a Posteriori (MAP) model and a marginal posterior probability density functions for gas saturation. After interpretation of the results we concluded that, due to its statistic nature, random noise can be differentiated from the signal if it has RMS amplitude up to the value measured in the vintage seismic data. In terms of coherent noise, in order to carry out a quantitative interpretation, the contamination RMS level needed to be at the same order of the difference volume RMS at the reservoir level.
AAPG Datapages/Search and Discovery Article #90217 © 2015 International Conference & Exhibition, Melbourne, Australia, September 13-16, 2015