M.G.

Lamont
1
, T.A. Thompson
1
and C. Bevilacqua
2
1
DownUnder GeoSolutions Pty Ltd
80 Churchill Ave
Subiaco WA 6008
2
Apache Energy Limited
Level 3, 256 St Georges Tce
Perth WA 6000
[email protected]
[email protected]
[email protected]
ABSTRACT
The aim of quantitative interpretation (QI) is to predict
lithology and uid content away from the well bore.
This process should make use of all available data, not
well and seismic data in isolation. Geological insight
contributes to the selection of meaningful seismic
attributes and the derivation of valid inversion products.
Uncertainty must be taken into account at all stages
to permit risk assessment and foster condence in
the predictions. The use of the Bayesian framework
enables prior knowledge, such as a geological model,
to be incorporated into a probabilistic prediction, which
captures uncertainty and quanties risk.
Nostradamus is a uid and lithology prediction toolkit
that forms part of a comprehensive QI workow. It utilises
a Bayesian classication scheme to make quantitative
predictions based upon inverted seismic data and
depth-dependent, stochastic rock physics models.
The process generates lithology and uid probability
volumes. All available information is combined using
geological knowledge to create a realistic pre-drill model.
Separately, stochastically modelled multidimensional
crossplots, which account for the uncertainty in the rock
and uid properties (based on petrophysical analyses
of well data), are used to build probability density
functions such as acoustic impedance (AI) vs Vp/Vs
and LambdaRho vs MuRho. These are then compared
to crossplots of equivalent inverted data to make
predictions and quantitatively update the geological
model. Individual probability volumes as well as a most-
likely lithology and uid volume are generated. This
paper presents a case study in the Carnarvon Basin that
successfully predicts uids and lithologies away from
well control in a way that effectively quanties risk and
reserves. Two of the three successful gas exploration
wells were drilled close to dry holes.
KEYWORDS
Inversion, case study, quantitative interpretation, gas,
probability, Carnarvon Basin, Mungaroo Formation.
INTRODUCTION
The Mungaroo Formation is a major exploration target
in the Carnarvon Basin, offshore Western Australia.
The Mungaroo Formation sandstones (referred to as
sands hereafter) were deposited during the Late Triassic
(Carnian-Norian) as part of an extensive uvial system.
This system comprises stacked braided channel belts that
attain thicknesses of several tens of metres, typically as a
complex network of interconnecting sandstones encased
by overbank shales, siltstones, carbonaceous shales
and thin coals in parts. The porosity and permeability of
sandstones from the Mungaroo Formation can be high and
are relatively well preserved with depth.
Tectonic activity initiated in the Early Jurassic led
to extensional faulting and block rotation resulting in
signicant erosion and the emergence of a structural
NNESSW horst and graben trend. Subsequent sea level
rises and subsidence following the cessation of rifting along
Australias northwestern margin led to the deposition of the
Cretaceous Muderong Shale which drapes the Mungaroo
Formation horsts and grabens. The Muderong Shale
consists of regionally thick marine shales with excellent
sealing capabilities. In this study area the Muderong Shale
overlies the subcropping Mungaroo Formation at a major
unconformity surface.
Gas migration into the Mungaroo Formation is believed
to have occurred through a combination of selective fault
and sand channel conduits, however, not all apparently
sealed sandstone channels were charged, possibly
because of complete isolation from the migration fairway.
Gas saturation in these sands is usually associated with
high seismic amplitudes, however, ambiguity exists due to
high seismic amplitudes associated with high porosity brine
sands and also carbonaceous shales. This study outlines
the comprehensive quantitative interpretation (QI) workow
that was successful at identifying the gas bearing sands.
Prior to the QI study four dry holes had been drilled in
the block. The initial QI study was based on these four dry
holes. Three signicant gas discoveries have now been
drilled. The discovery wells are close to and, in one case,
down dip from the previously drilled dry holes. The study
has been updated after each new well.
This paper was rst published in APPEA Journal 2008
1
Matthew G. Lamont
lead author
Drilling Success As A Result of
Probabilistic Lithology and Fluid Prediction
A Case Study in the Carnarvon Basin, WA
www.dugeo.com
Best Paper at
APPEA 2008
2
M.G. Lamont, T.A. Thompson and C. Bevilacqua
The aim of the study was to predict lithology and uid
content away from the well bore, and in particular to identify
gas sands. The process made use of all available data, not
just well and seismic data in isolation. Geological insight
contributed to the selection of meaningful seismic attributes
and the derivation of valid inversion products. Uncertainty
modelling was taken into account to permit risk assessment
and foster condence in the predictions. The use of the
Bayesian framework enabled prior knowledge such as a
geological model to be incorporated into a probabilistic
prediction, which captured uncertainty and quantied risk.
Nostradamus is a uid and lithology prediction toolkit.
It utilises a Bayesian classication scheme to make
quantitative predictions based on inverted seismic data and
depth-dependent, stochastic rock physics modelling.
Stochastically modelled multidimensional crossplots,
which account for uncertainty in rock and uid properties
(based on petrophysical analyses of well data), were used
to build 2D probability density functions (AI vs Vp/Vs).
These were compared to crossplots of equivalent inverted
seismic data to make predictions and quantitatively update
the geological model. Individual probability volumes as well
as a most-likely lithology and uid volume were generated.
The study incorporated petrophysics, rock physics,
geology, geophysics and uncertainty to successfully predict
uids and lithologies away from well control in a way that
effectively assists the determination of risk and reserves.
BACKGROUND
QI is concerned with making lithology and uid predictions
based on a limited amount of known data (the wells)
and widespread observed data (the seismic data). Prior
geological knowledge needs to be used to set the context
for the analysis. Figure 1 shows the QI workow used for
this project.
QI relies on the seismic observations being a direct
result of the impedance contrasts of the rock strata being
studied. That is, the relative seismic amplitudes need to be
preserved. Adequate preparation of seismic data for QI is
imperative, and in our observation this is often done poorly.
On this occasion good quality angle stacks were available,
therefore the seismic data preparation was conned to:
Amp||tude preserv|ng frequenoy enhanoement,
geared to both enhance frequencies generally, and
also to help balance the frequency content of the
angle stacks; and,
Ang|e staok a||gnment. The |nvers|on assumes that
each event is at the same time on each stack. This nal
alignment process applies a residual ne tuning.
PREPARATION OF THE ROCK PHYSICS MODEL
The impedance of rocks (and hence the impedance
contrast of different rocks) is due to their elastic properties.
Critical geologic factors that control elastic properties can
be related to both depositional environment and burial
history (Avseth et al, 2003). Quantifying depth dependence
imposes a meaningful geologic trend which constrains
uncertainty and the expected range of seismic responses.
Therefore, a rock physics model was needed which
described the elastic behaviour of all the possible lithology
and uid combinations that were expected (as interpreted in
the wells as signicant for the delineation of hydrocarbons)
as a function of depth.
The starting point was a detailed petrophysical
interpretation following on from the quality control (QC) and
preparation, including environmental corrections, of the
available well log data. Synthetic shear logs were generated
for two of the wells using petrophysical trend analysis.
An end member rock physics model was built around a
petrophysical analysis. An end member is dened as the
cleanest example of a lithology present and is indirectly
based on mineralogy (Duncan et al, 2004). Once the elastic
properties of the end members are known, the elastic
behaviour of any rock composed of different proportions of
these end membersany Vsand with any uid saturation
can be determined. Figure 2 shows an example of the end
member interpretation. The straw coloured picks are the
end member sands and the blue picks are the Mungaroo
Shale. Before being used, each pick is upscaled and
averaged to become one point on the crossplots shown in
Figures 3 and 4.
Depth-dependent end member elastic lithology and uid
property trends, based on available log data, were derived
along with their uncertainties. These are shown in Figures 3
and 4. The solid tted line is the trend, while the dotted line
on either side captures two standard deviations of scatter.
Quantication of the inherent scatter in end member rock
properties was essential to understand the range of seismic
responses and associated inversion derivatives that were
observed.
It is worth comparing the TVD below mud line (TVDBML)
vs Vp trends for the Mungaroo shale and the sand. At about
a depth labelled Y metres the trends cross over. What this Figure 1. The comprehensive QI workow used in this study.
3
Drilling Success As A Result of Probabilistic Lithology and Fluid Prediction A Case Study in the Carnarvon Basin, WA
Figure 2. Example well from the study illustrating the end member picks. Sands are straw coloured and the shales are aqua blue.
Figure 3. Shale trends for the two dominant shales. Each picked interval from the end member interpretation is upscaled and then
plotted as a single point on these cross plots. y indicates the pivot point above which shales are softer than sands and below which
shales are harder than sands.
Mungaroo Fm
gas sands
Mungaroo shales
T
V
D
B
M
L

(
m
)
V
s

(
m
/
s
)
D
e
n
s
i
t
y

(
g
/
c
m
3
)
Muderong shale
Mungaroo shale
Vp (m/s)
Gas
Brine
4
M.G. Lamont, T.A. Thompson and C. Bevilacqua
means is that above Y metres the shale is softer than the
sand. Below Y metres the shale is harder than the sand
(Figs 3 and 4). This results in completely different AVO
behaviour above Y metres to below it. In fact these trends
describe constantly varying AVO behaviour. Therefore, it was
imperative to have a depth-based rock physics model.
The analysis determined that the necessary end members
in this case were:
||tho|og|es
Quartz rich sand;
The Muderong and Mungeroo shales; and,
Coal/Carbonaceous shale.
F|u|ds
Gas; and,
Brine.
To capture the expected range of rock properties and
hence the expected range of seismic responses, for all
candidate lithology-uid combinations over the depth range
of interest, stochastic forward modelling was performed.
Gassmann substitution was an integral part of the modelling.
Gas saturations used for modelling were 0% (brine case)
and 75+-5% (high saturation gas case). A low saturation
gas was not considered in the nal analysis due to the poor
discrimination from high saturation gas and a low geological
risk for the presence of low saturation gas. For this study
10,000 forward models were produced for every depth step
of 4 m. The parameters for each model were determined
by the Monte Carlo sampling of the multidimensional rock
physics model. The rock physics model is comprised of the
trends that are displayed in the four separate cross plots,
however they should not be sampled from one trend at a
time because they are, in fact, coupled; they need to be
thought of as one space and a single sample drawn from
this multidimensional space to get the correct distribution
of properties.
The varying AVO behaviour with depth is further illustrated
by the stochastic modelling results shown in Figure 5.
Each point on these plots represents one of the 10,000
models. It is clearly evident that over a 600 m depth interval
the clusters have shifted both absolutely and relative to
one another. The ellipses on these plots represent the
two standard deviation contour of the probability density
functions (PDFs). That is, these gures are showing both
the stochastic modelling results and the resulting PDFs.
A standard workow used in many QI projects may, if
necessary, begin by using Gassmann substitution to alter
the well data so that the sands represent the all brine case.
In a similar way, the logs may subsequently be altered to
represent a particular hydrocarbon case. These logs can
then be cross plotted and the different lithologies and
uids identied. The cross plot can be partitioned into
Figure 4. Sand trends for the area. As for the shales, each point represents an interpreted, upscaled end member interval.
T
V
D
B
M
L

(
m
)
V
s

(
m
/
s
)
D
e
n
s
i
t
y

(
g
/
c
m
3
)
P
o
r
o
s
i
t
y
Vp (m/s) Vp (m/s)
5
Drilling Success As A Result of Probabilistic Lithology and Fluid Prediction A Case Study in the Carnarvon Basin, WA
different lithology/uid combinations and these cut offs
used to produce lithology/uid prediction volumes. At times
gradations are added stepping away from the centre of the
partitions or away from theoretical trends and the resulting
volumes are called probability volumes, although they are
not technically probability volumes.
Consider this case study where the relative behaviour
of the rocks is varying rapidly with depth: this standard
workow will base the partitioning of this space and the
subsequent lithology and uid volumes on the location of
the sands and shales in the available wells. If the worker
is lucky the lithologies in the available wells are adequately
sampled through their depths of interest, and hence an
averaging of the modelling results over the depth interval
results. This situation is reasonably well illustrated in Figure
6. Here the modelling results represent a 600 m depth
interval. Consider the point at the Vp/Vs ratio of 1.65 with
a AI of 8,500. In the X+600 m crossplot in Figure 5, this
point clearly represents a gas sand. In fact the Bayesian
update will give this point a probability of being a gas sand
in the 90s. Now consider this same point in Figure 6. This
point falls well within the ellipses of gas, brine and shale.
Probability volumes produced from this modelling will
assign roughly similar probabilities to all three when in fact
it is a gas sand. Thus it is imperative to consider depth
dependency of the impedance and velocity responses.
A fundamental output of the modelling, based on the
trends and their uncertainties, was depth dependent
multivariate rock property PDFs for each lithology/uid
class. The stochastic modelling also provided the following
information:
AvO a|ded |n the d|sor|m|nat|on and pred|ot|on of gas
sands;
Ouant|fy|ng depth dependenoe was |mportant for
optimal uid and lithology discrimination; and,
The range of unoerta|nt|es a||owed d|sor|m|nat|on of
gas bearing sands.
Figure 7 shows the relationship of porosity with depth as
well as the changing resolution with depth. It is apparent
that at shallower depths uids can be easily discriminated
(brine sand vs gas sand) and lithologies less so (shale vs
brine sand).
SPARSE SPIKE SIMULTANEOUS INVERSION
A multi offset, Bayesian wavelet derivation program was
used to estimate the wavelet and its uncertainty (seismic
noise level). This was a strong test of the amplitude delity
of the seismic data. It featured a fully integrated Bayesian
approach to the coupled uncertainties in wavelet estimation,
a process which is critical in an inversion study (Gunning
and Glinsky, 2006).
The wavelet extraction provided maximum likelihood
estimates of the wavelet, as well as multiple realisations
from the posteriorethe nal derived distribution(s) after
Figure 5. Stochastic forward modelling results. The four major uid/
lithology combinations are represented here. Ellipses represent the
two standard deviation contours around the probability density
functions. Notice that the 600 m depth step has resulted in the
entire four clusters being grossly translocated as well as the shale
and brine clusters swapping relative positions.
Figure 6. Stochastic forward modelling from the depths of X to X +
600 m combined. Consider the point at the position AI=8,500 and
Vp/Vs=1.65. On this plot the probability of it being hydrocarbon,
brine or shale is roughly equal. In Figure 5 the same point has an
overwhelming probability of being hydrocarbon.
V
p
/
V
s
V
p
/
V
s
AI
V
p
/
V
s
AI
Gas Sand
Brine Sand
Shale
Shale + Coal
Shale + Sand (75.0%)
[brine 3000 + gas 3000 (75%)]
Shale + Sand (75.0%) [brine 3000]
Shale
Shale + coal (15.0%)
Gas Sand
Brine Sand
Shale
Shale + Coal
Shale + Sand (75.0%)
[brine 3000 + gas 3000 (75%)]
Shale + Sand (75.0%) [brine 3000]
Shale
Shale + coal (15.0%)
6
M.G. Lamont, T.A. Thompson and C. Bevilacqua
all the observed data and prior information has been
consideredhighlighting the uncertainly in the wavelet
scaling and extent. The nal wavelets are shown in Figure
8. The wavelets were extracted from all wells and all angles
simultaneously and the nal wavelet is approximately zero
phase.
A geostatistical methodology was used to build the low
frequency model. It utilised all available information (wells,
horizons, trends and velocity eld) producing a stable model
away from well control while providing an exact match
at the well locations. Through the initial and subsequent
updating of the project, an appreciation was gained as to
the importance of the low frequency model in obtaining
accurate reservoir predictions away from well control. In this
case it was a difcult procedure due to the lack of conformity
of the geology. That is, the volume doesnt easily break into
layers, and therefore modelling extensive horizons at target
level was not possible. Understanding the uncertainty
in the model was essential and is an ongoing part of the
investigation.
A sparse spike inversion algorithm, SPIKE, was used
for the volume inversion. It is an AVO inversion which in
essence inverts for intercept, gradient and curvature (Shuey,
1985) in order to determine P impedance, S impedance
and density. Although the inversion can take any number of
input stacks, three were used on this occasion. From these
fundamental outputs, Vp/Vs, LambdaRho and MuRho
were also calculated. LambdaRho and MuRho are rock
property attributes that characterise the incompressibility
and rigidity of a rock respectively. These quantities can be
obtained from the common equations for Vp and Vs, which
are a function of Lames parameters (lambda and mu), the
modulus of rigidity and density (rho).
Figure 9 shows the detuning power of the inversion as
well as the resolution limits. This wedge modelling is based
on the wavelets and parameters from this study and hence
is an accurate representation of the resolution limits.
Figure 10 shows the synthetic to seismic ties and the
relative P and S impedance match to the well data. A good
match was obtained through the zone of interest. Figure 11
shows the absolute inversion tie at one of the wells, with a
good match obtained.
DEPTH CONVERSION
The rock physics model was depth based and the
resulting probability volumes were needed in depth so as to
be easily used for the planning of new wells and as input to
Figure 7. A plot of AI vs Vp/Vs for varying depths and porosities. Ellipses are illustrative of ~1 standard deviation of scatter. Ellipse rotation
is not accounted for in this gure. Fluid discrimination is good at all depth levels, however does decrease with increasing depth. Fluid and
lithology discrimination is maximised using both P and S attributes. Lithology discrimination increases with depth.
V
p
/
V
s
AI
P
o
r
o
s
i
t
y
7
Drilling Success As A Result of Probabilistic Lithology and Fluid Prediction A Case Study in the Carnarvon Basin, WA
Figure 8. Final wavelets (multi-well extraction). The bunching of the wavelet realisations illicits condence in the result.
Figure 9. This gure shows the resolution limit of the QI ow. A wedge model, populated with the rock properties from the project,
was used as a basis for the generation of near, mid and far seismic stacks. The simultaneous inversion was run on these stacks
followed by the lithology and uid classication procedure. It is clear that the procedure correctly predicts sands down to around 10 m
thicknesswell below the seismic tuning thickness.
FINAL WAVELETS MID STACK WAVELET REALISATION
8
M.G. Lamont, T.A. Thompson and C. Bevilacqua
the reservoir model building phase. Therefore it was logical
to take the inverted seismic products to depth before the
analysis stages. The required depth was depth calibrated to
well control as opposed to seismic depth.
The methodology was to take the available time/depth
pairs and, using the initial velocity model, convert the
times to depth. The differences in the depths at each point
can then be used to calculate velocity updates at these
times. These velocity updates are built into a 3D volume
using geophysically/geologically constrained geostatistics.
The time depth points were honoured while a stable and
geologically consistent velocity eld away from the control
was produced. Finally, the update is added to the initial
velocity model. This velocity model was then used to
depth convert all of the inversion products, which was
subsequently tied to the wells.
LITHOLOGY AND FLUID PREDICTION
Figure 12 is a plot made before the drilling of any of the
discovery wells. It combines rock property volumes from
the inverted seismic data with the ellipses representing
the rock property modelling results. Firstly, an anomaly
was identied. The points from each of the rock property
volumes in the region of the anomaly are the samples in
the cross plot. Then three polygons are drawn. The rst
represents a potential sealing shale, the second represents
a potential gas sand and the third a potential down dip
brine sand. The points in each polygon are then highlighted
on the crossplot and color coded accordingly. Next, the
ellipses representing the PDFs at the mid-point depth of
the corresponding polygon are superimposed over the top
of the plot. They show a remarkable alignment between
the model results and cross plotted rock property volumes.
This engendered condence in the inversion and modelling
procedures. The next step was a Bayesian update.
Lithology and uid prediction was based on a Bayesian
supervised classication scheme called Nostradamus. It
brought together prior information, including stochastic
modelling, inverted data, and interpretation to produce
lithology and uid probability volumes through the zone of
interest.
At the core of the scheme are multivariate PDFs, one for
each depth level of interest. Each lithology/uid combination
was represented in the PDFs and hence resulted in an
associated probability volume being produced. In addition,
a most likely lithology/uid volume is produced.
Unlike Avseth (2003) and Anderson (2003) who use a
Bayesian approach based on relative interface properties
(AVO), this method utilises absolute rock properties to
dene the possibilities of an inversion resolvable layer. This
volume based prediction limits the possibilities to individual
layers rather than interfaces between two layers.
Figure 13 shows a slice through the resulting gas
probability volume with the subsequent discovery well Figure 11. Acoustic impedance tie at one of the wells.
Figure 10. Relative impedance ties at one of the discovery wells.
NEAR STACK MID STACK FAR STACK
RELATIVE
P-IMPEDANCE
RELATIVE
S-IMPEDANCE
T
i
m
e
(
s
)
9
Drilling Success As A Result of Probabilistic Lithology and Fluid Prediction A Case Study in the Carnarvon Basin, WA
locations and the original dry holes. It starts to reveal the
nature of the high gas probability sands as well as showing
how close two of the discovery wells are to the original dry
holes.
Figure 14 shows the rst discovery well. It was drilled
close to and down dip from an earlier dry hole. The new
well encountered gas sands as predicted. The location of
the dry hole comes out as a ~10% chance of gas in the
probability volume. Figure 15 shows the gas probability
volume in 3D in the same area as Figure 14. It reveals the
channelised geometry of the gas sands as highlighted by
the gas probability volume.
Figure 16 shows the second discovery well. It was drilled
at 67 degrees to intercept multiple high gas probability
bodies. Again the gas came in as predicted.
Figure 17 is the third and, at the time of writing, most recent
discovery. This well is also close to a previously drilled dry well.
The gas probability volume has now been used as the
basis for a reservoir model.
CONCLUSIONS
Three signicant gas discoveries have been made in an
exploration permit following a comprehensive QI study. Two
of these wells were close to dry holes. In addition, one of
the discovery wells was down dip from an earlier dry hole.
The workow described here assimilates geology,
geophysics, petrophysics, rock physics and uncertainty.
Simultaneous inversion products were compared with
multivariate rock property PDFs to make probabilistic,
volume-based uid and lithology predictions using a
Bayesian framework.
A depth dependent rock physics model is essential. The
sparse spike inversion was able to detune the seismic data
to a large extent.
Probability volumes were built using a Bayesian
classication scheme utilising rigorously derived, depth
dependent, PDFs. They were not derived from empirical or
arbitrary lines drawn on crossplots.
Although this workow involves considerably more work
than the standard approach, it is scientically justiable and
has delivered remarkable success.
ACKNOWLEDGEMENTS
The authors wish to thank Apache Energy, Kufpec and
DownUnder GeoSolutions for allowing the publication of
this study. The authors also wish to acknowledge Anne
Locke, Bjorn Muller, Adam ONeill and Alex Pauli for their
tireless efforts in doing the majority of the technical work
for this project.
Figure 13. Locations with high gas probability in a 200 ms
window below Top Mungaroo. The map shows previous dry
holes and the three new discoveries.
Discovery 1
Dry hole
Discovery 2
Discovery 3
Dry hole
Figure 12. Comparison of rock properties from the seismic inversion with the PDFs (from the appropriate depth) produced from the rock
physics model.
V
p
/
V
s
AI
AI
10
M.G. Lamont, T.A. Thompson and C. Bevilacqua
Figure 15. The gas probability volume shows the complex
gas sand geometries around the rst discovery well.
Original well failed to intersect a gas reservoir.
Figure 16. A cross section through the gas probability volume at the
second discovery well.
Figure 14. Gas probability data provided condence to drill downdip of a 12 m brine sand resulting in the rst discovery.
FIRST DISCOVERY WELL EXISTING WELL
SECOND DISCOVERY WELL
11
Drilling Success As A Result of Probabilistic Lithology and Fluid Prediction A Case Study in the Carnarvon Basin, WA
Figure 17. Third well, third discovery. Arbline showing discovery well position along with previous dry hole.
REFERENCES
ANDERSON, T., MARTIN, P. AND KOSTER K.,
2003Quantication of uid phase probability in prospect
evaluation using a Bayesian approach. ASEG 16th
Geophysical Conference and Exhibition, Adelaide.
AVSETH, P., FLESHE, H. AND WIJNGAARDEN, A.J.,
2003AVO classication of lithology and pore uids
constrained by rock physics trends. The Leading Edge,
(October) 100411.
DUNCAN, G., JAMIESON, M., MORRISON, A. AND
GLINSKY M.E., 2004Lithology and Fluid Prediction
in Lightly Explored Basins. ASEG 17th Geophysical
Conference and Exhibition, Sydney.
GUNNING, J. AND GLINSKY, M.E., 2006Wavelet
Extractor: A Bayesian well-tie and wavelet extraction
program. Computers and Geosciences (32) 61936.
SHUEY, R.T., 1985A simplication of the Zoeppritz
equations. Geophysics, 50 (4) 60914.
APPROXIMATE WELL LOCATION DRY WELL
Matthew Lamonts academic qualications
culminated in a Ph.D. in Geophysics from Curtin
University of Technology. His rst class honours
discertation was on seismic stratigraphy and his Ph.D.
topic was seismic multiples. He is an adjunct associate
professor at Curtin University. His working career
spans over 15 years including working in Woodsides
geophysics group performing depth migrations and
depth conversions. He worked for BHPB as the
seismic processing and imaging technical lead in
Houston. Back in Australia with BHPB he worked
in the quantitative interpretation area in addition to
processing and imaging. Matthew started DownUnder
GeoSolutions in 2003. He continued with depth
imaging, reection tomography, depth conversion and
quantitative interpretation research and application.
Matthew was the ASEG distinguished lecturer in
2005. The subject of the talk and paper was spectral
decomposition. Member: ASEG, SEG, EAGE, APPEA
and PESA.
Troy Thompson received a PhD from Curtin University
of Techonology. His topic was automated prestack
event picking. He received a rst class honours degree
at Curtin working on anisotropy. Troy was a founding
partner in DownUnder GeoSolutions. Today he is
a principal geophysicist, managing the QI services
division, and a director of DownUnder. He has written
papers on a variety of subjects including anisotropy,
prestack autopicking and quantitative interpretation.
Member: ASEG, SEG, EAGE and PESA.
Carlo Bevilacqua graduated in 1979 from the West
Australian Institute Of Technology with a Bachelor
of Applied Science (Physics). After two years of
seismic acquisition and processing with Geophysical
Services Inc in Perth he joined Oil Company of
Australia in Sydney. Carlo then joined LASMO Energy
as a staff geophysicist in Brisbane, and relocated to
London and Rome. In 1990 Carlo joined Enterprise
Oil where he worked in serveral positions including
Chief Geophysicist (Rome), Team Leader Southern
North Sea (London) and International Exploration
Coordinator (London). Carlo joined Apache in 1999 in
Perth where he works as a senior geophysical advisor.
Member: ASEG, SEG, EAGE, APPEA and PESA.
D
U
G
S
3
1
0
2
8
9