Seismic data conditioning is central to inversion for rock properties, although it does not replace high-quality data acquisition and processing work. The objective for conditioning seismic before inversion is to ensure amplitude preservation while eliminating the post-processing residual effects such as anisotropy and noise, which should make the seismic data more AVO and inversion friendly.

The conditioning should begin with an analysis of the common midpoint (CMP) gathers for determining how the data can be improved.  The potential issues may be residual multiples, linear noise, random noise, gather flatness, and wavelet stretch.  Only after determining which issues are present in the data can an adequate workflow be designed.  An essential factor to consider while designing the workflow to be applied to the data is to not over-condition the data.  Additional processing does not necessarily mean a better product will be produced. The aim is to be conservative and, most importantly, preserve the AVO response while flattening the events as much as possible as well as adequately attenuating noise without generating artifacts.

Figure 1 shows various types of conditioning processes which can be applied to improve the quality of seismic data, subdivided into three segments: attenuation of residual noise, gather flattening and signal processing.  A preliminary step may be an analysis of signal-to-noise (S/N) ratio for identifying whether the data has noise issues which must be addressed.  If noise is in fact an issue, determining whether the noise is linear, random, or comes from multiples may be achieved by analyzing the gathers.  For example, if the data exhibit residual multiples, a weighted least squares radon transform is effective in modeling the coherent residual multiples. The application of such processes to model seismic events by moveout discrimination will aid in reducing interference from multiples with primary energy. Caution is encouraged in this process given the risk of removing usable signal, especially at near offsets where the moveout differentiation between primaries and multiples is difficult to achieve. The use of forward and reverse radon transform to convert data from time-space domain to Tau-p domain and vice versa, lies on the combination of seismic velocity parameters (low and high scalar versions of the best seismic velocities available) in order to accurately mute the zones with multiples in the Tau-p domain. For this reason, a thorough examination of the seismic velocities is required.  For remnant linear noise not of hyperbolic nature, several filter routines based on velocity or dip, frequency or wavenumber are available and may be utilized to remove such undesired events.

For interbed multiples, regular gap deconvolution with zero phase correction is an option which can be applied with a frequency bandwidth constraint.  For random noise issues, it is assumed that the stacking process will reduce or suppress non-coherent noise, although in specific environments such as carbonates, there is low fold and low transmission which has the effect of significantly lowering the S/N ratio. In this situation, a random noise attenuation technique may prove worthy for increasing the continuity of events.

Accurately predicting reservoir properties using simultaneous prestack inversion can be highly dependent on the degree of ‘flatness’ seismic gather events.  The primary cause for the need to apply gather flattening techniques to seismic data is residual NMO due to anisotropy, where its effects typically appear as ‘hockey sticks’ in the farther offsets.  If the data are stacked without correcting for residual NMO, the output will be both spatially and temporally mis-positioned.

Residual time shifts caused by unaccounted time shifts among the traces may also pose a threat to attaining success in AVO inversion studies. Trim statics is a commonly-used technique which uses maximum cross-correlation values with a model trace from different computation windows.  This process can dynamically shift data samples in time and preserve wavelet character across the entire offset range.  Residual Event Alignment or REVEAL (Schlumberger proprietary technique) is a more effective method which can be applied to angle stacks or CMP gathers.

For correction of CMP gathers, the process consists of first sorting the NMO corrected gathers into offset planes.  For each plane, a ‘raw’ displacement field is computed using the previous offset as the reference. Next, a final displacement field for each offset is calculated and applied by summing all the previous ‘raw’ displacements.  After all the time shifts have been taken into account, the final displacement field may be filtered in order to smooth existing abnormal values. Finally, the data is sorted back into the CMP gather domain. An example of a CMP gather before and after REVEAL is illustrated in Figure 2.  A robust method such as REVEAL for the correction of residual time shifts is especially important in PP PS and time-lapse inversion, where multiple independently-acquired datasets (i.e. baseline and monitor surveys) may need to be aligned.

The final segment of our seismic conditioning discussion focuses on signal processing techniques.  Many algorithms exist in industry today for enhancing and/or manipulating seismic data such as deconvolution, wavelet processing, and Q compensation; what may be used highly depends on the objective. For example, increasing frequency content may be of interest if the objective includes the interpretation of thin beds.

Wavelet equalization is a process where offset dependent wavelets are estimated globally within a seismic survey and can be matched to a reference wavelet at a certain offset. The resulting shape filters can then be applied to achieve a consistent wavelet across the offset in a gather.   In areas where significant amplitude changes are associated with sea-bottom depth changes (i.e. marine continental slope), an amplitude balancing technique may be applied to the seismic data.  In this process, seismic energy in a specified window around a reference horizon is computed for all traces, and then horizontally filtered for removing the effects of noise.  Each trace in the seismic data set is eventually rescaled according to the relation between the horizontally filtered energy level of the trace and the energy level of the reference trace.  For amplitude variations in a seismic survey which are not due to geology, a scaling technique may be used where statistically-derived amplitude scaling factors across an offset over a volume of data can be estimated. The process involves referencing to a common RMS amplitude level over a large window to derive scaling factors.

Seismic-driven characterization studies will usually benefit from seismic data conditioning work.  There are instances where the job performed during the data processing sequence is sufficient and no post-processing conditioning is required.  This depends highly on the quality of the processing algorithms available and the experience level of the seismic processor.