A Comparison of Inversion techniques for Estimating VP/VS from 3C3D Seismic Data

Paul Fyfe Anderson

Exploration and development geophysicists regularly make use of prestack analysis is seismic data in order to identify hydrocarbon accumulations in the subsurface in quantities that are economic and in reservoirs that can produce the hydrocarbons efficiently. One commonly used attributed derived from prestack analysis is the ratio of P-wave and S-wave velocities, or Vp/Vs, however there are a variety of methods available to derive these parameters. This thesis compares three established workflows for estimating Vp/Vs from 3C3D seismic data, and discusses some of the limitations of each method on both a synthetic model and an acquired 3C3D seismic survey from eastern Alberta, Canada.

The first method is a two step inversion where we first invert for P- and S-wave reflectivity from prestack PP gathers, followed by a second, decoupled inversion of these reflectivities to impedances. Another method condenses this into a single inversion, allowing estimation of P- and S-Impedance directly from prestack PP gathers. The third inversion is similar, in that it is a single-step inversion, however it also incorporates converted wave, or PS, data in an attempt to better constrain the solution by providing additional independent input.

Analysis of the model data has shown that the prestack inversion with converted wave data provides the most accurate estimates of Vp/Vs of the three methods compared, though the prestack inversion without converted wave data also appears to have far outperformed the two-step inversion in the zone of interest. However the model example also shows a high degree of sensitivity to the registration of the converted wave data to PP-time and that even small errors in this registration can degrade the quality of the inverted result.

In the case of the real seismic data from eastern Alberta, the prestack inversion without converted wave data has produced the most accurate estimate of Vp/Vs ratio in the reservoir interval. Interestingly it has been revealed that the two-step inversion process has actually created noise in the Vp/Vs volume resulting from the de-coupled second inversion step. Correlations to well logs at well locations indicate that the prestack inversion with converted wave data has outperformed the two-step inversion, despite the registration errors that are inherent in the model resulting from an isotropic assumption made in the processing of the converted wave data, which has significantly affected not only the registration of the converted wave data to P-wave time, but has also degraded the bandwidth and amplitude fidelity of the converted wave volumes.