Chapter 2

Modeling Geological Continuity

2.1 Introduction
2.1.1 Why Do We Need To Model Geological Continuity? Geological continuity is pos-
sibly the most important driver for subsurface flow, so it should be represented accurately
in a reservoir flow-simulation model at a level that is relevant for the particular appraisal or
development problem in question. Most often, a geological interpretation is required to
quantify geological continuity. Geological interpretations, however, are often qualitative or
descriptive in nature. For such interpretations to be useful, they must be made quantitative.
Geological information that cannot be made quantitative cannot be used in 3D reservoir
modeling and, hence, will not have an impact on the decision-making process.
This chapter presents various alternative geological-continuity models for modeling the
continuity of petrophysical properties and facies distributions. As discussed in Chapter 1, a
geological-continuity model is used to relate the data to each other as well as to unknown
properties at unsampled locations. The geological-continuity models are then used in Chap-
ter 3 to construct geocellular models reflecting the modeled geological continuity and being
constrained to well and seismic data at the same time.
Capturing geological complexity into simple models seems an impossible task, yet not
all geological detail matters for flow or needs to be explicitly modeled.A geostatistical tool-
box should, therefore, provide a spectrum of models, from simple parametric models (re-
quiring only knowledge about the rough characteristics of the subsurface continuity) to
more complex models (requiring prior detailed geological interpretation).
Building high-resolution geocellular models from wells and seismic is usually achieved
through a hierarchical approach, as shown in Fig. 1.2. In terms of property modeling, one
usually follows two steps:

.. Generate the geometry of the most important facies.

Populate each facies with petrophysical properties.

The main reasons for creating this hierarchy are as follows:

. Petrophysical properties often follow trends governed by the geometry of facies or

architectural elements such as channels. Accurately modeling such trends is impor-
tant for prediction of fluid flow and saturation patterns in the reservoir.
 18 Petroleum Geostatistics Modeling Geological Continuity 19

. Petrophysical properties within each facies are more homogeneous than the reservoir
as a whole. The statistical properties of petrophysical properties within each facies
curvilinear shapesare often hard to model with cell-based techniques. The object approach
consists of dropping directly onto the reservoir grid a set of objects representing different
are fairly stationary and can be modeled by well-established geostatistical methods. geological facies, each with their own geometry and relative spatial distribution. These ob-
jects are then moved around and locally transformed to match the local data (wells and seis-
In this chapter, two types of continuity models for describing facies geometry are dis- mic); see Chapter 3 for additional discussion.
cussed: object models and training-image models. For describing properties within each fa- In theory, there is no limitation to the type of shapes that can be modeled, but most cur-
cies or within a fairly homogeneous layer, the variogram model is used. rent applications focus on fluvial and submarine channel objects. The first task in an object-
based method is to establish the various types of objects (sinuous channel, elliptic shale,
2.2 Describing Facies cubic crevasse) and their mutual spatial relationship: erosion of one object by another, em-
2.2.1 Object Models. Also termed "Boolean models," these models import geologically bedding, and attraction/repulsion of objects. Next, the dimensions of each object need to be
realistic shapes and facies associations directly into the reservoir model by means of ob- quantified, usually in terms of a distribution of width, thickness, or width-to-thickness ratio,
jects; see Fig. 2.1 for some examples. Architectural elements with crisp geological and vertical cross section parameters, sinuosity, etc.
In the case of channel-type reservoirs (fluvial or submarine), various sources of infor-
mation can be considered:

. should correct the biases that come up when inferring 3D object properties from 2D
Outcrop studies of analog systems are the best source of information, although one

outcrops (smaller 3D objects are less likely to occur in 2D sections).

. Well-log data may provide information on object geometry and relate object shape
parameters from outcrop data to actual subsurface object shapes. Wells usually are
drilled in high-pay zones, which may introduce bias into the parameters derived from
wells. A parameter such as the proportion of channel vs. background facies may be
subject to bias when estimated directly from wells.

Compared to cell-based methods, object-based methods provide realistic shapes but are
harder to constrain to local reservoir data, such as dense well data, high-quality 3D/4D seis-
mic, and production data. Object shapes need to be "morphed" (i.e., their location and
geometry must be changed to match the local reservoir data; see Chapter 3). It is more dif-
ficult to perturb large objects to match abundant data than it is to generate cell-based mod-
els directly constrained to the same data. Therefore, object-based methods typically are ap-
plied with few wells and low-resolution seismic. They also are used in sensitivity studies
analyzing the impact of reservoir objects' connectivity to flow.

2.2.2 3D Training-Image Models. Geological complexity can rarely be described by a

limited set of object shapes. Outcrop data often display a large variety of geological het-
erogeneity types over many length scales. Geological features, from fine-scale barriers to
large-scale flow conduits, may need to be represented accurately in geocellular models to
produce realistic models for flow prediction.
The 3D training-image approach is a relatively new tool for geologists to communicate
their interpretations of geological heterogeneity style explicitly as a full 3D image. The 3D
training image is not a reservoir model; it is a conceptual rendering of the major hetero-
II
geneities that may exist in the subsurface. The aim of geostatistics, then, is to build reser-
II voir models that mimic the geological heterogeneity of the 3D training image and, at the
II

11
same time, constrain such models to actual location-specific reservoir data.
Fig. 2.1- Two examples of reservoir models built with object models: a fluvial channel reservoir An example of a 3D training image for an actual field is shown in Fig. 2.2a. Training im-
(top) and turbidite lobes (bottom). Courtesy of the Norwegian Computing Center and Alister ages may be defined at various reservoir scales, such as the bedding scale or the flow-unit
McDonald, Roxar.
I
scale (see Fig. 2.2b).
""
III
 20 Petroleum Geostatistics Modeling Geological Continuity 21

(a) Training image of a tidal-dominated reservoir Why Not Use the 3D Training Image as a Geocellular Model? The training image is a
with seven facies (courtesy ofChevronCOlp.)
mere concept; it depicts geological variability as interpreted from reservoir data and from
geological understanding of the reservoir. In this sense, the 3D training image is a model of
geological heterogeneity, just like an object model, but the training image has no "local ac-
curacy" (see Chapter 1) in the sense that it need not be constrained to any location-specific
reservoir data such as well-log, seismic, well-test, or production data, nor need the training
image have the same dimensions of the actual subsurface reservoir. A training image may
display the correct style of geological heterogeneity, but it lacks the local accuracy and data
consistency required for a reservoir model.
In reservoirs, particularly at the appraisal stage with only a few wells available, many al-
ternative training images should be produced, reflecting the inherent uncertainty about
the subsurface geological scenario. As discussed in Chapter 5, this set of alternative train-
ing images reflects a primary factor of uncertainty that needs to be accounted for in the
decision-making process.
What Are Sources for Obtaining 3D Training Images?
. aBoolean/process-based (Figs. 2.2a and 2.2b): The 3D training image is, in this case,
facies model generated with object-based methods; however, it is not constrained
to any local data.
. High-resolution seismic: Shallow seismic techniques can illuminate 3D heterogene-
ity near a 2D outcrop section or in neighboring or shallower, nonreservoir areas of an
(b) Training image of sinuous channels with point bars actual field.
obtained from a process-based technique
. Two-dimensional outcrop sections can be interpreted and interpolated with com-
puter-aided design (CAD) techniques to provide full 3D models.

Three-dimensional training images are not easy to come by because they force the reser-
voir modeler to be explicit about his prior vision of 3D spatial variability of facies or petro-
physical properties.

2.3 Describing Geological Continuity With a Variogram

2.3.1 Continuity as Captured by an Experimental Variogram. Variogramsare usedto
describe the geological continuity of "homogeneously heterogeneous" properties. This
means that variograms are best suited for describing the geological continuity of petro-
physical properties within relatively homogeneous layers or major flow facies. Variograms
should not be used to describe facies geometry unless the facies distribution itself is fairly
homogeneous. Therefore, variograms are typically used to describe the continuity of poros-
ity and permeability within layers or facies bodies.
To provide an intuitive understanding of the variogram, consider in Fig. 2.3 a set of syn-
thetic reservoir layers within which porosity is distributed. The experimental variogram is
a quantitative measure of spatial correlation and allows us to quantify/discriminate the vi-
sual differences between the various synthetic models in Fig. 2.3.
The simplest way of looking at spatial continuity is to measure the degree of linear cor-
Fig. 2.2- Training images generated with (a) an object-based technique (provided by Sebastien relation between the values of a single property measured at any two locations separated by
Strebelle, Chevron Corp.) and (b) a process-based technique (using a geomodeling software a certain 3D vector h, then plot that correlation vs. h.
program) without the need for these 3D concepts to be locally constrained to reservoir data. Denote by leU)the property located at location u=(x,y,z).In 3D space, the separation vec-
tor h between any two spatial locations u=(x,y,z) and u+h=(x+hx,y+hy,z+hz)is fully char-
acterized by a distance Ihland a direction ()represented by two angles: the azimuth and dip
angles (see Fig. 2.3f). The experimental (semi-) variogram is defined as
 Modeling Geological Continuity 23

1 N(h)

y(h) = 2N(h) ~ [z(u) - z(u+h) Y

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (2.1)

0.4
Compare behavior The variogram measures the average squared difference between two values z separated by
at Ihl~O,between
0.33
cases (a) and (b)
h. Plotting y(h) vs. h, with each h taken along the same direction e, visualizes the increase
t.: 026
in variability between two property values with increasing distance Ihl in that direction e.
0.20

0.13 This increase is expected because, on average, one expects that two values measured at in-
0.06 creasing distance from each other are likely to be more different or dissimilar. The vari-
1 0.00
ogram captures this increasing dissimilarity.
Variograms can be calculated for several directions (several combinations of azimuth and
dip angles) in 3D space. Because of the anisotropy of the geological continuity, the shapes
N/S direction of the experimental variograms depend on the direction along which they are calculated. To
I
I
I
illustrate this, consider the variograms in Figs. 2.3a through 2.3e:
I
I
. I
. The variogram increases from zero at the origin Ihl=Oand often reaches a plateau at
a certain distance. The distance at which this occurs is termed the range-(alsotermed
depfu ,
~'EW
NS::trJ .0
t sill ,j
10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 correlation length) for the direction considered. This range indicates the average ex-
:-r- ~
T tent of continuity/correlation along various directions and can be related to the size
Vertical direction
of dark and white patches, as indicated in Fig. 2.1b. The plateau is termed the sill and
corresponds to the statistical variance of the property.
.0100
. Often, such as in Fig. 2.3c, there is an apparent discontinuity on the y-axis of the var-
.0080
iogram plot, termed the nugget effect. The nugget effect reflects the geological vari-
y .0060 ability at scales smaller than the smallest experimental distance Ihl.
.0040J/~ Apparent discontinuity
A range that varies with direction indicates anisotropy in spatial continuity. In most reser-
~. .0020 voirs, one expects the range in the vertical/depth direction to be shorter compared to a hor-
. .0000 ,
.0 10.0 20.0 30.0 40.0 50.0 60.0 700 80.0 izontal direction. The variation of the range in various 3D directions typically displays an
Distance
ellipsoidal behavior. Such an ellipsoid can be summarized with six parameters: the ranges
along major, medium, and minor axes, and the rotation parameters defining its 3D orienta-
tion in terms of azimuth, dip, and plunge.
.0080 The variogram behavior at the origin (i.e., for small h) is an important factor to differen-
.0060
tiate geological continuity. Compare, for example, the variogram behavior at the origin in
Y Fig. 2.3a with Fig. 2.3b. A slower increase from Ihl=Oin Fig. 2.3a coincides with a

,
.0040

.0020
\
Variogram in horizontal
smoother-looking continuity in reservoir porosity.
The variograms calculated in Fig. 2.3 suggest a way to discriminate and quantify between
.0000
does not reach sill various patterns of continuity using a limited set of parameters, summarized from experi-
(e) Azimuth
.0 10.0 20.0 30.0 40.0 50.0 600 70.0 80.0 mental variograms:
Distance

U=(X,y,z) . The variation in correlation lengths/ranges along various directions.

z(u) = property z measured at u
. The amount of nugget effect and the behavior of the variogram for small distances Ihl.

Fig. 2.3-(a) Reservoir with smooth spatial continuity translates into a slower decrease of the
y(N/S) variogram at the origin; (b) the range in the vertical and horizontal direction can be associated
x (E/W)
with an average extent of black and gray patches in each direction; (c) a nugget effect indicates
small scale heterogeneity; (d) strong layering gives rise to a variogram in the layering direction
Fig. 2.3-(Caption on facing page.) that does not reach a zero; (e) notation convention.
 24 Petroleum Geostatistics
Modeling Geological Continuity 25

2.3.2 Variogram of Categorical Variables. A variogram can be calculated for either con- teristics for different ranges of values; for example, high permeability may be associated
tinuous or categorical variables. Fig. 2.4 shows an example of a variogram for a facies map with a specific direction with stronger spatial continuity than low permeability.
containing three facies (three categories), each exhibiting a different spatial variability. Fa-
cies I appears isotropic, while facies 2 and 3 show a distinct anisotropy, orthogonal to each 2.3.3 Issues of Stationarity and Trend. A note of caution should be made regarding the
other. A single variogram could not reflect the difference in continuity of each facies. To en- calculation of any statistics, whether a histogram, scattergram, or variogram. Such statis-
able distinction between the three facies, an indicator variable is assigned to each facies. tics are calculated on the basis of data gathered or pooled from some given 3D volume (e.g.,
The indicator of facies i is defined by assigning a value of "I" to all locations at which fa- a specific layer). To meaningfully interpret the resulting statistics and characteristics such
cies i occurs and a value of "0" elsewhere; see the bottom of Fig. 2.4. An indicator vari- as anisotropy and correlation length, the geological continuity within this 3D pool should
ogram, which is the variogram calculated for each of the three maps, now reveals the dif- be fairly similar; that is, in mathematical terms, it should be stationary (see Chapter I). It
ference in the anisotropic correlation length of each facies. would not make sense to pool data from two regions with vastly different geological con-
As suggested in the Introduction, variograms are of limited use for defining facies geom- tinuity.
etry; therefore, indicator variograms should be applied mostly to a continuous variable by The patterns of heterogeneity displayed in Fig. 2.3 do not show any major lateral or ver-
binning the variable into a number of classes (typically, three classes for low, medium, and tical discontinuities and, hence, are deemed stationary. In actual reservoirs, one would need
high values). This allows investigating if that variable may have different spatial charac- to determine the regions (compartments) with layers of similar geological heterogeneity.
Each such region or layer would display characteristic correlation lengths and anisotropy
directions yet still may display some vertical or horizontal trend of the reservoir property
under study. Variograms are, however, difficult to interpret in the presence of such a trend.
Facies map, 3 facies
.
Variogram
_'H'_.
"""""~-"""",,, To account for such a trend, one could decompose the spatial variability of the property
ZOOj

1.50.'
~ -..~t...: into two parts: a reasonably smooth trend component and a more random residual compo-
~ NW/SE direction nent around that trend. Trends are often modeled deterministically using seismic data (im-
LEW
1.00

.50
- NE/SW direction
pedance inversions, for example). They also can be modeled using appropriate geostatisti-
cal interpolation algorithms (kriging with trend).
.00
b 10.0 ZO.O TO 2.3.4 Obtaining Variograms From Well Data. Actual reservoir data are typically
Distance
sparsely measured along a few vertical and deviated wells. Variograms are then calculated
Indicator map, facies I Indicator map, facies 2 Indicator map, facies 3 from these well data, with (possibly) a prior coordinates transform into the depositional
space (see Fig. 1.2). Various textbooks in geostatistics tutor the reader on how this is done

I Facies 3
in practice. Most current software packages provide excellent visual and interactive ways
to calculate experimental variograms from well data within complex 3D structures.
When only a few vertical or slightly deviated wells are available, the variogram in the

"",
n '

' I
rest horizontal directions is often ill-defined; it appears to be a pure nugget effect (purely ran-
'L
dom) and therefore difficult to interpret. The vertical direction is typically more densely
.~~~.,
sampled, and a range value can be determined readily. Consider the variogram in Fig. 2.5,

Variogram
lfacies 1
1
Variogram facies 2
l which is obtained by "drilling" five wells into the reservoir on the left. The variogram in
the vertical direction has identifiable characteristics (correlation length around 15 units),
~_.
Variogram facies 3 while the experimental variogram in any horizontal direction appears as a pure nugget, de-
.ZOO
/'" --....
'ZOO '

.. " . ::::::::n: :::::.::::: .ZOO

,
'_':"""::~:::::
.,f..'
spite the fact that strong channel-type continuity exists in horizontal planes!
What Are Possible Solutions?
150
.100 I J

.050
f .150 II
.100.I
"
"
,,'
.. .150.;I
.100
'i,.'
J1 . Assume from prior expertise the missing variogram parameters such as horizontal
correlation lengths or anisotropy directions. This is better than accepting the default
.050 .050
.000
.000. .000. values of an interpolation software, which often amounts to assuming no spatial cor-
0 10.0 ZO.O 30.0 0 10.0 ZO.O 30.0 40.0 0 10.0 ZO.O 30.0 40.0
Distance Distance
relation (pure nugget effect) or perfect correlation (layer-cake model). No correlation
Distance
or perfect correlation are extreme cases of spatial correlation and are not necessarily
Fig.2.4-A categorical map (top)withthree imbricatedfacies. Theoverallvariogramdoes not in- safe or conservative choices. Practitioners often tryout various variogram models by
dicate significant anisotropy (middle), Three indicator facies maps derived from the three-facies performing stochastic simulation (i.e., generating synthetic reservoirs such as the one
map (bottom); the three indicator variograms display the anisotropy pattern specific to each shown in Fig. 2.3). The variogram corresponding to simulated images that "look rep-
component facies.
resentative" for the underlying field is then retained.
 26 Petroleum Geostatistics 27
Modeling Geological Continuity

0.33

0.26 00
<l)
";:::::
0.20 .....
<l)
p.,
0.13 0
.....
p.,
0.06 ~u
0.00 'uo
E p.,
0
.....
....-
<l) ;i
.0160 Porosity vargram based o"five wells p., 0
~..

~/
"0
or I=i 0
.0120J ".
r. Vertica! (jirection
.;::
"'id
~
OJ-OJ
en
~
00
<l)
:!:
C)
- y~--~ 'G ,5
§<"8 ~ Qj
Yo080Jk--/ "E: 8
.~ bO
-"'-0
4-<
0
"C
0
E
N45E direction t.) '" I=i "C
.~ ;:: 0 C
.0040 ~'"
.~ as
0 .;,f c
~~.~ "3 0
OJ OJ S :;::;
.0000 "':::i en as
o. 40. 80. .00
'uo :iu
120.
;:: "0
Distance 0 I=i iii
u ~ u
I=i E
Fig. 2.5-Vertical and horizontal variogram calculated from five wells drilled in a channel reser-
0
";:::::
~
..
as
C)
voir. While a clear continuity exists in the N45E-direction (channeling direction), the variogram S .2
is noisy in that direction. Variograms are, therefore, typically used to describe the variability
petrophysical properties within major facies, not the spatial variability of such facies.
of ";:::::
00
IJ:.1
~I
CD
C')
.....
N
<l) ciJ
§.
u:
Another common practice is to borrow variogram parameters from outcrop data, such ...<:1
U
as the anisotropy ratio. One also may consider borrowing horizontal variogram pa-
rameters directly from seismic data.

2.3.5 Obtaining Variogram Models: A Workflow. Fig. 2.6 presents an overview of a

three-stage workflow for quantifying geological heterogeneity with a variogram:

Pool data over stationary zones, removing clearly interpretable trends.

Calculate experimental variograms and interpret their characteristics.
Model the variograms using a few parameters such as nugget constant, ranges, and
anisotropy ratios (see the Advanced Reading Box).

2.3.6 Limitations of the Variogram. The application of traditional, variogram-based geo-

statistics in the area of reservoir modeling was essentially carried over from its original ap-
plication in the mining industry. While mining has abundant drill-hole data, oil and
 28 Petroleum Geostatistics 29
Modeling Geological Continuity

Advanced Reading Box

Variogram Modeling
Three-dimensional variogram modeling consists of fitting an analytical model de-
fined by a few parameters to the experimental variograms. In the indicator case (cat-
egorical variables), this requires fitting a variogram model for each facies' experi-
mental indicator variogram. There are two main reasons for doing so:
0.4

. Estimation and stochastic simulation methods require the variogram to be

known for all distances, not only those experimental distances between sam-
0.33

0.26 .~
ple data locations. 0.20

. There are restrictions on the type of variogram models that can be used. Spa-
tial estimation techniques provide, in addition to estimates, a quantification of (c)
0.13

0.03 (d)
0.00
the error of that estimate in terms of an error variance (see Chapter 3). The
error variance necessarily needs to be positive. This requirement translates into
conditions on the variogram analytical model.

The latter reason makes variogram modeling a challenging task. In practice, one will
y
make use of predefined variogram models for which the positivity condition is
known to be satisfied. These basic models essentially serve as basis functions from
which, by linear combination, more-complex variogram models can be derived to fit
the experimental data. The details of such modeling are outside the scope of this .~
.0000, , . , , "
work and are the subject of more technically oriented books. .0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0
Distance

Fig. 2.7- Three alternative geological scenarios exhibiting the same vertical and horizontal var-
iograms. The variable is porosity.

gas reservoirs often lack this amount of data, and a direct inference of the variogram pa-
rameters is often not possible. In practice, one rarely has the opportunity to model "clean- Adopting a variogram-based approach often may be appealing because defining the var-
looking" experimental variograms from actual reservoir data. However, lack of adequate iogram requires setting only a handful of parameters, as opposed to drawing explicitly a full
data is not a good reason to dismiss geostatistics and resort to the arbitrary pure nugget var- 3D picture of heterogeneity (training-image approach) or defining fully the geometry offa-
iogram model implicit to many off-the-shelf interpolation software packages. cies objects (Boolean approach). Nevertheless, an important note of caution should be
The major limitation of the variogram is its inability to model realistic geological features. made: as demonstrated by Fig. 2.7, variogram-based geostatistics shift the responsibility of
Reservoir models built solely with variograms as the geological heterogeneity quantifier are stating one's geological belief about heterogeneity from the reservoir modeler to the geo-
often deemed unrealistic, appearing "too synthetic" and too "homogeneously heteroge- statistical algorithm. While the variogram is not fully explicit about geological hetero-
neous." In addition, it is difficult to link variogram parameters to curvilinear geological geneity, any geostatistical algorithm that uses a variogram will generate 3D numerical mod-
shapes and horizontal and vertical architectural elements as interpreted from outcrop data. els with an explicit 3D heterogeneity model (as with any of the three models in Fig. 2.7).
To demonstrate this point, consider the three alternative synthetic reservoirs in Figs. 2.7a Often, such variogram-based geostatistical algorithms will generate unrealistically homo-
through 2.7c. The corresponding variograms in the horizontal and vertical directions are geneous reservoirs. Hence, the decision to choose a specific variogram-based algorithm is
shown in Fig. 2.7d. The large difference in geological variability and connectivity in Figs. no less consequential than the decision to choose certain object types along with shape pa-
2.7a through 2.7c does not result in any significant difference between variograms. Note rameters or to adopt fully explicit 3D training images.
that these experimental variograms are calculated from exhaustive data, a luxury never af-
forded in real applications! Even with exhaustive data, variograms are poor discriminators Nomenclature
of geological scenarios. This limitation arises because the variogram models correlation be- h = lag vector; notation used to describe the distance between locations u and u+h
tween only two spatial locations-it "explores the world using only two fingers"; therefore, N(h) = number of data pairs {zen), z(u+h)} used to calculate the variogram for 3D
it cannot capture any curvilinear geological patterns, hence the earlier recommendation of separation vector h
not using the variograms for modeling the geometry of major facies. u = location in 3D space with coordinates (x,y,z)
III
,-- 30 Petroleum Geostatistics

z(u) = unknown property z at location u

y(h) = variogram
e = direction

Suggested Reading
On Variogram Calculation and Modeling
Thefollowing referencesprovide useful suggestionsfor calculating and modeling vari-
ograms:
Armstrong, M.: "Improving the estimation and modeling of the variogram," Geostatistics
for Natural Resources Characterization, G. Verly et al. (eds.), Reidel, Dordrecht, The
Netherlands (1984) 1-20.
Deutsch, C.y. and lournel, AG.: GSLIB: The Geostatistical Software Library, Oxford U.
Press, Oxford, UK (1998).
Gringarten, E. and Deutsch, c.y.: "Methodology for Variogram Interpretation and Model-
ing for Improved Petroleum Reservoir Characterization," paper SPE 56654 presented at
the 1999 SPE Annual Technical Conference and Exhibition, Houston, 3-6 October.
Isaaks, RH. and Srivastava, R.M.: An Introduction to Applied Geostatistics, Oxford U.
Press, Oxford, U.K.
lournel, AG.: "Non-parametric estimation of spatial distributions," Mathematical Geology
(1983) 15, No.3, 445.
This paper provides thefoundation of indicator geostatistics.
Olea, R.A: "Fundamentals of semi-variogram estimation, modeling and usage," Stochas-
tic Modeling and Geostatistics-Principles, Methods and Case Studies, I.M. Yams and
R.L Chambers (eds.), AAPG Computer Applications in Geology (1994) No.3, 27.
Yao, T. and lournel, AG.: "Automatic modeling of (cross) covariance tables using fast
Fourier transform," Mathematical Geology (1998) 30, No.6, 589.
This paper uses Fourier transforms to make variogram modeling more automatic.

On Boolean Methods
Deutsch, c.y. and Wang, L: "Quantifying object-based stochastic modeling of fluvial
reservoirs," Mathematical Geology (1996) 28, No.7, 857.
One of the first practical algorithmsfor object modeling.

Thefollowing papers represent a selectionfrom the large body of work on object mod-
elsfrom the Norwegian School:
Haldorson, H.H. and Damsleth, E.: "Stochastic Modeling," lPT (April 1990) 404.
Holden, L et al.: "Modeling of fluvial reservoirs with object models," Mathematical Ge-
ology (1998) 30, No.5, 473.
Skorstad, A, Hauge, R., and Holden, L: "Well conditioning in a fluvial reservoir model,"
Mathematical Geology (1999) 31, No.7, 857.

On Training Images
Caers, 1. and Zhang, T.: "Multiple-point geostatistics: a quantitative vehicle for integrating
geologic analogs into multiple reservoir models," Integration of outcrop and modern
analog data in reservoir models, AAPG Memoir 80 (2004) 383-394.
This paper presents various suggestions and guidelines for constructing 3D training
images.