Stochastic Time-lapse Inversion of a CO2 Sequestration Synthetic Seismic Data

Abstract

The objective of this work is to assess the effect of noise and parameterisation on the performance of the stochastic time lapse inversion. To do so, a noise-free synthetic dataset created for a feasibility study of an actual CO2 sequestration project (CO2CRC Otway Project) was inverted and used as a baseline. Noise (random and coherent) was added to the seismic data, input parameters changed and the results were compared with the baseline case. The findings for wrong parameterisation cases were very encouraging and consistent with the theory. When random noise was added to the input seismic data the algorithm was able to recover the true model within an acceptable margin of error. However, addition of coherent noise affected the inversion result significantly. Only when the root-mean-square (RMS) amplitude level was comparable to the one in the difference volume the algorithm was able to actually differentiate the noise from the signal. These findings support the idea of a careful processing to avoid coherent noise and a judicious interpretation when it is unavoidable. Finally a new indicator was developed to calculate the improvement in detectability after the input of new data using the stochastic time lapse inversion.