Xin Fu¹, Sergio Romahn² and Kristopher A. Innanen¹

¹UNIVERSITY OF CALGARY, DEPARTMENT OF GEOSCIENCE, CALGARY, CANADA
²PETRÓLEOS MEXICANOS (PEMEX)
* CORRESPONDING AUTHOR EMAIL ADDRESS: XIN.FU1@UCALGARY.CA

ABSTRACT
The time-lapse seismic method is widely used to monitor reservoir changes, and in this effort, full-waveform inversion is now playing an increasingly active role. Especially in time-lapse full-waveform inversion, reducing non-repeatability between baseline and monitoring surveys is essential. In this paper, we compare the impact of different degrees and types of wavelet non-repeatability on three basic time-lapse inversion strategies, including the parallel strategy (PRS), the sequential strategy (SQS), and the double-difference strategy (DDS). Numerical examples indicate that DDS is much more sensitive to wavelet non-repeatability than other methods. Differences between the dominant frequency (or frequency band) of the baseline and monitoring wavelets strongly compromise PRS. The SQS is significantly less sensitive to this non-repeatability than the other methods. We address these findings by proposing a double-wavelet method to suppress the influence of wavelet disagreement on the resolution and interpretability of the time-lapse models. The approach is based on the convolutional relationship between the shot gather and the Green’s function. Numerical testing indicates the approach can effectively eliminate the wavelet non-repeatability and is stable to the receiver/source positions non-repeatability in time-lapse inversions when accurate prior estimates of the wavelets in the two surveys are provided.