Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

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

net/publication/238424417

Integrated geological and geophysical analysis by hierarchical classification:


Combining seismic stratigraphic and AVO attributes

Article in Petroleum Geoscience · November 2008


DOI: 10.1144/1354-079308-800

CITATIONS READS

8 189

6 authors, including:

Alexis Carrillat Raul Ysaccis


Schlumberger Limited Schlumberger Limited
47 PUBLICATIONS 160 CITATIONS 15 PUBLICATIONS 104 CITATIONS

SEE PROFILE SEE PROFILE

Jonathan Hall
hyres geoscience solutions limited
18 PUBLICATIONS 94 CITATIONS

SEE PROFILE

All content following this page was uploaded by Jonathan Hall on 29 June 2022.

The user has requested enhancement of the downloaded file.


Integrated geological and geophysical analysis by hierarchical
classification: combining seismic stratigraphic and AVO attributes
Alexis Carrillat1,*, Tanwi Basu1, Raul Ysaccis1, Jonathan Hall1, Amiruddin Mansor2 and
Martin Brewer2
1
Schlumberger DCS, 18th floor, West Wing, Rohas Perkasa, No. 8 Jalan Perak, 50450 Kuala Lumpur, Malaysia
2
Exploration Sabah, Petronas Carigali, Level 10, Tower 2, Petronas Twin Towers, Kuala Lumpur City Center, 50088
Kuala Lumpur, Malaysia
*
Corresponding author (e-mail: acarrillat@slb.com)

ABSTRACT: The benefits of seismic attribute classification in subsurface studies


have been published widely. The approach is usually the same and, in most cases,
driven by a two-step procedure – an unsupervised classification and a supervised
scheme where training is used to redefine classes based on well-log data flagging a
specific fluid or lithology. In parallel to the multi-attribute analysis, interpreters have
also benefited from recent advances in computing power, enabling the generation of
multi-trace or texture attributes.
In these two workflows, the focus is either on the seismic texture facies for seismic
stratigraphic purposes, or on the reservoir facies for fluid and lithology mapping.
This paper presents a case study in which both texture facies and fluid prediction are
linked by performing a hierarchical classification scheme whereby a multi-attribute-
based volume, which captures seismic texture information, is combined with
amplitude versus offset (AVO) attributes to map fluid response into a single,
coherent reservoir facies volume. This methodology is then applied for exploration
data screening in offshore Borneo in the Greater Samarang sub-block (East Baram
Delta, offshore Sabah, Malaysia). In this case study, the geological analysis, seismic
geomorphology, seismic stratigraphy and combined fluid response from AVO
attributes facilitate the development of new play concepts in the highstand system
tracts and in the morphology generated by incised valleys in shoreface deposits.

KEYWORDS: seismic attributes, interpretation, seismic stratigraphy, seismic facies, geomorphology

INTRODUCTION geomorphology studies and/or supervised 3D seismic strati-


graphic mapping that fully honour the stratigraphic and geo-
Classification of seismic attributes has been employed in the metric aspect of the seismic data. As a result, these approaches
petroleum industry for about 15 years, using methods based tend to preserve the depositional and structural information
either on artificial neural networks (ANN) (McCormack 1991) captured in the seismic data and, thus, geometry, shape, rela-
or on statistics such as Bayesian classification (Sønneland et al. tionship and connectivity of the geobodies in 3D space. How-
1994). Although focused initially on classification of horizon- ever, there are several constraints related to inherent limitations
based attributes derived from the Hilbert transform of the of the method when performing classification of seismic at-
seismic trace, or the well-known waveform classification based tributes. Some of these are addressed in this case study.
on one or two picked horizons, recently there are a growing
number of approaches moving away from the limitations of + Seismic facies classification results produce discrete values,
surface-based attributes. Nowadays, true three-dimensional although sometimes an associated uncertainty, or distance to
(3D) attribute analysis preserves the real nature of spatial the cluster centre, is also generated as a by-product of the
information in complex geological settings (Carrillat et al. 2002). classification.
These new interpretation and reservoir characterization work- + When running seismic stratigraphic mapping from 3D
flows benefit from blending or combining seismic 3D volume stratigraphic or texture attributes, the lithological and fluid
attributes to produce ‘hybrid’, ‘multi-attribute’ seismic facies, or response aspect is preferably absent or not considered
‘meta-attributes’ volumes (e.g. de Rooij & Tingdahl 2002). fully.
In addition, recent 3D applications have been steered + Conversely, when multi-attribute analysis is performed to
towards seismic texture attributes, also called multi-trace map reservoir properties, for example when using inversion-
attributes, for 3D seismic facies characterization. These derived attributes, such as acoustic impedance (AI), Pois-
attributes work regardless of the dip of the reflectors, data son’s ratio and density, the seismic geomorphology and
frequency, the rotation, or transition and scale of the attributes seismic stratigraphic facies information are not incorporated
(Randen et al. 2000). These multi-trace attributes allow seismic directly.
Petroleum Geoscience, Vol. 14 2008, pp. 339–354 1354-0793/08/$15.00  2008 EAGE/Geological Society of London
DOI 10.1144/1354-079308-800
340 A. Carrillat et al.

Fig. 1. Location map of the Greater Samarang sub-block and study area.

This case study addresses these limitations by applying a zone (Fig. 1). The East Baram Delta structural style is charac-
hierarchical approach in two steps. First, an unsupervised terized by growth faults and rollover anticlines with crestal
classification is run to generate a multi-attribute volume that collapse structure. The Samarang Field, located to the west of
captures seismic stratigraphy and texture information as a the Morris fault, is one of the most illustrative examples of the
discrete set of seismic facies. This seismic facies volume is then aforementioned structural style and most of the hydrocarbon
combined in a systematic way with amplitude versus offset accumulations are located in the back limb of the rollover
(AVO) attributes to map fluid response by a hierarchical anticline and within the collapsed crestal part.
classification-estimation scheme based on a relationship ident- To the east of the Morris fault, the Sumandak area is located
ified between the AVO attributes and the water saturation from on the footwall side, which is characterized by a large anticlinal
log data. The final result is a single, coherent volume that structure, known as the Glayzer Anticline, located in the
incorporates both the seismic stratigraphic framework and southeastern part of the sub-block (Fig. 1). The series deposited
reservoir fluid information. above that anticline are characterized by a mainly eastward
structural dip direction and show a well-developed incised
valley system known as the Sumandak Complex (Stump 2004,
GEOLOGICAL FRAMEWORK Petronas Carigali unpublished report).
The study area is located in offshore Borneo in the Greater
Samarang sub-block in the East Baram Delta Province, off- STRATIGRAPHY
shore Sabah, Malaysia (Fig. 1). The Baram Delta Province owes
its name to the wave-dominated and tidal-influenced delta that In terms of stratigraphy, the framework is built from regional
is fed by the still-active Baram River. The Baram Delta is studies and the most recent biostratigraphic data analysis
characterized mainly by northwestwards outbuilding and pro- (Petronas Carigali, pers. comm.) based on calcareous nanofos-
grading successive phases (Koopman 1996). The Baram Delta sils, foraminifera (planktonic and benthic) and palynological
depocentre developed mostly during the Early Miocene and is data. The results of graphic correlation analyses of these data
likely due to a fault-controlled depression that formed at the are consistent, but challenge traditional chronostratigraphic
intersection of two major and deeper located crustal faults, interpretations in the area. To the east of the Morris fault or
namely the West Baram Line to the west and the Jerudong– Sumandak area they indicate a thick Lower Pliocene section
Morris fault to the east (Madon 1999). The Morris fault is a resting upon thin, earliest Upper Miocene that is unconform-
left-lateral fault zone that is the major N–S- to NE–SW- ably underlain by Middle Miocene rock.
trending structural lineament visible in the Greater Samarang To the west of the Morris fault or Samarang area, only rare
sub-block; it subdivides the block into the Sumandak–Glayzer and inconsistent occurrences of Miocene calcareous nanofossils
area to the east and Samarang area to the west of the main fault have been observed at shallow well depths; however, they are
Geophysical analysis by hierarchical classification 341

Table 1. Correlation between Shell Stages with Carigali chronostratigraphy and the depositional sequences from the present study

West ↔ East
With respect to the Morris fault
Shell Stages Stratigraphic markers Hiatuses Age Sequences
D & above SBH 10-40 SQ 2 (2a)
E to H SBH 50 Pliocene SQ 2b or 2ab
Stage IVF SQ 3
H to S SQ 4
SQ 5 –5A
Stage IVE SBH 60-70
SQ-6
S & down Miocene
Stage IVD SBH 75-90 SQ-7
Stage IVC SQ-7A

likely to be reworked. The youngest occurrence of the plank- Wong Hin Fatt et al. 2006) and corresponds to the boundary
tonic foraminifera suggests that this section is early Pliocene in between Shell Sabah Offshore Stage 4E and 4F (Levell 1987).
age. Across the Morris fault, biostratigraphic data are very
similar and indicate a thick Lower Pliocene section resting
unconformably upon Upper Miocene. DEPOSITIONAL ENVIRONMENT
The Greater Samarang sedimentation style can be subdivided
into three major cycles. The first cycle, starts from Sequence 8
WELL CORRELATION AND STRATIGRAPHIC through Sequence 5a, formed during the end of rifting as a part
FRAMEWORK of post-rift infill related to opening of the South China Sea in
the offshore Sabah area. This cycle corresponds mainly to
A consistent chronostratigraphic correlation across the Morris inner-neritic to fluvio-deltaic sediments (Thies et al. 2006). The
fault was achieved based on biostratigraphic information, show- Sequence 8 to 5a package is overlain by a second cycle starting
ing with confidence the subdivision of the depositional from Sequence 5 through Sequence 2 that mainly formed as
sequences into two broad geological time-equivalent series, shallow-marine to lower coastal plain deposits. Sequence 1, in
namely the Miocene and Pliocene intervals (Table 1). Marker turn, overlies this and has limited geographical extent.
correlation was done in an integrated effort guided by the The interpretation of the depositional environment within
seismic stratigraphic interpretation and refined from well-log each sedimentary package was done using all available data in
shape analysis. It should be noted that lithostratigraphic corre- an attempt to overcome the difficulty of distinguishing unique
lation was never attempted for the present geological investi- patterns from log shape of gamma ray alone and relating those
gation. Seismic horizon interpretation was guided by the to depositional environments. A top-down first-pass approach
principles of seismic stratigraphy defining unique depositional was followed from macro- (seismic) to a micro-scale (core
surfaces using standard reflection termination and geometrical data); then a bottom-up second pass was made to ensure
relationship criteria. Based on the seismic sequence strati- consistency at the different scales. Macro-scale investigation
graphic approach, 15 seismic horizon picks were carried over from the seismic data provided the maximum information on a
the Greater Samarang sub-block, following the same nomen- regional scale for depositional environment due to its extensive
clature of seismic sequences defined by Stump (2004, Petronas data and apparent ease of identifying the stratal termination
Carigali unpublished report). The sequence set starts from patterns. Seismic facies from 3D seismic attribute analysis using
Sequence 2 at the top and goes down to Sequence 8. Each of stratal geometry helped to identify the continuity of different
these sequences reflects a major depositional and/or tectonic depositional units and their sand to shale-prone nature. Log
episode in the history of offshore Sabah. shapes from the well data were examined next to refine the
In this chronostratigraphic reference framework, it appears depositional unit definition and precision of interpretation.
that seismic stratigraphic Sequences 7, 6 and 5a were deposited
during the Miocene age while Sequences 5, 4, 3, and 2 were INPUT FROM CORE DATA
deposited during the Pliocene. Further subdivisions within the
broad packages (for example Sequences 7, 7a, 4, 4a, 3, 3a, 2b, Cored intervals limited to Sequence 4 were available from four
2ab, 2a) were introduced based on log-shape analysis together wells only. The distribution of core data introduces a bias due
with finer seismic stratigraphic criteria. to unbalanced sampling both geographically and in stratigraphic
Shale packages were distinguished between and within the coverage. However, the cores were used as a calibration point
sequences (irrespective of thicknesses) when they showed to decipher the correct depositional environment of Sequence
lateral continuity across the field. Shale packages show quite 4. This calibration was then used to guide the interpretation of
non-uniform distribution depending on their location with the rest of the sequences above and below in their correct
respect to the Morris fault. For example, to the east of the depositional and sequence stratigraphic perspective and in
Morris fault quite thick shaly sequences are evident acting as terms of predictive succession of depositional sequences. A
very effective seals, especially in the Sumandak complex. To the typical facies association is a coarsening-upward sequence
west of the Morris fault the Samarang field exemplifies good starting from the offshore transitional facies through lower to
quality sealing shales. Table 1 highlights the Mio-Pliocene middle shoreface and finally the foreshore/beach at the top of
boundary as established between Sequences 5 and 5a, which the sequence. The lower shoreface to offshore transitional
corresponds to hiatus SBH60 (Dr Bill Krebs pers. comm.; environment is characterized by hummocky cross-stratified
342 A. Carrillat et al.

8 to Sequence 5a were deposited during the Miocene, while the


packages above Sequences 5 to 2 were deposited during the
Pliocene. Sequence 5a and those below – the Miocene package
– were formed as a post-rift infill package related to the rifting
and opening of the South China Sea. In the region there were
several periods of fault reactivation resulting in repetitive rift
infill sequences, which characterize the typical half-graben,
formed during active rifting. The upper portion of each of these
cycles is characteristically sand-prone inner-shelfal to fluvio-
marine sediments (Thies et al. 2006). It is difficult to identify
classic depositional systems tracts within these rift infill stages.
However, they consist primarily of aggradational highstand
system tracts.
In the Pliocene depositional history, Sequences 5 and 4a
represent the prograding delta-front deposits of a highstand
showing well-developed, low-angle clinoforms that downlap on
to the top of Sequence 5a. Sequence 4 has been interpreted
from the core as shoreface sands and is seismically character-
ized by parallel, continuous, high-frequency reflection patterns.
This sequence is deeply eroded by a major incision, which is
filled by Sequences 3a and 3 formed during the transgression
and highstand as transgressive back-fills and early highstand
deltas. The incision is very well defined in the northern half of
the field in and around the Sumandak Complex. The sediments
corresponding to this incision must have been transported to
deeper offshore. Sequence 3 is overlain by early and late
highstand progradation of the shoreline in the form of lower
coastal plain, and deltas in Sequences 2b and 2ab. Near the base
of Sequence 2b local channel incisions are seen possibly
corresponding to distributary channels of the lower coastal
plain. Sequence 2a might indicate the onset of the next
lowstand since minor incisions in the sedimentary package are
observed at the top of Sequence 2ab, especially towards the
north. Sequences 4a and 5 represent prograding delta fronts of
Fig. 2. Locations of key seismic sections on representative time slice the previous highstand.
and well locations.
Interpretation of seismic Inline 2 (Fig. 4)
sandstone and mud-rich heterolithic rocks, while the lower Inline 2 is situated to the west of the Morris fault, intersecting
shoreface to middle shoreface is characterized by hummocky two oppositely dipping growth faults: the Deep Growth fault
cross-stratified sandstone and sand-rich heterolithic rocks. This and the Samarang fault. Based on limited biostratigraphic data,
unit is overlain by cross-bedded upper shoreface overlain by the whole sedimentary package on this north–south strike
bioturbated sandstone. The topmost unit is characterized by section is classified into the same two chronostratigraphic
organic hash layers alternating with massive or swash-laminated packages. All the sequences are well developed on the southern
sandstones, which typically characterize beach sandstone or the side of the Deep Growth fault and the Samarang fault whereas,
foreshore sediments. Hence, the sedimentology of the core on the northern side, only Sequence 2 down to Sequence 5a are
lithofacies, their distribution and vertical association, sedimen- well developed. Sequences below 5a are hard to interpret due to
tary structures and special characteristics show the Sequence 4 the active mud diapirism and low quality of the seismic signal
cored intervals to be wave-dominated, laterally continuous below Sequence 5a. To the south of the Deep Growth fault,
shoreface sands. Sequence 7 and 7a show sand-prone packages, while Sequence
6 is shale prone.
DESCRIPTION FROM SEISMIC LINES In the Pliocene, Sequence 4a shows consistent progradation
at the base, representing a prograding delta front and Sequence
Several inlines, crosslines and random lines were interpreted in 5 is a shale-prone package of the prodelta of the same
detail to build the sequence stratigraphic framework. The highstand delta. Sequence 4 is again seismically characterized by
locations of some of these lines are shown in Figure 2. parallel, continuous, high-frequency reflection patterns of the
Examples of two key seismic inlines (Inline 1; Fig. 3 and Inline shoreface sands. To the south, Sequence 4 is eroded by a minor
2; Fig. 4), one key seismic crossline (Crossline 1; Fig. 5) and one incision, which dies out conformably to the north. Sequence 3a
random line (Random 1; Fig. 6) are detailed hereafter and then and 3 formed as transgressive back-fills and early highstand
integrated to build a consistent stratigraphic framework. deltas, which are overlain by early and late highstand prograda-
tion of the shoreline in the form of lower coastal plain deposits
Interpretation of seismic Inline 1 (Fig. 3) in Sequences 2b, 2ab and 2a.
Inline 1 is located to the east of the Morris fault in the
Sumandak area along a north–south axis. The whole sedimen- Interpretation of seismic Crossline 1 (Fig. 5)
tary package on this north–south strike section clearly reveals Crossline 1 is situated to the north of the Samarang field,
two main packages based on their age of deposition. Sequence intersecting the Morris, the Haselfoot, the Deep Growth and
Geophysical analysis by hierarchical classification 343

Fig. 3. North–south Inline 1 showing the interpretation of seismic sequences east of the Morris fault. LCP, lower coastal plain; TST,
transgressive system tract; HST, highstand system tract.

the Samarang faults from east to west. The whole sedimentary areas are characterized by low-angle clinoforms and slump scars
package on this west–east dip section is highly segmented by all related to fast sedimentation and sediment failure. Several
these faults together with the central collapse structure transgressive surfaces have been interpreted and marked as
in Samarang area. The presence of these faults and mud dashed lines (Fig. 6), which indicate the boundaries between the
diapirism illustrates the tectonic complexity of the area. Later- successive deltaic packages. The high amplitude continuous
ally, two structural segments can be distinguished to the east reflectors indicate possible sand-prone distributary mouth bars.
and west sides of the Morris fault. The Miocene package is well Sequence 4 is characterized by a deep incision forming the
developed and thicker on the eastern side of the Morris fault. promontories within the incised valley complex (Sumandak
The package is truncated against the uplifted mud diapir in the Complex). The incised complex is filled by shale-prone trans-
middle of the line. The Pliocene package starts with the gressive back-fills and some early highstand deltas in Sequences
shale-prone Sequence 5 followed by the deltaic package of 3a and 3. The maximum flooding surface corresponding to this
Sequence 4a. Top-sets and some clinoforms of the delta are still sequence has been interpreted, and separates the shale-prone
visible on the eastern side, while on the western side they are transgressive back-fills below from early highstand prograding
unclear. Sequence 4 shows parallel, continuous, high-frequency sand-prone delta deposits above, characterized by small-scale
seismic reflections on both sides of the Morris fault, while clinoforms. The deltaic deposits are overlain by an early and
Sequence 3 and above show considerable differences in the late highstand progradation of the shoreline in the form of a
reflection patterns, illustrating different depositional styles for lower coastal plain marking the uppermost part of Sequence 3.
these two areas. Sequences 3, 2b, 2ab, 2a all show parallel, Above Sequence 3, lower coastal plain deposits and deltas are
continuous, high-frequency seismic reflections on the western the dominant environments in Sequences 2b, 2ab and 2a. The
side, while the eastern side shows similar deposition patterns as lower coastal plain characteristically shows continuous reflec-
in the previous lines (Inlines 1 and 2). tors in Sequences 2ab and 2a.
From Sequence 3a to 2b, top-sets of the deltaic package
Interpretation of seismic Random line 1 (Fig. 6) corresponding to the distributary mouth bars are well defined
Random line 1 is situated on the eastern side of the Morris fault and they seem to be marching offshore (Fig. 6) within strati-
and passes through two key wells used for calibration, one of graphic time, while the shoreline is prograding northwest, i.e. an
them having core data available in Sequence 4. The line has offshore direction. The stepwise basinwards progradation of
been flattened at the top of Sequence 2b. The Pliocene package the mouth bars is highlighted by black arrows on Figure 6.
starts with Sequence 4a and 5, which show well-developed long Some of the most offshore deltaic package might actually be
clinoforms. Remnant lowstand erosions are identified in the shelf-edge deltas (Fig. 6). This overall regressive trend of the
top-sets of the lower coastal plain deltas whereas the delta front basin correlates well with the regional sequence stratigraphy.
344 A. Carrillat et al.

Fig. 4. North–south Inline 2 showing the interpretation of seismic sequences west of the Morris fault. Abbreviations as Figure 3, except SMRG
fault, Samarang fault.

FOCUS ON SEQUENCE 4 coastal plain. In these particular situations the added effect of
In the depositional environment analysis and among all the submarine and subaerial erosions is a possibility.
sequences, more focus was put on Sequence 4, because of its Projecting the estimated coastal plain base level of Se-
trap potential represented by promontories sculpted in shore- quence 4 basinwards (dotted line on Fig. 7), and flattening on
face sands and topped by shale-prone transgressive back-fills. that reference surface along a random line defined in the
This sequence has been penetrated by 17 wells located in both canyon dip direction allows the restoration of Sequence 4
Sumandak and Samarang provinces. The continuity of the configuration at the time of deposition and enables a finer
reflectors, together with the core information, points to a seismic stratigraphic analysis (Fig. 7). This reconstructed
shoreface deposition of laterally continuous sheet-like sands, cross-section shows the depositional pattern of the overlying
which have been interpreted as highstand deposits. Sequence 4 unit (Sequence 3a) on top of the Sequence 4 unconformity. It
is eroded by a major incision, which is a marine erosional event indicates that the early coastal onlaps of Sequence 3a are very
that created the promontories or buried hill topography in the close to the coastal break of Sequence 4, meaning that most
Sumandak Complex (Fig. 1). In terms of geomorphology, the of the Sequence 4 palaeotopography was under water. It also
eastern and southern parts of the Sumandak area are character- clearly shows the successive system tracts of Sequence 3a
ized by a smooth topography while the central and western transgressive system tract topped by a maximum flooding
parts show rough terrains and deep incisions. This geomorpho- surface (MFS) and marking the base of the progradational
logical expression, together with seismic interpretation, strati- clinoforms of the next highstand. This later one is topped by
graphic framework and well data integration, leads to the a sequence boundary marking the base of the next lowstand
inferred coastline. Calibration points for the lower coastal plain system tract characterized by toplaps located downdip off the
environment, coming from well data located on the Glayzer shelf break of the previous highstand.
Anticline area and to the north of the Sumandak Complex, After projecting the estimated coastal plain base level of
indicate the shallow-marine environment of the deposits imme- Sequence 4 over the entire Sumandak area, the depth difference
diately above the top of Sequence 4. The promontory and hill from this conceptual horizon down to the erosional top of
topography defines canyons that are orientated mainly in an Sequence 4 was calculated; the result is the palaeotopography
east–west direction (Fig. 7). In areas of very high erosion (e.g. shown in Figure 8. It would represent the erosional relief of
northern part of the Sumandak Complex) the canyons actually Sequence 4, which could be the potential accommodation space
traverse the whole marine shelf and continue towards the available for the prodeltaic shale and sand–shale progradation
Geophysical analysis by hierarchical classification 345

Fig. 5. East–west Crossline 1 showing the interpretation of seismic sequences. Abbreviations as Figures 3 and 4.

of Sequence 3a and 3. Palaeotopographic restoration was done charge history of the area since each province may have its
using a constant interval velocity of 2600 m s1 for the shale own hydrocarbon kitchen and its particular migration style.
package of Sequence 3 and 3a. It shows that a relief in the order One of the major unconformities and a key calibration
of up to 500 m was sculpted in Sequence 4 by the erosional horizon is Sequence 5a. The top of this sequence is close to the
unconformity (Figs 7 and 8). Assuming that most of the Miocene–Pliocene boundary as interpreted from biostrati-
erosional relief of Sequence 4 was under water as mentioned graphic data on key reference wells. Based on that calibration
before, the map actually gives an indication of the palaeo- point, the estimated timing of the different tectonic phases is
bathymetry during deposition of Sequence 3a and 3. outlined below.
Tectonic reconstruction, shown in Figures 9a–d, suggests
STRUCTURAL ANALYSIS AND EVOLUTION that the Greater Samarang structures are related to four
tectonic phases, described below.
Using the seismic horizon interpretation obtained from the
seismic sequence stratigraphic approach, a qualitative analysis
of the structural evolution was made by successive flattening of Phase 1
depositional surfaces. Phase 1 is an extensional phase that can be associated with the
Considering first the deeper level around 2500 ms two-way opening of the South China Sea and a rapid sediment accumu-
travel (TWT) time, an additional set of SW–NE faults, lation, occurring from Late Eocene to Late Miocene (Ngah
referred to as deeper growth faults, can be observed. These 1990; Thies et al. 2006). It has been defined as an initial rifting
deeper growth faults dip SE and mainly affect Sequence 6 phase that developed a series of syn-rift, half-graben sub-basins
and below (Fig. 9a). Among the key structures, the Glayzer controlled by growth faults that trend NE–SW, which is the
Anticline is distinguished on the eastern part of the sub-block sub-perpendicular direction to the rifting motion, and filled by
by its axial trend towards north–south and plunges to the a stratigraphic column that goes from Sequence 8 to Sequence
north. The Morris fault and the Haselfoot fault, located in the 6 (Fig. 9a). The NE–SW growth faults observed in the study
centre of the Greater Samarang sub-block, delimit the lower area were generated during this initial phase of basin formation.
part of the central graben. This graben represents a crestal Growth faulting might have triggered shale mobilization that
collapse on the top of the Glayzer Anticline (Fig. 9b). Further produced the shale ridges identified in the area. As can be seen
to the west on the same dip section (Fig. 9b), the Deeper on Figure 9a, the shale ridge constitutes the landward flank of
Growth fault and the Samarang fault drape the shale diapir the Samarang Province and the basinward flank of the
zone where a shale ridge is developed, splitting the area into Sumandak–Glayzer Province. The growth faulting that charac-
two provinces that could be differentiated as Samarang Prov- terizes this phase is, at the end of this period, affected by a
ince and Sumandak–Glayzer Province. This division by a minor inversion which is recognized by the erosion/
shale ridge is important and could impact the hydrocarbon peneplanation at the top of Sequence 6.
346 A. Carrillat et al.

Fig. 6. NW–SE Random line 1 showing the interpretation of seismic sequences in the Sumandak area flattened on top of Sequence 2b.
Abbreviations as Figures 3 and 4, except MFS, maximum flooding surface; DMB, distributary mouth bar; DF, delta front; TS, transgressive
surface; LS, lowstand.

Phase 2 Samarang Anticlines; both of them could be related to


Phase 2 is a compressional phase in Late Miocene–Pliocene, de-watering of the shale diapir. It is believed that during these
which was produced by the tectonic inversion of the previous collapses, the Morris fault was generated and the Haselfoot and
growth fault system, forming the Glayzer Anticline and part of Samarang faults were accentuated. These collapses could also
the Samarang Anticline. This phase controlled the deposition have favoured the later charging of the hydrocarbon-bearing
from Sequence 5a to Sequence 4/Sequence 3 (Fig. 9b). The sands. The deposition of Sequence 3 is contemporaneous with
onset of this compressional phase is well documented by the the collapse period and could mark the transition to the next
onlapping relationship of Sequence 5a on the top of Sequence extensional phase; therefore a regional subsidence could have
6, which in turn locally shows erosional truncation (Fig. 9b). characterized the area by this time. This possible tectonic
The compression led to uplifting, tilting and subsequent ero- relaxing could have been synchronous to the flooding of the
sion. This tectonic instability is expressed by an east–west Sumandak area that induced the muddy infill of its incised
slump scar and listric faults as recognized on the northern side valleys complex. The resulting shale-dominated section of
of the Glayzer structure. It slides to the north parallel to the Sequence 3 provides the top seals for the hydrocarbons trapped
plunging direction of the Glayzer Anticline. The slump affected in the Sequence 4 shoreface sandstones of the Sumandak
mainly Sequence 5a and the lower section of Sequence 5 and it buried hills.
has its décollement level on top of Sequence 6. As long as the
compression progressed, the area was uplifted and underwent
subsequent erosion, evidenced by the incised valleys docu- Phase 3
mented at the top of Sequence 4 in the Sumandak area and, to Phase 3 is a second extensional phase, which created a
a small extent, in the southwest part of the sub-block; i.e. east rejuvenation of the basin and involved the deposition from
and west of the Morris fault. Palaeotopographic restoration of Sequence 2b to Sequence 2 during Pliocene time (Fig. 9c). The
Sequence 4 indicates that up to 500 m of sediments in some major extension is associated with the activity of the Morris
parts of the Sumandak area have been eroded. This erosion led fault that produced accommodation space in the central part of
to a series of hills in Sumandak, which later were overlain by the the area through the whole phase; while the extensional activity
shale-dominated Sequence 3 to create the Sumandak buried of the Samarang fault in the western part is restricted mainly to
hills. Another evidence of this uplift/erosion is the local angular Sequences 2b and 2ab time. Additional synthetic and antithetic
unconformity recognized in the southwestern part of the area at faulting to the Samarang and Morris faults happened at this
the top of Sequences 4 and 3 (Fig. 9b). The end of this phase time; this episode would reactivate the previous crestal collapse
is characterized by the crestal collapses of the Glayzer and faults and would create a greater subdivision of the half-graben
Geophysical analysis by hierarchical classification 347

Fig. 7. Random section along a south Sumandak Canyon flattened on estimated coastal plain base level at top of Sequence 4 (flattened on dotted
line). Abbreviations as Figures 3 and 6, except SB, sequence boundary; LST, lowstand system tract.

by multiple block-faulted compartments. This tectonic over- taneous attributes, is that they are sensitive to the quality of
print helped to define the geometry of the Samarang field, the seismic pick and signal consistency. In a structurally
located behind the main antithetic fault on the back-limb of the complex setting, as in the Samarang field, correlation across
Samarang Anticline (Fig. 9c). small fault blocks, energy defocusing below fault planes, and
signal: noise ratio impact directly on the quality of the seismic
Phase 4 correlation and therefore the extraction of seismic attributes.
In addition, given the depositional environment dominated by
Phase 4 is a regional compressional phase, which took place relatively homogeneous shoreface deposits, the sensitivity to
from the Pleistocene to the Recent and produced the second variation in the rock quality is low from a geophysical
tectonic inversion (Fig. 9d). This phase affected the top of perspective. The subtle variations of rock property can be
Sequence 2 and it has controlled the subsequent deposition overwhelmed by structural imprint and imaging issues if a
until the present time. This late inversion phase is believed to careful analysis is not made.
have played a critical role in the trapping mechanism of In this study, 3D seismic attributes were considered mainly
hydrocarbons in the Samarang field by sealing some of the as they capture the information in 3D space and can be
faults. Based on the regional stress regime, this inversion is rendered with transparency tools, visualized and inspected as
likely to provide the same potential for sealing as other faults needed along inlines, crosslines, time slices, stratal slices or as
having similar strike directions. Other evidence supporting geobodies. The approach was first to run unsupervised classi-
inversion is the erosion/peneplanation of seismic reflectors at fication of two sets of key attributes aiming at different
seabed and the pop-up features visible at some locations along objectives in the classification results and let the natural
the Samarang fault. clustering of these attributes reveal the data characteristics. This
unsupervised classification step is done before refining the
SEISMIC ATTRIBUTE ANALYSIS classification in a second step using supervised neural network-
based classification and estimation.
In the Greater Samarang study, seismic attributes were ex-
tracted to provide information about potential reservoir qual-
ity and hydrocarbon indicators for screening purposes. The
value of traditional Hilbert transform attributes (Taner et al. Unsupervised classification of seismic attributes
1979) has been demonstrated in many case histories as well as The unsupervised seismic attribute classification, using K-mean
wave shape classification using neural networks, where the clustering (Coléou et al. 2003) had two key objectives. The first
seismic trace is decomposed into components of amplitude was to characterize seismic reflection patterns or seismic
and frequency information, such as in VRS spectral decompo- textures in the seismic volume, and the second was to look for
sition attributes (Sønneland 1996). The key limitation with all indications of possible fluid response or direct hydrocarbon
interpreted surface-based attributes, in particular instan- indicators (Fig. 10).
348 A. Carrillat et al.

Fig. 8. Three-dimensional reconstruction showing palaeotopography at the time of deposition of Sequence 3a. LCP, lower coastal plain.

The seismic texture facies cube In this study both types of seismic texture attributes are used
The parameters and signal characteristics of a group of seismic to characterize seismic reflection geometries into a seismic
reflectors characterize a seismic facies, which differs from that texture facies volume. The first pass for seismic texture analysis
of a neighbouring set of reflectors. The lateral and vertical was made using the following set of attributes: Chaos, Variance,
distribution and associations of the seismic facies reveal pat- Flatness, Gradient Magnitude, and a discrete representation of
terns that can be interpreted in 3D space and provide infor- spectral attributes VRS (Sønneland 1996). The set providing the
mation about the geology and/or the depositional environment. best natural clustering of the data and avoiding correlated
Ideally, three-dimensional seismic facies mapping should attributes included Chaos, Gradient Magnitude, Flatness and
honour the stratigraphic and structural framework of the the discrete combination of spectral attributes. The final
seismic data. To automate this procedure, a computation is texture-facies classification, using unsupervised classification
made of the dip and azimuth throughout the data volume. This with K-mean clustering, mapped four distinct reflection tex-
information is incorporated in the multi-trace attributes to tures and texture associations (Fig. 10).
guide the 3D seismic texture analysis through the data volume 1. The laterally continuous, parallel seismic reflection facies
(Randen et al. 2000). Analogous to the interpreter’s eye that showing good reflectivity and low heterogeneity of the trace
follows the local orientation of reflectors and stratigraphy (coded red in the texture-facies cube).
across a seismic section, the texture attribute generation is 2. The laterally continuous to sub-parallel facies with relatively
dip-steered. These attributes are referred to as 3D texture good reflectivity contrast and low seismic heterogeneity of
attributes, since they are able to describe the reflector-geometry the trace (coded green in the texture-facies cube).
in a small 3D neighbourhood (local orientation-guided or 3. The laterally semi-continuous to wavy seismic reflection
multi-trace attribute). The combination of the 3D texture facies showing poor reflectivity contrast (coded blue in the
attributes, containing orientation and/or continuity infor- texture-facies cube).
mation, enables mapping of the stratigraphic patterns from 4. The laterally discontinuous to chaotic seismic reflections
seismic data. facies showing poor reflectivity contrast (coded grey in the
There are essentially two main categories of 3D texture texture-facies cube).
attributes (Carrillat et al. 2002). The first includes texture
attributes that portray kinematic features of the seismic traces, This seismic reflection classification method was used as an
such as local orientation, signal discontinuity, or unconformity. independent way to evaluate the lithological associations away
The second category includes generic texture attributes that from the wells based on the seismic reflection patterns. The red
capture dynamic features in the seismic signal, such as spectral class (laterally continuous, parallel seismic reflections) and the
representations or amplitude behaviour from spatial derivatives. green class (laterally continuous, sub-parallel with relatively
Geophysical analysis by hierarchical classification 349

Fig. 9. Structural framework and tectonic evolution showing the four tectonic phases interpreted in the Greater Samarang sub-block.

good reflectivity contrast) appear to be sand-prone from well (discontinuous to chaotic seismic reflections) appear shale-
cross-validation analysis, while the blue class (laterally semi- prone (Fig. 11).
continuous to wavy seismic reflections) and the grey class
The seismic amplitude-dependent facies cube
Using the same unsupervised classification method (i.e. without
calibration from well data), the algorithm was also fed with
attributes that are amplitude dependent, such as RMS ampli-
tude, pseudo-Relative Acoustic Impedance (trace integration
method) and reflection strength. Since these three attributes
show some degree of linear correlation, a principal component
analysis was run to select only the first two components as
inputs to generate a seismic facies cube capturing the potential
Direct Hydrocarbon Indicator (DHI). The unsupervised classi-
fication with K-mean clustering produced an ‘amplitude-
dependent’ facies cube, which clearly highlights the bright spots
in the seismic data (Fig. 10). Based on the natural clustering of
the input attributes, this ‘amplitude-dependent’ facies cube was
classified into four classes; which have been interpreted and
associated with the following facies (Fig.12).
1. a most likely fluid-related class based on high amplitude
and high RAI response (colour coded red in the facies
classification volume; Fig.12);
2. a most likely shale-prone class (colour coded grey in
Figure 12);
3. a most likely wet sand-prone class (colour coded light
blue);
4. an intermediate class associated to the wet sand-prone class
(colour coded yellow).
The interpretation of seismic facies classes is based on
Fig. 10. Unsupervised classification workflow of seismic attributes. analysis and cross-checking at the well locations between the
350 A. Carrillat et al.

Fig. 11. Comparison of seismic sections at gas well case. Saturation Swt (left track) and gamma ray log (right track) displayed along the well.

seismic facies and well-log responses. The unsupervised anomaly for the wet case. This model suggests up to 10%
approach reveals a strong potential for identifying DHIs based amplitude increase between the zero offset and the far offset
on the match obtained in the Sumandak area facies mapping (35). Based on that information, amplitude scaling between the
and also for discriminating between the shale-prone and sand- near stack and the far stack using an RMS window of 30
prone intervals (Fig. 11). samples (60 ms) was applied.
This ‘amplitude-dependent’ facies cube was used as a basis The same process was used for the gas and oil cases. The
for rapid screening of the Greater Samarang sub-block to synthetic gather analysis at the top Sequence 4 gas sand shows
identify potential areas of interest for further more detailed a negative reflection coefficient, a decrease in relative amplitude
analysis. and an increase in absolute amplitude with offset (class III
AVO anomaly).
The oil case analysis revealed a weak, negative, reflection
AVO modelling and analysis coefficient with a weak negative amplitude gradient at the top
AVO forward modelling was carried out on three cases – gas of Sequence 4, corresponding to the known oil sand.
case, oil case and wet case – in order to understand and
calibrate the near and far offsets stacks (Fig. 12). Both Vp and
Vs data were available from Dipole Sonic Imager (DSI) logs in Supervised and hierarchical classification of seismic
the three key reference wells. attributes and AVO volumes
The wet case was analysed on a well located to the west of The supervised classification of seismic attributes is a natural
the Morris fault over the available logged interval (Sequences 3 follow up to the unsupervised classification, where the neural
to 6). The Vp and Vs data derived from the sonic log were used, network is guided by training data established from existing
together with density, to generate a synthetic gather from 0 to well-log responses. In this study, a hierarchical, supervised,
35 using the Zoeppritz (1919) equations. AVO analysis was classification scheme, using an artificial neural network with
then performed at different stratigraphic levels of interest. The back-propagation through multi-layered perceptrons (MLP),
results from AVO analysis were compared with the near and far was used to incorporate the AVO attributes. Hierarchical
offset stacks, the full stack amplitude data and the ‘amplitude- means that existing classes defined during the unsupervised
dependent’ facies cube on well-tied sections. The modelled classification step are refined by finer segmentation (hierarchy)
AVO response supports the near and far offset stacks near the based on a new set of seismic attributes and log data. This
borehole in showing no hydrocarbon-related amplitude hierarchical classification scheme enables the combination of
Geophysical analysis by hierarchical classification 351

Fig. 12. Hierarchical classification workflow.

post-stack attributes – here texture or amplitude-dependent cube by selection of the class of interest (sand-prone or
facies cubes – with offset stack attributes, such as near stack, far highest DHI potential).
stack and AVO response [Far(Far–Near_scaled)] in a sys-
tematic way. During the unsupervised classification step, the In the case of the unsupervised ‘amplitude-dependent’ facies
multi-attribute based texture facies volume and the ‘amplitude- generated from principal component analysis of RMS ampli-
dependent’ facies volume have defined ‘containers’ that are tude, pseudo-RAI and reflection strength, the hierarchy starts
then populated with new information derived from the seismic from the class defined as the most likely to be sand-prone and
AVO attributes (Fig. 12). hydrocarbon sensitive. This class is then populated selectively
The hierarchical classification involved three key steps by the neural network estimation model for saturation derived
(Fig. 12): from the best AVO attribute trained on selected well-logs (Fig.
12). Here, the cross-plotting analysis of saturation log versus
1. offset stacks and AVO attribute analysis against saturation AVO attributes revealed that the best correlation was obtained
logs at the well locations and selection of key discriminat- from the AVO response cube [Far(Far–Near_scaled)].
ing attributes; In order to quantify the prediction error, the estimation
2. correlation analysis and supervised neural network estima- workflow was run iteratively with different sets of input well
tion (for non-discrete values as opposed to classifica- data and verified with blind well tests not included in the neural
tion) from best AVO seismic attributes and saturation log network training (Table 2). The blind well test results derived
set; from the supervised estimation workflow provided confidence
3. hierarchical classification using either the unsupervised in the AVO response cube to presence or absence of
texture facies cube or unsupervised ‘amplitude-dependent’ hydrocarbon-bearing sands, actually gas sand (Fig. 11). Out of

Table 2. Blind well test of attributes-based HC prediction

Number of wells in training set Prediction success in Sumandak area (%) Prediction success outside Sumandak area (%)
16 90 62
7 70 60
3 70 60
352 A. Carrillat et al.

Fig. 13. Sequence 4a leads in buried hill play and multi-attributes section display. The white ellipses highlight two potential prospects from
combined AVO signature results, texture facies, stratigraphic and structural setting.

the 17 wells used for the validation analysis, the success rate for 14 illustrates that fact where a wet well resulted from a relatively
prediction of hydrocarbon-bearing sands is 0.63. If one consid- promising area when AVO attributes were considered alone.
ers only the sequences shallower than Sequence 5a, the success On the other hand, when the results from both facies and
rate for prediction of hydrocarbon-bearing sand goes up to AVO, based on the hierarchical classification scheme are
0.72. The poorer, deeper success rate is due to a fairly weak considered, the drilled wet sands appear without ambiguity and
AVO response in the AVO attributes below 2000 ms TWT. the map clearly highlights where other potential plays are
Due to the semi-quantitative nature of the input AVO located (Fig. 14). Updip from the wet well, the anomaly brought
attributes, the method proved to be inconclusive when it comes up by the combined facies and AVO attribute map corresponds
to distinguishing between low and high gas sand saturation. to a drilled and proven oil and gas accumulation (Fig. 14).

RESULTS AND DATA SCREENING CONCLUSIONS


The result from the hierarchical classification of the ‘amplitude- This case study demonstrates the value of a combined method
dependent’ facies cube combined with the AVO response based on seismic geomorphology, seismic stratigraphy, and
attribute was used for guiding the data screening and final fluid response from AVO data calibrated with existing well data
interpretation of the sequence stratigraphic framework and to improve the prospect generation process in a consistent and
depositional environment. The AVO fluid response revealed by systematic manner. This integrated workflow facilitated the
the hierarchical classification was used in parallel with the development of new play concepts in the highstand system
original amplitude and the texture facies volume to find the best tracts and in shoreface deposits incised during the lowstands to
prospects in terms of AVO signature and optimal setting to form buried hills, and improved the risk evaluation of those
ensure geologically consistent and reliable sand and seal associ- plays.
ation, as interpreted from the sequence stratigraphic frame- Geological analysis provided detailed depositional environment
work. Examples of comparative analysis of each seismic facies interpretation distributed within a consistent stratigraphic and
volume with original amplitude and far offset stack cubes for structural framework driven by seismic sequence stratigraphy.
delineation of new prospective areas are illustrated in Figures The geophysical analysis was driven by the interpretation of
11 and 13. the depositional sequences. Sequences 8 to 5a are Miocene
In both the Samarang and Sumandak–Glayzer areas, top seal deposits characterized mostly by post-rift infill: shelfal to
capacity is a strong control on hydrocarbon distribution. Figure fluvio-marine deposits showing aggradational stacked
Geophysical analysis by hierarchical classification 353

Fig. 14. Sequence 4a top map showing leads in buried hill play from combined facies and AVO (top) and far stack envelope (bottom).

highstands. Sequence 5a is dominated by highstand prograding Analysis and mapping of the multi-attribute volumes,
delta deposits. Sequences 5 to 2 were deposited mostly during together with amplitude data and incorporation of the geologi-
the Pliocene to possibly Pleistocene Sequence 5 shows high- cal model within the sequence stratigraphic framework, enabled
stand delta and delta front, while Sequences 4a and 4 show the identification of leads in a consistent manner, reconciling
shoreface deposits and deltaic progradation with lowstand the geological framework from seismic texture attributes and
incisions. Above, Sequences 3a and 3 represent the next AVO attributes for hydrocarbon mapping.
transgressive fill, deltas and delta front to lower coastal plain
The authors wish to thank Petronas Carigali Sdn Bhd for permission
associated with shelf-edge deltas that fill in the massive incised to publish and show the data and the Petronas Carigali Exploration
valley system sculpted in Sequence 4. Sequences 2b to Sequence Sabah team for support.
2 are dominated mostly by lower coastal plain deposits and
delta fronts to shelf-edge deltas of late highstand.
The structural interpretation suggests that the depositional REFERENCES
sequences can be grouped into four main packages connected Carrillat, A., Randen, T., Sønneland, L. & Elvebakk, G. 2002. Automated
to the structural evolution of the Greater Samarang area. mapping of carbonate mounds using 3D seismic texture attributes. Society of
Starting from the oldest package, in the Miocene, an extension Exploration Geophysicists, Expanded Abstracts, 21, 552–555.
was observed along SSE-dipping faults followed by minor Coléou, T., Poupon, M. & Azbel, K. 2003. Unsupervised seismic facies
inversion, compression in the Late Miocene–Pliocene, with classification: A review and comparison of techniques and implementation.
The Leading Edge, 22, 942–953.
mud diapirism and crestal collapses, followed again by a de Rooij, M. & Tingdahl, K. 2002. Meta-attributes – the key to multivolume,
Plio-Pleistocene extension, this time along WNW-dipping multi-attribute interpretation. The Leading Edge, 21, 1050–1053.
faults and, finally, a recent inversion. Koopman, A. 1996. Structure. In: Sandal, D.T. (ed.) The Geology and Hydrocar-
Seismic attribute analysis was performed on post-stack data bon Resources of Negara Brunei Darusalam 2nd edn. Syabas, Bandar Seri
to deliver 3D seismic facies and texture (stratal geometries) Begawan, 61–78.
characterization via unsupervised classification. The unsuper- Levell, B.K. 1987. The nature and significance of regional unconformity in
the hydrocarbon-bearing Neogene sequence offshore West Sabah. Bulletin
vised multi-attribute classification was then refined by a hier- of the Geological Society of Malaysia, 21, 55–90.
archical supervised approach to incorporate AVO data (near Madon, M.B.H. 1999. Petronas Chapter 5 – Basin types, tectonostratigraphic
and far offset stacks) and to calibrate the multi-attribute provinces and structural styles. In: The Petroleum Geology and Resources of
response with well-log data. Malaysia. Petronas, Kuala Lumpur, 77–111.
354 A. Carrillat et al.

McCormack, M.D. 1991. Neural computing in geophysics. The Leading Edge, Taner, M.T., Koehler, F. & Sheriff, R.E. 1979. Complex seismic trace analysis.
10, 11–15. Geophysics, 44, 1041–1063.
Ngah, K. 1990. Structural framework of Southeastern Malay basin. PhD thesis, Thies, K., Ahmad, M., Mohamad, H., Bischke, R., Boyer, J. & Tearpock, D.
Imperial College, University of London, UK. 2006. Structural and stratigraphic development of extensional basins: a case
Randen, T., Monsen, E., Signer, C., Abrahamsen, A., Hansen, J.O., Saeter, T. study offshore deepwater Sarawak and Northwest Sabah, Malaysia. Search
& Schlaf, J. 2000. Three-dimensional texture attributes for seismic data and Discovery, Article 10103 (AAPG Annual Convention Calgary, June
analysis. Society of Exploration Geophysicists, Expanded Abstracts, 19, 668–671. 2005).
Sønneland, L. 1996. A method of processing seismic data signals (VRS/Orthogonal Wong Hin Fatt, R., Boyce, B., Krebs, W et al. 2006. Chronostratigraphic
polynomial spectral decomposition). PCT Patent Application No. WO/9837437. chart of sedimentary basins of Malaysia. In: Petroleum Geology Conference and
Sønneland, L., Tennebø, P., Gehrmann, T. & Yrke, O. 1994. 3D model-based Exhibition, Abstracts, Kuala Lumpur, 14, 82–84.
Bayesian classification. Society of Exploration Geophysicists, Expanded Abstracts, Zoeppritz, K. 1919. Erdbebenwellen VIIIB. On the reflection and propaga-
13, 510–511. tion of seismic waves. Göttinger Nachrichten, I, 66–84.

Received 14 January 2008; revised typescript accepted 12 May 2008.

View publication stats

You might also like