No abstract

A shear‐wave (S‐wave) VSP experiment was performed at Lost Hills Field, California, in an attempt to detect hydraulic fractures induced in a nearby well. The hydrofrac well was located between an impulsive, S‐wave source on the surface and a receiver well containing a clamped, three‐component geophone. Both direct and scattered waves were detected immediately after shut‐in, when the hydraulic pumps were shut off and recording started. The scattered energy disappeared within about an hour, which is consistent with other measurements that indicate some degree of fracture closure and leak‐off within that period. Although S‐wave splitting was evident, no change was detected in the fast wave (polarized parallel to the fracture). However, the slow wave (polarized perpendicular to the fracture) did change over a period of about an hour, after which the prehydrofrac wavelet shape was recovered. The fact that only the wave polarized perpendicular to the fracture was affected is a dramatic confirmation of both theoretical predictions and laboratory observations of S‐wave behavior in a fractured medium. Subtracting the prehydrofrac wavelet from the wavelets recorded within the first hour after shut‐in revealed scattered wavelets that were diminished and phase‐rotated versions of the incident (prehydrofrac) wavelet. Arrival times of the direct and scattered waves were matched by ray tracing. We accounted for the scattered‐wave amplitudes by using numerical solutions of S‐wave diffractions off of ribbon‐shaped fractures. Amplitudes derived from full‐wavefield Born scattering, however, did not match recorded amplitudes. The phase of the scattered wavelets was matched very well by Born scattering when the incident wavelet was input, but only for fracture lengths no larger than half those predicted from fracture‐simulator models. These results show that a carefully controlled experiment, combined with accurate modeling, can provide important information about the geometry of induced fractures.

No abstract

Compaction in the reservoir overburden can impact production facilities and lead to a significant risk of well-bore failures. Prevalent practices of time-lapse seismic processing of 4D data above compacting reservoirs rely on picking time displacements and converting them into estimated velocity changes and subsurface deformation. This approach relies on prior data equalization and requires a significant amount of manual interpretation and quality control. We have developed methods for automatic detection of production-induced subsurface velocity changes from seismic data. We have evaluated a time-lapse inversion technique based on a simultaneous regularized full-waveform inversion (FWI) of multiple surveys. In our approach, baseline and monitor surveys are inverted simultaneously with a model-difference regularization penalizing nonphysical differences in the inverted models that are due to survey or computational repeatability issues. The primary focus of our work was the inversion of long-wavelength “blocky” changes in the subsurface model, and this was achieved using a phase-only FWI with a total-variation model-difference regularization. However, we have developed a multiscale extension of our method for recovering long- and short-wavelength production effects. We have developed a theoretical foundation of our method and analyzed its sensitivity to a realistic 1%–2% velocity deformation. The method was applied in a study of overburden dilation above the Gulf of Mexico Genesis field and recovered blocky negative-velocity anomalies above compacting reservoirs.

We compared P‐ and S‐wave velocities and quality factors (Q’S) from vertical seismic profiling (VSP) and sonic log measurements in five wells, three from the southwest San Joaquin Basin of California, one from near Laredo, Texas, and one from northern Alberta. Our purpose was to investigate the bias between sonic log and VSP velocities and to examine to what degree this bias might be a consequence of dispersion. VSPs and sonic logs were recorded in the same well in every case. Subsurface formations were predominantly clastic. The bias found was that VSP transit times were greater than sonic log times, consistent with normal dispersion. For the San Joaquin wells, differences in S‐wave transit times averaged 1–2 percent, while differences in P‐wave transit times averaged 6–7 percent. For the Alberta well, the situation was reversed, with differences in S‐wave transit times being about 6 percent, while those for P‐waves were 2.5 percent. For the Texas well, the differences averaged about 4 percent for both P‐ and S‐waves. Drift‐curve slopes for S‐waves tended to be low where the P‐wave slopes were high and vice versa. S‐wave drift‐curve slopes in the shallow California wells were 5–10 μs/ft (16–33 μs/m) and the P‐wave slopes were 15–30 μs/ft (49–98 μs/m). The S‐wave slope in sandstones in the northern Alberta well was up to 50 μs/ft (164 μs/m), while the P‐wave slope was about 5 μs/ft (16 μs/m). In the northern Alberta well the slopes for both P‐ and S‐waves flattened in the carbonate. In the Texas well, both P‐ and S‐wave drifts were comparable. We calculated (Q’s) from a velocity dispersion formula and from spectral ratios. When the two Q’s agreed, we concluded that velocity dispersion resulted solely from absorption. These Q estimation methods were reliable only for Q values smaller than 20. We found that, even with data of generally outstanding quality, Q values determined by standard methods can have large uncertainties, and negative Q’s may be common.