SPE 57259
Mapping Reservoir Saturation with Seismic Resolution to Improve Reservoir Model
S.B. Reymond, L. Soenneland, A. Strudley, Schlumberger (Geco-Prakla)
Copyright 1999, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the 1999 SPE Asia Pacific Improved Oil Recovery
Conference to be held in Kuala Lumpur, Malaysia, 25–26 October 1999.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
Abstract
Improved oil recovery on mature fields relies on a reservoir
model accurate enough to identify by-passed oil and to
monitor fluid responses to production. Integration of seismic
3D volumes with stochastic realisations and flow simulations
in a single interpretation environment (Shared Earth Model)
allows asset teams to perform rapid seamless iterations leading
to a perfected representation of the reservoir (Fig. 1). Facies,
fault and fluid mapping from 3D/4D/4C seismic data has
moved one large step ahead with the advent of grid-based and
3D volume seismic classification (Fig. 2). Resulting
deterministic spatial probability estimates of reservoir
parameters are direct inputs to the reservoir model to constrain
large uncertainties present in geostatistical distribution of
reservoir parameters.
Introduction
The concept of reservoir management that can be defined as
the judicious use of various means available to maximise
benefits or economic recovery from a reservoir has become
common practice through large integrated databases. The
reservoir model of a mature field is the depository of all
exploration and production information. Its aptitude to
reproduce and predict the real is one of the main factor
influencing economical development strategies.
As the trend in the industry moves towards smaller
discoveries in combined stratigraphic and structural traps in
deeper and deeper waters, the costs of exploration and
production increases dramatically. This puts a serious
constraint on the number of wells that may be drilled to
estimate and produce the reserves in place.
Although relatively limited by the number of cells a
reservoir model can comprehend, its main issue remains the
spatial resolution of reservoir parameters. The viability of their
stochastic or deterministic distribution is directly related to the
amount and accuracy of inter-well data available. Reservoir
models are still commonly built using sparse well data and
gross structural information resulting from too rapid
interpretation of 2D and 3D seismic. This may lead to largely
under-evaluated uncertainty on the distribution of reservoir
parameters. The objective of this presentation is to
demonstrate the amount of information that can be extracted
from land and marine 3D seismic volumes and how it may be
used to improve the construction of an accurate reservoir
model.
In particular, the quality of 3D seismic data has improved
through better quantification of uncertainties at all stages from
acquisition (source signature and receivers positioning) to
processing (quantitative quality assurance of all processing
steps). At the inversion stage, the uncertainty estimate is
inherent to the geostatistical methods applied for the
determination of reservoir parameters.
The key information provided by 3D seismic, is to map the
spatial distribution of continuous (facies) and discontinuous
(faults) geological events. 3D seismic resolution is limited by
numerous factors of which frequency attenuation and velocity
increase with depth are probably the most dominant.
Feasibility studies based on seismic modeling are therefore
applied to estimate the expected resolution at the target level.
At a given reservoir location, the acoustic response is
influenced by a combination of the structural, stratigraphic and
pore fluid effects. Detailed seismic modeling is required to
establish their relative contribution.
Additional seismic data types may contribute to resolve
subtle acoustic differences and to distinguish between matrix
and fluid effects. Amplitude changes versus offset (AVO)
analysis resulting from specific processing of conventional 3D
seismic data has proven to be effective in separating gas from
heavier fluids (oil and water). More prone to resolve fluids are
full elastic wavefields recorded with four component sensors
(2D/3D 4C seismic). In addition to the standard geophone
(land) or hydrophone (marine) used to record the
compressionnal waves, the 4C recording systems includes 3
orthogonal geophones to record directional shear waves. Such
data is commonly calibrated by down-hole seismic acquisition
(VSP). In summary, this article illustrates how the mapping of
intra-well reservoir parameters can be improved by integrating
2 S.B. REYMOND, A. STRUDLEY AND L. SOENNELAND, SCHLUMBERGER GECO-PRAKLA
all seismic data types into a single inverted result using
classification.
Objectives and methods
When ignoring the surface transformation factor (Boi), the
formulation of the volume of hydrocarbon (HC) present in a
reservoir is expressed as: Vhc= ∑I (1- Sw(I)) ⋅ φ(I) ⋅ h ⋅A(I)
where (I) is a reservoir cell unit, Sw the water saturation, φ
the porosity, h the reservoir thickness and A the area of the
reservoir filled with hydrocarbon. The value (Vhc) is easily
computed from the reservoir model with a degree of
uncertainty on each parameter increasing with the distance to
the wells. Our purpose is to demonstrate how each of these
variables can be extracted from a 3D volume of seismic data
adequately calibrated by AVO analysis, 4C data and well
information and to show how significant the reduction on the
uncertainty can be.
The reservoir area (A) filled with hydrocarbon is
computed from multi-attributes seismic facies classification. A
binary class map indicates area of the reservoir filled by a
given fluid (light HC, heavy HC and water).
The saturation (Sw) is determined by seismic saturation
indicators (geostatistical distance maps of the probability of
hydrocarbon presence). The seismic attributes are calibrated at
the well ties for reservoir parameters (porosity, water
saturation, pressure, fluid types, etc…). The inversion to
saturation can be either based on simple regression or based on
a cell by cell Neural Network inversion.
The reservoir thickness (h) is obtained by constraining the
hydrocarbon seismic hypothesis map by the depth closure of
the top reservoir surface. A distinct fluid contact is assigned to
all non-connected reservoir compartments.
The porosity (φ) is obtained by inversion to acoustic
impedance using the best feasible approximation
(Combettes1). The input trend model of acoustic impedance is
based on simple or ordinary kriging of the well data.
Successively replacing each variables in the reservoir
model by their seismic equivalent will result in a 3D shared
earth model combining well and seismic data resolution scales,
validated by quantified uncertainty.
Seismic classification (facies and fluid mapping)
Seismic classification results may provide the area of the
reservoir filled with hydrocarbons (A), possible indications of
the water saturation distribution (Sw), fault mapping and fault
sealing analysis.
The input to seismic classification algorithms is typically
seismic attribute grids or attribute cubes. A set of relevant and
independent instantaneous and volume attributes are chosen to
solve the problem at hand. In particular, we have used a new
set of volume attributes computed on the polynomial
reconstruction of the observed seismic trace in a given
reservoir interval. These attributes are orthonormal and might
extract all the information content of the seismic trace.
Similarly, a set of 3D attribute cubes from migrated seismic
SPE 57259
data can be generated and used as input to the classification
process.
The complete set of seismic attributes is then displayed and
analysed in multi-dimensional attribute space to examine all
eventual data redundancy. Once the optimum set of attributes
is selected, unsupervised classification algorithms
(geostatistical or Neural Network) are used to define the
natural clustering present within the data (Fig. 2). The
resulting class map is then used together with well data to
interactively select reference locations that will define the N
number of classes that fully represent the reservoir lithology
and fluids distribution.
Supervised classification (using reference data) have two
type of outputs. The first is a binary class map of the user
selected N classes of seismic facies. Each class is calibrated by
the well measurements. The calibration step is obtained along
the well trajectories by computing the exact same set of
seismic attributes on the synthetic trace as on the observed
seismic wiggle trace. The area (A) of the class corresponding
to a given fluid type (HC) is a direct input into the
computation of the volume of hydrocarbon in place. The
second output of supervised classification is the probabilistic
distribution of reservoir parameters within each class obtained
(SHC=1-Sw for the HC class).
Class map validation can be obtained by ignoring some
wells in the classification process to compare measured
parameters with predicted classification results.
Seismic modeling might be applied in cases of presumed
non-unique inversion caused by several reservoir parameters
affecting the reservoir acoustic response (pressure,
temperature mostly in cases of steam injection, lateral changes
in porosity and fluids).
Seismic classification on AVO volumes and 4 components
data also contributes to a better assessment and calibration of
lithofacies and fluid types as mapped by 3D seismic
classification. AVO 3D analysis provides indirect
measurements of the shear term of the acoustic response.
Including AVO attributes in the classification process
contributes to discriminate between fluid types having large
density and velocity differences.
4 components seismic data provides direct measurements
of the Poisson’s ratio leading to quantitative calibration of the
AVO 3D analysis. It also contributes in increasing the vertical
and lateral seismic resolution of the reservoir as the measured
shear wave commonly travel at least twice slower as the
compressionnal waves.
Fault classification and mapping
Faults manifest themselves as spatial discontinuity in the
seismic 3D response. Such area of discontinuity can be
enhanced by processing and structural attributes. Structural
attribute grids and cubes are then input to fault classification.
The procedure is similar to facies and fluid classification
using, in this case, structural well log data (Dip meters, FMI)
to calibrate the probability maps of faults.
The established fault network can be further analysed to
assess the degree of sealing of each fault plane. The sealing
SPE 57259 MAPPING RESERVOIR SATURATION WITH SEISMIC RESOLUTION TO IMPROVE RESERVOIR MODEL
analysis computes a cost function by comparing the classified
seismic lithofacies and water saturation values across each
fault trace or plane. The result is a probability maps of the
distribution of sealing and non-sealing faults. All probability
maps resulting from seismic classification are in a format
ready for input into the reservoir model.
Reservoir volume analysis
Reservoir volume analysis validates the classification
results against the depth closure map. Regions of the
reservoirs below an estimated fluid contact (OWC) are given a
lower probability of hydrocarbon presence. In addition, the
process estimates the thickness (h) and volume of each
reservoir compartment filled with hydrocarbons. The
procedure can be constrained by a system of sealing and non-
sealing faults to produce a connectivity analysis that will
assign a specific probability of hydrocarbon presence to each
reservoir compartments.
Inversion for acoustic impedance
Finally, extraction of the distribution of porosity (φ) is
needed. With the assumptions that no lateral pressure changes
are observed within one fluid class, the remaining differences
in acoustic impedance are dominated by porosity effects.
Seismic inversion for acoustic impedance based on best
feasible approximation algorithm (BFA) gives the user more
flexibility to honour all constraints. Input data are an initial
model of acoustic impedance, acoustic impedance logs
computed from well data, a cube of interval velocity and a
seismic cube. The output will be the volume of acoustic
impedance that best fits all constraints (Fig. 3). The initial
model of acoustic impedance for the inversion can be
generated by interpolating well log data onto a trend model
(simple kriging) or directly from the well logs (ordinary
kriging). A set of stratigraphic and structural bedding
constraints can be honoured.
Integrated databases allow seamless access to all stored
data types. A multi-dimensional cross-plot tool present in the
suite of applications is used to establish regression functions
between measured porosity and acoustic impedance values
obtained from seismic inversion.
Input to the reservoir model
The purpose of this presentation is to demonstrate how
reservoir parameters can be extracted from seismic data and
included into the reservoir model. The final results from
classification are probability grids and cubes that need to be
adapted to the scale of the reservoir model. A series of
applications have been developed to upscale the seismic
resolution to the cell size of the reservoir model.
In addition to data integration, iteration between tools
becomes of key importance. An integrated environment allows
swift iterations between the observed and the modeled to
converge towards an optimum reservoir parameters
representation.
3
Case studies and results
Two case studies (including time-lapse and 2D/4C seismic
data) are used to illustrate the successive steps leading to
calibrated inversion of seismic data for 3D reservoir saturation
mapping.
Prediction for saturation have been confirmed by 2 new
horizontal wells reaching 4 reservoir compartments and 1 new
exploration well.
Conclusions
1. The seismic signal needs to be optimally captured by an
appropriate set of seismic attributes to map spatial distribution
of seismic facies with seismic classification algorithms.
Comparative studies show that polynomial reconstruction of
the seismic trace better captures subtle lateral changes in
seismic facies.
2. Seismic facies class and probability maps inversion to
reservoir parameters is done by well data calibration of given
reservoir parameters (i.e. saturation, porosity).
3. Additional inter-well seismic data calibration is obtained
by including complementary seismic data in the classification
scheme (AVO and 2D/3D 4C).
4. Fault classification is based on spatial discontinuity
enhancements. It defines the reservoir compartmentalisation
with indications of faults sealing capacity.
4. Probability of faults presence and seismic saturation
indicators (with uncertainty estimates) are direct input to an
improved reservoir model.
Acknowledgment
The authors thank Lars M. Pedersen and Martin Landroe
(Statoil) for their collaboration on the 4D Gullfaks project as
well as Statoil and partners for releasing the data. This work
was partially financed by the European Union’s Thermie
Research program (OG 117/94 UK).
References
1. Combettes, P. "Signal recovery by Best Feasible
Approximation". IEEE Transactions on Image Processing,
Vol.2, No. 2, Apr. 1993.
4 S.B. REYMOND, A. STRUDLEY AND L. SOENNELAND, SCHLUMBERGER GECO-PRAKLA SPE 57259

Observed Modeled
a1
Initial model
2D/3D 4C and 4D 2D/3D seismic Well log
seismic data modeling kriging

Shared
Earth Reservoir Feasible Seismic
Well log data a1
Model models set
a2
Interval
Flow velocity
Petrophysics simulations
Classification
system

Fig.1- Shared Earth Model data integration scheme with reservoir a2

parameters probability estimate output through the classification
system. Repeated iterations between the observed and the
modeled leads to optimised Shared Earth Model. Fig.3- General inversion scheme with Best Feasible
Approximation algorithm. 4 constraints define a feasible set
towards which each value of acoustic impedance contrast present
in the initial model will eventually converge.
Surface seismic Well log data

Seismic Calibration
attributes

Unsupervised Supervised
classification classification

Reservoir fault, facies and fluids

class and probability maps

Fig.2- Unsupervised and supervised classification system.

Surface seismic includes 2D/3D 4D and 4C seismic data. Seismic
attributes are grids and 3D cubes. Unsupervised classification
provides natural clustering analysis. Calibration of seismic
attributes with reservoir parameters is done at the well ties.
Supervised classification results produces quantified spatial
distribution of reservoir parameters.