In unconventional reservoirs with significant organic content, the Greenberg-Castagna (GC) S-wave velocity prediction method does not yield accurate S-wave velocity predictions, with observed mean errors varying from 6% to 16% in a variety of unconventional reservoirs rich in organic content. This is because kerogen content is not explicitly taken into account in the GC S-wave velocity prediction method. Two alternative approaches for bedding-normal S-wave velocity prediction from P-wave velocity and other well logs in relatively straight holes drilled in unconventional reservoirs are investigated. The first method is purely empirical, requiring minimal information such as P-wave velocity and total organic carbon content. This approach implicitly accounts for compositional and stress effects on mudrock elasticity. The second method can be classified as a hybrid technique, comprising the following three steps: (1) computing a nonkerogen phase P-wave velocity using the Vernik-Kachanov (VK) model, (2) determination of the nonkerogen S-wave velocity from the GC approach, and (3) using a simplified VK model to mix the nonkerogen matrix with nanoporous kerogen to predict the bedding-normal S-wave velocity of the organic mudrock. The second method explicitly takes into account all the variables that control elastic properties of organic mudrock reservoirs. Tests in nine wells from seven different oil and gas shale reservoirs indicate that both methods have prediction accuracy better than 3% error when input data are accurate.

In situ P- and S-wave velocity measurements in a variety of organic-rich shales exhibit P-to-S-wave velocity ratios that are significantly lower than lithologically similar fully brine-saturated shales having low organic content. It has been hypothesized that this drop could be explained by the direct influence of kerogen on the rock frame and/or by the presence of free hydrocarbons in the pore space. The correlation of hydrocarbon saturation with total organic content in situ makes it difficult to separate these possible mechanisms using log data alone. Theoretical bounding equations, using pure kerogen as an end-member component without associated gas, indicate that kerogen reduces the P- and S-wave velocities but does not in general reduce their ratio enough to explain the observed low velocity ratio. The theoretical modeling is consistent with ultrasonic measurements on organic shale core samples that indicate no dependence of velocity ratios on the kerogen volume alone. Sonic log measurements of P- and S-wave velocities in seven organic-rich shale formations deviate significantly (typically more than 5%) from the Greenberg-Castagna empirical brine-saturated shale trend toward lower velocity ratios. In these formations, and on core measurements, Gassmann fluid substitution to 100% brine saturation yields velocity ratios consistent with the Greenberg-Castagna velocity trend for fully brine-saturated shales, despite the high organic content. These sonic and ultrasonic measurements, as well as theoretical modeling, suggest that the velocity ratio reduction in organic shales is best explained by the presence of free hydrocarbons.

Utilizing laboratory measurements on dry and fully brine-saturated well-lithified shale reservoir rocks and comparing them to log data and theoretical modeling, we find no statistically significant intrinsic dispersion from seismic to sonic and laboratory measurement frequencies due to fluid effects. Under in situ stress conditions, the Gassmann zero-frequency P-wave velocity prediction for a Permian basin sample was within 0.3% of the measured velocity on the brine-saturated sample at an ultrasonic frequency. At deviatoric stresses ranging from 1.7 to 70 MPa on the same sample, the percent error in the Gassmann P-wave velocity prediction ranges from 0.3 to 2.2%. These results are consistent with the lack of significant dispersion in the reported velocities and theoretical predictions in the Cotton Valley shale. Based on the Biot–Gassmann model, the characteristic frequency in both formations occurs at ~10¹⁰ Hz. Applying a squirt flow model also predicts transition to the high-frequency regime occurring at ~10⁹ Hz for both formations. The comparison of the sonic log measurements for four different shale/tight rock formations to ultrasonic measurements taken from core samples under in situ stress conditions confirms these findings. To our knowledge, this is the first extensive comparison of the sonic log to ultrasonic core data measurements. For these rocks, there is no clearly observable or predicted significant dispersion due to fluid effects at in situ stress conditions within the frequency range for which measurements are made.

Calculating the change in the saturated bulk modulus of a saturated rock as fluid properties change requires a priori selection of an effective bulk modulus of the solid constituents. When mineral moduli are similar, theoretical bounds on the solid modulus are tight. However, when solid properties vary greatly, as in organic-rich shales, bounds on the effective solid modulus result in uncertainty in the predicted change in the saturated bulk modulus of the rock. We use a semi-empirical rock physics model utilizing the Brown-Korringa equation for polymineral assemblages and introduce three parameters to estimate the pore space compressibility, the dry frame compressibility, and the fractional position of the effective solid modulus relative to the Reuss and Voigt bounds. We optimize for these three parameters in seven organic shale formations and find that the Reuss bound for the effective solid material modulus best fits the data when organic content is high. Furthermore, we use this model to fluid substitute to 100% brine saturation and find Gassmann’s equation using the Hill average predicts similar saturated moduli to the semi-empirical Brown-Korringa rock physics model when volume fraction of solid organic matter is less than 5%. However, at higher organic contents, we find that the error using the Gassmann-Hill approach increases, and the semi-empirical Brown-Korringa model better fits the data.