See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/277392423

Seismic Inversion Methods and some of their constraints

Article  in  First Break · June 2004

DOI: 10.3997/1365-2397.2004011

CITATIONS READS

115 7,399

2 authors, including:

Paul C. H. Veeken
Wintershall
67 PUBLICATIONS   581 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Betic Cordilleras, Southern Spain View project

The development of metods for seismic interpretation and geological modelling View project

All content following this page was uploaded by Paul C. H. Veeken on 16 October 2017.

The user has requested enhancement of the downloaded file.

first break volume 22, June 2004 technical article

Seismic inversion methods and some of their

constraints
P. C. H. Veeken and M. Da Silva*

Introduction negative effects of the seismic convolution. It is in general suc-

The interest in seismic inversion techniques has been growing cessful in reducing multiple energy (Yilmaz 2001). During
steadily over the last couple of years. Integrated studies are acquisition of reflection datasets, the seismic signal is sent
essential to hydrocarbon development projects (e.g. Vazquez et through the earth (i.e. convolved with the earth filter) and
al. 1997, Cosentino 2001) and inversion is one of the means reflected back into the geophone (convolved with the record-
to extract additional information from seismic data. Various ing set-up filter). Deconvolution is meant to undo the defor-
seismic inversion techniques are briefly presented. Inversion mation caused by these two filters to the seismic response. A
replaces the seismic signature by a blocky response, correspon- spiked response is never achieved because of the band limited
ding to acoustic and/or elastic impedance layering. It facilitates nature of the seismic data (Veeken, in prep.). The decon filter
the interpretation of meaningful geological and petrophysical would need to be infinitely long, which is clearly not feasible.
boundaries in the subsurface. Inversion increases the resolu- A perfect job is therefore hardly ever possible and that is where
tion of conventional seismics in many cases and puts the study seismic inversion or stratigraphic deconvolution comes in.
of reservoir parameters at a different level. It results in opti- Ideally, an acoustic impedance interface should be repre-
mised volumetrics, improved ranking of leads/prospects, better sented by a single spike on the seismic trace. The contrast is
delineation of drainage areas and identification of ‘sweet described by the reflectivity formula:
spots’ in field development studies (e.g. Veeken et al. 2002).
R = ( U2V2 - U1V1) / (U2V2 + U1V1) (1)
The main steps in an inversion study are:
■ Quality control and pre-conditioning of the input data. Acoustic impedance interfaces in the real world are sometimes
■ Well–to-seismic match, zero-phasing of data in zone of positioned very closely together. As a result, the seismic
interest and extraction of the wavelet. response of the first interface overlaps in time with the energy
■ Running of the inversion algorithm with generation of triggered by the next interface. The result is interference of the
acoustic or elastic impedance cubes and extraction of individual seismic events on the recorded seismogram. A com-
attributes. posite seismic loop is registered in the geophone at the surface
■ Visualisation and interpretation of the results in terms of (Figure 1A). Under these circumstances it becomes difficult to
reservoir development. discriminate between the effects stemming from each interface.
Therefore, the apparent amplitude and frequency of reflections
The inversion methods are either deterministic or probabilistic should always be treated with care during the interpretation
and the approach can be post- and/or pre-stack. Inversion phase (Veeken, in prep.).
schemes generally use migrated time data as basic input. The It is necessary to undertake ‘stratigraphic deconvolution’
pre-stack method exploits AVO effects on migrated CDP gath- to undo some of the negative interference effects (cf Duboz et
ers. There is a trade-off between method/cost/time and quality al. 1998). This technique is also known as seismic inversion.
of inversion results. Feasibility studies with synthetic model- The seismic signature is replaced by a blocky response corre-
ling are recommended before embarking on an inversion or sponding to the seismic impedance layering. This type of pro-
AVO project (Da Silva et al., in prep.). cessing facilitates the interpretation of meaningful geological
The past track record has demonstrated the benefits of the and petrophysical boundaries on the reflection data (Veeken et
seismic inversion method. However, it should be realised that al. 2002). The input for seismic inversion is traditionally com-
the inversion procedure is not a unique process, i.e. there is no posed of time migrated seismic data (pre- or post-stack), a
one single solution to the given problem. Care should be taken wavelet and an optional initial earth model (velocities and
when interpreting the inversion results. Adequate data pre- densities; Figure 1B). Inversion allows the study of reservoir
conditioning is a prerequisite for quantitative interpretation of characteristics in greater detail, under favourable circum-
the end results. stances the resolution of the dataset is even increased.
Detection of thin beds is based on subtle changes in the shape
Deconvolution and seismic inversion. of a specific seismic loop (doublet), a feature that is usually
Seismic inversion is undertaken to complement conventional beyond the standard seismic resolution.
deconvolution processing. Deconvolution tries to undo some It is recommendable that prestack time migration data

* P.C.H. Veeken, 29 rue des Benedictins, 57050 Le Ban St. Martin, France
M. da Silva, 1 rue Léon Migaux, 91341 Massy CEDEX, France

© 2004 EAGE 15
 technical article first break volume 22, June 2004

Figure 1A Thin bed configuration gives

rise to a composite seismic response due
to time overlap and interference of the
reflected seismic pulses (modified after
Todd and Sangree, 1977).

Figure 1B The input for a model-driven seismic inversion consists of time migrated seismic data, a seismic wavelet, an option-
al impedance model and picked time horizons. Proper data conditioning is a pre-requisite to obtain reliable results. Seismic
inversion is never a unique process, i.e. there is not a single solution to the given problem, but several models may equally
well explain the recorded seismic response.

16 © 2004 EAGE
first break volume 22, June 2004 technical article

serves as input to the inversion scheme. The main advantage is Dutta 2002; Figure 2).
a substantially better velocity field and improved positioning Inversion is often done in conjunction with AVO analysis.
of the reflections. The corresponding interval velocity model is The combination of these geophysical study techniques aug-
in general more detailed than that obtained via the conven- ments the confidence in correct ranking of leads / prospects
tional seismic method exploiting Dix's formula (Dix 1955). and definition of ‘sweet spots’ in the HC accumulations
Moreover, this velocity field is better suited for evaluation of (Veeken et al. 2002). Such an approach reduces uncertainties
in-situ geo-pressures and fault compartment predictions (e.g. and drilling risks, which is an important aspect for optimising

Figure 2 The pre-stack seismic inversion results in a better velocity model. The interval velocities calculated via Dix's formu-
la are very smooth and therefore less reliable. It ensures the best stack, but does not yield necessarily the best interval veloc-
ity. The inversion involves a pre-stack time migration and this velocity model preserves better the original variability of the
velocity field (modified after Dutta 2002).

© 2004 EAGE 17
 technical article first break volume 22, June 2004

an exploration and hydrocarbon development strategy (Da impedance of a rock sequence is defined as the product of den-
Silva et al., in prep). A pre-stack inversion scheme incorporates sity and velocity (cf Sheriff 2002). The link between the seis-
AVO effects seen on CDP gathers. The partial stacks (PS) show mic and acoustic impedance (AI) cube is the seismic wavelet.
in many cases a characteristic difference in behaviour for the The wavelet is derived either directly from the seismic data or
amplitude of the top hydrocarbon reservoir reflection (AVO computed with the aid of available well data. The density and
effect). The P- and S-wave energy are crucial parameters to sonic logs in the well permit calculation of the AI response.
model the reflectivity of synthetic offset gathers. Calibrated sonic with checkshots and/or VSP data are needed
Compressional P-waves contain information on the lithology for the depth-time conversion of the vertical log scale. The real
and the porefill, while transverse S-waves are hardly influ- seismic trace at the well location is subsequently matched with
enced by the fluid contents. Absence of P-wave related DHIs the reflectivity trace computed from the well logs. This com-
on the corresponding S-sections is, for instance, a useful crite- parison yields the seismic wavelet.
rion to discriminate between a hydrocarbon and brine filled
reservoir (Stewart et al. 2003). Rock physical parameters like A typical inversion project generally comprises the following
the Poisson's ratio, rhomu, lambdarho are estimated from the steps:
P- and S-wave velocity variations. The prestack inversion gives ■ Quality control and pre-conditioning of the input data.
a better handle on the lithology, porosity, permeability and/or ■ Well–to-seismic match and compilation of a synthetic trace.
water saturation in the porefill of the rocks under investiga- ■ Zero-phasing of data in zone of interest and extraction of
tion. the wavelet. This step can be circumvented when the non
Stratigraphic deconvolution tries to put a simple spiked zero-phase effects are integrated in the wavelet used for the
reflectivity response at geological boundaries (lithological inversion.
changes) and the main reservoir interfaces (for instance, a fluid ■ Running of the inversion algorithm.
contact). This is often done by inversion of the seismic cube ■ Visualisation and interpretation of the results in terms of
into an acoustic impedance cube (Figure 1B). The acoustic reservoir development.

Figure 3A Mute function applied on a CDP gather. Data from a specific offset and time range is used for each time sample
in the processing. In this way the amount of noise on the traces is drastically reduced. Selection of the proper mute function
is critical in optimising the stacking results.
Figure 3B Comparison of a raw stacked seismic line and the same line whereby the mute function is applied on the CDP
gathers before the stack.

18 © 2004 EAGE
first break volume 22, June 2004 technical article

Figure 4A Automatic Gain Control function (AGC) is computed in a sliding window. The average amplitude within the win-
dow (A1) is determined and scaled to a specific value. This scaling factor (F1) is applied on the sample at the centre of the
window. Subsequent tapering of the multiplication factors in the overlap areas of consecutive time window positions ensures
a smooth curve for the gain factors on the whole trace for each time sample. The original amplitudes are thus altered in a
time variant manner.
Figure 4B 'Preserved amplitude' stack section compared with a normalised stack section. The preserved amplitude process-
ing gives a better representation of the gross lithology of individual seismic units and facilitates their correct interpretation
(data courtesy Pemex).

Seismic inversion is a somewhat confusing expression. The acquisition footprint is often difficult to remove in
Inversion in itself means to undo an operation, but here it 3D surveys, but it is an essential step for reconstituting the
represents the transformation of a seismic amplitude cube true amplitude character of a dataset. The aim is to preserve
into an acoustic (or elastic) impedance cube. There are sever- the real amplitude behaviour proportional to the acoustic
al ways to achieve this objective, as will be shown later on. impedance contrasts in the subsurface and eliminate any
Time lapse inversion of 4D seismics and reservoir behav- known amplitude distortions. These amplitude corrections
iour monitoring facilitate the extraction of saturation and should be implemented when it enhances the overall quality
pressure effects induced by hydrocarbon production (e.g. of the seismic dataset. Whitening of the amplitude spectra is
Gluck 2000, Oldenziel 2003). in this respect sometimes a dangerous operation that needs
careful evaluation before it is done. Whitening means that
Input data conditioning the amplitudes for all frequencies are adjusted to the same
It is important that the seismic input data are screened for output level. This artificial amplitude boosting is in many
their quality. In a pre-stack approach it means going back to cases rather uncontrolled and accompanied by a significant
the CDP gathers and making sure that the panels are clean. and unacceptable increase in noise level.
It might prove necessary to apply a more efficient mute func- A gain function is certainly useful, unless it destroys the
tion for this purpose (Figure 3). Simple bandpass filtering can preserved amplitude character of the dataset (Figure 4).
be very effective for improving the seismic data. Applying a Preserved amplitude processing implies that only a light gain
spiking random noise attenuator is a more sophisticated way function is utilised, applied in a fairly large time window.
to reduce the noise on the migrated stack. A 3D filtering The sliding window should preferably not be smaller than
approach will give even better results (Da Silva et al., in 1000 ms. Each time the sample on the seismic trace has a cer-
prep). tain multiplication factor assigned. Amplitude balancing is a

© 2004 EAGE 19
 technical article first break volume 22, June 2004

Figure 5 F-K filtering can be safely applied when it is sure that only the noise is affected by the operation. It will change the
amplitude value and take out the noise distortion. It restores the amplitudes to a more reliable value that leaves scope for
quantitative interpretation.

delicate process that should be done with care, especially tivity sections. The suppression test results may look differ-
when later quantitative interpretation is the ultimate goal ent in the pre-stack domain. Hence, combined pre- and post-
(Veeken, in prep.). stack testing is recommendable for multiple suppression.
Even velocity filtering is worthwhile considering on the Artefacts in the input data clearly will lead to unreliable
condition that the F-K operation cuts out only noise (Figure inversion results. Multiple suppression, amplitude balancing
5). Under such circumstances the F-K filter will indeed alter and true amplitude processing are closely related subjects
the amplitudes, but in a good sense. The latter has to be that highly influence the inversion end-results. A preserved
demonstrated by proper testing of the processing parameters. amplitude section often looks different from a section that
Multiple suppression and deconvolution are other issues that pleases the interpreter’s eye. The ideal input to seismic inver-
need to be addressed in order to clean up the data. Radon sion are of course amplitudes that are directly proportional
transform or Tau-P processing is an alternative option, a last to the subsurface reflection coefficients.
resort when other methods fail to enhance the dataset. The Proper data conditioning is essential for later quantitative
Tau stands for intercept time at the zero offset position. P is interpretation of the inversion end results, i.e. when reservoir
the slowness parameter of the ray in Snell's Law. Seismic characterisation and lateral prediction studies are required
noise is sometimes more easily removed in the Tau-P domain. (Da Silva et al., in prep). For that matter reprocessing of the
A drawback is that the transform back to the TX domain is seismic dataset may prove necessary; even complete re-
rather sensitive to errors, because the operator is not orthog- shoots are sometimes justified (e.g. Onderwater et al. 1996).
onal (Trad et al. 2003). Furthermore, the Radon Transform The original data was processed with a special target in mind
suffers from loss of resolution and aliasing, arising from and all parameters were tuned to this objective. The later use
incompleteness of information due to limited aperture and of the same seismic dataset in pre-stack analysis was unfor-
discretization of the data or time sampling (Querne, pers. tunately not always foreseen. Pre-stack time migration
com.). results in better positioned seismic energy and also the veloc-
In general, multiples are bad news because inversion will ity picking is more accurate. The current demands, made by
treat them as primaries. Aggressive de-multiple operators, the reservoir engineer on the quality of reliable output at all
however, may cause unwanted damage to the primaries. stages of the seismic processing, have increased the tasks of
Decisions on the correct trade-off are usually made on reflec- the geophysicist over the last decade. Detailed knowledge of

20 © 2004 EAGE
first break volume 22, June 2004 technical article

Figure 6 VSP and synthetic seismogram display for comparison of the seismic resolution. The upgoing waves are separated
from the downgoing waves by F-K filtering. Alternatively the first break time can be added to each trace to line-up the up-
coming energy. If it is subtracted, then the downgoing energy is horizontally lined up. The VSP trace has a frequency con-
tent that is better comparable to the surface seismic dataset. The energy around the first break time is relatively clear from
multiple energy and is used in the compilation of a corridor stack (modified after Hardage 1985 and Sheriff 2002).

© 2004 EAGE 21
 technical article first break volume 22, June 2004

Figure 7A Amplitude and phase spectra computed for a seismic survey via the Fourier transform. The difference in ampli-
tude spectra is shown before and after applying a phase correction to the seismic dataset. The bulk phase correction ensures
a better match in the zone of interest. The zero-phase aspect is maintained up to 60 Hz. Two wavelets have been computed;
the last one has a much better zero-phase character.
Figure 7B Bulk phase correction does leave the geometry of seismic reflections untouched, but changes their amplitudes in a
systematic way. A 180° phase shift leads to an opposite polarity.

22 © 2004 EAGE
first break volume 22, June 2004 technical article

the geological model is essential to obtain sound processing data will facilitate the proper identification of the position of
products. A multi-disciplinary approach is certainly advis- impedance interfaces. A zero-phase dataset is characterised
able to guarantee the best end result. The initiative of by a symmetrical wavelet, whereby the maximum of the cen-
Comparative Analysis of Seismic Processing or CASP (Ajlani tral lobe corresponds with the position of the AI interface.
et al. 2003) is a step in the right direction that takes a prop- The phase spectrum shows a near zero value for all signifi-
er balance between data quality, turnaround and costs into cant frequencies. The phase spectrum is obtained when a
account. Fourier Transform is performed (a.o. Mari et al. 1999). The
Fourier Transform decomposes the seismic trace into individ-
Wavelet extraction ual periodic sine wave functions. Each frequency has an
1. Well-to-seismic tie amplitude and phase assigned to its sine function. The results
The well-to-seismic tie is a crucial step in seismic interpreta- of this waveform decomposition is usually summarised in
tion (a.o. White and Simm 2003). In the stratigraphic inver- frequency spectra. The periodic waveform decomposition
sion scheme a comparison is made between the synthetic and can be performed on the seismic trace as well as on the seis-
the seismic trace at the well location. Several assumptions are mic wavelet itself.
made to derive the seismic signal or wavelet: The synthetic trace is computed from the calibrated sonic
■ Time bulk-shift of synthetic trace is correctly determined. and the density logs. The sonic is converted to a velocity log
■ Data cube is zero-phase. for this purpose. A reflectivity trace is computed and this is
convolved with the seismic wavelet to yield the synthetic
The latter requirement is not always needed, but zero-phase trace. The sonic logs measure the transit time between two

Figure 8A Bulk time shift is important to establish the right match between the synthetic and the seismic cube. Errors in the
time shift have a major impact on the calculated phase correction in the synthetic-to-seismic well matching operation.
Figure 8B Computation of a zerophasing operator that converts the seismic cube within a certain time interval to its near
zero-phase mode. The determination is done in a matching procedure, whereby the reflectivity trace and the seismic trace at
a well location are compared. The matching operation permits the extraction of a wavelet via cross correlation procedures.
Another extraction method is to design a least square error or Wiener filter. In this case the phase correction results in a
sharper wavelet with less side-lobe energy.

© 2004 EAGE 23
 technical article first break volume 22, June 2004

Figure 9 Match of phase rotated synthetic traces and a zero-phased seismic cube. The zero-phase condition is only valid in
the red inset computation time window (courtesy Pemex).

Figure 10 Integration of seismic traces is

done under the assumption that the densi-
ty is constant and equal to two. The high
frequency variations are based on the well
sonic logs and added to the seismically
derived velocity trend. All traces of a seis-
mic section are thus converted into pseudo
sonic logs (modified after Yilmaz 2001).

24 © 2004 EAGE
first break volume 22, June 2004 technical article

sensors in the measuring device clamped to the borehole These reflectivity spikes correspond much better to meaning-
wall. The frequency of the sonic signal is much higher than ful geological boundaries and internal reservoir interfaces
the seismic signal. The seismic method incorporates a larger (Van Riel 2000, Veeken et al. 2002).
horizontal velocity component due to the acquisition geome-
try. Hence, slightly different velocities are measured by these 2. Zero-phasing and phase rotation angle
two methods. The discrepancy between the sonic and seismic Many steps in seismic processing assume that the data are
velocities are usually corrected with the aid of a check shot zero-phase. Zero-phase means that the phase spectrum of the
or VSP survey (cf White and Simm 2003). The discrepancy is seismic is flat (between 10 and 60 Hz on Figure 7A). An
also known as the drift of the sonic log with depth. The operator is designed for zero-phasing of the seismic sub-cube
check shots allow time conversion of the sonic data. (Figure 8). The zero-phasing is normally achieved within a
The check shot survey is a simple acoustic recording of small time window of about 1 second TWT. If the window is
the first arrival travel time (one way time) between a shot selected too small, there are not enough sampling points to
close to the well head and various geophone positions in the perform a reliable calculation. The matching procedures
well. Care should be taken not to introduce artificial steps in comprises the following steps:
the calibrated sonic log at these checkshot times. This may ■ Determining and applying a bulk time shift for the well
lead to artificial reflections on the computed synthetic trace synthetic trace (Figure 8). This trace is computed from the
that do not exist in reality. If more than the first arrival is time-converted calibrated sonic and density logs.
recorded, than a Vertical Seismic Profile (or VSP) is obtained. ■ Comparison of the two traces yields a seismic wavelet and
The VSP data requires special processing to make a direct a zero-phasing operator is designed.
comparison with the seismic traces possible (a.o. Veeken, in
prep; Figure 6). Many times only a bulk phase rotation angle is calculated for
Sometimes the density is estimated from the P-velocity log the seismic cube (Veeken et al. 2002). The change in rotation
and the following relationship is used (Gardner et al. 1975): angle has an influence on the shape of the wavelet, because
the match between rotated seismic and the well reflectivity
(2) trace will be no longer the same. A bulk phase rotation
applied to all frequencies does not alter the geometry of the
The velocity is expressed in m/s and the density in g/cm3. seismic reflections, but influences the polarity aspect of the
The constant 0.31 is in fact lithology dependent and can be display (Figure 7B). The phase rotation optimises the fit
adjusted. Here it corresponds to sandy reservoir layers. between the synthetic trace and the seismic trace at the well
According to Faust (1951 and 1953) the velocity can be esti- location (Figure 9). The aim now is to design a phase rota-
mated from the resistivity log: tion whereby the wavelet is zero-phase in shape. This is equal
to a more symmetric wavelet, with pre-runners and post-cur-
(3) sors. Applying this bulk phase rotation to the dataset is in
fact equivalent to the effect of a zero-phasing operator on the
The Vp is expressed in ft/s, resistivity in ohm/ft and the max- seismic input data. After applying the phase rotation to the
imum burial depth z in feet. seismic cube, a new matching procedure is normally started
to extract the best wavelet. It is verified that the residual
A wavelet is now established by two techniques: rotation and time bulk-shift are small (< +-1ms and < 30
■ Applying cross-correlation techniques between the syn- degrees). The final wavelet is computed and the actual inver-
thetic and the seismic trace at the well location. sion of the seismic data can now begin.
■ Designing a shaping filter that permits transformation of The non zero-phase aspect of the data can also be direct-
the reflectivity trace into the seismic trace. ly incorporated in the wavelet used for the inversion. The
irregular shape of the non zero-phase wavelet compensates
The shape of the wavelet depends very much on the time for the fact that the data are not zero-phase. Amazingly it
window chosen. A stable wavelet is usually derived in a max- also can take care of a part of the systematic noise (e.g. rudi-
imum 1 second TWT time window. The change of the mentary multiple ringing).
wavelet is caused by the fact that the seismic signal gets pro-
gressively distorted on its way into the earth. Moreover, poor Various seismic inversion methods
and unreliable well logs may give rise to an erroneous Several methods are available to perform a seismic inversion.
wavelet and therefore quality control is a crucial issue The approach is either deterministic or probabilistic in
throughout the entire processing sequence. nature. The deterministic methods are represented by:
This wavelet is subsequently utilised to perform the seis- ■ Simple integration of the seismic traces.
mic inversion, whereby the seismic traces are transformed ■ Sparse spike inversion.
into blocky seismic impedance traces. The spiked response is ■ Coloured inversion.
expressed by the limits of these blocked impedance units. ■ Model driven inversion.

© 2004 EAGE 25
 technical article first break volume 22, June 2004

Figure 11 Coloured inversion of seismic data exploits the frequency spectra of the logs. An average amplitude spectrum of
the input seismic is computed to derive an inversion operator. This operator is applied to the seismic cube, so that its ampli-
tudes are in better agreement with the well data. A coloured inversion example is given for seismic traces around a well. The
lithological units are more easy to recognise on the coloured inversion sections. The coloured seismic inversion is a fast
method, but the results are rather imprecise and only suitable for a quick look approach. For more accuracy a different inver-
sion approach is highly advisable (courtesy CGG).

The stochastic inversion scheme uses a statistical description (Lindseth 1979, Yilmaz 1987). A high-frequency velocity
of the subsurface to do the inversion. Uncertainties in the component is derived from the sonic log data in a well or
input model are quantified and these are also retained in the from several wells. This velocity field is interpolated between
end results. the control points. A basic trend is established between the
The input of an inversion exercise consists of post- sonic log and the seismically derived velocity field; thus the
and/or pre-stack data. The pre-stack method exploits AVO high-frequency trend is approximated.
effects in the dataset. As stated already, it is important that It is assumed that there is no variation in density. The
the data are as clean as possible with only a limited amount seismic response is then directly translated in a vertical
of amplitude distortion if quantitative interpretation is the low-frequency velocity trend for each of the seismic
objective (Da Silva et al., in prep). The pre-stack data should traces. The high-frequency velocity component is added to
be properly migrated. this velocity trend and a pseudo sonic log trace is derived
under the assumption that the density is 2.0. In this way
Deterministic inversion all seismic traces are inverted to velocity changes and the
1. Simple integration of the seismic traces. inverted traces are called pseudo synthetic sonic logs. This
Simple integration of seismic traces has been carried out in type of trace inversion is a poor man's job that can be
earlier years (1970-80s) to obtain reflection coefficients done better.
under the assumption that the density is constant and equal
to 2.0 (Figure 10; Yilmaz 2001). 2. Coloured inversion
The seismically derived velocity trends are limited in their The coloured inversion method is based on a special filter-
frequency contents and to compensate for this phenomenon, ing technique. The amplitude spectrum of the well log in the
it is possible to construct a ‘synthetic sonic log section’ inversion window is compared with that of the seismic

26 © 2004 EAGE
first break volume 22, June 2004 technical article

dataset (Figure 11). An operator is designed bringing the seis- forms the seismic trace into an assumed acoustic impedance
mic amplitudes in correspondence with those seen in the well. equivalent. The assumption is made that the seismic input
This operator is subsequently applied to the whole seismic cube is zero-phase, which is hardly ever the case. Again, it is
cube (Lancaster and Whitcombe 2000). A crossplot is made a quick and rather imprecise inversion method.
between the amplitude and the logarithm of the frequency to
compute the operator. A linear fit is performed to calculate an 3. Sparse spike inversion
exponential function fα and this serves as a shaping filter (cf In this method the seismic trace is simulated from a mini-
Walden and Hosken 1985, Velzeboer 1981). This filter trans- mum number of AI interfaces (or reflectivity spikes) that will

Figure 12 Sparse spike inversion uses a minimum number of acoustic impedance interfaces to model the subsurface reflectiv-
ity. The algorithm was initially working on a trace by trace basis that caused some instability in the inversion results. A 3D
approach is now adopted and further constraints for the solution are provided by the low frequency variation observed by
the well control. It is interesting to note that the wavelet used in this study is not zero-phase (modified after Ronghe and
Surarat 2002).

© 2004 EAGE 27
 technical article first break volume 22, June 2004

reproduce the real seismic response when convolved with the mic trace by a pseudo acoustic impedance trace at each CDP
wavelet. Amplitude, time position and number of the reflec- position (Pendrel and Van Riel 1998). The sparse spike
tivity spikes are not always realistic, i.e. not necessarily cor- assumption implies, however, that a thin bed geometry will
responding to the geological constraints. If a starting model not always be mimicked in the most optimal way.
is not available, the spikes might be placed in an unrealistic The zero-phase requirement can be circumvented by
way and still generate a synthetic that highly resembles the choosing a compound wavelet for the inversion, thus compen-
real seismic trace. The recursive method uses a feedback sating the non zero-phase aspect of the input data (Figure 12).
mechanism to generate a more satisfactory output. The multi-trace approach results in a much better stability of
The inversion algorithm was initially only working on a the computed solution. Sophisticated model-driven sparse
trace-by-trace basis, but now a multi-trace approach is spike inversion give more realistic output. Many times the
implemented. The inversion solution may vary considerably interpreter gets away with the sparse spike approximation, but
from trace to trace, thus making the reliability of the output in the majority of cases yet a better solution is needed.
weaker. A low frequency AI variation trend can be imported
to obtain more appropriate results and get a better conver- 4. Model-driven inversion
gence for the found solutions from trace to trace. The con- In this method a simple initial AI model is perturbed and a
strained option uses a low frequency model as a guide synthetic trace is computed using the seismic wavelet. The
(Figure 12). The low frequency variation is estimated from difference between this synthetic trace and the real seismic
blocked well logs and this gives much better results (e.g. trace is determined (cf. Cook and Sneider 1983, Fabre et al.
Ronghe and Surarat, 2002). The inversion replaces the seis- 1989, Gluck et al. 1997). The AI model with a very small dif-

Figure 13A Model driven seismic inver-

sion method. A simplistic initial strati-
graphic model (macro and micro) is
convolved with the wavelet to obtain a
synthetic response that can be com-
pared with the actual seismic trace. The
macro layers are formed by the mapped
TWT horizons. Microlayers are auto-
matically introduced into the macro-
model to define a grid cell volume -
based on inline, crossline and microlay-
er – for storing constant AI values. The
AI model is perturbed, whereby the dif-
ference between inverted trace and the
seismic trace is reduced until a small
threshold value is reached (Veeken et al.
2002).
Figure 13B Arbitrary seismic line
through various wells. The inversion
result below shows anomalous layers
with a decrease in acoustic impedance.
Red represents high and blue the lower
AI values. Computational constraints
compel that the macro layers are con-
tinuous over the survey area and inter-
polation has been done in the fault zone
polygons to fill the gaps in the grids.
Inversion results in the fault zones are
therefore physically meaningless (cour-
tesy Pemex).

28 © 2004 EAGE
first break volume 22, June 2004 technical article

ference is retained as solution (Figure 13). A simulated spikes is utilised in the modelling. Normally these layers are
annealing technique using a Monte Carlo procedure is 5-7 ms TWT thick. Several parameters can be set for use in
applied (Goffe et al. 1994, Duboz et al. 1998). This tech- the inversion algorithm and corridors define boundaries for
nique shows a resemblance to the growth of crystals in a the amount of variation (Figure 14). The method is robust
cooling volcanic melt (Ma 2003). It starts with a reflectivity and a real 3D inversion algorithm is applied (Coulon et al.
model Mo and computes the difference with the seismic 2000, Veeken et al. 2002). The latter is important for the sta-
input data after convolution with a wavelet. The model is bility of the retained solution.
perturbed and a new model Mn is simulated, for which the The model-driven inversion gives satisfactory results,
same difference is established. The two differences are com- even when well control is limited and the seismic quality is
pared and if the misfit for f(Mn) is smaller than that for Mo, rather poor. It is also possible to derive a wavelet straight
than the Mn model is unconditionally accepted. If not, than from the seismics by auto-correlation techniques. Even non
the Mn model is accepted but with a probability P = exp zero-phase wavelets can be utilised as described earlier on.
(- f(Mn) - f(Mo)/T), whereby T is a control parameter Well control is not always completely honoured by this
(acceptance temperature). This acceptance rule is known as method, but a great advantage is that the seismic data are the
the Metropolis criterion (Metropolis et al. 1953). The guide for the inversion. The averaging effect of the 3D
process is repeated a large number of times, until a very small approach gives rise to small discrepancies at the well loca-
residual difference (or threshold value) is found. tions that are in fact quite acceptable.
Computation of cost functions enables the determination of Another model-driven method (Invermod) makes use of
a real regional minimum for these differences. Principal Component Analysis (PCA, Helland-Hansen et al.
The initial AI model is made up of macro-layers defined 1997). The principal component method computes a stan-
by the shape of the seismic mapped horizons. Micro-layers dard response from which the input can be generated by
are automatically introduced in this macro model. It pro- applying specific weighting factors. These weighting factors
vides a stratigraphic grid cell volume together with the inline are extrapolated over the study area to allow predictions out-
and crossline subdivision for storing constant AI values. The side the control points. The inversion needs an a-priori start-
use of micro-layers ensures that an adequate number of ing model. The structural frame of the model is based on the

Figure 14 Comparison of well data and inversion results. On the right hand side the seismic wiggles are overlying the
coloured AI values. Lateral changes in AI are of interest to determine fluid fill in reservoir sands. Inversion may even lead
to detection of the behaviour of thin beds beyond seismic resolution. Subtle changes in seismic loop shape (doublets) are
exploited in this manner.

© 2004 EAGE 29
 technical article first break volume 22, June 2004

Figure 15 Probabilistic inversion uses

quantification of uncertainties attached
to the inversion input data. Probability
density functions (PDF) are defined
and an earth model is simulated, which
is perturbed to minimise the discrepan-
cy between the modelled and measured
seismic data. The inversion results are
summarised in P10, P50 and P90
maps. Individual realisations are often
dangerous to interpret. It is more useful
to determine stable areas in the suite of
simulations. The confidence for the
reliability of the inversion results in
these zones is increased. The statistical
variability in geometry of geological
bodies is taken into account in this
inversion scheme (modified after
Torres-Verdin et al. 1999).

shape of the mapped time horizons. The initial model, with classification of the wells helps in selecting the simulation
the separate velocity and density distribution, is built by PCA approach. Probability Density Functions (PDFs) are estab-
from the well log data. Weighting factors for the standard log lished for each grid point and these are utilised to perform a
response in the studied area are determined via a linear inter- random simulation (Torres-Verdin et al. 1999; Van der Laan
polation. The results of the convolution with the seismic and Pendrel 2001). The input for the PDF determination
wavelet are compared with the seismic traces and the Vp-rho comes from: well logs, spatial properties (variograms) and
models are perturbed to reduce the discrepancy. The velocity lithological distributions (Figure 15). The stochastic algo-
and density are hereby modelled separately, which can be rithm calculates for each simulation a synthetic trace, com-
tricky sometimes. pares it with the real seismic trace, and accepts or rejects it.
A simulated annealing process is utilised. The number of
Stochastic inversion solutions is reduced in this way and probability maps are
Geostatistics are used to build complete subsurface reservoir produced to assess the amount of uncertainty. The retained
models or realisations. Simulation is done on a local level as simulations are examined on their variance. If they closely
well as globally on the totality of the generated model (Haas resemble each other, then the prediction is rather good and
and Dubrule 1994, Dubrule 2003). All models honour the the confidence in the scenarios is increased.
well data. The simulation can be based on variations per The earth models show a high resolution variability com-
pixel or objects. The architecture of reservoirs is usually clas- parable to that found in the wells. The proposed variability
sified in various ways (Weber and Van Geuns 1990) and this is clearly beyond seismic resolution. When production data

30 © 2004 EAGE
first break volume 22, June 2004 technical article

Figure 16A Elastic impedance logs

computed for the various offset angles.
The top of the gas reservoir is well
expressed at the larger incidence angles.
The Vp/Vs ratio also shows a remark-
able break at this contact.
Figure 16B Estimation of Vp and Vs
velocities. A polynomial trend is pro-
posed by Castagna (1993) that is
adaptable to sandy, silty, shaley and
carbonate reservoirs. These estimations
are needed to calculate the Elastic
Impedance attribute at the different
incidence angles (Connolly 1999).

are integrated in the modelling constraints, then the value of P10-P90 maps can be somewhat misleading as the outcome of
the simulated model is increased accordingly. Production his- an individual simulation is not always realistic, e.g. very rapid
tory matching and pressure monitoring gives an indication changes in geological style that are an artefact of the working
for the connected reservoir volume and its permeability. method. It is better to use maps that are some sort of an aver-
A probability volume is generated for grid points with age and concentrate on the areas that are rather stable in sever-
porosities above 10%, using the simulation histograms. al simulations. The predictive value of the simulation is strong-
Subsequently geobodies are outlined, where the probability ly reduced in areas with great variability in outcome of the real-
is for instance above 70% for the porosity to be higher than isations. The averaging procedure has, however, a negative
10%. As more wells are drilled in the petroleum system, the effect on the high resolution aspect of the proposed solution.
best matched simulations are retained to further refine the
predictions (Sylta and Krokstad 2003). Pre-stack inversion
The drawback of the method is that the interpreter has to Elastic inversion (EI)
quantify the uncertainties in a realistic way. This a tedious and The method exploits AVO effects in the pre-stack domain and
somewhat precarious task. Areas without proper well control uses Vp, Vs and density information. Shear waves may yield
are difficult to predict and certain assumptions have to be valuable information on the lithological distribution (Pendrel
made. There is a cumulative increase in prediction error. et al. 1998). The S-waves mainly interact with the rock frame
Another problem is selection of a suitable presentation of the work, whereas the P-waves are influenced by the porefill and
results. Normally only the P10, P50 and P90 cases are retained rock matrix. The Vs is often estimated in wells using the
in the output and the rest of the realisations are ignored. These Castagna formula (Castagna and Backus 1993, Figure 16).

© 2004 EAGE 31
 technical article first break volume 22, June 2004

Figure 17 A flow chart for elastic inver-

Well data Seismic CDP gathers sion processing. The procedure is
rather complex and time consuming.
Vp Vs rho PS Near PS Mid PS Far PS stands for partial stacks and ZP for
zerophased seismic volume. The ‘EI 1
Connolly EI computation Spectral harmonisation trace’ is the EI response computed from
the well data at the centre position of
EI Near trace EI Mid trace EI Far trace PS Near PS Mid PS Far
Near offset range; the EI 2 and 3 traces
are corresponding with the Mid and
Far. The procedure gives access to rock
Well
Welltotoseismic
seismicmatch
match
for physical parameters like the attributes
forPS
PSMid
Mid
Rp, Rs, Ip, Is, Vp/Vs, rhomu and rho-
lambda.
Phase
Phaserotation
rotation
1

Wavelet Mid ZP PS Mid ZP PS Near ZP PS Far

Inversion 2 3

Inversion
Simple initial EI model

Inversion

EI
EIMid
Midcube
cube EI Mid Model
Time grids Complex EI model 1
microlayers
Wavelet Near
Complex EI model 2

Wavelet Far
EI
EINear
Nearcube
cube

EI
EIFar
Farcube
cube
Model-driven EI inversion scheme

The full Zoeppritz equations give a description of the This function F(t) is now called the Elastic Impedance, in
amplitude behaviour of reflector with offset (Zoeppritz analogue to the acoustic impedance concept. The angle
1919). These equations are tedious to work with and there- dependant P-wave reflectivity is also approximated by the
fore an approximation is introduced, valid under certain well known simplified description of the Zoeppritz equa-
conditions (e.g. Aki and Richards 1980; Shuey 1985). The tions:
Shuey equation gives an approximation of the Zoeppritz
reflectivity that is good up to 30-35° incidence angle. The R(T) = A + B sin2 T+ C sin2 T tan2 T   (5)
assumption is made that Vp is approximately twice Vs and
the higher terms are dropped due to the 30° angle of inci- A = 0.5 ( ∆Vp/Vp + ∆ρ/ρ)
dence condition. When working with pre-stack data, it is B = 0.5 ( ∆Vp/Vp ) – 4 (Vs/Vp)2 ( ∆Vs/ Vs) – 2 (Vs/Vp)2
necessary to make a better estimation by taking into (∆ρ/ρ)
account the difference in Vp and Vs. This is done by adopt- C = 0.5 ( ∆Vp/Vp )
ing a so-called elastic approach. Connolly (1999) intro-
duced the concept of Elastic Impedance. He defined a func- ∆Vp = Vp2 – Vp1
tion F(t) that is incidence angle dependent and related to Vp = (Vp1 + Vp2 ) / 2
the P-wave reflectivity in the following manner:
Combining the two expressions (4) and (5) results in the
R(T = (F(ti) - F(ti-1)) / (F(ti) + F(ti-1)) (4) Elastic Impedance being equal to:

32 © 2004 EAGE
first break volume 22, June 2004 technical article

making the necessary source and cable corrections, spherical

(6) divergence correction, static and topographic corrections,
prestack time migration, multiple and noise suppression,
Here K is a constant, that is taken to be equal to the average DMO, NMO and residual NMO in order to get the best
of (Vs/Vp)2. Vp and Vs are expressed in m/s and the density migrated stack section. The amplitude should preserve its
in gr/cm3. This type of EI computation is performed on pre- proportional aspect to the original subsurface acoustic
stack gathers and takes into account the changes in Vp, Vs impedance contrasts or reflection coefficients. Visual inspec-
and density as well as AVO effects. The approach is accurate tion and quality control tests play a key role in securing that
for small to moderate impedance changes. Dropping the this objective is optimally achieved.
third term in the Shuey equation in formula (5) - which is less The benefits of the elastic impedance calculations are
accurate, but faster - is equivalent to replace the tan2 θ by clearly demonstrated by Figure 16. It can be easily seen that
sin2 θ in the Connolly equation (6). the larger offset angles give a better discrimination at the top
of the gas reservoir. Also the Vp/Vs ratio shows a distinct
(7) break at this interface. The EI0 is corresponding to the
acoustic impedance AI (= ρ Vp). If K = 0.25 then the EI90 =
There are some assumptions that need to be fulfilled for this (Vp/Vs)2.
formula (7) to be correct: The EI seismic attribute is the basis for performing an
■ Two term NMO approximation is correct. elastic inversion similar to conventional acoustic inversion
■ Dix's velocities are often required in the raytracing going processing. In acoustic impedance inversion, a wavelet (shap-
from the offset to the angle of incidence domain. Dix's ing filter or cross-correlation techniques) is established
equation (Dix 1955) is valid for the following situation: between the AI trace from the well and the recorded seismic
• A layer cake geometry. trace. In the elastic inversion, wavelets are derived for differ-
• Offset is smaller than the depth of the reflector. ent incidence angle traces of EI(θ) and the corresponding
■ Angle of incidence θ is smaller than 30-35°, so that the migrated partial stack trace. A schematic EI workflow is pre-
Shuey approximation is correct (Shuey 1985). sented in Figure 17. The diagram is rather complex and it
■ Transverse isotropic medium. reflects the amount of effort needed to get to the goal.
■ Correctly balanced pre-stack amplitudes. There are other formulae to approximate the EI, e.g. log-
■ Amplitudes are proportional to sin2 θ. arithmic approach or also a less common non-linear function
(Tarantola 1984, 1986; Pica et al. 1990). The logarithmic
The last point implies taking out the acquisition imprint by function avoids tedious exponential descriptions and is valid
under the condition that K = 0.25 (Figure 18):

(8)

whereby IP = ρ Vp.

All these formulae have their own assumptions and valid-

ity range. It means that the elastic inversion results are often
only usable in a qualitative way, as the absolute value of the
inversion is not necessarily correct. For quantitative interpre-
tation more processing efforts are needed, like exploiting the
full set of Zoeppritz equations.
The far offsets of these EI cubes give detailed information
on the fluid contents (Figure 19). Attributes like Rp, Rs, Ip,
Is, Vp/Vs, rhomu and rholambda are easily calculated. The
rhomu and lambdarho (cf Goodway et al. 1997) are derived
from the following expressions:

Rhomu = Is2 (9)

Lambdarho = Ip2 – 2Is2 (10)
Figure 18 The Elastic Impedance approximation as intro-
duced by Connolly (1999). The logarithmic simplification is The Mu (µ) and Lambda (λ) are known as the Lame’s con-
done under the assumption that K = 0.25. The main advan- stants. Mu is the rigidity factor. The Lambda parameter is the
tage is that the logarithmic EI representation avoids an awk- incompressibility and contains the fluid information. Note
ward exponential notation. that it is not possible to get access to the density and veloci-

© 2004 EAGE 33
 technical article first break volume 22, June 2004

Figure 19 The integrated approach of

Elastic impedance sections an elastic inversion incorporates the
AVO effect on the traces of the migrat-
EI 10 degrees
ed CDP gather and uses Vp and Vs
information. Various angle dependent
EI cubes and other seismic attributes
are examined on anomalies that may
correspond to lithology, porosity or
fluid changes. The elastic inversion
EI 20 degrees
yields several seismic reservoir attrib-
utes. Crossplots demonstrate their cor-
relation with porefill changes.

EI 30 degrees

EI inversion attributes

Imp S LambdaRho

RhoMu Vp/Vs

ty values directly for each separate layer via an inversion perturbed P- and S-reflectivity models. The method is
scheme. It is always necessary to make an estimation of their described in detail by Ma (2002) and his article serves here
individual contribution to the total change in impedance of as a guide. The approach is basically model-driven. The
the various layers later on. inversion is achieved by applying a simulated annealing tech-
The calculated attributes are studied in detail and the nique (Ma 2002). This is opposed to a genetic technique,
attention is usually focused on anomalies. Well plots in zones considering the biologic evolution as a basis for the guided
of interest are utilised to carry out quantitative interpretation Monte Carlo approach (e.g. Mallick et al. 1995).
and to perform reservoir parameter predictions. These The Aki and Richards formula (1980) gives access to the
parameters can also be obtained via AVO analysis, but that approximate P-wave reflectivity at the various offsets in the
calculation is less reliable (Cambois 2000). pre-stack domain.

2. Simultaneous inversion R(T) = 0.5( ('Vp/Vp) +('U/U)) – 2(Vs/Vp)2 (2 ('Vs/Vs) +

Simultaneous inversion is calculating synthetic gathers from ('U/U))sin2 T+ 0.5 ('Vp/Vp) tan2 T (11)

34 © 2004 EAGE
first break volume 22, June 2004 technical article

Figure 20A Simultaneous inversion is

basically model-driven in approach.
The Aki and Richards approximation
of the Zoeppritz equations is used to
compute the reflectivities at different
offsets. P- and S-wave acoustic imped-
ance models are perturbed in order to
compute a new reflectivity and synthet-
ic CDP gathers. These are compared
with real seismic data and the differ-
ence is minimised by a simulated
annealing technique. The advantage of
the simultaneous inversion is that it
avoids the limitation in the degree of
the angle of incidence and the same
time model is applicable for all offsets
(modified after Ma 2002).
figure 20B Cross section across simul-
taneous inversion results. The proposed
new well location was later successful
in finding additional commercial
hydrocarbons (courtesy Fugro-Jason
Geoscience).

done under the following assumptions:

where Vp is the average P-velocity between two uniform half- - The earth has approximately horizontal layers.
spaces, Vs is the average S-velocity and r is the average density. - Each layer is described by both acoustic and shear imped-
It can be rewritten in terms of P- and S-wave impedances: ances.
The reflection coefficients can be calculated for the n-th layer
R(T) = (1 + tan2 T('Ip/2Ip) - 8 (Vs/Vp)2 sin2 (T('Is/2Is)) - in the starting model in the following way:
(tan2 T(Vs/Vp)2 sin2 T ('U/2U) (12)
'Ip/2Ip = (Ipn – Ipn-1) /(Ipn + Ipn-1) (14)
The assumption is now made that the relative changes in
(∆Vp/Vp), (∆Vs/Vs) and (∆ρ/ρ) are small and the incidence 'Is/2Is = (Isn – Isn-1) /(Isn + Isn-1) (15)
angle θ is much less than 90°. This means that the second
order terms can be ignored (Fatti et al. 1994): (Isn + Isn-1) / (Ipn + Ipn-1) (16)

These functions are utilised to compute the reflectivity at all

R(T) = (1 + tan2 T('Ip/2Ip) - 8 (Vs/Vp)2 sin2 T('Is/2Is) (13) angles in the formula above. This is convolved with the
wavelet to obtain a synthetic gather. The synthetic gather is
The background Vs/Vp relationship should be known to compared with the original seismic CDP gather and the mis-
solve this equation. If the Vp and Vs models are not good fit is computed (Figure 20A). The model is subsequently per-
approximations of the earth model, then the linear inversion turbed and a new comparison made to reduce the misfit. The
will give incorrect results. Ma (2002) proposes therefore to convolution assumes a plane wave propagation across the
substitute the average Vs/Vp by the average Is/Ip values. This boundaries of horizontally homogeneous layers and does not
Is/Ip value is not coming from an a-priori model but is take into account geometrical divergence, non-elastic absorp-
derived individually for each inversion iteration. Inversion is tion, wavelet dispersion, transmission losses, mode conver-

© 2004 EAGE 35
 technical article first break volume 22, June 2004

sions and multiple reflections. These issues should have been et al. 2000). It is a computing intensive method; therefore it is
addressed in the data pre-conditioning step. often combined with other non-linear estimation and correla-
The inversion transforms the seismic cube into reflectivi- tion schemes. Neural network training is a powerful tool to
ty cubes for different offset ranges. Cross sections and layer convert seismic data into pseudo well log traces (e.g. Banchs
maps are convenient to visualize lateral changes in the inver- and Michelena 2002). Interval confidence calculation is done
sion results (Figure 20B). The advantage of the simultaneous in order to adopt a more statistical approach.
inversion is that there are few constraints on the validity of The first four inversion methods described above are
the computation caused by the offset angles used. A simple known as acoustic impedance inversions. The model-driven
initial Vp model serves as input, summarising the expected inversion is the best suited for poor data quality surveys with
low frequency variation. This approach guarantees stability limited well control (estimated sonic, estimated density, no
in the end-result. A density model is constructed applying checkshots?) as the seismic data itself is the basic guide for
Gardner’s estimation (Gardner et al 1975) or a very simple the inversion. Elastic inversion is labour intensive and should
Vp/Vs model is used (as approximation a constant value of only be undertaken when a feasibility study has demonstrat-
2). A further advantage is that the same time model is appli- ed its benefits. A shear sonic is needed or estimated, but even
cable for all offset angles. The calculations are usually done that is not always necessary according to Cambois (2002).
for six discrete offset angle ranges. The following attributes His suggested fluid factor basically boils down to the behav-
are computed: Vp, Vs, rhomu, lambdarho and Vp/Vs. iour of the far offset stack.
The two term Aki and Richards formula is frequently
written in a simplified form (5). It gives access to the RP and
RS reflectivities for the zero offset in the following way: A word of caution
As a word of warning: ‘Seismic inversion is not a unique
Rθ = A + B sin2 θ + C sin2 tan2 θ process’. There are several AI models that generate similar
synthetic traces when convolved with the wavelet. The num-
A = 0.5 ( ∆Vp/Vp + ∆ρ/ρ) ber of possible solutions is significantly reduced by putting
B = 0.5 ( ∆Vp/Vp ) – 4 (Vs/Vp)2 ( ∆Vs/ Vs) – 2 (Vs/Vp)2 constraints on the modelling and, in doing so, most plausible
(∆ρ/ρ) scenario is retained (Veeken et al. 2002, Da Silva et al., in
C = 0.5 ( ∆Vp/Vp ) prep). The support of other investigation techniques, like
AVO analysis and forward modelling, increases the confi-
A is the intercept, B is the gradient in AVO analysis and C is dence in the inversion results. Even a negative correlation is
the AVO curvature. If Vp / Vs = 2, then for the zero degree important information as it results in an increase of the risk
incidence angle (Russell et al, 2003) the two term approxi- attached to the prospect. It may sound a bit controversial,
mation is valid and : but ultimately it will reduce the drilling risk on the well pro-
posals because of the better ranking criteria.
RP 0 = A (17) Seismic velocities are sensitive to the presence of gas in a
rock sequence. A 5-10% gas saturation has already a tremen-
RS 0 = (A-B) /2 (18) dous impact on the seismic response. It will lead to AVO and
AI anomalies, but these are non-commercial. The extent of
The density of the rock for the P- and S-wave is the same, but the mapped anomalies should therefore always be treated
the velocity contrasts change. with some care. AVO and inversion anomalies are likely
Also a linearised Bayesian approach has been adopted in related to the maximum distribution of hydrocarbons.
an AVO inversion scheme (Buland and Omre 2003). The Seismic inversion depends heavily on the proper integra-
Bayes theorem exploits the conditional probability principle. tion of well data. Incorporation of anisotropy effects in the
The posterior solution is given by a Gaussian probability inversion scheme will improve the quality of the output data
density function, whereby the calculations are based on a (Rowbotham et al. 2003). This is especially important when
Monte Carlo simulation. The linearisation is made possible dealing with deviated holes.
by assuming weak impedance contrasts for the Zoeppritz Seismic inversion is gradually becoming a routine pro-
equations. A drawback is that noise in the data has a detri- cessing step in field development studies as well as for
mental effect on the uncertainties in the solution. exploration purposes. Even time lapse inversions are now
conducted (Gluck et al. 2000, Oldenziel 2003). The AI
3. Pre-stack waveform inversion attribute is gradually replacing the normal amplitude seis-
This method is based on a wave equation forward modelling mic representation as inversion is becoming an integral part
algorithm. Numerous earth models are fitted to the pre-stack of the reservoir characterisation workflow (cf Latimer et al.
response of the individual traces. The wave equation is run 2000, Van Riel 2000). The positive track record of case his-
with converted wave energy, interbed multiples, transmission tories clearly demonstrates the added value of this type of
losses and P-wave reflections (Benabentos et al. 2002, Mallick seismic analysis.

36 © 2004 EAGE
first break volume 22, June 2004 technical article

Conclusions sion. Geophysics, 68, 1, 185–198.

Inversion replaces the seismic signal by a blocky impedance Cambois, G. [2000] AVO inversion and elastic impedance.
response. Various seismic inversion schemes are available, Expanded abstracts, 70th SEG Annual Meeting, Calgary, 1-4.
each having their own advantages. Input data conditioning is Castagna, J.P. and Backus, M.M. [1993] Offset dependent reflec-
an important step when quantitative interpretation is the tivity – Theory and practice of AVO analysis. SEG, Tulsa,
ultimate goal. Model-driven inversion is robust, even when Investigations in Geophysics 8, 348.
dealing with poor data sets. A true 3D approach stabilises Connolly, P. [1999] Elastic impedance, The Leading Edge, 18, 4,
the output. The probabilistic method quantifies uncertainties 438-452.
in the geological model and incorporates these in the end Cook, D.A. and Sneider, W.A. [1983] Generalized linear inversion
results. The shared earth model is suited as input to reservoir of reflection seismic data. Geophysics, 48, 665-676.
simulation packages. The pre-stack approach exploits AVO Cosentino, L. [2001] Integrated reservoir studies. Technip, Paris,
effects present in the seismic dataset. All of these methods 310.
have their own restrictions and limitations. Feasibility and Coulon, J.P., Duboz, P. and Lafet, Y. [2000] Moving from seismic
synthetic modelling studies are recommended before starting to layered impedance cube and porosity prediction in the Natih E
an inversion and/or AVO project. member. GeoArabia, 5, 1, 72-73.
There is a trade-off between work involved / cost / time Da Silva, M., Rauch-Davies, M., Soto Cuervo, A. and Veeken, P.
and quality of the end product. Seismic inversion is not a in prep. , Pre- and post-stack attributes for enhancing production
unique process, i.e. there is not a single solution to the given from Cocuite gas reservoirs. 66th EAGE Annual Conference,
problem. In other words: several AI or EI models may equal- Paris [2004]
ly well explain the same seismic response. Numerous studies Da Silva, M., Rauch-Davies, M., Soto Cuervo, A. and Veeken, P.
have already proven the added value of inversion processing in prep. , Data conditioning for a combined inversion and AVO
for better definition of prospects and leads, improved DHI study. 66th EAGE Annual Conference, Paris, [2004]
detection, delineation of sweet spots with better porosity and Dix, C.H. [1955] Seismic velocities from surface measurements.
permeability characteristics. All these aspects allow optimisa- Geophysics, 20, 68-86.
tion of field development plans with improved volumetric Duboz, P., Lafet, Y. and Mougenot, D. [1998] Moving to layered
estimations and make for more reliable economic forecasts impedance cube: advantages of 3D stratigraphic inversion. First
possible. Break, 17, 9, 311-318.
Dubrule, O. [2003] Geostatistics for seismic data integration in
Acknowledgements earth models. EAGE distinguished instructor series 6, 282.
We are indebted to CGG, CMG, Pemex and Jason Dutta, N.C. [2002] Geopressure prediction using seismic data:
Geoscience companies for granting permission to use their Current status and the road ahead. Geophysics, 67, 6, 2012-
data. We thank our immediate colleagues for sharing their 2041.
working ideas. In particular we want to thank Dr E. Mendez, Fabre, N., Gluck, S., Guillaume, P. and Lafet, Y. [1989] Robust
M. Rauch-Davies, H. Bernal, G. Velasquez, A. Marhx, J. multichannel strati-graphic inversion of stacked seismic traces,
Camara, N. Van Couvering, R. Walia, J. Helgesen, M. 59th Annual Meeting SEG, 943.
Querne, Y. Lafet, J.M. Michel, C. Pinson, S. Addy, R. Fatti, J.L., Smith, G.C., Vail, P.J., Strauss, P.J. and Levitt, P.R.
Martinez, JL. Gelot, P. Van Riel and P. De Beukelaar [1994] Detection of gas in sandstone reservoirs using AVO analy-
(Solegeo) for their contributions. The comments of the sis.: A 3-D seismic case history using the Geostack technique.
reviewers have been highly appreciated. Special thanks is Geophysics, 59, 1362-1376.
given to Dr M. Bacon for his constructive remarks. Faust, L.Y. [1951] Seismic velocity as a function of depth and geo-
logic time. Geophysics, 16, 192-206.
References Faust, L. Y. [1953] A velocity function including lithologic varia-
Aki, K. and Richards, P.G. [1980] Quantitative seismology, the- tions. Geophysics, 18, 271-287.
ory and methods. Freeman, San Francisco, 557. Gardner, G.H.F., Gardner, L.W. and Gregory, A.R. [1974]
Ajlani, G., Al Kaabi, M. and Suwainna, O. [2003] Comparative Formation velocity and density - The diagnostic basics for strati-
analysis ( CASP): a proposal for quantifying seismic data process- graphic traps, Geophysics, 39, 770-780.
ing. The Leading Edge, 22, 1, 46-48. Gluck, S., Juve, E. and Lafet, Y. [1997] High resolution imped-
Banchs, R. E. and Michelena, R.J. [2002] From 3D seismic attrib- ance layering through 3D stratigraphic inversion of post stack
utes to pseudo-well-log volumes using neural networks: practical seismic data. The Leading Edge, 16, 1309- 1315.
considerations. The Leading Edge, 21, 10, 996-1001. Gluck, S., Deschizeaux, B., Mignot, A. Pinson and Huguet, F.
Benabentos, M., Mallick, S., Sigismondi, M. and Soldo, J. [2002] [2000] Time-lapse impedance inversion of post-stack seismic data.
Seismic reservoir description using hybrid seismic inversion: a 3D Expanded abstracts, 70th Annual Meeting SEG, Calgary, 1509-
case study from the Maria Ines Oeste Field, Argentina. The 1512.
Leading Edge, 21, 10, 1002-1008. Goffe, W.L., Ferrier, G.D. and Rodgers, J. [1994] Global optimi-
Buland, A. and Omre, H. [2003] Bayesian linearized AVO inver- sation of statistical functions with simulated annealing. Journal of

© 2004 EAGE 37
 technical article first break volume 22, June 2004

Econometrics, 60, 65-100. impedance inversion. First Break, 21, 53-57.

Goodway, W., Chen, T. and Downton, J. [1997] Improved AVO Russell, B.H., Hedlin, K., Hilterman, F.J. and Lines, L.R. [2003]
fluid detection and lithology determination using Lame’s petro- Fluid property discrimination with AVO: A Biot-Gassmann per-
physical parameters: LambdaRho, MuRho and Lambda/Mu spective. Geophysics, 68, 1, 29-39.
fluid stack from P and S inversions. Canadian Society of Sheriff, R.E. [2002] Encyclopedic dictionary of exploration geo-
Exploration Geophysicists, Abstracts, 148-151. physics. SEG, Tulsa, 429.
Haas, A. and Dubrule, O. [1994] Geostatistical inversion – a Shuey, R.T. [1985] A simplification of the Zoeppritz equations.
sequential method of stochastic reservoir modelling constrained Geophysics, 50, 609-614.
by seismic data. First Break, 12 , 11, 561-569. Stewart, R.R., Gaiser, J.E., Brown, R. and Lawton, D.C. [2003]
Hardage, B.A., [1985] Vertical seismic profiling - a measurement Converted-wave seismic exploration: applications. Geophysics,
that transfers geology to geophysics: in Berg O.R. and Woolverton 68, 1, 40-57.
D.G. (eds) [1985,] Seismic stratigraphy II: an integrated approach Sylta, O. and Krokstad, W. [2003] Estimation of oil and gas col-
to hydrocarbon exploration, AAPG Memoir 39, AAPG, Tulsa, umn heights in prospects using probabilistic basin modelling
13-34. methods. Petroleum Geoscience, 9, No. 3, 243-254.
Helland-Hansen, D., Magnus, I., Edvardsen, A., and Hansen, E. Tarantola, A. [1984] Inversion of seismic reflection data in the
[1997] Seismic inversion for reservoir characterisation and new acoustic approximation. Geophysics, 49, 1259–1266.
well planning in the Snorre field. The Leading Edge, 16, 3, 269- Tarantola, A. [1986] A strategy for non linear elastic inversion of
273. seismic reflections data. Geophysics, 51, 1893-1903.
Lancaster, S. and Whitcombe, D. [2000] Fast track "coloured" Torres-Verdin, C., Victoria, M., Merletti, G. and Pendrel, J.
inversion. Expanded abstracts, 70th SEG Annual Meeting, [1999] Trace-based and geostatistical inversion of 3-D seismic
Calgary, 1572-1575. data for thin sand delineation: An application in San Jorge Basin,
Latimer, R.B., Davison, R. and Van Riel, P. [2000] Interpreter's Argentina. The Leading Edge, 18, 9 , 1070-1077.
guide to understanding and working with seismic derived acoustic Trad, D., Ulrych, T. and Sacchi, M. [2003] Latest views of the
impedance data. The Leading Edge, 19, 3, 242-256. sparse Radon Transform. Geophysics, 68, 1, 386-399.
Lindseth, R. [1979] Synthetic sonic logs - a process for strati- Van der Laan, J. and Pendrel, J. [2001] Geostatistical simulation
graphic interpretation. Geophysics, 44, 3-26. of porosity and risk in a Swan Hills reef. Expanded abstracts,
Ma, X. [2002] Simultaneous inversion of prestack seismic data 71th SEG Annual Meeting, 1588-1591.
for rock properties using simulated annealing. Geophysics, 67, Van Riel, P., [2000] The past, present and future of reservoir char-
1877-1885. acterization. The Leading Edge, 19, 8, 878-881.
Mallick, S. [1995] Model-based inversion of amplitude variations Vazquez, R., Mendoza, A., Lopez, A., Linares, M. and Bernal, H.
with offset data using a genetic algorithm. Geophysiscs, 60, 939- [1997] 3-D seismic role in the integral study of the Arcabuz-
954. Culebra field, Mexico. The Leading Edge, 16, 12, 1763-1766.
Mallick, S., Lauve, J. and Ahmad, R. [2000] Hybrid seismic inver- Veeken, P.C.H., (in prep.), Seismic stratigraphy, basin analysis and
sion: A reconnaissance tool for deep water exploration. The reservoir characterisation. Elsevier, Amsterdam, 350.
Leading Edge, 19, 38-43. Veeken, P.C.H., Rauch, M., Gallardo, R., Guzman, E. and Vila, R.
Mari, J.L., Glangeaud, F. and Coppens, F. [1999] Signal process- Villasenor [2002] Seismic inversion of the Fortuna National 3D
ing for geologists and geophysicists. Editions Technip, Paris, 480 survey, Tabasco, Mexico. First Break, 20, 5, 287-294.
Metropolis, N., Rosenbluth, A., Rosenbluth, M., Teller, M., and Velzeboer, C.J. [1981] The theoretical seismic reflection response
Teller, E. [1953] Equation of state calculations by fast computing of sedimentary sequences. Geophysics, 46, 843-853.
machines. Journal Chem Phys, 21, 1087-1092. Walden, A.T. and Hosken, J.W.J. [1985] An investigation of the
Oldenziel, T. [2003] Time lapse seismic within reservoir engineer- spectral properties of primary reflection coefficients. Geophysical
ing. Thesis, Delft University of Technology, 204. Prospecting, 33, 400-435.
Onderwater, J., Wams, J. and Potters, H. [1996] Geophysics in Weber, K.J. and Van Geuns, L.C. [1990] Framework for con-
Oman. GeoArabia, 1, 2, 299-324. structing clastic reservoir simulation models. Journal of Petroleum
Pendrel, J., Stewart, R.R. and Van Riel, P. [1998] Interpreting sand Technology, 42, 10, 1248-1297.
channels from 3C-3D seismic inversion. Expanded Abstracts, White, R. and Simm, R. [2003] Tutorial: Good practice in well
71th SEG Annual Meeting, 1588-1591. ties. First Break, 21, 10, 75-83.
Pica, A., Diet, J. and Tarantola, A. [1990] Non linear inversion of Yilmaz, O. [1987] Seismic data processing, Society of
seismic reflection data in a laterally invariant medium. Exploration Geophysicists, Investigations in geophysics, 2, Tulsa,
Geophysics, 55, 284-292. SEG, 526.
Ronghe, S. and Surarat, K. [2002] Acoustic impedance interpre- Yilmaz, O. [2001] Seismic data analysis, Volume 1 and 2. Society
tation for sand distribution adjacent to a rift boundary fault, of Exploration Geophysicists, Investigations in geophysics, 10,
Suphan basin, Thailand. AAPG Bulletin, 86, 10, 1753-1771. Tulsa, SEG, 2027.
Rowbotham, P., Marion, D., Eden, R., Williamson, P., Lamy, P. Zoeppritz, K. [1919] On the reflection and propagation of seis-
and Swaby, P. [2003] The implications of anisotropy for seismic mic waves, Erdbebenwellen VIIB, Gottinger Nachrichten I, 66-84.

38 © 2004 EAGE

View publication stats