Seismic Data Preconditioning for Improved Reservoir

Characterization (Inversion and Fracture Analysis)

Darren Schmidt *, Hampson-Russell, Calgary, AB, Canada
[email protected]
Alicia Veronesi, Hampson-Russell, Calgary, AB, Canada
[email protected]
Franck Delbecq, Hampson-Russell, Calgary, AB, Canada
[email protected]
Jeff Durand, formerly Hampson-Russell, currently Shell, Calgary, AB, Canada
[email protected]

Summary
Advanced seismic techniques such as prestack inversion or azimuthal analysis are now
routinely adopted for quantitative reservoir characterization of rock properties and fracture indicators,
however the quality of the input seismic data is critical to such advanced analyses. Frequency content,
random and coherent noise, amplitude-preserving processing and gather flatness must be addressed
at the preconditioning stage. Improved results and increased confidence follow from careful well-driven
modeling and QCs.

Introduction
The results of prestack inversion and fracture analysis are ultimately dependent on the quality of
the input seismic data. An optimal result is produced when the seismic data are rich in frequencies
(both high and low), relatively noise free, of consistent amplitude, and the signal events are aligned (to
the expected AVO modeled response). These properties should be spatially stable, which can be
evaluated through map QCs. Care must be taken to not overly process the data, as this may remove
signal that contains important information.
To evaluate the suitability of the seismic data, prestack modeling at well locations is done. This
may be full wave reflectivity modeling (in which primaries, converted waves and multiples are all
modeled) or convolutional Zoeppritz-type synthetics (in which only primaries are modeled). Comparing
the real seismic to the well synthetic response is the best quality control step. This is usually only
practical at the latter stages in processing once the seismic signal is relatively clean and coherent. For
this reason the processes described in this discussion are applied after prestack migration.

Method
It is recommended to start with a wide range of QCs to analyze the data quality. A series of
maps are produced such as average and dominant frequency content, RMS amplitude over a large
window, and AVO cross-correlation correlation. Maps are particularly important since they are
diagnostic of the spatial distribution of data properties.
Seismic reservoir characterization assumes the input seismic has an invariant wavelet,
frequency and amplitude consistency, and good azimuth distribution across the survey (for anisotropic
analysis). Experience has shown that when the seismic does not satisfy these assumptions,
preconditioning (with stringent QCs) is helpful.

GeoConvention 2013: Integration 1

 For example, a difference of approximately 20 Hz in the average frequency was observed
between two nearby well locations that could not be accounted for by obvious near-surface effects
(Figure 1). Preconditioning (spectral balancing, in this case) was done to correct this.

Figure 1: Maps of average frequency before (left) and after (right) preconditioning. Extracted wavelets and
amplitude spectra before preconditioning are shown at two well locations (left). The map QC after preconditioning
indicates the average frequency is both higher and more laterally consistent (note: the wavelet shown was used
for the inversion).

Well logs are used to evaluate the seismic data quality through well ties and phase analysis.
Full wave reflectivity and convolutional Zoeppritz synthetics are then compared to the real seismic
gather. Reservoir characterization techniques are based on the Zoeppritz equations (or their
approximations) as the forward model. If the processed gathers differ too much from the Zoeppritz
synthetics, data preconditioning is most likely needed. Full wave synthetics, which include multiples and
PS conversions, can help further evaluate what processes should be applied to the gathers. For
example, it is possible to assess the influence of the multiple energy by turning on or off multiple
generation in the full wave algorithm.
For azimuthal QCs, COCA (Common Offset Common Azimuth) and CACO (Common Azimuth
Common Offset) displays are important, and should be reviewed to assess the presence of anisotropy
(HTI) in the survey. Spatially, the quality of the azimuth distribution is evaluated using azimuth versus
offset crossplots. Synthetic models can also be constructed versus azimuth and QC’d as above (Roure
et al., 2011).
Once the evaluation is complete, a preconditioning workflow can be designed and implemented
with the above QCs again to verify improvement. With the current computing power, inversion and
fracture analysis techniques themselves can now be used as QC steps and be (re)run to evaluate the
conditioning processes. It is recommended at least one fast track inversion should be run on the
gathers without preconditioning to set a benchmark

GeoConvention 2013: Integration 2

 Conditioning Processes
The importance of data preconditioning can be seen in the difference between Figures 2 and 3,
which show density results from prestack inversion. Understanding the complex heterogeneity of this oil
sands play is essential for successful reservoir development. The resolution (both high and low) has
been enhanced through balancing the amplitude spectra. Image quality was also improved through
temporal alignment of signal events in the gathers through VTI corrections, and subsequent application
of trim statics. Coherent noise was reduced by using a high-resolution Radon transform to attenuate
linear and parabolic multiple energy, while random noise was reduced with a 3D FXY projection filter.

Figure 2: Prestack inverted density prior to data preconditioning. The inserted well is a blind control. The
survey is from the Kinosis oil sands area (data courtesy of Nexen Inc. / CNOOC).

GeoConvention 2013: Integration 3

 Figure 3: Density from prestack inversion of preconditioned seismic. All inversion parameters are the
same; the only changes are the input seismic data and the associated offset dependent wavelets.

Often overlooked is the importance of the information contained in the low frequencies. Model-
based inversion schemes use well logs to construct the low frequency model to account for the missing
low frequencies in the seismic. When the model has to fill in more frequencies that are missing from
the low end of the seismic bandwidth, the inversion result becomes driven more by the wells and less
by the seismic. To alleviate this, AVO-friendly broad band spectral balancing (Nagarajappa and
Downton, 2009) can be applied. The result will be an increase of high frequency signal for increased
resolution, and more low frequency information so that the inversion can be directed by the seismic as
opposed to single point location well logs. Also, the amplitude spectra of the seismic is now laterally
consistent (cf. Figure 1), hence the assumption of wavelet stability is more valid.
Preconditioning to attenuate random noise may be required. It is not essential to have a
perfectly noise free seismic data as most inversion schemes model the data, hence random noise (that
does not fit the model) is discarded in the residual. Coherent noise is generally attenuated well during
conventional processing. If still present, the use of prestack noise attenuation schemes such as FXY
deconvolution and Radon transform filtering have proven quite effective. Methods and parameters
need to be tested, as each seismic dataset is unique. There may be cases where coherent noise is not
readily visible on the seismic data, however, and after inversion it can become quite apparent,
especially with the rainbow color palettes often used for display.
For fracture characterization azimuthal information needs to be preserved. Processes such as
noise attenuation should be done in an AVAz (Amplitude Variation with Azimuth)-friendly manner.
Instead of using a two-dimensional filter on a CMP gather, it is recommended to apply a three-
dimensional filter to an individual COV (Common Offset Vector) tile gather. For example, 3D F-Kx-Ky
filtering can be applied to COV tiles for footprint attenuation. To test parameters, the process must be
run for all individual COV tiles (which often number in the hundreds for land 3D geometries) and then
sorted back into CMP gathers for QC and stack.
Small-scale alignment of coherent signal events on gathers is generally accomplished through
the use of (time-variant) trim statics. However, if VTI or HTI patterns are observed on the seismic data,

GeoConvention 2013: Integration 4

 then additional anisotropy analyses and corrections can be done. Processes such as orthorhombic
velocity analysis (Wang and Wilkinson, 2012), high-density bispectral velocity picking, and VVAz
(Velocity Variation with Azimuth) investigations provide additional information about the subsurface (for
example, estimation of eta field or fracture intensity (Delbecq et al., 2013)). After applying these
corrections, any remaining small misalignment can be addressed through a final pass of trim statics.
An additional benefit of reviewing the velocity model is the optimization of offset to angle
conversion, since AVO equations are a function of incidence angles. By combining well logs (where
present) and seismic velocities (to fill shallow and deep sections) a more accurate velocity model is
created. It is also important to note that the offset to angle conversion should be done from a floating
datum rather than a flat processing datum. Significant error has been observed when this is not done;
underestimating the maximum incidence angle by up to 15 degrees (Figure 4).

Figure 4: Angle gathers before (left) and after (right) preconditioning. The zero offset synthetics trace from is
shown in red. On the left, angles were computed from processing datum. On the right, angles were computed
correctly from surface; useable now out to 45 degrees (dark blue). Area is Kinosis oil sands (Nexen / CNOOC).

Variations in near-surface conditions often result in the loss of amplitude over certain zones.
Barring geologic changes, the amplitude of a reflection event should remain relatively consistent across
a 3D survey. To correct for these surface-related amplitude changes, the RMS amplitudes are
calculated over a large window on the stack, then applied to each gather to normalize the lateral
variations. Alternatively, more detailed methods could also be used, such as modeling the expected
response from well data (in offset or azimuth), then applying an angle-dependent scaling function to the
seismic gathers.

Conclusions
The quality of any seismic reservoir characterization volume is restricted by the quality of the
input data. Often, results can be greatly improved by proper preconditioning of the prestack seismic
data. The method described in this paper uses multiple QC products (maps, synthetics, anisotropic
analyses and the inversion techniques themselves) to evaluate the seismic prior to the reservoir
analysis. The QCs can be used to design a data preconditioning sequence which may include

GeoConvention 2013: Integration 5

 frequency and amplitude balancing, noise attenuation, anisotropic velocity analysis, and signal event
alignment. This entire process is non-unique and, if necessary, iterative.
Seismic data QCs and preconditioning provide insight and confidence in the input seismic data
and reservoir characterization final results. When properly evaluated and applied, the result is better
quantitative reservoir characterization and improved interpretation. This was demonstrated through a
density inversion.

Acknowledgements
The authors wish to express their appreciation to Nexen Inc. and CNOOC for permission to
show data images, and to thank Dragana Todorovic-Marinic and Dave Gray for their input. Thank you
to all clients who not only provided feedback, but also pushed to develop different techniques to solve
challenges as they were encountered. Special thanks to the members of the Hampson-Russell
Services team.

References
Delbecq, F., Downton, J., and Letizia, M., 2013, A math-free look at azimuthal surface seismic techniques, CSEG Recorder,
38, 20-31.
Gray, D., Schmidt, D., Nagarajappa, N., Ursenbach, C. and Downton, J., 2009, An azimuthal-AVO-compliant 3D land seismic
processing flow, Expanded Abstracts of the 2009 Joint CSPG/CSEG/CWLS Conference
Gray, D. and Wang, S., 2009, Towards an optimal workflow for Azimuthal AVO, Expanded Abstracts of the 2009 Joint
CSPG/CSEG/CWLS Conference
Li, X., 2008, An introduction to common offset vector trace gathering, CSEG Recorder, 34, 28-34.
Nagarajappa, N., and Downton, J., 2009, AVO compliant spectral balancing, Expanded Abstracts of the 2009 Joint
CSPG/CSEG/CWLS Conference
Roure, B., Collet, O., and Downton, J., 2011, The reflectivity response of multiple fractures and its implications for azimuthal
AVO inversion, Expanded Abstracts of the 2011 Joint CSPG/CSEG/CWLS Conference
Roy, B., Anno, P., and Gurch, M., 2008, Imaging oil-sand reservoir heterogeneities using wide-angle prestack seismic
inversion, The Leading Edge, 11, 1192-1201.
Schmidt, D., Gray, D., Li, X., Trad, D. and Downton, J., 2009, Improving the Image: 5D interpolation and COV gathering of a
megabin survey, Expanded Abstracts of the 2009 Joint CSPG/CSEG/CWLS Conference
Wang, S. and Wilkinson, D., 2012, Imaging for unconventional resource plays using an orthorhombic velocity model: SEG
Technical Program Expanded Abstracts 2012
Yilmaz, 0., 2001, Seismic Data Analysis: Processing, Inversion, and Interpretation of Seismic Data

GeoConvention 2013: Integration 6