True vector processing, which exploits the particle-motion information inherent in a multi- component seismic recording, has the potential to yield more accurate P- and S-wave images than conventional scalar processing when P/S wavefield cross-contamination occurs. The simplest example of using particle motion to distinguish between wave types is polarisation filtering, a single-trace vector method borrowed from earthquake seismology. This does not, however, generally lead to effective wavefield separation for surface reflection data due to significant interference between the different waves in a typical exploration-seismic recording. More effective vector processing is possible if particle-motion information is coupled with the familiar tools of frequency and velocity filtering. Four closely related multi-trace, multi-component processing algorithms, namely Multiple Signal Classification (MUSIC), Spectral Matrix Filtering (SMF), Integrated Wavefield Separation Analysis (IWSA) and Parametric Inverse Modelling (PIM), have been examined and evaluated. The theoretical inter-relationships between these various vector processing schemes have been clearly demonstrated. A number of technical modifications have been introduced to optimise the performance of the parametric vector methods for surface seismic applications. Comprehensive practical trials, using both synthetic and real surface reflection data, have revealed new information about the relative performances and suitability of the parametric vector separation algorithms.
The frequency-space domain parametric equations form the common underlying mathematical basis of the MUSIC, SMF, IWSA and PIM vector schemes. The seismic recording is assumed to comprise a finite number of plane waves, modelled in terms of their Fourier components and two frequency-independent parameters, namely slowness and polarisation. MUSIC, IWSA and PIM differ only in their method of estimating the wavefield slowness and polarisation parameters. The parameter-estimation process employed by MUSIC is based on the spectral estimator function produced via eigeanalysis of the spectral matrix. IWSA recovers wavefield slowness and particle-motion information through eigenanalysis of a transfer matrix that relates the Fourier spectra of data at one receiver to those at an adjacent receiver. PIM solves for the desired wavefield parameters using a nonlinear inversion scheme to minimise the error between the observed seismic data and the modelled data. Once parameters for each of the waves are determined, wave separation is accomplished using the least-squares solution to the parametric equations. SMF differs from the other three parametric schemes in that it does not recover actual wavefield parameters. Instead, separation via SMF is achieved through projection of the multi-component seismic record into the signal vector space.
Previous practical demonstrations of these parametric vector methods have been largely confined to non-seismic, borehole, and single-component surface seismic applications. A wide range of synthetic and real data examples have been used here to evaluate the parametric vector schemes for multi-component surface seismic applications. Iterative separation is shown to be critical to the success of parametric vector processing of real seismic reflection data. The performance of the separation schemes is closely tied to the accuracy of the local plane-wave assumptions of the techniques. Practical controls that influence the success of separation include selection of the data-analysis window, design of the iterative separation process, choice of the frequency bandwidth, design of the spectral and transfer matrix smoothing operators, and provision of initial wavefield parameter estimates. Despite requiring considerably less user-input than the other parametric schemes, both IWSA and SMF are found to be unsuitable for real datasets that contain numerous waves and/or significant random noise contamination. Synthetic trials demonstrate that MUSIC is feasible for real-data applications. However, its comparatively long run-time means MUSIC is generally not the favoured separation approach. The PIM algorithm is the preferred separation scheme for exploration-seismic applications. This vector method has been successfully applied to a multi-component dynamite shot record and an OBC common-receiver gather. In situations where the non-linear inversion process of PIM is sidetracked by high-amplitude noise, or the desired wavefields exhibit similar vector properties, improved separation results can sometimes be achieved using MUSIC.
Practical demonstrations suggest that MUSIC and PIM are particularly suited for improved shallow P/S imaging. Vector processing aimed at recovering reflection energy from ground-roll contaminated data has demonstrated the potential of integrated scalar and vector processing to improve recovery of signal through high-amplitude surface waves. Where significant P/S cross-contamination does not occur, production vector processing may not be cost effective. However, even in such situations, vector analysis provides a powerful tool for validating the assumptions required to produce P- and S-wave sections via scalar processing of multi-component data. Due to the high degree of interactivity required to select suitable processing parameters, implementation of the parametric vector methods in a highly-automated production environment is not yet viable. The window-based methodology employed by the parametric vector processing schemes suggests a particular affinity for specialised target-horizon P/S imaging subsequent to reconnaissance exploration. Future research, focused on demonstrating practical situations for which vector processing significantly enhances P/S imaging, will stimulate further methodological refinements, leading to more widespread application.