Geologic Modeling

and Mapping
COMPUTER APPLICATIONS IN THE EARTH SCIENCES
Aseries edited by Daniel F. Merriam

1969 - Computer Applications in the Earth Sciences

1970 - Geostatistics
1972 - Mathematical Models of Sedimentary Processes
1981 - Computer Applications in the Earth Sciences: An Update of the 70s
1988 - Current Trends in Geomathematics
1992 - Use of Microcomputers in Geology
1993 - Computerized Basin Analysis: The Prognosis of Energy and Mineral Resources
1996 - Geologie Modeling and Mapping
Geologic Modeling
and Mapping

Edited by
Andrea Forster
GeoForschungsZentrum Potsdam
Potsdam, Germany

and
Daniel F. Merriam
University of Kansas
Lawrence, Kansas

Plenum Press • New York and London

 Library of Congress Cataloging-in-Publication Data
On file

Proceedings of the 25th Anniversary Meeting of the International Association for

Mathematical Geology, held October 10- 14, 1993, in Prague, Czech Republic

ISBN-13: 978-1-4613-8017-7 e-ISBN-13: 978-1-4613-0363-3

DOl: 10.1007/978-1-4613-0363-3

© 1996 Plenum Press, New York

Softcover reprint of the hardcover I st edition 1996
A Division of Plenum Publishing Corporation
233 Spring Street, New York, N. Y. 10013

10987654321

All rights reserved

No part of this book may be reproduced, stored in a retrieval system, or transmitted in any form
or by any means, electronic, mechanical, photocopying, microfilming, recording, or otherwise,
without written permission from the Publisher
 PREFACE

This volume is a compendium of papers on the subject, as noted in the

book title, of modeling and mapping. They were presented at the 25th
Anniversary meeting of the International Association for Mathematical
Geology (IAMG) at Praha (Prague), Czech Republic in October of 1993.
The Association, founded at the International Geological Congress (IGC)
in Prague in 1968, returned to its origins for its Silver Anniversary
celebration. All in all 146 papers by 276 authors were offered for the
165 attendees at the 3-day meeting convened in the Hotel Krystal. It was
a time for remembrance and for future prognostication.

The selected papers in Geologic Modeling and Mapping comprise a broad

range of powerful techniques used nowadays in the earth sciences.
Modeling stands for reconstruction of geological features, such as
subsurface structure, in space and time, as well as for simulation of
geological processes both providing scenarios of geologic events and how
these events might have occurred. Mapping stands for spatial analysis of
data, a topic that always has been an extremely important part of the
earth sciences. Because both modeling and mapping are used widely in
conjunction, the book title should reflect the close relation of the subjects
rather than a division.

Here, we bring together a collection of papers that hopefully contribute

to the growing amount of knowledge on these techniques.
All of the authors are experts in their field so the reader will have
authoritative papers on the various aspects of the subject. Some time has
elapsed since the Prague meeting, time used by the contributors of the
book to incorporate additional and new ideas into their offerings. We
want to thank all the authors for patience and help in preparing the book.
We also thank the many reviewers who through their efforts improved the
content and presentation of the papers. LeaAnn Davidson of the Kansas
Geological Survey helped to prepare the final layout and her assistance

v
Vl PREFACE

is greatly appreciated. Cora Cowan, also of the Geological Survey,

assisted with preparation of the index.

We hope the readers will enjoy the content and catch some of the flavor
of the Silver Anniversary Meeting of the IAMG in Praha - those who
attended enjoyed and learned much.

Andrea Forster
 CONTENTS
Introduction, by D.F. Merriam ......................... IX

Sedimentary process simulation: a new approach for

describing petrophysical properties in three
dimensions for subsurface flow simulations,
by 1 Wendebourg and lW. Harbaugh. . . . . . . . . . . . . . . 1

Modeling of multi component diagenetic systems, by

R.Ondrak .................................. 27

Modeling petroleum reservoirs in Pennsylvanian

strata of the Midcontinent, USA, by
W.L. Watney, lA. French, and W.J. Guy. . . . . . . . . . . .. 43

Thermal modeling at an ancient oro genetic front with

special regard to the uncertainty of heat-flow
predictions, by U. Bayer, B. Lfulenschlol3,
1. Springer, and C.v. Winterfeld ................... 79

Effective transport properties of artificial rocks - means,

power laws, and percolation, by o. Kahle and
U. Bayer ................................... 95

Three-dimensional modeling of geological features

with examples from the Cenozoic Lower Rhine
Basin, by R. Alms, C. Klesper, and A. Siehl .......... 113

Volumetrics and rendering of geologic bodies by

three-dimensional geometric reconstruction
from cross sections or contour lines, by
H. Schaeben, S. Auerbach, and E.U. Schutz . . . . . . . . . .. 135

VII
Vlll CONTENTS

The effect of seasonal factors on geological

interpretation of MSS data, by Do Yuan, J.Eo
Robinson, and MoJ. Duggin 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 153

Reconstruction of the Leduc and Wabamun rock salts,

Youngstown area, Alberta, by NoL. Anderson
and RJ. Brown 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 175

Use of the computer for the structural analysis

of the Ordovician sedimentary basin in
Estonia, by A. Shogenova 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 199

Pairwise comparison of spatial map data, by DoFo Merriam,

u.e. Herzfeld, and A. Forster 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 215

Applications of spatial factor analysis to multivariate

geochemical data, by Eoe. Grunsky, Qo Cheng,
and F.Po Agterberg 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 229

Geostatistical solution for the classification problem with

an application to oil prospecting, by J. Harff,
R.A. Olea, J.e. Davis, and Goe. Bohling 0 0 0 0 0 0 0 0 0 0 0 0 0 263

Transition probability approach to statistical analysis

of spatial qualitative variables in geology, by
J. Luo 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 281

An intelligent framework for geologic modeling

applications, by L. Plansky, K. Prisbrey,
Co Glass, and L. Barron 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 301

Contributors 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 323

Index 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 327
 INTRODUCfION

This book on Geologic Modeling and Mapping contains 15 papers by 38

authors. Although the book is divided into two parts - modeling and
mapping - many of the papers could be in either part or both. This
suggests that modeling and mapping go hand-in-hand in today's research
and indeed they do. Both also are heavily dependent on graphics to
convey the author's intent to the reader and these papers reflect that
dependence.

The first paper on SEDSIM, a sedimentary simulation programming

system, was used by Wendebourg and Hamaugh to describe petrophysical
properties of subsurface flow simulations in 3D. This approach has
practical applications in describing aquifers and oil and gas reservoirs.
They conclude from their studies that '... interpretations of simulation
results require careful consideration of assumptions and boundary
conditions ... and an understanding of the ... geological framework ...
represented by the simulation.'

Modeling of multi component, large-scale diagenetic systems is the subject

of the paper by Ondrak. This approach, based on several assumptions
and simplifications, helps us to understand the complex story of
dissolution and precipitation of minerals in porous media. The model
allows simulation of the time and place of porosity and mineral
distribution in a sandstone; Ondrak illustrates the modeling technique with
several theoretical examples representing different geological settings.

Watney, French, and Guy direct their attention to modeling of petroleum

reservoirs with an example from the Pennsylvanian of the Midcontinent
(USA). They note that this type of modeling is limited by the available
data and knowledge of the reservoir. This approach to modeling can be
qualitative, process oriented, or quantitative, and they conclude that a
combination of these approaches probably provides the best model.

IX
x INTRODUCTION

Bayer and otben modeled the thermal conditions and heat flow for an
ancient orogenetic front. They concentrated on the interaction of different
factors and how they change the geothermal field and the uncertainties
involved in extrapolation of near-surface conditions to depth. They
concluded that modeling of complex situations under uncertainty allows
a better understanding of various alternative scenarios.

Power law equations, related to transport properties in a two-component

system, as applicable to the frequencies of the material and statistical
measurements of anisotropy is the basic contention of Kahle and Bayer.
Their two-component system provides a simple idealized model for
complex anisotropic geological models.

Three-dimensional modeling of geological surfaces and bodies is the

subject of the paper by Alms, Klesper, and Siehl. They illustrate their
modeling by examples from the Cenozoic of the Lower Rhine Basin in
Germany. The concepts and software systems used in the modeling for
handling 2D and 3D data and the graphic output is outlined in their
presentation. The graphics give an excellent 3D perspective to the
results.

A three-dimensional method for the geometric reconstruction of geologic

data is described and discussed by Schaeben, Auerbach, and Schutt.
They use parallel cross sections or contour lines to accomplish the
reconstruction, and then display the results graphically in 3D. Their
method is automatic and requires no interaction by the user.

All of these papers stress the problems involved with modeling and the
limits of available data, understanding the processes, and the wide range
of interpretations that can be given to the results. All of the authors
agree that this approach is useful and helpful in understanding geological
conditions as known today and that with continued refinement and
resolution, the simulation can be better and perhaps even more exact.
Undoubtedly, we are the threshold of many exciting and valuable
possibilities in the field of modeling.
INTRODUCTION Xl

The second part of the book is concerned with mapping or spatial

analysis, and many of these papers stress practical (or economic)
applications. As will be seen in these papers, as with modeling, graphics
is an all important aspect.

Leading off in this part of the book is a paper on MSS data collected by
LANDSAT, which were used to determine the temporal variations for an
area in Nevada. Yuan, Robinson, and Duggin describe their statistical
analysis of the remotely sensed data by discriminant analysis to recognize
different lithologies and seasonal variations.

The paper by Anderson and Brown explores ways to reconstruct the

location in the subsurface of the now gone Devonian salt units removed
by dissolution. Thickness maps and seismic profiles were utilized in the
reconstruction of the former extent of the salt. They used their model to
interpret the processes and timing of the dissolution.

Shogenova uses a computer in a structural analysis of an Ordovician

sedimentary basin in Estonia. Trend analysis was applied to a series of
structural and stratigraphic maps to help in determining regional trends
and isolating local features. From this analysis, she then was able to
interpret better the basin's history.

Pairwise comparison of spatial (map) data is accomplished effectively by

computing a similarity coefficient and constructing a resultant map
according to Meniam, Herzfeld, and Fdrsier. This gives insight into the
predictability of one map to another, a quantitative descriptor for
classification, and additional data for interpretation. They describe this
approach and illustrate it by an example using geological, geophysical,
and geothermal data for an area in southeastern Kansas.

Grunsky, Oleng, and Agiemerg describe and apply spatial factor analysis
to determine multivariate relationships in geochemical data, illustrated by
examples from British Columbia in Canada. They determined that the
regional geochemical trends are associated with the underlying
lithological variations. Because of this association, the technique can be
xu INTRODUCTION

used as an aid in the exploration for mineral deposits in areas that have
been adequately sampled.

Additional information on their continued study on exploration for

mineral deposits is provided in the contribution by HarlT and othen. The
additional information is to determine by discriminant analysis which
geological observations are the most informative. This technique is
illustrated by an example of the distribution of oil and gas fields in a
petroleum area of western Kansas. They note that the method can be
used for '... planning of exploration activities because it provides the
geologist with information about favorable targets ... '

A Markov chain model is used by Luo for analyzing a set of spatial data
described by discrete states. The model consists of five steps: definition
of data; analysis; simulation; assessment of simulation results; and
graphical representation of results. Methods of Markov analysis and case
studies demonstrate this approach.

Plansky and othen use an artificial intelligence (AI) and expert systems
(ES) approach as a framework for constructing system models. AIlES,
and this includes neural networks, has been used more and more recently
in the study of earth systems. They give a practical example from the
mining industry as an application of their approach.

These papers give the reader a good broad background and detailed
examples on the subject of Geologic Modeling and Mapping. Both
academic and practical approaches are discussed by the authors. The
references cited in the papers will give a sweep of available published
material on the subj ects for in depth reading. All in all the papers give a
good idea as to the subjects covered during IAMG's Silver Anniversary
meeting in Praha but the subjects are updated as of mid-1995.

To all users of this book - good reading, creative thinking, and productive
results!

Daniel F. Merriam
Geologic Modeling
and Mapping
 SEDIMENTARY PROCESS SIMULATION: A NEW
APPROACH FOR DESCRIBING PETROPHYSICAL
PROPERTIES IN THREE DIMENSIONS FOR
SUBSURFACE FLOW SIMULATIONS

Johannes Wendebourg
Stanford University, Stanford, California
and
Institut Franfais du Petro/e, Rueil-Malmaison, France

John W. Harbaugh
Stanford University, Stanford, California

ABSTRACf

Subsurface fluid flow is critically dependent on the 3D distribution

of petrophysical properties in rocks. In sequences of sedimentary rocks
these properties are strongly influenced by lithology and facies
distribution that stem from the geologic processes that generated them.
Three types of simulators are contrasted that represent variations of
petrophysical properties: stochastic simulators, stratigraphic-form
simulators, and sedimentary process simulators. The first two generally
require closely spaced well information or seismic data and can be
"conditioned" to accord with the data, but neither can represent the
influence of depositional processes directly. By contrast, process
simulators do not require closely spaced data, but they generally cannot
be forced to accord closely with the data. The sedimentary process

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 1
2 WENDEBOURG AND HARBAUGH

simulator described here, known as SEDSIM, provides 3D geometric

forms and spatial distributions of grain sizes of alluvial, fluvial, and
deltaic deposits that are controlled by information supplied as initial and
boundary conditions, including the initial topographic surface, and fluvial
and sediment discharge volumes through time. The resulting sediment
distributions can be compared directly with maps and sections based on
well data and seismic surveys, but they can also be transformed into
estimates of porosity and permeability, thereby placing them in form for
direct use with subsurface flow simulators. Several recent applications of
SEDSIM in generating descriptions of groundwater aquifers and
hydrocarbon reservoirs and their use in subsurface flow simulations are
presented. Comparisons between complex actual and simulated 3D
sequences suggest that statistical descriptions of simulated sequences
could be used as input to stochastic simulators. This would combine the
advantages of stochastic simulators that can condition simulations to field
data, with the advantages of process simulators that treat geometric forms
and flow properties of sequences interdependently and represent the
development of sedimentary facies through space and time.

INTRODUCflON

Three-dimensional representation of petrophysical properties is

needed for studies that involve subsurface fluid flow at various scales and
stages. The main applications include the exploration for and production
from oil and gas reservoirs, and in contaminant remediation of
groundwater aquifers.
In hydrocarbon exploration, we need to know the properties of
carrier beds if they exist and the migration paths through which
hydrocarbons may have moved before accumulating in traps. Given
enough information about the geologic structure and petrophysical
properties of beds in the "area, we could simulate the 3D motions of
migrating hydrocarbons to forecast where they may be trapped.
In exploitation drilling, petrophysical properties of rocks
interpreted from well logs and cores, and from production records guide
the location and engineering features of wells. Opportunities to extract
hydrocarbons from sedimentary sequences depend strongly on the 3D
distributions of porosity and permeability within them, but rarely are
these properties interpreted in appropriate 3D detail. Well logs and
SEDIMENTARY PROCESS SIMULATION 3

production records provide petrophysical information at and near well

bores, but extrapolation in suitable detail beyond well bores is
exceedingly difficult if conventional subsurface geological procedures are
employed.
In aquifer remediation, small-scale sedimentary heterogeneities
may determine the rates and transport direction of contaminants in the
subsurface, and it is therefore important to understand their spatial
distribution in choosing effective clean-up procedures. Contaminants not
only pollute the groundwater, they also interact with the sediments where
dispersion and adsorption processes are as important as advective
transport. Without a detailed spatial description of the petrophysical
properties that affect flow within the aquifer, these processes cannot be
modeled correctly.
Simulators are needed to generate these distributions, for they are
difficult or impossible to obtain in 3D detail otherwise. Three main types
of simulators are available: (1) stochastic simulators, (2) geometric-form
simulators, and (3) sedimentary process simulators. Once a simulator is
selected, the challenge is to adapt it to obtain responses in suitable accord
with information provided by real data such as existing wells.

Stochastic Simulators

Stochastic simulators use geostatistical procedures to generate

constrained random variations of reservoir properties in two or three
dimensions (Isaaks and Srivastava, 1989; Deutsch and Joumel, 1992).
They are useful in dealing with groundwater flow problems as well as in
managing production from oil and gas reservoirs, and can be classified
into discrete and continuous subtypes (Haldorsen and Damsleth, 1993).
Discrete or object-based simulators are useful for representing
shapes and petrophysical properties of relatively large-scale sedimentary
bodies such as channels, deltaic lobes, and shale barriers, and they usually
employ rules for the dimensions, shapes, and orientations of these
sedimentary bodies (Geehan and others, 1986). By contrast, continuous
or sequence-based simulators are useful for representing small-scale
variations in petrophysical properties that vary continuously in three-
dimensions. They may be used to represent these properties within larger
sedimentary bodies previously generated with discrete simulators
(Rudkiewicz and others, 1990). Furthermore, both are capable of
4 WENDEBOURG AND HARBAUGH

according closely with data provided by wells and seismic data (Doyen
and others, 1991).
Both discrete and continuous stochastic simulators may assume
statistical stationarity, which implies that the gross statistical properties
of aquifer or reservoir beds do not change spatially. Stochastic simulators
are also sensitive to the manner in which spatial continuity is represented,
both in the choice of an appropriate mathematical model and in the
choice of correct model parameters. For example, Murray (1992)
represented permeability variations of selected oil reservoirs in the
Powder River Basin of Wyoming with stochastic simulations whose
results were conditional upon information from wells, but determined that
none of the alternative and equally probable responses were reasonable
because of the lack of continuity between beds with high permeabilities,
suggesting that the geostatistical model was inadequate to represent
variations in geological conditions. Murray then used an additional
stochastic technique termed "simulated annealing" that introduced
connectivity in the model results, thereby better approximating actual
geological variations.

Geometric-form Simulators

Geometric-form simulators are purely deterministic, contrasting

with stochastic simulators that depend on successions of random numbers.
Geometric-form simulators are generally used to produce 2D vertical
sections that display the geometrical forms of sedimentary sequences in
accord with concepts of sequence stratigraphy that have been derived
largely from interpretations of seismic sections (Strobel and others, 1990;
Lawrence and others, 1990). Geometric-form simulators are well suited
to represent the geometrical details of sedimentary units (parasequences)
that usually are stacked one upon another and form as functions of the
space available for deposition (accommodation space), which is a function
of the interplay between eustatic variations in sealevel, tectonic
subsidence, and sediment supply. Geometric-form simulators thus
incorporate geological concepts directly, in contrast to stochastic
simulators that generally do not represent geological concepts. The use of
geometric-form simulators has stimulated the development of "inverse"
methods for determining variations in eustatic sealevel and tectonic
subsidence rates from the geometrical forms and facies relationships
based on information from outcrops and wells (Lessenger, 1993).
SEDIMENTARY PROCESS SIMULATION 5

Geometric-form simulators are not directly usable for use with

flow simulators. First, they are generally 2D, which limits their use with
2D versions of flow simulators. Second, geometric-form simulators do
not provide information directly about variations in porosity and
permeability needed for flow simulators. These and other petrophysical
properties must be assigned subsequently, usually being based on
comparisons between actual facies and those generated by geometric-form
simulators.
There is an additional limitation in that geometric-form simulators
usually are operated to obtain responses that accord with continuous
seismic sections, but the degree of resolution provided by seismic sections
may be inadequate to permit accurate facies interpretations to be made,
so that subsequent assignment of petrophysical properties may be very
uncertain. This point has been elucidated by Shuster and Aigner (1994)
who have generated synthetic seismic sections from geometric-form
models whose input and boundary conditions were completely known.
These synthetic sections were then interpreted by geologists in terms of
their apparent geologic history and facies distribution, yielding a
qualitative measure of the uncertainty involved in choosing model
parameters, as well as in interpreting facies from seismic sections.

Sedimentary Process Simulators

Sedimentary process simulators link sedimentary processes with

sedimentary features and materials and operate in accord with the
physical laws and empirical relationships that govern transport,
deposition, and erosion of clastic sediment. These processes are
represented with equations that are solved numerically, and the numerical
sequences obtained as solutions represent the responses of the processes
through time and 3D space, producing sequences of beds whose
properties can be represented by spatial variations in grain sizes. These
variations in tum can be transformed into 3D distributions of porosities
and permeabilities for use with 3D flow simulators. Sedimentary process
simulators thus provide direct linkages between depositional environments
and processes that create sedimentary sequences, and their subsequent
ability to transmit pore fluids.
Some early sedimentary process simulators employed the diffusion
equation to represent sediment transport, and involved the assumption that
6 WENDEBOURG AND HARBAUGH

the rate of sediment transport is related to gradients of depositional

slopes, yielding smoothly varying representations of sedimentary
sequences in two or three dimensions (Harbaugh and Bonham-Carter,
1970). The diffusion equation for diffusion-based simulators is a
macroscopic equation that can be derived from the integration of
open-water flow equations together with empirical relationships for
sedimentation and erosion, and that is applicable mainly to intermediate
to large-scale descriptions of sedimentary sequences. The previously
mentioned early simple simulators have been improved recently by using
several lithologies (Rivenaes, 1992).
Process simulators based on the hydrodynamics of sediment
transport treat the nonlinearities of sediment transport directly and
therefore yield much more complex and realistic representations of
sedimentary sequences than the other types of simulators (Tetzlaff and
Harbaugh, 1989; Martinez and Harbaugh, 1993). However, the use of
process simulators requires a fundamentally different approach in dealing
with sedimentary sequences. Instead of employing reconstructions such
as isopachous and lithofacies maps based on well logs and seismic
sections, process simulators' responses are controlled by environmental
data that are supplied as input, including topography at the beginning of
a simulation, fluid and sediment discharge rates by streams during the
experiment, and wave directions and intensities. Little or no
environmental information can be extracted directly from well or seismic
data, and therefore most of it must be estimated from modem depositional
processes (Lee and others, 1991; Martinez and Harbaugh, 1993) or
indirectly inferred from petrographic information (Wendebourg, 1991).
Obtaining information for controlling process simulators is a major
challenge and will require improved measurement of process rates in
modem depositional environments.
The largest problem in using sedimentary process simulators is
that they may not accord exactly with specific sedimentary sequences at
known locations. For example, a large deltaic complex may be known
from logs of scattered wells and intersecting seismic lines. We can use a
process simulator to create a deltaic complex of similar form and level of
detail, but the details are unlikely to match. The reason, of course, is that
the process simulator responds in terms of the processes represented.
Thus, users of sedimentary process simulators are faced with the
fact that the simulators cannot be conditioned to small-scale features
observed in wells or seismic data. We can affect the processes that are
SEDIMENTARY PROCESS SIMULATION 7

represented in the model adjusting its boundary conditions so that the

fluid and sediment discharge rates of inflowing rivers are changed, or sea-
level rises or falls, or the heights and attack angles of incoming waves are
changed, but we cannot command the processes to respond so that certain
depositional features are reproduced in detail. Some of these process
parameters are more or less stable whereas others need to be adjusted
because of their sensitivity in controlling grain-size distributions and
geometric forms of beds produced in experiments. By making judicious
adjustments, users can focus on critical ranges for initial water depths,
wave heights and directions, or fluvial discharge rates for water and
sediment, and test their sensitivities with regard to simulated stratigraphic
sequences.

Comparison of the Three Types of Simulators

In spite of their fundamental differences, the three types of

simulators intergrade insofar as their performance is concerned. For
example, although process simulators are totally deterministic in their
formulation, they usually seem to behave randomly because of complex
interactions between processes and materials, even though such apparent
random behavior is not explicitly generated. By contrast, stochastic
simulators yield random responses by design because of their dependence
on sources of random numbers. Paradoxically, numerical procedures for
generating sequences of random numbers (which are more properly
termed pseudorandom numbers) are actually deterministic, further blurring
the contrast between process simulators and stochastic simulators.
Stochastic simulators have an important advantage in that they can
represent uncertainty in making 3D projections of facies and petrophysical
properties that are expressed numerically. On the other hand, they have
a disadvantage in that they lack geological "insight" by contrast with
process simulators. Importantly, both stochastic simulators and process
simulators share the quality of irreversibility and are therefore known as
"forward" models. Their irreversibility sterns either from their explicit
random formulation or their representation of actual processes, both of
which are inherently irreversible.
Geometric-form simulators may produce geometrical forms of
sedimentary features similar to those of process simulators, but geometric-
form simulators deal only with the gross geometrical forms of
sedimentary bodies, and are not well suited for predicting petrophysical
8 WENDEBOURG AND HARBAUGH

properties because they are usually based on larger-scale seismic data

used for 2D representation.

smSIM AS A PROCESS SIMULATOR

The simulator described here for generating 3D sedimentary

sequences is named SEDSIM, an abbreviation for SEDimentary Sequence
SIMulation. SEDSIM represents most of the major processes that create
and modify sedimentary systems in which clastic sediment predominates
(Tetzlaff and Harbaugh, 1989; Martinez and Harbaugh, 1993; Wendebourg
and others, 1993). They include sediment erosion, transport, and
deposition as embodied in fluvial and wave processes, eustatic sealevel
change, isostatic compensation, tectonic uplift and subsidence,
compaction, and submarine slope failure (Table 1). In eroding,
transporting, and depositing clastic sediment in fluvial environments,
SEDSIM employs the Navier-Stokes equations for open-channel flow that
are solved using a Lagrangian discretization method (Tetzlaff and
Harbaugh, 1989). Flow is coupled with empirical sediment-transport
equations that control erosion, transport, and deposition of up to four
gram SIzes.
The processes represented in SEDSIM can be grouped in two
main categories, namely autocyclic processes and allocyclic processes.
Autocyclic processes are mutually interdependent upon each other and are
"internal" to SEDSIM. Autocyclic responses are inherent in virtually all
coupled systems, whether mechanical, biological, or geological. Basic
autocyclic processes in SEDSIM include flow in open channels, erosion,
transport, and deposition of clastic sediment, wave activity, and isostatic
subsidence. More complex autocyclic processes represented by SEDSIM
are derived from combinations of these basic processes, such as channel
avulsion, delta-lobe switching, and slumping.
Allocyclic processes, by contrast, are imposed externally, and
while they affect the autocyclic processes, they are not mutually affected
in turn. Allocyclic processes represented in SEDSIM include eustatic
sealevel change, tectonic subsidence, and climatic variations represented
as changes in external boundary conditions, such as change in wind
directions and intensity, and change in volume and geographic locations
of the fluid and sediment discharge volumes of inflowing streams (Fig.
1).
SEDIMENTARY PROCESS SIMULATION 9

Table 1. Overview of SEDSIM's modules listing their purposes, types of

input, and spans of time involved when each module is used.

SEDSIM PURPOSE INPUT FREQUENCY OF

MODULE ITERATIONS
NAME
FLOW Flow of surface water Start and end time, Seconds to minutes
and transport of tlow timestep,
sediment empirical transport
coefficients
SOURCE Discharge of water Source location, Days to years
and sediment volume of water and
sediment and grain
size distribution
through time

WAVE Redistribute sediment Main wind direction, Minutes - hours

by wave action transport rate, depth
of mobile bed

SEALEVEL Change eustatic Relative sealevel 102-10 3 years

sealevel curve

TECTONIC Vertical faulting Rate of elevation 102-10 3 years

change of basement
through time

SLOPE Redistribute sediment Maximum stable Days to years

by slumping slope per grain size

ISOSTA Subside basement due Flexural rigidity of 103-106 years

to sediment load lithosphere

COMPACT Reduce sediment Porosity-load 103-106 years

thickness due to function for each
compaction grain size

In SEDSIM, sedimentary sequences and other operations are

defined with respect to a gridwork of 3D cells that are indexed in the two
horizontal directions by rows and columns. Each cell is of fixed size
laterally during an experiment, with each cell's horizontal dimensions
being the same, thus forming a square. The vertical height of an
individual cell is variable and usually much less than that of the
horizontal dimensions. The number of rows and columns, and the number
of layers vertically may vary greatly, depending on the objectives of the
experiment and the level of detail desired. In typical studies, gridworks
range from 20 columns and rows and a few layers to more than 100 rows
and columns and more than 50 layers thus they may involve hundreds of
thousands of grid cells. Depositional sequences are stored in files as 3D
10 WENDEBOURG AND HARBAUGH

Discharge function

Initial lopography

a;
"
Sea level curve

/
SEDSIM

~
, Teclonic subsidence rales

~
~

'""'
Q)

Q)
.~
Ci! E
a;
a:
Time
A

Time- and IIlhostratigraphic record

Figure 1. Schematic diagram showing types of inputs for SEDSIM simulations,

clockwise from lower left: A, eustatic sealevel fluctuations over span of time represented
in experiment; B, initial topography of area of simulation; C, discharge rates for water
and clastic sediment over time span of experiment; D, frequency distribution of grain
sizes of clastic sediment supplied during experiment (they may represent geological
characteristics of watershed); E, orientation of faults and tectonic subsidence rates; F,
SEDSIM's output represented by lithostratigraphic record over time span represented in
experiment.

numerical records that provide both lithostratigraphic and

timestratigraphic descriptions that may be displayed in 3 D and in color
on modem graphics workstations (Ramshom and others, 1992).
SEDSIM has been used over a range of scales, with horizontal
resolutions ranging from several tens of meters to several hundreds of
meters. Vertical resolutions are usually much finer, ranging from a few
SEDIMENTARY PROCESS SIMULATION 11

centimeters to tens of meters. The selection of horizontal and vertical

scales and the number of cells employed depends on the objectives and
detail desired, including the amount of geologic time to be represented by
each sedimentary layer that is generated. The aggregate geologic time
represented in individual experiments may range widely, from several
years to several hundreds of thousands of years. Comparisons of the
geologic time represented in simulated sedimentary sequences with that
of actual sedimentary sequences is difficult. The time represented in
actual sequences may be substantially greater because SEDSIM represents
only those intervals required for individual depositional events, such as
the formation of channel deposits during floods, thereby ignoring intervals
between floods when little erosion or deposition occurs.
SEDSIM has been applied to modem fluvial deposits (Lee and
others, 1991), to beach environments (Martinez and Harbaugh, 1993), to
Quaternary alluvial and deltaic complexes in California (Koltermann and
Gorelick, 1992; Wendebourg, 1994), Wisconsin (Webb and Anderson,
1991), and Norway (Tuttle and others, 1995), to Cretaceous and Tertiary
Gulf Coast sediments (Tetzlaff and Harbaugh, 1989; Wendebourg, 1991;
Martinez and Harbaugh, 1993), and to Jurassic clastic hydrocarbon
reservoirs in the North Sea (Griffiths, 1994a, 1994b).

INITIAL AND BOUNDARY CONDmONS

REQUIRED FOR SEDSIM'S OPERAnON

In an experiment, SEDSIM must be provided with initial

conditions at the outset of the experiment, as well as with boundary
conditions that may vary over the span of geological time represented by
the experiment. The most important boundary conditions include the form
of the topographic surface at the outset of the experiment and variations
in the volumes of fluid and sediment volumes supplied by inflowing
streams during the span of the experiment.
Evidence for the importance of initial topography is provided by
ultra-high resolution seismic surveys involving young deltaic
environments. Chiocci (1994) describes Holocene deposits on the
Calabrian shelf of the Eastern Tyrrhenian Sea, Italy, where postglacial
sealevel rise and local tectonic uplift have influenced the deposition of·
deltaic and nearshore sediments whose geometric forms and facies
distribution have been strongly dependent on the underlying topography.
12 WENDEBOURG AND HARBAUGH

Here, the topography consisted of a narrow, steep shelf on which

depocenters marked the locations of river mouths, which moved landward
during transgression. In addition, scattered bioherms served as dams that
locally trapped suspended sediment.
An example of how a submerged topography has influenced
geometric form and grain-size distribution is provided by the Cretaceous
Woodbine Formation of East Texas (Wendebourg, 1994). Topographic
features, such as the Edwards reef trend and subtle undulations of the
submerged shelf, influenced the deposition of isolated lens-shaped bodies
of sediment that today serve as oil and gas reservoirs (Fig. 2). In
SEDSIM's simulation of these sediments, the Edwards reef trend served
as an important sedimentation boundary, influencing variations in bed
thickness and in the areal distribution of grain sizes (Fig. 3).
In applying SEDSIM to an actual area, initial topography for the
stratigraphic sequence being simulated can be determined from well data
(Koltermann and Gorelick, 1992) or 3D seismic data (Griffiths, 1994b).
Although the present structural form of a sedimentary sequence may be
defined in detail by seismic reflectors, changes that stem from
postdepositional compaction and structural deformation must be
subtracted if a reflecting horizon is to represent its topographic form
during deposition (Wendebourg, 1994).
An ancient topographic surface may be readily defined if it
involves the boundary between bedrock and overlying poorly consolidated
sediment. An example is provided by coarse-grained sediments deposited
as an ice-contact delta in the Holocene at Gardermoen in southern
Norway (Tuttle and others, 1995). About 9500 BC during an interval
spanning about 100 years, the retreating Scandinavian glacier dumped
about 4.4 km 3 sediments ranging in grain size from cobbles to silt,
thereby drowning an area of about 50 km2 whose original topography
involved differences in elevation of about 150 m (Fig. 4).
To yield meaningful results, SEDSIM must be provided with
appropriate rates for discharge volumes of fluid and sediment for each
inflowing stream over the time span represented in the experiment. These
discharge rates can be defined at boundaries either in form of point
sources such as rivers, or in form of line sources, as for example at
marine boundaries. Fluid discharge is coupled with sediment discharge,
usually expressed in form of suspended load within the water column and
as bed-load transported along the stream bed, beach, or ocean floor.
SEDIMENTARY PROCESS SIMULATION 13

10km
~ >15
>30m
m
<15 m

Figure 2. Structure contour map showing reconstructed ancient depositional topography

at base of Cretaceous Woodbine Formation in Polk and Tyler Counties, East Texas.
Map is based on three seismic lines whose traces are shown by solid lines. Contour
interval is 20 m. Plus signs near edges of overall map denote places where additional
data points were provided to constrain surface appropriately. Three shades of gray
represent isopachs of local variations in thickness of sandstones that serve as oil
reservoirs and are based on information from wells shown (Foss, 1979). Inset map
whose outline is represented by bold lines denotes area of SEDSIM simulation
experiment. Arrows denote main sediment transport directions and locations of sources
of fluid and sediment in experiment. Results of experiment are shown in Figure 3 (from
Wendebourg, 1994).
14 WENDEBOURG AND HARBAUGH

_____----;App(lJJ~. trace 01
/ EOYioillds fee4 "'WId

........
Thickness ]m]

Mean grain SlZ8 [mm]

.'.".."
l001S
~,r.()2'5 -
o 12..Q.11!I
0.5-10 009-t1_f2
0("" o to-O.09

A 10km B

Figure 3. Maps produced in SEDSIM simulation experiment involving deposition of

Woodbine sandstones that serve as oil reservoirs: A, isopachous map; B, and facies map
showing areal distribution of mean grain sizes. Area of experiment is denoted with bold
lines and labeled "SEDSIM Model" in Figure 2. Contour lines represent submerged
topography after deposition, with contour interval of 50 m. Thicker and coarser
sandstone deposits formed in simulation occur near Edwards reef trend, according with
actual thickness variations of Woodbine sandstones shown in Figure 2 (from
Wendebourg, 1994).

Although modern rates can be obtained from direct measurements of fluid

and sediment discharge rates, such rates are not directly available for
ancient deposits and must be determined indirectly.
Although there are several methods to determine paleodischarge
rates (Koltermann, 1993), one specifically developed and applied to
SEDSIM involves the "similar basin" approach (Koltermann and Gorelick,
1992) that assumes that variations between modern climatic zones can
serve as approximations for climatic changes in the past. Koltermann and
Gorelick compiled flow and sediment discharge data from modern
topographic basins in California's Coast Ranges and utilized flood
statistics to generate a climate history for the last 600,000 years for the
Alameda Creek drainage basin in Central California (Fig. 6). Flood
statistics in warm and dry interglacial periods of the past correspond with
flood statistics of topographic basins today in southern California,
whereas cooler and wetter glacial periods correspond with basins today
SEDIMENTARY PROCESS SIMULATION 15

2 km I

Figure 4. Map showing geographic features of ancient Gardennoen delta in southern

Norway represented in SEDSIM simulation experiment involving ice-contact delta
deposits. Streams emanating from glacier dumped sediment through five portals
indicated by arrows labeled as sediment sources. Contours show bedrock topography
based on borehole and seismic data that served as initial topography in simulation
experiment. Contour lines are in m above present sealevel with interval of 10m. Margin
of actual deltaplain deposits is shown with heavy dashed line, and margin of simulated
delta plain deposits is shown with heavy solid line. Light gray denotes area covered by
glacier during creation of delta, whereas medium gray denotes simulated deltaic
deposits. Hills not covered by sediment correspond with areas where bedrock crops out
at present. Modified from Tuttle and others (1995).

in northern California. Such comparisons between present and past are

feasible because of geomorphic similarities between the different basins
of the Coast Ranges.
16 WENDEBOURG AND HARBAUGH

COMPARING RESULTS OF SEDIMENTARY SIMULATORS

WITH ACTUAL DATA

The most challenging aspect of using sedimentary process

simulators involves comparing their results with actual data. Because we
cannot regulate process simulators so that their results accord closely with
the details provided by wells, seismic sections, or outcrops, we must
judge whether the responses are appropriate by making indirect
comparisons. Graphic responses obtained in simulations can be especially
adapted to facilitate comparison, such as generating simulated logs of
grain-size variations at locations where wells occur (Fig. 5). Comparisons
are influenced by the limited vertical resolution of a simulation and the
number of discrete grain sizes used in simulations. In SEDSIM only four
grain sizes are employed, so the degree of concordance needs to be
eval uated accordingly.
Thus far, assessing the degree of accord has largely involved
visual comparisons of maps and sections and logs, and is influenced by
the level of detail represented in simulated sections and maps, which in
turn is limited by the grid resolution both laterally and vertically. Details
observed in sections based on outcrops or correlations between closely
spaced wells are likely to be too fine to be represented in sections
generated in simulations. For example, small channels or crevasse splays
seen in the actual section shown in Figure 7A cannot be seen in the
simulated section (Fig. 7B) because the actual processes that generated
these details have not been simulated.
Although direct visual comparisons are straightforward, they do
not enable us to measure the degree to which a simulation's results accord
with actual data. Quantitative comparison can be based on statistical
descriptions of simulation results (Tuttle and others, 1995). Furthermore,
statistical descriptions can be used as input to stochastic simulators either
directly in form of stationary parameters such as means and variogram
statistics, or as nonstationary parameters involving trends of geometric
forms and spatial distribution of facies (Beucher and others, 1992).
When comparing actual and simulated sequences, we can also
make use of other relationships. In both simulated and actual sequences,
subsurface fluid flow is constrained by the spatial distribution of
petrophysical properties. In actual sequences these properties may be
estimated from production tests, core analyses, and petrophysical logs
SEDIMENTARY PROCESS SIMULATION 17

Cooler, 65
Wetter

Mean Daily
Discharge
(m 3/s)

Warmer, K Paleoclimate Curve

Drier 0
A
1200

Mean Rood
Discharge
(m3/s)

600

o B
10' ~-----.-----.------.-----~------r-----.
Suspended Load

105
Sediment
Discharge
(ml/yr) 10.

Bedload

200 400 600

Thousand Years Before Present

Figure 5. Graphs showing variations: A, in climate; B, fluid discharge rates; and C,

sediment discharge rates for alluvial fan deposits formed by Alameda Creek (San
Francisco Bay region, California) during last 600,000 years. Discharge rates for fluid
and sediment shown in graphs Band C were used as input to SEDSIM simulation
experiments whose results are shown in Figure 7B. Note that variations in both
discharge rates include overall cyclic variations that reflect major changes in climate and
smaller variations or "noise." representing stochastic variations in flood frequencies
(from Koltermann and Gorelick, 1992, reproduced with permission of the American
Association of the Advancement of Science).
18 WENDEBOURG AND HARBAUGH

Qj Qj
> .. 0
..!!! .. 0 § 200
C!I .. 0
QI ItJ ~
VI VI
QI
[l QI
>
o >
.g 200
.8
C!I
190

E E
.: .:
c: c:
o .2 180
'';::;
n;
~
C!I
o .. >
QI
iii iii
190 -+-.......-,.......-,.......---...---...-- 170 -+----r..,....---.--r--r--r-----r---.
'"
ci
0 g '"o
ci ci
-
0
ci
ci
'" . 0

Median grain size diameter in mm Median grain size diameter in mm

o Simulated grain size

.. Measured grain size

Figure 6. Example plots of two locations in Gardermoen delta of

Figure 4 comparing variations in median grain size between actual and
simulated sequences with respect to elevation. Note limited variability
of simulated grain sizes because only four grain sizes are used in
SEDSIM, as well as limited vertical resolution of simulated layers
(from Tuttle and others, 1995)

whereas in simulated sequences, these properties can be derived from

depositional processes in the form of grain-size distributions.
Two examples involving subsurface fluid flow can be cited: (1)
Petrophysical properties derived from process simulation and used with
groundwater flow simulations of the aquifer shown in Figure 7B have
yielded in spatial distributions of variations in hydraulic head that
compare favorably with actual data derived from water wells
(Koltermann, 1993). (2) In an application involving oil migrating into
reservoir beds at South Belridge field of the San Joaquin Valley of
California, simulations helped determine the degree to which lithologic
heterogeneities influenced migration, where facies interpretations,
petrophysical logs, and oil saturations from cores of producing wells
permitted simulations of depositional processes to be coupled with
subsequent simulations of subsurface fluid flow (Wendebourg, 1994).
Details of a permeability-grain size function for South Belridge are
shown in Figure 8, where a log-linear function has been fitted to scattered
SEDIMENTARY PROCESS SIMULATION 19

North South

8
5;;~~~L ,.... Skm

Figure 7. Cross sections showing: A, actual deposits interpreted from information from
wells; and B, simulated deposits that form Alameda Creek alluvial fan. Length of
section is 16 Ian and thickness of deposits in center is about 235 m. Gray shades denote
variations in mean grain sizes, dark gray denoting coarse sediment and light gray
denoting fine sediment. Ticks in upper cross section (A) indicate position of wells that
provided information on which cross section is based. Note smooth representation of
grain size variations in simulated cross section (B) as compared with complex detail of
actual deposits in cross section (A) (from Koltermann and Gorelick, 1992, reproduced
with permission of the American Association of the Advancement of Science)

data points for subsequent use in simulating flow. Figure 9 compares oil
saturations with respect to permeability for measurements from cores
versus those generated using a two-phase reservoir infilling simulator
(Wendebourg, 1994). Although sediments with very high and very low
permeabilities accord relatively well, the spread of simulated saturations
for sediments with intermediate permeabilities is greater than those
20 WENDEBOURG AND HARBAUGH

10000
i 0
0-
0
I
I 'U
•
61,,,"
~
~<:tI
"
1
0
0 " c 1" 0
p "_: a a 00 o~

-E
a 0 "00...0 11100 O.n tI
1000 0-
--.0 110

-
'0
- .....
n
~
00 'C 0
. u
d'o
0 o•
oi -
~

"Q, a

In
a
>-
.-::
i!I", a rfl
CIJ
0
U
00 a
a
§~

&.
100
:.actI
-
D-

.•.
Q)
- ..!!. I
~

E
I-
_Q-' 0
I'"
u U o

Q)
c.. ~ ~ 0
10
;
I

,~
I
cO

o 20 40 60 80

Hydrocarbon saturation in percent

Figure 8. Permeability versus mean grain size in reservoir beds in South Belridge oil
field. Larger solid circles denote averages based on measurements in cores (Miller and
others, 1990) whereas small squares denote data values used in SEDSIM simulation
experiment (from Wendebourg, 1994).

measured in cores. One explanation is that the representation of the

reservoir in the simulation does not sufficiently reflect the true
sedimentary heterogeneity and therefore grain sizes resulting from the
simulator are too homogeneous and thereby overestimate the actual
permeabilities. More realistic permeabilities would shift simulated
permeability values downward and therefore improve the degree of
accordance. Also it is possible that the simulation may not have reached
an equilibrium state in its representation of the reservoir, as compared
with present conditions in the actual reservoir (Wendebourg, 1994).
SEDIMENTARY PROCESS SIMULATION 21

1000
il L I I

-
-
j ~ i!
"C IL l !!
E l~ 11 II
100
>-
:!::
.c
1 j

I ! jj
ctS
G.I II ii
! i
E
10.
10
G.I
a. ,;
I
I
!
:
!
: I

I, !!
1
I

1
i IJ
f f 1i
I I

0.1 L---.l--L-l......L!....!.
11...Ll...l...-----l---'-......:.....L-1J.I. . L..!.,;.!..!..-
I ----!-----L.....J........l.."--"--'-W
0.001 0.01 0.1 1.0

Mean grain size (mm)

Figure 9. Permeability versus oil saturation in reservoir beds in South Belridge field
involving reservoir infilling study conducted with SEDSIM. Solid circles denote
measurements based on cores (Miller and others, 1990) whereas squares represent values
obtained at individual grid cells generated by SEDSIM and subsequently used in
simulaton of subsurface multiphase flow. Note that variations in simulated saturations
in cells with high permeabilities are similar to those measured in cores, whereas
variations in simulated saturations in grid cells with intermediate permeabilities are
higher than those in cores (from Wendebourg, 1994).

CONCLUSIONS

Depositional process simulation is a procedure that can provide

3D descriptions of aquifers and oil and gas reservoirs in terms of the
events that formed them. Because process simulators are irreversible and
involve working forward through geologic time, it is difficult to constrain
them directly to observations. Furthermore, they are regulated by initial
and time-dependent boundary conditions involving rates of processes that
22 WENDEBOURG AND HARBAUGH

must be defined a priori. In spite of these requirements, process

simulators have important advantages in that they yield the 3D geometric
forms of sedimentary deposits and their lithologic properties
simultaneously. By combining depositional process simulators with
subsurface fluid-flow simulations, reservoir and aquifer flow models can
be obtained that are concerned with the effects of sedimentary
heterogeneities on flow. Flow simulations also provide additional ways
of calibrating depositional simulations, but they also increase the number
of alternatives that can be obtained as responses.
In addition to direct visual comparisons between simulated and
actual sequences, it is appropriate to use statistical descriptions of
simulated sedimentary sequences and flow properties derived from them.
Process simulations may be used to generate inputs for geostatistical
simulators, either directly as descriptions of the spatial distribution of
grain sizes, or in form of trends in the spatial distribution of facies. The
latter is possible because depositional process simulators store histories
of events and can represent facies trends simultaneously through space
and time, properties of fundamental importance in geology. Finally,
interpretations of simulation results require careful consideration of
assumptions and boundary conditions that control the simulator, based on
an understanding of the regional and local geological framework of the
area and sequence of beds represented in the simulation.

ACKNOWLEDGMENTS

We thank our colleagues of Stanford's SEDSIM project who

contributed to this paper, including Paul Martinez, Young-Hoon Lee,
Christoph Ramshorn, Christine Koltermann, and Don Miller. In addition,
Kevin Tuttle of the University of Oslo and Cedric Griffiths of the
University of Adelaide contributed to our work. Finally, we thank
Philippe Joseph of the French Petroleum Institute (IFP) for reviewing the
manuscript.
SEDIMENTARY PROCESS SIMULATION 23

REFERENCFS

Beucher, H., Galli, A, LeLoch, A, Ravenne, C., and Heresim Group,

1992, Including a regional trend in reservoir modeling using the
truncated Gaussian method, in Soares, A, ed., Geostatistics, Troia
1992, Vol. 2: Kluwer Academic Publ., Dordrecht, p. 555-566.
Chiocci, F.L., 1994, Very high-resolution seismics as a tool for sequence
stratigraphy applied to outcrop scale - examples from Eastern
Tyrrhenian margin HolocenelPleistocene deposits: Am. Assoc.
Petroleum Geologists Bull., v. 78, no. 3, p. 378-395.
Deutsch, C.Y., and Journel, AG., 1992, GSLib - Geostatistical Software
Users Manual: Oxford University Press, New York, 340 p.
Doyen, P.M., Guidish, TM., and de Buyl, M., 1991, Lithology prediction
from seismic data, a Monte-Carlo approach, in Lake, L.W.,
Carroll, H.B., and Wesson, TC., eds., Reservoir characterization
II: Academic Press, London, p. 557-564.
Foss, D.C., 1979, Depositional environment of Woodbine sandstones,
Polk County, Texas: Gulf Coast Assoc. Geol. Societies Trans., v.
29, p. 83-94.
Geehan, G.W., Lawton, TF., Sakurai, S., Klob, H., Clifton, TR.,
Inman, K.F., and Nitzberg, K.E., 1986, Geologic prediction of
shale continuity, Prudhoe Bay field, in Lake, L.W., and Carroll,
H.B., eds, Reservoir characterization: Academic Press, London,
p. 63-82.
Griffiths, c., 1994a, SEDSIM's application in simulating the Brent
lower-shoreface Rannoch Formation of the North Sea's Viking
graben: Proc. SEDSIM Affiliates Meeting, Stanford University,
Stanford, California, 27 p.
Griffiths, C., 1994b, SEDSIM's application in simulating a mid-Jurassic
North Sea submarine apron fan for prospect risk analysis: Proc.
SEDSIM Affiliates Meeting, Stanford University, Stanford,
California, 21 p.
Haldorsen, H.H., and Damsleth, E., 1993, Challenges in reservoir
characterization: Am. Assoc. Petroleum Geologists Bull., v. 77,
no. 4, p. 541-551.
Harbaugh, J.W., and Bonham-Carter, G., 1970, Computer simulation in
geology: Wiley-Interscience, New York, 575 p.
Isaaks, E.H., and Srivastava, R.M., 1989, An introduction to applied
geostatistics: Oxford University Press, New York, 561 p.
24 WENDEBOURG AND HARBAUGH

Koltermann, C.E., and Gorelick, S.M., 1992, Paleoclimatic signature

in terrestrial flood deposits: Science, v. 256, no. 5065, p. 1775-
1782.
Koltermann, C.E., 1993, Geologic modeling of spatial variability in
sedimentary environments for groundwater transport simulations:
unpubl. doctoral dissertation, Stanford University, 250 p.
Lawrence, D., Doyle, M., and Aigner, T., 1990, Stratigraphic simulation
of sedimentary basins: concepts and calibration: Am. Assoc.
Petroleum Geologists Bull., v. 74, no. 3, p. 273-295.
Lee, Y.H., Haskell, N.L., Butler, M.L., and Harbaugh, IW., 1991,
3-D modeling of Arkansas river sedimentation (abst.): Am. Assoc.
Petroleum Geologists Bull., v. 75, no. 3, p. 619.
Lessenger, M., 1993, Forward and inverse simulation models of stratal
architecture and facies distributions in marine shelf to coastal plain
environments: unpubl. doctoral dissertation, Colorado School of
Mines, 182 p.
Martinez, P.A., and Harbaugh, IW., 1993, Simulating nearshore
environments: Pergamon Press, Oxford, 265 p.
Miller, D.D., McPherson, IG., and Covington, T., 1990, Fluviodeltaic
reservoir, South Belridge field, San Joaquin Valley, California, in
Barwis, IH., McPherson, IG., and Studlick, IR., eds., Sandstone
petroleum reservoirs: Springer, New York, p. 109-130
Murray, C.I, 1992, Geostatistical applications in petroleum geology
and sedimentary geology: unpubl. doctoral dissertation, Stanford
University, 230 p.
Ramshorn, C., Ottolini, R., and Klein, H., 1992, Interactive three-
dimensional display of simulated sedimentary basins (abst.):
Eurographics Workshop on Visualization in Scientific Computing
(Clamart, France), 1 p.
Rivenaes, IC., 1992, Application of a dual-lithology, depth-dependent
diffusion equation in stratigraphic simulation: Basin Research v.
4, no. 2, p. 133-146.
Rudkiewicz, IL., Guerillot, D., and Galli, A., 1990, An integrated
software for stochastic modelling of reservoir lithology and
property with an example from the Yorkshire Middle Jurassic, in
Buller, A.T., Berg, E., Hjelmeland, 0., Kleppe, I, Torsaeter, 0.,
and Aalsen, 10., eds., North Sea oil and gas reservoirs II:
Graham and Trotman, London, p. 400-406.
SEDIMENTARY PROCESS SIMULATION 2S

Shuster, M.W., and Aigner, T., 1994, Two-dimensional synthetic

seismic and log cross sections from stratigraphic forward models:
Am. Assoc. Petroleum Geologists Bull., v. 78, no. 3, p. 409-43l.
Slingerland, R., Harbaugh, J.W., and Furlong, K., 1994, Simulating
clastic sedimentary basins: Prentice Hall, Englewood Cliffs, New
Jersey, 220 p.
Strobel, J., Soewito, F., Kendall, C.G., Biswas, G., Bezedk, J., and
Cannon, R., 1989, Interactive (SEDPAK) simulation of clastic and
carbonate sediments in shelf to basin settings: Computers &
Geosciences, v. 15, no. 8, p. 1279-1290.
Tetzlaff, M., and Harbaugh, J.W., 1989, Simulating clastic sedimentation:
Van Nostrand Reinhold, New York, 202 p.
Tuttle, K., Wendebourg, J., and Aagaard, P., 1995, A process-based
depositional simulation of the coarse-grained Gardermoen delta,
Norway (abst.): Geol. Soc. London Conf. on Quantification and
Modeling of Spatial Patterns in Permeable Rocks (Scarborough,
England), 1 p.
Webb, E., and Anderson, M.P., 1991, Simulating the grain-size
distribution ofWisconsian age glaciofluvial sediments: application
to fluid transport (abst.): Am. Assoc. Petroleum Geologists Bull.,
v. 75, no., 3, p. 689.
Wendebourg, J. 1991, 3-D modeling of compaction and fluid flow of
the Woodbine progradational sequence, Polk County, Texas (abst):
Am. Assoc. Petroleum Geologists Bull., v. 75, no. 3, p. 69l.
Wendebourg, J., Ramshorn, C., and Martinez, P.A., 1993, SEDSIM3.0
User's Manual, Proc. SEDSIM Affiliates Meeting, Stanford
University, Stanford, California, 73 p.
Wendebourg, J., 1994, Simulating hydrocarbon migration and
stratigraphic traps: unpubl. doctoral disseration, Stanford
University, 258 p.
 MODELING OF MULTICOMPONENT DIAGENETIC
SYSTEMS

Robert Ondrak
GeoForschungsZentrum Potsdam, Potsdam, Gennany

A BSlRACf

A computer model is presented, which permits the modeling of

large-scale diagenetic evolution of sandstones resulting from mineral dis-
solution and precipitation. The model includes various mineralogical
components and dissolved species which are major components of many
sandstones. The model allows the simulation of temporal and spatial
evolution of porosity and mineral distribution in a sandstone layer. Pos-
sible application to diagenetic problems is illustrated by several theoreti-
cal examples, representing different geological environments.

INlRODUCflON

Modeling diagenesis is a complex task because various processes

may interact. In principle, however, two mechanisms determine the dia-
genetic evolution of rocks: mechanical compaction and chemical diage-
nesis. Several models have been published during the last decade, con-
cerned with the numerical modeling of sandstone diagenesis (e.g. Wood
and Hewett, 1982; Leder and Park, 1986; Chen and others, 1990; Nagy
and others, 1990; Gouze and Coudrain-Ribstein, 1993). The models
steadily increased in complexity from simple a-dimensional quartz models

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 27
28 ONDRAK

(e.g. Leder and Park, 1986) to complex, multidimensional reaction-

transport models (Chen and others, 1990). The model presented here
focuses on regional diagenetic alterations, which are the result of the
interaction of mineral reactions and fluid flow in sandstones. In a flow-
driven diagenetic environment, the regional distribution of the flow sys-
tem usually is inhomogeneous, varying in space as a function of the local
porosity and permeability distribution. In addition, mineral dissolution or
precipitation changes porosity and permeability with time, which, in tum,
modify the flow system. The interaction between the flow system and the
diagenetic alteration of the rock is illustrated by computer simulations.

mEORETICAL BACKGROUND OF mE MODEL

Modeling chemical diagenesis requires the numerical simulation of

water-rock interactions. Depending on the mineral composition of the
rock, a large number of minerals and dissolved species can be involved.
The attempt to consider all possible components provides a model too
complex to be handled numerically. The present example, therefore, is
limited to the dominant constituents of many sandstones: quartz, calcite,
albite, and kaolinite as mineral species. For each mineral a chemical re-
action has to be defined describing the interaction between the pore fluid,
the mineral, and the resulting dissolved species [Reaction (1 )-(4)].

Si0 2 + 2H2O <=> H 4Si0 4 (1)

CaCO) + H 2CO) <=> 2HCO)- + Ca2+ (2)
NaAISi)08+4H+ + 4HzO <=> Na++AP+ + 3H4Si0 4 (3)
A1 2SizOs( OH)4+6H+ <=> 2AP+ + 2H4Si0 4 + H 2O (4)

The interaction between a gaseous phase (C0 2) and the pore fluid
also causes the formation of a dissolved species [Reaction (5)].

(5)

The reactions describe the mineral-fluid interaction, or in general

the phase transition between the solid, the fluid, and the gas phase. In
addition, a complete chemical description of the system requires the inter-
MODELING OF DIAGENETIC SYSTEMS 29

action of the dissolved species and the ions in the pore fluid [Reaction
(7)-(8)]. In the presence of carbonic acid these are:

H2CO) <=> HCO)· + H+ (6)

HCO)· <=> CO/· + H+ (7)
H4Si0 4 <=> H)Si0 4• + H+ (8)
H2O <=> OR + H+ (9)

For each chemical reaction an equation can be formulated describ-

ing its equilibrium condition. The stability constants of the reactions
differ nonlinearly as a function of temperature and pressure. Figure 1
shows the temperature dependence of the different stability constants
along the vapor pressure line illustrating that temperature is one of the
major controlling factors for water-rock interaction processes.

o 50 100 150 200

Tempera1Lfe in·C

Reaction (1) Reaction (2) Reaction (3) Reaction (4) Reaction (5)
-?t<-
Reaction (6)
-+- -+-
----
Reaction (7)
-+-
Reaction (8)
--
Reaction (9)
~

Figure 1. Temperature dependence of stability constants of reaction (1)-(9).

30 ONDRAK

Lastly, it has to be considered whether the model should be regard-

ed as an open or a closed system with respect to certain components. In
general, the model describes an open system into which species can be
imported or exported by fluid flow. But water-rock and gas-water inter-
actions are phase transitions which only take place if the specific phase
is present. Therefore, the system may become closed with respect to a
particular component if this component is removed from the system re-
sulting from dissolution. The diagenetic model must handle this modifi-
cation of the physical system and it must ensure a correct mass balance
for the entire system. The correct mass balance is achieved by logical
separation of phase equilibria and phase transitions, which permits to
model the time dependent diagenetic evolution of rocks (Ondrak, 1992,
1993; Ondrak and Bayer, 1993).
The dissolution and precipitation of minerals not only alters the
mineralogical composition of the rock, but also the petrophysical proper-
ties, for example conductivity and permeability of the rock. The modifi-
cation of the physical properties changes the flow regime of the pore
fluid, which determines the transport rates of the dissolved species.
Therefore, the flow system and the diagenetic evolution of the rock in-
teract. To study the coupling between the two systems the chemical re-
action model is combined with a two-dimensional flow model in which
the flow rates vary spatially and temporally as a function of the evolving
physical properties. The combination of the two modules allows to sim-
ulate the interaction of fluid flow, chemical evolution of the pore fluid,
and mineral reactions with respect to the regional diagenetic evolution of
a sandstone.

MODEL CALCULATIONS

The following section outlines the possible application of the model

to an idealized sandstone. The following schematic geological scenario
is assumed for the simulations. The thickness of the sandstone layer is
small with respect to its lateral extension. Therefore, its thickness can be
reduced to a two-dimensional surface (Fig. 2). The sandstone is bounded
by almost impermeable claystone layers from above and below. Pore
fluid, therefore, can only flow within the surface. Within the model a
pressure gradient is defined between the left and right border of the layer
MODELING OF DIAGENETIC SYSTEMS 31

causing a fluid flow from left to right (Fig. 2), whereas the lateral
boundaries are sealed; the layer is 40 km long, 30 km wide, and is locat-
ed at a depth of 3 to 2 km. Temperature along the layer changes from
110 °C at the deepest position to 65 °C at the highest. Both temperature
and burial depth remain constant throughout the calculations. Permeabil-
ity and porosity of the layer differs spatially, representing the heteroge-
neity or'natural rocks, whereas the mineralogical composition is consid-
ered to be more uniform but changes from one example to the other. The
three layers in Figures 3-10 represent the same layer at different time
slices: at the beginning of the calculations (0 my), in the middle (5 or 2.5
my), and at the end (10 or 5 my).

Sediment Layer

Figure 2. Schematic geological model for simulations. Pore water flow through
sandstone layer is from left to right. For model, sandstone body is represented by
2D surface.

Example 1

In the first example the diagenetic evolution of a calcite cemented

quartz sandstone is considered. The initial flux ranges between 15 and
20 m/y depending on the local permeability of the layer. The fluid en-
tering the layer from the left side is in chemical equilibrium with respect
to temperature, quartz, calcite, and a low CO 2-concentration. The condi-
tions remain constant throughout the simulated time period. Within the
layer, the chemical system is considered a closed system with respect to
32 ONDRAK

CO 2• The amount of carbonic acid, therefore, can only be modified by

calcite dissolution or precipitation [cf. Reaction (2)]. As the fluid flows
updip along the layer it cools which changes the chemical equilibria of
the dissolved species. The equilibrium concentration of quartz decreases
with declining temperature (Fig. 1) causing the precipitation of silica
cement. For the carbonate equilibrium the opposite trend is the situation.
Calcite solubility increases with declining temperature. Therefore, calcite
cement is dissolved continuously by the upward moving pore water. The
effects on porosity are illustrated in Figure 3, which shows the initial
porosity distribution and its temporal evolution after 5 and 10 million
years. Most apparent is the overall porosity increase during the simulated
time period. Near the left and right boundary of the model this trend is
reversed towards the end of the simulations. For a detailed consideration
of the interacting processes, the temporal evolution of the different
mineral species is illustrated in Figures 4 and S.
Figure 4 shows the changes of calcite cement resulting from disso-
lution and fluid flow. The moving fluid dissolves the pore filling cement
because of its temperature dependent leaching capacity. Therefore, disso-
lution proceeds not homogeneously throughout the layer, but is most
prominent where the highest temperature variations occur along the flow
path. This is the situation, in particular near the right and left boundaries,
because of the increased inclination of the layer. The calcite dissolution
capacity is not only temperature controlled, but also depends on the
chemical composition of the pore fluid. This effect can be seen in the
central part of the layer, where calcite dissolution proceeds slowly
because of the small temperature variations. In this part of the sandstone,
calcite removal becomes important towards the end of the simulated time
period. In this final phase, the fluid increasingly is undersaturated with
respect to calcite because of the complete removal of calcite cement
farther downstream and the dissolution proceeds into the central region
of the layer following zones of high flow rates. Tn addition, the calcite
dissolution increases the permeability of the layer, thereby, increasing
flow rates and mass transport. The higher mass transport, in tum,
enhances the dissolution process causing an interaction of fluid flow and
mineral reactions.
MODELING OF DIAGENETIC SYSTEMS 33

% porosity

Figure 3. Porosity evolution of artificial calcite cemented sandstone for time period
of 10 my. Bottom layer shows porosity distribution at beginning of simulation,
whereas middle and top layer show porosity distribution after simulated time period
of 5 and 10 my. Flow rates range between 15-20 m/y, and CO 2-concentration is low.

% cement
11.29

10.16

9.03

7.90

6.18

5.65

4.52

3.39

2.26

1.13

0.00

Figure 4. Temporal and spatial evolution of silica cement distribution of sandstone

layer shown in Figure 3.
34 ONDRAK

The interaction between fluid flow and diagenesis also can be illus-
trated by the development of the silica cement (Fig. 5). The large amount
of newly formed silica cement in the left part of the layer, which corre-
sponds to the zone of early calcite cement removal, is interesting. Here,
silica precipitation is more effective in destroying porosity than is gained
by calcite dissolution. The timing of increased silica precipitation (Fig. 5)
coincides with the complete removal of the calcite cement (Fig. 4). The
evolution of the calcite and the silica cement also is reflected in the
porosity development (Fig. 3). In Figure 3 an initial porosity increase can
be seen in this part of the layer, which is destroyed again towards the end
of the simulated time period. The initial porosity increase is followed by
porosity destruction, which indicates that the enhanced flow rates
resulting from the permeability increase by calcite dissolution promoted
the strong silica precipitation. The spatial variations in silica precipitation
are connected to mass transport variations and varying temperature
changes along the flow path. The results illustrate the interaction
between fluid flow and diagenetic evolution of sandstones in a flow-
dominated system. High-flow rates enhance the mass transport of
dissolved components. The increased mass transport, in tum, promotes
the dissolution or precipitation of minerals depending on the surrounding
conditions. The mineral reactions alter the permeability of the rock
which controls mass transport. The interaction and feed back of the
processes results in interesting diagenetic patterns even in the situation of
a relatively simple calcite cemented sandstone.

Example 2

The following example shows the same sandstone layer with the
difference that the CO 2 concentration is two orders of magnitudes higher
than in the previous example. The setting could be compared to an envi-
ronment with active CO 2 production resulting from the maturation of or-
ganic matter or with an area where juvenile CO 2 occurs. In the present
example, flow rates were reduced by approximately one order of magni-
tude having values of 1-2 mfy. The diagenetic evolution of the sandstone
layer differs from the previous example because of the modified boundary
condition. Again, an overall porosity increase can be observed (Fig. 6),
which correlates with zones of high-flow rates.
MODELING OF DIAGENETIC SYSTEMS 35

0/0 cement
10.0

Figure 5. Temporal and spatial evolution of calcite cement distribution of sandstone

layer shown in Figure 3.

With respect to the previous example, the main difference is the

lacking porosity decrease in the left part of the layer towards the end of
the simulated time period. Because of the low-flow rates of 1-5 m/y,
silica cementation is too small to compensate the porosity increase caused
by calcite dissolution. The diagenetic evolution is dominated by calcite
dissolution, which is increased strongly by the high CO 2 concentration of
the pore fluid. Figure 7 shows the temporal evolution of the calcite
cement distribution corresponding to the porosity evolution of Figure 6.
In general, the picture is comparable to the previous example with the
main difference that changes occur at smaller flow rates than before. The
distribution of the silica cement is not illustrated because the amount of
newly generated silica cement is negligible. Silica precipitation reaches
a maximum of 1% in some areas, which has little effect on the overall
picture.
Example 2 illustrates that the diagenetic evolution of a sandstone is
not controlled simply by temperature changes and mass transport but by
all factors which alter the chemical composition of the pore fluid. In the
present model, the high CO 2 concentration of the pore fluid is the domi-
nant feature which enhances calcite dissolution. In the following exam-
ple, the effect of feldspar dissolution on the model is studied as an other
factor influencing the diagenetic evolution of the model sandstone.
36 ONDRAK

% porosity

Figure 6. Porosity evolution of artificial calcite cemented sandstone for time period of
10 my. Bottom layer shows porosity distribution at beginning of simulation, whereas
middle and top layer show porosity distribution after simulated time period of 5 and 10
my. Flow rates range between 1-2 m/y and CO 2 concentration is high.

% cement
10.00

Figure 7. Temporal and spatial evolution of calcite cement distribution of sandstone

layer shown in Figure 6.
MODELING OF DIAGENETIC SYSTEMS 37

Example 3

In the final example, the mineralogical composition of the sandstone

layer differs from the previous ones, because albite is distributed homoge-
neously in the sandstone, in addition to calcite and quartz. Furthermore,
it is postulated that the dissolution of albite will result in the formation
of kaolinite as the new aluminium-bearing mineral phase. The fluid en-
tering the layer is in equilibrium with calcite and with a relatively high
CO 2 concentration but not with respect to albite. Figure 8 illustrates the
temporal and spatial porosity evolution of the layer. The most prominent
feature of this example is an overall porosity decrease of the sandstone.
It is most pronounced along the zones of high initial porosity, and moves
as a cementation front along the flow direction through the layer. In
order to understand the processes causing this porosity evolution one has
to look at the temporal and spatial distribution of the different mineral
phases.
Figure 9 illustrates the temporal evolution of the albite distribution
in the layer. It can be seen that albite is dissolved by the entering pore
fluid which, although in chemical equilibrium with calcite, contains
enough acidity to dissolve albite. A relatively sharp reaction front de-
velops breaking up into different fingers as a result of the varying flow

% porosity

Isolines: Temperature

TIME

Figure 8. Porosity evolution of artificial sandstone for time period of 5 my. Bottom
layer shows porosity distribution at beginning of simulation, whereas middle and top
layer show porosity distribution after simulated time period of 2.5 and 5 my.
38 ONDRAK

% cement
10.0

Isolines: Temperature

TIME

Figure 9. Temporal and spatial evolution of albite distribution of sandstone layer

shown in Figure 8.

rates in the sandstone. At the dissolution front the leaching capacity of

the pore fluid is consumed and, therefore, feldspar cannot be dissolved
anymore. In the model, dissolution of albite buffers the fluid and inhibits
mineral reactions farther down stream of the reaction front. This is a
simplification, which neglects the small alterations resulting from tem-
perature variations along the flow path. However, these changes are too
small to be noticed in the figures.
The dissolution of albite causes a shift in pH towards an alkalic en-
vironment making the fluid supersaturated with respect to calcite which
then starts to precipitate (Fig. 10). Similar observations of feldspar disso-
lution in the presence of calcite cement have been described by Siebert
and others (1984) from investigations of thin sections. But the precipita-
tion of calcite stops as soon as the albite has disappeared. Calcite starts
to dissolve again, because of the increasing dissolution capacity of the
cooling pore fluid.
According to Reaction (3) the dissolution of albite increases alu-
minium and silica concentration in the pore fluid. Following Reactions
(1) and (4) an increased aluminium and silica concentration will result in
MODELING OF DIAGENETIC SYSTEMS 39

% cement

[solines: Temperature

TIME

Figure 10. Temporal and spatial evolution of calcite cement of sandstone layer
shown in Figure 8.

the formation of silica and kaolinite cement. The precipitation of these

cements cause the overall porosity reduction shown in Figure 8. On the
scale of the model the major amount of silica cement can be contributed
to the dissolution of the feldspar present in the rock, whereas mass trans-
port and temperature related changes are negligible. However, it must be
kept in mind that this result is valid only for the special assumption of the
model. For other geological settings the results may look different as
shown by Examples 1 and 2.

CONCLUSIONS

The previous examples illustrated the interaction of mineral phases

with the pore fluid and the flow system under various boundary condi-
tions. The results should not be seen purely from their quantitative value,
but also their qualitative aspect should be considered. A major problem
for quantitative modeling is the quality of the thermodynamic data on
which the calculations rely. Already little uncertainties in the data can
modify drastically the results of such highly nonlinear dynamic system.
In the present model the acidity of the pore fluid is related closely to the
carbon dioxide content of the fluid. A main source of CO 2 is the matura-
40 ONDRAK

tion of organic material (Siebert and others, 1984). New models in or-
ganic geochemistry (e.g. Burnham and Sweeney, 1989) allow to quantify
the amount and the time of CO2 production. The next step to better un-
derstand the interaction of organic and inorganic diagenesis could be ob-
tained by the combination of diagenetic modeling and maturation models.
Another aspect which becomes interesting in systems with high-flow rates
is the kinetics of the reactions. This point is not considered in the model
because the flow rates are considered to be small enough for chemical
equilibrium to be established. This consideration obviously must be
modified when flow rates are increased. The model, therefore, cannot be
applied to environments with flow rates of meters per day, where the
equilibrium assumption is no longer justified. Therefore, the present
model is limited to flow rates of centimeters to meters per year, which al-
low the establishment of chemical equilibrium (Ondrak, 1993). Within
this framework it can be used as a tool to test diagenetic concepts derived
from petrologic studies against chemical and physical first principles. But
it should be kept in mind that the model is based on a number of assump-
tions and simplifications, which may have to be rethought if the
simulation results do not provide plausible results with regard to
geological and mineralogical observations. For this reason, the present
model can be only one more step towards a better understanding of the
complexity of diagenetic processes. It will evolve further as the
comprehension of the processes grows. However, it provides a tool to
obtain a better understanding of the complex geological system.

ACKNOWLEDGMENTS

Thanks are due to U. Bayer, T. McCann, and D.F. Merriam for

critical reading of the manuscript.

REFERENCES

Burnham, A.K., and Sweeney, 1.1., 1989, A chemical kinetic model of

vitrinite maturation and reflectance: Geochim. Cosmochim. Acta, v.
53, no. 10, p. 2649-265~
MODELING OF DIAGENETIC SYSTEMS 41

Chen, W., Ghaith, A, Park, A, and Ortoleva, P., 1990, Diagenesis

through coupled processes: modeling approach, self-organization,
and implications for exploration, in Meshri, lD., and Ortoleva, P.
1, eds., Prediction of reservoir quality through chemical modeling:
Am. Assoc. Petroleum Geologists Mem. 49, p. 103-130.
Gouze, P., and Coudrain-Ribstein, A, 1993, Quantitative approach to
geochemical processes in the Dogger aquifer (Paris Basin): from
water analyses to the calculations of porosity evolution, in Don~, A
G., and others, eds., Basin modelling: advances and applications:
NPF Spec. Publ., v. 3, Elsevier, Amsterdam, p. 343-35l.
Leder, F., and Park, W.C., 1986, Porosity reduction in sandstone by
quartz overgrowth: Am. Assoc. Petroleum Geologists Bull., v. 70,
no. 11, p. 1713-1728.
Nagy, KL., Steefel, C.F., Blum, AB., and Lasaga, AC., 1990, Dis-
solution and precipitation kinetics of kaolinite: initial results at
80°C with application to porosity evolution in a sandstone, in
Meshri, lD., and Ortoleva, P.I, eds., Prediction of reservoir quality
through chemical modeling: Am. Assoc. Petroleum Geologists
Mem. 49, p. 85-10l.
Ondrak, R, 1992, Mathematical simulation of water-mineral interaction
and fluid flow with respect to the diagenetic evolution of sand-
stones, in Kharaka, Y.K, and Maest, AS., eds., Water-rock in-
teraction: Proc. 7th Intern. Symp. Water-Rock Interaction: Balkema,
Rotterdam, p. 1187-119l.
Ondrak, R, 1993, Untersuchungen zur mathematischen Simulation der
Sandsteindiagenese und der damit verbundenen Veranderungen der
Porositat und Permeabilitat: Berichte des Forschungszentrum Julich,
v. 2770, 99 p.
Ondrak, R, and Bayer, u., 1993, Dissolution and cementation in basin
simulation, in Harff, 1, and Merriam, D.F., eds., Computerized
basin analysis: the prognosis of energy and mineral resources:
Plenum Press, New York, p. 59-82.
Siebert, RM., Moncure, G.K, and Lahann, RW., 1984, A theory of
framework grain dissolution in sandstones, in McDonald, D.A., and
Surdam, RC., eds., Clasitc diagenesis: Am. Assoc. Petroleum
Geologists Mem. 37., p. 163-175.
Wood, IR, and Hewett, T.A, 1982, Fluid convection and mass transfer
in porous sandstones-a theoretical model: Geochim. Cosmochim.
Acta, v. 46, no. 10, p. 1707-1713.
 MODELING PETROLEUM RESERVOIRS IN
PENNSYLVANIAN STRATA OF THE
MIDCONTINENT, USA
W. Lynn Watney, John A. French, and Willard 1. Guy
University of Kansas, Lawrence, Kansas

ABSTRACf

Improved characterization of petroleum reservoirs must include

better geologic models in order to predict quantitative attributes of
reservoir units. By necessity, petroleum reservoir prediction and
modeling must make both interpolations and extrapolations from limited
data. Several approaches to modeling of sedimentary rocks include
descriptive or conceptual (qualitative) geologic models and geostatistical,
simulation (process), and visualization (quantitative) models. Each type
of model has its advantages and limitations, including the appropriate
scale of application, data requirements, and knowledge as to how the
reservoir was formed. All of the models compliment one another,
providing views of complex reservoirs from different perspectives.
Quantitative modeling potentially can create a more coherent, integrated
view of the reservoir than qualitative conceptual models. An optimum
model probably includes a combination of approaches based on the extent
and type of knowledge about the reservoir.

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 43
44 WATNEY, FRENCH, AND GUY

The examples of stratigraphic simulation and 3D visualization

models presented in this paper are based on multilayered oolite units
developed in the Upper Pennsylvanian, carbonate-dominated Swope
Formation depositional sequence in both the shallow subsurface of eastern
Kansas and in the Victory Field in southwestern Kansas. The models
discussed in this paper include local and regional scales and quantitative
process and visualization types.

INlRODUcnON

Stratigraphic modeling is a powerful tool that can improve our

ability to understand the complexities of petroleum reservoirs, and thereby
assist in the recovery of oil and gas from those reservoirs. Types of
models can be classified into (1) conceptual models -- a geologic
interpretation or construct; (2) correlation models -- a manual
interpretation of the spatial association of geologic units; (3) interpolation
models -- a machine generated association of spatial data, that is a
visualization; (4) forward simulations -- machine generation of geology
based on input of processes; and (5) inverse models -- machine derivation
of process parameters from geologic data. Two forms of modeling are
presented in this paper: two-dimensional stratigraphic simulations and
visualization modeling. These approaches, which include examples at
different scales, are discussed in turn in the next sections.

Stratigraphic Simulation Modeling

Stratigraphic forward simulation modeling begins as an inductive

exercise to ascertain the processes behind sedimentation and the resulting
stratigraphic succession. Initially, a quantitative data set is obtained to
infer and characterize the processes and develop a conceptual model such
as the one for Upper Pennsylvanian carbonates in the upper part of the
Midcontinent, U.S.A. (Fig. 1). The steps in database development
include:
(a) identify, describe, correlate, map, and interpret 3D geometries
of regional and locally significant temporally distinct genetic
units (such as depositional sequences) (Table 1);
(b) characterize shelf paleotopography and subsidence history;
MODELING PETROLEUM RESERVOIRS 45

(c) conduct inverse modeling procedures to estimate timing and

rates of sediment accumulation (e.g., time-series analysis of
logs and lithologic data).

Time-- ..

200-500 ka
North - -"

About
300ft
Subsidence (90m)
<0.01 mil 000 yrs
Transcontinental
Pennsylvanion Arch
Midconlinent
shelf profile

1
+---- - - 500·600 ml (800-960 m) ----+

Subsidence
0,04 m/looo yrs
Arkoma and
Anadarko
basins Gondwana

Figure 1. Conceptual model depicting processes that dynamically interact to produce

4th-order depositional sequences Upper Pennsylvanian cyclic strata of Midcontinent.

Sequence-stratigraphic analysis is used to define the temporally

distinct packages needed in the simulation model. A nonscalar sequence
stratigraphic terminology is used in which the duration and order
classification of the cycles follows that of Goldhammer, Oswald, and
Dunn (1991) (Table 1). These unconformity-bounded cyclothems (5-25
m thick) possess all of the components of Vail's seismic depositional
sequences except thickness and duration (Y oule, Watney, and Lambert,
1994). These are classified as 4th-order units based on estimated
durations of 0.25 to 0.5 Ma (Heckel, 1986~ Klein, 1990). Cyclothemic
4th-order depositional sequences are fundamental mapping units in the
Pennsylvanian of the Midcontinent, as evidenced by their designation as
formations in formal stratigraphic nomenclature (Zeller, 1968). Fourth-
order cycles can be correlated easily with wireline logs. Thicknesses and
contrasting lithologies of component units that comprise these cycles
46 WATNEY, FRENCH, AND GUY

generally can be distinguished on well logs (Fig. 2). Biostratigraphic

information derived mainly from conodonts and ammonoids is being used
to resolve independently individual 4th-order depositional sequences as
well. These sequences are correlatable regionally, and because they are
bounded by laterally extensive unconformities, they can be considered as
temporally distinct units (i.e. depositional sequences) for purposes of
paleogeographic reconstructions, interpretation of detailed facies
relationships, and for modeling.

13 T25S-R21E

s:
Ft M
100

1-'·"-
30

Member of the
Oennls Umeslone
80
Stall< Black Shale

T C&nville WmHlone
Gal_, 20

T
sn.",
'c; Upper Balhany Falls 60
::J ~11e

'GlOo
;:: :.c
00
Lower Bethany Fall.
Oi Gl C<o Oolrte
.§ g Gl~
<.!l.2l 40
.-.:: Q)
<I}::J
~ Lower Bo1Ilany FBIII 10
OCT
OoQ) u; Umas!ona
~<I}
....L Hushpuckr1.y Bloclt Sh
20

~
MiIddle Cleek Lst
Elm Brar>eh S..I.

Sn!abat Limest0J'l8 Mbr

Gamma of H&nha Ume51cne
a a
Ray

Blocky M..Jdstone
(RJleosol) ~OI
Sequence
Umestone (Jvbinly W:x:kestone BoundoIies
O"ld M.Jddy R:x:kstone)
Umestone (Jvblnly Grcinstone
and v.el1-v.ashed R:x:kstone) Mnor
Sequence
M..Jdslorre and Shale BoundoIies
Qncludes Ajeosds)
A'1osphatic Black Shale

Figure 2. Unsealed gamma ray-neutron wireline log annotated with Swope sequence
and its component genetic units: Middle Creek Limestone = transgressive unit;
Hushpuckney Shale = condensed section; Bethany Falls Limestone = regressive
(forced) unit; Galesburg Shale = capping paleosol in the location of this well. Wavy
lines are subaerial unconformities.
MODELING PETROLEUM RESERVOIRS 47

Other scales of unconformity-bounded sequences are present,

including 3rd-order cycles consisting of sets of five to seven 4th-order
depositional sequences. In some situations where 4th order cycles are
thick, 5th-order unconformity-bounded minor cycles are recognized, thus
the stratal packaging is complex. Any model must resolve units down to
the fifth order, because these are flow units fundamental to reservoir
performance and reservoir prediction.
The stack of unconformity-bounded oolite and wackestone
packages developed within the Bethany Falls Limestone of the 4th-order
Swope sequence modeled in this paper are classified as 5th-order minor
cycles. These cycles exhibit transgressive bases and regressive tops, with
basinward shifts leading to subaerial exposure interpreted as evidence of
forced regression by eustacy. These minor cycles are therefore not
parasequences even though the thicknesses of these units might suggest
such an interpretation.

Table 1. Cycle orders (modified after Goldhammer, Oswald, Dunn, 1993)

Cycle Order Duration Am(!litude RiselF all rates Seguence terminology This study
(million (meters) (mIka)
yrs., Ma)
1st order >100 <.01
2nd order 10-100 50-100 0.01-0.03 Supersequence
3rd order 1-10 50-100 0.01-0.1 Sequence Sequence Set
4th order 0.1-1 1-150 0.4-5 Sequence, Cycle Cycle, Cyclothem
5th order 0.01-0.1 1-150 0.6-7 Parasequence, Cycle Minor Cycle

The correlatable cyclicity observed in these strata is attributed to

changes in accommodation space effected by changes in relative sealevel.
These changes in accommodation are interpreted as a response to an
interplay of tectonism, eustasy, and sediment supply. Simulation
modeling is used to examine the relative contributions of these causal
mechanisms.

Parameter Definition

Specific procedures to derive parameters used in the simulation

modeling are divided into six categories (Table 2). These parameters
were discussed in Watney, Wong, and French (1991).
48 WATNEY, FRENCH, AND GUY

Table 2. Parameters used in modeling Pennsylvanian cyclothems.

(I) Characterization and interpretation of temporally distinct

stratigraphic units;
(II) Estimates of subsidence across Kansas shelf during the
Upper Pennsylvanian;
(III) Estimates of elevation across the Pennsylvanian shelf and
magnitude of relative sealevel change;
(IV) Characteristics of Pennsylvanian eustacy;
(V) Sealevel history estimated from Quaternary analog;
(VI) Sediment accumulation rates from Recent and Quaternary
analogs.

The values of the parameters used to build the simulations include:

(1) 4th-order sealevel rise and fall ranging to maximum of 100 m

characterized by rapid rise and gradual fall, patterned after late
Quaternary eustatic fluctuations;
(2) elevation differences across the depositional surface of up to
100 m based on study of regional isopach, facies, and
decompaction (135%) of the Pleasanton clastic platform;
(3) sediment accumulation rates determined by water depth and
slope of depositional surface (energy level) ranging under 2
m/thousand years (m/ka) and based on Quaternary analogs
(Fig. 3);
(4) subsidence ranging between 0.02 mlka and 0.2 mlka
determined through backstripping;
(5) the maximum depth at which carbonates accumulated
(euphotic zone) was set at 35 meters, using modem data to
help constrain this value.

2D Stratigraphic Simulation Modeling (KANMOD)

Our initial geologic simulation model of carbonate sedimentation

was developed for a personal computer using one-dimensional output
generating a single stratigraphic succession at one location. This ID
model incorporated parameters derived from characterization of Upper
Pennsylvanian carbonate-dominated depositional sequences as well as data
on sedimentation and stratigraphy derived from Quaternary and Recent
MODELING PETROLEUM RESERVOIRS 49

analogs (Watney, Wong, and French, 1991) (Fig. 3). The ID model has
user defined input of eustatic sealevel curve, subsidence rate, and
carbonate sediment accumulation rate. The output is a time vs. elevation
plot of the sealevel curve, initial depositional surface, and new sediment
surface. Rapid sealevel rise leads to attenuation of sediment
accumulation rate. A maximum depth for carbonate accumulation is
gIven. No erosion occurs during subaerial exposure.

lOW-eMtgy wave base

- - - HIgh-Ene'W Cond<'''''''
___ Low-Energy CQnd,'.OI'IS

25

30

I I I I
0.5 075 10 125 1.5 1.75 2.0
025
Sedimentation Rate (mlka)

Figure 3. Carbonate sediment accumulation rate vs. water depth used in 2-D model.

The second phase of simulation model development was an

expansion of this preliminary one-dimensional simulation into a two-
dimensional model (French and Watney, 1990). Programming was done
by 1. French using the Pascal language; the program is tentatively titled
KANMOD. This stratigraphic simulation model generates carbonate
sediments according to water depth and energy levels related to
depositional slope. This model is used to simulate the development of a
set of oolite shoals that occur as part of individual 5th-order minor cycles
developed in the 4th-order Swope sequence in southeastern Kansas (Fig.
2). This study area is sef'Ving as a shallow subsurface to outcrop analog
50 WATNEY, FRENCH, AND GUY

site to oil and gas fields occurring in south-central and southwestern

Kansas. Shelf conditions, elevation, slope, and general deposition setting
were apparently similar in these two areas even though they are separated
by distances of 200 to 400 mi.
KANMOD generates up to 11 different carbonate facies,
including carbonate skeletal grainstone and oolite, which is the most
abundant reservoir type in the Missourian carbonate-dominated
depositional sequences being studied. Depositional rates are significantly
reduced by rapid relative sealevel rise (subsidence + eustasy) such as
those inferred for the modeled Pennsylvanian strata. Essentially,
sedimentation can not keep up with rapid rising sealevel. Also, sediments
are slowly eroded during subaerial exposure.
The model generates a color 2D image of the resulting
stratigraphy, as well as more detailed individual stratigraphic profiles at
selected locations defined by the user in the modeled profile. The
computer screen output can be captured to obtain a hard copy using
commercial screen capture programs. Input parameters include
subsidence rates, eustatic sealevel changes, and initial depositional
topography. Sediment accumulation rates are inferred from modem
sediment analogs defined via depth-rate relationships. Subsidence rate
can be changed across the area that is being modeled. A eustatic sealevel
curve can be drawn using the keyboard, read from a file, or generated
through the use of convolved sinusoids such as Milankovitch orbital
parameters. The initial shelf configuration on which deposition takes
place can be drawn on the screen or input from a file based on
information interpreted from mapping and facies analysis.
The horizontal and vertical scales of the graphical output can be
changed for successive runs to maximize the image. Parameters that can
be varied include the duration of individual runs and the vertical
exaggeration of the output. Smaller areas of the output can be selected
and expanded to full screen by drawing a box around the area of interest
and repeating the run. The second screen then is filled with the boxed-in
area.
Run length of the 2D model can be differed within two basic
scales: (1) 100,000 to approximately 2 million years; or (2) longer term,
5+ million years for 3rd-order sequence modeling. Time steps differ with
run length; shorter runs are composed of shorter time steps. At each time
step during a model run, a pixel of a given "facies" is "deposited."
MODELING PETROLEUM RESERVOIRS 51

Model output is comprised of a set of these pixels. Model runs typically

range from 3.5 minutes to 5 minutes on a 386-33 IBM-compatible
computer.
The basic stratigraphic unit that is generated by the model is the
depositional sequence. All of the primary types of genetic units that
comprise these depositional sequences can be produced, including
flooding units, condensed sections, late-highstand (or forced-regressive)
units containing complex internal stratigraphy, minor cycles within the
depositional sequence, high-energy grainstone units, and subaerial-
exposure surfaces. Lowstand, transgressive, and highstand systems tracts
can be developed under the proper conditions. Geometries usually
observed are produced including progradation oflate highstand, regressive
genetic units with toplap and downlap. Time lines can be included at
user-selected intervals to identify where the depositional surface was at
various times during development of the deposit.
A high-resolution data file also can be generated during the
model run capturing pixel information. The size of the file is limited by
the available RAM. These files include information from each time step
across the entire two-dimensional model "area," including facies and
thickness information as well as temporal data, such that the duration of
the hiatuses and the relative ages of unconformities can be captured.

1WO-DIMENSIONAL SIMUlATION OF STACKED OOLITES

Overview

The southeastern flank of the Pleasanton siliciclastic platform in

eastern Kansas is a pronounced depositional slope (Fig. 4). At least three
temporally distinct sets of strike-oriented oolite buildups were deposited
along this basin-facing slope in the Bethany Falls Limestone Member
(regressive phase exhibiting forced regression) of the 4th-order scale
Swope depositional sequence. Interpretation of data suggests that the
oolites onlap landward and downlap basinward. In some locations several
of the oolites form what is interpreted as a shingled stack of oolites
representing separate episodes of accumulation during falling sealevel.
Some of the bounding surfaces separating the oolites are subaerially
exposed. Subaerial exposure of these surfaces are suggested by thin
blocky claystones with rhizoliths along some of these surfaces and light
 Base Nuyaka Creek - Base L V"I
tv

_1 T--~

_J
___I
r-
,
r---r/ ' I

L_ i ~.

,.

-';,. fI' J, t?W .~f ~UI' lP ..,. If.,. In '1,r\\)~n¢:rt~JI.;G.J;ft::t-·[;f;;::~·vr.~• .' J' l C;~ '1""I~:S·jlAIJ" .. 1m J~
.i
Figure 4. Isopachous map of Pleasanton Group showing large clastic wedge in eastern Kansas. 10-ft
contours with gray levels distinguishing 30 ft (9 m) increments. Thicknesses up to 200 ft (61 m) in area
of southward prograding deltaic platform. White bar is location of cross section in Figure 5. Platform
thins gradually to west against Nemaha Uplift. Abrupt southern margin is interpreted as a depositional
!
~
slope. Southward beyond this margin is time equivalent sediment-starved interval reflecting lower shelf
adjoining deeper actively subsiding foreland basin (Arkoma Basin). Several subsequent cycles are draped
over this topography including Swope sequence, the focus of this paper. ~
MODELING PETROLEUM RESERVOIRS 53

isotopic excursions of carbon and oxygen profiles. These isotopic profiles

shows depletion of heavier carbon at the position immediately below
several of the apparent exposure surfaces, suggesting meteoric exchange
with the limestone. In some situations, the base of the oolites is marked
by thin zones of probably slightly deeper water subtidal carbonate,
suggesting that deposition of the oolitic units was punctuated by minor
marine transgressions.
A similar stack of oolitic units in the Bethany Falls Limestone is
developed in the Victory Oil Field, which is located 480 km to the west
in Haskell County, Kansas. Victory Field is located roughly along
depositional strike of the near-surface occurrences, based on regional
mapping of the Swope sequence. Results of 3D visualization modeling
of this sequence in Victory Field in southwestern Kansas are presented
in a later section of this paper.
The 2D simulation modeling program, KANMOD, was used to
generate a dip-oriented cross section of the Swope sequence in the
shallow-subsurface analog area in southeastern Kansas. The cross section
is datumed on a persistent marker that occurs below the Pleasanton clastic
wedge. The Pleasanton thins southward toward the basin along this line
of section. The slope shown here is interpreted as a minimum
approximation of depositional slope at the time the Swope sequence was
deposited.
An initial examination of the oolite might suggest that the oolite
represents the culmination of a simple regressive, shallowing-upward
succession. However, multiple cores taken near this cross section and up
and down the slope reveal an oolitic grainstone that is punctuated by
multiple thin skeletal packstones and subaerial exposure surfaces resulting
in a stacking of these units (French and Watney, 1993). The correlations
of the oolites shown on the cross section between the various shelf
positions is conjectural, but suggests the development of different oolite-
bearing minor cycles as accommodation is gained or lost. Internal
correlation of these high-frequency minor cycles is not definitive.
The resulting stratal pattern is an en echelon or imbricate set of
oolitic units. Although the oolitic facies occupies the same general
stratigraphic position in the Bethany Falls Limestone, it is interpreted to
be comprised of multiple, temporally distinct packages that were
deposited as the result of relative changes in sealevel. This has
significant implications for reservoir characterization and estimating fluid
flow in these strata because the bounding surfaces and intervening
54 WATNEY, FRENCH, AND GUY

lithologies usually lack permeability. Each stratal package exhibits

genetically different oolites with differing textures, adding further to the
heterogeneity. The stratigraphic simulation is an attempt to generate this
stratal pattern via inferred geologic processes.

Results of Simulation

The Swope sequence was simulated based on a dip-oriented cross

section (Fig. 5). The input parameters were derived from this local
setting including slope angle and relief. A sealevel curve was defined
based on interpretation of the facies successions of the Swope sequence
and adjoining units in core, outcrop, and wireline-Iog data obtained in the
area (French and Watney, 1993).
Three separate oolitic units were generated from three separate
stillstands of sealevel that were part of an overall regression (caused by
falling relative sealevel) that occurred near the end of deposition of the
Swope sequence (Fig. 6). The latter portion of the sealevel curve that
generated the entire Swope sequence is shown in the figure. The
distribution of oolite in the computer model resembles that depicted in the
cross section, with shingled oolites stepping down the depositional slope.
This model output includes time lines for every 30,000 yrs. This is
instructive to show the time separation of the oolites and the downlap
relationships of the oolite with contemporaneous deeper water facies.
Also, while an oolite unit is being deposited on a low shelf elevation, the
upper shelf is undergoing subaerial exposure. The convergence of time
lines in a basin ward direction (downlap) occurs in the black, organic-rich
marine shale in this sequence (the Hushpuckney Shale) (Fig. 2). This
convergence of time lines in this shale reflects a condensed section.
Water depth during the accumulation of this black shale is interpreted to
be below the effective photic zone. The depth of the photic zone in this
model run in Figure 2 is set at 35 m. Thus, limited or no carbonate
accumulation occurred, and because of the lack of evidence of significant
clastic influx accumulation rates were set low in the model (.01 m/ka).
The simulation results suggest that carbonate deposition occurred along
updip shelf locations during equivalent intervals of time when black shale
accumulated on the lower shelf and basin. The updip carbonate section
essentially downlaps onto the black shale.
 West East ~
KGS McGee II XNW KGS Ermal 'I KGS Highbaugh.l KGS Green Aaes'l BEA KGS Womelsdorl'l KGS Emerson'l o
tJ
I- 5.0 mi(B.O km) + 3.B mi{S.1 km) -f- 2.4 mi(3.9 km) + 2.0 mi(3.2 km) + 2.1 mi(3.4 km) -j- 2.5 mi(4.0 km) -I- 3.0 mi(4.8 kmj -l tn
C
Z
o
~
~
Mound Valley ot"""
tn

~
~
rn
tn
~
re;,?~OomlnnnUY oolilk: gra1nslone
o......
•• • and packstone
SkeleleJ pockslon. ~
rn
~ 00011nondy CoO,bonalo
wackoslono

Dominantly sandslone and

[2) Sllilley sandstone Her1ha sequence
~~~:~:roshalo ond
~ 50 It
Soquence Boundaries: (15 m) '"""
• . • Unconlormlly .J
11.-;~·"""1 - ~."'11 CorrelativB Conform Datum Nuyaka Creek black shale

Lltl"lofaeies Boundary
Block Shel.

Figure 5. Diagrammatic core and outcrop cross section Swope sequence comprised of Bethany Falls and Mound Valley
limestones. Dip-oriented cross section identifies three discrete, shingled oolite layers in Bethany Falls Limestone and VI
VI
additional oolite Mound Valley Limestone.
 V\
0\

-~
;
(')
~::c:

Figure 6 . Stratigraphic simulation of three oolitic grainstones developed along slope as result of three minor sealevel
~
oscillations (stillstands) during overall sealevel fall . Time lines are white dashed lines cutting through facies indicating
contemporaneous facies.
~
 ~~'"'~~~'2
BoIJoton Co., Ks ~
o
., tJ
. \ tr1
; '\ ~~ - .-
zot:
i /\ IZ
:·2----
~/ f' \
tg
:~1 Sea-Level ~~
Curve ~
~ o
t"""
tr1

~
~
C/)
tr1
~.I"
~
20M I o.....
I I I
:;d
400 450 500 C/)
sJ ~do .H~O ?tiD 2~0 arlo 3~O 20k ...
HIGH ENERGV FACIES LOW ENERGV FACIES SL RISE )6"/ka (no

_~-5
<~ Meter
Meters
_
_
<2 Meters
2-8 Meters
energ~

_
considerations)
~-20 Meters
-
_ 2 0 - 35 Meters
5-~5 Meters _ 8 - 2 0 Meters
_ ~5-20 Meters _ 2 0 - 2 7 Meters G@n@ral/other
_ 2 0 - 2 7 Meters _ 2 7 - 3 5 M.. ters iII Storn washovers
_ 27-35 Meters Subaerial Unc:on
_ > 3 5 Meters
_ >35 Meters Sea-level pOSition

Figure 7. Regional 2D stratigraphic simulation of KANMOD. Swope 4th-order carbonate-dominated

sequence depicting deposition across depositional topography created by Pleasanton clastic wedge
(underlying uniform gray interval). Section extends north-south across northern Midcontinent (500 km,
~
310 mi).
58 WATNEY, FRENCH, AND GUY

A regional stratigraphic simulation of the Swope sequence in

Figure 7 depicts the sequence along a 500 km-Iong south-to-north cross
section spanning the carbonate-dominated shelf across the northern
Midcontinent. The model includes the relatively steep Pleasanton slope.
The model run includes only two stillstands in sealevel during late-stage
regression of the Swope sequence, leading to formation of only two
oolitic units on the break in slope. The simulation includes an expanded
stratigraphic profile of the Swope sequence containing the two oolites.
The location of this profile on the model section is identified by the
arrow.
An actual measured stratigraphic section obtained from a core on
the shelf break is shown in the boxed area alongside the simulated
section. The simulated section resembles the actual section in terms of
facies type, thickness, and stratigraphic succession. Another simulated
succession is shown higher on the shelf. Here the Swope sequence is
much more attenuated than to the south. The landward, higher shelf
locations had more limited accommodation and thus thinner succession
in general.

Conclusion of Simulation Results

The simulation model corroborated the conceptual model inferred

from observations in the analog area by adjusting input parameters. The
simulation results have helped to constrain the a priori concept. The
model has successfully generated the observed stratal patterns via inferred
geological processes. The procedure aids in prediction of the oolite
petroleum reservoirs in the sense that the stratal template to be likely
caused by geological processes that affect other analogous parts of the
Kansas shelf. Accordingly, a more intelligent and focused search can be
made to locate similar stratal geometries and lithofacies. The timing and
magnitudes of the sealevel stillstands seem strongly to control the
occurrence and stacking of the oolites. The calibration of this relative
sealevel change is the purpose of the near-surface analog site, where cores
can be taken to resolve the distribution of units and reconstruct a stacking
geometry. Regional studies also are needed to ascertain the shelf
configuration and relative elevations of other locations and to develop a
robust relative sealevel curve.
MODELING PETROLEUM RESERVOIRS 59

Relative sealevel change involves both tectonics and eustacy.

Thus, sealevel histories interpreted from two widely separated areas are
expected to differ depending on the local tectonic history and relative
changes in elevation and subsidence history during accumulation of the
stratigraphic interval of interest. This is the challenge that can be
addressed through integrated, multiscaled studies, and sensitivity analyses
of the critical parameters.

11IREE-DIMENSIONAL VISUALIZATION
OF OOLmC RESERVOIR IN VICfORY FIELD

Overview

Three-dimensional visualization is a form of modeling, whereby

a framework of attributes of a rock volume is viewed using computer
graphic displays. Correlations of attributes between wellbores as in these
examples are simply linear interpolations within predefined layers. The
definition of this layering determines how the attributes are interpolated
between the wells. The primary layers are the depositional sequences,
with internal layers assigned to best approximate how the rock was
deposited. Geostatistical models are available to make extrapolations
conditioned to the available well data. Computer power and graphic
capabilities provide real-time 3D viewing and manipulation of the rock
volume. Stratamodel (version 3.1), running on an SGI Personal Iris, is
the software used in this example. Stratamo del is a stratigraphic
geocellular modeling program that can display various well attribute
information interpolated into a 3D volume displayed as surfaces, panels,
cross sections, and solids. Key stratigraphic surfaces such as correlatable
bounding surfaces of temporally distinct 4th-order cycles are used to
provide an initial framework.
Wireline-Iog data in Victory Field contain critical information for
recognizing bypassed oil and evaluating improved oil recovery potential.
Visualization techniques are shown to help in identifying and
characterizing the reservoir heterogeneity.
The dominant reservoir rock in Victory Field is oolitic grainstone.
The 4th-order Swope sequence is the focus of this example. It is
comprised of apparent 5th-order cycles analogous to those described in
the near-surface analog site in southeastern Kansas. The visualization
60 WATNEY, FRENCH, AND GUY

model indicates subtle packaging within a thick (up to 18 m) oolitic unit,

with internal layers intricately stacked and shingled. Porosity is
dominantly oomoldic, and is interpreted to have resulted from freshwater
dissolution associated with early subaerial exposure. The porosity
apparently was enhanced by later dissolution and crushing. The crushing
of the fabric is inferred to represent the product of early burial
compaction of the more weakened areas of the dissolved oolite. Locally,
these pores are occluded by late-stage calcite and dolomite cements which
further adds to the complexity. The porous and permeable rock is
spatially limited as isolated pods in the thicker lobes of oolite. These
pods are perhaps related to sites where greater volumes of undersaturated
meteoric water percolated through the unit (Watney and others, 1993;
1994).
Effective pay is recognized in this complex pore system using
"Super" Pickett plots. Typically, the porosity cutoff is high (>16%) and
water saturation is low «20%). Together these parameters indicate a low
bulk volume water (porosity times water saturation) in the pay intervals.
Stratigraphic intervals with these cutoffs and reservoir thickness
exceeding 1 m are considered to be effective pay. Visualization of these
calculated parameters was accomplished using Stratamodel. The
information together is used to assess the potential for bypassed oil and
gas accumulations. Utilizing these techniques, an estimated 3 million
barrels of oil are believed to be behind pipe in Victory Field (Watney and
others, 1994).
A 7.5 sq mi area was selected from Victory Field for 3D
visualization modeling (Figs. 8 and 9). Forty-seven wells were digitized
from this area representing about 15,000 m of log traces from an interval
extending between several cycles above and below the Bethany Falls
Limestone. Log type types used included gamma ray, sonic, neutron and
density porosity, and induction resistivity. The area is located in the
northern part of Victory Field, encompassing one of the two large oolite
lobes developed in the Bethany Falls Limestone. The particular
objectives of the field study were to:

(1) establish best-fit stratigraphic layering and

compartmentalization of the oolitic reservolf usmg
visualization modeling;
 a=
otj
SlJOPE
tT1
t:
z
o
'"d
tT1
;d
o~
tT1

~
~
en
tT1
~
o
'"""4
~
en

Figure 8. Isopachous map of Swope sequence in Kansas. Contour interval is 10 ft (3 m). Shading is
every 20 ft (6 m) . Medium gray area is location of thicker oolitic grainstones. Width of this region is
narrower in southeastern Kansas because of steepness of shelf edge. Thick black area in southeastern
Kansas includes clastics added by southerly source, lapping against depositional topography. Analog site 0\
and Victory Field are identified. -
62 WATNEY, FRENCH, AND GUY

Victo r y Oll Field

.
l"

. . ·· "· . · ·· ··
-I

v .>
.,.
.
·· . . .. . . '" · " · I,
~

.>

.,.
" ·
0

, .,
·
" ·· .
· . " .·
..· " • 0

.,. . . · ·
. · '" ·· . ·· · "" . .,. ·..
I 0

""
I
I ~

<>
<>
,
· · TYPE

· ".
I

'" " ' \,.. <>

0-
• a,'

..
I
I '" . l"- · · ,,~:~e 10 '" Ca •

" .
P- , v

. · " ·· ·· . ·
0 o
I

..
I
.,.
" , .,..,.
.
I ~ ~,
r.::
"
" 0
· ·<1\>
· ··..,.·· . · . 1-_ · ··· h'"
<> ..
0

.,.
· .1' · · · .,. .
!
Figure 11 •
'"
J
· "· " · " ·· ·
" I·".
" .~

.> <>
"
· ·
0
~

· ···· ·
" "
0 0

"
-
0 .. 0

~ : ."
0

· '" I
<>
<>1 r ...
"
· ·
<>
·
<>

I
~

'"
,,, I
<>
. .'" <>
"
:lot" ".
1»'5.1,;1 (toLal:1ICIJ. :nnrrrr

Figure 9. Index map to Victory Field identifies wells used to make 3D visualization
with Stratamodel software.

(2) calculate and display variables (such as water saturation and

sonic-minus-density porosity) as color attributes to show
interparticle and total porosity, bulk volume water (Sw x
porosity), and visual displays of an empirical estimate of
separate vug porosity developed by Lucia and Conti (1987);
MODELING PETROLEUM RESERVOIRS 63

(3) apply filters to target areas of the field with specific reservoir
properties based on selected porosity-water saturation cutoffs;
(4) systematically examine variables in the modeled reservoir
volume to assess controls and perhaps enhanced views of areas
to be targeted for additional oil recovery;
(5) compare this visual, empirically derived model of a reservoir
to the simulations of a temporally equivalent oolite in
southeastern Kansas.

Computer elevation grids of key stratigraphic surfaces from

Victory Field were read into the program. Each datum serves as an event
surface, which in this study are divisions between depositional sequences
and the genetic units that comprise them. Additional layers are
generated within these temporally distinct packages for use in
interpolating attributes. The layering template can be selected that most
closely fits the internal stratigraphic framework:

(1) onlap -- layers parallel the upper event boundary;

(2) offlap -- layers parallel to an inclined internal surface that is
assigned to approximate progradation;
(3) truncation -- layers of equal thickness parallel the lower event
boundary where thickening and thinning of interval is
accomplished by adding or truncating layers at the top event
surface, that is the upper surface is an unconformity;
(4) proportional -- layers are a constant number in interval, each
varying in thickness in proportion to overall thickness

The regressive limestones were best treated by truncation whereas the

shales and flooding units were assigned layers that onlap.
The layers define the correlation template for interpolation
between the wells. The grid consists of square cells with a spacing of
180 ft (90 m). The cells have the same thickness as the layers, set at a
constant 2 ft (0.7 m) for the Bethany Falls Limestone. As the unit
thickens, layers are added to the top of the unit. This is referred to as
truncation layering. Each cell retains the values of the attributes assigned
by interpolation from the wells. The grid size (60 x 60 cells) and number
of layers (739) determine the size of the model that is manipulated by
Stratamodel. The total number of cells in the Victory Field model is
approximately 2.6 million.
64 WATNEY, FRENCH, AND GUY

Digitized well data, sampled every 0.5 ft (0.15 m), were read into
Stratamodel. The program reassigns (averages) the vertical well data to
values for each layer, for example, into 2 ft (0.61 m) layers in the oolite
example. The values of the attributes in the layers then are interpolated
using a weighted distance function, analogous to the surface gridding
procedure. The search radius and other parameters can be selected to
conform with that of the imported grids.
An attribute model was built from Victory Field log data
transforming the raw log values to new composite variables, for example,
water saturation and bulk volume water. An option termed geobody also
was used to determine the amount of connectedness between cells that
met certain criteria, for example, cutoff values of porosity and water
saturation.

Results of Visualization

The visualization modeling conducted for Victory Field well-log

data included construction of a series of cross sections between wells.
These sections facilitated both relating the model back to the original well
data and comparison with other analyses such as petrography (Figs. 10
and 11). Panels of multiple, regularly intersecting cross sections were
generated that serve as the primary method for presentation of 3D
properties (Figs. 12 through 15).
Working with the computer, models can be rotated and zoomed
in and out. Attributes are selected for display and color values assigned
(scaled or given default values). This tool is particularly powerful when
attributes are filtered to isolate specific intervals, layers, areas, or
attributes in a certain range.
The fine color resolution and distinction provides an enhanced
stratigraphic view of the layering of the oolitic reservoir within the
Bethany Falls. The truncation method provides the most coherent
layering of the interpolated model. Attempts of other types of layering
such as parallel to base and proportional exhibit more lateral
heterogeneity within the layers. The truncation method builds layers from
the base of the limestone upward with layers parallel to the base as
previously described. This truncation layering appears to be the most
coherent because it most closely approximates the stratigraphic layering
MODELING PETROLEUM RESERVOIRS 6S

Figure 10. Southeast-to-west cross section generated using Stratamodel software

showing interpolated gamma ray between wells. Section is indexed in Figure 9. Even-
toned gray units are shales separating cycles of limestone. Swope sequence is
highlighted.
66 WATNEY, FRENCH, AND GUY

Figure 11. Southwest-to-northeast structural cross section showing interpolated sonic

porosity. Section indexed on map in Figure 9. Bethany Falls Limestone is identified.
2-ft thick cells involved in making interpolations are highlighted in this presentation.
Truncation model for generating cells made best fit, approximating how limestone was
deposited. Higher porosity (lighter shade of gray) thins markedly to northeast.
MODELING PETROLEUM RESERVOIRS 67

Figure 12. 3D panel, looking northwest, showing interpolated gamma ray for Bethany
Falls Limestone. Lowest gamma-ray limestone corresponds to medium gray at top of
interval. This is location of oolitic grainstone. Notice how it thickens in southeastern
area.
68 WATNEY, FRENCH, AND GUY

Figure 13. 3D panel looking northwest showing interpolated density porosity in

Bethany Falls Limestone. Light shaded layers are in upper Bethany Falls and
correspond to oolitic grainstone intervals. Even gray on edges is no data where density
logs are lacking. Some of layers appear to lap out into structure.
MODELING PETROLEUM RESERVOIRS 69

Figure 14. 3D panel of filtered (>20%) interpolated density porosity for Bethany Falls
Limestone looking to northwest. Onlap of porous carbonate layers to northwest is
indicated.
70 WATNEY, FRENCH, AND GUY

Figure 15. 3D panel of calculated bulk volume water (BVW) for Bethany Falls
Limestone using sonic porosity, filtered at 10% cutoff. Areas with low BVW (medium
gray) are thought to be prospective reservoir rock with significant oil or gas saturation.
MODELING PETROLEUM RESERVOIRS 71

that is represented by the well logs. Thus, the selection of layering

procedure is necessarily an iterative procedure.
The filtering option is used to isolate portions of the reservoir that
appear to contain effective pay. Filtering can be used to isolate
individual sequence, layers, or range of attribute values. Using this
approach, a previously undetected isolated porosity pod was located in
Victory Field, located slightly down dip of the structural crest on the
southeast flank of the modeled area (Fig. 14). The fact that the area is
isolated from porous intervals updip suggests that hydrocarbon may be
trapped there.
Three separate porous layers in the Bethany Falls Limestone are
recognized in the model (Fig. 14). The porous layers are oolite as
confirmed by cuttings and core. The lower porous layer laps out updip
onto the structure, the middle layer extends over the crest of the structure,
and the top porous layer onlaps out on the flank of the structure similar
to the lower layer. It therefore seems that the current structure was a
paleohigh. Other evidence such as general stratigraphic thinning suggest
that Victory Field was a localized positive structure.
Base and top of the oolite succession are subaerial exposure
surfaces and suggest a higher order cyclicity than the general 4th-order
cycle. Other internal surfaces that define the suggested layering are not
expressed clearly as exposure surfaces. Isotopic analyses are needed to
examine if these surfaces were indeed exposed. A relative rise and fall
of sealevel may have lead to development of the stacked oolites.
The onlap/offlap relationships of the oolites in the Swope
sequence at Victory Field resemble the shingled offlapping oolites
modeled in the analog sites in southeastern Kansas. The implications of
the simulation modeling in the analog area suggested that this type of
minor cyclicity could express itself elsewhere where a similar shelf
configuration presented itself (slope, shelf elevation that is appropriate for
sealevel history to play out, and depositional conditions). The Victory
Field area is such as site. The correlation of temporal equivalency of
these minor cycles between the areas is currently not possible because the
minor cycles are currently below biostratigraphic resolution. Regardless,
a subdued basin-facing slope is suggested to have influenced the
formation and distribution of several small-scale oolite units in Victory
Field in a manner similar to the outcrop analog site in eastern Kansas.
Other down dip and overlooked "sweet spots" in other fields along this
shelf trend may be located and extend the life of these fields.
72 WATNEY, FRENCH, AND GUY

The areas of low bulk volume water shown in a visualization

(Fig. 15) compare closely with favorable (effective reservoir) bulk volume
water cutoffs for oomoldic porosity as defined from crossplot analysis.
This off-structure pay is analogous to the development of flanking
effective pay in the underlying Sniabar Limestone reservoir in nearby
Northeast Lemon Field. The implications for similar traps and
compartments in other fields or wildcat locations beyond existing fields
and in new pays is significant. In general, stratigraphic
compartmentalization coupled with diagenetic overprinting is important
in defining the location of remaining reservoirs and establishing reservoir
continuity .
Slices through the individual layers of the reservoir further
illustrate the lateral onlap/offlap relationships of the porosity in the oolite
layers (Fig. 16). Cells that were connected were used to define a
geobody. The surface of this resulting volume is displayed in Figure 17.
It reveals the basic tripartite division of the oolite reservoir created by the
layering.
Lateral and vertical heterogeneities are many in the Lansing-
Kansas City Groups including multiple cycles, subzones, and layers
within the cycles. Each cycle and possibly each subcycle can have
different reservoir properties due to separate, temporally distinct
depositional environments, and differences in diagenesis. Three-
dimensional modeling initially is expensive and time consuming,
especially with regard to data preparation, but the ease in examining the
reservoir, drawing inferences, and communicating with other collaborators
is facilitated through the use of such techniques. The modeling provides
an excellent vehicle to communicate the nature of reservoir heterogeneity
and can be helpful in later design of recovery methods. Also, this
software permits the geologic model to be scaled up and exported to a
reservoir simulator.

CONCLUSIONS

Quantitative geologic modeling is an integral component in

improving the understanding and prediction of petroleum reservoirs.
Different modeling approaches have inherent advantages and
disadvantages, but can complement each other when used judiciously.
Future modeling efforts probably will combine components of several
MODELING PETROLEUM RESERVOIRS 73

Figure 16. Slice map of density porosity view looking northwest showing layer from
lowest porous grainstone. Porosity laps out westward onto structure.
74 WATNEY, FRENCH, AND GUY

Figure 17. Geobody showing connected cells within Bethany Falls using density
porosity with 25% cutoff. This view looks southwest showing layer numbers. Lateral
discontinuities are noted on this gray-scale presentation.
MODELING PETROLEUM RESERVOIRS 75

types of modeling to obtain the desired ranges of scale of application and

optimization of the available data.
Stratigraphic simulation modeling operates on estimates of
process variables, producing output that resembles actual stratigraphic and
facies data. Experiments using this modeling produces insights as to how
the strata were deposited and aids in improving predictions utilizing
sequence stratigraphic analysis. The nature and distribution of temporally
distinct depositional sequences are fundamental to successful utilization
of this type of modeling because timing of events is of paramount
importance.
Visualization modeling as described in this paper utilizes 3D
presentation and manipulation of the stratigraphy to help reconstruct
complex reservoir geometries. Again, the approach is based on an
established framework of temporally distinct stratigraphic units in which
quantitative reservoir attributes are interpolated thorough various methods.
The interpolation is an iterative, empirical approach. The derived model
can provide additional insight and understanding that is useful in reservoir
prediction and compliments the simulation models.

ACKNOWLEDGMENTS

Johannes Wendebourg and John Davis are thanked for their

reviews of this paper, their comments were very helpful and improved the
paper significantly. lC. Wong is thanked for implementing Stratmodel.
Ajit Verma and Tracy Gerhard are thanked for their assistance In
digitizing well logs and formatting information for Stratamodel.

REFERENCES

French, lA., and Watney, W.L., 1990, Computer modeling of Mid-

Continent cyclothems and its application in the prediction of
hydrocarbon reservoirs (abst.): Am. Assoc. Petroleum Geologists
Bull., v. 74, no. 5, p. 657.
76 WATNEY, FRENCH, AND GUY

French, lA., and Watney, W.L., 1993, Integrated field, analog, and shelf-
scale geologic modeling of oolitic grainstone reservoirs in the
upper Pennsylvanian Kansas City Group in Kansas, in Lineville,
B., ed., Reservoir characterization III: PennWell Books, Tulsa,
Oklahoma, p. 983-993.
Goldhammer, RK., Oswald, E.l, and Dunn, P.A., 1991, Hierarchy of
stratigraphic forcing: Example from Middle Pennsylvanian shelf
carbonates of the Paradox basin, in Franseen, E.K., Watney,
W.L., Kendall, C.G.St.C., and Ross, W., eds., Sedimentary
modeling: computer simulations and methods for improved
parameter definition: Kansas Geol. Survey Bull. 233, p. 361-413.
Goldhammer, RK., Lehmann, P.l, and Dunn, P.A., 1993, The origin of
high-frequency platform carbonate cycles and third-order
sequences (Lower Ordovician El Paso Group, West Texas):
Constraints from outcrop data and stratigraphic modeling: Jour.
Sed. Pet., v. 63, no. 3, p. 318-359.
Heckel, P.H., 1986, Sea-level curve for Pennsylvanian eustatic marine
transgressive-regressive depositional cycles along midcontinent
outcrop belt, North America: Geology, v. 14, no. 4, p. 330-334.
Klein, G.deY., 1990, Pennsylvanian time scales and cycle periods:
Geology, v. 18, no. 5, p. 455-457.
Lucia, FJ., and Conti, RD., 1987, Rock fabric, permeability, and log
relationships in an upward shoaling, vuggy carbonate sequence:
Texas Bur. Economic Geology, Geol. Circ. 87-5, 22 p.
Watney, W.L., Wong, lC., and French, lA., 1991, Computer simulation
of Upper Pennsylvanian (Missourian) carbonate-dominated cycles
in western Kansas (United States), in Franseen, E.K., Watney,
W.L., Kendall, C.G.St.C., and Ross, W., eds.: Sedimentary
modeling: computer simulations and methods for improved
parameter definition: Kansas Geol. Survey Bull. 233, p. 415-430.
Watney, W.L., French, lA., Guy, W.l, Carlson, R, and Wong, lC.,
1993, Geological characterization of oolitic grainstone reservoirs
in the Upper Pennsylvanian Lansing-Kansas City groups in
Victory Field and near-surface analog in Kansas, in Reservoir
description workshop: Energy Research Technology Transfer
Series 93-1, p. 159-190.
MODELING PETROLEUM RESERVOIRS 77

Watney, W.L, and others, 1994, Depositional sequence analysis and

sedimentologic modeling for improved prediction of
Pennsylvanian reservoirs, Final Report, U.S. DOE Contract
DOEIBC/14434-13, 212 p. + attachments, 212 p. +figs.
Youle, J.C., Watney, W.L., and Lambert, L.L., 1994, Stratal hierarchy and
sequence stratigraphy -- Middle Pennsylvanian, southwestern
Kansas, U.S.A., in, Klein, G.D., ed., Pangea: paleoclimate,
tectonics, and sedimentation during accretion, zenith, and breakup
of a supercontinent: Geol. Soc. America Spec. Paper 288, p. 267-
285.
Zeller, D.E., ed., 1968, The stratigraphic succession in Kansas: Kansas
Geol. Survey Bull. 189, 81 p.
 THERMAL MODELING AT AN ANCIENT
OROGENETIC FRONT
WITH SPECIAL REGARD TO THE UNCERTAINTY
OF HEAT-FLOW PREDICTIONS

U. Bayer, B. LiinenschloB, 1. Springer, and C. v. Winterfeld

GeoForschungsZentrum Potsdam, Potsdam, Genncmy

ABSTRACf

The geothermal field of the upper crust results from a variety of

interacting processes in combination with the transport properties of
rocks. The study attempts to illustrate how the various factors may inter-
act and how they may modify the thermal field. The main result focusses
on the uncertainties related to the extrapolation from the near-surface field
into greater depth. It is shown that a unique solution may not be
guaranteed. However, the modeling approach allows to weight different
scenarios, and perhaps may help to localize exploration wells at the most
informative point.

INTRODUcnON

The thermal field frequently is considered conductive, sometimes

with the addition of heat sources within the crust, which are the result of
radioactive decay. Toth (1962) illustrated that hydrostatic forces related
to topography may induce a deep-reaching pore fluid flow. The
importance of convective heat transfer has been discussed repeatedly,

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 79
80 BAYERAND OTHERS

especially with regard to sedimentary basins (Bethke, 1985; Bjoerlykke,

Mo, and Palm, 1988; Clauser and Villinger, 1990; Smith and Chapman,
1983). The examples usually are related to basin and graben situations
with a sedimentary filling. The anisotropy of the rocks (permeability,
thermal conductivity) is frequently ignored, a feature, which becomes of
special interest in tectonically deformed areas, such as orogenetic fronts.
We will illustrate by example how rock properties and especially
greater permeability zones may modify the thermal field down to several
thousands of meters at a regional level. In terms of a conductive heat
flow, the anisotropy of rocks is of minor importance as long as the tem-
perature distribution is considered. However, if one considers the vertical
heat flow, then the models indicate that the fluxes are affected by the
inhomogeneity of subsurface structures. Because the vertical heat flow is
one of the "measurables" in boreholes, the models can provide an impor-
tant tool to understand observed subsurface "thermal anomalies". Models
may allow the reduction of the uncertainty of predictions of the thermal
field, a problem that has occurred in the context of the location of wells,
which have been positioned with regard to anomalies of the geothermal
field, for example the KTB, Urach, and Soultz (Haenel, 1982; 10bmann
and Schulz, 1992; Kappelmeyer and others, 1991).
The paper resulted from an interdisciplinary approach. UB. de-
signed the project, B.L. is responsible for the conceptual model, J.S. did
the numerical modeling and designed appropriate programs, C.v.W. pro-
vided the basic geological model.

TIlE GEOWGICAL MODEL

The area under consideration is located near Aachen (Germany),

and all modeling is based on a cross section through the northern border
of the ancient Variscian orogenetic front that extends into the ancient
"molasse" basin. This geological cross section (Fig. lA) is part of a bal-
anced cross section, which was studied in detail by v.Winterfeld (1994).
It summarizes all the currently available surface and subsurface data.
Nevertheless, the cross section itself provides a geological model which
may change as our knowledge and our understanding of oro genetic pro-
cesses increase. A somewhat different view of the geological setting was
given by Knapp (1980). Within this study we take the geological model
of v. Winterfeld (1994) as granted because any future modifications may
THERMAL MODELING 81

modify only the regional aspects, but not the principal considerations of
this paper.

Wurm Syncline Inde nappe

.." .
NW SE

NN
.1

·1 '1

:~=~~:
.~
~
. ~

.,
.'~~~=~~~~ ...,
-8 IS] ~~:'zojc • Gedinnian
~

.
•'() L"S..'9 Givclian • Frasnian 0 Famennian [!]] Dinantian • Namurian • Westfalian ,,~ldl"3

NW SE

.1.,," .L..-_ _ _ _

12lt
----'-~----''___'--

S,:r.

Lithology number
___ ~ _ _ _ _ _ _ _ _ _ _ _ ____l.

o Direction 01 struClural anisotropy . ,....

Figure 1. A (top), Geological structure of Variscian front near Aachen (v. Winterfeld,
1994); B (bottom), conceptual model of Variscian front. Different main lithotypes are
indicated. Physical properties are summarized in Table 1 corresponding to material
number. Hatching indicates direction of first principal axis of conductivity tensors.

The cross section illustrates the main features of the Northwestern

Rhenish Massif with its thin-skinned tectonic setting of nappes, which are
related to the Aachen detachment and extend into the northern foreland
transition zone. The main tectonic structures have been formed by late
Variscian compressional deformation. Remarkable structural elements are
the low-grade metamorphic parts of the Eifel nappe in the southeast, the
repetitions and tectonic discontinuations of the stratigraphic units by blind
thrust imbrications, and the overthrusting and nappe implacement. Further
82 BAYERAND OTIIERS

structural elements are folding, cleavage, and the depression of the

marginal molasse area (for details see v. Winterfeld, 1994).
The lowermost strata consist of Cambrian and Ordovician units of
predominantly slates and quarzites that have been folded by the
Caledonian orogeny. Above these strata, a sequence of Devonian and
Carboniferous sedimentary units follows. The lower Devonian sequence
is represented by arenitic Old Red facies. In the Middle Devonian to
Lower Carboniferous a marine shelf facies developed with limestones,
claystones, and calcareous sandstones. Finally, in the Upper Carbonifer-
ous coaly molasse sediments were deposited. A description of the re-
gional geology is given by Knapp (1980).
The repeated stratigraphic sequences include alterations of various
lithotypes as well as tectonic foliation in several areas. Both structures
induce anisotropies of the physical properties. The transport coefficients
especially have to be considered as tensors with different values parallel
and perpendicular to layering and foliation (Ondrak, Bayer, and Kahle,
1994). Furthermore, hot springs at Aachen indicate increased
permeabilities along several stratigraphic units and fault zones, for exam-
ple detachment lines. The fault zones locally are related to Devonian
carbonates that may be highly permeable from karstification, even at
greater depth.

CONCEPTUAL AND NUMERICAL MODEL

The geological model discussed previously provides many strati-

graphic and tectonic details. However, in order to model transport pro-
cesses, geological information has to be translated into physical proper-
ties, and the scales have to be adapted to the software and hardware
capabilities, that is a conceptual model has to be developed (Welte and
Yalcin, 1987; Welte,1989). This transformation process introduces a
certain "roughness" to the geological model, which is dictated by the
modeling environment. In our special situation, we selected the main
tectonic units as the principal structural components (Fig. lB). On this
level, all the information about stratification and foliation is lost. To
overcome this problem "effective transport properties" have been intro-
duced that account for the modeling scale and the anisotropy of the rocks.
The direction of the first principal axis of the permeability and conduc-
tivity tensors is indicated in Figure lB.
THERMAL MODELING 83

The values for the transport coefficients parallel and perpendicular

to the stratification have been estimated as the weighted arithmetic and
harmonic means of the lithotypes involved. The weights are simply the
thicknesses of the strata. A more detailed discussion of such estimations
is given by Ondrak, Bayer, and Kahle (1994). For foliated units, labora-
tory data have been used to determine the transport tensors. The physical
property data used for all the models considered here are summarized in
Table 1. They are compiled from various sources and include data spe-
cific to the area as far as available (Schiirmeyer and Wohlenberg, 1985;
Bless and others, 1981; Haenel, 1983; Heitfeld and others, 1985), and
average data for the typical rock types as present in the literature (e.g.
Angenheister, 1982; Freeze and Cherry, 1979; ms, 1991; Somerton,
1992). These estimated physical properties certainly are a sensible point
of the entire model. However, a model is a simplification of nature, and
has to be modified if the specific knowledge increases. Within the context
of this study, the true values of the physical properties are of secondary
importance.

Table 1. Physical property data for lithotypes used in model calculation.

Distribution of mateiral parameters in model is indicated by lithology
number in Figure lB.

Lithology 1 2 3 4 5 6 7 8 9 10
number
U £W/KmJ 3.469 2.049 1.981 1.772 2.17 2.294 2.325 2.518 2.45 2.05
All £W/KmJ 3.565 2.08 2.065 2.041 2.22 2.344 2.371 2.538 2.475 2.05
kJ. [mDarcyJ 0.001 0.012 0.05 0.053 0.065 0.096 0.109 0.324 0.244 2.344*
kll [mDarcI,J om 0.228 0.252 0.305 0.464 1.056 1.299 1.754 2.366 2.344*
A [IlW/m J 1.306 1.404 1.389 1.296 1.278 1.03 0.938 0.719 0.537 -
* variable 20, 200, and 2000 mDarcy

The maximum thermal conductivity (3.6 W/Km) and the minimal

permeabilty (0.001 mDarcy) are present in the Cambrian and Ordovician
units. The lowest thermal conductivity (1.8 W/Km) occurs in the
Westfalian strata of the foredeep due to interbedded coaly layers. The
highest permeabilities are related to the carstic and dolomitic limestones
of the Devonian (2.4 mDarcy), and to the fault systems.
The control parameters of the models are the mean thermal gradi-
ents known from well data (Schiirmeyer and Wohlenberg, 1985;
Domstadter and Sattel, 1985; Graulich, 1969; Haenel and Staroste, 1988;
84 BAYERAND OTHERS

Hurtig, 1992), and data from the hot springs at Aachen that have been
studied in detail by Pommerening (1992) and Langguth and Plum (1984).
The geochemical studies indicate an origin of these springs at 120-130°C,
a control parameter which has to be approached by the model.
The calculations have been accomplished by finite element meth-
ods using commercial software as well as especially designed and devel-
oped software which accounts for the coupling of heat conduction and
forced convection. At the present state of the modeling approach we
ignore possible effects of free convections, which could be induced by
density variations of the pore fluid under the variable depth dependent
Pff-conditions. Finally, it is assumed that heat conduction and pore fluid
flow are in a stationary state. In this situation the coupled system of par-
tial differential equations has the following form:

Darcy flow q =-!(grad p + pgez )

Jl
Mass balance div q = -di{ ~ (grad p + pge,)) = 0
Energy balance div{A grad T) - P c q grad T + Q =0

where
q Darcy flow rate,
k penneability tensor,
Jl viscosity,
p pore fluid pressure,
p pore fluid density,
g acceleration of gravity,
ez vertical unit vector,
A thermal conductivity tensor,
T temperature,
c specific heat of pore fluid,
Q heat production.

Another problem in a two-dimensional finite scenario is the

boundary conditions. They effectively determine the special solution of
the partial differential equations, and the models may be biased by
boundary effects. For our models, we select a constant average annual
surface temperature of 10°C and a zero hydrostatic pressure along the
THERMAL MODELING 85

surface. At the northern and southern faces. the boundaries are closed for
both types of fluxes. These closed boundary conditions provide some
disturbances in their vicinity, however in the southern region, the
permeability of the rocks, in generally, is extremely low so that the
boundary effects remain small. At the northern end, the section actually
approaches a fault that may be considered closed and impermeable.
Towards the lower boundary at 10 km depth we assume that the
permeability approaches zero while a heat flux is entering from beneath.
This heat flux has to be determined for different variants of the models
in order to approach the control data.

MODELING RFSULTS

Figures 2 to 6 illustrate the modeling results within the area with

regard to effects of heat conduction, heat production, and hydrostatically
forced convection.

10 35 60 85 - - - --
110 135 160 185 210 235°C

Figure 2. Two models with conductive heat transport. A (top), Simple conductive heat
transport with basal heat-flow density of 60 mW/m2;B (bottom), heat conduction com-
bined with heat production corrresponding to data given in FigurelB and Table 1 and
basal heat-flow density of 47.5 mW/m2•

Figure 2A illustrates the classical approach of basin analyses (e.g.

Yukler, Cornford, and Welte, 1978) with a heat flux entering the area
86 BAYERAND OTHERS

from the lower boundary (60mW/m2). The isotherms resulting from this
model reflect the variable conductivities of the different rock types.
However, in this simple setting, a series of one-dimensional models also
would provide acceptable results as long as we consider the temperature
field only. But the anisotropies become of importance if the heat fluxes
are considered, as will be discussed next.
The model of Figure 2A implies a nearly linear thermal gradient,
which is slightly modified by the thermal conductivities of the various
rock types. However, its extrapolation to greater depths would collide
with our understanding of the temperature at the lithosphere/astenosphere
boundary. The classical method to overcome this discrepancy is to as-
sume heat production within the crust. Figure 2B illustrates how the ther-
mal field is modified if radioactive decay is assumed to be a source for
heat production. In order to adapt the near-surface heat-flow field to the
control data, the heat flow at the lower boundary has to be reduced to
about 47.5 mW/m2. The temperatures increase now less rapidly with
depth, the deeper parts are cooler, and the gradient declines with depth
as required by geophysical reasoning.
The shortcoming of the conductive models is that they cannot ac-
count for the hydrothermal springs at Aachen. In order to have hot
springs at the Aachen thrust one has to tum on pore water fluxes. Figure
3 gives an impression of the flow system that arises if the permeabilities
along the fault systems are set to 2.4 mDarcy, a rather small value. Here
the influence of the anisotropy becomes clear, the pore flow directions
correspond closely to the layering (cf. Fig. IB).
In Figure 4 we summarize the shifting of the thermal field
resulting from forced convection for two different permeabilities along the
fault systems. As required, the isotherms bend upward near the Aachen
thrust fault. In the recharge area, in contrast, the isolines bend downward
causing a cool upper crust, a feature that is known from this area
(Graulich, 1969). The heat flow at the lower boundary is the same as in
the simple conductive model of Figure 2A. In parts, it may have to be
adjusted as our knowledge about surface heat fluxes increases.
A critical review of these modeling results shows that they do not
yet reflect the details of the hydrothermal conditions at Aachen. Geo-
chemical thermometers indicate that the water of the springs was heated
originally to about l20-BO°C. In our models the best results are achieved
with permeabilities between 20 to 200 mDarcy along the fault system. In
this situation the permeable zone is located near the 100 to 110°C
THERMAL MODELING 87

Figure 3. Flow field of hydrostatically forced pore water flow. Permeability along
faults is set to 2 mDarcy.

10 35 60 85 - -- -
110 135 160 185 210 -
235 °C

Figure 4. Thermal field under influence of hydrostatically forced pore fluid flow. Ther-
mal boundary conditions are same as in Figure 2A. Permeabilities at fault zones are 200
mDarcy (top) and 2 Darcy (bottom), respectively.

isotherm. The model can be adapted further to the control data by

introducing a variable heat flux at the lower boundary of the model. In
Figure 5 the basal heat flow increases from 60mW 1m2 at the northern
boundary to 85 mW/m2 at the southern boundary. This boundary condi-
88 BAYERAND OTIIERS

tion lifts the thermal field towards the surface and the possible source
area of the springs now is located near the 120°C isotherm, whereas the
surface heat fluxes are still near the observed values. The trend in basal
heat flux now parallels the trend in near-surface heat fluxes, however, it
is contrary to the near-surface mean thermal gradient, which in the model
as in nature decreases towards the south. These complex patterns result
from the complex interaction of basal heat flux, thermal conductivity, and
pore fluid flow. Although the model may not approach reality, it
approaches the complexity of reality and indicates that control data from
a deep borehole (Walter and Wohlenberg, 1985) would be necessary to
proof or improve the model.

10 35 60 85 -- ----
110 135 160 185 210 2350C

Figure 5. Thermal field under influence of forced pore fluid flow (as in Fig. 4A, 200
rnDarcy along faults) and variable basal heat-flow density (60 mW/m1 at northern
boundary, increasing to 85 mW/m1 at southern end).

UNCERTAINTIF.S IN PREDICflNG 11IE GEOTHERMAL FIELD

At the end of the last section we were faced with the fact, that
there is no definitive test for even a "best fit" model if nonlinear interac-
tions are considered. In this section we will consider the uncertainties
associated with extrapolations to depth in some detail.
One of the main control parameters is the near-surface field of
heat fluxes. One could assume that the degrees of freedom can be reduced
as the knowledge of the near-surface field increases. One possibility
would be to locate a sequence of shallow wells along the profile, which
then are studied in detail with regard to the heat flux by separating it into
its conductive and convective components (Clauser and Villinger, 1990).
Next, we will illustrate that this approach may fail if the medium is
inhomogeneous and anisotropic.
THERMAL MODELING 89

Figure 6 shows the vertical heat flux for the two conductive
models with and without heat production as given in Figure 2. The heat-
flow representation corresponds to a high-pass filter that is applied to the
thermal field, thus it enhances small variations. In our example, these
variations are the result of the different thermal conductivities of the
rocks, although there may be numerical errors that confuse the patterns -
an aspect which will not be considered within this context.
Figure 6 illustrates that the two models may be calibrated so that
the near-surface heat-flow pattern agrees almost perfectly. Correspondence
was achieved simply by modifying the heat input at the lower boundary,
as previously discussed. Down to about 1000 m the heat-flow patterns are
so similar, that they actually cannot be distinguished. The same is true for
the near-surface temperature field down to about the same depth as can
be illustrated by taking the difference between the temperatures of the
two models. Based on current knowledge of near-surface heat fluxes, the
two models are indistinguishable. This result reflects our current
uncertainty about the temperature distribution in the upper crust.
More important is that in a one-dimensional well that would be
drilled into the model's scenario the inhomogeneities caused by the
anisotropic rock properties easily could be misinterpreted as convective
effects so that the attempt to separate conductive and convective compo-
nents could fail. Studies and interpretations of the thermal field, therefore,
should be accompanied by numerical models. The modeling approach
allows to test the sensitivity of the system with regard to various types of
disturbances and allows to locate areas with the highest potential for
relevant additional information.

CONCLUSIONS

Geothermal studies usually rely on one-dimensional models and

on the concept of heat conduction. Most geothermal maps are compiled
on this basis (Haenel and Staroste, 1988; Hurtig, 1992). In our study,
modeled conductive heat transport results in variations of temperature
gradients and heat-flow densitiy. In addition, forced convective and heat-
generating processes can transform strongly the temperature distribution
calculated from thermal conduction. The models allow to separate and
investigate the effects of thermal conduction, forced convection and
90 BAYERAND OTHERS

45 . 5 4, 8
-- - - - -
. 5 51 . 5 54 . 5 57 . 5 60 . 5 63 . 5 66 . 5 69 . 5 72 . 5 mi'! / m-

Figure 6. Vertical heat-flow densities for conductive models of Figure 2. A (top), No

heat production, basal heat-flow density 60 mW/m2; B (bottom), heat production ac-
cording to Table 1, basal heat-flow density 47.5 mW/m2.

radiogenetic heat generation, which occur coupled in nature. On the other

hand, the attempt to model the geothermal field at an ancient orogenetic
front indicates that near-surface data are not sufficient to understand the
thermal structure of deeper parts of the upper crust. Identical surface heat-
flow density can be derived from different temperature fields.
Heat conduction and forced convection provide an interacting sys-
tem. The complexity of the models can be increased if the transport
coefficients depend on the PfI conditions, and if free convection is
introduced as an additional force into the system. In such highly nonlinear
systems, we may enter a "new world" of possible phenomena.
Besides the uncertainties the models allow to study various
scenarios in advance to the location of a well. Although the models do
not allow a definite prediction of the thermal subsurface structure, they
enhance the differences between various alternative scenarios. These
differences may help to localize exploration wells at points with the
highest information content.
THERMAL MODELING 91

ACKNOWLEDGMENTS

The study is part of a project funded by the "Bundesministerium

fur Forschung und Technologie" (BMFT), Projekt No. 0326761A.

REFERENCFS

Angenheister, G., ed., 1982, Landolt-Bomstein. Numerical data and

functional relationships in science and technology. Physical properties
of rocks, v. 1a: Springer-Verlag, Berlin, 373 p.
Bethke, C.M., 1985, A numerical model of compaction-driven ground-
water flow and heat transfer and its application to the paleohydrology
of intracratonic sedimentary basins: Jour. Geophys. Res., v. 90, no.
B7, p. 6817-6828.
Bjoerlykke, K, Mo, A., and Palm, E., 1988, Modeling of thermal convec-
tion in sedimentary basins and its relevance to diagenetic reactions:
Marine Petrol. Geology, v. 5, no. 4, p. 338-35l.
Bless, M.J.M., and others, 1981, Preliminary report on Lower Tertiary -
Upper Cretaceous and Dinantian-Famennian rocks in the boreholes
Heugem-lila and Kastanjelaan-2 (Maastricht, the Netherlands):
Meded. Rijks Geol. Dienst, N.S. 35-15, p. 333-415.
Clauser, c., and Villinger, H., 1990, Analysis of conductive and convec-
tive heat transfer in a sedimentary basin, demonstrated for the
Rheingraben: Geophys. Jour. Intern., v. 100, no.3, p. 393-414.
Dornstadter, J., and Sattel, G. ,1985, Thermal measurements in the
research borehole Konzen, Hohes Venn (West Germany): N. Th.
Geol. Palaont. Abh., v. 171, no. 1-3, p. 117-130.
Freeze, R.A., and Cherry, lA., 1979, Groundwater: Prentice-Hall, Inc.,
Englewood Cliffs, New Jersey, 604 p.
Graulich, J. M., 1969, Eaux minerales et thermales de Belgique: XXIII
Intern. Geol. Congress, v.18, p. 9-15.
Haenel, R. 1982, The Urach geothermal project (Swabian Alb, Germany):
E. Schweizerbart'sche Verlagsbuchhandlung, Stuttgart, 419 p.
Haenel, R., 1983, Geothermal investigations in the Rhenish Massif, in
Fuchs, K, Gehlen, K von, Malzer, H., Murawski, H., and Semmel,
A. eds., Plateau uplift. The Rhenish shield - a case history: Springer
Verlag, Berlin, p. 228-246.
92 BAYERAND OTHERS

Haenel, R, and Staroste, E., eds., 1988, Atlas of geothermal resources in

the European Community, Austria and Switzerland: Verlag Th.
Schaefer, Hannover, 74 p.
Heitfeld, K.-H, Voltz, H, Yii, S., Kopf, M., and Krapp, L., 1985,
Geotechnical rock tests research borehole Konzen, Hohes Venn (West
Germany): N. lb. Geol. Palaont. Abh., v. 171, no. 1-3, p. 195-205.
Hurtig, E., ed., 1992, Geothermal atlas of Europe: Haack Verlag, Gotha,
165 p.
IES (Integrated Exploration Systems) GmbH, 1991, PDIIPC I-Dimen-
sional Simulation Program, User Manual Version 2.2: liilich, Ger-
many, 264 p.
lobmann, M., and Schulz, R, 1992, Temperatures, fractures and heat
flow densitiy in the pilot-hole of the Continental Deep Drilling
Project (Oberpfalz, Germany): Scientific Drilling, v. 3, no. 1-3, p. 83-
88.
Kappelmeyer, 0, and others, 1991, European HDR project at Soultz-sous.
Forets, general presentation: Geothermal Science and Technology, v.
2, no. 4, p. 263-289.
Knapp, G., 1980, Erlauterungen zur geologischen Karte der nordlichen
Eifel 1 : 100 000: GLA Nordrhein-Westfalen, Krefeld, 155 p.
Langguth, H-R, and Plum, H, 1984, Untersuchung der Mineral- und
Thermal quell en der Eifel auf geothermische Indikationen: BMFT-
Bericht T84-019, Nichtnukleare Energietechnik, 176 p.
Ondrak, R, Bayer, u., and Kahle, 0.,1994, Characteristics and evolution
of anisotropic geologic media, in Kruhl, J., ed., Fractals and dynamic
systems in geosciences: Springer Verlag, Berlin, p. 355-367.
Pommerening, 1., 1992, Hydrogeologie, Hydrogeochemie und Genese der
Aachener Thermal quell en: unpubl. doctoral dissertation, RWTH
Aachen, 169 p.
Schiirmeyer, 1., and Wohlenberg, 1., 1985, The anisotropy of the thermal
conductivity of Ordovician rocks from the research borehole Konzen,
Hohes Venn (West Germany): N. Jb. Geol. Palaont. Abh., v. 171, no.
1-3, p. 131-143.
Smith, L., and Chapman, D.S., 1983, On the thermal effects of ground-
water flow, 1. Regional scale systems: Jour. Geophys. Res., v. 88,
no. Bl, p. 593-608.
Somerton, W.H, 1992, Thermal properties and temperature-related
behavior of rock/fluid systems: Elsevier Science Publishers B.V.,
Amsterdam, 257 p.
THERMAL MODELING 93

Toth, 1., 1962, A theory of groundwater flow in small basins in central

Alberta, Canada: Jour. Geophys. Res., v. 67, no. 11, p. 4375-4387.
Walter, R., and Wohlenberg, J., 1985, Proposal for an ultra-deep research
borehole in the Hohes Venn Area (West Germany): N. Jb. Geol.
Palaont. Abh., v. 171, no. 1-3, p. 1-16.
Welte, D.H., 1989, The changing face of geology and further needs:
Geologische Rundschau, v. 78, no. 1, p. 7-20.
Welte, D.H., and Yalcin, M.N., 1987, Basin modeling - a new compre-
hensive method in petroleum geology, in Mattavelli, L., and Novelli,
L., eds., Advances in organic geochemistry: Organic Geochem., v.
13, p. 141-151.
Winterfeld, C. von, 1994, Variszische Deckentektonik und devonische
Beckengeometrie der Nordeifel - Ein quantitatives Modell
(profilbilanzierung und Strain-Analyse im Linksrheinischen
Schiefergebirge): unpubl. doctoral dissertation, RWTH Aachen, 309
p.
Yukler, M.A., Comford, C., and Welte, D., 1978, One-dimensional model
to simulate geologic, hydrodynamic and thermodynamic development
of a sedimentary basin: Geologische Rundschau, v. 67, no. 3, p. 960-
979.
 EFFECTIVE TRANSPORT PROPERTIES OF
ARTIFICIAL ROCKS -
MEANS, POWER LAWS, AND PERCOLATION

O. Kahle and U. Bayer

GeoForschungsZentrum Potsdam, Potsdam, Germany

ABSTRACT

Two-component systems provide simple idealized models for com-

plex anisotropic geological structures. Such artificial media are studied in
terms of their "effective" transport coefficients. It is shown that means
and power laws provide empirical descriptions for the transport tensor and
relate it to statistical measurements of the composition and structure of
the medium. General anisotropic media are related to statistically isotropic
media. The isotropic medium is well described by an iterative mean for
low contrasts of conductivities. This iterative mean is derived from the
arithmetic and harmonic means. For high conductivity contrasts rules of
percolation theory come to application which expand the classical
concepts of percolation theory. At the percolation threshold both accesses
fall together, as is illustrated by appropriate power laws for the effective
transport coefficients.

INTRODUcnON

The estimation of "effective" transport properties is a classical

problem associated with modeling of transport processes in porous rocks.
In nature, transport properties differ within a wide range of length scales
which cannot be incorporated in numerical models. In addition, most

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 95
96 KAHLE AND BAYER

rocks are anisotropic on a large scale, an aspect which should be included

into models because it modifies the flow path. Thereby, the dimension of
the model is of secondary order, it occurs in microscopic pore models as
well as in megascopic basin models.
Generally, there are two ways of modeling phenomena in rocks.
The first approach is based on the classical equations of transport or on
the continuum model. The second one is based on network models of
pore space and fractured rocks, that is the models are at the smallest
scales with regard to the physical properties. Based on these microscopic
models modem concepts of the statistical physics of disordered systems
can be used to derive the macroscopic properties of the system. In this
context the concepts of fractals and percolation play important roles
(Sahimi, 1993).
A typical way to represent the average effects of small-scale flow
in coarse grid reservoir or basin models has been termed pseudo relative
permeabilities (Kyte and Berry, 1975; Stone, 1991). These "effective"
permeabilities usually are generated by a stepwise coarsening process.
The results derived from fine grid models then are used as input for
coarser grid models. A variety of techniques has been used in the past to
achieve the coarsening process. These techniques range from simple
statistic averages which are inaccurate in real applications to detailed
numerical simulations that are consuming much computation time. An
alternative strategy is real-space renormalization, where the effective
property is calculated recursively for increasingly coarser grids from small
groups of cells. This process continues until the entire grid is reduced to
a single block (King, Muggeridge, and Price, 1993).
The aim of our work was to determine simple rules from detailed
simulations which allow an easy calculation/estimation of the effective
transport property derived from two-component models which resemble
sedimentary sequences. The property under consideration can be different
as long as the basic transport equations are the same, for example
permeability, heat conductivity, and diffusion coefficients are all related
to the Laplace equation under stationary transport conditions.
The paper provides power laws for the effective transport
coefficients which are derived from a pseudoization method which is
applied to a domain in space consisting of two materials with distinct
properties. Special attention is drawn on anisotropic distributions of the
materials within the domain, and a simple algorithm allows a change in
EFFECTIVE TRANSPORT PROPERTIES 97

the internal structure of the distribution from an isotropic to a perfectly

layered system.

GEOMETRIC MODELS OF ARTIFICIAL ROCKS

Figure 1 illustrates a set of artificial, computer-generated

sedimentological patterns or rocks. These artificial rocks form the base for
the modeling approach. The frequency of both components is identical
within all blocks, however, the ordering (stratification, lamination, or
layering) increases from the upper left to the lower right corner. The
generation process is rather simple: square planes (bricks of height 1 and
side lengths l) are distributed randomly within the modeling area, a 3-
dimensional cube of size 603 'pixels'. If the 'bricks' have a side length 1
= 1 (1 pixel), then the medium is statistically homogeneous and equiva-
lent to site percolation models (Stauffer, 1985). In this special situation
the concepts of percolation theory such as percolation threshold, cluster-
size distribution, and fractal properties can be applied.

.
". ~.'

t~WA~ ·
.. .. ,
~.:-', ~'''')
:x;:,.,;.
.
': ·..·.t.
'~I-' .' .
~
'.'I~\
~ .

. ' .:..tit
!' ~~~:
........ .... .

"

Figure I. Examples of generated artificial media with I: 1 ratios of two components.

Structural anisotropy increases from upper left to lower right comer, generating lengths
1,2,5, 10,20, and 50 pixels.
98 KAHLE AND BAYER

However, if planes of sizes 1 > 1 are used, the properties of the

artificial media change. The main reason is that some of the randomly
placed objects may overlap. These overlappings are counted once in
determining the density distribution. The "events", therefore, are not
independent, and the filling mechanism is somewhat modified. Within this
study we will consider only fillings with squared plane elements, a
process which provides structures similar to sedimentological patterns.
The two components within the model area should not be considered as
pore space and rock matrix, but rather as the distribution of two
lithological types such as clay and sandstone, or as two mineral phases
with different conductivities. In natural systems the pore systems remain
connected at porosities much below the percolation threshold of our
model. In the situation of pore models, other approaches, such as the
grain consolidation model (Schwartz and Kimminau, 1987), are more
successful. Another way would be to replace our site oriented approach
by a bond-oriented one (Wong, Koplik, and Tomanic, 1984).

UI
~o~-----------------------------------------------.
SO 20 10 S 2 I
20
--- - ---+.;-- - ' - ~
length of generating line
10

IL-----~ __ ~ __________ ~ ____ ~ ________ ~~~~~~

o os 0 ,1 02 03 os

1-p

Figure 2, Power law dependence of characteristic length L from frequency of higher

conductive material (P2) and different generating lengths I.

The medium discussed can be characterized statistically by

introducing the mean length Lj.; for continuous sequences of the two
materials j=I,2 in every direction i=x,y,z. The generating lengths 1;
(lengths of the square planes) and the mean lengths L; are correlated by
a power law (Fig. 2):

(1)

whereby
EFFECTIVE TRANSPORT PROPERTIES 99

(2)

The parameter a is a characteristic constant (a=O.58 .. ) for this type

of model. The parameter 10 is the basic unit for the grid, for example one
pixel. If the generation length equals 10 in all directions, then the equation
simplifies to

(3)

The power in Equation (1) expresses the fact that the medium may
be rescaled to the statistically isotropic medium [Eq. (3)] if it has been
generated by statistically independent events (no overlapping),
independently of the size of the objects used. Note that there is no cross
dependence between the different directions and the mean lengths. Both
components are distributed symmetrically concerning the frequencies of
the components [Eq. (2)], but they are not symmetrical with regard to the
distribution of the lengths itself. The geometry, in addition, can be
characterized by some invariant parameters such as the inner surface S
and various measurements of anisotropy which have been discussed in
Ondrak, Bayer, and Kahle, (1994). These parameters can be helpful in
order to relate the model parameters to measurable data.

CALCULATION OF EFFECfIVE TRANSPORT COEFFICIENTS

In the situation of a stationary transport in porous media, the

mathematical description may simplify to the Laplace equation which
accounts for fluid flow, thermal conductivity, and diffusion processes in
homogeneous media. For our heterogeneous medium we will consider
fluid flow as a special situation, and we assume that Darcy's laws holds
in absence of gravity effects (Fick's law for the diffusion processes):

q=-KVp (4)

where q is the Darcy velocity [m/s], K the coefficient of permeability (K

= klh), k the intrinsic permeability of the rock [cm2], h the dynamic
viscosity of fluid [Poise, kg/ms], andp the pressure potential [Pa, kg/ms2].
The equation for conservation of mass (continuity equation) is
100 KAHLE AND BAYER

(<PP ) =V .pq + e (5)

with P density of fluid [kg/m3], ~ porosity of rock, e generation density of

fluid [kg/m3 s]. For incompressible fluid and constant temperature in the
domain, that is p = const. and for the absence of sources, e = 0, this
equation reduces to

V'lj=O (6)

Together with (4) we get the elliptical differential equation

V . (XVp) =0 (7)

which has to be solved together with appropriate boundary conditions.

Given a solution of Equation (7), the effective permeability can be com-
puted by use of Equation (4). These modeling results resemble laboratory
measurements and provide the database for the approximate methods
discussed next.
If K is a known function of the space variable, the usual theory of
linear elliptic equations can be applied. However, the geometrical gener-
ation process (Fig. 1) leads to a random field of K -values. The perme-
ability at each site corresponds to the material component at that site.
This progress initially provides two delta-functions describing the fre-
quency distribution of K in space. For further investigations it is easy to
generalize the delta-distributions to log normal distributions in order to
incorporate fluctuations of the conductivity within each material compo-
nent. On the other hand, it is known that for broad distributions of con-
ductances the transport is dominated by those conductances with magni-
tudes greater than some characteristic value, at which the set of conduc-
tances forms an infinite connected cluster. Transport in such systems
reduces to a percolation problem with this characteristic value and,
therefore, resembles the approach by delta functions.
We use a finite difference method to solve (7) and to calculate the
average effective value of K for the entire modeling domain. The flow is
driven by a constant pressure gradient between two opposite faces of the
cube. The other boundaries are unpermeable. This definition of the
boundary conditions allows an easy calculation of K for the stationary
situation, and Equation (4) provides a tool, in addition, to prove the
EFFECTIVE TRANSPORT PROPERTIES 101

quality of the numerical solution. The total flux has to be constant

through any cross-section perpendicular to the flow direction.
The initial conductivities along a bond between two sites is taken
as the harmonic mean of the values at the sites, an approach used in flow
models for porous media (Kinzelbach, 1986). Figure 3 gives an example
of the calculated pressure potential for such a stationary solution and
illustrates how the anisotropic properties of the rock modifies the flow
pattern.

A B

Figure 3. Example for stationary solution of transport problem. Distribution of two

materials is given in A, potential is elucidated in B. Flow shows typical features of
fingering. Ratio of conductivities is 100: 1, and black areas in A indicate component of
higher conductivity.

The calculations were made for different frequencies of the compo-

nents, a wide range of conductivity contrasts (several orders), and various
structural anisotropy. The principle axes of the transport tensor K are
determined by calculating the flow normal and tangential to the
geometrical structure. These calculations are sufficient because of the
symmetry of the planes used to generate the layering. Some calculation
results are given in Ondrak, Bayer, and Kahle (1994), by example, which
elucidate the influence of anisotropy.

REPRESENTATIVE VOLUMFS

In order to achieve accurate results with a minimum computation

effort the modeling domain has to be sufficiently large to keep the statis-
102 KAHLE AND BAYER

tical fluctuations small enough. The error related to the size of the do-
main should be of equal order similar to the error of the numerical flow
calculations. Figure 4 shows frequency distributions of the calculated
effective conductivities for modeling domains which range from size 603
down to a single site (size P) for a statistically isotropic and an aniso-
tropic medium. It becomes clear that a cube size of 13 produces two
delta-functions located at the two conductivities associated with the two
materials. For larger domains the distribution broadens and then fuses into
a single peaked distribution around a well defined mean value. Finally,
the distribution transforms into a single peak with small variance. But the
isotropic and anisotropic situations are distinguished by a significant
difference in the representative domain size for which the deviations from
the effective value are negligible.

' ,OJ t." , .1. 1.21 '.1' ,.» .,.,. ...,

B conductivity

Figure 4. Two sets of frequency distributions of effective conductivity for isotropic

medium A and anisotropic one B based on different domain sizes. Conductivities of two
components are kJ = 0.05 and k] = 0.95; frequency of component 2 is p] = 0.4. For
anisotropic medium generating length is I = 10, and flow is parallel to structure (com-
pare to Fig. 3).
EFFECTIVE TRANSPORT PROPERTIES 103

In the isotropic situation a domain size of 323 is sufficient to

equalize the random fluctuations of the generation process. An anisotropic
structure requires a larger domain size to reach the same balancing. For
the example shown in Figure 4B, the standard deviation of the effective
value will be lower than 1% if the domain sizes are greater 50 3 (Fig. 5).
It can be expected that the required domain sizes are proportional to L 2 j'
The flow calculations and the determinations of K used below were ail
done with a domain size of 60 3 • This is a good compromise between good
statistics and calculation time.

0 .• 5

• • • •
0 .• 0 C slandard de'llalion

.~ • mean conducllvity
•
• • • • • •
0.:15
·s'"
• •
-
.a
~
·s
'p
g
0.30

0.25

D.2O
• •
."
§ 0.15
Ii 0.10
E
0 .05

0 .00
&0 50 <0 32 25 20 16 12 10 8 6 3
bi~ap size

Figure 5. Standard deviation and effective conductivity for anisotropic medium of Figure
4B for different representative volumes.

MEANS AND POWER LAWS FOR EFFICIENT TRANSPORT

COEmCIENTS

A general aim is to describe the effective mean conductivity by

simple relations in terms of measurable features of the medium. The
effective conductivity, per definition, is some type of a mean taken over
the different subdomains under consideration. Here, we will illustrate that
this concept together with basic measurables, similar to the frequency of
materials and simple measures of structural anisotropy (L), is sufficient
for an approximate description of the effective mean conductivities.
First, we consider the most simple situation of a medium of infinite
structural anisotropy, that is the medium is completely layered. This
situation is understood fully, the medium can be imagined as consisting
of resistors all in series for flow perpendicular to the layering and in
104 KAHLE AND BAYER

parallel for flow in direction of the structure. The thickness of the layers
is unimportant, and the effective conductivity K can be approximated by
the weighted arithmetic and harmonic means:

K = P1k1 + Pi2:: A (for flow II structure), (8)

(9)

These values provide upper and lower bounds for the effective
conductivity .
Another situation considered is the statistically isotropic medium
which will be discussed in some detail. A number of formulations have
been inventoried for the effective conductivity (Prakouras and others,
1978) with contrary results in several instances. These equations incor-
porate many effects and corrections which are not suitable to describe our
artificial medium. We propose the following access: An isotropic medium
can be imagined as a mixture of areas of complete layering where flow
is either normal or tangential to the structure. Thus, we expect that the
effective conductivity is a mean bounded by Equations (8) and (9). Such
a effective conductivity can be constructed by the following iteration
process (Beckenbach and Bellmann, 1965; Schoenberg. 1982):

n - f(k n-l k n-l P ) = H

k1 - 1'2'2- (10)
k 2 - g(k1
n - n-1 k n-1 P ) = A
'2'2-
(11)

k1n-1 < k1n < k2n < k2

n-1
'
(12)

with the limit

lim k1n = lim k2n ::M(k1,k2,P2) (13)

~oo n-+oo
EFFECTIVE TRANSPORT PROPERTIES 105

This iteration process provides some analogy to real space normalization,

however, the initial means are taken over the entire domain.

For equal frequencies of the two components (PI = P2 = 0.5) the

iterative mean M is equal to the geometric mean G = k/lk/2 (see
Schoenberg, 1982, for a theoretical derivation). However, for different
ratios of the components there are small, but significant differences
between M and G, whereby M provides a better approximation for the
effective conductivity than the usual proposed geometric mean. This
averaging technique provides sufficient approximations only at low con-
ductivity contrasts between the two components. Figure 6 illustrates the
relative error between the computed effective conductivity and the ap-
proximation by the iterative mean. Obviously, the conductivity contrasts
should be lower than 20: 1 in order to keep the error below 2%. Surpris-
ingly, the relative error vanishes at the percolation threshold for all

... ..... "'.................. "' ............... ..

0.8

.
0.6

0 .4

.... ~ .. _ .... , .. , .......M···... - ... --."rc ........... ............ t<..

02
.X"".n -"
~ 0 .0. " --"~~.'-"'"
:::e At~ '!! .. ........... t:::::~ ::: . .... - -
__ "
• ••• " ...... .. .... . .jl:' " •• ••
~
ralio 01 condudiv ity k, : k,
-0.2
___ '3
-0.4 < 1.10
· · · ~ -·' : 100
____ ' ; I00J
-0.6 .......... ,·,0000
-+-- , , 'o(x)co
-0.8 P. - 1: 1000000
- 1.1CXXXJXO
.. ,,> ... 1:-
.1.0+--~~l,--_---+-I1~-o----+-----<---------~
a 0.1 0 .2 0 .3 0 .. O. ~ 0 .6 0 .7 08 09
portion 01 hlghel conWCliv componenl p,

Figure 6. Relative difference between calculated conductivity and iterative mean for
different ratios of conductivity.

conductivity contrasts. At this point the effective conductivity K is

exactly described by M (Fig. 7)

(14)
106 KAHLE AND BAYER

~ 1. oE..J
ral lO ~ COf'lOvdl\l'lly 1c: , kJ,
~ _ _ 1J
s:
____ 1,'0
~ LOE""
1.100
.. 10(- •• , 1000
___ 'I(,'I()IX)

........ , IOOOCO

'''''.
-+-, 1000000
- l .' OCIOCX)OO

---,;-
I QE·7l--.....,c:.-+-----..L--+-Z---_ _- + -_ _ _- - + - - - + - - -_ _ ~

0' 02 OJ 04 O~ 05 07 o. o.
pc:1f1.IQfl ot hlghQr condudlv CO'nponQnI P,

Figure 7. Normalized effective conductivity calculated from flow model for different
ratios of conductivities.

At the percolation threshold the statistically isotropic medium is known

to have a fractal structure, and the correlation length of the clusters
formed by material 2 becomes infinite. At this point the assumption of a
mixed medium as defined previously seems to be valid. This fact could
be an interesting field for further investigations, at this point the result
provides an additional motivation to analyze the effective conductivity in
terms of means.
The approximations by means fail at some distance of the
percolation threshold and for high conductivity contrasts. The medium
then is better described by the concepts of percolation theory. It is
obvious that simple means cannot explain effects such as the percolation
threshold - if one component is a perfect isolator the means would be
zero independent of the frequency of the material. In this special example
our model structure corresponds to the standard medium discussed in
percolation theory, and the average conductivity above a typical threshold
obeys a power law equation:

(15)

(16)
EFFECTIVE TRANSPORT PROPERTIES 107

with Pc - 0.317 and D - 1. 72 for our models. Percolation theory predicts a

site percolation threshold of 0.3116 for the simple cubic lattices (Stauffer,
1985). The difference, between the theoretical value and the one we
determined may be explained by calculation errors and the fact that the
size of the representative volume increases theoretically to infinity at the
percolation threshold (pc). Formulae such as Equation (15) have been
discussed repeatedly in a similar context (Staufer, 1985; Peitgen, JUrgens,
and Saupe, 1992). Actually, the identical expression was determined for
a percolation model derived from log normal distributions by hard
clipping.
Percolation theory is related to the situation of a 0: 1 conductivity
contrast. It describes a singular situation similar to the iterative mean M
[Eq. (14)]. In geological applications one may be faced with two or even
more components with finite conductivities. If both components are
conductive in our model, however, with high conductivity contrasts
(>20:1, see previous discussion) then Equation (15) has to be modified.
We determined that the following equation is a sufficient approximation
to describe a generalized isotropic medium if the frequency of the mate-
rial with higher conductivity (P2' k 2) is above the percolation threshold:

,p> Pc (17)

The indices are selected in a way so that the higher conductive material
has the index 2. Pc is the percolation threshold for material 2. K'(pJ
denotes the normalized effective value at Pc:

(18)

In this example, the exponent D is not a single constant, but it is a func-

tion of conductivity contrast. The terms (k2-k1) and k1 scale the term in
square brackets into the interval 0-1. If the conductivity contrast increases
towards infinity, Equation (17) approaches asymptotically Equation (15),
that is the percolation case. This illustrates that the classical percolation
situation really should be taken as a singular situation which, however,
can be accessed by taking the limit toward an infinite conductivity
108 KAHLE AND BAYER

contrast. Although Equation (17) describes the effective conductivity for

finite conductivity contrasts, it depends on the percolation threshold. The
discussed limit has to be taken from frequencies above the percolation
threshold. The formula is only valid for P2 > Pc. We have not determined
yet a similar simple transformation for the region below the percolation
threshold. It seems that the limit from the area below the percolation
threshold is of a different nature than the limit from above the percolation
threshold, so Pc may mark a real discontinuity even in the generalized
situation discussed here.
At last we consider the generalized anisotropic medium. It may be
understood as an intermediate medium between the extreme situations of
the statistically isotropic medium with the effective conductivity I and the
perfectly layered medium with its effective conductivities A and H. The
mean I can be calculated from Equation (17) as far as this equation is
valid, or the value computed from the numerical flow model may be
used. Thus, the effective conductivity should be limited by the corre-
sponding limit values:

for flow" to structure, (19)

H~ KJ.~ I for flow .1 to structure, (20)

These equations provide some analogy to Equation (12), and by an

heuristic argument one may assume that the effective conductivity is once
more a mean between the given bound, at least as long the conductivity
contrasts are low enough. A first approach to describe K as a function of
(H, I, A) is the weighted geometric mean which also provides a power
law:

flow" to structure, (21)

flow .1 to structure, (22)

These equations, indeed, approximate the data as Figure 8 illustrates. The

figure shows the data points and the associated power law correlation
lines for conductivity contrasts from 1: 1 to 20: 1. The parameter given for
EFFECTIVE TRANSPORT PROPERTIES 109

every curve is the structural anisotropy P as defined in Ondrak, Bayer,

and Kahle (1994):

(23)

It should be mentioned that the approximation fails for higher

conductivity contrasts especially at high structural anisotropies, a situation
encountered earlier during the discussion of the iterative mean for statisti-
cally isotropic media. At high anisotropies calculation errors may play an
important role which, in addition, bias the approximation. These errors
may be related to the fact that the representative volume increases rapidly
with increasing anisotropy. The parameters a and ~ depend on the degree
of anisotropy and the frequency of the materials. This dependency, again,
is well described by a power law :

a=r, (24)

where P denotes the structural anisotropy. The power x should be a

complex function of material frequencies. Again, a simple functional
relationship has not yet been determined and the situation close to the
percolation threshold needs further research.
In summary, we determined that means and power laws provide
good and fast approximations for effective conductivities of statistically
isotropic and for anisotropic media within some limits. At least, the
geometric mean can be understood as another power law so that a concise
approach seems possible in order to describe media with a wide range of
conductivity contrasts. A special and, perhaps, interesting result is that the
iterative mean provides a perfect approximation for statistically isotropic
media at the percolation threshold. This result introduces a 'fix point'
which could provide a base for further studies.

CONCLUSIONS

The paper provides power law equations which relate the effective
transport properties of a randomly generated two-component medium to
llO KAHLE AND BAYER

the frequencies of the materials and to simple statistical measurements of

anisotropy. This approach allows 'easy estimations' of the transport tensor
for geological media by approximate functions and for different degrees
of anisotropy of the medium.

de-gr .. 01 "1'UC'Utai anatropy

_ )0 '0 ~ ) I..l I
... ",:" ..... €-~~*-4-
0.7 0.75 0,8 0,85 0.9 0 95
VA

HIK

0,9

0.8

0,7

0_6
d~, . . oIlftr\lC'!.urv.1 Mishopy
- )0 10 5 ) U 1
..... ·8...... -i~.4}.*-w-
0,5
0.5 0.6 0.7 0.8 0,9
8 HI'

Figure 8. Effective conductivity (K )for anisotropic media related to arithmetic mean

(A), harmonic mean (ll) and effective conductivity of statistically isotropic medium (1)
by a power law. Points denote values calculated from flow model, lines show power law
K=(l)a
fits; (a) - - for flow II structure and (b) -H=(H)~
- for flow 1. structure.
A A K 1

The random generation mechanism of the media includes the

statistically isotropic medium. For such media with infinite conductivity
EFFECTIVE TRANSPORT PROPERTIES 111

contrasts, the classical descriptions used in percolation theory apply. In

extension to percolation theory it was possible to determine an unbiased
description for arbitrary conductivity contrasts at the percolation thresh-
old, given by the limit of an iterative mean, combining the arithmetic and
harmonic means. In addition, this iterative mean provides good
approximations for any composition of the medium as long as the con-
ductivity contrasts are low. In situations of high conductivity contrasts an
extension of the concepts of percolation theory, including the iterative
mean as a parameter, is best suited to describe the effective transport
properties if the material with higher permeability has a frequency P > Pc.
The rules of percolation theory, therefore, are not only applicable to black
and white media, but can be extended to describe a continuous space of
permeability contrasts.
The anisotropic medium was related to the isotropic and the
perfectly layered medium by a simple power law respectively the
weighted geometric mean. This approximation holds also for low
conductivity contrasts. This result may provide a first step towards
transformation rules which allow to map anisotropic media onto an
equivalent problem involving statistically isotropic media. Such rules
would allow to reduce the space of possible structures drastically.
A main result, however, is that the point of percolation threshold
Pc also is meaningful in the context of media with finite conductivity con-
trasts. The coincidence with the perfect description by the iterative mean
at the point of percolation can be considered an analogy to the geometric-
arithmetic mean algorithm which forms a base for the theory of elliptical
integrals (Beckenbach and Bellmann, 1965). Here could be an interesting
connection between different theories such as percolation, fractals,
effective medium theory, and classical calculus.

REFERENCES

Beckenbach, E.F., and Bellmann, R., 1965, Inequalities: Springer-Verlag,

New York, 198 p.
Dikow, E., and Hornung, u., 1991, A random boundary value problem
modelling spatial variability in porous media flow: Jour. Diff.
Equations, v. 92, no. 2, p. 199-225.
Hilfer, R., 1992, Local-porosity theory for flow in porous media: Phys.
Rev. B, v. 45, no. 12, p 7115-7121.
112 KAHLE AND BAYER

King, P.R, Muggeridge, A.H., and Price, W.G., 1993, Renormalization

calculations of immiscible flow: Transport in Porous Media, v. 12,
p.237-260.
Kinzelbach, W., 1986, Ground water modeling: Elsevier, Amsterdam, 333
p.
Kyte, J.R, and Berry, D.W., 1975, New pseudo functions to
control numerical dispersion: Soc. Pet. Eng. AIME Jour., v. 15, p.
269-276.
Morland, L.W., 1992, Flow of viscous fluids through a porous
deformable matrix: Surveys in Geophysics, v. 13, p. 209-268.
Ondrak, R, Bayer, U., and Kahle, 0., 1994, Characteristics and evolution
of artificial anisotropic rocks, in Kruhl, IH., ed., Fractals and
dynamic systems in geoscience: Springer-Verlag, BerlinlHeidelberg,
p. 355-367.
Peitgen, H.O., Jurgens, H., and Saupe, D., 1992, Chaos and fractals-new
frontiers of science: Springer, New YorklBerlinlHeidelberg, 220 p.
Prakouras, A.G., Vachon, RI., Crane, RA., and Khader, M.S., 1978,
Thermal conductivity of heterogeneous mixtures: Intern. Jour. Heat
Mass Transfer, v. 21, no. 8, p. 1157-1166.
Sahimi, M., 1993, Flow phenomena in rocks: from continuum models to
fractals, percolation, cellular automata, and simulated annealing:
Review of Modem Physics, v. 65, no. 4, p. 1393-1534.
Schoenberg, 1.1, 1982, Mathematical time exposures: Math. Assoc.
America, 270 p.
Schwartz, L.M., and Kimminau, St., 1987, Analysis of electrical con
duction in the grain consolidation model: Geophysics, v. 52, no.
10, p.1402-1411.
Stauffer, D., 1985, Introduction to percolation theory: Taylor & Francis,
London and Philadelphia, 89 p.
Stone, H.L., 1991, Rigorous black oil pseudo functions: Proc. 11th SPE
Symp. on Reservoir Simulation, p. 57-68.
Wong, P., Koplik, J., and Tomanic, J.P., 1984, Conductivity and
permeability of rocks: Phys. Rev., v. B30, p. 6606-6614.
THREE-DIMENSIONAL MODELING OF GEOLOGICAL
FEATURES WITH EXAMPLES FROM THE CENOWIC
WWER RHINE BASIN

Rainer Alms, Christian Klesper, and Agemar Siehl

Geologisches Institut der Universitiit Bonn, Bonn, Gennany

ABSTRACf

Concepts and software systems for interactive free form modeling of

geologically defined geometries are presented, which utilize the up-to-date
availability of high performance techniques for data handling, 2D and 3D
interpolation and computer graphics. Under control of a programmable
interface, commercial programs (ONTOS, GMP) and research
developments (GEOSTORE, GRAPE, GOCAD) have been implemented
in a workstation network using Remote Procedure Call for data exchange.
This system has been developed using examples from the Lower Rhine
Basin.

INTRODUCfION

The present structure of geological surfaces, bodies, and property

distributions is a product of interacting processes during geological
evolution. The grade of complexity of the pattern depends on the number
of structure-forming effects superimposed on one another. Complexity

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 113
114 ALMS, KLESPER, AND SIEHL

usually increases with the passing of time, but traces also may fade out
and vanish. The recent assemblage of surfaces and bodies contains all
information we have, and the geologist must recognize the dynamics of
its evolution in the actual stage of development. The assemblage defines
boundary conditions for all derived models, for example dynamic
modeling of sedimentary and structural basin development.
The fabric of rocks intersected by many surfaces of different meaning
thus is the key in revealing the nature and interaction of earlier processes,
tracing them back through time, understanding how the structure was
formed, and how it probably will develop in the future. This characteristic
geoscientific process of inverse reasoning is outlined in Figure 1 (Siehl,
1993), but it is well known that inversions do not have unique solutions.
This is the main reason why we make geometric models in geology and
test them against various aspects of reality: to determine the most likely
solution.
Geological data typically are sparse compared to the complex
structures they are derived from and are supposed to describe, that is, the
essential parts of the structure we want to model are not supported by
data as they are hidden and cannot be verified. Therefore, the first step
in modeling always is the problem of pattern recognition with few
constraints. Usually the geologist must decide between contradictory
assumptions. Thus, in addition to the usual inspection of geological maps
and sections, drawings of alternative versions in perspective views,
especially when structures are complicated, are needed. Modem
computer-aided geometric design techniques combined with high-
performance computer graphics can be powerful tools that assist the
expert in the investigation of unknown domains. Various types of graphic
representation, animation, and even stereoscopic views contribute
decisively towards the rendering and understanding of modeled virtual
geological objects in three and more dimensions.
When a geologist makes use of such modeling systems, he/she needs
an interface and interactive tools to enter the conceptual geometric model
together with the supporting data into the computer. He/she must bring
in the background knowledge in order to formulate geometric hypotheses
by which geological processes a surface might have been formed in
regions where the data support is insufficient. This can be regarded as a
type of intelligent interpolation and extrapolation, which up to now could
not be achieved by an algorithm. Rendering of the designed structure
permits the control of all geometrical consequences of the initial
THREE-DIMENSIONAL MODELING 115

Inverse Reasoning in Geological Modeling

interaction
--'"
geological --. present
of geological
evolution
~

situation
processes

inverse reasoning:

( processes ( series of scenes )... ( snapshot )

'-------~

DYNAMIC (n-D) KINEMATIC (4-0) GEOMETRIC (3-D)

SIMULATION MODELING RECONSTRUCTION

forward modeling: backward modeling: data modeling:

- math. formulation -erosion -GIS
- calibration - tectonic deformation - database
- error estimation - compaction -CAGD
- model improvement - sedimentation - visualization

Figure 1. Inverse reasoning in geological modeling.

assumptions. Each of the designed elements and then the entire assembled
model are tested subsequently in regard to geological plausibility and
goodness of fit to the primary and derived data. This modeling process
is one of interactive learning and increased insight into the spatial
interrelationships of geological surfaces and bodies. It leads to better
understanding of the history of formation which, in tum, is incorporated
in the improvement of the model to increase structural resolution.
When, finally, this process has gone through several cycles, it may
result in a consistent and up-to-date model of subsurface reality according
to the state of knowledge, with a high-predictive power for practical
purposes.
It is stored in a databank for application such as the execution of
updated cartographic editions of maps and sections. One application. of
such models is the construction of geological maps by intersecting the
assembly of geological surfaces with a terrain model and testing the
116 ALMS, KLESPER, AND SIEHL

resulting pattern against field and subsurface data (Siehl and others,
1992).
Such models consequently lead to the second step of inversion: the
kinematic backward modeling to access paleogeological states,
rearranging and smoothing out the deformed and disrupted pile of strata.
Actual three-dimensional backward modeling of sedimentary basin
evolution is now within reach, true to scale and evolution rates, and can
provide details of sedimentation, burial, compaction, and subsequent
structural development.

DESIGN OF TIlE SYSTEM

An interactive computer system for modeling geological geometries

in 3D (Fig. 2) requires the following essential functionalities:

• storage and retrieval of large 3D data sets with their topological

structure,
• generation of irregular triangular and tetrahedral meshes,
• interactive visualization and graphic edition in 3D,
• intersection algorithms for 3D objects,
• 3D interpolation algorithms,
• numerical evaluation of model properties.

Many of these features are realized in a variety of available programs.

However, there exists no single 3D-GIS which completely meets the
fundamental requirements for adequate geometric modeling of geologic
structures in 3D. Procedures of database management, interactive
visualization, methods for mesh generation, and 3D interpolation are
rapidly developing. To be able to adapt the appropriate software product
in each situation, an open system with a programmable user interface was
created which
• establishes a problem-oriented user interface with graphics support of
database management,
• ties different programs together,
• manages data exchange, and
• provides an interface to implement algorithms for new and specific
requirements.
THREE-DIMENSIONAL MODELING 117

Interactive Geological Modeling

interaction
visualization test consistency and
graphical edition program parameter control dynamic plausibility

modeling tool
input output
mesh generation
interpolation
generation of complex geo-objects
balanced backward deformation
model concept, maps, sections
knowledge, parameters, perspective views,
vector and time series,
raster data, space-time sections
distributed databases

3D/4D-database
raw data
model context
model data
model versions
time dependent stages

Figure 2. Interactive geological modeling.

The object-oriented 3D graphical programming system GRAPE

(GRAPE, 1993) was used to build the framework (Fig. 3). The system
kernel provides the possibility of integration of user defined projects with
specific functionalities. Our GEOCON project (Alms, Klesper, Siehl,
1993) allows the utilization of other software systems such as GOCAD
and GMP. In combination with powerful visualization facilities, newly
developed modules such as a 3D cursor, interactive stereo mode, and
volume discretization based on tetrahedrons increase its usability.
Interfaces to all 3D modeling software tools used are implemented to hide
data exchange behind the user interface.
As in our situation, most of the databanks are located on machines
other than those with the modeling tools. Remote procedure calls (RPC)
can be used to simplify and accelerate the online data exchange between
118 ALMS, KLESPER, AND SIEHL

local and remote programs. A RPC-connection is established by a calling

sequence from the modeling tool to the remote database. The window of
the database-management system will be displayed on the local machine
where the data are selected by the user. The selected data are then sent
to the local modeling tool for further processing. Such connections enable
direct data exchange between the running programs without the creation
and storage of data files.
For the database management of 3D objects, the ONTOS database-
management system (ONTOS DB, 1992) is used. GEOSTORE (Bode,
Breunig, and Cremers, 1994) is a tool kit that serves as an interface for
3D geological data to ONTOS and offers specific functions for geological
data. A special GIS functionality is implemented to guide the formulation
of spatial oriented retrieval queries by the visualization of the database
content, for example the position of cross sections and fault lines in map

Configuration of Hard- and Software

with Functionalities

[~_ _ _ _ _. SUN
_ UN_IX_:~ _ _ _ _J workstation network
+SGI .

GRAPE
Project GeOCON
GOCAD GMP

model dimension 40 3D 3D modeling and

geometry representation vIr vector raster visualizing
attribute handling yes yes yes
CAGD yes yes no

User Interface

RPC interactive data exchange

GEOSTORE

database management
ONTOS

Figure 3. Configuration of hardware and software with functionalities.

THREE-DIMENSIONAL MODELING 119

view. It can provide cartographic orientation of the stored data and

facilitates the selection of special objects. The user may obtain more
information by selecting visualized objects on the screen and browsing
deeper into the data structure. A specific geologic function is the
possibility of defining any cross section through the three-dimensional
data structure stored in the database and their visualization. Replacing
objects in the database is controlled by a set of user-defined integrity
constraints to maintain the consistency of the database.

PROCEDURES FOR SURFACE AND VOLUME MODELING

Models based exclusively on regular grids and on conventional CAD

methods are not suitable for modeling the geometry of complex
geological objects (Mallet, 1993). For mesh generation and interactive
surface design, the program GOCAD (GOCAD report, 1993) is integrated
in a manner as described. GOCAD is a program especially devoted to
geological requirements using the Discrete Smooth Interpolation (DSI)
method to fit surfaces interactively to points which represent complex
structures. By this method surfaces are modeled as sets of triangular
facets, minimizing a global roughness criterion. Local structures are
modeled introducing constraints, for example fixing nodes, directing
intersection lines, vectorial linking of surface nodes to other objects. This
enables the user to include his/her knowledge and ideas immediately in
the interpolation process and to visualize and analyze the results directly
in 3D. Space-filling tetrahedrons correspond to the flexibility of irregular
triangulation and provide the discretization for subsequent application of
finite element methods for backward modeling of deformation.
The volume modeling program GMP (GEOCALC/GEODRAW, 1992)
is used to calculate the distribution of attributes through space. The
volume is discretized with a regular 3D grid using 2D grids to represent
stratigraphical boundaries. Attribute values are interpolated with a
"minimum tension" algorithm at the nodes of the grid. Defining the cutoff
level for attribute ranges, the distribution can be displayed as isosurfaces.
More complex structures have to be modeled separately by different
volumes which must be merged after calculation. Three-dimensional grid
spacing must be equal for all different volumes building one model.
Description of shapes and internal structures of the volumes cannot be
120 ALMS, KLESPER, AND SIEHL

done by the 3D grid discretization itself. Additional surfaces for bounding

and clipping are necessary to define parts of the volume.

GEOWGICAL SETTING

We have developed our system in conjunction with the geological

modeling of actual examples from the Lower Rhine Basin (LRB). The
LRB is situated in NW Germany and the Netherlands and consists of a
large-scale Cenozoic graben system filled with Oligocene to Quaternary
sediments. In this location the LRB links the Upper Rhine Graben
southwest of the Rhenish Massif and the North Sea Rift (Zagwijn, 1989).
It can be regarded as the northwestern branch of an intraplate rift system
with a triple junction located in the Mainz Basin (Ahorner, 1975; lilies,
1977) (Fig. 4).
The basement of the LRB is part of the Central European Variscan
foldbelt and consists of the same Devonian and Carboniferous rocks as
the surrounding Rhenish Massif. Permian and Mesozoic sediments of low
thickness are known from drillholes and some outcrops at its
southwestern boundary in the Eifel Mountains. Graben subsidence started
in Middle Oligocene and continues today. The deepest parts of the LRB
subsided more than 1200 m below present sealevel, while the adjacent
Rhenish Massif was uplifted to several hundreds of meters (Fig. 5).
Extension tectonics with a slight dextral wrench component created a
characteristic block fault pattern striking in general NW-SE and dipping
55 0 to 65 0 to the west, whereas the strata are dipping eastwards.
Considerable deposits of lignite occur in the deltaic to fluvio-
lacustrine series of Miocene age, overlain by the gravel and sand of
braided river systems. Extensive exploration by drilling as well as by the
mining of brown coal have provided an excellent three-dimensional
database for the reconstruction of basin development.

EXAMPLE AND DATABASE

The database are lithological cross sections (horizontal scale 1: 10 000,

vertical scale 1:2 000) from the Rheinbraun AG mining company. The
general orientation of the sections is SW-NE with a parallel distance
between 1.5 to 4.5 km (Fig. 6).
THREE-DIMENSIONAL MODELING 121

We have used these interpreted cross sections and not the original
well data because the primary data are extremely heterogeneous in
resolution and quality. Rigorous interpretation by the Rheinbraun mine
geologists, who also provided background knowledge not available in
borehole logs, has been essential in providing a consistent database for
3D modeling.
Before digitizing, the sections must be inspected carefully and
reinterpreted. This involves testing selected boundaries between strata in
respect to their consistency as well as generalizations where horizons
pinch out or where their assignment is not clear. A complete encoded
description of the topological interrelations is attached to each data point
before the set is stored in ONTOS via GEOSTORE.
The plausibility of the digitized geometry can be tested by visualizing
the vectors as polygons and comparing them with the original cross
sections. Additionally an automatic check of the consistency in regard to
its explicit geometric accuracy relative to the internally defined data
model is done by GEOSTORE using the topological information attached.

SURFACE REPRFSENTATION TECHNIQUES

To represent surfaces, irregular triangulation meshes are used. As

mentioned previously they provide a better fit to the irregular boundaries
of geological objects than rectangular grids or voxel discretization.
Depending on the data distribution and the nature of the surfaces, two
alternatives in designing the meshes were considered:

• a direct triangulation, in which each data point represents exactly a

single node of the mesh, or
• an indirect meshing with the construction of the net on the basis of a
convex hull around the data points; the triangulation then is fitted to
the points.

With the given irregular data distribution in clusters along the section
lines, direct meshing will evoke extremely adverse mesh geometry which
will produce artifacts during further processing. For this reason indirect
meshing was applied which yielded a more regular net geometry. The
initial triangulation mesh is fitted closely to the points using the 3D
interpolation algorithm DSI of GOCAD. Special constraints while
122 ALMS, KLESPER, AND SIEHL

modifying the triangulations have been applied to match the mesh exactly
to the designer's ideas, for example:

• fixing of areas for local modification,

• minimizing distances between surface and data point, or
• directing intersection lines along faults.

DRtir15

Par i s
Bas i n

4' 'S' i8 '

Figure 4. Location of Lower Rhine Basin (LRB) within generalized geological

framework. Marked area is enlarged in Figure 6.
THREE-DIMENSIONAL MODELING 123

Rur Block Erfl Blocl, Cologne Block

..
1 QUllternor)" ~ rllier lerrKe,
l ' lIIIIJ R@,,\~I'" 6;. Rotton 1ms:
·tOOO
C P:loC"~ne ~ HIIIUplaH!:S Fo:-r.uthoo
:II ~ ~~
~ Miocene tmll~dtn .·ormahcn
~ P":..IiiiI \lIlt Forrnatlon
O:IROC'C!:"U!' ZJ C'olo(l:f:l' rorm ltlo:1 k older
om 0 10
Corbo",!erous
De\'onum
EZdJ
~ ba.!'I'ement

Figure 5. Generalized section trending SW-NE through center of Lower Rhine Basin,
for location of A and B see Figure 6.

There generally are two approaches In modeling faults as

discontinuities of bedding planes:

• creation of one coherent triangulation net which is deformed to a

sharp edge along the intersection lines of faults and strata, or
• modeling fault and stratigraphic surfaces separately as independent
objects which are related to each other by special operations (for
example intersections) or constraints (for example movement vectors).

For the mere visualization of surfaces, the first method is adequate.

However, for detailed structural analysis or for the intended 3D balanced
backward modeling, the second method must be used. Here the
geometrical elements can be assigned to specific geological events, and
it is possible to experiment with alternative hypotheses of structural
development, the most common way in which geological models are
formed.
The construction of a set of faults and stratigraphic surfaces shown in
Figure 7 will be demonstrated by some examples.
124 ALMS, KLESPER, AND SIEHL

~~ open cas t lllmcs ~ basement

Cl'oss - src lions \ faulLs al basp of Tfrtiary

Figure 6. Area under investigation is located in southeastern part of Lower Rhine

Basin, Thirty-eight cross sections (parallel lines) represent database of model. Section
shown in Figure 5 is marked with A and B.
THREE-DIMENSIONAL MODELING 125

------------------------------- -

Flow Chart of Modeling Geologically Defined Geometries

Interpreted geological situation
,........ basic data: points, attributes, topology, semantics

r- m faults n bedding surfaces

~------~----------------------------r_------~~
~ ~
selection of fault data ~ I selection of bedding surface data
i
generation of polygons and meshes I Igeneration of polygons and meshes
+
~ ~1_nte_rs_e_Ctl_On_b~y,fa-ult-s------~Jr
lit meshes to data J ~
1 l fit meshes to data

Intersection by other faults I +

l mesh Improvement
1
lID! fault assemblage
'-------,~--...!:..-.....I
~s_t_ra_ti_gr_a_Ph_ic__y__se-q-u-en-c-e--~~
~
U
-1'----_ _ consistant
geometrically _gelogical
_model _ _L
1

Figure 7. Construction of a complex fault and strata assemblage.

FAULT SURFACFS

As all faults are younger than the bedding planes which they intersect
and the general information of the fault position is given by the database,
the construction of the surface assemblage starts with fault modeling. To
construct a fault surface, the respective data points will be selected with
GEOSTORE from the database, followed by automatic triangulation with
GOeAD based on a convex hull around the selected points (Fig. 8).
Design of this convex hull depends on the data distribution and the
3D orientation of the surface to be constructed. A virtual plane can be
defined by the user through a normal vector or by determining the surface
with a minimum sum square distance to all points. The bounding polygon
then is seen as the enclosure of the proj ection of the data points on the
plane.
After construction of each fault, the intersection of faults with each
other must be considered. With increasing complexity of the fault pattern,
interactive control by the designer becomes more and more essential.
126 ALMS, KLESPER, AND SIEHL

After numerical checks of the approximation quality and a visual

comparison with known fault patterns, for example from isovalue maps,
the fault surfaces can be stored as triangulation nets in ONTOS.

SlRATIGRAPIDC SURFACFS

The construction of stratigraphic surfaces is based on a convex hull

around the data points. In this situation the initial plane defining the
convex hull normally has subhorizontal orientation. The mesh is
intersected by the previously constructed faults and is refined
subsequently in several stages. Improvement of the approximation quality
of the intersected stratum plane to the data points will be accomplished
with DSI under consideration of several constraints. In particular the
intersection lines of the layer must be fitted to the corresponding fault
surfaces (Fig. 9).
These processes normally can be done automatically, but with
increasing complexity of the pattern, it again is necessary to check and
to correct the mesh interactively.
If the stratigraphic surfaces are more or less flat and nearly parallel,
and the faults are subvertical and not too complex in their intersection
pattern, a faster and simpler procedure for designing the stratigraphic
layers can be applied. In this situation, a reference surface will be
generated as a plane covering the maximum extension of the vertical
projection of all other surfaces. This surface is intersected with the faults
and fitted to the data points of one horizon in the center of the sequence.
It then will be duplicated, vertically shifted to the levels of all other
horizons, and fitted to their respective data points. The stratigraphic
surface is finished interactively by adding new faults or clipping the
surfaces according to the extension of the data supporting points.

ANALYZING mE MODEL

The practical investigation and rigorous testing of modeled 3D-surface

structures normally is done, for example, by 2D visualiziation of isovalue
maps with fault patterns (Fig. 10).
Two-dimensional isovalue maps provide direct comparison with
conventional maps and sections. Improvement and updating of the
THREE-DIMENSIONAL MODELING 127

Figure 8. Digitized cross sections in southern part of Erft Block with triangulated fault
surfaces.

computer model is continued until its consistency is satisfactory relative

to the supporting data and the given assigment.
An important component of analyzing the model and model evaluation
is interactive 3D visualization. The combination of different
functionalities, such as translation, rotation, zooming, transparency, colors,
128 ALMS, KLESPER, AND SIElll..

a)

t'0/
"0

c)

/ c?--4\
~/
, ~
.

o data point ~ stratigraphic urface

o free node fault line

Figure 9. Successive steps in fitting stratigraphic surface dissected by three faults to

data points in oblique view and in schematic sections on left.
THREE-DIMENSIONAL MODELING 129

and lighting, or animated stereo views, allows the inspection and

understanding of complex 3D objects and derived models. Some examples
of 3D visualizations are shown in Figures 11 and 12.

Erft Block
base of stratum 7

) 1'50.00

....
·1150. 00

( # ~SO.OO m

I X

Figure 10. Depth line contour map of base of stratum 7.

DISCUSSION

Computer-aided geometric modeling in 3D is not yet usually

employed in geology because of the following reasons:

• relevant data are inaccessible to electronic processing.

• data exchange between 3D modeling systems is not properly
supported.
• much interactive input lS needed to model complex geologic
structures.
130 ALMS, KLESPER, AND SIEHL

Figure 11. Oblique view from NW into Lower Rhine Basin. Lower layer represents
base of Tertiary according to Hager (1988) with structural units "Krefeld Block," "Ven10
Block," and "Rur Block." Upper layer represents digital terrain model with horizontal
resolution of 50 x 50 m. Deepest parts of Rhine River Valley are about 30 m
above sealevel, highest of Eifel mountains about 200 m.

Figure 12. View from SW to model space. Gray base of Tertiary with triangular mesh.
In area of "Erft Block" some of strata and faults with cross sections are modeled. Upper
surface indicates Pliocene "Hauptkies-Forrnation" (Fig. 5).
THREE-DIMENSIONAL MODELING 131

• integration of expert knowledge still is done by direct manual

interaction. This may introduce an unwelcome subjective component.
• geometric models are often too simple in relation to complex reality.
• methods of evaluation and error estimation of geometrical models are
still rudimentary.

To model complex geological structures in 3D effectively, it is

desirable to represent the geometrical and topological relationships as
exactly and completely as possible in the data structure managed by the
database system. Only then can local modeling and updating of parts of
the structure be performed maintaining both the consistency of the entire
model and the integrity of the database. This requires the design of an
open, modular system with modeling tools interoperating closely with
databases. Different functionalities implemented in distributed modeling
systems can be tied together by an interface which gives online graphical
control of the modeling process. Interactive data exchange for raw data
and model data must also be managed if the systems and databases are
on different machines.
Geometric modeling is normally done for a particular region,
depending on the problem to be solved and the available data. In the
course of further treatment, combination and merging of such models
might be of great importance, but is seldom possible. It raises the
problem of appropriate up and downscaling involving the subsequent
refinement of structures and integration of additional elements, on one
hand, as well as the generalization and omission of details on the other.
A critical point in 3D modeling is the recognition of objectives and
the purpose of a particular model. Data models, as defined in Figure 1,
demand high spatial resolution and close-to-nature complexity, because
such models may be made to solve practical problems in predicting
subsurface structures for exploration, mining, or geotechnical purposes.
When bringing in kinematic or even dynamic components into the
modeling, spatial resolution and complexity normally is reduced. In these
situations we may be merely in the state to model simplified structures
in order to understand basic principles of interacting processes and
geological evolution.
The excellent quality of available graphical rendering sometimes may
conceal the fact that in the individual case we know very little about
accuracy and precision of our models.
132 ALMS, KLESPER, AND SIEHL

ACKNOWLEDGMENTS

We would like to express our appreciation to the Rheinbraun AG

mining company for providing the database of the Erft Block. Substantial
financial support was received by the German Science Foundation (DFG)
within the framework of the special research project "SFB 350:
Wechselwirkungen kontinentaler Stoffsysteme und ihre Modellierung".

REFERENCES

Ahorner, L., 1975, Present-day stress field and seismotectonic block

movements along major fault zones In Central Europe:
Tectonophysics, v. 29, p. 233-249.
Alms, R., Klesper, C., and Siehl, A, 1994, Geometrische Modellierung
und Datenbank-entwicklung fur dreidimensionale Objekte:
Nachrichten aus dem Karten- und Vermessungswesen, Reihe I, v. Ill,
P 9-16.
Bode, T., Breunig, M., and Cremers, AB. 1994, First experiments with
GEOSTORE, an information system for geologically defined
geometries: submitted to Proc. Workshop on Advanced Research in
Geographic Information Systems (Monte Verita, Ascona).
GEOCALC/GEODRAW, 1992, Geologic modeling program, DYNAMIC
GRAPHICS, INC., reI 5.0.
GOCAD Report, 1993, Association Scientifique pour la Geologie et ses
Applications, verso 9.3, 1, p. 1-231; 2, p. 233-487.
GRAPE, 1993, GRAPE - A Graphical programming environment for
mathematical Problems: Universitat Bonn, SFB 256; Universitat
Freiburg, Inst. f. Angew. Mathematik, verso 4.0.
Hager, H., 1988, Die Unterflache der tertiarzeitlichen Schichten: Geologie
am Niederrhein, GLA Krefeld, p. 37.
Illies, IH., 1977, Ancient and recent rifting in the Rhinegraben: Geol. en
Mijnbouw, v. 56, p. 329-350.
Mallet, IL., 1992, GOCAD - a computer aided design program for
geological applications: three-dimensional modeling with Geoscientific
Information Systems: NATO ASI, C 354, p. 123-142.
ONTOS DB, 1992, User Guide (Release 2.2) ONTOS INC. (Feb. 1992).
Siehl, A, 1990: Geological maps by interactive spatial modeling: Three-
dimensional computer graphics in modeling geologic structures and
THREE-DIMENSIONAL MODELING 133

simulating geologic processes, Freiburger geowissenschaftliche

Beitrage, v. 2, p. 95-96.
Siehl, A, 1993, Interaktive geometrische Modellierung geologischer
Flachen und Korper: Die Geowissenschaften, v. 11, p. 342-346.
Siehl, A., Riiber, 0., Valdivia-Manchego, M., and Klaff, J., 1992,
Geological maps derived from interactive spatial modeling, in Digital
map series in geosciences to geo-information systems: Geol. Jb.,
A122, p. 273-289.
Zagwijn, W.H., 1989, The Netherlands during the Tertiary and
Quaternary: Proc. Symp. on Coastal lowlands~ geology and
geotechnology (Univ. Utrecht): Kluwer Acad. Publ., Dordrecht, p.
107-120.
 VOLUMETRICS AND RENDERING OF GEOLOGIC
BODIES BY THREE-DIMENSIONAL GEOMETRIC
RECONSTRUCTION FROM CROSS SECTIONS OR
CONTOUR LINES

Helmut Schaeben, Stephan Auerbach, and Esther U. Schlitz

Universitiit Trier, Trier, Gennany

ABSTRACT

In applied earth sciences a three-dimensional object may be given by

a set of parallel cross sections. More specifically, the body is given by a
set of contour lines, which may represent the intersection of its boundary
surface(s) with the plane of the cross section thus distinguishing the
body's interior and exterior, or a given value of some property of the body.
Graphically displaying the body's geometry and topology, that is its
internal structure in terms of contour lines discriminating the body's
interior and exterior, or quantifying some of its features, for example
arbitrary intersections or volumes corresponding to given contours,
essentially requires to reconstruct the corresponding boundary surfaces of
the three-dimensional body.
A computer-aided method of geometric reconstruction of closed
three-dimensional bodies from their cross sections or contour lines,
respectively, is presented, the essentials of which are implicit
representation of surfaces and bodies, and set-valued interpolation. This
approach suggests to approximate the interior of a three-dimensional body
as an entity and thus allows the recovery of bodies of arbitrary shape and
topology. In particular, a cross section may comprise more than one
closed contour line.

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 135
136 SCHAEBEN,AUERBACH,ANDSCHUTZ

Set-valued interpolation is performed numerically by consecutively

solving a sequence of univariate interpolation problems. These problems
are solved by applying quadratic splines of class C1 and employing their
favorable properties of local support, smoothness, accuracy, and
preservation of monotonicity.
The method is automatic and does not require any interaction by the
user. Its capability to reconstruct complex geologic bodies from their
contour lines for rendering and volumetrics is exemplified using synthetic
data.

INTRODUCTION

This contribution is concerned with the problem of reconstructing a

three-dimensional body (cf. Auerbach and Schaeben, 1990a, 1990b) in
terms of its outer and inner boundary surfaces given by a collection of
planar contours representing cross sections through the body. This
problem has important applications beyond geosciences. The suggested
method allows immediate applications to computer-aided graphical
display or volumetrics once the body is reconstructed. It applies to bodies
of arbitrary shape and topology and thus to complex geologic bodies.
With respect to its range of applications it compares favorably with
methods based on spatial triangulation (cf. Meyers, Skinner, and Sloan,
1992).
Rendering folded and faulted geological surfaces from series of cross
sections is described in Klein, Pflug, and Ramshom (1989). They apply
interactive, that is user-controlled trial and error, spatial triangulation
between digitized geological cross sections and corresponding
interpolation algorithms to accomplish visual display of complex
geological surfaces.
The review on computing volumes (Walsh and Brown, 1992) is
summarized as follows:
* step method (contour areas x change in thickness): severely
underestimates
* trapezoid rule (first-order Newton-Cotes numerical integration):
slightly overestimates
* pyramid rule: slightly underestimates
* combination of trapezoid and pyramid rule
VOLUMETRICS AND RENDERING 137

* Simpson's rule (second-order Newton-Cotes numerical integration,

three terms): requires evenly spaced contours and an even number
of contours
* 3/8 rule (second-order Newton-Cotes numerical integration, four
terms): requires evenly spaced contours and the number of contours
be a multiple of 3.

For details of numerical integration the reader is referred to (Press and

others, 1990).

SET-VALUED INTERPOLATION

Let B C 1R3 be a closed and bounded body in the three-dimensional

Euclidian space 1R3. Its intersections S;, i =0, ... , n + 1, with n + 2 parallel
planes, which are labeled by levels Zi according to a suitably selected
coordinate system are defined as

B (Zj) = Sj = {(x, y, z) E Biz = zd c B c IR 3 , i = 0, ... , n + 1 . (1)

These cross sections are closed and bounded; however, it is not required
that they are simply topologically connected. It is assumed that So = Sn+1 =
o and Si '# 0, i = 1, ... , n. The boundaries C = reSi) of Si, here termed
contour lines, are assumed to be given as polygons, for example polygons
which may be thought of as representing digitally recorded contouring
lines of a map.
The problem is to reconstruct the body B from the contours reSi), that
is to determine a body iJ c 1R3 , which interpolates the cross sections for
each level Zi, iJ (z;)= S;, i = 0, ... , n + l. Because these interpolation
conditions involve sets rather than numbers, the suggested method may be
considered as "set-valued interpolation".
The following implicit representation of the body B is used. Let d(a,a)
denote the usual Euclidian distance in the plane IR2

d[(x,y,z),(x',y',z)] = V(x - x')2 + (y - y')2 (2)

138 SCHAEBEN, AUERBACH, AND SCHUTZ
Then, the distance of a point (X,y,Zi) to any nonempty set Si corresponding
to level Zi is defined by

d[(X,y,Zi),Si] = (x',y',Zi)ESi
inf d[(x,y,zd,(x',y',Zi)] (3)

For a given point (x,y) E IR2 and for every cross section level Zi the
distance d i of the point (X,y,Zi) E S? to the contour C = nSi) is given by

{
d((x, y, Zi), Sf) if (x, y, Zi) E Si
di(x, y) i=l, ... ,n (4)
-d((x,y,zd,Si) if (x,y,zd E Sf

={
d((x, y, Zi), r Si) if (x, y, Zi) E Si
i =1, ... ,n (5)
-d((X,y,Zi),rSi) if (X,y,Zi) E Sf

where SF is the complement and nSa the boundary of Si' di(x,y) is well
defined for i = 1, ... , n and distinguishes the interior and exterior of Si just
by its sign. Because So = Sn+1 = ~, do(x,y) and dn+l(x,y) are not defined
initially, they have to be set equal to a constant appropriatly chosen
according to some heuristics, or determined according to some more
sophisticated methods.
The sign of di(x,y) indicates whether (X,y,Zi) is a point inside or
outside the body B at intersection Si, d i = 0 if (X,y,Zi) E r (SJ
The point of view of the three-dimensional reconstruction problem is
changed now from the "horizontal" intersection planes to "vertical" lines
in the following manner (see Fig. 1). For a given point (x,y) E IR2 the
sequence {Zi, di(x,y)}i = I, ... , n may be thought of as data sampled along lines
(X,y,Zi) E Si through the body B orthogonal to its planes of intersection.
Every sequence of data gives rise to a univariate interpolation problem.
The univariate function g(o;x,y) is a solution of an individual interpolation
problem if

(6)
VOLUMETRICS AND RENDERING 139

% 7

% 6

% 5

d4
%4

% 3

% 2

% 1

(X,y)

Figure I. While two-dimensional view induced by given "horizontal" intersection planes

at levels Zj is employed to determine d;(x,y), (x,y) E IR2, one-dimensional view along
"vertical" lines (xo,Yo,z), Z E IRI emphasizes to interpret dj(x,y), i = I,,,., n, as given
values read off univariate function g(z;xo,yo) : g(Zj;Xo,yo) = dj(xo,yo) with parameters Xo,Yo.

Thus, there is an interpolant g(o;x,y) for any point (x,y) E IR2. These
interpolants provide an implicit representation of the reconstructed body
A

B by

(x, y, z) E iJ if and only jf g(Zj x, y) ;::: 0

(x, y, z) rt iJ if and only if g(Zj x, y) < 0 (7)

140 SCHAEBEN,AUERBACH,ANDSCHUTZ
which can be employed easily for rendering. It is emphasized that the
~

reconstructed body B depends only on the sign of the univariate

interpolants g(o,;x,y). The boundary [' (B) is represented by

r(.B) = {(x, y, z) E IR3 I g(z; x, y) = O} (8)

Thus, the initial three-dimensional problem has been reduced to a

sequence of rather simple univariate problems.
~

By definition it is obvious that the reconstructed body B interpolates

~

the cross sections Si, that is B (zD = Si, i = 1, ... , n (cf. Gmelig Meyling,
1988).
Moreover, the following theorem by Levin (1986) gives a
convergence order for the distance between reconstructed and original
body in terms of the cross-section distance h, dependent on the
~

smoothness of the body's contours. It thus guarantees that B is a

mathematically reasonable approximation of B. Let B be a closed body in
the three-dimensional Euclidian space whose boundary reB) is composed
of a finite number of surfaces, which are Ck -continuous and mutually
disconnected. Let Zi = Zo + ih, 0 ~ i ~ n + 1, be the level of the cross
sections and 8 any positive constant. For each h > 0 consider any point
(i, l,i) with minO<i<n I i - Zi I ~ 8, which is a point of disagreement
between B and the reconstructed B

(9)

Then

(10)

For the theorem and its proof see Levin (1986).

VOLUMETRICS AND RENDERING 141

VOLUMETRICS

An immediate application of set-valued interpolation is the estimation

of the three-dimensional volume of B by the volume of the reconstructed
body B, which can be calculated according to

voI3(B) f'V voI3(.8) = 11k dzdxdy = ll1 1f3(x,y,z)dzdxd y

= 111 1{(x,y,z)e R3 Ig(z;x,Y)2:: o}dzdxdy

= 11 voh( {z I g(z; x, y) ~ O} )dxdy

= 11 l(x, y)dxdy (11)

Here 1h( xJl,z ) denotes the indicator with respect to the set Band l(x,y)
denotes the total length of segments along the vertical line {(x,y,z) I Zo ~ Z
~ Zn+l} for which g(z;x,y) ~ O.
This estimate of the volume of the body B is supposed to be superior
to the estimate

vob(.8) = 1 voh( {(x, y) I g(z; x, y) ~ O} )dz (12)

provided by the step method (cf. Walsh and Brown, 1992).

THE INTERPOLANT g: UNIVARIATE HERMITE

INTERPOLATION AND QUADRATIC C l SPLINES IN BEZIER
FORM

As should be obvious, the reconstructed body B essentially depends

on the univariate interpol ants g(o;x,y). More specifically, it sensitively
depends on the roots of the interpol ants g(o;x,y), for which they change
their sign. Therefore, the interpol ants should not only satisfy the
requirements in terms of local support, smoothness, and accuracy, but
they also should be variation-diminishing and shape-preserving to yield
good results.
142 SCHAEBEN,AUERBACH, AND SCHUTZ
Local support of the interpolant g implies that it only is affected
locally by changes in the data. This feature provides serious numerical
advantages when a cross section should be omitted or additionally
included. The required smoothness of g should correspond to the assumed
smoothness of the body to be reconstructed. For most practical
applications it should be sufficient that g is continuously differentiable,
that is gEe' ([zo, zn+l])' According to Levin's theorem (1986) the scheme
of set-valued interpolation is of second-order accuracy if the interpolant g
is of second-order accuracy, that is if the interpolant exactly reproduces
linear polynomials (including constants), then the set-valued interpolation
exactly reproduces straight cylinders and cones with their axis of
symmetry in (x,y) = (0,0). The requirements of variation-diminishing and
shape-preserving aim at an interpolant g(z; 0,0), which does not have more
roots than the polygon is defined by {Zi, die 0,0) }i =', ... , n'
Furthermore, the interpolant should be monotone nondecreasing on
intervals [Zi,Zi+d with di ::;; di+1 and monotone nonincreasing on [Zi,Zi+d
with d i ~ d i+,; this feature is referred to as comonotonicity with the data by
Gmelig Meyling (1988). These requirements usually are met by additional
constraints concerning the values of the derivative of the interpolant g(z;
0,0) at the data points (Zi, di(o,o», i = 1, ... , n. An intuitively appealing and
actually efficient choice for the slope estimates reflecting the shape of the
original data is given by

d ~( x,y ) -_ di+1(X,y) - di-l(X,y) , i=

I 1, ... ,n (13)
Zi+l - Zi-l

which in tum results in a Hermite interpolation problem

g(Zi; x, y) = di(X, y), i = 1, ... , n
g'(Zi; X, y) = d;(x, y), i = 1, ... , n (14)

Proper candidates for the interpolant g are splines, which may be

thought of as piecewise polynomials and satisfy all requirements
discussed previously. What springs to mind are quadratic splines because
they provide a C' interpolant g and easy control of the sign of the first
derivative, which is a piecewise linear function.
VOLUMETRICS AND RENDERING 143

An early account of shape-preserving quadratic splines was given by

Schumaker (1983). He suggested two-point Hermite interpolation on the
interval [Zi,Zi+l]. However, a quadratic spline on [Zi,Zi+tl, which has only
three parameters cannot generally solve the Hermite interpolation problem
which requires to adjust four parameters. But a unique C l quadratic spline
exists on [Zi,Zi+tl with a simple knot at Si, Zi < Si < Zi+l, which actually
solves the Hermite interpolation problem (Gmelig Meyling, 1988).
Usually, Si =(Zi + zi+d/2 applies.
However, the splines provided by this constructive solution are not
generally comonotone. Comonotonicity is satisfied only if the estimates of
the first derivatived;(o,o)have the proper sign and are sufficiently small in
absolute value. Thus, the initial estimates d; provided by (13) usually
need to be repeatedly improved (cf. Gmelig Meyling, 1988).
Here, a different construction of a quadratic spline satisfying all
constraints, that is interpolating the given data and the estimated slopes,
and being comonotone is suggested.
A quadratic spline may be represented by its Bezier control points
(Farin, 1988). Thus, the basic idea is to construct the three Bezier control
points for each interval [Zi,Zi+tl, i = 0, ... , n. In the following, two cases
have to be distinguished (see Fig. 2).

d
..... -.
".
.... :

,(
... ~
:' ~

, .
.·:t i
Zo jz, ,/ z, Zo z, ~z. z
-_ .... ...
, ..iI
~
'

convex case non convex case

Figure 2. Univariate interpolation problem to determine furction g(z;xo,yo) interpolating

given data (z;, d;(xo,yo». i =O, ....n + 1. estimated slopes dj(xo.yo), i =0, ... , n + 1. and
being comonotone; dashed line depicts polygon defined by {z;. d;( o.o)}; =I. .... n'
144 SCHAEBEN, AUERBACH, AND SCHUTZ
Case (i) applies if the polygon {Zk, dio,O)}k=i_I""'i+21S convex, case
(ii) applies otherwise.
Under the assumptions of case (i) the first and last Bezier control
points b o, b 2 are given by the two data points b o = (Zi, diCo,o)) and b 2 =
(Zi+l, di+l(o,o)) themselves, the intermediate Bezier control point b l is
given as the point of intersection of the lines through (Zi, diCo,o)) with
sloped;(o,o)and (Zi+l, di+1(O,O)) with sloped;+l (0,0) respectively (see Fig. 3).

Figure 3. Geometrically constructive determination of intermediate Bezier control point

in case (i) of convex polygon {Zk> dk(o,o)} k= ;·1, .. ;+1,

If ,; denotes the abscissa of this intermediate Bezier control point b l , then

Zi < ,; < Zi+l holds.
Considering case (ii) this elementary geometric construction does not
provide a useful intermediate Bezier control point because either the two
lines do not intersect at all, or the point of intersection does not
correspond to a spline function satisfying all required constraints.
=
Therefore, in a first step of case (ii) an auxiliary point b2 b~ is defined
by its abscissa ';i = (Zi + zi+d12 and its ordinate
VOLUMETRICS AND RENDERING 145

= (di(o.o)+di+ 1(0.0)) /2
A

d (0.0) provided by linear interpolation between

(Zi, di(o.o)) and (Zi+l, di+1(O,O)). Further, the slope at this point has to be
defined. Its value will control essentially the shape of the spline function.
Any value larger (smaller) than the slope of the linear interpolant, that is
the straight line through (zi,di(o,o)) and (Zi+l, di+1(0,0)), if it is positive
(ne~ative), is a reasonable choice. Then, the intervals [Zi, (;] and [(i, zi+d
both satisfy the assumptions of case (i). Therefore, in a second step, the
procedure of case (i) is applied to each interval (see Fig. 4).

'\"'"
'\ (zH-2,d ...2)
~=(Zi+Z;'1)/2 \"'"

Figure 4. Geometrically constructive determination of intermediate Bezier control point

in case (ii) of nonconvex polygon {Zk> dk(o,O)}k = ;-1..;+2 by applying case (i) to convex
polygons {(z;_1> d,_I(O,O», (z;, di(o,o», (Si' d (0,0», (Zi+lo di+l(o,o»)and {(z;,di(o,o»,
(S;,d (0,0», (Z;+10 d;+I(o,o», (Z,+20 d;do,o)}, respectively.

The quadratic spline thus defined on [Zo, Zn+I1 is C 1_ continuous, co-

monotone, and calculating its roots is straightforward.
As the boundary surface f(B) is implicitly given by g(z;x,y) = 0, the
roots of the interpolant g(z;x,y) are required to render the body B; more
specifically, calculating the roots of the quadratic g(z;x,y) on intervals
[Zi, zi+d with g(Zi;X,y) x g(Zi+l ;x,y) < 0 is required.
146 SCHAEBEN,AUERBACH, AND SCHUTZ
NUMERICAL REALIZATION AND COMPUTATIONAL ASPECT

Gridding

Obviously, in any practical application the interpolant g(z;x,y) can be

evaluated only for some discrete points (XhY/) E IR2. Therefore, a
rectangle fxo, x K ] x [Yo, yd = R c IR2 is defined such that Si c R, i =
1, ... ,n, and then gridpoints (XhY/) E R are defined according to equidistant
partitions of the intervals [x o, x K ] and [Yo, yd, respectively, by

Xk = Xo + kh l , k = 0, ... ,[( (15)

YI = Yo + [hz, [= 0, ... , L (16)

The assumption of equidistant partitions has been made for the sake of
notational simplicity only. All numerical operations are confined to grid
points (XhY/,Zi) E IR3.
A contour in the Zi plane is given in "digitized form" as an ordered
sequence of a finite number of points to constitute a closed polygon,
which in turn defines the cross section Si. For each point (XhY/,Zi) it is
checked whether (XhY/,Zi) E Si or (Xk,y/,Zi) E SF by calculating its
-minimum distance from all contours of the cross section at level Zi
according to Equations (4).
Then, the complete data di(Xk,y/), d;(XhY/), i = 1, ... , n, of each Hermite
interpolation problem to be solved are available.

Roots of the interpolant g(Z;Xk,Y/)

Because only the roots of the interpolant g(Z;XhY/) are of major

interest it is checked whether di(XhY[) x di+1(XhY/) < 0. In the affirmative
situation there is a root in [Zi, zi+d. Its actual numerical determination is
then straightforward.

The only problem left is to define do(XhY[) and dn+1(Xk,y/)

appropriately. To clarify the role of do, dn+1 let (XhY[) E IR2 be fixed. If
d\(XhY../) > 0, then do(XhY/) essentially determines the smallest of the roots
VOLUMETRICS AND RENDERING 147

z* of the corresponding interpolant g(z;Xk.Y/); analogously, if dn(xk.ya > 0,

then dn+l(xk.Y/) essentially determines the largest of the roots z* of the

°
corresponding interpolant g(Z;Xk,y/). Thus, it is reasonable to define
do(Xk,y/), dn+I(Xk,y/) such that do(xk.Y/) < 0, dn+I(Xbya < as So = Sn+1 =~. If

°
d 1(Xk,y/) < 0, the actual value of do(XbY/) does not matter at all if do(xk.Y/) <
0; analogously for dn(Xk,y/) < and dn+l(xk.ya < 0.
The situation, when the numerical values of do(xk.Y/) and dn+l(xk.Y[)
actually matter, poses a problem of extrapolation. Levin (1986) suggested
a global value independent of (Xk.Y/)

(17)

where deS) denotes the diameter of the set S = U[~1 Sj ,whereas Omwa
(1987) and Gmelig Meyling (1988) favor local linear extrapolation

dO(Xklyt) ~ min(2d1 (xbyt) - d2 (XklYI), -E) (18)

dn+dxb yd = min(2dn (xJ:, yt) - dn - 1 (Xk, yd, -E) (19 )

where f denotes a small positive number.

Here it is emphasized that in special applications other values may be
more useful; thus, it may be reasonable to set

(20)

for example, when the intersections Sj' Sn coincide with the "horizontal"
boundary surfaces of the body B.

Volumetrics

For the computation of the volume of B, the integration formula

simplified for the discrete situation applies

vob(B) = 111(X,y)dxdy
1\ L
= hlh2 L L voll({Z I g(Z;Xklyt) ~ O}) (21 )
k=1 /=1
148 SCHAEBEN, AUERBACH, AND SCHUTZ
where vol I ({ Z I g(z;xt,yD ;;:= O}) is the sum of the lengths j+
Z 1- Zj of
intervals with g(Z;Xk,yt) ;;:= * * ] when Zj* denote the roots of
0 for Z E [ Zj,Zj+l
g(z;xt,Yt).

Rendering

The set of points {(xt,Yt,Z *) I g(z *;xt,Yt) = O} C 1R3 can be used to

visualize the body, its outer and inner boundary surfaces by computer
graphics involving voxels (Samet, 1990a, 1990b).

PRACTICAL EXAMPLES AND APPLICATIONS

In a first simple example, a sphere of unit radius is given by its

contour lines in several cross sections. In this example, the plane of any
intersection does not contain more than one (closed) contour line.
Numerical performance of the suggested scheme for approximating
the volume of the unit sphere is summarized in Table 1.

Table 1. Numerical performance for approximation of volume of unit

sphere: calculated value (relative error in percent: ranking of
performance) [cpu time units]; true volume 4.188790.

# gridpoints 8x8 13 x 13 23 x 23
# cross sections
(# points per
counterline)
5 (36) 4.149 (0.94%:5) [0.11] 4.396 (4.96%: 14) [0.28] 4.417 (5.45%:18) [0.88]
5 (72) 4.138 (1.20%:7) [0.21] 4.375 (4.46%:13) [0.57] 4.409 (5.26%: 17) [1.78]
9 (36) 4.043 (3.45%: II) [0.18] 4.253 (1.55%:8) [0.44] 4.270 (1.94%:10) [1.52]
9 (72) 4.028 (3.81%:12) [0.38] 4.237 (1.17%:6) [0.99] 4.263 (1.79%:9) [3.07]
19 (36) 3.979 (5.00%:15) [0.38] 4.201 (0.30%:2) [1.00] 4.208 (0.46%:4) [3.10]
19 (72) 3.977 (5.04%:16) [0.77] 4.178 (0.23%: I) [2.02] 4.207 (0.45%:3) [6.28]

For various choices of (i) the total number of gridpoints, (ii) the total
number of cross sections, and (iii) the total number of points per
counterline numerical performance is compared and ranked in terms of
relative error in percent and cpu time units. It should be noted that the
VOLUMETRICS AND RENDERING 149

two-dimensional discretization of the planes of intersection and the

distance of the sections should conform with one another to yield good
results (cf. Levin's theorem). All results can be judged to be in good
agreement with the "true" theoretical value of 4/3 7r = 4.188.
In a second example, a torus with R = 500, r = 250 is considered to
demonstrate the method's versatility. The planes of intersection have been
selected parallel to a plane containing the torus' axis of largest inertia.
Thus there are planes of intersections which contain more than one, that is
two contourlines, the interior of each corresponds to the interior of the
body to be reconstructed. Selecting two different discretizations (i) seven
equidistant cross sections and a lOx 10 square grid the approximate
volume 6.287E+08 was calculated (see Fig. 5);

G~ i"'-.
~
i"""-

/"

(
~
r-
~ D
(
,.",.

D
.~
D
( LUlIl
-> ".

.....
D
I
V
mu ~

.........
~ t>

Figure 5. Approximately of volume of torus with R == 500, r == 250 from contour lines
seven equidistant planes of intersection with grid of 10 x 10 points resulting in
approximated volume of 6.287 E+08, true volume is 2,(- Rr2 =6.168 E + 08.
150 SCHAEBEN, AUERBACH, AND SClfOTZ

(ii) 17 equidistant cross sections and a 20 x 20 square grid the

approximate volume 6.099E+08 was calculated (see Fig. 6), which are
both in sufficiently good agreement with the "true" theoretical value 2 rr?
Rr2 = 6.168 E+08 of the volume of a torus with R = 500, r = 250.
Had the planes of intersections been selected orthogonal to the torus'
axis of largest inertia, then cross sections with more than one, that is two
nested contour lines would have been generated with the interior of the
inner contour lines corresponding to the body's exterior. In this situation
the final volume of the torus is given by the difference of the volumes of
the two bodies represented by the outer and inner contour lines,
respectively, that is by applying the algorithm successively to each set of
outer and inner contour lines, respectively. The numerical results do not
differ significantly if the number of intersecting planes and the number
of grid points are chosen appropriately .

Figure 6. Approximation of volume of torus with R = 500, r = 250 from contour lines in
17 equidistant planes of intersection with grid of 20 x 20 points resulting In
approximated volume of 6.099 E + 08, true volume is 2 n? Rr2 = 6.168 E + 08.
VOLUMETRICS AND RENDERING 151

CONCLUSIONS

The suggested method compares favorably with methods, which are

more restrictive with respect to feasible shapes and topologies of the
bodies to be reconstructed. Because the problem of reconstructing a
complex body of nested partial bodies can be reduced to a sequence of
three-dimensional problems, and because each three-dimensional problem
can be reduced to a sequence of rather simple univariate problems,
parallel processing provides a promising option.

REFERENCES

Auerbach, S., and Schaeben, H., 1990a, Surface representations

reproducing given digitized contour lines: Math. Geology, v. 22, no.
6, p. 723-742.
Auerbach, S., and Schaeben, H., 1990b, Computer-aided geometric design
of geologic surfaces and bodies: Math. Geology, v. 22, no. 8, p. 957-
987.
Farin, G., 1988, Curves and surfaces for computer aided geometric design
- a practical guide: Academic Press, Inc., New York, 334 p.
Gmelig Meyling, R.H.J., 1988, Three-dimensional reconstruction from
serial cross sections by set-valued interpolation: Memorandum 754,
Faculty of Applied Mathematics, Univ. of Twente, Enschede, The
Netherlands, 30 p.
Klein, R., Pflug, R., and Ramshom, C., 1989, Shaded perspective views
by computer: a new tool for geologists: Geobyte, v. 4, no. 4, p. 16-24.
Levin, D., 1986, Multidimensional reconstruction by set-valued
approximation: IMA Jour. Numer. Anal., v. 6, no. 2, p. 173-184.
Meyers, D., Skinner, S., and Sloan, K., 1992, Surfaces from contours:
ACM Trans. Graph., v. 11, no. 3, p. 228-258.
Omwa, A.A., 1987, Reconstruction of a closed three-dimensior:al body by
set-valued interpolation: Rept. Faculty of Applied Mathematics,
Univ. of Twente, Enschede, The Netherlands, 42 p.
Press, W.H., Flannery, B.P., Teukolsky, S.A., and Vetterling, W.T., 1990,
Numerical recipes: the art of scientific computing: Cambridge
University Press, Cambridge, 702 p.
152 SCHAEBEN, AUERBACH, AND SCHUTZ
Samet, H., 1990a, The design and analysis of spatial data structures:
Addison-Wesley, Reading, Massachusetts, 496 p.
Samet, H., 1990b, Applications of spatial data structures: computer
graphics, image processing and GIS: Addison-Wesley, Reading,
Massachusetts, 480 p.
Schumaker, L.L., 1983, On shape preserving quadratic spline
interpolation: SIAM Jour. Numer. Anal., v. 20, no. 4, p. 854-864.
Walsh, J., and Brown, S., 1992, Computing areas and volumes from
contour maps: Geobyte, v. 7, no. 3, p. 44-47.
 THE EFFECf OF SEASONAL FACTORS
ON GEOLOGICAL INTERPRETATION OF MSS DATA
D.Yuan
Desert Research Institute, Las V egas, Nevada, USA

J. E. Robinson
Syracuse University, Syracuse, New York, USA

M. J. Duggin
SUNY College of Environmental Science CD1d Forestry, Syracuse, New
York, USA

ABSTRACf

Temporal variations are considered to be important factors in their

effects on remotely sensed image data and discrimination analysis of the
images as a basis for geologic interpretation. Some systematic temporal
factors may be corrected by instrument adjustment but random and
stochastic factors are not necessarily systematic. However, statistical
methods can be used to determine the effect of the stochastic temporal
factors. In this study, 29 MSS subscenes from northern Lincoln County,
Nevada images were obtained and nine different lithologies were sampled.
Two linear statistical models related to the stable background and
seasonal factors were constructed and tested. The results suggest that (1)
the band brightness values resulting from different lithologies are
statistically different~ (2) the seasonal effects do exist in a statistical
sense; (3) the seasonal factors are such that the bands show higher values
in summer and fall, but lower in winter and spring~ (4) the seasonal
effects are consistent within the four MSS bands for any given lithology

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 153
154 YUAN, ROBINSON, AND DUGGIN

area; and (5) the seasonal factors are consistent for different types of
lithology within any given MSS band. These results provide statistical
support for lithological classification using remotely sensed data.

TIME FAcroRS

Duggin (1985) summarized the systematic and random factors that

limit the discrimination and quantification of terrestrial features using
remotely sensed radiance. If the band brightness from one ground target
at one time is considered, the factors affecting the measured values result
from the passage of radiant energy to the target, the reflection of radiance
from the target, the passage of reflected target radiance to the detector,
the effect of the detector, the effect of cloud and haze, and the effect of
background radiance. The systematic variation across an image can be
corrected but not the random variation. In a later study, Duggin,
Sakhavat, and Lindsay (1987) examined systematic and random variations
in recorded radiance for two LANDSAT Thematic Mapper bands of
images obtained over a uniform agricultural area before and after harvest.
They determined that the variations were large enough to be considered
in the discrimination of the terrestrial cover types. The stated result also
is true for lithological discrimination using satellite images, as long as the
area has sufficient vegetation cover (Duggin and Robinove, 1990).
If the band brightness from a ground target in a fixed location is
considered, the variation of the recorded radiance varies with respect to
time. This variation primarily depends on the nature of the changing
factors such as vegetation cover, cloud, snow, rainfall, and other random
effects. If the detector, emitted energy, and satellite conditions are
assumed constant when compared with that resulting from naturally
changing factors, the variation across a series of images because of
equipment factors is small and can be ignored. Therefore, factors
affecting the band brightness from a constant ground target can be
thought of as "stochastic".
The effect of vegetation varies on a seasonal basis, however, there
are many uncertain factors, such as precipitation, temperature, and number
of days of sunshine. Thus, the effect resulting from vegetation should be
both "seasonal" and "stochastic". The snow effect is mainly "stochastic".
Because there is more snow in winter than in other seasons. Snow also
is a "seasonal" factor. Rainfall may be random, however, in the study
INTERPRETATION OF MSS DATA 155

areas, there is more rain in winter and spring than in summer and fall.
Cloud and atmospheric turbulence usually are random factors, but some
time dependent variation may need to be considered. If the band
brightness from a ground target in a series of images is considered to be
consistent, then the variation of the radiance will show both "seasonal"
and "stochastic" characteristics. These characteristics should be detectable
by statistical hypotheses testing methods. In order to simplify the
discussion, only the seasonal factors are considered. However, the
monthly, daily even hourly factors could be considered in the same way
as long as there are enough images to perform statistical testing. Table 1
gives some of the time differing factors affecting the band brightness of
MSS data. Note that the "systematic time factors" here are not the
"systematic factors" that are dependent on the observation systems in the
usual sense that have been discussed by authors such as Duggin (1985)
and Duggin and Robinove (1990). Such factors also are stochastic in that
they show some statistical trend or periodicity. Factors in this study seem
to be systematic only when large numbers of images from a single area
are examined.
It is necessary to estimate the effects of time factors, or more
specifically, to estimate the seasonal factors, and determine their
dependency on such factors as the frequency spectra of the indicating
bands, the lithological type, and even the interaction of bands and
lithologies. For this, it is necessary to construct statistical models.
In order to test the seasonal effects, a series of 29 MSS subscene
images from northern Lincoln County, Nevada were collected. These
images were recorded in the period from 13 September 1972 to 16 June
1977. The study area is centered approximately at 114°30' West
longitude, 30°20' North latitude, extending about 23.5 km on each side.
The image size is 300 x 300 pixels with each pixel representing a 79 by
79 m2 ground area after rectification. Figure 1 shows the coverage of
these images and the selected subscene study area. Robinove and Chavez
(1978) studied the temporal variation of the whole scene albedo of the
same data set. Table 2 lists some basic information about this data set.

SURFICIAL GEOWGY

The study area is typical southwest Valley and Range province

topography with elongated north-south mountain ranges and intervening
156 YUAN, ROBINSON, AND DUGGIN

Table 1. Summary of time factors affecting received lithological radiance.

Type of time Subclasses Characteristics Examples

factors
Systematic Trend time Long tenn, linear, Greenhouse effects
time factors factors invisible
Long tenn, cyclic, Sunspot activity
invisible
Cyclic time Yearly (seasonal) Vegetation, Precipitation,
factors Obvious to invisible Snow, Wind, Cloud
Monthly, maybe Lunar variation
invisible
Daily Sun angle, Brightness of
the sky
Stochastic Catastrophic N ondirectional, Floods, Snow Stonn
time Factors stochastic time obvious visible
factors Directional, obvious SlidelEarth Quick,
visible Human Activity
Non-catastrophic Visible Precipitation, Snow,
stochastic time Clouds
factors
Invisible Atmospheric Disturbance

The whole
scene of the images
The srudy area

Figure 1. Approximate location of selected study area and area covered by satellite
images used for study.
INTERPRETATION OF MSS DATA 157

Table 2. Basic information about image data stack used for the study
(data from Robinove and Chavez, 1978).

Date of Average And Minimum Brightness Remarks

expo B4 B4 B5 B5 B6 B6 B7 B7
ave min ave mm ave min ave min
720913 47 13 50 6 50 4 23 0
721124 54 9 50 1 47 0 19 0 snow widespread
721230 60 6 57 5 51 3 21 0 snow widespread
731014 36 8 39 1 39 0 17 0
740623 55 16 61 8 61 7 28 0
740816 48 11 53 7 53 3 24 0
740903 47 12 51 6 50 5 21 0
741202 25 1 24 0 24 0 10 0 snow on Mtns
750609 41 9 57 9 66 0 28 0 scattered clouds
750627 42 11 57 0 66 0 28 0
750802 40 10 55 9 63 3 26 0
751215 49 11 48 5 45 2 18 o snow widespread
760120 23 9 22 3 27 2 9 o snow on Mtns
760129 18 4 25 2 27 0 11 o snow on Mtns
770410 39 8 53 8 58 6 24 o snow on Mtns
760516 41 10 57 11 65 6 28 o snow on Mtns
760621 42 10 59 9 67 5 29 0
760709 41 11 58 10 64 6 26 0
760805 43 12 46 6 47 3 20 0
760823 38 11 40 5 42 3 17 0
760901 35 6 49 4 54 2 22 0
761007 25 4 35 2 40 0 16 0
761016 29 9 29 3 30 2 13 o clouds and wind trace
761025 25 5 34 3 37 0 15 o scattered clouds
761130 15 1 21 1 24 0 10 0
761218 14 1 19 0 21 0 8 0
770405 35 9 47 6 52 4 21 o snow on Mtns
770423 40 9 54 8 60 4 25 0
770616 41 9 56 8 64 4 27 0
158 YUAN, ROBINSON, AND DUGGIN

dry valleys. The southern boundary of the subscene of approximately 11

miles north of Pioche. Climate and vegetation are related and
altitude-dependent. Vegetation is scarce in the valleys, but is considerable
on the mountains and uplands. The higher mountain areas with steppe
vegetation have more rainfall, colder winters, and cooler summers. The
lower areas are covered by sagebrush and the higher areas by single-leaf
pinon and juniper forests (Tschanz and Pampeyan, 1970).

(. .!. .!...J identified farmland area

Figure 2. Brief lithological map and training area distributions. "Blocks" on map are
training areas of correspondent lithologies. Piece of farmland in Lake Valley is
identifiable from all images. Boundary differs with respect to seasons.

The eastern two-fifths of the subscene lies on the western side of

the Wilson Creek Range, which is one of the highest and most
inaccessible ranges in Lincoln County (Tschanz and Pampeyan, 1970).
The part of this range in the subscene area is composed entirely of .
volcanic rocks. The western three-fifths of the subscene is Lake Valley.
The majority of the valley surface consists of older alluvium deposits.
The northwestern of the valley is recent lake sand deposits composed of
material eroded from the nearby mountains. Alluvium fans and aprons
occur continuously around the mountain front.
INTERPRETATION OF MSS DATA 159

About 60 percent of the rocks exposed in the study area are

Tertiary and Quaternary sedimentary deposits~ the remaining 40 percent
are volcanics. There are only a few patches of Cambrian and Ordovician
sedimentary rocks. The northwestern portion of the subscene is the Lake
Valley playa deposit. The eastern part of the sub scene is Tertiary volcanic
rocks that flank the southeastern side of Mount Wilson. There are 10

Table 3. Mean band brightness value by season for lithologies.

Lith. Season N Band 1 Band 2 Band 3 Band 4

QL All Season 29 51.28 61.98 62.13 25.79
Spring 3 47.44 65.62 68.49 28.41
Summer 7 58.74 78.21 82.87 35.31
Fall 11 51.54 60.95 60.19 25.45
Winter 8 45.81 47.84 44.28 16.96
QOL All Season 29 36.42 44.75 47.29 19.86
Spring 3 29.18 41.95 47.84 20.32
Summer 7 36.03 50.20 58.10 25.26
Fall 11 32.48 40.25 42.91 18.52
Winter 8 44.89 47.20 43.65 16.82
TVT All Season 29 28.56 30.90 40.79 18.85
Spring 3 44.19 54.65 63.16 26.10
Summer 7 24.93 30.65 48.21 23.59
Fall 11 22.53 23.57 33.74 16.80
Winter 8 34.19 32.31 35.60 14.79
IVY All Season 29 25.61 29.24 35.80 15.56
Spring 3 21.96 29.94 39.33 17.14
Summer 7 28.97 38.29 48.72 21.34
Fall 11 24.66 27.19 33.40 14.70
Winter 8 25.34 23.86 26.49 11.06
TG All Season 29 24.69 27.96 34.27 14.92
Spring 3 21.99 28.84 37.69 16.25
Summer 7 28.38 36.66 45.73 20.06
Fall 11 25.02 27.19 33.12 14.52
Winter 8 22.02 21.06 24.52 10.48
160 YUAN, ROBINSON, AND DUGGIN

Table 3. (Cont.) Mean band brightness value by season for lithologies.

Lith. Season N Band 1 Band 2 Band 3 Band 4

TKVU All Season 29 21.09 21.85 29.99 14.12
Spring 3 20.80 25.99 36.53 16.76
Summer 7 23.52 28.14 41.45 19.92
Fall 11 20.34 19.91 27.64 13.36
Winter 8 20.09 17.48 20.73 9.10
OEOP All Season 29 27.01 30.75 35.93 15.89
Spring 3 25.35 32.53 40.61 17.89
Summer 7 29.87 39.41 47.92 21.42
Fall 11 25.56 28.06 32.04 14.32
Winter 8 27.11 26.21 29.04 12.47
CAMU All Season 29 27.02 31.33 38.16 16.87
Spring 3 24.25 32.68 42.35 18.54
Summer 7 31.05 40.57 51.05 22.73
Fall 11 27.90 31.62 37.66 16.83
Winter 8 23.31 22.35 26.00 11.17
JASP All Season 29 25.65 29.05 36.58 16.17
Spring 3 24.03 31.80 41.70 17.78
Summer 7 27.95 36.10 47.92 21.35
Fall 11 26.12 29.13 36.69 16.62
Winter 8 23.58 21.74 24.60 10.40

mappable lithological units occurring on the surface in the study area.

Classified lithologies are: Pliocene younger lake beds (QI), Pliocene older
alluvium deposits (Qol), Tertiary younger volcanic rocks, undivided,
including intravolcanic sedimentary rocks and perlite (Tvy), Tertiary tuffs
and tuffaceous sediments (Tvt), Tertiary granite stocks and dikes (Tg),
Tertiary volcanic rocks, undifferentiated (TKuv), Middle Ordovician
Eoreka Quartzite (Oe), Lower Ordovician Pogonip Group (Op), Upper
Cambrian Limestone and dolomite (-G-u) and a small patch of Jasperoid
(Tschanz and Pampeyan, 1970). Because two of them (Oe - Middle
Ordovician Eoreka Quartzite and Op - Lower Ordovician Pogonip Group)
were difficult to locate and distinguish from each other in the images,
they were combined to form a "mappable unit" (Oe+Op or Ope in the
figures) for this study. A simplified map of lithology and training area
INTERPRETATION OF MSS DATA 161

selection for the study area is shown in Figure 2. A detailed description

of the lithological units is given in Tschanz and Pampeyan (1970). The
mean seasonal band brightness values for the 9 lithologies are calculated
and listed in Table 3. In order to actually view the temporal change over
the selected temporal images, 14 of the 29 subscene images are converted
into albedo images and displayed in Figure 3.

BAND-TIME INTERACflON TEST

The average brightness values for the 9 lithological units for the
4 bands from 29 MSS images in this study area have been computed and
plotted against time of exposure in Figure 4. Although there are some
irregularities, it is obvious that the time variant pattern for the 4 bands for
any of the given lithologies are similar. Two time curves are considered
to be "similar" or "in phase" if they reach their peaks and troughs almost
simultaneously.
In order to analyze the effects of temporal factors on the band
brightness for the lithologies, a specific testing procedure had to be
designed. Observation of the band brightness values - time plots for
different lithologies (Fig. 4) suggests that the effect resulting from time
factors could be approximated by their linear portion. For example,
considering the time effects to different bands of a given lithology, the
suitable statistical model would be:
(1)

Where BVLlijk is the k'th received band i radiance at time period j for
lithology 1, III is the background value for lithology I, BAND j is the
effect of band i, SEASONj is the effect of time period j, BSij is the
interactive affect of band i and time period j, E jjk is the random error for
the individual image. The subscripts i = 1,2,3,4~ j=Spring, Summer, Fall,
Winter and k various depending on the number of images in
correspondent period. We want to test the hypotheses:

for any i and j. (2)

If H Olbs could not be rejected, we will further test:

162 YUAN, ROBINSON, AND DUGGIN

for any i=I,2,3,4. (3)

and

HOls: SEASO~=O, for any j=I,2,3,4. (4)

The hypothesis H Olbs ' H 01b ' and HOIs can be tested by the SAS
general linear model procedure GLM. Table 4 gives the tables for
analysis of variance for the 9 lithological units. In GML procedure, the
program first checks the validity of the model (i.e. compute the F-value
for the model and compute the probability that the theoretic F-value is
greater than the observed F-value). After the validity of the model is
checked, the program then continuous to check the effects of the
contributing factors (i.e. band passes and seasons). The former checks the
effect of the result of "MODEL," whereas the later checks the effect of
"F ACTOR". For example, for lithology QL (Quaternary Lake Deposit),
the F-value for checking the model is 10.88, the computed Prob (Ftheoretic
> Fobserved) = 0.0001. This suggest that the model (is 1) valid for lithology
QL. The same conclusion could be obtained for the rest 8 lithologies. All
the computed the Prob (Ftheoretic > Fobserved) are smaller than 0.001 (8 of
them are equal to 0.0001, and one of them (QOL) equals 0.0006). This
suggests that all the 9 models are valid even at 99% significance level.
The rest of the hypotheses can be tested in the same way.
For hypothesis (2) H 01bs : BS ij = 0, the computed Prob (Ftheoretic >
Fobserved) for the 9 lithologies range from 0.0770 (TKVU) to 0.9162
(QOL), that is 0.5758 (QL), 0.9162 (QOL), 0.l137 (TVT), 0.4877
(TVY), 0.2816 (TG), 0.0770 (TKVU), 0.7141 (OEOP), 0.2787 (CAMU),
0.4936 (JASP). 8 out of 9 lithologies with Prob (F theoretic> F observed) > O.l
TKVU (0.0770 ) is the only exception. Actually, all are greater than 0.05.
This suggests that at a given significance level, say 95%, the band-season
interactions are insignificant statistically, regardless of the lithology. Also
it can be inferred that the BAND-SEASON interaction is small compared
with the effects directly the result of the BAND and SEASON factors in
all situations. In other words, the seasonal effect is consistent for the
different bands for any given lithology. This result can be observed from
the band value v.s. time plots for each of the lithologies because the
different bands have similar variant patterns with respect to time. For
example, they have similar variation periods and attitudes, and reach the
INTERPRETATION OF MSS DATA 163

Figure 3. 14 albedo images converted from LANDSAT MSS images obtained from
13/911972 to 16/6/1977.
 164 YUAN, ROBINSON, AND DUGGIN

; 120 ---0- b4_qUn..n __ b6_ql_mcan

- bS_ql_....." _ b7_ql_m..n
:100
w 10
~

~60
D <10

1091 1464 1830

ftumbc:r or d. y. (rom 71lJJ1l
I~ r---=-----~----~----------__;
bO_qol_m.... _ b6_qol_mun 70r---=-~----------------------,
---0- bO _ " - _ _ b6_"-",,,""
.? 131 bS_qol_mcan _ b7_qol_mcan ; 60 - bS_,,-mun _ o7_'l-m ...
: HlJ • SO

~ 10 w 40
~

~60 ~
-.:
30
.Q 40 D 20
."
] 20 5 10
"" D

732 1091 1464 1830

number or ciay. rrom 71lJ/71 number ot dol,., from 7123112
80r---=-~~--------------------,
. 100 r------------------------------,
--e-- b4 _M_mCAl\ ---It- b6_M_mc.a.n
bS_M_m..n _ b
~ SO

r
]
~ 60
~

40

20
.Q

1091 1091 1464 1130

numb~r or dlY. rrom 7113/72. "umbtr 01 day. (rom 711311'1

w
6Or------------------------------,
---0- b4_"""'_m ... _ _ b6_"""'_m<Cl
10r-------------------------------,
b4_cpe_mc:an ____ bo6_CIpCI_man
~ so
_ bS_""",_mcan _ b7 Ikvu_mun ~ 10 bS_opc_mQn -.- b7_opc_mC&ll
-: 60
40
SO
30

...~:
~

flO
] to 1. 20
... o~~~~~~~~~~~~~~~
... 10

o 1091 1464 1330 366 1091 1464 1330

ntunber o( d,y. from 7123/72 I'lumber or day. (reM 7IDln
70r-----o-~,--------------------------, 70r---~--------------------------__,

..
b4.J"p_mcatl
, 60 ; 60 bS ....Jup_tnc.an
• 50 -: SO
40 ~
40

~ 30

..,... 20
'" S
...: 10
'"
10
O~-L~~-L~~~~~_L~~~~

1091 1464 lIl0 0 732 1091 1464 l~3O

number Qr dlY' from 7113/n number or day_ (rom 71ZJ172

Figure 4. Band brightness value variant pattern for lithologies. Where b4_ql_mean
stands for mean band 4 brightness value of lithology Ql, and so forth. "camu" stands
for Cambrian carbonate rocks, "ope" stands for Ordovician rocks. Other lithologies see
explanation in text.
 INTERPRETATION OF MSS DATA 165

Table 4. ANOVA for band - time interaction tests (for 9 lithologies).

A. ANOV A classification level information.

CLASS LEVELS VALUES
BAND 4 BAND 4, BAND 5, BAND 6, BAND 7
SEASON 4 SPRING, SUMMER, FALL, WINTER
NUMBER OF OBSERVATIONS IN DATA SET = 116
DEPENDENT VARIABLE: BAND VALUE

B. ANOV A table by lithology.

LITHO- EFFECT SOURCE DF SS MS F- PR
LOGY VAL. >F
QL MODEL MODEL 15 36636.98 2442.47 10.88 0.0001
ERROR 100 22450.94 224.51
TOTAL 115 59087.92 513.81
FACTOR BAND 3 25462.91 8487.64 37.81 0.0001
SEASON 3 9464.63 3154.88 14.05 0.0001
BAND· SEASON 9 1709.44 189.94 0.85 0.5758
QOL MODEL MODEL 15 16174.30 1078.29 2.98 0.0006
ERROR 100 36175.41 361.75
TOTAL 115 52349.71 455.21
FACTOR BAND 3 13334.96 4444.99 12.29 0.0001
SEASON 3 1440.01 480.00 1.33 0.2699
BAND· SEASON 9 1399.32 155.48 0.43 0.9162
TVT MODEL MODEL 15 13981.17 932.08 7.52 0.0001
ERROR 100 12401.60 124.02
TOTAL 115 26382.77 229.42
FACTOR BAND 3 7061.97 2353.99 18.98 0.0001
SEASON 3 5087.24 1695.75 13.67 0.0001
BAND· SEASON 9 1831.97 203.55 1.64 0.1137
TVY MODEL MODEL 15 9580.77 638.72 6.86 0.0001
ERROR 100 9308.77 93.09
TOTAL 115 18889.54 164.26
FACTOR BAND 3 6223.96 2074.65 22.29 0.0001
SEASON 3 2562.69 854.23 9.18 0.0001
BAND· SEASON 9 794.12 88.24 0.95 0.4877
166 YUAN, ROBINSON, AND DUGGIN

Table 4 (Cont.)
B. (Cont.) ANOVA table by lithology.
LITHO- EFFECT SOURCE DF SS MS F- PR>F
LOGY VAL.
TG MODEL MODEL 15 8841.06 589.40 11.76 0.0001
ERROR 100 5010.93 50.11
TOTAL 115 13851.99 120.45
FACTOR BAND 3 5667.87 1889.29 37.70 0.0001
SEASON 3 2615.56 871.85 17.40 0.0001
BAND· SEASON 9 557.63 61.96 1.24 0.2816
TKVU MODEL MODEL 15 6506.04 433.74 10.59 0.0001
ERROR 100 4093.84 40.94
TOTAL 115 10599.88 92.17
FACTOR BAND 3 3667.61 1222.54 29.86 0.0001
SEASON 3 2174.40 724.80 17.70 0.0001
BAND· SEASON 9 664.04 73.78 1.80 0.0770
OEOP MODEL MODEL 15 9114.25 607.62 5.97 0.0001
ERROR 100 10172.01 101.72
TOTAL 115 19286.26 167.71
FACTOR BAND 3 6281.13 2093.71 20.58 0.0001
SEASON 3 2199.01 733.00 7.21 0.0002
BAND· SEASON 9 634.11 70.46 0.69 0.7141
CAMU MODEL MODEL 15 11336.09 755.74 11.45 0.0001
ERROR 100 6602.33 66.02
TOTAL 115 17938.42 155.99
FACTOR BAND 3 6920.21 2306.74 34.94 0.0001
SEASON 3 3678.20 1226.07 18.57 0.0001
BAND· SEASON 9 737.68 81.96 1.24 0.2787
JASP MODEL MODEL 15 9711.32 647.42 7.05 0.0001
ERROR 100 9189.71 91.90
TOTAL 115 18901.03 164.36
FACTOR BAND 3 6240.34 2080.11 22.64 0.0001
SEASON 3 2692.85 897.62 9.77 0.0001
BAND· SEASON 9 778.13 86.46 0.94 0.4936

peaks and troughs almost at the same times. Their variations are
harmonic.
INTERPRETATION OF MSS DATA 167

For hypotheses Botb: BANDi = 0, the tests show that at a 95%

significance level the hypothesis should be rejected (Prob (F theoretic >
Fobserved) = 0.0001 for all the lithologies). These suggest that each lithology
usually will have different values for different bands.
The test does produce anomalies for Bois: SEASONj = O. For 8
out of 9 lithologies Prob (Ftheoretic > Fobserved) < 0.001, which suggests that
the SEASON factors do have an effect at 99% level on the band
brightness. However, the computed Prob (Ftheoretic> FobserveJ = 0.2699 for
Qol, which suggests that the seasonal effect for lithology Qol is not as
important as it is for other lithologies. However, there is some
contribution. Generally, the seasonal factors do have some effect on band
brightness for ground lithologies and in most situations, this effect is
strong and should not be ignored.

LITHOLOGY-TIME INTERACI10N TEST

The foregoing suggests that band brightness values for different

lithologies differ statistically in MSS data and the seasonal effects should
also have some effect on the variation for a given band and a given
lithology. In Figure 5, the mean band brightness values for nine
lithologies are plotted by seasons for the four bands. The actual mean
seasonal band brightness values are listed in Table 3. For example, in the
first diagram of Figure 5 (band 4), Quaternary Lake Deposits (QI) has the
highest band brightness value for all the seasons with average value
51.28. Tertiary Volcanics Undifferentiated (TKvu) has the lowest band
brightness value for all the seasons with average value 21.09. The same
regularity can be observed in the other diagrams of Figure 5. The general
tendency for all the rocks is that they have a brighter value in summer
and fall, whereas have darker value in winter and spring for all four
bands. Also, the recent sedimentary rocks (QI, Qol) have higher band
brightness values than the older sedimentary rocks (Op+Oe, {;-u),
whereas the older sedimentary rocks have greater band brightness values
than volcanic rocks with possible exception of Tvt. The Tvt exception
could be due to physical properties, mixed Tertiary tuffs and tuffaceous
sediments, or due to sampling errors.
In order to test the temporal effect due to different lithologies for
a given band, the time factors are grouped into four seasonal factors in
the following lithology - time interaction model:
168 YUAN, ROBINSON, AND DUGGIN

BYL..:",=b.+UTH +SEASONJ+LS81.+E81"... (5)

5.,.. I 8

Where BVLgijk is the kith received band i radiance at time period j for
lithology g, bi is the background value for band i, LIm. is the effect of
lithology g, SEASONj is the effect of time period j, LSgj is the
interactive affect of lithology g and time period j, Egjk is the random error
for the individual image. The subscripts g = 1,2, ... , 9; j=Spring, Summer,
Fall, Winter, and k various depending on the number of images in the
corresponding period. It is necessary to test the hypotheses:

Hl12bB: LSu=O, for any i and j; (6)

HCY1JI: UTHj=O, for any i=1,2,3,4; (7)

HC1J.r: SEASO~=O, for any j=1,2,3,4. (8)

Model (5) and hypotheses (6), (7), (8) can be tested by applying
the interactions between seasonal factors and lithologies. The SAS GML
procedure is used for testing these hypotheses. Table 5 lists the results for
analysis of variance for the four bands. The SAS program tests the
validity of the models first and then checks the effects resulting from
concerning factors. The following is a brief interpretation of these
results.
All the models for the four bands based on (5) fit well within the
Prob (Ftheoretic > FobserveJ = 0.0001. Thus, there is no problem in accepting
model (5). The results for testing Ho2bs: LS ij = 0, for any i and j show
that Prob (Ftheoretic > FobserveJ = 0.8521, 0.2546, 0.4551, 0.6666 for MSS
band 4 to band 7 respectively. This suggests that, for a given significant
level 95%, these lithology-time interactions are statistically insignificant.
In other words, seasonal effects on band brightness do not depend on the
actual lithology from which the radiance is obtained. Rather, seasonal
factors are consistent in spite of different lithologies. This is an important
result obtained from the statistical analysis as it provides the foundation
for the possibility of extracting the "pure" radiance from the lithology out
of a mixture of lithologies and other temporal factors. According to the
statistical results, the temporal effects could be considered as time
 INTERPRETATION OF MSS DATA 169

Band 4 brightness value by seasons

..
60
• I?l I:lJ

..
b4_ql_me.an b4 _IYlJTIe.an b4_ope_mean
PJ [] b4_ILme.an C b4_CAmu_mean

•
::J b4_qol_mean

> m b4_lVy_mean b4_tkvu_mean I:l b4jasp_mcan

.
<II
<II

C
50

.cc.o
..,
;:
40
.,p

.
."
C

= 30

20

10
Spring Summer Fall Winter All seasons

Band 5 brightness value by seasons

.
80
•
r.I
b5_ql_me.an
b5_qol_mean
12 bS_IYl_mean
[] b5_ILmean
a b5_ope_mcan
[J b5_CAmu_mean

•
::J

'"
>
70 B bS_lVy_mean b5 _tkvu_mean ta bSjosp_mean

..
<II
'"
:c'<I>"
60

..,;: 50
II>

.
."
.:

=
40

30

20

10
Spring Summer Fall Winter All seasons

Figure 5. Mean seasonal band brightness value by lithologies (explanation see Fig. 3).
 170 YUAN, ROBINSON, AND DUGGIN

Band 6 brightness value by seasons

.. 80 •
121
b6_ql_mean
b6_qol_mean
ra
D
b6_tvt_mean
b6_tLmean
a b6_Qpe_mean
[] b6_e&mUfte&n

..., •
::I
;;;
m b6_tvy ftean b6_lkvu_mean IZl b6j .. p_mean

.
.,
C!
70

~60
L:
""
~ SO
..,
""
I':::: 40

30

20

10
Spring Summer Fall Winter All seasons

Band 7 brightness value by seasons

. •
P.I
b7_ql_mean
b7_qol_mean
EI b7_lvt_mean
0 b7_I&-mean
EJ b7 _ope_m:.>n
D b7 _e&mu _mean
-=."
'"'"
30 EI b7_tvy_mean
• b7 _tkvu _mean [J b7j .. p_mean

'"C!
~
...
""
.... 20

"" C!

="
10

Spring Summer Fall Winter All seasons

Figure 5. (Cont.) Mean seasonal band brightness value by lithologies (explanation see
Fig. 3).
INTERPRETATION OF MSS DATA 171

Table 5. ANOVA for lithology - time interaction tests (for 4 MSS bands).

A. ANOV A: CLASS LEVEL INFORMATION

CLASS LEVELS VALUES
SEASON 4 SPRING SUMMER FALL WINTER
LITH 9 CAMU JASP OEOP QL QOL TG TKVU TVT TVY
NUMBER OF OBSERVATIONS IN DATA SET - 261

B. ANOV A table by band.

BAND EFFECT SOURCE DF SS MS F- PR
VAL. >F
4 MODEL MODEL 22947.45 655.64 35 0.0001
ERROR 225 44185.79 196.38
TOTAL 260 67133.24 258.20
FACTOR SEASON 3 555.83 185.28 0.94 0.4204
LITH 8 19105.27 2388.16 12.16 0.0001
SEASON*LITH 24 3286.36 136.93 0.70 0.8521
5 MODEL MODEL 35 44957.63 1284.50 7.94 0.0001
ERROR 225 36383.18 161.70
TOTAL 260 81340.81 312.85
FACTOR SEASON 3 6805.74 2268.58 14.03 0.0001
LITH 8 33541.74 4192.72 25.93 0.0001
SEASON*LITH 24 4610.15 192.09 1.19 0.2546
6 MODEL MODEL 35 42198.54 1205.67 9.06 0.0001
ERROR 225 29926.51 133.01
TOTAL 260 72125.05 277.40
FACTOR SEASON 3 17923.66 5974.55 44.92 0.0001
LITH 8 21052.60 2631.58 19.79 0.0001
SEASON*LITH 24 3222.28 134.26 1.01 0.4551
7 MODEL MODEL 35 7597.20 217.06 9.95 0.0001
ERROR 225 4910.07 21.82
TOTAL 260 12507.27 48.10
FACTOR SEASON 3 4170.51 1390.17 63.70 0.0001
LITH 8 2980.18 372.52 17.07 0.0001
SEASON*LITH 24 446.51 18.60 0.85 0.6666
172 YUAN, ROBINSON, AND DUGGIN

dependent optical filters under which each lithology retains its own
contribution.
For testing Ro2b: LI~ = 0 for any i, the computed Prob (Ftheoretic
> F observeeJ =0.0001 for all the four models of the MSS bands. This
suggests that the lithology effects are significant at 99% level. In other
words, different lithologies do have different band values in a statistical
sense. This is the foundation for lithological classification by using
specific band brightness values for MSS data.
For testing Ro21: SEASONj = 0, the computed Prob (Ftheoretic >
Fobserved) = 0.0001 for band 5, 6, and 7. This suggests that the seasonal
effects are significant for band 5, band 6, and band 7 at the 99% level.
This suggests that the lithology effects and the seasonal effects can be
considered significant at 90% level. However, the Prob (Ftheoretic > Fobserved)
= 0.4204 for band 4 suggests that the seasonal factors do not have a
significant effect on band 4 brightness in a statistical sense. In other
words, seasonal factors have a weak effect on band four.
Summarizing the results, it can concluded that the time factors are
consistent within different bands (except possibly band 4) for any given
lithology, and are consistent for different lithologies for any given band.
This provides a foundation for the possibility of extracting the lithological
background from a series of actual images. However, determining a
reasonable representation of the lithological trend from the temporal
images remains a problem that must be considered.

CONCLUSIONS

Statistical results obtained in this study can be summarized as

follows
(1) The different lithologies in the study area have different mean
values of band brightness for each band. This provides the foundation for
statistical classification based on the brightness values. Without the
certainty of these differences any classification for lithologies based on
band brightness values would not be significant.
(2) The received radiance or band brightness values show
different mean values for each band for any given lithological formation.
This suggests that band brightness from individual band represents
different physical properties of the material.
INTERPRETATION OF MSS DATA 173

(3) Seasonal factors usually cause important changes in band

brightness. The rocks usually have higher band brightness values in
summer and fall and lower values in winter and spring. Presumably these
variations are the results of a clearer and drier atmosphere along with less
vegetation in summer and fall, whereas there is greater rainfall and more
vegetation in winter and spring.
(4) The seasonal factors do not have strong interaction with
different bands. Thus, it can be concluded that the seasonal factors are
consistent for individual bands and lithologies.

REFERENCES

Duggin, M.J., 1985, Factors limiting the discrimination and quantification

of terrestrial features using remotely sensed radiance: Intern. Jour.
Remote Sensing, v. 6, no. 1, p. 3-27.
Duggin, M.J., Sakhavat, H., and Lindsay, J., 1987, The systematic and
random variation of recorded radiance in a LANDSAT thematic
mapper image: Intern. Jour. Remote Sensing, v. 6, no. 7, p.
1257-1261.
Duggin, M.J., and Robinove, C.J., 1990, Assumptions implicit in remote
sensing data acquisition and analysis: Intern. Jour. Remote
Sensing, v.ll, no.l0, p.1669-1694.
Robinove, C.J., and Chavez, P.S. Jr. 1978, LANDSAT albedo monitoring
method for an arid region: Presentation at the AAAS Intern.
Symp. Arid Region Plant Resources, Lubbock, Texas.
Tschanz, C.M., and Pampeyan, E.H., 1970, Geology and mineral deposits
of Lincoln County, Nevada: Nevada Bureau of Mines Bull. 73,
187p.
RECONSTRUCTION OF THE LEDUC AND WABAMUN
ROCK SALTS, YOUNGSTOWN AREA, ALBERTA

Neil L. Anderson
University of Missouri-Rolla, Rolla, Missouri, USA

R. James Brown
University of Calgary, Calgary, A Iberta, Canada

ABSTRACf

There are four relatively thick (>20 m) bedded Devonian rock salt
units in the Youngstown area, south-central Alberta, Canada (T2S-3S, RS-
20W4M), those of the Cold Lake Formation, Prairie Formation, Wabamun
Group, and Leduc Formation. There is no substantive evidence that the
former two Middle Devonian rock salts (Cold Lake and Prairie) have
been leached post-depositionally in the study area. However, the latter
two Upper Devonian rock salts (Leduc and Wabamun) have been leached
extensively in places, and are preserved now as irregularly shaped bodies
of variable areal extent, having maximum net thicknesses on the order of
45 m and 40 m, respectively.
The dissolution of the Wabamun and Leduc rock salts has caused
the overlying strata to subside, more-or-Iess on a one-to-one basis. As a
result, at any well location in the study area there is a direct correlation
between the relative subsidence of a specific geologic horizon and the net
thickness of all rock salt dissolved after the deposition of that horizon.
This direct relationship between dissolution and subsidence is the key to
reconstructing the paleodistribution of the Wabamun and Leduc rock salts
and determining the timing and mechanisms of leaching.

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 175
176 ANDERSON AND BROWN

In this paper, a suite of present-day and original net-thickness

maps for the Leduc and Wabamun rock salts in the Youngstown area of
Alberta are presented. These maps were generated using well-log control.
They are based mostly on the interval thicknesses of the residual rock
salts and the encompassing Wabamun and Leduc intervals, and variations
in structural relief at the Wabamun and Leduc levels and along the
overlying post-Devonian horizons. Interpretation of the maps of net salt
thickness supports the thesis that salt dissolution in the study area was
initiated or enhanced by some or all of four principal processes: (I) the
near-surface exposure of the rock salt, which resulted from the erosion of
the overlying Paleozoic sediment during the pre-Cretaceous hiatus~ (2) the
partial dissolution of underlying rock salt~ (3) regional faulting or
fracturing during the mid-Late Cretaceous~ and (4) glacial loading and
unloading.
In support of the present-day net-salt-thickness maps, two seismic
profiles also are presented. The first line images an isolated remnant of
Wabamun rock salt~ the second crosses a linear Wabamun salt-dissolution
feature. Leaching at the second location was initiated during the mid-Late
Cretaceous, probably as a result of faulting or fracturing.

IN1RODUcnON

The residual rock salts of the Wabamun Group and Leduc

Formation (Upper Devonian) in the Youngstown area, south-central
Alberta (Figs. 1 to 4), are preserved as irregularly shaped bodies having
maximum net thicknesses on the order of 45 m and 40 m, respectively.
Both of these rock salts have been leached extensively and were
deposited more widely than their present-day distribution might suggest
(Anderson, 1991~ Anderson and Brown, 1991a, 1991b, 1992~ Anderson,
Brown, and Hinds, 1988b; Anderson, White, and Hinds, 1989; Hopkins,
1987; Meijer Drees, 1986; Oliver and Cowper, 1983). Indeed, in a study
of the adjacent Stettler area (T30-45, RIO-25W4M), Anderson and Brown
(199Ib) conclude that about 40 m of net Wabamun rock salt probably
was deposited throughout the entire Youngstown area (T25-35, R5-
20W4M).
In our study of the Stettler area, we mapped the present-day
distribution and reconstructed the paleodistribution of the Wabamun rock
salt (Anderson and Brown, 199Ib). On the basis of this suite of maps of
RECONSTRUCTION OF ROCK SALTS, ALBERTA 177

Alberta
Age
Central Plains ·
Serles ~o~
~0
Era PERMIAN
Ma 286

1.6 o 320 PENNSYLVANIAN

t--
o ~

!
5.3

Hand 23.7 >

II:
2
o
« z

~
rJl
Hills w rJl
~
II: o ~
36.6 W
I-
57.8
~ 360
Paskapo~
66.4
.J!.a~
.Exs.!,1aw "\ Upper
Edmonton \ 1st Upper Wabamun~
Group ~ White Winterburn
Belly River! Speckled Cam rose
ea l"arK Shale
Q.
-=
~

~
g. 2nd
(!)
-g
e Colorado White Q) Leduc
~ Shale Speckled .0
'8 Duverna
-5 1-_ _ _vShale Cooking Lake
I~ 374 Late z 0
iii Colorado Viking!
j Shale Pelican
rJl
::>
...J
I
Waterways « 0
Late ow CD
~ .... Slave PI.
.... Ft.
Z N
0
J:Jsh.scale one..Vg;.5 0 w
o ~1.~1 Vermillion >
w ...J

? Joli Fou
Lower
~
W
Prairie 0 0:
II:
o o Middle
~§ Upper
o
N
Winnipegosis
o 'E
§ el------i rJl ~ Contact Rapid~
W
~~ Lower ~ ITI Cold Lake
r-----
387
~I
lErnestina

IZ
detrlta -'-/'
!. J
144 Early J 408 Early
~

I U/ 0 ~ L'otsberg
L SILURIAN
lLL «~ Granite Wash 438
z
«
Fernie
Middle II:
..,
::> ~ U
:>
208 Middle 0
a
505 a:
L~ 0
z
- J',-

.:ElM. ./
Finnegan
Deadwood ~ «
a:
CD
.J;;"arlie Middle ~
570
~
.....-:::::£ 5
Basal/sst

Figure 1. Stratigraphy of central Alberta (modified after AGAT Laboratories, 1988).

In Youngstown study area (Fig. 2), relatively thick Devonian rock salt is preserved
within several stratigraphic intervals: Cold Lake Formation, Prairie Formation, Leduc
Formation, and Wabamun Group. In study area, Upper Leduc and Wabamun rock salts
have been dissolved extensively; however, there is no conclusive evidence that
underlying Middle Devonian Cold Lake or Prairie rock salts have been leached.
178 ANDERSON AND BROWN

T35

T30

T33

T32

r31

T30

T29

T28

TV
A'
T26

T2S

R20 R19 RIB Rl1 R16 RIS RIO R13 R12 Rll RIO R9 AS R7 A6 RS
WOM W4M W4M W4M

Figure 2. Isopachous map (in meters) depicting present-day net thickness of Leduc salt
in Youngstown area. Net salt thicknesses are based on Leduc well control where
available. Elsewhere thicknesses are based on estimates of salt-related subsidence at
shallower levels. Areas of thickest salt (>30 m) are finely stippled. Only larger lakes are
shown.

net salt thickness, we concluded that dissolution of the Wabamun salts in

the Stettler area occurred episodically, although in some places (e.g. near
the northeastern margin) more-or-Iess continuously from the late
Paleozoic to the present. This is a result of several principal processes. In
support of these conclusions Anderson and Brown (1991b) demonstrate
that:

(A) On a local scale, the Wabamun interval ranges by up to 40 m (Fig.

3). These variations can be correlated directly to the thickness of the
residual salt, which supports the interpretation that about 40 m of
Wabamun rock salt was deposited in the Stettler study area and that
this salt was extensively leached.
RECONSTRUCTION OF ROCK SALTS, ALBERTA 179

(B) Seismic and well-log data illustrate clearly that post-salt strata drape
across residual Wabamun and Leduc salt. These observations
indicate that dissolution and related subsidence have occurred.

(C) Generally, Wabamun salt is not encountered along the Wabamun

subcrop edge (Fig. 3). This observation supports the interpretation
that subsidence originated along the exposed Wabamun outcrop
during the pre-Cretaceous hiatus.

(D) Basinward of the Wabamun sub crop edge, the Wabamun salts in the
Stettler area have been leached preferentially along NNE-trending
(and orthogonal) lineaments. This pattern is consistent with the
theses that dissolution was initiated in places by regional faulting
during the mid-Late Cretaceous and is self-sustaining. Leaching
along these shear zones is envisioned as a cyclic and continuous
process whereby the collapse of overlying strata enhances vertical
permeability, thereby providing a conduit for unsaturated water and
facilitating additional dissolution (Anderson and Brown, 1992~
Anderson, Brown, and Hinds, 1988a).

(E) In places, present-day drainage patterns are correlatable to areas

where one or both Upper Devonian salts are thin or absent
indicating that dissolution probably has occurred in the post-
Pleistocene, possibly as a result of glacial loading and unloading
(Anderson and Brown, 1992~ Anderson and Cederwall, 1993~
Anderson and Knapp, 1993).

We have followed up these previous studies of salt dissolution in

south-central Alberta and analyzed the leaching of the Wabamun Group
and Leduc Formation salts in the Youngstown area (T25-35, R5-20W4M).
Following the methodology of Anderson and Brown (1991b), we have
reconstructed the distribution of these Upper Devonian salts at various
times since their deposition (Figs. 2, 3, 5, 6, and 7). These
paleoreconstructed net-salt-thickness maps are consistent with previous
studies of salt dissolution in south-central Alberta and some of the large-
scale mechanisms of salt dissolution reported by Anderson and Cederwall
(1993) and Anderson and Knapp (1993) in their studies of other western
Canadian Devonian rock salts.
180 ANDERSON AND BROWN

ORIGINAL DIS1RIBUTION OF TIlE LEDUC AND WABAMUN

SALTS

Leduc Salts

The lithology of the Upper Devonian Leduc Formation varies

significantly within the Youngstown area of Alberta. Near the edge of the
East Ireton Shale Basin (Fig. 2), the Leduc is dolomitized fringing reef.
Shelfward (to the east), the Leduc becomes increasingly evaporitic and
consists predominantly of interlayered dolomites, anhydrites, and in places
a basal halite (rock salt) unit (Belyea, 1964).

T35

T34

T33

T32

131

T30
~
-'1(}: •
Youngstown
.; Plover T29
tole

T28

.'olJ8(:IIY Cr T27
tfl Reu:~'''vO'1
T26

T25
.00

R20 Rl9 RIB Rl7 R16 R15 Rl4 R13 Rl2 Rl1 RIO R9 RB R7 R6 R5
~M ~M ~M ~M

Figure 3. Isopachous map (in meters) depicting our interpretation of present-day

distribution of Wabamun salts. Net salt thicknesses are based on Wabamun well control
where available. Elsewhere, thicknesses are based on estimates of salt-related subsidence
at shallower levels. Areas of thickest salt (>30 m) are fmely stippled.

In order to estimate the original distribution of the Leduc salts (also

referred to as the Cairn salts) in the Youngstown study area (Fig. 2) and
following the methodology described by Anderson and Brown (1991b),
we have constructed a suite of four maps: (1) net Leduc salt based on log
RECONSTRUCTION OF ROCK SALTS, ALBERTA 181

control - sonic, density, and caliper only (Fig. 2 shows a modified

version of this map); (2) Leduc isopach (not shown); (3) present-day
Leduc structure (Fig. 8); and (4) restored Leduc structure (Fig. 9). This
map suite indicates that there are direct correlations between relative
structural relief at the top of the Leduc, thickness of the Leduc Formation,
and net thickness of Leduc salt dissolved in post-Leduc time. For
example, outside the interpreted Leduc salt basins (Fig. 5), the top of the
Leduc (Fig. 8) is consistent with the restored Leduc structure map (Fig.
9), indicating that little, if any, Leduc salt was deposited in these areas.
Within the confines of the salt basins, however, the present-day Leduc
structure values are lower than the restored Leduc structure values, except
at those well locations where thick residual salt is preserved (i.e. locations
of little, if any, dissolution). Our interpretation is that, within the confines
of the salt basins, the difference between the present-day and restored
Leduc structure at any well location is a reasonable estimate of the
thickness of Leduc salt that was dissolved at that well location in post-
Leduc time. On the basis of this premise, the original distribution of the
Leduc salt within the Youngstown study area (Fig. 5) was reconstructed.

-'" -'"'"
.....
....:. r.:.
MO
'"aO '"aO
'"~ '"rio ~
0
;::
~
0 It)
co> co>
6 <0 6
M
6 ..;.
0
'"~
W - 0

o
Restored Mannville
Mannville
200 Restored Wabamun
Wabamun
Restored Leduc

1
~ 600
400 - -.........~ Leduc
,---::-+- Original Leduc Salt
Preserved Salt
." Cooking Lake
'"5l
.Q
Beaverhill Lake
iii 800

1000

I I
10km
1200

Figure 4. Geologic section from Youngstown study area illustrating discontinuous nature
of Wabamun and Leduc salts. Both present-day and reconstructed profiles are displayed.
Wells incorporated into geologic section are highlighted in Figure 2.
182 ANDERSON AND BROWN

T35

T34

T33

T32

T31

T30

T29

-1',,0'
Drumheller T28
~LirHe'rsh
-.;:.:0 loke
T27

T26

T25

R20 RI9 Rl8 Rll Rl6 Rl5 RI4 Al3 RI2 All RIO A9 RB A7 R5
W4M W4M W4M W4M

Figure 5. Isopachous map (in meters) depicting original distribution of Leduc salt in
Youngstown study area. Net salt thicknesses are based on Leduc well control where
available. Elsewhere, thicknesses are based on estimates of salt-related subsidence at
shallower levels. Area of greatest salt thickness (>40 m) is finely stippled.

More specifically, at each well location where Leduc control was

available, the difference between the present-day Leduc structure (Fig. 8)
and the restored Leduc structure (Fig. 9) was calculated. At those well
locations where Leduc salt is no longer present, this difference is
interpreted to represent the original salt thickness. At those control points
where salt remains, the original salt thickness was estimated by summing
the thickness of the salt remnant and the calculated difference. At those
shallow well locations where Leduc control is not available, the
contouring of the present-day Leduc structure map was constrained by
interpretive contouring trends and patterns of structural relief along post-
Leduc horizons.
RECONSTRUCTION OF ROCK SALTS, ALBERTA 183

T35

~Fbr~'f
·~loe
,
Chain .",
lakes .1)
'MOowlOng
~lako

HancJh.1/,
Lo.le:e
Colemon
A . ..,' PlOver
. oke
clI,.IOgS\own

take

) SO"ye r•
ReservoI,.

us

R20 R19 R18 Rl7 Rl6 R15 R14 RI3 RlI RIO R8 R7 R6 R5
W4M W4M W4M W4M

Figure 6. Isopachous map (in meters) showing our interpretation of distribution of

Wabamun salts at end of Viking time. This map supports interpretations that: (i) about
40 m of Wabamun rock salt was deposited throughout Youngstown study area; (ii)
earliest phase of dissolution initiated along Wabamun subcrop during pre-Cretaceous
hiatus; and (iii) salt-dissolution front established along Wabamun subcrop migrated to
southwest thereafter at variable rates and in self-sustaining manner. Because significant
dissolution occurred only around Wabamun subcrop up to end of Viking time, this map
also represents that original salt distribution where Wabamun is preserved. Area of
greatest salt thickness (>40 m) is finely stippled.

Wabamun Salts

The Upper Devonian Wabamun Group in the Youngstown area is

subdivided into the Stettler and Big Valley Formations (Fig. 1). The
Stettler consists predominantly of interlayered dolomites, anhydrites, and
residual rock salt; the Big Valley is composed of green shales and
fossiliferous limestones (Andrichuk and Wonfor, 1953; Wonfor and
Andrichuk, 1953).
184 ANDERSON AND BROWN

The Wabamun rock salt (Stettler Formation) in the Stettler area

(T30-45, RIO-25W4M) has been mapped by Anderson, Brown, and Hinds
(1988b) and Anderson and Brown (199Ib), who conclude that about 40
m of rock salt was deposited throughout the Youngstown area, but that
much of this salt was post-depositionally leached. In order to test this
previous interpretation, we have constructed a suite of four maps: (I) net

T35

~o"':11
aj
.......... ~tolce

Chom .~
loke\ .0

~
~OWI,ng
",D . lokI!..'

T31
,.
~Hondh,'h
\)Lok(t T29
1\coremon
~to~e
fJ Oo~lond
lOke
T28

T27

T26

T25

R20 RI9 RI8 RIl RIG RIS RII RIO R9 RS Rl A6 R5

W4M W4M W4M W4M

Figure 7. Isopachous map (in meters) showing our interpretation of distribution of

Wabamun salt at end of Colorado time. Areas of thickest salt (>30 m) are fmely
stippled. Comparison of Figures 3, 6, and 7 supports thesis that extensive phase of salt
dissolution was initiated by regional faulting during Colorado (mid-Late Cretaceous)
time.

Wabamun salt based on log control - sonic, density, and caliper only
(Fig. 3 shows a modified version of this map); (2) Wabamun isopach; (3)
present-day Wabamun structure; and (4) restored Wabamun structure.
Outside the postulated Leduc salt basins (Fig. 5), there is a direct
correlation between the structure at the top of the Wabamun, the
thickness of the Wabamun interval, and the thickness of the residual
Wabamun salt. More specifically, at well locations outside the interpreted
RECONSTRUCTION OF ROCK SALTS, ALBERTA 185

Leduc salt basins (Fig. 5), the top of the Wabamun is consistently up to
40 m lower than the respective restored Wabamun structure. At control
well locations, the difference between these two structure contour maps
when added to the thickness of the residual Wabamun salt is consistently
about 40 m. At control locations within the Leduc salt basins, the
difference between the two Wabamun structure contour maps, plus the
thickness of any residual Wabamun salt, less our estimate of the thickness
of Leduc salt that was dissolved in post-Wabamun time, is consistently
about 40 m. These relationships support the thesis that about 40 m of

r31

T26

T25

A20 RI9 RIS AI7 RIS RIS RI4 R13 RI2 Rl1 RIO R9 RS R6 RS
~M _M ~M W4M

Figure 8. Structure map (in meters) of subsea depth to top of Leduc. Bold dots highlight
those sections for which there is deep well control, that is, to top of Leduc. In those
areas where deep well control is absent, contouring of present-day Leduc structure was
constrained by apparent local trends and structural patterns along shallower horizons.

Wabamun salt was deposited uniformly throughout the Youngstown study

area. At those shallow well locations where Wabamun control is not
available, the contouring of the present-day Wabamun structure map was
constrained by interpretive contouring trends and patterns of structural
relief along post-Wabamun horizons.
186 ANDERSON AND BROWN

PRESENT-DAY DISTRIBUTION OF TIlE LEDUC

AND WABAMUN SALTS

Leduc Salts

The present-day distribution of the Leduc salt (Fig. 2), was

estimated principally from three maps: (I) original Leduc salt thickness
(Fig. 5); (2) present-day Leduc structure (Fig. 8); and (3) restored Leduc
structure (Fig. 9). At each well control point, the present-day salt-

T32

r3'

T30

T29

T28

T27

T26

T25

R20 R,9 R'8 R'7 R'6 R,5 R'4 R,3 R,2 R" R'O R9 RB R7 R6 RS
W4M W4M W4M W4M

Figure 9. Restored Leduc structure map (in meters). Ideally, contours represent pattern
of structural relief that would be observed if all of original Leduc salt (dissolved in post-
Leduc time) were replaced.

thickness estimate (A L) is equal to the original (end Leduc time) salt

thickness (B L ) less the difference between the restored (CL ) and present-
day (D L) Leduc structure values. Following Anderson and Brown
(199Ib), we use the formula:
RECONSTRUCTION OF ROCK SALTS, ALBERTA 187

At those shallow well locations where Leduc control is not

available, the contouring of the present-day Leduc structure map was
constrained by local interpretive trends and structural relief along
shallower horizons. At all available deep well control points, our present-
day salt-thickness estimate (A L) and the actual thickness of the residual
salt (according to the well-log data) are effectively the same. The original
and present-day thicknesses of the Wabamun salts were estimated
concurrently with our analysis of the Leduc salts. This was necessary in
order to ensure that our Leduc and Wabamun net-salt paleodistribution
maps were compatible.

Wabamun Salts

The present-day distribution of the Wabamun salt (Fig. 3) was based

largely on the analyses of three maps: (1) original salt thickness; (2)
present-day Wabamun structure; and (3) restored Wabamun structure.
Although a map of the original salt thickness is not shown here, Figure
6 is almost identical to it because up to the end of Viking time only slow
dissolution occurred and it was localized around the Wabamun subcrop
edge. At each well control point our estimate of present-day net salt
thickness (Aw) is equal to the original salt thickness (40 m) plus our
estimate of the thickness of Leduc salt dissolved in post-Wabamun time
(BW) less the difference between the restored (CW) and present-day
(DW)Wabamun structure values. Following Anderson and Brown (1992b),
we use the formula:

AW =BW - (CW -DW)'

At all available deep well control points, our present-day salt-

thickness estimate (Aw) and the actual thickness of the Wabamun salt
remnant (according to the well-log data) were effectively the same. In
those areas where Wabamun control is sparse, the contouring of the
present-day Wabamun structure map was constrained by apparent local
trends and structural patterns along shallower horizons.
188 ANDERSON AND BROWN

PALEODISTRIBUTION OF THE LEDUC AND WABAMUN SALTS

Methodology

The paleo distributions of the Leduc and Wabamun salts were

estimated following the three steps outlined here:

Step 1. The subsea depths to the eight horizons listed here (Fig. 1) were
determined at about 3000 well locations within the study area; (increasing
depth and age, top to bottom):
(A) top Lea Park;
(B) top Colorado;
(C) top Second White Speckled Shale (Second Specks);
(D) top Viking;
(E) top Mannville;
(F) base Cretaceous (top Paleozoic unconformity);
(G) top Wabamun;
(H) top Leduc.

Step 2. Both present-day and restored structure maps were drafted for
each of these eight structural data sets. (Figs. 8 and 9 are the present-day
and restored maps for the Leduc.) Ideally, the restored map for a
particular horizon represents the pattern of structural relief that would
exist if we replaced all of the Leduc and Wabamun salts that were
dissolved after deposition of that particular horizon. Differences in
structural relief between corresponding present-day and restored structure
maps are therefore estimates of the thickness of all of the salt (both
Leduc and Wabamun) removed after the deposition of the relevant strata.

Step 3. The paleodistribution of the Leduc and Wabamun salts were

determined at the following times following the described methodology:
(A) end Leduc (Fig. 5);
(B) end Wabamun (Fig. 6);
(C) end Mannville;
(D) end Viking (Fig. 6);
(E) end Second Specks;
(F) end Colorado (Fig. 7);
(G) end Lea Park;
(H) present-day (Figs. 2 and 3).
RECONSTRUCTION OF ROCK SALTS, ALBERTA 189

The net salt thicknesses at these times were estimated on the basis
of: (1) the original salt-distribution maps (Figs. 5 and 6); (2) the present-
day structure maps (Fig. 8); and (3) the restored structure maps (Fig. 9).
For example, the total thickness of Leduc salt at the end of Viking time
(EL) was calculated to be equal to the present-day thickness of the Leduc
salt (FL) plus the difference between the restored (GL ) and present-day
(Hd Viking structure values less the thickness of Wabamun salt leached
in post-Viking time (IL ). We use the equation:

In an analogous manner, we concurrently estimated the thicknesses

of the Leduc and Wabamun salts at the other selected times during the
Cretaceous. At most of our control well locations, this process was
relatively straightforward (or at least constrained) for one of several
reasons: (1) Leduc salts were not deposited (Fig. 5) and hence only
Wabamun salts were leached; (2) thick (on the order of 45 m) Leduc salts
were preserved (Fig. 2) and probably only Wabamun salts were leached;
(3) the thicknesses of the Wabamun and Leduc salt remnants were known
from well control; and (4) both salts had been leached totally from the
section. The agreement between our present-day net-salt-thickness
estimates and the actual thicknesses of the salt remnants (as determined
from well control), and the compatibility of the suite of reconstructed salt-
distribution maps indicates that the paleoreconstructed salt-thickness
estimates probably are reliable to within 10m or less. Our estimates of
present-day net salt thickness are probably reliable to within 5 m.

Paleoreconstruction

Selected maps showing the past and present-day distribution of the

Leduc and Wabamun salts are depicted in Figures 2, 3, 5, 6, and 7. On
the basis of the full suite of such reconstructed net-salt-thickness maps we
have concluded:

(1) The Leduc salt in the study area was deposited in two, possibly
interconnected, fault-controlled, restricted basins on the shelfward
side of the developing Leduc fringing-reef complex (Fig. 5). The
maximum thickness of the Leduc salt in the study area was on the
order of 45 m. There are several reasons for concluding that the
190 ANDERSON AND BROWN

Leduc salt basins were fault-controlled: (i) prominent diffraction

patterns and vertical offsets along near-basement events (pre-
Leduc) are imaged on seismic data from the Leduc salt-basin
areas; (ii) the interpreted Leduc salt basins are oriented more-or-
less parallel to the edge of the fringing-reef complex (Fig. 5),
indicating that both features could have been influenced by
movement along preexisting orthogonal planes of weakness; (iii)
the western edges of the Leduc salt basins are interpreted to be
abrupt rather than gradational, a feature more characteristic of a
fault-controlled basin than of a gradual change in the depositional
environment; (iv) the dissolution of both the Wabamun and Leduc
salts during the mid-Late Cretaceous initiated along an orthogonal
set of lineaments; and (v) with respect to the western salt basin,
dissolution of the Leduc salts has been most extensive near the
basin margins, perhaps as a result of the reactivation of the faults
that originally controlled the development of the basin.

(2) About 40 m of Wabamun salt was deposited uniformly throughout

the Youngstown study area (Fig. 6). This conclusion is based on
our observation that there is a direct correlation between structure
at the top of the Wabamun, thickness of the Wabamun interval,
and net thickness of Wabamun salt dissolved in post-Wabamun
time. All three values vary, in a relative sense, by 40 m or less.

(3) The dissolution of the Leduc and Wabamun salts was initiated and
enhanced by some or all of four principal processes: (i) near-
surface exposure, as a result of erosion during the pre-Cretaceous
hiatus; (ii) regional faulting/fracturing during the mid-Late
Cretaceous; (iii) glacial loading and unloading (Fig. 10); (iv)
partial dissolution of the underlying salts.

(4) With respect to erosion during the pre-Cretaceous, the earliest

phases of Wabamun salt dissolution initiated along the Wabamun
outcrop edge during the pre-Cretaceous hiatus (Fig. 6). As
indicated on the suite of reconstructed maps, the dissolution front
that was established along the sub crop edge migrated towards the
southwest in a self-sustaining manner. An isolated remnant of
Wabamun rock salt is seen in the seismic section of Figure 11.
The observations that the Mississippian (top of the Paleozoic
RECONSTRUCTION OF ROCK SALTS, ALBERTA 191

sub crop ) and earliest Cretaceous (Mannville) horizons drape across

the residual salt support the interpretation that dissolution along
and immediately to the southwest of the Wabamun sub crop
occurred during Early Cretaceous time (Fig. 7).

(5) The orthogonal patterns displayed on the suite of Wabamun and

Leduc salt paleo distribution maps (Figs. 2, 3, 5, 6, and 7) support
the interpretation that dissolution fronts developed along a suite of
orthogonally oriented (NNE-WNW) regional fault planes during
the mid-Late Cretaceous. The apparent expansion of these areas
of dissolution through time indicates that the dissolution fronts
migrated laterally. It is noteworthy that the edges of the fringing
reef complex, as well as edges of the Leduc salt basins, are
consistent with the strikes of the hypothesized faults (Fig. 5),
indicating that possibly these lineaments were reactivated planes
of weakness. A seismic profile across an interpreted NNE-trending
fault is presented as Figure 11. The observation that most of the
dissolution occurred after the deposition of the Second Specks and
prior to the end of Colorado time supports the interpretation that
faulting was mid-Late Cretaceous in age (Fig. 1).

(6) Several lakes to the west of the postulated Leduc salt basins
(Gough, Sullivan, and Dowling; Fig. 3) are situated in areas where
the Wabamun salts are thin or absent; and all of the larger lakes
in the west Leduc salt basin area (Plover, Antelope, Contracosta,
Coleman, Oakland, and the Berry Reservoir; Fig. 2) are situated
in areas where the Leduc salts have been extensively leached.
These relationships indicate that significant salt leaching has
occurred in Recent time, possibly in response to glacial unloading
(Anderson and Brown, 1991b, 1992; Anderson and Knapp, 1993).

(7) Cold Lake and Prairie Evaporite rock salts of the Middle
Devonian Elk Point Group are preserved within the study area.
Although we have seen no evidence that these salts have been
leached here, it is possible that the dissolution of these underlying
rock salts could have triggered the leaching of the Wabamun or
Leduc salt. Within the confines of the postulated Leduc salt
basins, the dissolution of the Wabamun salt could have been
triggered by the leaching of the underlying Leduc salt. Such
192 ANDERSON AND BROWN

leaching could have occurred at any time after the deposition of

the Wabamun salt.

(8) The dissolution of the Leduc and Wabamun salts has occurred at
various times during the geologic past and, at least in places, has
been more-or-Iess continuous since deposition. Several trigger
mechanisms have been identified and it has been concluded that
leaching is self-sustaining. With respect to self-perpetuation, we
note that the established dissolution fronts do not advance at a
uniform rate. These observations indicate that a number of
secondary factors influence salt dissolution. Consideration should
be given to effects of features or processes such as regional
tectonism, periods of emergence, underlying reefs, the differential
compaction of pre-salt sediment, uneven loading and unloading,
gypsum to anhydrite conversion (and vice-versa), facies changes
within both the salt and encompassing strata, the local
hydrological and geochemical environment and changes therein,
and the effects of oil and gas activity.

(9) The timing of salt dissolution is of significance to the

explorationist for several reasons: (i) stratigraphic traps can form
where reservoir facies were either preferentially deposited or
preserved in salt-dissolution lows; (ii) reservoir facies can develop
in high-energy environments such as topographic highs that are
controlled by salt edges or remnants; (iii) structural traps can form
where reservoir facies are draped across salt remnants or collapse
features; and (iv) salt remnants can be misinterpreted as reefs,
faults, or other structural features (Anderson, 1991; Anderson and
Brown, 1987, 1991a; Anderson, Brown, and Hinds, 1988a, 1988b;
Anderson, White, and Hinds, 1989).

(10) The methodology developed by Anderson and Brown (1991b), and

employed in this study, is based on the premise that the
dissolution of subsurface salt causes the contemporaneous collapse
of the overlying strata. The thickness of leached salt and the
magnitude of subsidence are assumed to be equal. Although the
methodology is relatively robust, as evidenced by the suite of
compatible, consistent, restored structure maps, in places it might
not adequately account for processes such as: (i) nonuniform
 RECONSTRUCTION OF ROCK SALTS, ALBERTA 193

w Line 102 E
sholpOint number
241 221 3tI 182 161 142 121 102 81 81 41 23

0.0

-Colorado

- Second Specks
- Viking
- Mannville
c:;-
ell - Mississippian
.e 1.0 Wabamun
.g
ell
- near·top salt
ii~.~ near·base salt
IreIDn

- Elk Point marker

1 km
2.0
postulated fault/fracture plane
I

Figure 10. West-east seismic profile across NNE-trending salt-dissolution feature (I33,
R19W4M; Fig. 3) in Youngstown area of Alberta (after Anderson and Brown, 1992).
Dissolution is thought to have initiated along reactivated shear zone during mid-Late
Cretaceous. Subtle time-structural relief along shallowest events may be indicative that
leaching and subsidence has occurred in Recent time in response to glacial loading and
unloading.
194 ANDERSON AND BROWN

primary deposition~ (ii) erosion~ (iii) differential compaction~ (iv)

horizontal strain associated with collapse~ (v) differential
compaction of in fill (compensation) sediments~ (vi) salt flow~ (vii)
dissolution of underlying salts~ and (viii) faulting.

w E

0.0

O.S

Colorado mkr
· Second SpecI<s

• MaonviUe
· MISsissippian
• Wabamun

1.0
• BHL mkr

• Prairie

1.5 • seismic bsrn!

o kin

Figure 11. West-east seismic profile across isolated remnant of Wabamun rock salt
(T34-35, RIIW4M; Fig. 3) from Youngstown area of Alberta (after Anderson, Brown,
and Cederwall, 1994). Salt-bearing interval is imaged as moderately high-amplitude
peak-trough sequence. Relative to those areas where rock salt has been leached, salt-
bearing interval is characterized by an anomalously thick Wabamun interval, up to 8 ms
of velocity pullup along pre-salt events, and up to 25 ms of relative relief at
Mississippian level. These observations indicate that up to 40 m of residual salt is
preserved locally.

SUMMARY

About 40 m of Wabamun Group salt were deposited throughout

the Youngstown area~ the Leduc salts (up to 45 m thick), in contrast, are
thought to have been deposited only within two principal, fault-controlled,
restricted salt basins on the shelfward side of the developing Leduc
fringing-reef complex. Both salts have been dissolved extensively and
more-or-Iess continuously since deposition and by a variety of processes.
RECONSTRUCTION OF ROCK SALTS, ALBERTA 195

Our investigations indicate that dissolution of the Wabamun and

Leduc salts was initiated and enhanced by some or all of four principal
processes: (1) the near-surface exposure of these salts, as a result of the
erosion of the overlying Paleozoic sediment during the pre-Cretaceous
hiatus; (2) the partial dissolution of underlying salts; (3) regional
faul ting/fracturing during the late Cretaceous; and (4) glacial loading and
unloading.
Here, we have presented a suite of reconstructed salt-distribution
maps in order to substantiate the theses of continuous leaching and
diverse mechanisms. We have discussed our reconstruction methodology
and concluded that it is relatively robust. We are hopeful that this
technique can be applied to other areas where the effects of leaching are
not masked by processes such as erosion, faulting, compaction, or salt
flow.

ACKNOWLEDGMENTS

We wish to thank Renaissance Energy, Gulf Canada Resources,

Crestar Energy and Poco Petroleums, all of Calgary, for their generous
support of this and related studies, and to acknowledge the skilled
draftsmanship of Bill Matheson (Matheson Graphics Services, Calgary).

REFERENCES

AGAT Laboratories, 1988, Table of formations of Alberta: AGAT

Laboratories, Calgary.
Anderson, N.L., 1991, Dissolution of the Wabamun Group salt:
exploration implications, in Cavanaugh, T.D., ed., Integrated
exploration case histories, North America: The Geophysical
Society of Tulsa, Spec. Publ., p. 179-210.
Anderson, N.L., and Brown, R.I., 1987, The seismic signatures of some
western Canadian Devonian reefs: Jour. Can. Soc. Exploration
Geophysicists, v. 23, no. 1, p. 7-26.
Anderson, N.L., and Brown, R.I., 1991a, A seismic analysis of Black
Creek and Wabamun salt collapse features, western Canadian
sedimentary basin: Geophysics, v. 56, no. 5, p. 618-627.
196 ANDERSON AND BROWN

Anderson, N.L., and Brown, RI., 1991b, Reconstruction of the Wabamun

Group salts, southern Alberta, Canada, in Cavanaugh, T.D., ed.,
Integrated exploration case histories, North America: The
Geophysical Society of Tulsa, Spec. Publ., p. 145-178.
Anderson, N.L., and Brown, RI., 1992, Dissolution and deformation of
rock salt, Stettler area, southeastern Alberta: Can. Jour.
Exploration Geophysics, v. 28, no. 2, p. 128-136.
Anderson, N.L., and Cederwall, D.A., 1993, Westhazel General
Petroleums Pool: Case history of a salt-dissolution trap in west-
central Saskatchewan, Canada: Geophysics, v. 58, no. 6, p. 889-
897.
Anderson, N.L., and Knapp, RW., 1993, An overview of some large-
scale mechanisms of salt dissolution: Geophysics, v. 58, no. 9, p.
1375-1387.
Anderson, N.L., Brown, RI., and Hinds, RC., 1988a, A seismic
perspective on the Panny and Trout fields of north-central Alberta:
Can. Jour. Exploration Geophysics, v. 24, no. 2, p. 154-165.
Anderson, N.L., Brown, RI., and Hinds, RC., 1988b, Geophysical
aspects of Wabamun salt distribution in southern Alberta: Can.
Jour. Exploration Geophysics, v. 24, no. 2, p. 166-178.
Anderson, N.L., White, D.G., and Hinds, RC., 1989, Woodbend Group
reservoirs, in Anderson, N.L., Hills, L.V., and Cederwall, D.A.,
eds., Geophysical atlas of western Canadian hydrocarbon pools:
Can. Soc. Exploration Geophysicists and Can. Soc. Petroleum
Geologists, p. 101-132.
Anderson, N.L., Brown, R.I, and Cederwall, D.A., 1994, A seismic
analysis of erosion, salt dissolution and structural relief at the
Paleozoic subcrop in the Sullivan Lake area, south-central
Alberta, Canada: Can. Jour. Exploration Geophysics, v. 30, no. 1,
p. 51-59.
Andrichuk, 1M., and Wonfor, IS., 1953, Late Devonian geologic history
in Stettler area, Alberta, Canada: Alberta Soc. Petroleum
Geologists, News Bull. 1, no. 12, p. 3-5.
Belyea, H.R, 1964, Woodbend, Winterburn and Wabamun groups, in
McCrossan, R.G. and Glaister, RP., eds., Geological history of
western Canada: Alberta Soc. Petroleum Geologists, p. 66-88.
RECONSTRUCTION OF ROCK SALTS, ALBERTA 197

Hopkins, IC., 1987, Contemporaneous subsidence and fluvial channel

sedimentation: Upper Mannville C pool, Berry Field, Lower
Cretaceous of Alberta: Am. Assoc. Petroleum Geologists Bull., v.
71, no. 3, p. 334-345.
Meijer Drees, N.C., 1986, Evaporitic deposits of western Canada: Geol.
Survey Canada Paper 85-20, 118 p.
Oliver, T.A., and Cowper, N.W., 1983, Wabamun salt removal and shale
compaction effects, Rumsey area, Alberta: Can. Petroleum
Geology Bull., v. 31, no. 3, p. 161-168.
Wonfor, IS., and Andrichuk, IM., 1953, Upper Devonian in the Stettler
area, Alberta, Canada: Alberta Soc. Petroleum Geologists, News
Bull. 1, no. 9, p. 3-6.
 USE OF THE COMPUTER FOR THE STRUCTURAL
ANALYSIS OF THE ORDOVICIAN SEDIMENTARY
BASIN IN ESTONIA

Alla Shogenova
Estonian A cademy of Sciences, Tallinn, Estonia

ABSTRACf

Structural analysis of the Ordovician sedimentary basin in Estonia

and estimation of fracture zones with small amplitude was made with a
computer with an example from the Rakvere oil shale and phosphorite
deposit. Trend analysis was the main applied method. A database
containing the spatial coordinates of 500 boreholes and depth of 12 stages
and formations, interpreted from gamma-ray and resistivity logs, was
created using a Lotus spreadsheet. A BASIC program was written to read
data from the database, calculate the depths and thicknesses of the
appropriate formations, and calculate the trend coefficients. The first-
degree trend coefficients were computed for modeling regional structure
maps (Estonian Homocline) and regional isopachous maps of the
sedimentary units. Structure and isopachous maps of 12 stages and
formations in the Ordovician carbonate bedrock were made using kriging.
The result of the structural and thickness trend analysis showed four new
zones with rather small amplitudes established for the Rakvere deposit:
three fracture zones with subsidence of beds and thinning, and one fold
zone with the thinning over on anticlinal fold.

GeologiC Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 199
200 SHOGENOVA

lNTRODUCflON

The geological structure of Estonia is essentially the Estonian

Homocline - a regional structure on the south slope of the Baltic Shield,
situated in the northwestern part of the Russian Platform (Suveizdis,
1979). A large amount of data is available in Estonia (20,000 boreholes)
showing the structure of the relatively thin sediments (less than 1 km) in
the early Paleozoic basin. Structural analysis of the area was completed
traditionally by hand. At this time the main structural features of this
region were displayed, and zones of disturbances were outlined (Puura,
Vaher, and Tuuling, 1986, 1987). The Estonian Homocline slopes gently
to the south (2.9 - 4.4 mlkm). The slope is complicated by major and
minor linear disturbances and fracture zones, caused by fault movements
in the crystalline basement. Most disturbed areas are complicated by
karsting, and carbonate rocks are dolomitized. As is known from the
experience of exploitation of the Baltic oil shale and phosphorite basin
(Fig. 1), fracture zones - nonamplitude disturbances, or those with rather
small amplitudes -are the most dangerous for mining (Gazizov, 1971;
Heinsalu and Andra, 1975). They usually are not located during detailed
prospecting of deposits. For locating zones of disturbances, resistivity
method with dipole configuration of electrodes was used (Vaher, 1983,
1986), but it did not provide exact and detailed information about location
and structure of these zones.
Computerization of the structural analysis of the Ordovician
carbonate bedrock, overlying the commercial seam of phosphorites
(pakerort Stage), including the commercial seam of the oil shale
kukersites (lower part of Kukruse Stage), was completed by the example
of the Rakvere deposit. Trend analysis of the structural surfaces and
thicknesses of carbonate rocks revealed the zones of· disturbances,
characterized the structure of detected zones, compared their structure at
different stratigraphical levels, and allowed conclusions to be made about
geological processes which led to the formation of these zones.

lNPUTDATA

For the structural analysis of the Rakvere deposit, the data from
gamma-ray logs in 500 boreholes, resistivity logs in 290 boreholes
measured by Geological Survey of Estonia were used. The logging data
STRUCTURAL ANALYSIS OF A SEDIMENTARY BASIN 201

STIJDY AREA

I",
-
~
".~' ~J
,
... .."' ....
,
Phosphorite deposits -
50 Ian '-,,~
.
()

I I '"-

Figure l. Location of study area (Rakvere oil shale and phosphorite deposit).

available in the analog form were interpreted by a technique established

by Shogenova (1989a), which permits determination of the depths of 12
stages and formations (Table 1) in the Ordovician succession and with
accuracy high enough to the reveal the low-amplitude zones.
The analysis of fracturing of carbonate rocks by using these
logging data has been described in earlier publications (Shogenova,
1989b, 1991a, 1991b).
F or compiling the database, the spreadsheet from Lotus software,
Autocad program, and BASIC programs were used. It is easy to use the
spreadsheet for borehole data. The coordinates of 500 boreholes were
digitized using a digitizer and Autocad program. They were imported to
the database using the BASIC routine. The data from the wells and depth
of stages and formations in the boreholes were input from the keyboard
or digitized directly from the logs (Table 1).
After completion of the database it was essential to correct all
errors. The sources of errors in this situation may be the result of the
following: (1) incomplete depths of some formations in the database
202 SHOGENOVA

because of the absence of resistivity logs in some boreholes; (2) thinning

of upper formations in some boreholes or their total absence as a result
of erosion (for example in the north and northeast of the described
region); (3) mistakes in the elevations of the boreholes; (4) inaccurate
location of boreholes on the map.
If some of the stages or formations are absent in the upper part of
the section, those depth are marked by zero and in the middle or lower
parts of the section by the value of the previous depth. This permits the
BASIC routine, used for exporting data from database, to eliminate those
boreholes with an absence of the data.

Table 1. Stratigraphic scheme of study area according to

Hints and others (1993). * - stages and formations
included in database.

:::E CI) CI)

~ ~ r.il r.il
REGIONAL u FORMA- u
~ STAGES
til
>-
~
r.il r.il
STAGES nONS
~
CI) til
~
CI)
§
til
~
Quaternary deposits Q
Up- ASHGILL * Saunia O-:sn
per Nabala O:z.Jnb Paekna Opk
U
Rakvere 01rk Ragavere 0118'
~
u.:i Oandu 010n Hirmuse °2hr
0 Z *
-
CARADOC Keila O)d Keila O.jd
N
-< ~
JOhvi * OJh Johvi °ih

-
0 0 Vasavere °2VS
u.:i u Idavere 01id
0 Tatruse 021t
~
*
< :> LLANDEILO
Kukruse * 01 kk Viivikonna 02VV
01ub Korgekallas * O.jcr
~
~ 0 Uhaku
Vao * °2Va
Q ~ Lasnamagi 011s
:><
~
~
LLANVIRN Aseri * 01as Aseri 02as
~ 0 Napa 02np
< Kunda °1.2kn Loobu °ib
u.:i * Sillaoru Olsi
~
ARENIG Volhov * 0lv1 Toila Olli
~
0 Latorp * °llt Leetsu Olil
~ TREMADOC Pakerort Olpk Tiirisalu Ollr
STRUCTURAL ANALYSIS OF A SEDIMENTARY BASIN 203

In this manner every borehole is described in the database by the

following parameters: N - number of borehole, X, Y - spatial coordinates
of boreholes, elev. - elevation of the borehole, and depth of 12 stages or
formations of the Ordovician carbonate section (Table 1), and Quaternary
sediments.
To export data from the database, it was translated from the Lotus
format (.WKI file) into the Data Interchange Format (.DIF file).

TECHNIQUE OF 11IE SPATIAL ANALYSIS

The problem of the spatial data analysis has been described in

publications by Davis (1986), Nikitin (1986), Phillip and Watson (1987),
Merriam and Jewett (1988), Bohnam-Carter (1989), and Harff, Lange,
Olea (1992). The possibilities of using the computer mapping procedures
were tested with different gridding algorithms applied in commercial
packages and using BASIC programs. The best results and those closest
to the hand-contoured interpolation were obtained using a kriging
algorithm with selection of the search radius corresponding to the object
under study. Therefore, the kriging algorithm was used for mapping
different parameters.
The main mathematical method which was used for the spatial and
isopachous analysis was trend analysis as described by Davis (1986).
The regional structure of the studied area is the Estonian Homocline and
the structural and thickness data were modeled by the first-degree
polynomial trend surfaces.
For the spatial analysis in the sedimentary basin, the following
techniques were applied: (I) modeling of the structure maps, their three-
dimensional views using the kriging algorithm; (2) computing of the first-
degree polynomial trend coefficients with gridding on the map of the
same density as structure maps and plotting of the first-degree trend
surfaces; (3) plotting of the residual maps constructed by subtracting of
first-degree trend surfaces from structure maps; (4) plotting of the
isopachous maps for every formation and for sums of formations using
a kriging algorithm; (5) computing of the first-degree trend coefficients
with gridding of the map of the same density as the isopachous maps and
plotting of the first-degree trend thicknesses; and (6) plotting of the
residual isopachous maps constructed by subtracting first-degree trend
thicknesses from the isopachous maps.
204 SHOGENOVA

To export data from the database and calculate the trend coef-
ficients, a BASIC program consisting of subroutines and working in an
interactive mode with the following menu was used:
(1) Read data from .DIF file
(2) Calculate Z and print X,Y,Z on display
(3) Save X,Y,Z in .DAT file
(4) Read X,Y,Z from any .DAT file
(5) Calculate trend coefficients and residuals
(6) Calculate of statistical characteristics
(7) Exit.
Subroutine number 2 of this menu has three subroutines:
(1) Calculation of depth
(2) Calculation of thickness
(3) Calculation of sum of depths.
Numbers 5 and 6 of this menu are subroutines as described by Davis
(1986).
After applying this program all the files and trend coefficients are
computed for the gridding and plotting of maps and three-dimensional
views by the commercial package according to the described sequence of
procedures.

SPATIAL ANALYSIS OF mE ORDOVICIAN CARBONATE

ROCKS

Spatial Structural Analysis

Structure of the 12 stages and formations (Table 1) in the

Ordovician carbonate basin was studied by the structure contour, trend,
and residual maps and three-dimensional views obtained using the
described technique (Figs. 2 and 3). The residual maps, constructed by
subtracting first-degree trend surfaces from the structure maps, show
zones with subsidence (Figs. 2C, 3C). From the maps of the first-degree
trend surfaces, the regional strike and dip of formations were defined. The
regional NE-strike azimuth of formations changes from 83° (OIIt) (Fig.
2B) to 86.5° (021g)(Fig. 3B). The regional SE dip of formations is
changing from 2.96 mlkm (021g) (Fig. 3B) to 3.6 mlkm (02kk).
The negative anomalies on the residual structure maps define
zones with subsidence or synclinal folds of structures. The positive
STRUCTURAL ANALYSIS OF A SEDIMENTARY BASIN 205

8
2
4
U -2
-' -6
-8 -)0
-14
-18

-24 -22
-28 -26
A B

8
4
0
-4
.. -8
- 12
-16
-2tJ

0 2 4 6 II km
C

Figure 2. A, Structure contour map of Rakvere deposit, drawn on bottom of Latorp

Stage (lower reference surface of section). Borehole control shown by white points.
Contours in meters from sealevel; B, first-degree trend surface of Latorp Stage; C,
residual map, constructed by subtracting first-degree trend surface from structure
contour map of Latorp Stage (A). Contour interval is 2 m. (D) Three-dimensional view
drawn on bottom of Latorp Stage.

anomalies on these maps reflect anticlinal folds. The well-known Aseri

zone of disturbances (west part of the study area) is observed clearly by
the more dense location of contours on structure maps and on residual
maps as well. But zones with small amplitudes are determined on the
structure maps with difficulty - usually by structural noses, seldom by
isometric anomalies (Figs. 2A, 2C, 3A, 3C). On the residual maps,
constructed by subtracting the first-degree trend surface from structure
maps, zones with subsidence up to 5 meters, determined by contours
with negative values, are revealed (Figs. 2C, 3C).
Zones with small amplitudes revealed on the residual maps change
with depth and are not in similar locations in different zones; some
decrease upwards and others are emphasized.
206 SHOGENOVA

-t

o
-4

~ -8

-12
-16
-20
c o 2 4 6 II kill
D

Figure 3. A, Structure contour map of Rakvere deposit, drawn on bottom of Rilgavere

Formation (upper reference surface of section). Contours in meters above sealevel.
Original borehole control shown in Figure 2A. On blanked area Rllgavere Formation is
eroded. B, First-degree trend surface of Rllgavere Formation; C, residual map,
constructed by subtracting first-degree trend surface from structure contour map of
Rilgevere Formation (A). Contour interval is 2 m. D, Three-dimensional view drawn on
bottom of Rilgavere Formation.

Spatial Thickness Analysis

Using the described techniques, isopachous maps of stages and

formations of the Ordovician carbonate basin (Figs. 4A, 5A), of different
parts (Fig. 6C), and the widespread ones (Fig. 6A) were compiled, maps
of the first-degree trend thicknesses (4B, 5B) and residual thicknesses
were produced (Figs. 4C, 5C, 6B, 6D). As it observed on the maps (Figs.
4B, 5B), the regional trends of thickness and the amplitudes of these
features are distinguished for different formations. The data conform
generally to the described directions of thickness changes and their
amplitudes (Puura, 1987). The change of these parameters is connected
with different paleogeographical conditions· of sedimentation.
It should be noted that thicknesses of the lower stages OIIt (Fig.
4A-4C), Olvl, 01_2kn, and 02as (Table 1) in the study area have
STRUCTURAL ANALYSIS OF A SEDIMENTARY BASIN 207

undergone relatively slight changes and are distinguished weakly from the
regional background. The amplitude of the alteration of the background
thickness is there not more than 1 meter. Directions of thickness
alterations coincide only in the Latorp and Volhov Stages and are
distinguished in all other formations. The amplitude of the regional
changes of thickness rise upwards and reach in 02va, 02kr, 02kk to some
meters. As they compensate each other, the amplitude of regional
alteration of their total thickness reaches to three meters. Also the high
negative residuals of thicknesses (thickness loss) are observed in these
formations, equal to 1.5 meters in 02va, up to 3.0 meters in 02kr and
02kk. The rocks of the upper formation 02id and O~h are characterized
by the common features of the regional background - nearly coinciding
directions of thickness increase from southwest to northeast and also a
small change of regional thickness, being about 0.8 meters in the area
under study in 02id and 0.7 m in O~h. In some places in these

1.8

1.6

1.6 1.4

1.2 1.2

0.8 1.0
A II

0.8

OA

0.0

-0.4

-0.8

C 0 2 3 41un

Figure 4. A, Isopachous map of rocks of Latorp Stage. Contour interval is 0.4 m.

Original borehole control shown in Figure 2A. B, First-degree trend thickness of rocks
of Latorp Stage. Contour interval is 0.2 m. C, Residual map constructed by subtracting
first-degree trend surface from isopachous map of rocks of Latorp Stage. Contour
interval is 0.4 m.
208 SHOGENOVA

13 13.S
13.0
11 12.S
9 J 12.0
, 7 11 .S
11.0
5
10.5
3 10.0
9.5
B

Figure S. A, Isopachous map of rocks of Keila Stage (in north and northeast, in valley
of Kunda River, rocks are partly eroded). Contour interval is 1 m. Original borehole
control shown in Figure 2A. B, First-degree trend thickness of rocks of Keila Stage.
Contour interval is O.S m. C, Residual map constructed by subtracting first-degree trend
surface from isopachous map of rocks of Keila Stage. Contour interval is 1 m.

formations, decreasing thickness of 1.5 meters in comparison with the

background thickness is observed. The same value of thickness loss is
noted in 02k1, increasing up to 6 meters in the north and northeast (in the
valley of the Kunda River) because of erosion processes (Figs. SA-SC).
The rocks of the upper formations are eroded completely in the
northeastern part of the study area (Fig. 3A). The erosion of the rocks of
the Keila Formation has affected the calculation of its first-degree trend
thickness (Fig. SB). As a result of this erosion in the southwestern comer
of the residual map thickness of 02kl, we see negative values (Fig. 5C).
The isopachous map of the 01lt-02hr (Fig. 6A) shows difference
between two structural maps, drawn on bottom of the Latorp Stage,
covering the commercial seam of the phosphorites (Fig. 2A), and bottom
of the Ragavere Formation, the first reference surface in the upper part
of the section (Fig. 3A). On the residual isopachous map, constructed by
subtracting the first-degree trend thickness from the isopachous map of
0llt-02hr, zones of thinness are revealed (Figs. 6B ,6C). Comparing them
STRUCTURAL ANALYSIS OF A SEDIMENTARY BASIN 209

with the Aseri dislocation, which is indicated on this map by thinning of

up to 8.5% of the total thickness of the carbonate beds, we can conclude
that the other anomalies of decreasing thickness (thickness losses up to
4.4%) reflect also zones of disturbances. Comparison of the isopachous
map of the beds 01lt-02hr (Fig. 6A) with the total isopachous map of the
O2vat0 2kM-0 2kk (Fig. 6C) has shown that the main part of the thinness
in the basin is the result of karst which takes place in these formations,
including the commercial seam of the oil shale kukersites (02kk).

1.5

0.5
,
-0.5

-1.5

-2.5

...3.5

-1

-2

...3

·4
0 2 .. 6 tlkm D

Figure 6. A, Isopachous map of beds O)lt-0:lzr. Original well control shown in Figure
2A. Contour interval is 0.5 m. On blanked area rocks of some upper formations are
eroded. B, Residual map constructed by subtracting first-degree trend surface from
isopachous map of beds O)lt-O:lzr.Contour interval is 0.5 m. C, Isopachous map of beds
of 0zvd-Ozkk (Ozva+O:/cr+Ozkk). Original borehole control shown in Figure 2A.
Contour interval is 0.5 m. D, Residual map constructed by subtracting first-degree trend
surface from isopachous map of beds of 0zvd-Ozkk. Contour interval is 0.5 m.

Results of the Structural Analysis

After collating data onto one map, the axes of the subsidence, axes
of thinness, axes of the low-resistivity anomaly zone (Fig. 7), and the
210 SHOGENOVA

analysis of the structure and isopachous maps of all formations, we can

observe the following:
(1) In the study area, except for the known Aseri dislocation, four
new zones with small amplitudes were established: three fracture
zones with the subsidence of the beds 2-5 meters and thickness
loss up to 4.4% (1-3) and one folding zone (4) with thickness loss
of up to 7.3% on its anticlinal fold.
(2) These zones partly coincide with the axes of the low-resistivity
anomaly zones.
(3) Thickness losses in the zones of disturbances, which are
reflected by the negative anomalies on the residual isopachous
maps, are caused by karsting intensified by movement along faults
in the crystalline basement.
(4) The greatest amplitude of the subsidence is defined by the
bottom of Ollt - the lower reference surface, which covers the
commercial seam of the phosphorites.
(5) The karst cavities of the greatest size may occur in the rock of
021g-02_3nb (up to 4,0 m), in 02kk, 02kr (up to 3.0 m). The karst
cavities of up to 1.5 m may occur in 01_2kn, 02Va, 02id, O~h,
02kl. From these formations possibility of occurrence of karst
cavities is the greatest in 02Va. In the rocks of Olvl and Ollt the
location of karst cavities is the least possible.
(6) The investigated Ordovician carbonate beds may be subdivided
by the rigidity of the rock mass into four intervals, whereas the
most difficult mining conditions were defined in the oil-shale
containing interval 02kk-02kr.

CONCLUSIONS

Computerization of a structural analysis of log data by applying

a kriging algorithm for the gridding procedure and trend analysis for
subtracting the regional background have permitted:
(1) completion of a comparative analysis of all structural surfaces
and of stages and formations in the Ordovician carbonate basin;
(2) characterization of regional trends of structural and thickness
alterations;
(3) zones of disturbances to be revealed especially those with
small amplitudes and erosion zones;
STRUCTURAL ANALYSIS OF A SEDIMENTARY BASIN 211

(4) description in detail of structures of these zones and

determining places of karst location;

o
o

o
o
o

o
o

o 1 2 3 " km
GJl [ZJ2 [2J3 ~4 ~5 [Z]6 ~71ZJs ITJ9
Figure 7. Scheme of axes of located zones of disturbances and low-resistivity anomaly
zones.
Explanat~on: 1 - boreholes; 2 - axes oflow-resistivity anomaly zones by Vaher (1983);
3 - axes of homocline fold of Aseri zone of disturbances by Vaher (1983); 4 - uncertain
borders of Kantku1a zone of disturbances (puura, Vaher, and Tuu1ing, 1987); 5 -
erosional cut of buried valley of Kunda River; axes of zones, revealed by: 6 -thickness
loss of rocks of 02hr-Ollt, 7 -subsidence of bottom of Rllgavere Formation, 8 -
subsidence of bottom of Latorp Stage; 9 - numbers of revealed zones of disturbances
with small amplitudes.

(5) to estimation of the rigidity of the rock mass, on the basis

of the established characteristics of the formations (Shogenova,
1991a,1992); and
(6) allowed forecasting the possibility of the karsting in differ-
ent carbonate formations.

The results and example of using the computer for the structural analysis
in a sedimentary basin in Estonia permits recommendation of the
application of such a technique for all of Estonia and also for other
regions with similar geological conditions.
212 SHOGENOVA

ACKNOWLEDGMENTS

The author would like to thank the colleagues from the Geological
Survey of Estonia for making available geological and geophysical data,
my supervisor Prof. Vaino Puura, my colleagues from Institute of
Geology, Estonian Academy of Sciences who helped me in preparation
of this paper and editors Andrea Forster and Daniel F. Merriam for
helpful suggestions and remarks.

REFERENCES

Davis, I.C., 1986, Statistics and data analysis in geology: John Wiley
& Sons Inc., New York, 646 p.
Bonham-Carter, G.F., 1989, Comparison of image analysis and geo-
graphic information systems for integrating geoscientific maps, in
Agterberg, F.P., and Bonham-Carter, G.F., eds., Statistical analysis
in the earth sciences: Geol. Survey Canada Paper 89-9, p. 141-
155.
Merriam, D.F., and Jewett, D.G., 1988, Methods of thematic map
comparison, in Current trends in geomathematics: Plenum Press,
New York, p. 9-18.
Gazizov, M.S., 1971, Karst i ego vliyanie na gomye raboty (Karst and its
influence on the mining works): Nauka, Moscow, 204 p.
Harff, I., Lange, D., and Olea, R., 1992, Geostatistics for computerized
geological mapping: Geol. Jb., A 122, p. 335-345.
Heinsalu, u., and Andra, H., 1975, Treshtshinovatost' v rayone slancevyh
shaht Estonii i geofizicheskie metody eyo issledovaniya. (Jointing
in the Estonian oil shale basin and geophysical research methods
for its study): Valgus, Tallin (In Russian with English summary),
116 p.
Hints L., Meidla T., Gailite L.-J., and Sarv L., 1993, Catalogue of
Ordovician stratigraphical units and stratotypes of Estonia and
Latvia: Estonian Acad. Sciences, Tallinn, 62 p.
Nikitin, A. A. , 1986, Teoreticheskie osnovy obrabotki geofizicheskoy
informacii. (Theoretical basis of the treatment of geophysical
information): Nedra, Moscow, 342 p.
STRUCTURAL ANALYSIS OF A SEDIMENTARY BASIN 213

Phillip, G.M., and Watson, D.F., 1987, Neighborhood discontinuites in

bivariate interpolation of the scattered observation: Math. Geology,
v.19, no. 1, p. 69-74.
Puura, V., ed., 1987, Geology and mineral resources of the Rakvere
phosphorite-bearing area: Tallinn Valgus Publisher (In Russian
with English summary and explanation to the Figures), 211 p.
Puura, V., Vaher R., and Tuuling, I., 1986, Tectonics, in Puura, V., ed.,
Geology of the kukersite-bearing beds of the Baltic oil shale
basin: Valgus Publishers, Tallinn (In Russian with English
Summary), p. 55-62.
Puura, V., Vaher R., and Tuuling, I, 1987, Tektonika. (Tectonics), Puura,
V., ed., Geology and mineral resources of the Rakvere
phosphorite-bearing area: Valgus Publisher, Tallinn (In Russian
with English summary and explanation to Figures), p. 90-104.
Shogenova, A., 1989a, Raschlenenie ordovikckoy karbonatnoy tolshtshi
na Rakvereskom mestorozhdenii fosforitov po dannym
skvazhinnoy geofiziki. (Detailed subdivision of the Ordovician
carbonate beds in th~ Rakvere phosphorite deposit by geophysical
logging): Proc. Academy Sciences Estonian SSR, Geology, v.38,
no.l (In Russian with English summary), p. 10-13.
Shogenova, A., 1989b, Ocenka narushennosti ordovikskoy karbonatnoy
tolshtshi na Rakvereskom mestorozhdenii fosforitov po dannym
skvazhinnoy geofiziki. (Fracturing estimation of the Ordovician
carbonate beds on the Rakvere phosphorite deposit by geophysical
logging): Proc. Academy Sciences Estonian SSR, v. 38, no. 3 (In
Russian with English summary). p. 118-123.
Shogenova, A., 1991a, Ocenka ustoychivosti gornogo massiva
Rakvereskogo mestorozhdeniy a fosforitov po dannym skvazhinnoy
geofiziki. (Rigidity estimation of the rock mass on the Rakvere
phosphorite deposit by geophysical logging.) Proc. Estonian
Academy Sciences, v. 40, no. 3 (In Russian with English
summary), p. 104-111.
Shogenova, A., 1991b, Primenenie EVM dlya detal'nogo strukturnogo
analiza Rakvereskogo slance-fosforitovogo mestorozhdeniy a po
dannym skvazhinnoy geofiziki. (Use of the computer for the de-
tailed structural analysis of the Rakvere oil-shale and phosphorite
deposit by geophysical logging): Primenenie EVM pri reshenii
zadach geologii poleznyh iskopaemyh Estonii: AN Estonii, Tallinn
(In Russian), p. 49-65.
214 SHOGENOVA

Shogenova, A., 1992, Struktura Rakvereskogo mestorozhdeniy a fosforitov

po geologo-geofizicheskim dannym (s ispol'zovaniem
modelirovaniya na EVM). (Structure of the Rakvere phosphorite
deposit by geological and geophysical data by computer analysis),
Abstract of candidate thesis: Sankt-Petersburg Mining Institute,
Sankt-Petersburg, (In Russian), 22 p.
Suveizdis, P.I, ed., 1979, Baltic tectonics: Mokslas, Vilnius (In Russian),
92 p.
Vaher, RM., 1983, Tektonika fosforito-slancevogo basseyna severo-
vostochnoy Estonii. (Tectonic of the phosphorite and oil shale
basin of North-Eastern Estonia): Abstract of candidate thesis,
Institute of Geochemistry and Geophysics, Belorussian Academy
Sciences, Minsk (In Russian), 22 p.
Vaher, RM., 1986, Primenenie metoda soprotivleniy dlya vyyavleniya
zon treshtshinovatosti v Severo-Vostochnoy Estonii. (Application
of the resistivity method for revealing fracture zones in north-
eastern Estonia): Proc. Academy Science, Estonian SSR, Geology,
v. 35, no. 4 (In Russian with English summary), p. 146-155.
 PAIRWISE COMPARISON OF SPATIAL MAP DATA

Daniel F. Merriam
University of Kansas, Lawrence, Kansas, USA

Ute C. Herzfeld
Universitiit Trier, Trier, Gemzany

Andrea Forster
GeoForschungsZentrnm Potsdam, Potsdam, Gemzany

ABSTRACf

Map comparison/integration of spatial data has become important

recently with the plethora of data being generated and accumulated. In
many instances it is of interest to compare/integrate these data. The
simplest comparative procedure is to overlay two maps and visually
compare the areas of similarity and differences. This cursory examination
is quick, but important aspects of the comparison may be overlooked.
The simplest integration procedure is to overlay several maps and note
the areas of correspondence on a resultant map. This type of
comparison/integration is visual and subjective; quantatively, the degree
of similarity may be expressed either (1) as a coefficient of overall
similarity or (2) as a resultant map. The similarity coefficient gives an
indication as to the goodness of correspondence and the resultant map
shows where. Several techniques of pairwise map comparison have been
utilized recently including bivariate and multivariate statistics, probability,
AIlES (Boolean logic), algebraic, and fuzzy set theory. A geothermal

GeologiC Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 215
216 MERRIAM, HERZFELD, AND FORSTER

data set from southeastern Kansas is used to demonstrate pairwise map

comparison with other spatial geological and geophysical data and to
interpret local and regional conditions.

INTRODUcnON

The problem of quantifying map comparisonlintegration has been

of interest to geologists for many years. The easiest way to compare two
maps, of course, is to overlie them on a light table and note the areas that
are similar and dissimilar based on some criterion (or set of criteria). To
compare two maps where measured units are the same, one map can be
subtracted from the other to produce a resultant map (e.g. an isopachous
map), which then can be interpreted from the pattern that shows the
similarity between the two maps. Where the two map units are different,
a crossplot of the variables on each map can be used to compute a
correlation coefficient. These first ways to accomplish the task were
straightforward and simple.
With the advent of remote sensors and automatic data acquisition,
these simplistic comparisons were not feasible anymore. Those working
in remote sensing began to explore ways in which to analyze literally
billions of items of data routinely collected in space from airborne and
space platforms. Geophysicists also had to devise new ways of
integrating diverse data obtained from the thousands of miles of land and
sea profiles collected during exploration for mineral resources.
Geographers were faced with organizing all types of spatial data and
comparing and integrating them. Geologists, likewise, have been
struggling with masses of data generated by automated equipment
designed to provide adequate data for today's sophisticated three-
dimensional studies.
There are several reasons to determine the relation of the map
variables, not only to each other, but between maps. It may be desirable
to know the (1) predictive value of one map to another, (2) evaluate the
map similarity for classification, or (3) use the information for geological
interpretation (Merriam and Jewett, 1988). For whatever reason to
comparelintegrate, the technique should be simple and fast. We suggest
one such approach in this paper where we do a pairwise comparison of
geological, geophysical, geothermal, and other spatial data to determine
the predictive value of one map to the others.
PAIRWISE COMPARISON OF SPATIAL MAP DATA 217

PREVIOUS WORK

Many papers have appeared in the literature in the past several

years related to map comparison/integration of geological data. A good
summary on the subject is given by Merriam and Jewett (1988)~ other
sources include Unwin (1981), Gold (1980), and Davis (1986).
An early work was by Mirchink and Bukharsev (1959) who took
corresponding data points from maps and plotted them on a scatter
diagram to determine the degree of accordance. Later, Ribeiro and
Merriam (1979) used scatter diagrams and regression curves to determine
the relationship of one map variable to another.
As early as 1966, Merriam and Sneath proposed using the
coefficients from low-degree polynomial trend surfaces of structural
surfaces as descriptors of those surfaces. The values were used as input
into a clustering routine and plotted as dendrograms showing the overall
similarity of one map to another. Merriam and Lippert (1966) noted the
relation of residue values from trend analysis in determining map
similarity. Thrivikramaji and Merriam (1976) used the trend-surface
coefficients from thickness maps for input into a clustering routine in a
study of the structural development of Kansas during the Pennsylvanian.
Miller (1964) proposed comparing maps using the values of the computed
trend-surface matrices, and Rao (1971) used the least-square equation.
Robinson and Merriam (1972) used values at grid nodes on
spatially filtered maps to determine the relation of one map to another.
Robinson, Merriam, and Burroughs (1979) reported on a technique to
relate the location of oil production to geologic structure. Merriam and
Robinson (1981) utilized a point-by-point comparison to obtain a
coefficient of resemblance in comparing spatially filtered maps to each
other. Eschner, Robinson, and Merriam (1979) and Robinson and
Merriam (1984) constructed 'resultant' maps to compare spatially filtered
maps to each other.
In 1988 Merriam and Sondergard published on a technique they
devised to determine the relationship between one map and another by
computing a spatial Reliability Index. Brower and Merriam (1990) used
several multivariate algorithms, including principle component analysis
(PCA), principal coordinates, Q-mode vector analysis, correspondence
analysis, and clustering to determine the relationship between gridded
structural maps. They followed up this initial work in 1992 (Brower and
Merriam, 1992) with a method utilizing correlation coefficients to
218 MERRIAM, HERZFELD, AND FORSTER

compare adjacent points on thematic maps. Merriam (1992) published an

example of pairwise comparison utilizing similarity coefficients
represented by dendrograms of standardized structural, stratigraphic,
topographic, and geophysical data. Forster, Merriam, and Brower (1993)
have compared geological and geothermal map data using cluster
analysis.
A computer program, MAPCOMP, to perform a weighted
comparison of two or more maps using a matrix algorithm was published
in 1988 by Herzfeld and Sondergard. This algebraic algorithm, which
produces a resultant map showing areas of similarity and dissimilarity,
has been used successfully in several subsequent studies involving the use
of geological, geophysical, and geothermal data (Herzfeld and Merriam,
1991; Merriam, Fuhr, and Herzfeld, 1993 ; Forster and Merriam, 1993;
Merriam, Herzfeld, and Forster, 1993). Herzfeld and Merriam (1995)
followed up this approach by suggesting optimizing techniques to
automate the process for the selecting the 'best fit' for the algebraic
algorithm.

TIlE DATA

The data for this study were compiled for an area in southeastern
Kansas comprising Chautauqua County between T32S and T35S and R8E
and R13E; an area of approximately 620 sq mi (1600 sq km). The
subsurface data were collected from the files at the Kansas Geological
Survey and consisted of petrophysical logs of all types, sample logs,
drillers logs, and scout tops. A structural map on top of the Mississippian
was constructed as well as bottom-hole temperature (BHT) maps for the
Lower Pennsylvanian, Mississippian, and Cambro-Ordovician Arbuckle
Group.
Geophysical data, aeromagnetic (Yarger and others, 1981) and
gravimetric (Lamb and Yarger, 1985), were digitized on a 3-mi grid from
published maps. Precambrian data were used from material by Cole
(1962). Background information for the study was obtained from
Merriam (1963) and other Kansas Geological Survey publications.
PAIRWISE COMPARISON OF SPATIAL MAP DATA 219

11IEBACKGROUND

A variety of techniques have been utilized to compare spatial data

- each practical for different conditions. These techniques are based
either on a point-by-point comparison to obtain a coefficient of
resemblance or some combination of spatial data to produce a resultant
map. The approaches that have been used in map comparison, in addition
to bivariate and multivariate statistics, include probability (Brower and
Merriam, 1990), AIlES (Boolean logic) (Bonham-Carter and Agterberg,
1990; Agterberg, Bonham-Carter, and Wright, 1991; Maslyn, 1991),
algebraic (Herzfeld and Sondergard, 1988; Herzfeld and Merriam, 1991),
and fuzzy set theory.
Similarity coefficients and resultant maps can be computed from
the raw data, gridded data, or a set of descriptors derived from the
original data (Fig. 1). Where the data are measured in different units,
they first must be pretreated to standardize them. The similarity measure
usually is a correlation or distance coefficient and has the general form
(Unwin, 1981)

area over which phenomena are located together

ca =---------------------------------
total area covered by the two phenomena

The coefficients computed between each pair of maps then can be

tabulated in a row,column (r,c) matrix and the relationship between the
variables or samples displayed in a dendrogram for visual ease in
interpretation.
Resultant maps can be constructed by noting the relation on a
point-by-point correspondence between maps or using an interpolated
grid. The values then are contoured for visual interpretation.

INTERPRETA nON AND RESULTS

The geological, geophysical, and geothermal maps for our area in

southeastern Kansas were analyzed by first computing correlation
coefficients and constructing dendrograms and then by using the algebraic
map-comparison program developed by Herzfeld and Sondergard (1988).
One data subset consisted of seven maps, constructed just for Chautauqua
County, digitized on a 3-mile grid. This data set consisted of two
220 MERRIAM, HERZFELD, AND FORSTER

A c B
[§-~ -~

. ····l
[B •
.......
. .......
.. . DESCRIPTORS
...

Figure 1. Flowdiagram of spatial analysis from data input to graphic output.

structural, two geophysical, and three geothermal maps. Regression

analysis was performed pairwise on the data sets collected from the seven
maps and the results summarized in a similarity matrix (Table 1). These
data then were displayed graphically as a dendrogram (Fig. 2).
From the matrix (Table 1), it is obvious that the highest
relationship (0.91) is between the Precambrian configuration and structure
on top of the Mississippian. Gravity is related to the Precambrian
configuration (0.70) and Mississippian structure (0.75). Although the
PAIRWISE COMPARISON OF SPATIAL MAP DATA 221

Table 1. Correlation coefficients computed for pairwise comparison of

7 maps of Chautauqua County based on gridded interpolated values. Best
match is 0.91 between Precambrian configuration and Mississippian
structure. Results displayed in Figure 2 as dendrogram.

Pennsylvanian
Temperature

Mississippian
Temperature .34

Arbuckle
Temperature .14 .31

Mississippian
Structure .31 .30 .08

Gravity .33 .34 .16 .75

Aeromagnetics .03 .01 .10 .00 .03

Precambrian .21 .23 .05 .91 .70 .01

Configuration
Penn. Miss. Arb. Miss. Aero- Pre-
Temp. Temp. Temp. Struct. Gravity mag. Camb.

Aeromagnetic

Arbuckle Temp.

Miss. Temp.

Penn. Temp.
_____---Ir-
Gravity -
Miss. Structure

Precambrian

1.0 0.9 O.S 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0

Figure 2. Dendrogram based on correlation coefficients determined between pairwise

comparison of maps. 1.0 is perfect match; 0.0 is no match. Dendrogram based on the
WVGM method (Sneath and Sokal, 1963). Index map shows location of Chautauqua
County in southeastern Kansas.
222 MERRIAM, HERZFELD, AND FORSTER

gravity is related to structure, the aeromagnetics are not related to either

the structure or geothermics. This confirms the interpretation that the
aeromagnetics reflect anomalies within the Precambrian basement. This
relationship has been noted in others studies in eastern Kansas (Merriam,
Herzfeld, and Forster, 1993). The geothermal data for the three
stratigraphic units are not related highly to each other nor to the structural
and geophysical variables (Forster, Merriam, and Brower, 1993).
The original data sets (except for the aeromagnetic data) are
contoured and shown in Figure 3. Casual inspection of the patterns
displayed on these maps show little obvious similarity. The three
geothermal maps have a series of 'highs' and 'lows' but not all in the same
geographic location. There is a subtle increase in temperature to the west
that corresponds to the increased depth as a result of the westward dip of
the units. The gravity map also shows an overall increase in values from
east to west that coincides with the westward dipping structural surfaces.
Both the Precambrian and Mississippian surfaces, which dip to the west,
have numerous small features superimposed on them. Thus, without a
similarity coefficient, it would be difficult to determine the relationship
of one map to another from just a visual inspection of the contour pattern.
Utilizing the algebraic map-comparison program developed by
Herzfeld and Sondergard (1988), a pairwise comparison was made of the
data sets. Results of this comparison of the six maps are shown in Figure
4. The most obvious feature of this analysis is the high similarity
between the Precambrian-Mississippian structure maps (where the values
are 0.10 or less) over most of the area. The next best relationship is that
of the Precambrian-Gravity match. These results reenforce the results
obtained from the computation of the correlation coefficients, but show
spatially where the similarities are high. The similarities of the pairwise
comparison of the geothermal maps to each other and their relation to the
other map variables is spotty. However, the relation of Mississippian
temperature and structure shows that the similarity is close locally,
especially where the structure is positive and the temperature is higher.
Because some oil and gas fields in southeastern Kansas are associated
with positive structural features, it is reasonable to assume for these
structures a higher temperature will occur on the anticlines. This relation
of geothermal anomalies to the occurrence of petroleum in Kansas has
been noted by Ball (1982) and Merriam and Forster (1995). The
Pennsylvanian temperature map is difficult to interpret because of the
considerable thickness of the unit and BHTs collected from different
PAIRWISE COMPARISON OF SPATIAL MAP DATA 223

depth levels. The problems with the BHT maps are discussed in more
detail in Forster and Merriam (1993).

Pennsylvanian Temperature Arbuckle Temperature

Mississippian Temperature Gravity

MissiSSippian Structure Precambrian Configuration

Figure 3. Contour of original3-mi gridded data used for pairwise comparison. Contour
Interval (CI) = 2.5"F for temperature maps; 2 mgs for gravity map; 20 ft for
Mississippian structure map; and 50 ft for Precambrian configuration map. Area is
Chautauqua County, Kansas (f.32 S. - T. 35 S., R. 8 E. - R. 13 E.), approximately 620
sq mi. Location of area shown on Figure 2 inset map. Contouring by SURF ACE III
(Sampson, 1988).
224 MERRIAM, HERZFELD, AND FORSTER

Pennsylvanian-Arbuckle Temperature Pennsylvanian-Mississippian Temperature

Miss. Structure-Miss. Temperature Mississippian-Arbuckle Temperature

Precambrian-Mississippian Structure Precambrian Configuration-Gravity

Figure 4. Contoured pairwise comparison of map variable using algebraic MAPCOMP

(Herzfeld and Sondergard, 1988). Small values indicate high similarity; high values
poor similarity. Areas of high similarity (0.10 or less) are outlined. Contour Interval
(Cl) is 0.10 and 0.05. Area is Chautauqua County, same as in Figure 3. Amount of
correspondence is given in Table 1 by similarity coefficients - between Precambrian-
Mississippian structure (0.91 %); between Mississippian structure and gravity (0.75%);
and between Precambrian and gravity (0.70%). Areas of high similarity shaded.
PAIRWISE COMPARISON OF SPATIAL MAP DATA 225

CONCLUSION

Pairwise comparison of spatial map data offers a good, quick

method of analyzing the relationship between a series of maps of different
parameters. By combining a quantitative assessment on the degree of
similarity and a technique of outlining areas of high similarity, a good
understanding on spatial relations can be gained to help in the geological
interpretation.

ACKNOWLEDGMENTS

We would like to thank Rick Brownrigg of the Kansas Survey for

help in processing the data. Partial funding for the study was provided
by Conoco Oil Company and the Wichita State University Geology
Research Fund.

REFERENCES

Agterberg, F.P., Bonham-Carter, G.F., and Wright, D.F., 1990, Statistical

pattern integration for mineral exploration, in Gaal, G., and
Merriam, D.F., eds., Computer applications in resource estimation,
prediction, and assessment for metals and petroleum: Pergamon
Press, Oxford, p. 1-21.
Ball, S.M., 1982, Exploration application of temperatures recorded on
log headings - an up-the-odds method of hydrocarbon-charged
porosity prediction: Am. Assoc. Petroleum Geologists Bull., v. 66,
no. 8, p. 1108-1123.
Bonham-Carter, G.F., and Agterberg, F.P., 1990, Application of a
microcomputer-based Geographic Information System to mineral.-
potential mapping, in Hanley, J.T., and Merriam, D.F., eds.,
Microcomputer applications in geology, IT: Pergamon Press, Oxford,
p.49-74.
Brower, J.C., and Merriam, D.F., 1990, Geological map analysis and
comparison by several multivariate algorithms, in Agterberg, F.P.,
and Bonham-Carter, G.F., eds., Statistical applications in the earth
sciences: Geol. Survey Canada Paper 89-9, p. 123-134.
226 MERRIAM, HERZFELD, AND FORSTER

Brower, lC., and Merriam, D.F., 1992, A simple method for the
comparison of adjacent points on thematic maps, in Klirzl, H., and
Merriam, D.F., eds., Use of microcomputers in geology: Plenum
Press, New York and London, p. 227-240.
Cole, YB., 1962, Configuration of top Precambrian basement rocks in
Kansas: Kansas Geol. Survey Oil and Gas Invest. 26, map.
Davis, lC., 1986, Statistics and data analysis in geology (2nd ed.): John
Wiley & Sons, New York, 646 p.
Eschner, T.R, Robinson, lE., and Merriam, D.F., 1979, Comparison of
spatially filtered geologic maps: summary: Geol. Soc. America Bull.,
pt. 1, v. 90, p. 6-7.
Forster, A, and Merriam, D.F., 1993, Geothermal field interpretation in
south-central Kansas for parts of the Nemaha Anticline and flanking
Cherokee and Sedgwick Basins: Basin Research, v. 5, no. 4, p. 213-
234.
Forster, A, Merriam, D.F., and Brower, lC., 1993, Relationship of
geological and geothermal field properties: Midcontinent area, USA,
an example: Math. Geology, v. 25, no. 7, p. 937-947.
Gold, e., 1980, Geological mapping by computer, in Taylor, D.RF., ed.,
The computer in contemporary cartography: John Wiley & Sons,
Chichester, p. 151-190.
Herzfeld, UC., and Sondergard, M.A, 1988, MAPCOMP - a FORTRAN
program for weighted thematic map comparison: Computers &
Geosciences, v. 14, no. 5, p. 699-713.
Herzfeld, Ue., and Merriam, D.F., 1991, A map-comparison technique
utilizing weighted input parameters, in Gaal, G., and Merriam, D.F.,
eds., Computer applications in resource estimation, prediction, and
assessment for metals and petroleum: Pergamon Press, Oxford, p.
43-52.
Herzfeld, UC., and Merriam, D.F., 1995, Optimization techniques for
integrating spatial data: Math. Geology, v. 27, no. 5, in press.
Lamb, C., and Yarger, H.L., 1985, Absolute gravity map of Kansas:
Kansas Geol. Survey Open-File Rept. 85-15, map. Scale 1:500,000
Maslyn, RM., 1991, Boolean grid logic - combining expert systems with
grid nodes (abst.): Proc. GeoTech/GeoChautauqua '91, Denver
GeoTech, Inc., Denver, CO., p. 316.
Merriam, D.F., 1963, The geologic history of Kansas: Kansas Geol.
Survey Bull. 162, 317 p.
PAIRWISE COMPARISON OF SPATIAL MAP DATA 227

Merriam, D.F., 1992, Geological interpretation of integrated thematic

spatial data: Geologisches Jahrbuch Reihe A, Hf. 122, p. 233-24l.
Merriam, D.F., and Forster, A., 1995, Subsurface temperature anomalies
as a key to petroleum-producing areas in the Cherokee and Forest
City Basins, eastern Kansas? (abst.): Am. Assoc. Petroleum
Geologists Regional Meeting (Tulsa, Oklahoma), 1 p.
Merriam, D.F., and Jewett, D.G., 1988, Methods of thematic map
comparison, in Merriam, D.F., ed., Current trends in
geomathematics: Plenum Press, New York and London, p. 9-18.
Merriam, D.F., and Lippert, R.H., 1966, Geologic model studies using
trend-surface analysis: Jour. Geology, v. 74, no. 3, p. 344-357.
Merriam, D.F., and Robinson, IE., 1981, Comparison functions and
geological structure maps, in Future trends in geomathematics: Pion
Ltd., London, p. 254-264.
Merriam, D.F., and Sneath, P.H.A., 1966, Quantitative comparison of
contour maps: Jour. Geophysical Res., v. 71, no. 4, p. 1105-1115.
Merriam, D.F., and Sondergard, M.A., 1988, A reliability index for the
pairwise comparison of thematic maps: Geologisches Jahrbuch, v.
AI04, p. 433-446.
Merriam, D.F., Fuhr, B.A., and Herzfeld, U.C., 1993, An integrated
approach to basin analysis and mineral exploration, in Harff, J., and
Merriam, D.F., eds., Computerized basin analysis; the prognosis of
energy and mineral resources: Plenum Press, New York and
London, p. 197-214.
Merriam, D.F., Herzfeld, U.C., and Forster, A., 1993, Integration of
geophysical data for basin analysis: Proc. 3rd Intern. Brasilian
Geophysical Soc. Congress (Rio de Janeiro), v. 2, p. 1025-1030.
Miller, R.L., 1964, Comparison-analysis of trend maps, in Computers in
the mineral industries, pt. 2: Stanford Univ. Publ., Geol. Sci., v. 9,
no. 2, p. 669-685.
Mirchink, M.F., and Bukhartsev, V.P., 1959, The possibility of a
statistical study of structural conditions: Doklady Akad. Nauk SSSR
(English translation), v. 126, no. 5, p. 1062-1065.
Rao, S.V.L.N., 1971, Correlations between regression surfaces based on
direct comparison of matrices: Modem Geology, v. 2, no. 3, p. 173-
177.
Ribeiro, J.C., and Merriam, D.F., 1979, Quantitative analysis of
depositional environments (Aratu Unit, Reconcavo Series, Lower
Cretaceous) in the Reconcavo Basin, Bahia, Brazil, in Gill, D., and
228 MERRIAM, HERZFELD, AND FORSTER

Merriam, D.F., ed., Geomathematical and petrophysical studies in

sedimentology: Pergamon Press, Oxford, p. 219-234.
Robinson, IE., and Merriam, D.F., 1972, Enhancement of patterns in
geologic data by spatial filtering: Jour. Geology, v. 80, no. 3, p.
333-345.
Robinson, IE., and Merriam, D.F., 1984, Computer evaluation of
prospective petroleum areas: The Oil & Gas Jour., v. 82, no. 34, p.
135-138.
Robinson, IE., Merriam, D.F., and Burroughs, W., 1979, A quantitative
technique to determine relation of oil production to geologic
structure in Graham County, Kansas, in Davis, IC., and deLopez,
S.L., eds., Computer mapping for resource analysis (Mexico City):
Inst. de Geografia de la UNAM and Kansas Geol. Survey, p. 256-
264.
Sokal, RR, and Sneath, P.R.A., 1963, Principles of numerical taxonomy:
W.R. Freeman and Co., San Francisco, 359 p.
Thrivikramaji, K.P., and Merriam, D.F., 1976, Trend analysis of
sedimentary thickness data: the Pennsylvanian of Kansas, an
example, in Merriam, D.F., ed., Quantitative techniques for the
analysis of sediments: Pergamon Press, Oxford, p. 11-21.
Unwin, D., 1981, Introductory spatial analysis: Methuen, London and
New York, 212 p.
Yarger, H.L. and others, 1981, Aeromagnetic map of Kansas: Kansas
Geol. Survey Map M-16. Scale: 1:500,000
 APPLICATIONS OF SPATIAL FACTOR ANALYSIS
TO MULTIVARIATE GEOCHEMICAL DATA

E.C. Grunsky
Geological Survey Branch, Victoria, British Columbia, Canada

Q. Cheng
University of Ottawa, Ottawa, Ontario, Canada

F.P. Agterberg
Geological Survey of Canada, Ottawa, Ontario, Canada

ABSTRACf

Spatial Factor Analysis (SPF AC) is a technique used to determine

multivariate relationships from the auto- cross-correlation relationships of
spatially referenced data. The method derives linear combinations of
variables with maximum autocorrelation for a specified lag distance .
.Eigenvectors are calculated from a covariance quotient matrix derived
from auto-cross correlation estimates at lag 0 and distance d. SPF AC has
been applied to regional geochemical (moss mat) sampling data collected
in the southern part of Vancouver Island, British Columbia, Canada. The
results of the analysis show that regional geochemical trends and
interelement associations change with lag distance and orientation and can
reflect the underlying spatially based lithological variations. The
technique also has been applied to lithogeochemical data in the Sulphurets
area of British Columbia, Canada. Zones of alteration associated with
porphyry copper and gold mineralization have been shown to be distinct
spatially from the background.

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 229
230 GRUNSKY, CHENG, AND AGTERBERG

IN1RODUcnON

Traditional approaches to multivariate analysis such as principal

components analysis do not consider the spatial relationships of the data.
Multivariate relationships of spatially based data have been studied by
Agterberg (1966, 1974), Myers (1982, 1988), Royer (1988), Switzer and
Green (1984), Wackernagel (1988), Grunsky and Agterberg (1988, 1992),
Grunsky (1989). Grunsky and Agterberg (1988, 1992) have shown that
the application of a technique termed, SPFAC, produces linear
combinations of variables that have maximum correlation based on auto-
cross-correlation estimates at specific lag intervals.
Multivariate analysis has been applied to a wide range of
geochemical studies and techniques for assessing geochemical data are
well documented (Howarth and Sinding-Larsen, 1983; LeMaitre, 1982).
Geochemical data usually are correlated and within a given sampling
regime, several populations of data may represent different geochemical
processes. The study of multielement geochemical data has been studied
in a spatial context using a variety of techniques. Wackernagel (1988)
contrasted the results of a principal components analysis between the
variance-covariance matrix and a spatial covariance matrix based on
variogram estimations using geochemical data from a regional soil survey.
Bourgault and Marcotte (1991) used a spatial form of the Mahalanobis
distance (multivariable variogram). The technique was applied to a
lithogeochemical survey over a pluton. Bellehumeur, Marcotte, and
lebrak. (1994) have studied the spatial properties of principal components
from two fractions of a stream sediment survey in Quebec.
The spatial factor technique employed here differs from the
described approaches in the following ways:

(1) SF A determines linear combinations of variables and their

significance to the spatial pattern from a given direction and some
specified lag distance, d. Thus, spatial relationships can be evaluated
as a function of lag distance and direction.
(2) The SPF AC technique uses the auto- cross-correlation
relationships based on the signal at lag zero as opposed to the
covariances of the variables.

The technique of SPF AC is described by Grunsky and Agterberg

(1988, 1992). Details of the method will not be described here. The
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 231

application of SPF AC as used in this study employs the use of the two
point model only, that is, only one orientation is considered in the
analysis.
Given a vector of variables, X, with lag 0 auto- cross-correlations
of Ro , and lag d auto- cross-correlations of Rd where

lall a12 ....... . alml IFdll Fd12 ....... Fdlml

la21 a22 ....... . a2m l IFd21Fd22 ....... Fd2ml
RO= I. I Rd= I I
I. I I I
laml am2 ....... . amml IFdml Fdm2 ....... Fdmm l

F dij represents estimates of auto- cross-correlations of variables Xi and Xj

for lag d using a function F. If a is a column vector with m unknown
elements, then a'x has lag 0 auto- cross-correlations of a'ROa and lag d
auto- cross-correlations of a'Rda. uRda can be maximized subj ect to the
constraint that a'Roa= 1.

The solution satisfies:

or

where '" is the Lagrange multiplier and a IS an eigenvector of the

transition matrix U such that

For m variables Ud can be decomposed into p separate spectral

components. Estimates of the overall noise of the auto- cross-correlations,
and a measure of the goodness of fit (squared multiple correlation
coefficients, R2) to estimate the predictive power of each component can
be computed as outlined in Grunsky and Agterberg (1988, 1991b). For
the approach to be valid, Rd and RO must be positive definite. A
correction procedure can be applied to modify R d, or Ro such that they
are positive definite (Grunsky and Agterberg, 1988, 1991b).
The two-point model is determined by a single vector that
determines the relationship between locations using the underlying
statistical model:
232 GRUNSKY, CHENG, AND AGTERBERG

J d + E'·I
Z'·I = Z'·U

where Z'i and Z'j are row vectors consisting of standardized values of
variables i and J; Ei is a row vector consisting of residuals. If Sj
represents a column vector of the signal values corresponding to Z, then
premultiplication of both sides of the given equation by Sj yields:

S·Z'·=S·Z'·Ud + S·E'·
J I J J J I
Each column of U d represents a set of regression coefficients by which
the value of a variable at location i is predicted from the values of all
variables at location j. If the residual variance for variable Zk is written
as a\, then the k-th column ofUd has a 2R ol as its variance-covariance
matrix. The variances of the variables are proportional to the elements
along the main diagonal of this matrix. Grunsky and Agterberg (1991a,
1991 b) have provided computer programs for the implementation of the
spatial factor technique.

MODEIS OF AUTO- CROSS-CORRELATION

The spatial factor procedure requires that estimates of auto- cross-

correlation be obtained for the variables over the spatial range of the data.
Auto- cross-correlation relationships can be expressed such that a random
variable X k (k=I, ... ,m), with zero mean, at location xi' in geographical
space is related to every other location Xj by some function F where,

x·1 = F(d·IJ·)x·J + y.I

Both i and j go from 1 to N where N denotes total number of

observations (Agterberg, 1970). F(dij) is a function of distance dij
between these two points. The function is considered to decay to zero at
the maximum distance of the "neighborhood" for which it is defined. The
residual Yi is the realization of a random variable Yi at point i. It satisfies
E(Yi) = 0 and Yi is assumed to be independent of Xj.
The selection of functions used to estimate auto- cross-correlation
is dependent on the nature of the spatial relationships. A number of
models can be used such as the spherical, Gaussian, or exponential
models. These functions have specific characteristics that are useful for
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 233

describing the behavior of regionalized variables (Journel and Huijbregts,

1978). A specific requirement of autocorrelation functions is that they
satisfy the following conditions. Given the autocorrelation of a random
variable, rx ' and lag, d: a) Irid)1 ~ 1~ b) rid) = rx(-d). As well,
functions that obey the properties required of regionalized variables
within defined limits also can be applied. Quadratic functions were used
as estimates of spatial auto- and cross-correlation by Agterberg (1970)
and Grunsky, and Agterberg (1988, 1989).
In this study, three functions of autocorrelation (quadratic,
exponential, and Gaussian) were used to define the spatial variation of the
irregularly spaced multielement geochemical data sets. For each variable
and pair of variables, estimates of auto- cross-correlation were computed
for specified lag intervals representing the lag coefficients of a
correlogram. From these experimental correlograms, the parameters for
each function then were calculated using the downhill simplex method of
multidimensional minimization (Press and others, 1986, p. 289) whereby
successive estimates of the function parameters are selected based on the
sum of squares between the observed autocorrelations and the function
estimates. The parameter estimates converge as the sum of squares
converges to a minimum.
SPF AC can make use of any designated function specified to
create values for Ro and Rd. In this study SPF AC was run using four
types of estimates of auto- cross-correlation. Three estimates were based
on the exclusive use of quadratic, exponential, and Gaussian functions. A
fourth estimate was made using a mix of functions (mixed model) based
on the minimum sum of squares criteria. For each pair of variables the
function whose parameters satisfied the minimum sum of squares was
selected. For most variables, the exponential function usually met the
minimum sum of squares criterion

STUDY AREAS

Two areas were selected for study. Southern Vancouver Island

was selected as an area to evaluate the use of SPF AC for interpreting
regional geochemical survey data. The Sulphurets area in the northern
part of the province was selected to evaluate the use of SPF AC in a
mineralized area.
234 GRUNSKY, CHENG, AND AGTERBERG

The mmimum distance between sample sites on southern

Vancouver Island ranges from 0.07 to 9.65 kilometers with a median
distance of 3.65 kilometers. The average minimum distance between
sample sites in the Sulphurets area ranges from 1 to 670 meters with a
median minimum distance of 55 meters.
The data from regional geochemical sampling program in the
southern part of Vancouver Island contains information about regional
geological processes such as major lithologic differences and structural
trends of major lithologies. Anomalous samples that reflect local
processes were not included in this study because of the poor
autocorrelation of such data as a result of low-sampling densities.
Data were analyzed using FORTRAN programs prepared by
Grunsky and Agterberg (1991a, 199Ib). In addition data from the
southifD part of Vancouver Island were analyzed and displayed using
Splus (Statistical Sciences, 1993). Data from the Sulphurets area were
displayed using SPANSTM (INTERA TYDAC, 1993).

GEOWGY OF SOUmERN VANCOUVER ISLAND

The southern part of Vancouver Island lies at the southern end of

the Insular Belt of the Canadian Cordillera (Fig. 1). It comprises three
fault-bounded terranes which were assembled in the late Eocene (N.W.D.
Massey, pers. comm., 1994). The bulk of the island is underlain by the
Wrangellia terrane. This thick Paleozoic to Mesozoic sequence is made
up of three major volcanic cycles - the oceanic island-arc volcanics of the
Devonian Sicker Group, the oceanic flood-basalts of the Upper Triassic
Karmutsen Formation and the Lower Jurassic arc volcanics of the
Bonanza Group. Coeval with the Bonanza Group are the plutons of the
Island Intrusive Suite and the mid-crustal level metamorphic rocks of the
West Coast Complex. The volcanic sequences are separated by the clastic
sediments and limestones of the upper Paleozoic Buttle Lake Group and
limestones and shales of the Quatsino and Parson Bay Formations.
Sandstones, shales, and conglomerates of the Upper Cretaceous Nanaimo
Group lie unconformably on the older rocks. Wrangellia has undergone
a complex structural history, with frequent rejuvenation of previous
structures, imparting a prominent northwesterly aligned structural fabric.
The present map pattern is dominated by the northwesterly trending
contractional faults of the Tertiary Cowichan fold and thrust system.
 ~ Carmanagh Group
~
Geology of Southern Vancouver Island ~ Metchosin Igneous Complex

~ Flores Volcanics CZl

273000 E
5434000 N Nanaimo Group
~o
B "T.l
0IITl Leech River Complex CZl
~
.....
, '" ,
~ ..... Pacific Rim Complex :>
Bonanza Group
d
... 'j. -:"~
FXill
~ ~
~ Parson BaylOuatsino Forrmtions "T.l
:>
(')
0-- - Karmutsen Forrmtion
,..;j
1:: 11:: Buttle Lake Group
o ~
tt~~ Sicker Group

fm~l Westcoasl Crystalline Complex

273000 E 502000 E
I Island Plutonic Suite
5330000 N 5330000 N
mi
CZl
~
CZl
-
N
I.H
Figure 1. Geology of southern part of Vancouver Island, British Columbia. See text for detailed explanation of map codes. VI
236 GRUNSKY, CHENG, AND AGTERBERG

The Leech River Complex is separated from Wrangellia by the

San Juan - Survey Mountain Faults. It comprises Jura-Cretaceous marine
sediments and basaltic rocks, regionally metamorphosed and deformed to
schists and amphibolites. The ophiolitic basalts and gabbros of the
Metchosin Igneous Complex have been thrust beneath the southern end
of Vancouver Island along the northerly dipping Leech River Fault. The
post-accretion, clastic sediments of the Eocene-Oligocene Carmanagh
Group unconformably overlie all older rocks along the southwestern
coast.

GEOWGY AND AU-ASSOCIATED ALTERATION

OF mE SULPHURETS AREA

The Mitchell-Sulphurets mineral district in northwestern British

Columbia is about 120 km2 in size. It is noted for extensive alteration
zones associated with porphyry copper and molybdenum systems as well
as other types of gold and silver mineralization. The area is underlain by
Jurassic volcanics and sedimentary rocks, primarily from the Hazelton and
Stuhini Groups. The central part of the area is underlain by so-called
pyritic altered rocks (Fig. 2) that host most of the potassic, sulphidic, and
silicic alteration zones. The contacts of the pyritic altered rocks with the
Stuhini Group to the west and the Hazelton Group to the east consist of
faults striking NS to NE-SW and NS to NNW-SSE, respectively. Most of
the geochemical anomalies in the area are spatially related to the
alteration (Cheng, Agterberg, and Ballantyne, 1994).
Vein-stockwork systems occurring in the pyritic altered rocks (Fig.
2) show both brittle and ductile deformation which probably took place
at relatively great depths. Most volcanic and sedimentary rocks in the
study area are cut by subvolcanic porphyritic intrusions of dioritic,
monzonitic, syenitic, and low silica granitic composition. Relatively large
granitic intrusions are located in the western part of the study area and
smaller dioritic intrusions in the southern and eastern parts. The pyritic
alteration, extensive quartz vein stockworks and several types of gold
mineralization zones (Cu-Au, Mo-Au, Cu-Mo-Au, and As-Au-Ag)
probably are related genetically to the emplacement of these porphyry
alkalic intrusions. Some Au mineralization zones may be related to
younger faults (Alldrick, 1987; Alldrick and Britton, 1988; Kirkham,
1990; Kirkham, Ballantyne, and Harris, 1989, 1990, 1992).
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 237

130°15'
k::::::.1 Pyritic Allered Rocks

~ Stockwork
I -- I Quart z vein

-,

o 2000
m
°
130 10'

Figure 2. Alteration zones in Mitchell-Sulphurets district.

Gold and Au-associated elements and oxides show spatial

correlation with the intrusive rocks and the faults. The main purpose
object of applying SPFAC method in the Sulphurets area is to study these
relationships.

REGIONAL GEOOlEMICAL SAMPLING

PROGRAM - SOUTIlERN VANCOUVER ISLAND

The southern part of Vancouver Island, (National Topographic

Series Maps 92B and 92C) was sampled as part of a regional
geochemical reconnaissance sampling program (Matysek and others.,
1990). Moss mats were sampled and analyzed for their geochemical
responses in place of traditional stream sediments. Moss mat sampling is
238 GRUNSKY, CHENG, AND AGTERBERG

an alternative to sampling stream sediments in areas where fine sediment

material generally is lacking in stream beds, as is the situation for
Vancouver Island. Moss mats capture fine sediment and are plentiful in
Vancouver Island streams (Matysek and Day, 1987). Quality control and
sample preparation procedures are discussed in Matysek and others.
(1990).
A total of 599 sites were sampled for moss-mat sediment at an
average sampling density of 1 site per 10.3 km 2 . The sample sites are
shown in Figure 3. Moss mat samples weighing 1-2 kg were scraped from
logs and boulders which occur within the active stream channel. The
samples were analyzed for 22 elements using a variety of determination
techniques as documented in Matysek and others. (1990). Because the
sample sites follow stream beds, the locations of the samples do not cover
the map area in a uniform manner. As a result some lithologies have not
been sufficiently sampled.

(
'::..:. :: ....... "':::"!" ....... , . :., '.: ....., .... \~... .
5420 N , . !....'" .... ; .......... "" ~ "'" t • \' .. ...
I.: : .... '\.. ... .. •'~ :
,..l ~·u-" ....... .. -. ... =.,'-""
t "',( ••~' .,~t..;':

T
t~·
". I ,. :.-."":, '" ........
5400 ·'.~t
"'ltt •.... ~ Ill' • t··...
:. J.' • .... 1 " .',
......
..
.. I .. '(-.............1 ~ ; . '" :

. "....... .
0
0
0 "" I ..1 t', ; " .. ..... .. ..
"';:; 5380 .. ., .......' .. "' ,) ... , ..
", ": ., ," ....... !\ ··-.. to ". ..
_t~

. ......
~ -.,..,.: ..... ..
- ,........, ~. ~,
..
E
5360 ' .. ! .......... t.

50km
5340

300 350 400 450 soo

m x 1000

Figure 3. Sample locations, southern part of Vancouver Island. Sample sites follow
drainage network.

PRELIMINARY ASSESSMENT OF DATA

SOUTIlERN VANCOUVER ISLAND

Regional geochemical survey data can be difficult to interpret. A

significant difficulty in the interpretation of geochemical data is the
nonnormal nature of their distributions. Meaningful estimates and
interpretations of nonnormal distributions is unlikely if statistical
measures of the distributions are made under the assumption of normality.
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 239

In many distributions, the skewed nature of the data can be overcome by

applying a suitable transformation that shifts the values of the distribution
such that it becomes normally distributed. The method selected for the
Southern Vancouver Island study was that of applying Box-Cox
generalized power transformations using a method outlined by Howarth
and Earle (1979). The Box-Cox family of transformations are defined as:
A. A.
y=(x - 1)/ for A. > 0
y=ln(x) for 1..=0

The presence of outliers in the data poses a problem particularly

for determining an appropriate value of A.. Campbell (1986) has
developed a method for determining a suitable value of A. based on the
presence of outliers but this method was not used in this study. One way
to minimize the effect of outliers, is to trim the data at the 99, 98, and 95
percentile rankings and compute values of A. for each group of data. This
approach was taken for this study. Table 1 lists the values of A. for a
selected group of elements calculated and the subsequent value that was
used. The selection of an appropriate value of A. in the situation of
outliers is problematic.

Table l. Values of A. and replacement values for selected elements.

Element L.L.D n<L.L.D Trim A Replacement Value
Pb 2.00 256 0.98 Log 1. 00
eu 2.00 0 0.98 0.2
Ni 2.00 0 0.98 0.25
Co 2.00 0 0.98 0.5
Mn 5.00 0 0.98 Log
Fe 0.02 0 0.98 Log
U 0.50 13 0.98 0.3 0.04
As 1. 00 216 0.98 Log 0.39
v 5.00 0 0.98 0.2
Cr 5.00 0 0.98 0.3

L.L.D. - Lower Limit of Detection

n<L.L.D. - number of samples less than the detection limit.
Trim - the maximum percentile used for determining the value of A
A - Value for Box - Cox power transformatlon
For A=O, the natural logarithm lS used for the transformation.
Replacement value determined only for censored distributions.

Another problem is the occurrence of censored distributions.

Censored distributions result when a portion of the sample population is
less than some specified value (i.e., the detection limit). The problem of
censored data becomes more apparent when means and covariances are
240 GRUNSKY, C:EIENG, AND AGTERBERG

required. The use of a single substituted value biases the computation of

the moments of the distribution. However, if the distribution is assumed
to be normal, then the replacement value of the censored data and
parameters of the distribution (mean, variance) can be estimated from the
portion of the distribution that is not censored. Estimates of the
distribution parameters can be obtained using the EM algorithm
(Dempster, Laird, and Rubin, 1977) and is discussed by Campbell (1986),
and Chung (1985, 1988). Chung (1985) and Campbell (1986) have
published computer programs for the statistical treatment of geochemical
data with observations below the detection limit. The assumotion of
normality is essential for the EM algorithm to work. Campbell (1986)
invokes an algorithm to transform the data to normality using Box-Cox
transformations as described in the previous section. More recently,
Sanford, Pierson, and Crovelli (1993) have developed a method that
allows the calculation of a suitable replacement value based on a
maximum likelihood approach.
Q-Q plots were made for 22 of the elements for both the
untransformed and power transformed data The plots were compared for
assurance that the power transform was appropriate and to evaluate the
degree of censoring of the data. Elements including Ag, Cd, and Bi are
highly censored which results in difficulties for meaningful estimates of
auto- cross-correlations. An example of the effects of the Box-Cox power
transformation for the element, Cu, is shown in Figure 4.
Untransformed Lambda=.2

co
,.
I
I"-
o
0
E
~
~

g;
CD
:l
()
<1 g
II)

o
...
C'l
N

-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Quantiles of Standard Normal Quantiles of Standard Normal
Figure 4. Effect of Box-Cox power transformation for Cu, southern Vancouver Island.
A, Q-Q plot of original data for Cu. Distribution is not normal and can affect
calculations of means and variances. B, Q-Q plot of transformed data for Cu. Note
straightline indicating distribution is near normal.
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 241

In the situation of constant sum data, the problem of

transformations and outlier detection are more problematic (Barcelo,
Pawlowsky, and Grunsky, 1994). The use of Box-Cox power
transformations can achieve normality for the marginal distributions;
however, the multivariate distribution may not be multivariate normal.
The application of log-ratios as outlined by Aitchison (1986) overcomes
the problem of the interpretation of constant sum data, however the
resulting log-ratio distributions may themselves not be normal and thus
may be difficult to interpret.
Subsequent to determining the most appropriate transformation for
each element, the technique outlined by Chung (1985) was used for
determining suitable estimates of the means and standard deviations of the
transformed elements. From the estimates of the distribution parameters,
replacement values were calculated as described by Sanford, Pierson, and
Crovelli (1993). Table 1 lists the values of A, and the subsequent
replacement values that were used for the censored elements, Pb, U, and
As.
During the initial evaluation of the data, the elements were
examined for their potential usefulness for deciphering geochemical
trends. The use of scatterplot diagrams with all possible pairs of elements
indicated that some elements show poor correlation with others. This was
the situation for the elements such as Sn and Ag. These elements were
eliminated for any further consideration.
Principal components analysis was applied to the transformed data.
The results indicated that there is a clear distinction between the mafic
igneous rocks, the felsic plutonic rocks and the metasedimentary
sequences. Statistical summaries of the data by Matysek and others
(1990) show that many of the elements display distinct differences in
population distributions as a function of lithology. Elements that typically
reflect the mafic volcanic lithologies of the Sicker Group volcanics, mafic
rocks of the Coast Plutonic Complex, the Karmutsen basalts, and the
Metchosin Group pillow basalts (Fig. 1) include Co, Cr, Cu, Ni, Fe, and
v. Elements that reflect felsic volcanic and intermediate to felsic intrusive
lithologies include, As, Pb, and U. These elements represent lithologies
such as the felsic Island Plutonic Complex. Sedimentary rocks which
include the Sooke, Carmanah, Nanaimo, Leech River, Pacific Rim, and
Buttle Lake groups are less obvious to distinguish solely on the basis of
10 selected elements that were used for the study. Differences between
the mafic lithologies are described by associations of Co, Cr, Cu, Ni, Fe,
242 GRUNSKY, CHENG, AND AGTERBERG

and V. Iron and Cu usually are associated with mafic volcanic and
intrusive rocks of the area including some phases of the Westcoast
Crystalline Complex, probably representing assimilated mafic protolith.
Elements such as Ni, Co, and Cr usually are restricted to the Sicker,
Karmutsen and Bonanza volcanic groups and the Metchosin Igneous
Complex.

FSTIMATFS OF SPATIAL AUTO-

CROSS-CORRELATION

Subsequent to the correction procedures outlined previously, auto-

cross-correlation values were determined for a range of lag intervals and
orientations. Examination of the spatial distribution of the data suggested
that a lag interval between 10 and 40 kilometers would reflect the most
of the spatial processes. From Figure 1 it can be seen that the dominant
tectonic and lithologic trends in the area are in the NW -SE direction for
the northern and eastern parts of the area. An E-W trend predominates in
the southwestern part of the area and some N-S trends are observed for
some units in the eastern part of the area. Thus, the orientations were
subdivided into N-S, NE-SW, E-W, and NW-SE directions. Experimental
auto- cross-correlation correlograms were determined for the four
orientations with an angle of tolerance of 15° and lag intervals ranging
from 10 to 40 kilometers. The data generated from these correlograms
then were used to determine the parameters for the three models of
autocorrelation as discussed earlier.
An examination of correlograms and signal to noise ratios of the
elements indicated that the following elements would be the most suitable
for study: Cu, Pb, Ni, Co, Mn, Fe, D, As, V, Cr. As described previously,
these elements reflect mafic and felsic igneous processes to a significant
extent at the scale of sampling and size of the area. Other elements such
as Ag, Au, Bi, W, Sn, and Sh tend to he censored highly or have such
high nugget effects at the scale of sampling, that it is difficult to use
these elements to detect spatial geochemical processes. The elements, Sh,
Au, Ag, Cd, Mo, Sn, Bi, Hg, W, F, and L.O.I. were not included in the
study because the low-sampling density resulted in large nugget effect
values. Auto-cross-correlograms for these elements were difficult to
interpret.
 Cu [Transformed]
E-W NE-SW
q q
OuadfllLlc Funcuon

U~I'1I·&I FunctIOn
G&uMlan FunellOtl ("')
~
\i ObWIYed Collet.llon ~
~
o 1 T o
~
tiL T
, . T·. "'! ~
"'! .. \~ '''' h~ \ p' • '. tn
o o
''1;
f'\.. - .... ,~ I., .
. ',1"1 rf J~"", '~'" ',1 ,'" " ,
' I'1 '. • • ...}J
• ~",.~ ~
N "'! tn
9 9 '"d
o 2 3 4 o 2 3 4
Lag (km) x 10 Lag (km) x 10 ~
N-S NW-SE
~
'"rj
q q >
("')
~
I~t
(£) ~ ~
o o \" r.,
"
...
"'.t.. It •

N
.' •... ,
"'!
o o
.. .''-..r,... .j'-r-.':,t\
. . ;"'.,~\. ....... tn
~
......
N
tn
N
9 9
o 2 3 4 o 2 3 4
Lag (km) x 10 Lag (km) x 10

Figure 5. Experimental correlogram and 3 function model fits for Cu, southern Vancouver Island. Note changing signal to noise
ratio and decay rate for different orientations. Sa: Correlogram for E-W direction with signal to noise ratio of --0.6. 5b: N
Correlogram for NE~SW direction with signal to noise ratio of --0.3 . 5c: Correlogram for N-S direction with signal to noise ratio ~
of --0.5. 5d: Correlogram for NW-SE direction with signal to noise ratio of --0.7.
244 GRUNSKY, CHENG, AND AGTERBERG

An examination of the auto correlations of the 10 selected elements

showed that correlations were strong in some directions and weak in
others. Elements that show strong autocorrelation signals in the NW -SE
direction include Cu, Ni, Co, Mn, Fe, and V. Elements that show strong
autocorrelation in the N-S direction include U, As, and Cr. Figure 5
displays plots of the auto-correlation functions derived for Cu over the
four different orientations. The figure shows that Cu is distributed
anistropically with the longest decay in the NW -SE direction and the
shortest decay in the NE-SW direction. The figure also shows plots of the
three function fits (quadratic, Gaussian, and exponential) to the
correlogram. The correlograms and functions indicate a decay to zero in
the range of 25 km for the E-W, NE-SW, and N-S directions. The
correlograms decay to values below zero for the these three directions and
indicate that there is a lack of stationarity. For the NW-SE orientation the
correlogram and functions decay to zero at a distance greater than 40 km.

LITHOGEOOIEMICAL DATA FROM THE SULPHURETS AREA

About 1233 surface outcrop samples (Fig. 6) were collected along

mapping traverses for mineral deposit research by RV. Kirkham, S.B.
Ballantyne, and D.C. Harris of the Geological Survey of Canada during
the period 1986 to 1990. Both altered and unaltered rocks were sampled.
The samples were analyzed chemically for approximately 30 trace
elements and 13 oxides (Ballantyne, 1990; Ballantyne and others, 1992).
The lithogeochemical data were stored using the Geographical
Information System SPANS (INTERA TYDAC, 1993) for preparation of
contour maps and integration with other types of data.
Statistical characteristics of the lithogeochemical data were studied
using various multivariate analysis procedures including factor analysis
and cluster analysis showing that As, Sb, Ag, Pb, Zn, Cu, Mo, F, Nb, S,
Cd, Se, Eu, K 2 0, and Si0 2 are correlated positively whereas Cr, V, Ce,
Co, Tb, Fe, T, Sr, Sm, W, La, MnO, MgO, and CaD are correlated
negatively with Au. Contour maps for these elements show that most Au-
associated elements have higher concentration values in two separate
zones. The first zone strikes SN to NE-SW and lies close to the western
contact between the pyritic altered rocks and the Stuhini Group, and the
other one strikes SN to NW -SE along the eastern contact with the
Hazelton Group. Au-associated elements in the first zone include Au, Cu,
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 245

Mo (to the north), K 20, and Al 20 3 ' and those in the second zone As,
Sb, Ag, and Si0 2 . The higher temperature element association in the first
zone may be the result of being in closer proximity to intrusive rocks and
greater depth below the topographic surface at the time of origin.

.•
13d']6" •

.'~
... ".

• •".). .'
"',..
'f.,
. \
-... ..." .-
~ ..-
-: L • .,.".
~. '.' r
" ... -'1"':'-""'-,.:.. .
I ~• •

~...

: I'· • i. tt!
.' ",.... "'!AI
l.,.,·· . .~L
\. • ,.*"-,.......
.,.
*..
• • • • • • ' -:-. ' . ' .

..... ... •• J"

I.
I ••
• ,.' • •
.....
•
•• fa .,.
... 1- i
t
IiIt· •

,.
56'<lY • .. t • 'f.. t':." '.. • ...

. ....
. . ..... - ...
... • ,. '-" ....t •

..
II • • • .... . . . . . . • •
:
~Ia·.· .,,:
• t •• t • ... .'
\, ,~.
....
~"" ..,,"~... .
\ ..~. . :. ., 1:" ~ :
... t .• •• .....
. , ••
. . :.. i.··.· .
.... .tjt.1.,.
"-I

... •
oNto!; t
•
\, ...
, .. I~. ~
•••
.~.
~
..
..... ••

.
at lit •.•• --: ~
•J
. ...
• ''''.. 'til •
~\.r.;i·
.t .t ••• • -1\.'

.. ......
:.. )
~ ·1
o 2000

Figure 6. Sample location map, Sulphurets area.

APPLICATION OF SPATIAL FACfOR ANALYSIS

Southern Vancouver Island

SPF AC was applied to the 10 selected elements for the four

orientations and at lag distances from 10 to 40 km at 5 km intervals using
the quadratic, exponential, Gaussian, and mixed models. This resulted in
96 spatial factor results for subsequent interpretation. From these analyses
only the mixed model results are presented here.
246 GRUNSKY, CHENG, AND AGTERBERG

The results of the analysis indicated that the significance and

relationships of the elements change as a function of lag distance and
orientation. As discussed in Grunsky and Agterberg (1988), the squared
multiple correlation coefficient (R2) provides a measure of significance of
each variable. Values of R2 can be calculated for the variables over the
transition matrix V and for each of the spectral components Vi' Ideally,
the values of R2 should not exceed a value of 1. Figure 7 shows the
values of R2 for the transition matrix V over the 7 lag distances and 4
orientations. The figure shows that as the lag distance increases, there is
generally, a corresponding decrease in the value of R2. Of particular
interest is the change in relative significance of variables as a function of
orientation.

[MIxed Model] [E-W]

MnF. U As V C,

[NE-SW]

~r------------------------------'

[ -S]

[NW-SE]

Cu PI> N; Co Mn F. U As V Cr

Figure 7. Multiple R2 correlation coefficients for mixed model SPFAC. Each line
represents R2 coefficients for lag distance (in kilometers) indicated at top of figure.
Note that for N-S orientation, Ne, Co, and Cr exceed values of 1.0. This suggests that
collinearity may interfere in determination of meaningful results.
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 247

Comparison of the R2 coefficients in Figure 7 illustrates that as

orientation changes, the relative significance of the variables changes. In
the E-W and NW -SE directions the relative significance of the elements
are closer to each other than in the N-S and NE-SW directions. In the
NW-SE and E-W directions Cu, Ni, Co, Mn, Fe, V, and Cr share a
similar relative significance. In the N-S and NE-SW directions, this is not
the situation. In the N-S direction Ni, Co, and Cr have significantly
greater R2 values than the other elements and in the NE-SW direction Cu,
Ni, and V are substantially greater than the other elements. The
associations of these elements reflect the orientations of various
lithologies with geochemical responses in preferred directions.
The spectral components Vi, explain which elements are
associated in a manner similar to principal components analysis. The
components generally are fewer than the number of the elements. In this
study the number of components seldom exceeded three. An examination
of the spectral components as a function of lag distance and orientation
indicate that several geochemical associations with defined spatial
structures can be distinguished.
The geochemical sampling pattern over the area does not reflect
adequately the distribution of the lithologies that exist on the geological
map. Thus, the interpretation of the geochemical patterns must be made
with the knowledge that the sampling does not represent the complete
geological picture. The results of the SPF AC indicated that as lag
distance and orientation change, specific associations of elements were
obtained. Not all of the element associations derived from the analysis
can be interpreted in a geological context. This is most probably the
result of the nature of the moss mat sediment that may be a mix of one
or more lithologies, making interpretation difficult as well as an
insufficient sampling density to adequately represent the spatial extent of
the underlying lithologies. An evaluation of the most significant
component for each of the results for the 7 lag distances and 4
orientations provided the following interpretations.
From the application of the spatial factor methodology several
patterns and elemental associations have been identified in addition to the
two examples as described. These associations generally are described for
each orientation as follows.
As shown in Figure 7, from the R2 coefficients, the dominant
elements for the E-W direction are Mn, Fe, As, Cr, V, V, and Ni.
Generally, as lag distance increases, the overall significance of the
248 GRUNSKY, CHENG, AND AGTERBERG

variables decreases however the relative significance of the variables

remains approximately the same. Examination of the most significant
component for the 7 lag distances indicated that the dominant lithologies
highlighted by this orientation are the felsic intrusive rocks of the Island
Plutonic Suite and the West Coast Crystalline Complex that contains a
mix of felsic to mafic plutonic rocks with varying amounts of assimilated
supracrustal material. At the lag distance of 40 km, the association does
not seem to be present. This most likely reflects the spatial extent of
these rocks in terms of the samples used to derive the auto- cross-
correlation coefficients.
In the NE-SW direction, Cu, Ni, Cr, and As are the dominant
elements for all of the lag distances with the exception of the 40 km
distance. Typically the plotted scores of these components highlighted
areas of mafic provenance including the Karmutsen, Bonanza, and Sicker
volcanics as well as sediments with mafic components such as the Leech
River Complex contammg metavolcanics and mafic derived
metasediments, and parts of the West Coast Crystalline Complex
intrusions. The elements usually associated with these components are Ni,
As, and U. At the 40 km lag distance the dominance of V highlights the
distinction between the Leech River Complex and the Nanaimo Group
sediments with Co, Fe, V-bearing mafic volcanic rocks. Figure 8A
graphically displays the R2 coefficients for the dominant second
component at the 25 km lag distance. Nickel, Cr, and U are the most
significant elements. Figure 8B shows the corresponding amplitude
vectors. Nickel and Cr contribute to the positive scores and U, V, Cu, and
Pb contribute to the negative scores. Figure 9 shows a plot of interpolated
scores for this component. Positive scores of the samples outline the
Leech River metasedimentary rocks, and the Bonanza and Sicker
volcanics. The negative scores outline the Westcoast Crystalline Complex
and portions of the Metchosin Igneous Complex.
From Figure 7, Ni, Co, Cr, and Cu are shown to be the most
significant elements for the N-S direction. A comparison of scores for the
7 lag distances shows that at shorter lag distances ( < 30 km) mafic
volcanics of the Karmutsen and Sicker volcanic assemblages, and mafic
derived metasediments of the Leech River Complex are highlighted. For
lag distances greater than 35 km, the West Coast Crystalline Complex is
highlighted.
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 249

[Mixed Model] 25 km NE-SW

o .
o
Cu
- _

Pb
:.A
. . . . . . ..
£:1 C~ JL.:.::::::~..:..·..~~
C L ..

Ni Co Mn Fe U As V Cr
NE-SW

f-
:g 1 c~ ... C..l~ .......................
<y ~
---------_
~
................... =. :-><.."""
..................
.....
;rS 1
:g ,...........,./ -.- ~---- ----.,......-..,
- .. '
~ -".,' ", .
B
. Cu Pb Ni Co Mn Fe U As V Cr

Figure 8. A, Multiple R2 correlation coefficients for mixed model SPFAC in NE-SW

orientation at lag 25 km. For dominant second component, Ni is most significant
element for this orientation and lag distance. Components I and 3 are insignificant
relative to second component. B, Amplitude vector loading for mixed model SPFAC
in NE-SW orientation at lag 25 km. Loadings indicate that positive scores are
associated with samples containing Ni and Cr.

For the NW-SE direction, Fe, Co, Cu, Ni, V, and Cr are the most
significant elements (Fig. 7). The components for this orientation and the
7 lag distances contrast the mafic volcanic and igneous rocks from the
sedimentary and felsic plutonic suite. Distinctions between the various
volcanic rocks also is highlighted at the 20 km lag distance, whereby Mn-
Fe enriched volcanics are distinguished from Ni-Cr enriched volcanics.
The NW-SE direction is the dominant direction for many of the volcanic
lithologies in the area. This also is demonstrated in Figure 7 where most
of the elements associated with the ~fic volcanics (Sicker, Karmutsen,
and Bonanza Groups) have similar R values. These lithologies also are
dominated by Ni, Co, Cr, Cu, Fe, and V. There are 5 components
associated "'4th this lag distance and orientation. Figures lOA and lOB
show the R and amplitude vectors for the results at the 20 km lag
distance. Figure lOA shows that Co, Cu, Fe, and V are the dominant
elements and Figure lOB shows that the negative loadings are associated
with the elements of mafic associations (Cu, Co, Fe, and V). The first
component describes the dominant NW-SE trends of the Karmutsen,
Bonanza, and Sicker Group volcanics; the slight positive loading of Pb
is displayed in Figure 11 where positive scores highlight areas of Pb
 N
VI
o
[NE-SW] 25 km Component 2

. .- l:-.
N
..,j-
L() N
,-..
0
0 0
0 ..,j-
0 L()

x
E ro
r0
J
L() ~
Ci'
c
..c
T ~
..- I.D
Score 00
.... r0
0
z L()

- 2 -1 a 2 3
r~
..,j-
50 km
J
r0
lO
;
9
30 35 40 45 50
Eosting (m x 10000)
~
>
Figure 9. Contour map of first component for NE-SW direction at lag 25 km. For this orientation positive scores ~
associated with Ni and Cr enriched samples outline Leech River metasedimentary complex, Bonanza Group ~
volcanics and parts of Sicker Group volcanics are outlined. Samples with negative scores outline portions of
Westcoast Crystalline Complex and parts of Metchosin Igneous Complex. ~
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 251

[Mixed Model] 20 km NW-SE

Cu Pb Ni Co Mn Fe As v Cr
NW-SE
~ c~ C.2 .... 5~~ _CA - C 10.-

~~.~.'.4;:·~ ~;d~~c
o B
Cu Pb Ni Co Mn Fe u As v Cr

Figure 10. A, Multiple R~ correlation coefficients for mixed model SPFAC in NW-SE
orientation at lag 20 lan. For this lag and orientation there are 5 components and at least
three of them indicate significant contribution from at least one element. First
component indicates that Co, Fe, and Cu are most significant elements at this lag and
orientation. Third component shows that Mn, Ni, and Cu are dominant elements and
second component indicates that Mn and Ni are significant. B, Amplitude vector
loadings for mixed model SPFAC in NW-SE orientation at lag 20 kIn. Note varying
associations of As, Co, and V which describe differences between volcanic assemblages,
metasedimentary assemblages and plutonic complexes.

enrichment (felsic intrusives) and the negative scores outline the volcanics
of the Karmutsen and Sicker Groups and the Metchosin Igneous
Complex. The second component outlines associations, for positive
loadings of elements that correspond with the Leech River metasediments,
and the Westcoast Crystalline Complex. Negative loadings are associated
with felsic igneous rocks. The third components represents a unique
association of Ni, Co, Cr, Cu Mn, V, Fe, and As which describes most
of the volcanic rocks and mafic derived sediments in the Leech River
Complex and Bonanza Group.

APPLICATION OF SPATIAL FAcroR ANALYSIS IN TIlE

SULPHURETS AREA

Because of anisotropy of the spatial distribution of Au and Au-

associated elements in the study area, auto- cross-correlation coefficients
were estimated for different directions (E-W, NE-SW, NW-SE, and N-S)
with lags extending to lOOOm. Because of relatively large differences in
 N
VI
[NW-SE] 20 km Component N

N
....t-
11) N

0
-----
0 0
0 ....t-
O 11)

x
E ~ J
11)
CJ'I ~
C
L
.....
I-
Score ~
00
0 e .:;-;:;.
·t
z I I
~j -6 -2 a 2 4 6 v~
v 50 km
n
11)
~
vQ

30 35 40 45 50
Eosting (m x 10000)
~
>
Q
Figure 11. Contour map of first component for NW-SE direction at lag 20 km. Components outline element
associations that distinguish major volcanic groups; mafic derived metasedimentary assemblages and igneous
~
complexes.
~
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 253

elevation, true distances between sample locations were computed from

digital elevation data and these were used instead of their horizontal
projections. The unit of lag was set equal to 12.5 meters with a tolerance
of 12.5m. The search angles were OO±90° (isotropic case), OO±15° (E-W),
45°±15° (NE-SW), 90 o±15° (NS), and 135°±15° (NW-SE). Before
computation of the auto- cross-correlation coefficients, the trace-element
determinations were log-transformed and the oxides subjected to
Aitchison's log-contrast transformation. Three models (quadratic,
exponential, and Gaussian) were used for fitting auto- and cross-
correlation functions. The elements and oxides have auto- cross-
correlation functions that differ for different orientations as shown for Au
in Figure 12 where the exponential model is used. In general, the spatial
correlation remains relatively strong for longer distances in the trend
direction than across trends. For example, Au, Ag, As, Sb, and Si0 2
show relatively strong spatial correlation in the SN to NW -SE directions,
Mo and W in the E-W to NE-SW directions, and K 20 and Cu in the SN
to NE-SW directions.
LogAu [Exponential Model] [Sulphurets]

OlrectJon
- - - N·S

NE·SW

E·W
NW·SE

0.0 0.2 0.4 0.6 0.8 1.0

Lag (km)

Figure 12. Exponential correlogram functions for log Au in four different directions.

Spatial factors were calculated for groups of 10 trace elements

(Zn, As, Mo, Ag, Cd, Sb, Ba, W, Au, and Cu) and 11 oxides (Si0 2,
Ti0 2, A1 20 3, total iron, MnO, MgO, CaO, Na20, K 20, CO 2, and
P 20 5). These calculations were carried out for the isotropic case and the
four different directions (E-W, NE-SW, NS, and NW-SE) with lag
254 GRUNSKY, CHENG, AND AGTERBERG

distances increasing from 100m to 900m in steps of 100m. For each

orientation and lag distance two to four major spatial factors could be
distinguished. Examples are given in Figures 13 and 14 and Tables 2 and
3. Figure 13 is a contour map of the scores for oxides in the NW -SE
direction) with lag distance equal to 400m. The coefficients of the
corresponding amplitude vector are given in column 4 (A-NW400) of
Table 2. Silica and (negative) sodium have the largest amplitudes in this
situation. These two oxides also have the largest squared correlation
coefficients (R2) (column 5, R-NW400, Table 2). The pattern of Figure 13
outlines primary areas of silicic alteration and shows spatial correlation
with the occurrences of quartz veins and vein stockwork systems (see Fig.
2). After Si and -Na in Table 2 (NW400), -Mg and CO 2 have the largest
amplitudes and correlation coefficients for the pattern of Figure 13.

Figure 13. Contour map of third spatial factor, 11 oxides, NW-SE direction, 400m lag,
delineates silicic alteration.

To illustrate the dependence of the preceding results on direction

and distance, amplitudes and squared correlation coefficients are shown
for similar silicification factors with 400m lag in the north-south direction
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 255

(N400 in Table 2) and 600m lag in the NW-SE direction (NW600). This
indicates that the negative loading of Mg increases in magnitude with
clockwise rotation toward the north whereas the relative contribution of
CO 2 increases with longer distances in the NW -SE direction.

Table 2. Amplitude (A-NW400) and squared correlation (R-NW400)

coefficients for spatial factor of Figure 12 in comparison with coefficients
for similar factors with 400m lag in NS direction (columns 2 and 3) and
600m lag in NW-SE direction (last two columns). Largest values are
highlighted.
A-N400 R-N400 A-NW400 R-NW400 A-NW600 R-NW600
Si 1.150 0.677 1.089 0.035 0.782 0.016
Ti 0.201 0.063 -.239 0.003 -.244 0.003
A1 -.020 0.001 0.139 0.002 -.365 0.009
Fe 0.043 0.001 0.311 0.006 0.031 0.000
Mn 0.381 0.279 0.344 0.012 0.570 0.030
Mg -.918 0.901 -.602 0.017 -.654 0.018
Ca -.238 0.034 -.067 0.000 0.395 0.007
Na -.343 0.106 -.918 0.047 -.918 0.042
K -.076 0.007 0.038 0.000 -.275 0.004
CO? 0.157 0.020 0.588 0.024 1. 053 0.068
P -.122 0.323 - .115 0.001 -.361 0.011

Figure 14. Contour map of third spatial factor, 10 trace elements, NS direction, 400m
lag, delineates gold mineralization.
256 GRUNSKY, CHENG, AND AGTERBERG

Table 3. Amplitude (A-NW400) and squared correlation (R-NW400)

coefficients for spatial factor of Figure 13 in comparison with coefficients
for similar actors with 400m lag in NW-SE direction (columns 2 and 3)
and 600m lag in NS direction (last two columns). Largest values are
highlighted.
A-NW400 R-NW400 A-N400 R-N400 A-N600 R-N600
Zn -.040 0.000 0.035 0.000 0.369 0.004
As 1. 015 0.100 0.557 0.032 0.799 0.014
Mo 0.537 0.030 0.169 0.003 0.479 0.004
Ag 0.240 0.004 0.076 0.000 -.098 0.000
Cd 0.876 0.073 0.352 0.012 0.949 0.018
Sb 0.788 0.051 0.308 0.010 0.093 0.000
Sa -.019 0.000 0.110 0.005 0.200 0.003
W -.112 0.003 -.073 0.001 -.188 0.001
Au 1.452 0.235 0.521 0.034 1. 467 0.058
Cu -.449 0.031 -.181 0.003 -.080 0.000

A second example for trace elements is illustrated in Figure 14.

It shows the spatial factor for regional Au mineralization with lag of
400m in the NS direction. This pattern is correlated positively with As
and Au (Table 3). The same type of spatial factor could be distinguished
for 400m lag in the NW -SE direction and 600m lag in the NS direction.
In the latter two situations, Cd is positively associated with Au and As.
The examples of Figures 13 and 14 illustrate that SPFAC is
helpful for determining associations between trace elements and oxides.
In this respect its role is similar to that of factor analysis. However,
SPFAC has the added advantage of allowing these associations to change
as a function of distance as well as direction.

CONCLUDING REMARKS

Results of the spatial factor method (SPF AC) applied to Southern

Vancouver Island show that for the scale of sampling, multielement
geochemical patterns are associated with a particular spatial process
related to underlying lithological variation, or variation in stream
sediment composition. The interpretation is dependent on the spatial
continuity of the elements and a suitable sampling density to ensure
minimum nugget effects. The mafic igneous volcanic and intrusive
lithologies, felsic intrusive rocks, and metasediments have characteristic
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 257

geochemical signatures that are highlighted at various orientations and lag

distances. Generally, lithological patterns that are indicated by the spatial
factor scores are consistent in shape for similar orientations, however the
extent of the patterns change with changing lag distance. Lithologies such
as the Leech River metasediments show to their maximum spatial extent
at the shorter lag intervals. As the lag distance increases the pattern
decreases due to the decreasing value of auto- cross-correlations.
The analysis of the Sulphurets area shows that specific zones of
alteration can be distinguished readily using the spatial factor technique.
Specific geochemical processes that have preferred directions can be
isolated using this methodology and can assist in mapping favorable
horizons.
The application of SPF AC provides more detail on the
multielement relationships oflithologies than can be described using more
conventional methods. In the application of the technique to a regional
sampling program, multielement associations of regional trends can be
factored providing that they have been adequately sampled. In areas of
more detailed sampling, as shown in the Sulphurets area, significant
geochemical trends that are associated with alteration and mineralization
can be enhanced using the spatial factor technique.

ACKNOWLEDGMENTS

This work has been supported by the Geological Survey Branch,

British Columbia Ministry of Energy Mines and Petroleum Resources. We
are grateful for the assistance of Nick Massey and Steve Sibbick of the
Geological Survey Branch, for information about the geology and
geochemistry of the Southern Vancouver Island area.

REFERENCES

Agterberg, F.P., 1966, The use of multivariate Markov schemes In

petrology: Jour. Geology, v. 74, no. 5, pt. 2, p. 764-785.
Agterberg, F.P., 1970, Autocorrelation functions in geology, in
Merriam, D.F., ed., Geostatistics: Plenum Publ. Co., New York,
p. 113-142.
Agterberg, F.P., 1974, Geomathematics: Elsevier, Amsterdam, 596 p.
258 GRUNSKY, CHENG, AND AGTERBERG

Aitchison, 1, 1986, The statistical analysis of compositional data:

Chapman and Hall, New York, 416p.
Alldrick, D.l, 1987, Geology and mineral deposits of the Salmon River
Valley, Stewart area, NTS 104A and 104B: British Columbia
Ministry of Energy, Mines and Petroleum Resources, Geology
Survey Branch, Open File Map 1987-22, 1:50,000.
Alldrick, D.l, and Britton, 1M., 1988, Geology and mineral deposits of
the Sulphurets area: British Columbia Ministry of Energy, Mines
and Petroleum Resources, Open File Map, 1988-4, 1:50,000.
Ballantyne, S.B., 1990, Geochemistry of Sulphurets area, British
Columbia (abst.): Program with Abstracts, Geol. Survey Canada,
Minerals Colloquium, 1 p.
Ballantyne, S.B., Shives, R.B.K., Harris, D.C., Plouffe, A., Judge, A., and
Pilon, lA., 1992, An integrated approach and model for the
discovery of blind Cu-Au porphyry systems: Geol. Survey
Canada, Minerals Colloquium, poster.
Barcelo, C., Pawl owsky , v., and Grunsky, E.C., 1994, Outliers in
compositional data: a first approach, in Proc. Intern. Assoc. Math.
Geology, Ann. Meeting (Mount Tremblant, Quebec, Canada), p.
21-28.
Bellehumeur C., Marcotte. D., and Jebrak, M., 1994, Multielement
relationships and spatial structures of regional geochemical data
from stream sediments, southwestern Quebec, Canada: Jour.
Geochemical Exploration, v. 51, no. 1, p. 11-35.
Bourgault, G., and Marcotte, D., 1991, Multivariable variogram and its
application to the linear model of coregionalization: Math.
Geology, v. 23, no. 7, p. 899-928.
Campbell, N.A., 1986,. A general introduction to a suite of multivariate
programs: CSIRO Division of Mathematics and Statistics, Perth,
Australia, unpubl. rept., unpaginated.
Cheng, Q., Agterberg, F.P., and Ballantyne, S.B., 1994, The separation
of geochemical anomalies from background by fractal methods:
Jour. Geochemical Exploration, v.51, no. 2, p.109-130.
Chung, C.F., 1985, Statistical treatment of geochemical data with
observations below the detection limit, in Current Research, Part
B: Geol. Survey Canada Paper 85-1B, p. 141-150.
Chung, C.F., 1988, Statistical analysis of truncated data in geosciences:
Science de la Terre, Inform. Geol. (Nancy), v. 27, p. 157-180.
APPLICATIONS OF SPATIAL FACTOR ANALYSIS 259

Dempster, A.P., Laird, N.M., and Rubin, D.B., 1977, Maximum

likelihood from incomplete data via the EM algorithm: Jour.Roy.
Statist. Society, Ser. B, v. 39, p. 1-38.
Grunsky, E.C., and Agterberg, F.P., 1988, Spatial and multivariate
analysis of geochemical data from metavolcanic rocks in the Ben
Nevis Area, Ontario: Math. Geology, v. 20, no. 7, p. 825-861.
Grunsky, E.C., 1989, Spatial Factor Analysis: a technique to assess the
spatial relationships of multivariate data, in Agterberg, F.P, and
Bonham-Carter, G.F. eds., Statistical applications in the earth
sciences: Geol. Survey Canada Paper 89-9, p. 329-347.
Grunsky, E.C., and Agterberg, F.P., 1989, The application of spatial itr
analysis to unconditional simulations with implications for mineral
exploration, in Proc. Applications of Computers and Operations
Research in the Mineral Industry (APCOM) 1989 Chapter 20, p.
194-208.
Grunsky, E.C., and Agterberg, F.P., 1991a, FUNCORR, a FORTRAN 77
program for computing multivariate spatial autocorrelation:
Computers & Geosciences, v. 17, no. 1, p. 115-131.
Grunsky, E.C., and Agterberg, F.P., 1991b, SPFAC: a FORTRAN 77
program for spatial multivariate analysis: Computers &
Geosciences, v. 17, no. 1, p. 133-160.
Grunsky, E.C., and Agterberg, F.P., 1992, Spatial relationships of
multivariate data: Math. Geology, v. 24, no. 6, p. 731-758.
Howarth, R.J., and Sin ding-Larsen, R., 1983, Multivariate analysis,
(Chapter 6), in Howarth, R.I., ed., Statistics and data analysis in
geochemical prospecting, v. 2, in Govett, G.J.S., ed., Handbook of
exploration geochemistry: Elsevier, Amsterdam, 437 p.
Howarth, R.J., and Earle, S.A.M., 1979, Application of a generalized
power transformation to geochemical data: Math. Geology v. 11,
no. 1, p. 45-62.
INTERA TYDAC Technologies Inc. 1993, Reference manual of SPANS
Version 5.2
Joumel, A.G., and Huijbregts, C.J., 1978, Mining geostatistics: Academic
Press, London, 600 p.
Kirkham, R.y., 1990, Sulphurets area, British Columbia: Geology (abst.):
Geol. Survey Canada, 1990.
Kirkham, R.V., Ballantyne, S.B., and Harris, D.C., 1989, Sulphurets area,
British Columbia: geology, geochemistry, and mineralogy (abst.):
Geol. Survey, Canada, Minerals Colloquium, 1989.
260 GRUNSKY, CHENG, AND AGTERBERG

Kirkham, RV., Ballantyne, S.B., and Harris, D.C., 1990, Sulphurets,

British Columbia: Geology, geochemistry, and mineralogy of a
deformed porphyry copper, molybdenum and precious metal
system (abst.): Geol. Survey Canada, Minerals Colloquium, 1990.
Kirkham, RV., Ballantyne, S.B., Harris, D.C., Henderson, IR,
Henderson, M.N., and Wright, T.O., 1992, Lower Jurassic
Sulphurets porphyry Cu-Au system, northwest British Columbia.
(abst.): Geol. Surv. Canada, Minerals Colloquium, 1992.
LeMaitre, R W., 1982, Numerical petrology, statistical interpretation of
geochemical data: Elsevier, New York, 281.p.
Matysek, P.F., and Day, S.1., 1987, Geochemical orientation surveys:
northern Vancouver Island, fieldwork and preliminary results, in
Geological fieldwork, a summary of field activities and current
research: British Columbia Ministry of Energy Mines and
Petroleum Resources Paper 1988-1, p.493-502.
Matysek, P.F., Gravel, I.L., Jackaman, W., and Feulgen, S., 1990, British
Columbia regional geochemical survey, stream sediment and water
geochemical data, Victoria/Cape Flattery - NTS 92B / 92C: British
Columbia Ministry of Energy Mines and Petroleum Resources
RGS Report 24, Geol. Survey Canada, Open File Rept. 2182, 1
Map Booklet, 1 Data Booklet
Myers, D.E., 1982, Matrix formulation of co-kriging: Math. Geology: v.
14, no., 3, p. 249-257.
Myers, D.E., 1988, Some new aspects of multivariate analysis, in Fabbri,
A.G., Chung, C.F., and Sinding-Larsen, R, eds., Statistical
treatments for estimation of mineral and energy resources: Proc.
NATO Conf (11 Ciocco, Italy), Reidel, Dordrecht, p. 669-687.
Press, W.H., Flannery, B.P., Teukolsky, S.A., and Vetterling, W.T., 1986,
Numerical recipes, the art of scientific computing: Cambridge
Univ. Press, Cambridge, 818p.
Royer, I.-I., 1988, Geochemical data analysis, Fabbri, A.G., Chung, c.F.,
and Sinding-Larsen, R, eds., in Statistical treatments for
estimation of mineral and energy resources: Proc. NATO Conf. (11
Ciocco, Italy), Reidel, Dordrecht , p. 89-112.
Sanford, RF, Pierson, C.T., and Crovelli, RA., 1993, An objective
replacement method for censored geochemical data: Math.
Geology, v.25, no. 1, p. 59-80.
APPLICATIONS OF SPATIAL F ACTOR ANALYSIS 261

Statistical Sciences Inc., 1993, SPLUS Reference Manuals, Version 3.2:

Seattle, Washington.
Switzer, P., and Green, AA, 1984, MinIMax autocorrelation factors for
multivariate spatial imagery: Stanford University, Dept. Statistics,
Tech. Rept. No.6, 14 p.
Wackemagel, R., 1988, Geostatistical techniques for interpreting
multivariate spatial information, in Fabbri, AG., Chung, C.F., and
Sinding-Larsen, R, eds., Statistical treatments for estimation of
mineral and energy resources: Proc. NATO Conf. (11 Ciocco,
Italy), Reidel, Dordrecht, p. 393-409.
 GEOSTATISTICAL SOLUTION FOR THE
CLASSIFICA TION PROBLEM WITH AN
APPLICATION TO OIL PROSPECTING

Jan Harff
Institutfur Ostseeforschung Warnemunde, Warnemunde, Germany

Ricardo A. Olea, John C. Davis, and Geoffrey C. Bohling

Kansas Geological Survey, Lawrence, Kansas, USA

ABSTRACT

Mathematically modeled thematic maps can relate productivity to geol-

ogy and display the probability of occurrence of mineral deposits.
Stochastic simulation, a geostatistical technique that regards a surface as an
outcome of a random function, is used for nonparametric characterization of
the uncertainty involved in the estimation of the Mahalanobis' distances at
un sampled locations. This avoids overly smooth maps and precludes the
need for distributional assumptions to assess misclassification probabilities,
problems encountered in earlier studies that described the relations between
multivariate observations and rock type by distances and treated them as
univariate variables.
In the stochastic approach, discriminant analysis is used to simultane-
ously partition the multidimensional space occupied by the observations and
select the most informative geological variables. Clusters are identified by
regarding each observation as a point in space with as many dimensions as
variables. Mahalanobis' distances to the cluster centroids at all geographic
locations are estimated using sequential Gaussian simulation. By generating
multiple realizations of the distances, the most likely distances and their es-
timation uncertainties can be assessed. Finally, using Bayesian relations,

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press. New York, 1996 263
264 HARFF AND OTHERS

probabilities are adjusted for misclassification and sampling bias. The tech-
nique is used to demonstrate that results of stochastic simulation agree well
with actual distributions of fields in an oil-producing area in western
Kansas, USA; the map of probability of drilling a producing well corre-
sponds with previous results using kriging to map Mahalanobis' distances.

INTRODUCTION

The interpolation between sampling sites irregularly distributed in an

area of interest is a spatial estimation problem in the construction of geologi-
cal maps in mineral exploration and in marine geology. In these situations
the problem reduces to extracting the maximum information from a minimal
number of observations by preparing maps covering the entire area of in-
vestigation. Lately mathematical models and computers have become widely
used for the construction of such geological maps.
Geostatistics, based on the theory of regionalized variables, has pro-
vided valuable tools for handling univariate and multivariate geological vari-
ables (Journel and Huijbregts, 1978; Myers, 1982). Instead of interpolating
the multivariate geological feature fields, Harff and Davis (1990) proposed
to take advantage of the relations between geological features. This is done
by describing the relations between multivariate observations and types or
classes of rocks or rock sequences using distance measures which are in-
versely proportional to the similarity to these types. The problem is reduced
to the handling of these distances as univariate variables. Regionalized clas-
sification involves a unification of the theory of classification of geological
objects developed in the former Soviet Union (Rodionov, 1981) with the
theory of regionalized variables (Matheron, 1970). Harff, Davis, and Olea
(1992) used kriging for the interpolation of the distances. This method,
however, produces overly smooth maps and requires that distributional as-
sumptions be made in order to assess misclassification probabilities.
The main purpose of this paper is to present the use of stochastic simu-
lation for a nonparametric characterization of the uncertainty involved in the
estimation of the Mahalanobis' distances at unsampled locations (Journel,
1989; Deutsch and Journel, 1992). In addition to the construction of maps,
we address the problem of data interpretation related to a special geological
task. In Harff, Davis, and Olea (1992) mapping was done to assess
exploration risk. Although the present case study focuses on hydrocarbon
GEOSTATISTICAL CLASSIFICATION 265

exploration, the methodology is completely general and can be applied

equally well to other geological mapping problems and interpretation of
geological data.

STOCHASTIC SIMULATION FOR NONPARAMETRIC

CHARACTERIZATION OF UNCERTAINTY

Thematic maps conveniently display the likelihood of occurrence of

mineral deposits by relating a productivity variable P with information
provided by geological variables G. The prediction of P by G is based on
probabilistic techniques and the existence of a sample subset comprising
measurements of both P and G. Regression is not adequate because the
correlation between P and G typically is weak. Instead, if each observation
is regarded as a point in G-dimensional space, clusters can be identified that
have a one-to-one correspondence with a partition of P.
The first step in our procedure is a systematic selection of the most
informative variables in G concurrently with optimal partitioning of both P
and the multidimensional space defined by the retained variables in G. The
search is based on discriminant analysis and in a first stage ignores those
observations without production information. The best possible partition of
P and G is one in which all the observations in one cluster in G belong to
one and only one partition of P.
Once the partitions are decided, the correlation matrix and mean vector
for each cluster in G are calculated. At this point those observations without
production information enter the analysis. The correlation matrix and mean
vector are used to calculate the Mahalanobis' distances from all observations
in G to the centroids of the clusters. Observations without production are
assigned to the closest G class and the calculation of the correlation matrix
and mean vector is repeated using this time all observations in G. The
calibration stage of the procedure ends by recording, for each G cluster, the
proportion of observations that belong to each P partition.
The next stage is estimation of Mahalanobis' distances at all geographic
locations. The estimation is done by stochastic simulation, a geostatistical
technique that regards a surface as an outcome of a random function. By
generating multiple realizations of the Mahalanobis' distances, the most
likely distances and their estimation uncertainties can be assessed. At any
location, the likelihood of membership in any cluster G is given by the
266 HARFF AND OTHERS

proportion of times the simulated distance to cluster G is smaller than the

distance to the alternative cluster.
Using Bayesian relationships, the final step is adjustment of the prob-
ability of belonging to any P class for misclassification and for sampling
bias.

A CASE STUDY

The methodology is easier to understand if each step is illustrated with

an example. We will use data on petroleum geology in the western part of
the State of Kansas, USA. The study relates to oil deposits in the Lansing
and Kansas City Groups of the Pennsylvanian System. These consist of
cyclic marine limestones with interbedded dark gray, marine shales. A geo-
logical overview is given by Merriam (1963). The most prolific deposits are
along the Central Kansas Uplift (Fig. 1), where production has reached a
mature stage. This tectonic unit has influenced the entire sedimentary pro-
cess during Paleozoic time. To the west of the Central Kansas Uplift lies the
Hugoton Embayment, a part of the Anadarko Basin. For more information
about development of hydrocarbon exploration and exploitation in Kansas,
see Watney (1984) and Watney and others (1989).

Figure 1. Oil fields in Kansas and study area (nonshaded).

GEOSTATISTICAL CLASSIFICATION 267

For this study, we employed a data set containing stratigraphic, struc-

tural, and petrophysical information from 1046 wells penetrating the
Lansing and Kansas City Groups. Because productivity was available as an
indicator that does not give the amount produced, we were forced to parti-
tion the productivity variable P into only two classes: productive and
nonproductive. Ninety-five of the wells were classified as oil wells (class
"oil") and 753 were classified as dry holes (class II dry "). The balance are
wells with unknown production status. Figure 2 shows the area of
investigation with the locations of oil, dry, and unclassified wells. These
data can be regarded as a censored sampling. The purpose of the case
study is to reproduce the more detailed portrait of the reality shown in
Figure 1 by using only the censored data. The successful completion of
the exercise should encourage readers to use the method when a sparse
sample is all that is available to characterize a prospective area.
Each observation has information about 16 variables. Remember that
we want to have as many partitions in the geological variable space as we
have production classes. Thus in this case study we are interested in a
partition into two classes. After running a discriminant analysis, we
concluded that ten of the variables were either redundant or irrelevant,
reducing the number of prediction variables of interest to six.
Although a six-dimensional space is smaller than a 16-dimensional
space, the dimensions are still too numerous to permit graphical display of
six-dimensional clusters. An illustration such as Figure 3 is an alternative
to the display of the six-dimensional space; it shows the distances to the
two centroids after an anamorphosis that forces each of the sets of
distances to follow a normal distribution with mean zero and a variance of
one. Even though the relative locations of the points in the figure are
different from those in the six-dimensional space, the display is instructive
because it is still possible to keep the points from each partition separated.
The boundary between the two partitions is the S-shaped curve in the
middle of the display. The curvature of the boundary is the result of
anamorphosis and the small displacement is an effect of differences in the
covariance matrices of the two clusters.
268 HARFF AND OTHERS

.. ..... .. ..... ·· . · . ..... . .

. ,a. •
•• 0 0 ~:o . ,a.••• ~ '!/
~ ~

.. . . .
0 0

.' .· . ..
.' 0

. .. .
DO

o. 0.+.
4' 0'"
.,a.,a..,

..'" .
~

. '" · .
+ .~
• ~ . 4 -
• OJ.
•'f" .0. gO
. ...
0
0
0 • -II>.

. 4
0
o Q ... 0 a o!.o 0 o"~+ 0
a

·· · . · . -
0
0:0.°"'°00 : .. 0 0 o 00+0 .~'b
0
o 000 0 0 I.,a. J +! ++ 'I> ° 8
0 ,,0
~ 0 4. °
" ,a. 0

· ~~M. oOo.c CI . . . . Oo A ++o-t, 0

.+.· •
00
0
8 +' o. ° '.:. o·~ .. • g~ + ~ ...
.:t-
0+ '.'
~
0

oo:oo~ 1i4 : ~~ -
·.
'. Co 000 0 0 ••
g~ M-g ,;~ o A~
1.A. •
00 Cb 0 0 0 0 0
0 0 o •
o .& .& +
g."
q.P
0\
0 ... CI 000 0 00. 0
0

o• 0
0 0 0. D+: ~ ,0'· •
0 0°
4 0. 1+00."00 + o· ~. I- j 0 0 0 110 •
0
o 0 0 4- + 0 0 0 00

,
(I

0
~f, +Oo o~ o~ +Il;10

. ·
0 a oo.
I'"•
0.
to
0' " .0 0 00 1_.... 0 0
0
++ + + ~+ •• -
0 ocC
+
0. 0 •
8 00 .t
.+00
tOO
t
0:°
o\.ot ootP
0 0++
0 D·
+Go: g
00

°0
+ + +:
+
o0 : \ .& 0 0 +0 o. 00 00 ~"'t;.. 0 Cit 0 ~ 0
QOO } #°0+0
+ I.....
0
°l_+: ..+ '0

·
0 0
g ++ +. ~t -+

.. . ..
0 o 0
0 0 0 0
.0+ •
00 0.0
+Oo 0 0
+0 0
0
0

0 0 0
l-+ + i + +

.
o 0 0 , 0 Ooo.q..o. 0+ 0101 oCl. 0 -r- 0 .; +0

· ·.
0 oot:Po.+o+oooogoc tr&A~:J: .o 00+
° 1,- :f o 0+' + 0".. • } 0 ,\+ ,\ t~... 0

.
0
o~og qt o~o~~+""b + If' 0o+ 0° 0
0
0 0'" 080+/+0 0Q .. 0+ + 0 0 :0 0 00 •
~ .. +
°if +' 0 0 \0
o o-S +t
0 tOO 0 CI ". 0 a

J
Q
II 0 + 0 ~ 1-f 0 00

• +CI 0 0 .r+" .,.A.+ CI -fa 0

o
°
0
0
+ °
J
0 -0 0
0+ 0

.J

+ Unclassified
I 50 km I
,a. Produci 9 Well o Dry Hole

Figure 2. Map of drillhole locations.

Figure 3 shows the final partition obtained as a refinement of a first

partition, taking into account the 198 wells whose production status is not
known. After the first pass, all wells (including those with an unknown
production status) were assigned to the closest cluster, the locations of the
centroids were recalculated using all 1046 wells, and the wells were reas-
signed to the closest new centroids. The final count of oil, dry, and unclas-
sified wells on each of the two partitions in the geological variable space is
given in the lower part of Figure 3. Partition G2 contains 72% of the pro-
ducing wells, thus is the predictor for the target class "oil." By default,
partition ~-containing 63% of the dry wells-is the predictor partition for
the target class "dry." The 95 unclassified wells in G1 were considered as
dry wells and the 103 in G2 as oil wells. Figure 3 can be regarded as a
graphical display of the calibration stage.
GEOSTATISTICAL CLASSIFICATION 269

N
o
~

-1

-2

-3

AnDI

... Producing Well • Dry Hole + Unclassified

Por1ition G
1 2
P 27 68
D 474 279
U 95 103

cor coef (AnOI. An02) - -0.04

Figure 3. Observations plotted in anamorphosed distance space.

Table 1 contains the means and the covariance matrices for both clus-
ters in the six-dimensional geological space after recomputing the location of
the centroids of G1 and G2 using all 1046 observations. Changes in the
assignments, means, and covariances due to the changes in the centroids
obtained in the first pass were of minor importance.
270 HARFF AND OTHERS

Table 1. First and second moments.

Val"iahle Mean
G1 G2
HEEB, Surface elevation of the Heebner Shale, m -349.7 -353.2
K-BP, 1l1ickness base of the K-zone-top Pennsylvanian, m 106.2 41.08
I, Thickness of the I-zone, m 5.354 4.307
PorH, Thickness of porous carbonate in H-zone, m 0.7814 0.5804
PorI, Thickness of porous carbonate in H-zone, m 0.5912 0.3901
Ga], Maximum gamma-ray radiation in J-zone, API units 172.9 165.7

Total number of observations 596.0 450.0

Covariance matrix for cluster G1

HEEB K-BP I PorH PorI GaJ

HEED 9721.317 2323.5028 -85.2216 -19.2278 -8.8930 837.647
K-BP 2323.503 1037.2876 2.8061 -3.9420 -0.8282 234.769
I -85.222 2.8061 9.4949 0.3457 1.1737 30.839
PorH -19.228 -3.9420 0.3457 0.9378 0.1284 -1.422
PorI -8.893 -0.8282 1.1737 0.1284 0.9323 3.901
GaT 837.647 234.7694 30.8390 -1.4217 3.9011 3279.531

Covariance matrix for cluster G2

HEEB K-DP I PorH PorI GaJ

HEEB 1670.551 125.899 -21.70941 -9.15971 -4.61448 659.331
K-BP 125.899 738.311 14.32330 -4.13676 -3.55293 839.535
I -21.709 14.323 2.48585 0.02744 0.14306 17.223
PorH -9.160 -4.137 0.02744 0.66596 0.07759 -5.858
Pori -4.614 -3.553 0.14306 0.07759 0.40483 -1.535
GaT 659.331 839.535 17.22314 -5.85837 -1.53457 4916.712

The next stage is mapping the two sets of Mahalanobis' distances.

Because the univariate distribution of the distances is strongly skewed, a
nonnal score transfonnation is in order for more accurate geostatistical
estimation. As shown in Figure 3, there is no correlation between the
anamorphosed distances, indicating that a simpler estimation of each
anamorphosed distance separately is as accurate as the simultaneous
estimation of both variables. Here we resorted to geostatistical simulation to
explore the potential of the technique to assess the probability of mis-
classification. The appeal of simulation is the avoid~nce of distributional
GEOSTATISTICAL CLASSIFICATION 271

assumptions about the estimation errors that are required when using
universal kriging to map the anamorphosed distances (Harff, Davis, and
Olea, 1992). Structural analysis of the variables produced the semi-
variograms in Table 2 and Figure 4.

Table 2. Semivariogram models.

Variable Model

Exponential (9,0.26) + Exponential (435,0.74)

Exponential (6,0.16) + Exponential (315,0.84)

The tectonic pattern of the Central Kansas Uplift causes a drift in the
data along the NE-SW direction. The semivariograms are calculated" for the
NW-SE direction perpendicular to the trend. Using the sequential Gaussian
method (Deutsch and Journel, 1992), we simulated a set of equally probable
realizations of distances AnDl (x) and AnD2 (x). Figures 5 and 6 show the
first five realizations per anamorphosed distance out of a total of 101 real-
izations. Instead of a continuous coverage each realization is made from a
grid of 111 by 76 nodes. The seed used to simulate both sets of
anamorphosed distances is the same. For comparative purposes the figures
also include the result of an interpolation using universal kriging (Journel
and Huijbregts, 1978). We apologize for the black and white illustrations
and refer the reader to Harff and others (1993) for color renditions.
Figure 7 illustrates the calculation of the probability of membership in
partition G2 for a specific node in the lower-left corner of the study area,
after back-transforming all realizations to the original distance space. To
eliminate any possible effect of the simulation seed, the realizations were
compared with an offset of one simulation. The value of Dl (x) for the first
simulation and the value of D2 (x) for the second simulation provided the
first point for Figure 7, the value of Dl (x) for the second simulation was
paired with the value of D2(X) in the third simulation to provide the second
point in the crossplot, and so on until pairing Dl (x) from the 100th
simulation with D2 (x) from the 10 1st simulation for a total of 100 pairs.
272 HARFF AND OTHERS

The proportion of pairs for which d2 < dl + 3.84 provides the value of the
probability p(G2 (x)) at the lowerleft node, 24% in this example. The value
3.84 accounts for the difference in covariance matrices of the clusters used
in the calibration. By repeating this process for the remaining 8435 nodes,
we obtained a 111 by 76 grid for p( G2 (x)) which is displayed in Figure 8.
The NW-SE striking zone of elevated probabilities clearly outlines the
Central Kansas Uplift where most of the oil deposits of western Kansas are
located.

1.00 AnD1

0.80
..
E 0.60
'"
0,
.2

...
Cii
>
·E 0.40
"
(/J

0.20

0.00 ,
o. 40. 80. 120. 160.

Lag, kilometers

1.00 AnD2

0.80

E 0.60

..
~
01
.2
Cii
>
·E
"
(/J
0.40
..
0.20

0.00 , I , I

o. 40. 80. 120. 160.

Lag, kilometers

Figure 4. Semivariogram models.

GEOSTATISTICAL CLASSIFICATION 273

c D

F
3.
,.
IS

I~

lJ.ti(·

00
()fij.t

12

,.""
3."

Figure 5. Anamorphosed distance AnDl (x) to centroid of dry partition CI .

A, Universal kriging; and B t11rough F. realizations.
274 HARFF AND OTHERS

c D

Figure 6. Anamorphosed distance AnD2 (x) to centroid of oil partition G2 .

A, Universal kriging; and B tllrough F, realizations.
GEOSTATIS TICAL CLASSIFICATION 275

o ~~~~~--~~~~~~

o 40 80
Distance 1

Figure 7. Detennination of probability of belonging to oil partition for lower left node of
area of study.
276 HARFF AND OTHERS

90.0~
50.0
10.0

I 50 km I

Figure 8. Percentage probability of belonging to oil partition G2 .

To estimate the probability of encountering oil in a well, the fonnula of

the total probability can be used:

p( oil(x)) = p( c; (x))p(oilIG1 ) + p(G2 (x))p(oillG2 )

= p(oillGt ) + {p(oilIG 2 ) - p(oilIGt)}p( G2 (x))

which is a linear transfonnation of p( G2 (x)). The coefficients p( Dill Gj ) can

be estimated using Bayes' fonnula:

( '11 G ) _ p( oil)p( Gj Ioil)

POt j - p(oil)p(Gjloil) + (l-p(oil))p(Gjldry)

Table 3 contains the conditional probabilities p(Gjloil) and p(Gjldry),

which are computed directly as proportions in the tabulation in Figure 3.
Consequently, the calculation of a map with the probability of hitting oil
GEOSTATIS TICAL CLASSIFICATION 277

depends ultimately on a knowledge of the global probability of finding

production. In the absence of this information, a sensitivity analysis can be
performed. Figure 9 is a map of the probability of drilling a producing well
obtained using the value p( oil) = 0.42 taken from a publication by Watney
and others (1989). Note that Figure 9 is simply a rescaling of Figure 8.
An actual count of all wells reaching the Lansing-Kansas City in Barton
County gives a value of p(oil) =0.69. Barton, approximately 100 Km to
the northwest of the lower right corner of Figure 9, is the only county
entirely inside the area with p( oil) ~ 0.55. A comparison of Figures 8 and 9
with the map in Figure 1 shows that the general trend in the distribution of
hydrocarbon deposits is reflected by the current study. The deposits linked
with post-Mississippian-pre-Middle Pennsylvanian structures of the Central
Kansas Uplift are reasonably recognizable. The oil favorability in the
Hugoton Embayment in the central part in the region of investigation is
marked by elevated probabilities in this zone.

Table 3. Conditional probabilities.

Geologic partition
G1 G2
Production class
Oil 0.284 0.716
Dry 0.629 0.371

CONCLUSIONS

The study provides a contribution to computerized geological mapping

and interpretation of geological data. In order to make assumptions about
the oil favorability of an entire area of investigation it is necessary: (1) to
study empirically the relationship between oil-bearing (target variable) and
easily measurable geological features (predictor variables) using a relatively
small random sample of wells drilled in the region, and (2) to provide data
vectors of the predictor variables assigned to drilling or measurement points
covering the area of investigation so that a reasonable interpolation is
possible. Stochastic simulation is suggested for the interpolation. The
method is designed as an iterative procedure: by including more information
in a stepwise manner, the results become more reliable. The method can be
278 HARFF AND OTHERS

55'0~
40.0
30.0

I 50 km I

Figure 9. Percentage probability of belonging to productive class.

used for planning of exploration activities because it provides the geologist

with information about favorable targets, although it cannot replace
geological thinking and modeling.

ACKNOWLEDGMENTS

The authors thank Prof. A. Journel, Stanford University, USA, for

valuable scientific advice and discussions; and express their appreciation to
the Deutsche Forschungsgemeinschaft, the U.S. National Science
Foundation (INT-9111646) and the Kansas Geological Survey for the
support of this research project.
GEOSTATIS TICAL CLASSIFICATION 279

REFERENCES
Deutsch, C.V., and Journel, A.G., 1992, GSLIB-Geostatistical Software
Library and user's guide: Oxford University Press, New York, 314 p.
Harff, J., and Davis, J.C., 1990, Regionalization in geology by
multivariate classification: Math. Geology, v. 22, no. 5, p. 573....;.588.
Harff, l, Davis, lC., and Olea, R.A., 1992, Quantitative assessment of
mineral resources with an application to petroleum geology:
Nonrenewable Resources, v. 1, no. 1, p. 74-84.
Harff, I, Davis, IC., Olea, R.A., and Bohling, G.C., 1992, Computer-
gestiitzte geologische Kartierung und Rohstoffperspektivitat: Die
Geowissenschaften, OktoberlNovember 1993, p. 375-379.
Journe1, A.G., 1989, Fundamentals of geostatistics in five lessons:
American Geophysical Union, Short Course in Geology 8,
Washington, D.C., 40 p.
Journel, A.G., and Huijbregts, C. l, 1978, Mining geostatistics: Academic
Press, London, 600 p.
Matheron, G., 1970, La Theorie des Variables Regionalisees et ses
Applications: Les Cahiers du Centre du Morphologie Mathematique de
Fontainebleau, Fascicule 5, Ecole Nationale Superieure des Mines,
Paris, 211 p.
Merriam, D.P., 1963, The geological history of Kansas: Kansas Geo!.
Survey Bull. 162, 317 p.
Myers, D.E., 1982, Matrix formulation of cokriging: Math. Geology,
v. 14, no. 3, p. 249-257.
Rodionov, D.A., 1981, Statististiceskie Resenija v Geologii: Nedra,
Moscow, 211 p.
Watney, W.L., 1984, Recognition of favorable reservoir trends in Upper
Pennsylvanian cyclic carbonates in western. Kansas, in Hyne, N.L.,
ed., Limestones of the Mid-Continent: Tulsa Geo!. Society Spec. Pub!.
2, p. 201-246.
Watney, W.L., Newell, K.D., Collins, D.R., and Skelton, L.H., 1989,
Petroleum exploration in Kansas-past trends and future options: Proc.
Eighth Tertiary Oil Recovery Conference, TORP Contr. 10, Lawrence,
Kansas, p. 4-35.
 TRANSITION PROBABILITY APPROACH TO
STATISTICAL ANALYSIS OF SPATIAL
QUALITATIVE VARIABLES IN GEOLOGY

Junfeng Luo
Gesellschaft fur wasserwirtschaftliche Planung und Systemforschung
mbH, D-12526 Berlin, Germany

ABS1RACf

A spatial qualitative variable can be expressed by {z(x) E {Sl,S2···Sm} I

x E R}, where {Si} represents a set of discrete states and x the
coordinate vector in a two- or three-dimensional space R The descriptive
aspect of geology as well as the qualitative information from basic
geological investigations force the community of mathematical geology
inevitably to handle such qualitative variables as rock types, sedimental
phases, ore types, etc. when modeling geological phenomena connected
with them. A spatial Markov chain, or transition probability approach,
provides an appropriate formalism to model geological spatial phenomena
characterized by qualitative and discrete state variables. Methods of
Markov conditional simulation (estimation) as well as case studies are
given to demonstrate the proposed approach.

INlRODUCfION

The analysis and modeling of geologic phenomena described by

qualitative variables for example rock types, sedimentary facies, ore, or
contaminant types, etc. are of special interest and importance to such
practical problems as:

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 281
282 LUO

o spatial distributions of certain rock types in underground (for

related questions in environment geology, hydrology, or petroleum
geology)

o spatial distribution patterns of different ore types as a basis for

mine planning;

o environment monitoring of the distribution of contaminants, for

example certain heavy metals.

In the context of mathematical modeling of geological phenomena

Krumbein and Graybill (1965) distinguish geological data or variables as
nominal, ordinal, and numeric type, that is three types, which correspond
to qualitative (for example rock types), semiquantitative (hardness scale
of minerals), and quantitative (geochemical survey) variables respectively.
A spatial qualitative variable z(x) can be defined as a type of variable
which takes, in nature, a set of nonnumeric discrete states as values, that
IS

(1)

where x represents the coordinate (vector) in a two- or three-dimensional

region Rand {Si} is a finite set of m distinct discrete states considered,
for instance, {dolomite, sandstone, limestone, shale} describes a simple
geologic section of four sediment types. It is apparent that among the
discrete states there exists generally no such order relation as Si > SJ
(i:;tj) or vice versa. That implies the difference Z(xI)-z(x2)
(z(xI)' z(x2) E {Si} and Xl' x2 E R) has no numerical meaning. This
important property of qualitative variables distinguishes them from
(measured) quantitative variables.
In the last years, geostatistical methods based on the theory of
regionalized variables are used widely to model geological spatial
phenomena. In geostatistics a regionalized variable z'(x) refers usually to
a measured quantitative variable given by (cf. Iournel and Huijbregts,
1978; or Olea, 1991)

{ z'(x) E (a, b) or [a, b] I X E R} (2)

where (a, b) or [a, b] represents a real value interval with a < b. In the
situation of a spatial phenomenon described by a set of discrete states, for
SPATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 283

example rock types, or lithofacies, geostatistical methods can be

employed only under certain previsions and restrictions. Whereas Markov
chain models provide a straightforward formalism to handle such discrete
variables.
This paper suggests that a transition probability approach or a spatial
Markov chain model provide an appropriate formalism to describe as well
as to model those geological spatial phenomena characterized by discrete
variables z(x).
First, it is argued that (k-step) transition probabilities are appropriate de-
pendence measures for the spatial auto- and cross-correlations among the
m states. The analysis of the transition matrices provides the information
about spatial dependence and expected homogeneous range, discontinuity,
ergodic property of states considered. Furthermore, the transition
probabilities can be used straightforward for conditional simulation or
estimation of unknown z(x). Methods of Markov conditional simulation
(estimation) as well as case studies are given to demonstrate the proposed
approach.

DFSCRIPTION OF SPATIAL VARIABILITY OF

QUAUITA~EVARIABLFS

A quantative description of the spatial variability of different discrete

states is the first step towards modeling. In following discussions, let z(x)
represent some geological phenomenon, for example, the spatial
distribution of rock types in a certain region R B={ z(xiY I B=l...N} will
be used to denote a initial set of samples (observations) of z(x):
z(xB) E { S }... Sm }, xB E R S(h)= { z(x~, z(x n +h) I a= 1...N(h)} refers to
a set of spatial sample pairs from the basic sample set B, that is z(x~,
z(xn+h) E { z(xB) I B=l...N} and their locations are separated by direction
vector h.
Now we suppose that z(x~=Si' z(xn+h)=Sj' without loss of generality.
To describe the qUalitative variability between z(x~ and z(xn +h), an
intuitive way is to interpret it directly as a state transition of z(x) from Si
at xn to Sj at xn+h, and along direction h and with step length Ihl.
Denote nij(h) the total transition number from Si to Sj o~~erved in the
spatial sample set S(h), a m by m transition count matrix N =(nijCh») can
be obtained

i,j=1...m (3a)
284 LUO

In In

nj.(h) = L njk(h) , n .. (h) = L nk·(h) i,j=1...m (3b)

k=\ J k=\ J

In In In

n .. (h) = ~ nij(h) = ~ nj.(h) = ~ nj(h) = N (h) (3c)

IJ I J

Equations (3) summarize complete state transitions in S(h)={z(xa +h),

z(xn> I a=1...N (h)}. Because a transition from Si to S· in direction h
implies also a transition from Sj to Si in the opposite direction -h, two
estimated transition matrices can be obtained from Equations (3) as
follows

p*(h) = ( p *jjCh) ) = ( nij(h)/nj.(h) ) i,j=l ... m (4a)

p*(-h) = (P*jjC-h) ) = ( nijCh)/n.jCh) ) j,i=1...m (4b)

with
o~ p*ijC±h) ~ 1, Lj p* jjC±h) = 1 for i,j=1...m (4c)

Suppose that z(x) is a realization of a spatial random function with

Markov property along lines in direction h (Switzer, 1965, p.l860; Luo,
1~93, p.18; s~e also the Appendix). Under the hypothesis of homogeneity,
Pi/h) and p ji(-h) are valid estimates of transition probabilities Pi/h) =
p( Z(x+h)=Sj I Z(x)=Si ) and Pi-h) = p( Z(x)=Sj I Z(x+h)=Sj )
respectively. Whereas pjj(h) describes the spatial auto-correlation of state
Si, PijCh) (i 7= j) reveals the cross-correlations between any two different
states Si and Sj. The complete spatial structure of z(x) under consideration
then can be studied through the structure analysis based on the so-called
Markov diagrams of k-step transition probabilities PijCkh) for different step
k as well as for diverse directions h (see the following case study).
Remarks: The basic idea of geostatistical approachs to characterize the
spatial variability of a regionalized variable z'(x) given by Equation (2),
is to define some spatial continuity measures based upon the qUll1lJitative
difference between z'(x) und z'(x+h) (cf. Joumel, 1988). A wi2ely used
measure is the semivariogram y(h) = 112 E { (z'(x+h) - z'(x» } under
the intrinsic hypothesis, which is estimated by the experimental
semivariogram (Joumel and Huijbregts, 1978, p.12; Luo, 1984, p.l30;
Olea, 1991, p.25)
SPATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 285

1 N(h)
y*(h) = - L (z'(xa +h)-z'(X a )2 (5)
2N(h) a=l

If a qualitative variable z(x) is specified by Equation (1), it is apparently

difficult as well as inappropriate to apply Equation (5) directly, unless
some transformation f on z(x) is made. In other words, the discrete state
space 0= {SI",Sm} should be mapped by finto another space of numeric
type 0', so that Equation (5) can be applied toj(Z(x». For this purpose
Gaussian Anamorphosis (e.g. de Fouquet and others, 1989) and indicator
function (Joumel, 1983) usually are used. Whatever transformation is
applied, a question remains, namly, to what extent j(z(xa,+h» - f(z(xaJ)
represents the purely qualitative difference between z(xa,+h)=Sj and
z(xaJ=Si'

MODELING BASED ON mE TRANsmON PROBABILITIES

A modeling of spatial variability, for example conditional simulation or

estimation of the spatial distribution of different states is possible, as soon
as the transition matrices ( Pij(kh) ) for different lags and directions h have
been estimated from available data. There are several algorithms for
conditional Marlcov simulation or estimation of the spatial phenomena
(Luo, 1993a). In general, conditioning indicates that modeling procedures
are directly controlled by known states and corresponding state-transition
probabilities. Principly, a modeling procedure of an unknown z(xo) (Xo
for a point, a cell or a block in R) is performed in the following
sequences

(1) Detect n known neighboring cells of z(Xo), z(xa)' a=L.n. Let

Piaj(h~=p(z(Xo)=Sj I z(xJ=SiJ, z(xJ=Sia E {SI",Sm} and ha=xa-Xo·
specifies then tlie probability of the occurrence of Sj at Xo,
conditioned by Sia at xa;

(2) CalculateVs = ( PsI".Psj"'Psm)' the average transition vector of

the transition probabilities Piaj(ha) corresponding to n known
z(xa ), a=L.n:

Psj = lin Lia=l...n Piaj(ha) (for j = L.m) (6)

286 LUO

Vs provides an approximation to p( z(xo) = Sj I z(xa) = Sia'

a= 1...n), the conditional distribution, upon which conditional
simulation or estimation can be achieved by

(3a) Simulating state Sj for z(xo), in accordance with the Monte-Carlo

Method, or by

(3b) Assign SjO to z(xo), such that PsjO is maximal among

{Psj' j= 1...rn } (estimation).

A CASE STUDY

To illustrate the performence of the transition probability model, a case

study is given with the emphasis on the modeling spatial pattern of a
discrete variable z(x) defined by Equation (1). Figure IA shows a known
geological profile section of four glacial deposits of Quaternary age. The
four deposits: gravel, sand, mixture of sand and clay, and clay are
subsequently represented by Si i=1...4. For the sampling, the 480m long
section with 60m in depth is represented by a regular grid of 96 x 60
cells of size 5 x 1m2 in such a way that each cell contains one and only
one of the 4 defined states Si (cf. Fig.IB).
In this way we have a gridding representation of the geological section
considered and a basic sample set { z(xJ3) I J3=1...N=5700} on the 96 x 60
grid nodes, where xJ3=(x, y) is the coordinate vector of middle point of
corresponding cells. Table 1 summarizes the observed frequencies of each
state, that is ni' along with estimated state probability p *j. To describe
quantitatively the deposit (state) alternations along the vertical (y-) and
the horizontal (x-) direction, transition count matrices [Eq. (3a)] along
with the corresponding transition probability matrice [Eq. (4)] are
calculated. Table 1b gives only the I-step transition count and probability
matrices, wheras k-step transition matrices p(kh), k=2,3 ... can be derived
easily by recursive using p(kh)=p(h)*p«k-l)h) because of the Markov
property and Chapman-Kolmogorov Equation (e.g. Feller, 1968; Ferschl,
1970).

Spatial Structure Analysis of the Markov Model

Based on k-step transition probabilities, we now can quantify the spatial

variability of each deposit (state) as well as spatial correlations between
SPATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 287

Table 1a. Estimates of state probability of Sj (5 x 1m2-grid)

n·1 1320 2738 308 1394

.2292 .4753 .0535 .2420 P*(S·)=

1
n·I / 2:.1 n·1

Table lb. Transition probability matrice (5 x 1m2-grid)

p(hx )=(p ijCh x)) Il hxl=5m p(-hx)=(PijC-h x))

~ .0191 .0053 .0153 ~ .0153 .0115 .0092

.0074 .9841 .0074 .0011 .0092 ~ .0070 .0029
.0502 .0635 ~ .0000 .0240 .0685 ~ .0000
.0087 .0058 .0000 .9855 .0145 .0022 .0000 ~

p(h y)=(p ijChy)) I Ihyl=lm p(-hy)=(Pij(-hy))

.7720 .1076 .0530 .0674 .7720 .0667 .0235 .1379

.0333 ~ .0106 .0454 .0537 .9107 .0254 .0102
.1006 .2175 .6818 .0000 .2273 .0909 ......QJUJl. .0000
.1306 .0194 .0000 .8501 .0638 .0861 .0000 ~

Table 2. Structure parameter of Markov-Model (5 x 1m2)

S 1(gravel) S2 (sand) S3 (mixture) S4 (clay)

E[DjCx)] 132m 286m 49m 321m

E[Dj(Y)] 4.4m 11.2m 3.1m 6.7m
djCx) 650m 1300m 325m 1500m
djCy) 16m 32m 16m 25m
p(Sj) .2292 .4753 .0535 .2420
7!j (x) .2262 .4837 .0541 .2359
7!j (y) .2330 .4664 .0544 .2461
1-Pj/x) .0379 .0175 .1031 .0156
1-PjjCy) .2280 .0893 .3182 .1499
288 LUO

18 . 8 Explanal ion

l1li gravel

..
fBj sand

~mjxture
e.B
clay

18m

-Ie .• Sm
~2:se . 8 ~73B.B

Figure lA. Geologic profile section of four glacial deposits of Quaternary age.

30 .0 ~-'-:""':--'-~"-T"-~"'-)""-1""T" ' T--r'--r"';~'-"T--r"T'''T-T''T I

..... ..... ..... f' .... 1' .... :' ....;..... :' .... r"" ...... ....... "To",
~ ~ ' :"'" T' ... :' ..... .. '.. ..... .....
~ ~' ~ ~ ~ ~ ~ [=::J SIdle I

24 .0 l'JIttI;~""":''''':''''' : '''''j'' . ~ ......( ... ' .... ;. .....:-- ... j- .... j......!... ~ .. "1' c::::J SIdle 2
" :'" .. ..: ,.. ~., ,.. ~ ..... ~ .. .. .i" .. i-" .• i····· ,;" ,. -
... .. .......:. .... .:......... .....
. ,:.
E]Sldle3

: :
~
18.0
,.. ... ~ ... ,,~ ... ,:, ... , ~" .. .. ... ~ .. , ... .....-~. . ._--.- .. .... _ S late 4

. .. : ... .. : .. . .. ~ ..... ; .. .. ,: ..... ~ .. . -. : . .. -, .:......~ .....:......:. ....., .. , - -=- . .... ; ..... :. .. . . . ~ ..... ~ ~
12.0
..... ... : .. :.. L... i ..... L .. L... L. .. L .. L .. l.. ....L ...L. ....L... L .. L... \!
. : : :; ::" - . : ;'
6.0
I ..
.. ': .. ... :.. ... :.... , :'. --~ ... -:- .... !' .... ! '.... ~.
.....,~. .... :.: ..... :......
.: '.
,........ :.. ......': ... .........
:. ~
.....
......;. ....··T'·· ..~.. ·.. ·~ ....r. ··:· ...·'i'····:·····:·
.;......:...............: .. .
. .. " .:.,
.
.
. ..
~,
.
., '
.

.J. , ... : ..... !..... J ... .. ~ .. ....~ ..... ~ •• , •• ~ ......; •••

. . .. .

-6.0

-12.0

-180

-24 .0
-:::::.::::: '::::: ;::::: :.::: ::: i::::: ;.:::::: 1.':: ::: l.: ::::1:::::i.': ::: L:: l. ~~::f.:.::::r:" ~.......
.... :... .. ~ .... :.
....." ;: .... .-; : " :"
: ::: :
I I I I I I I I I
4300.0 4350.0 44000 4450 .0 4500 .0 4550.0 4600.0 4650 .0 4700 0

Figure lB. Gridded representation of geologic section (96 x 60 cells of size 5xlml).
SPATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 289

them. Let us look at the Markov diagrams (Fig.2A-D), where k-step

transition probabilities Pij(kh) against corresponding steps k are graphically
presented. The two diagrams in Figure 2A, for instance, document the
variation of transition probabilities P1j(kh),j = 1,2,3,4 along the horizontal
(x-) and vertical (y-) direction respectively. Here we observe:

(1) as step k (distance) increases, transition probability P12(kh) and

p 14 (kh) increase, whereas p 11 (kh) decreases. Opposite to that,
P13(kh) shows few variation. This implies that state 1 (gravel) has
a high spatial correlation with state 2 and 4 (sand and clay), but
few with state 3 (mixture);

(2) if step k remains small, then P11(kh) > P12(kh) > P14(kh) > P13(kh).
But as soon as step k exceeds certain values that is
kx < 24 (= 120m) ky < 5 (= 5m), the diagram shows that P11(kh)
< P12(kh) at first, then P11(kh) < P14(kh). This phenomenon implies
that if we observe state 1 occurring at position x [i.e.
z(x) = gravel], then, for relative smaller distance kh, that again
state 1 occurs at position x+kh [i.e. z(x+kh) = gravel] is more
possible than that other states occur. Generally, the larger the
distance, the less the possibility of occurrence of state 1 at
position x+kh unless there is a period. If the distance kh exceed
certain values, then it is more possible that state 2 (sand) occurs
at position x+kh [i.e. z(x+kh) = sand] because P11 (kh) < P12(kh);

(3) furthermore, if step k is large enough such that kx > 130

corresponding to 650m, and ky > 16 corresponding to 16m, then
the four transition probabilities P1j(kh), j = 1,2,3,4 become almost
invariant and P1/kh) -> Pj. In this situation, the spatial auto- and
cross-correlation among the four states disappear;

(4) as h -> 0, there occurs p 11 (k~) -> 1 as it should be, but p 11 (k~)
does not. This implies there exists some discontinuity of state 1
(gravel) along the y-axis (vertical) for smaller distance.

Similarly, we can quantify and analyze variations of the other states

based on transition probabilities P2j(kh), P3j(kh), and p1i(kh) and the
corresponding Markov-diagrams (Fig.2B-D). To quantity the spatial
continuity of each deposit, the probability of k-time continuous occurrence
of state Si, p(Ti=k), can be used (e.g. Krumbein and Dacey, 1969)
290 LUO

t rP(~k:)II::::::::::::::::::::::::::::~l
0 ..
-
1 .-...
1III.la, jh ..
" . , " II.. tit
lu
I...
~ ,.".,b. fl-S III tal
--Go- IK·d u ( h III a..)
I ...... . _.'N.UJ
.............. k'
Q.O (r-______ ~-- _____

ao 100 tlO 200 2&0 300

atap k (1 atep • 6ml
Initial atate • 93 (Mlxtu .. )
360 400 460 600 no
c to 20 30 .0
atap k (1 atep • 1m)
Initial atate • 93 (Mlxtu,.)
10
'"

tIP(~k:)II::::::::::::::::::::::::::::~l
-,,_III,rl.6,,',1 ... IMlot.... I........ ,
-.. pt)dnll4 ....:a) . . . . .IoJ<I& I ....... .

60 100 110 200 260 300

otep k (1 atep • 6m)
:no .&00 .eo 600 no 0 tel 20 30 .0
ttap k (t atep • 1m)
10 to

Initial atate • 94 (Clay) Initial atate • 94 (Oay)

Figure 2. Markov diagrams of A, state 1 (gravel); B, State 2 (sand); C, state 3 (mixture);

and D, state 4 (clay).
SPATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 291
k-I .
p(Ti=k) = (1 - Pii(h» • (Pii(h) (for I = 1...4) (7)

p(Ti=k) describes practically homogeneous spatial extension of state Si

and also can be against distance k graphically presented in Figure 3.
There we observe, in general, as distance k increases, p(Ti=k) for
i=I,2,3,4 decrease. But the decreasing rates for each deposit Sj are
different. For instance, for given probability value of 0.6, it seems that
clay and sand have larger spatial extension than gravel as well as that for
the mixture of sand and clay [for example along the horizontal,
P(T4= 170m)=p(T2=160m) =p(T 1=55m)=p(T3=20m)=0.6]. In comparision
of p(Tj=k) along the x-axis (horizontal) and the counterpart along the y-
axis (vertical)., it is easy to see that the anisotropic variability of the
deposits is quantified [e.g. for deposit gravel SI
Px(T 1=55m)=py(T1=2m)=0.6]. Furthermore, we can calculate the expected
value of the extension of state Si' E(Di) by

(8)

As summary, we have a set of structure parameters (similar to those in

geostatistics) given in Table 2, where index j stands for deposit Sj, x and
y in parenthesis represent along the x- and the y-axis respectively. E[DjO]
is the expected value of the extension of Sj. 1ljO = limkh_>ooPij(kh), the
limited value of transition probability Pilkh) and 1lJ O is comparable
directly to the corrsponding state probabihty p(Sj)' ~O is the so-called
memory duration such that 1lJ O= Pij( ~O ) is satIsfied at the first time,
and hence djO quantifies the correlation range of Sj. I-Pii Oprovides the
information on the spatial continuity or variabihty or Sj as kh -> 0
(similar to the so-called "nugget-effect").

Conditional Simulation of Spatial Distribution of Deposits

Now, based on the transition matrices ( Pi'(kh) ) mentioned previously,

the spatial distribution pattern of different deposits can be modeled. For
the purpose of demonstration, we make use of seven drillholes in the
geological section (Fig.l) to condition or control the modeling procedure.
This is known from drillholes which are used as starting states for the
simulation. The modeling procedure extends sequentially from the bottom
to the top of the section: the lowest "layer" is conditioned by every
known deposit state at the seven drillhole positions and modeled first.
The modeling of the second lowest "layer" then is controlled by
292 LUO

along x - direction (horizontal) along y - dlrection (vertical)

prob{TJ'k l
'~~------~====~
o. al<oQru.l ~ ~T.)

0 .8 ........ p(T.~ 01 .u - UI'IO

60 lOO 160 200 260 300 3&0 400 460 600

Figure 3. Diagram of probability of k-time continuous occurrence of Si' P(Ti=k)

y
[J Cells unknown &
to be simulated

Cell known
with state 1

~~~~,~:::::::::::~,:_=~,.~~"'+-+-t--H-:
+~' -::=:=::::~+-t--il-t­
Cell known
with state 2

i~;=~
Cell known
with state 3
Wi '-:-'"-:-'-:- X
o

k=2

y+. direction b

1 1±ll±IIIEIIFIII
Uk) (I, k)
k=l
1++1 1 1 1++11'1 1 1.... 1
/, iq iK
)("-direction ----+ x+-direction

Figure 4. Description 2D-Markov simulation conditioned by drillholes

SPATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 293

every known deposit state at the seven drillhole positions of the second
lowest "layer" as well as by those "known" (simulated) states of the
lowest "layer". Such modeling procedure continues until the entire section
will have been simulated (Fig.4).
Figure 5A and 5B show a simulated section (stochastic modeling using
Monte-Carlo method) and an "estimated" one [i.e. assigning state S·o to
an unknown z(:xo), such that transition probability PsjO is maximal]. nere
one can see that the spatial pattern of the original section has been well
reproduced.

CONCLUDING REMARKS

Markov chain model provides an appropriate approach to modeling

geologic spatial phenomena described by a set of discrete states. Under
the Markov model, the spatial variability of different ore types are
interpreted intuitively and straightforward by spatial state alternations or
transitions, which then are described by corresponding transition probabi-
lities. A reliable estimation of transition matrices, therefore, is the first
and important step towards the Markov modeling.
The structure analysis based on the k-step transition matrices p(kh)
provides not only such structure parameters such as memory duration, 1-
Pii(h) if Ihl is small and limited transition probability 1lj=lim lkhl_>oo Pij(kh).
In fact, the Markov diagrams of k-step transition probabilities Pij(kh)
against step k reveal the complete spatial auto-correlation and cross-
correlation structure of the problem under consideration.
Conditional simulation or estimation of the spatial variability or
distribution of different ore types is practicable as soon as the transition
matrices ( Pij(kh) ) for different lags and directions h have been estimated.
The applicatIon of the spatial Markov chain model consists of the follow-
ing five steps:

(1) state definition and encoding of the sample data;

(2) Markov model analysis and structure analysis based on k-step

transition probabilities;

(3) conditional Markov simulation or estimation based on transition

probabilities;
294 LUO

30.0
. . . . .. . . . ~ . . . .. . .
c::J State I
24.0 c::J State 2
State 3
18.0 _State 4

12.0

6.0
_ ... :..... :..... 1. .... , ..... , ..... :..... :. . .. <. .... .:. .....:. .• <.. .... .:.-: .... .:. .....; .•... : .. ••.
. .: _::: ~ :: ;:::"
o ..
_ ••••• : • • ••• ~ ••••• : ••••• ~ ••••• ~ . ,_ '_:" _ •• ~ ••••• ':' •••• ~. " I I . :• • • • • • : • • •

. . ...

-6.0

-\2.0

-t80 I

: . : ; : :: :: :: : J
-24,0
-:::::j:::::t::::T::T:::F::r~T:::l::::;:r:::F:::l:::T:::[:r::::!:::::f:::::f:
4300.0
I -
4350.0
r -
4400.0
-, - ~-,.....-
4450.0
-=-....-.
4500,0
,- -4550.0 I -I
4600.0
1
4650.0
~ (
4100.0
.

Figure 5A. Image produced by conditional Markov simulation.

30.0
c::J State 1
~St41e2
24.0
State 3
18.0 _State4

12.0

6.0

.0

~.O

-\2.0

-\8.0

-24.0

Figure 5B. Image produced by "Markov estimation".

SP ATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 295

(4) assessment of simulation results; and

(5) graphical representation of the results.

A detailed discussion of algorithms for conditional Markov simulation

as well as its applications are given in Luo (1993a). Although
geostatistical methods are used mostly for the modeling of spatial
phenomena, they are only employable under certain provisions if a spatial
phenomenon is described by a set of discrete states, or especially, of
qualitative states like rock types or lithofacies. In this situation, the
Markov model shows not only the advantages, but the transition
probability formalism can be used directly to estimate the bivariate
probability distribution defined for a set of class indicator variables (Luo
and Thomsen, 1994), because there are, under certain prerequisite, certain
relations between transition probabilities and the cross- and variograms
of the class indicators (Luo, 1993a, 1993b).

ACKNOWLEDGMENTS

This work was completed in the institute of geology, geophysics, and

geoinformatics of the Free University of Berlin. I thank Prof. Dr. W.
Skala who encouraged and supervised the author during the research work
which, for the most part, was supported by a scholarship from the Free
University of Berlin. The author also thanks those who have helped his
work, especially Dr. H. Burger and Dipl. Math. A. Thomsen for
discussions and useful suggestions.
296 LUO

REFERENCFS

Bartlett, M.S., 1971, Physical nearest neighbor models and non-linear

time series: Jour. Appl. Prob., v. 8, no. 2, p. 222-232.
Bartlett, M.S., 1975, The statistical analysis of spatial pattern: Chapman
and Hall Ltd, London, 90 p.
Bennett, R.I, 1979, Spatial time series - analysis - forecasting - control-:
Pion Ltd., London, 674 p.
Besag, IE., 1972a, On the correlation structure of two-dimensional
stationary processes: Biometrika, v. 59, no. 1, p. 43-48.
Besag, IE., 1972b, Nearest-neighbour systems and the auto-logistic
models for binary data: Jour. Roy. Statist. Soc., B34, no. 1, p. 75-83.
Besag, IE., 1974, Spatial interaction and the statistical analysis oflattice
systems: Jour. Roy. Statist. Soc., B36, no. 2, p. 192-236.
Besag, IE., 1986, On the statistical analysis of dirty pictures: Jour. Roy.
Statist. Soc., B48, no. 3, p. 259-302.
Chung, K.L., 1967, Markov chains with stationary transition probabilities
(2nd ed.): Springer-Verlag, Heidelberg, 301 p.
de Fouquet, C., Beucher, H., Galli, A., and Ravenne, c., 1989,
Conditional simulation of random sets - application to an argilaceous
sandstone reservoir, in Armstrong, M. ed., Geostatistics v. 2: Kluwer
Academic Publ., Dordrecht, p. 517-530.
Feller, W., 1968, An introduction to probability theory and its
applications, Vol. 1 (3rd ed.): Wiley International Edition, New York,
509 p.
Feller, W., 1971, An introduction to probability theory and its
applications, Vol. 2 (2nd ed.): John Wiley & Sons, New York, 669
p.
Ferschl, F., 1970, Markovketten, Lecture notes in operations research and
mathematical systems: Springer-Verlag, Berlin, 168 p.
Journel, A.G., 1983, Nonparametric estimation of spatial distributions:
Math. Geology, v. 15, no. 3, p. 445-468.
Journel, A.G., 1988, New distance measures: the route towards truely
non-Gaussian geostatistics: Math. Geology, v. 20, no. 4, p. 459-475.
Journel, A.G., and Huijbregts, Ch., 1978, Mining geostatistics: Academic
Press, London, 600 p.
Krumbein, W.C., and Graybill, F.A., 1965, An introduction to statistical
models in geology: McGraw-Hill Book Co., Inc., New York, 475 p.
SPATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 297

Krumbein, W.C., and Dacey, M.F., 1969, Markov chain and embedded
Markov chains in geology: Math. Geology, v. 1, no. 1, p. 79-96.
Luo, J., 1984, On some problems of the estimation of experimental
variograms: Jour. Geoscientific Information, no. 4, Wuhan, China, p.
130-137 (in Chinese).
Luo, J., 1993a, Konditionale Markov-Simulation 2-dimensionaler geologi-
scher Probleme: Berliner Geowiss. Abh., D, 4, Berlin, 103 p.
Luo J., 1993b. Beziehungen zwischen raumlichem Markov-Modell und
Ko-Indikator-Systemen sowie deren praktischen Bedeutungen, in
Peschel, G., ed., Beitrage zur Math. Geologie und Geoinformatik,
Band 5. Neue Modellierungsmethoden in Geologie und
Umweltinformatik: Verlag Sven von Loga, Koln, p. 37-43.
Luo, J., and Thomsen, A., 1994, Direct estimation of the bivariate
probability distribution of a regionalized variable from its spatial
samples: Science de la Terre Ser. Inform. Geol., v. 32, p. 115-123.
Olea, RA., ed., 1991, Geostatistical glossary and multilingual dictionary:
Oxford Univ. Press, New York, 177 p.
Ripley, B.D. 1988, Statistical inference for spatial processes: Cambridge
Univ. Press, Cambridge, 148 p.
Switzer, P., 1965 A random set process in the plane with Markovian
property: Ann. Math. Stat. v. 36, no. 6, p. 1659-1863.
298 LUO

APPENDIX: THEORETICAL NOTE ON MARKOV CHAIN

MODElS

About Markov chain theory there is a lot of literature available, for

example Chung (1967), Feller (1968, 1971), or Ferschl (1970). The
following are basic definitions of Markov process/chain models.
Denote {Z(t) E nit E T} a random process: n is the state space which
can be (a, b) or [a, b], a continuous interval of real values with a < b, or
be a set of discrete states {S I'" Sm}' T is the one-dimension parameter
space (e.g. for time T=[O,+oo) ). Suppose Z(t i ) = zi E n, i=1...n". be
values (states) of Z(t) corresponding to time points t1 <t2".<tn
respectively. Z(t) is a Markov process if the conditional probabilities, for
2 ~ q ~ n, satisfy the following identity (Ferschl, 1970)

p(Z(tq)~q I Z(tq_I)=Zq_I".Z(tl)=zl) = p(Z(tq)~q I Z(tq_I)=Zq_l) (lA)

when Q=(a, b). If Q={Sl",Sm}or {Sl",Sm"'} there is (Feller, 1968)

A Markov chain is a special Markov process defined by Equation (lB).

When n = {Sl",Sm}' Z(t) is a finite Markov chain. The conditional
probability relation specified by Equation (lA) or (lB) also is termed
Markov property. In Equation (lB), suppose Z(tq)=Sj' Z(tq_I)=Si, Si,Sj E
{SI",Sm}' and "C=tq-tq_l , then

Pij(tq_I,"C) = p(Z(tq_l + "C)=Sj I Z(tq_I)=Si) = p(Sj lSi) (lB')

Equation (lB') is termed the transition probability of process Z(t) from

Si at t.9- 1 to Sj at t q. "C is the transition step length. If Pij(tq_l , "C)=Pij("C) for
any Si,Sj E {Sl",Sm} holds, then Z(t) is a Markov chain with
homogeneous transition probabilities. In this situation the state transition
behavior of the Markov chain is described completely by the m by m
transition probability matrix P('t)=(Pij<'t), or P=(Pij) for given step-length
'to
Now consider {Z(x) E n I x E R}, a random function or a random field
defined in a two- or three-dimensional parameter space R There are two
ways to define a spatial Markov chain model. By one of them, the 2D-
Markov property is formulated in the frame of a "nearest-neighbor
system" (cf. Bartlett, 1971,1975; Besag, 1972a; Bennett, 1979). There
SPATIAL MARKOV CHAIN FOR GEOLOGIC MODELING 299

Z(x) is considered on a regular 2D-grid. Let Zk I be a realization of Z(x)

at the grid point (k,l). The Markov property is 'defined by the following
conditional probabilities (Besag, 1972b, 1974):

or by a more general formula (Ripley, 1988)

P(Zk II ZU v' (u,v) E {Zlk's neighboring points}) (2B)

" ,
Here {Zk /s neighboring points} can enclose also the four diagonal grid
points (k± 1,l± 1). Markov chain models satisfying Equation (2A) define
a group of Markov random functions or Markov fields which are used
mainly in image processing for such purposes as classification and
estimation of pixels (Besag, 1986; Ripley, 1988).
Another definition of a spatial Markov chain model leads back to an
early work of Switzer (1965), who had constructed a plane random
process with m~2 states, such that the alternation of states along any
straightline on the plane considered has the Markov property. From his
results, another type of spatial Markov chain model can be derived.
Suppose {Z(Xq)=Zq E {St",Sm}, q=1...n} be a set of samples ordered
along a straightline in such a way that Xq-xq_t=aqh, where h is the
direction vector and al<a2< ... <an are real constants. Z(x) is a spatial
Markov chain if

One of the obvious advantages of the Markov model defined by

Equation (3A) is that regarding a certain direction h, Z(x) can be
considered as a one-dimensional Markov chain. As a result, the complete
theory of Markov chains/processes can be applied to investigate the
spatial structure of Z(x) regarding to different directions under
consideration. In fact, the known geologic knowledge or the spatial
configuration of available data allows to analysis of the spatial structure
of Z(x) merely along certain directions.
 AN INTELLIGENT FRAMEWORK FOR GEOLOGIC
MODELING APPLICATIONS

Lee Plansky, Keith Prisbrey

University of Idaho, Moscow, Idaho, USA

Carl Glass
University of A rizona, Tucson, A rizona, USA

Lee Barron
University of Idaho, Moscow, Idaho, USA

ABSTRACf

This paper provides a framework that can be used for constructing

system models, simulations, or actual devices - real time or otherwise.
The framework constructs, in the limit, can be intelligent, adaptive,
systems or devices with a common architecture that may be used in a
variety of applications. Application examples are given from
characterization and exploration. A representation is introduced that
generalizes the concept of an artificial intelligent agent, or being. The
organization for intelligent systems given in this paper may be applied to
systems modeling or control in: waste characterization (buried, stored, or
processed), site characterization and restoration, materials processing and
handling, regulatory compliance monitoring, environmental monitoring,
waste storage or repository modeling, site selection, and facility
monitoring and control.

Geologic Modeling and Mapping

Edited by A. Forster and D. F. Merriam, Plenum Press, New York, 1996 301
302 PLANSKY AND OTHERS

IN1RODUCflON

The earth sciences, mining and materials processing industries are on

the threshold of a technological revolution which could lead to the full
automation of exploration, characterization, mining, and materials
processing. Integrated, or hybrid, knowledge systems, working together
with intelligent sensors and adaptive devices are one of the tools that can
help realize this revolution. An integrated knowledge system is an
assemblage of computer programs which may work together with
electronic hardware and uses the technologies of artificial intelligence
including: expert systems, neural networks, classical procedural
computing, databases, and information reporting systems. Such systems
are new. A recent integrated application in metallurgy uses neural
networks and expert systems to control electric arc furnaces in steel
manufacturing (Kehoe, 1992). This integrated application is estimated to
save two million dollars per year per furnace, where the furnace was a
state-of-the-art system before the improvement! The benefits come from
using neural networks together with expert systems in the furnace
controller. Such developments imply that profit or national defense
restrictions may keep many of them from ever being published,
subsequently hindering economic growth. Figure 1 shows one way in
which neural, expert, and procedural systems may be integrated. A neural
network (N) at any node in the diagram may be replaced by a procedural
(P) or expert system (E) system, model, or module. There are a finite
number (9) of interfaces that must be considered in the construction of
such an integrated system, they are: N:N. N:E, N:P, E:P, E:E, E:P, P:N,
P:E, and P:P.

DEFINmONS

Neural Networks

Neural networks are a tool of artificial intelligence. Their uses

include: function mapping or fitting, pattern matching, feature extraction,
and completion of missing data. They are being used in modeling
systems, including rule generation, when information is limited. Neural
networks, also known as a parallel distributed processing (PDP) network,
neural computing network, or neural net, is a parallel information
 EHPERT SYSTEM NEURRL NE11IIORK o PROCEDURRL SYSTEM
o o ~
R
ti-4

KI..Y ~
~

i
~
Io"!j

~
~
~
~

NIIIE:
Rny Module Could Ret Rs The
EHec:utiue or Control Module w
o
Figure 1. Arbitrary integrated knowledge system. w
304 PLANSKY AND OTHERS

processing network composed of :many simple processing elements.

Neural networks can be a complement to expert systems. They can play
an important role in the analysis and modeling of the underlying complex
processes represented by the data. Neural networks offer a way to
generate rules from large amounts of data resulting from the mutual
interaction of complex processes. They are effective when the number of
rules governing a system or set of processes is large or where data are
missing or noisy. Normally, enough data must be available for model
construction and model validation when applying neural networks. A
neural network can generate rules from the relationships its learns from
the data. This ability can be used in place of an expert, especially when
dealing with complex proce~ses where a domain expert would be
expensive or not available.

Expert Systems

Expert systems are software systems which use automated reasoning

procedures or rules to draw conclusions. The most widely used expert
system is the so-called rule-based expert system. The heart of the system
is the inference engine that uses the rules to draw conclusions from a
knowledge base of information and facts. The rules, information, and
facts are obtained from human experts in the area discipline. Expert
systems cannot be used outside of the domain of their knowledge base.
The system usually has an explanation module which explains why or
how it arrived at a particular result. In a rule-based expert system, the
rules the expert system shell uses to search the knowledge base are in the
form of the IFTHEN statements: IF hardness IS 7.0 AND (the) color is
white THEN (the mineral is) quartz. The words IF, THEN, and AND are
fixed and part of the inference engine. The words hardness, color, 7.0,
and white are user defined variables and their values, or data. The word
quartz is a rule conclusion.

Procedural Systems

Procedural systems are software systems that consist of a main

calling routine and a set of subroutines, or modules, the sum of which
perform one or more predefined functions. Procedural systems usually are
written in higher level computer languages such as Pascal.
INTELLIGENT FRAMEWORK FOR MODELING 305

TECHNICAL CONSIDERATIONS

True intelligence requires the ability to adapt to changes. Intelligent

systems may be intelligent in the sense that they are programmed to
respond in a given way to a known event(s) or event sequence. They,
however, are not necessarily adaptive. This applies to artificially
intelligent hardware-software-based systems -- without adaptability they
respond in a fixed, predetermined manner. Thus, many of today's
artificial intelligence methods are simply another type of algorithm,
procedure, or model. The key to real intelligent systems lies in making
the system(s) adaptive. This requires the implementation of concepts from
biology, psychology and the cognitive, learning, and adaptive sciences.
These areas are important to real long-term systems applications.
In developing integrated knowledge systems, expert systems can play
an important role as the primary way of control, with a sensor system
providing the input data to a neural network or its procedural
representation, such as a computer program written in the C, Pascal, or
FORTRAN computer language and integrated in new or existing systems.
An expert system could be replaced by procedural programs, especially
if the control functions are simple and the knowledge base or data needed
are small. Neural networks excel over classical approaches to sensor data
analysis, function fitting, and feature and pattern recognition applications.
Neural networks in integrated systems permit the designer to analyze
nonlinear processes for which no simple analysis techniques exist. They
enable rule, procedure, and function generation of complex processes
which "classical" expert systems can not do. This is of interest in process
control applications and mining automation. Neural networks can analyze
and fit noisy measurement data and determine features, patterns, and
functions in the data with ease. They can complement or replace expert
systems. New techniques or other software or procedural algorithms also
may be included in the framework of this paper. Additional technical
considerations for implementing integrated knowledge systems are given
in Glass and others (1993). Workers at the U. S. Bureau of Mines,
Spokane Research Center are actively developing a hybrid AI roof bolting
system for coal mining safety application (Bevan and Hill, 1994;
Hoffman, 1994; Howie and Nichols, 1994; Signer and others, 1993).
306 PLANSKY AND OTHERS

APPUCATIONS

Integrated knowledge systems are applied here to the conceptual

design of a system for the exploration, characterization, identification, and
evaluation of earth materials, minerals, soils, or ores in the format of an
intelligent, adaptive exploration systems for earth, near-earth objects, or
planetary resources. This scenario is selected while the requirements for
extraterrestrial systems are severe and demand more ingenuity. They
inspire innovation, are challenging, and the need for success makes the
impossible, possible. They imply an interdisciplinary approach to problem
solving that compels the designer to forge reliable interfaces among
different technologies. Humans are assumed not to interact. Such a system
also may malfunction, be damaged, lost, or simply disposed. A second
example is taken from mining and the characterization of blast fragment
sizes using neural networks. This application and its solution is an
example of one that may be integrated readily and developed into an
intelligent product for market. Many of the concepts and technologies,
covered, or implied by this work can be used directly in geologic as well
as in industrial applications. A proposed organization for an AI Agent
concludes this section.
Applications for integrated knowledge systems include exploration,
characterization and mobile (or stationary) monitoring, control, or
automatic maintenance systems for use in mining. Exploration and
characterization generally require the use of different detection techniques
for the target(s) (detection objectives) in a given geologic environment.
Measurement becomes more difficult as the geologic environment
changes. The nature of different geologic, ore, or waste materials may
complicate matters - each material is generally different in properties and
behavior. This is in contrast to, but supported by, the present ability to
develop such units or devices for specific industrial applications, as
witnessed by the arc furnace example noted. Of particular importance to
the present work are modular, generic intelligent sensors or sensor
systems; such systems have been developed. Modular, intelligent,
integrated measurement systems would allow multiuse and
interchangability of costly sensor and data-acquisition systems. They have
their own computer processors with self-test and repair functions and the
ability to include multiple sensors for arbitrary application in different
environments. This work also implies the need for intelligent, look-ahead
sensor systems.
INTELLIGENT FRAMEWORK FOR MODELING 307

Characterization and Exploration

Identification of a material implies characterization which requires

property measurements and sensor systems. If the material is unknown it
is named upon characterization. It then is classified as a known material.
Characterization involves determining the nature or attributes of a
material and determining its properties. Properties can be taken in the
broad sense to include the behavior of a material under external forces.
Characterization leads to identifying a material, assuming it is a known
one. Exploration is the systematic examination of a region with the goal
of increasing existing knowledge of the region. It implies that a target
material is sought together with geologic information about the search
region that can be used to help locate or identify the material. Geologic
information is used in evaluating the potential for the material and its
worth.
Given a collection of known materials X = {Xl, X2, X3, ... , Xm},
two problems relevant to characterization and exploration are: (1)
identify an unknown material Uk or, (2) determine a known material Xk.
In the latter situation, Xk is a target material(s), or the valuable
objective(s) of the search. Characterization requires the determination of
the properties of the materials to be identified. The properties of a target
material usually are known or defined ahead of time. The
characterization, identification and evaluation of minerals, rocks, or ores
consists of the steps:

1. Characterize the target material

1a. Measure the target material properties
2. Identify the target material
2a. Determine if known material
3. Inventory the target material
4. Evaluate the target material resource value
5. Record and report results.

This can be formulated into a characterization, identification, and

evaluation procedure as shown in Figure 2. The procedure assumes that
the target material is available, that is, exists or has been located. Its
steps can be included in a more general exploration, identification, and
evaluation procedure (for known materials) as shown in Figure 3, namely:
308 PLANSKY AND OTHERS

1. Characterize
Target Material

1a. Measure Target

Material Properties

2. I dentify Target
Material

yes

3. I nuentory Target
Material
no

4. Eualuate Target
Material Ualue

Enter Amount
s. Record and and Ualue
Report Results .....- - - - '

Figure 2. Characterization and identification.

INTELLIGENT FRAMEWORK FOR MODELING 309

I. Specify Target
Material
...._ _ _ _... Ia. Specify Character
of Target Material

2. Find Target 2a. Characterize

Material Candidate Candidate Material

2b. Identify Target

Continue
Searching Material

yes

3. Inllentory Target no
or Candidate
Material

no
4. Ellaluate Target
or Candidate
Material Ualue

5. Record and
....--1 Report Results ,...- ....- ...

Figure 3. Examination or exploration and characterization and assessment.

310 PLANSKY AND OTHERS

1. Determine target material(s)

la. Specify character of target material(s)
2. Locate target material(s), (i. e.: locate candidates for a match)
2a. Characterize the candidate material(s)
2b. Identify the target material
2c. Examine candidate material(s) for a match
3. Action if no match (not located, continue or stop search)
4. Action if match (located)
4a. Inventory the target material
4b. Evaluate the target material resource value
5. Record and report results.

The procedure shown in Figure 3 requires the definition and

properties of the target material. The step "Find Target Material" can
represent any search operation or search procedure. These two procedures
summarize the two problems noted previously: identifying a known
material and locating a known material. The basic questions addressed by
both procedures are: Is Uk one of the known materials Xk in the set X
and if it is, which one? If a material to be identified is not in the set of
known materials, it is characterized and added to the set X of known
materials and its properties are recorded. If a material to be located is not
located or detected, the search either continues or ends. At the end of a
search, an inventory of materials identified or located during the search,
including any located target material(s) of potential use, would be
produced and a report developed. This inventory then could be used for
resource evaluation and future mission planning. Such proceduralized
operations for specific applications may be automated as part of special
purpose adaptive learning algorithm(s) and be part oflarger, more general
adaptive systems.
A potential orebody, if located, is assessed and its reserves and
potential value are recorded for future or immediate mining. Mining and
concentration would normally take place at the location where the
orebody is located. The operations of extraction, processing, fabrication,
and transport could take place at any arbitrary location. These latter
operations may occur repeatedly at different times and in different
sequence for different end products. This is summarized in Table 1. Table
2 is a summary of areas to be considered for exploration through end-use.
Figure 4 shows the typical sequence followed in exploration or
examination. Figure 5 shows the typical sequence followed in the
INTELLIGENT FRAMEWORK FOR MODELING 311

postexploration or post-examination period to mine, extract, transport, and

fabricate products for use. Note that characterization implies monitoring
but not the converse, monitoring requires characterization; similarly,
measurement enables testing.

Table 1. Steps in mining and post-mining activities.

Mine Extract Process Fabricate Transport

Mine X X X X X

Extract X X X

Process X X

Fabricate X

Transport X X X

PQ~~ibl~ A~DQn~ b~arJing lQ ~ng Mamrial U~~-->

Blast Fragment Size Analysis

Barron and others (1994) applied neural networks to differentiating

and understanding the relationships between mining blast fragment
distributions and orebody textures, layering, and crack patterns. Blasting,
crushing, grinding, and milling efficiencies could be improved by
understanding such relationships (Barron and others, 1994). The problem
selected was to recognize fragment size in "muck" piles from digitized
images despite daily changes in perspective, lighting, ore size, and ore
characteristics. Ideally, blasted ore fragments in truck loads must be
sorted by hand to determine size distributions, a prohibitive task. The
problem is to determine if neural networks can do the job easier. The
procedure used and results are described next (modified after Barron and
others, 1994).
312 PLANSKY AND OTHERS

Table 2. Exploration considerations for Earth or space-based mining.

Goal: Explore or examine, define objectives or target materials before

proceeding.

1.1. Locate Target Body or Source of Target Material

1.1.1. Possible Regions to Observe in Transit - List
1.1.1.1. Space Plasmas
1.1.1.2. Above Body Surface
1.1.1.3. Surface Topography
1.1.1.4. Structural Features
1.1.1.5. Surface Materials
1.1.1.6. Deep Regions
1.1.1.7. Minerals
1.1.1.8. Rocks
1.1.1.9. Ore Materials
1.2. Possible Actions
1.2.1. Characterize (implies monitoring is possible)
1.2.1.1. Measure Properties
1.2.1.1.1. Overall
1.2.1.1.1.1. volume or extent
1.2.1.1.1.2. orientation
1.2.1.1.1.3. shape
1.2.1.1.1.4. surface form
1.2.1.1.1.5. structure
1.2.1.1.2. Temporal
1.2.1.1.3. Spatial
1.2.1.1.4. Physical
1.2.1.1.5. Chemical
1.2.2. Identify
1.2.3. Inventory
1.2.3.1. Amount
1.2.4. Evaluate
1.2.4.1. Uses
1.2.4.2. Value
1.2.5. Record and Report
1.2.6. Change Location
1.2.7. Locate Next Target Body
1.3. Post Exploration
1.3.1. mining and concentration
1.3.2. extraction
1.3.3. fabrication
1.3.4. transport
1.3.5. use
 Obseruation Object(s), ~
Mission Knowledge Base, ~
Definition Goals, [uents, Rctions ~
Module(s)

~------------------~~j.-------------------~ i
Data Transmission
EHplore ~
j I-rj
[Hamine o
System Rrriual Characterize ~
HOME Refurbishing -------...J Sys t em
at Identify
BRS[ . Mission 1---,.o""J Depa rt u re -. --. 8
Objectlue Inuentory
Preparation Rssess
Report

DONE_TH I S_OBJECTI UE:

I
NeHt Objectiue or
Return to Home or
Discard System
CoN
......
CoN
Figure 4. Typical examination, characterization, and assessment sequence.
 IoN

.::.
-
Obseruation Object(s),
Mission Knowledge Base,
Definition Goals, Euents, Rctlons
Module(s)

=::== =
Data Transmission

'1 MINE_RESOURCE:
System EHtracl,
Rrriual
IIOME ) _ . \ Refurbishing ...
-,.. at Transport,
BRSE Mission
....... System
Departure 1-.
Objectiue Fabricate,
Preparation Use

til
~
DONE_TIll S_OBJECTI DE: ~
NeHt ObJecllue or
Return to Home or ~
Discard System
j
til
Figure 5. Typical post-examination. characterization. and assessment sequence for mining to transport.
INTELLIGENT FRAMEWORK FOR MODELING 315

Procedure

Ore piles were photographed and the photographs were digitized. The
ore piles were sorted by hand to give the neural network training sets.
The established back propagation (BP) neural network algorithm was
used. The BP network was trained to recognize correctly ore size by
class: 60%, 70%, 80%, 90% finer than 8 inches. In addition to the ability
of the neural network to classify correctly the training set, the predictive
ability was tested. The procedure included taking characteristic
"fingerprints" from the digitized images using the method described by
Bottlinger and Kohlus (1993). A two-dimensional Fast Fourier Transform
was taken of the digitized image. The Fourier coefficients were shifted to
give a large "D.C. Gain" peak in the center of the image. From that
center point, the coefficients were averaged over concentric (square)
rings. Plotting the average versus ring number gives a unique fingerprint
that relates directly to a size distribution in the original image. Figure 6
shows such a typical set of fingerprints taken from the second concentric
ring to the thirtieth.

Results

The neural network could be trained to associate a multitude of

images to four size distribution classes. Eighty pictures from two different
lighting regimes were correctly classified. Furthermore, forty pictures not
in the training set were predicted correctly with 98% accuracy. The one
misc1assified picture was later determined to have digitizing errors.
Different network configurations and fingerprints were examined.
Most network configurations seemed to work well. The best network
configuration had six input nodes, sixteen hidden layers, and· one output
node. We used a nonlinear log sigmoidal saturation function, and a
modified back propagation training algorithm from the MATLABiIl
software system. The best results were obtained using a single network
to split the four categories of images into two subcategories, and then two
more networks to split each of the two subcategories. This meant that the
first net divided the images into a fine subcategory containing both 80%
and 90% finer than 8 inches, and a coarse subcategory containing both
60% and 70% finer than 8 inches. The fine subcategory then was divided
by a second net into final categories of 80% and 90% finer than 8 inches.
Likewise, a third net was trained to recognize final categories in the
316 PLANSKY AND OTHERS

Ten Fingerprints from Coarse and Fine Blast Fragments

1-=--~--'---'---'---'---'---'----'---'--1

0.9

~0.8
.!::!
iii
§ 0.7
o
~
~0.6
;g
~0.5
u
...
CD
.§ 0.4
~
"0
~0.31=-- _ _- - - - - -_ _ _ _ _ ___
~
CD
~0.2

0J~ ?> sa e::::: j

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Harmonic Number (Ring Number) of Averaged Fourier Coefficient

Figure 6. Typical fingerprint results.

coarse subcategory of 60% and 70% finer than 8 inches. Various sections
of the fingerprint were used, all with relatively good success. The first six
positions of the fingerprint were sufficient for training.
The pattern recognition ability of neural networks is sufficient to
determine ore size in a variety of lighting conditions. Lighting conditions
can change because neural networks can be repeatedly trained on a
calibration pile. Errors introduced by shadows, overlapping of rocks and
particles, and other variable lighting conditions are automatically
corrected.

An AI Agent Based System Foundation

This subsection describes an intelligent exploration, examination or

characterization system based on the "Andral Model" concept (Plansky,
1992), this structure is similar to the "Black Box-Thin Wire" concept in
(Rich, 1990).
INTELLIGENT FRAMEWORK FOR MODELING 317

Intelligent Systems

Adaptation implies the ability to respond to an unknown event or

event sequence. In addition, the response must be an intelligent or good
one -- good being defined ahead of time. Adaptation also requires the
ability to do task planning and task execution. This is a fundamental and
important difference between an intelligent being or an "AI Agent" (or
Agent), and a so-called "intelligent" system which responds to a series of
predetermined events - the latter cannot learn by itself, it must be trained
in some way or be given the ability to train itself (thus tending toward
intelligence). A subsystem known as an adaptive learning unit (ALU) is
required for real intelligent, adaptive systems. An organization for a
typical ALU is given in Figure 7.
Vere and Bickmore (1990) have constructed an operating prototype
of an Agent for observation and assessment of underwater craft. They
conceive of an Agent to be an integrated artificial intelligence device that
acts in an environment, its adaptive nature being implicit. The Agent
(HOMER) can communicate, reason, plan, perceive, and act in or on its
environment. The prototype operates in a simulated, marine environment.
Vere and Bickmore (1990) maintain, as we do, that such systems are
possible by a resolute effort, with the needed funds and patience on the
part of management. Their contribution is of extraordinary significance.
The prototype device can: communicate with humans in a vocabulary of
800 English words; use a moderate level of grammar; access an episodic
memory; reflect on events; talk, receive commands, answer questions
from a human; and, synthesize and execute an action plan. It is a
LISPbased expert system with procedural systems as support routines.
The structural essence of their Agent is embodied in the work of the
present authors.

A General Model

An intelligent construct, the "Andral Model", was first proposed in

(Plansky, 1992). Figure 1 is a variant of this construct. Variants of its
organization implemented as an adaptive, distributed, artificial intelligent
(DAI) system is shown in Figures 8 and 9. Figure 8 shows the upper
level organization of a system for exploration/examination or
characterization using this concept as an integrated knowledge system as
discussed in the first example.
 w
00
-
Mission
Knowledge Base,
Definition (-
Goals, Euents, Rctions, Function and Support Library
Module(s)

EUENT
LERRNING
MONITORING RND RCTION
EHECUTIUE / - PLRNNING RCTION
AND
EUENT HANDLING
RDRPTRTlON
-L /
• ts
~
00
; 1 EUENTS ~
!.......... "i ·················DRTR RCQUISITION
~-, INFORMRTION INPUT
"
~
'"
MODULAR, INTELLIGENT SENSORSYSTEM(S)

Figure 7. Typical adaptive learning unit. I

INTELLIGENT FRAMEWORK FOR MODELING 319

The Redefinition of an AI A gent

A simple organization and definition for the AI Agent concept is

given in (Plansky, 1992). The proposed definition allows Agents of
different types and component parts or functions, equivalent to what is
shown in Figure 1 complemented by an adaptive learning unit (ALU). It
is readily extended to multiple Agents and families of Agents and puts
the concept of the AI Agent in a common framework for many
applications - small or large, real-time, near real-time, or normal
batch-type processing systems. It also includes such systems as HOMER
ofVere and Bickmore (1990). An improvement of the proposed definition
is: An AI Agent as an ALU and a combination of one or more of:

Parallel and/or sequential andral models

Intelligent sensor systems
Devices
Control systems
Knowledge and information databases
Function and/or software support libraries
Communication systems.

CONCLUDING REMARKS

Human-like intelligent, adaptive systems can be used in many facets

of geologic work. They are the next step in removing man from the
hazardous environments encountered. The challenge before us is in the
immediate development from off-the-shelf hardware and software of
integrated knowledge-based systems. If we do not act in this direction,
other technologists will - some of the most profitable applications are
obvious improvements or extensions of existing systems as shown by the
second example in this paper.
It is not easy to change the technological baseline in some industries.
The Old Testament (Job Chapter 28) tells of a time when slaves and poor
men hung from ropes in darkness using oil lamps for light to search for
riches in the Earth. Similar situations exist in our world today - circa
8,000 years later. The greatest advances have been in our century. The
next century, hopefully, will outshine the last 8,000 years and approach
the dream of geologists, mining engineers, and their fellow scientists of
320 PLANSKY AND OTHERS

fully automated investigation, resource exploration, evaluation, mining,

transport, and processing tools. The door is beginning to open now.

RDRPTIUE
Target Material
LERRNING
Specification UNIT

Computation Inuentory

I nformation and Data Eualuate

Report

Property
Measurement
and Sensors

SENSDR(S)

Figure 8. Organization of exploration and characterization system.

INTELLIGENT FRAMEWORK FOR MODELING 321

Intelligent

V
.Jcontrol
System

lntellige~t. I?ata
AcqUisition
System

Sensors at
Measurement
Interface CONTROL
FEEDBACK

o
o
KEY Neural Systems

Expert Systems
D Procedural Systems

Figure 9. First order, or organization level, of adaptive knowledge system.

ACKNOWLEDGMENTS

Authors Plansky and Barron would like to thank authors Prisbrey and
Glass for their support during their advanced degree programs. The first
author would like to thank Dr. Dan Merriam for his continued support
and encouragement over time and continents, on the lonesome path of a
doctoral student - may we all practice and reflect this dedication towards
our field and others.
322 PLANSKY AND OTHERS

REFERENCES

Barron, C.L., and others, 1994, Neural network pattern recognition of

blast fragment size distributions: lour. Particulate Science and
Technology, v. 12, no. 3, p. 235-242.
Bevan IE., and Hill, I.R.M., 1994. Automating coal mine roof bolting:
U. S. BuMines, Spokane Research Center, in press.
Bottlinger, M., and Kohlus, R., 1993, Particle shape analysis as an
example of knowledge extraction by neural nets: Particle and Particle
Systems Characterization, Deutsche Institut fuer Lebensmitteltechnik,
e. V., v. 10, no. 5, p. 275-278.
Glass, C.E., and others, 1993, Complementing the trend in automated
mining systems: Proc. Second Intern. Symposium on Mine
Mechanization and Automation, p. 609-624.
Hoffman, M.P., 1994, Computer-based monitoring and control of a
masttype roof drill: U. S. BuMines, Spokane Research Center, in
press.
Howie, W.L., and Nichols, T.L., 1994, Telemetric acquisition of dynamic
drill-rock interaction data: Proc. 40th Intern. Symposium of the
Instrumentation Soc. America, p. 341-352.
Kehoe, B., 1992, EAF controller passes intelligence test: Iron Age, v. 8,
no. 3, p. 28-29.
Plansky, L.E., 1992, Application of integrated knowledge systems to the
metallurgy of zircaloy4: unpubl. doctoral dissertation, Univ. Idaho,
172 p.
Rich, E., 1990, Expert systems and neural networks can work together:
IEEE EXPERT, v. 5, no. 5, p. 5-7.
Signer, S.P., and others, 1993, Using unsupervised learning for feature
detection in coal mine roof: Engineering Applications of Artificial
Intelligence, v. 6, no. 6, p. 565-573.
Vere S., and Bickmore, T., 1990. A basic agent: Computational
Intelligence, v. 6, no. 1, p. 41-60.
 CONTRIBUTORS

Frits P. Agterberg, Geological Survey of Canada, 601 Booth Street,

Ottawa KIA OE8, Canada

Rainer Alms, Geologisches Institut der Universitat Bonn, Nul3allee 8, D-

53115 Bonn, Germany

Neil L. Anderson, Department of Geology & Geophysics, University of

Missouri-Rolla, Rolla, Missouri 65401, USA

Stephan Auerbach, Fachbereich Geowissenschaften, Quantitative

Methoden, Universitat Trier, D-54286 Trier, Germany

Lee Barron, College of Mines and Earth Resources, Department of

Metallurgy and Mining Engineering, University of Idaho,
Moscow, Idaho 83843, USA

Ulf Bayer, GeoForschungsZentrum Potsdam, Telegrafenberg A17, D-

14473 Potsdam, Germany

Geoffrey C. Bohling, Kansas Geological Survey, University of Kansas,

Lawrence, Kansas 66047, USA

R. James Brown, Department of Geology & Geophysics, University of

Calgary, Alberta T2N IN4, Canada

Q. Cheng, Department of Geology, University of Ottawa, Ontario KIN

6N5, Canada

323
324 CONTRIBUTORS

John C. Davis, Kansas Geological Survey, University of Kansas,

Lawrence, Kansas 66047, USA

M.J. Duggin, Department of Forest Engineering, SUNY College of

Environmental Science and Forestry, Syracuse, New York 13210,
USA

Andrea Forster, GeoForschungsZentrum Potsdam, Telegrafenberg Al 7, D-

14473 Potsdam, Germany

John A. French, Kansas Geological Survey, University of Kansas,

Lawrence, Kansas 66047, USA

Carl Glass, College of Engineering and Mines, Department of Mining and

Geological Engineering, University of Arizona, Tucson, Arizona
85721, USA

Eric C. Grunsky, British Columbia Geological Survey, 1810 Blanshard

Street, Victoria, British Columbia V8R 3H4, Canada

W.J. (Bill) Guy, Kansas Geological Survey, University of Kansas,

Lawrence, Kansas 66047, USA

John W. Harbaugh, Department of Geological and Environmental

Sciences, Stanford University, Stanford, California 94305, USA

Jan Harff, Institut fOr Ostseeforschung Wamemiinde, Seestral3e 15, D-

18119 Rostock-Wamemiinde, Germany

Ute C. Herzfeld, Fachbereich Geowissenschaften, Quantitative Methoden,

Universitat Trier, D-54286 Trier, Germany

Olaf Kahle, GeoForschungsZentrum Potsdam, Telegrafenberg A17, D-

14473 Potsdam, Germany

Christian Klesper, Geologisches Institut der Universitat Bonn, NuBallee

8, D-53115 Bonn, Germany
CONTRIBUTORS 325

Junfeng Luo, Gesellschaft fUr wasserwirtschaftliche Planung und

Systemforschung mbH, D-12526 Berlin, Germany

B. LUnenschloB, GeoForschungsZentrum Potsdam, Telegrafenberg A17,

D-14473 Potsdam, Germany

Daniel F. Merriam, Kansas Geological Survey, University of Kansas,

Lawrence, Kansas 66047, USA

Ricardo A. Olea, Kansas Geological Survey, University of Kansas,

Lawrence, Kansas 66047, USA

Robert Ondrak, GeoForschungsZentrum Potsdam, Telegrafenberg A17, D-

14473 Potsdam, Germany

Lee Plansky, College of Mines and Earth Resources, Department of

Metallurgy and Mining Engineering, University ofIdaho, Moscow
83843, USA

Keith Prisbrey, College of Mines and Earth Resources, Department of

Metallurgy and Mining Engineering, University ofIdaho, Moscow
83843, USA

Joseph E. Robinson, Department of Geology, Syracuse University,

Syracuse, New York 13244, USA

Helmut Schaeben, Fachbereich Geowissenschaften, Quantitative

Methoden, Universitat Trier, D-54286 Trier, Germany

Esther U. Schlitz, Fachbereich Geowissenschaften, Quantitative Methoden,

Universitat Trier, D-54286 Trier, Gennany

Alla Shogenova, Institute of Geology, Estonian Academy of Sciences,

Tallinn, Estonia

Agemar Siehl, Geologisches Institut der Universitat Bonn, NuBallee 8, D-

53115 Bonn, Germany
326 CONTRIBUTORS

Jom Springer, GeoForschungsZentrum Potsdam, Telegrafenberg A17, D-

14473 Potsdam, Germany

W. Lynn Watney, Kansas Geological Survey, University of Kansas,

Lawrence, Kansas 66047, USA

Claus v. Winterfeld, GeoForschungsZentrum Potsdam, Telegrafenberg

Al 7, D-14473 Potsdam, Germany

Johannes Wendebourg, Institut Fran~ais du Petrole, 1&4 Avenue de Bois-

Preau, F-92506 Rueil-Malmaison Cedex, France

D. Yuan, Desert Research Institute, 755 E. Flamingo Road, Las Vegas,

Nevada 89119, USA
 INDEX
Adaptive knowledge system, 321 basin analysis, 85
adaptive learning unit, 318 basin models, 96
aeromagnetics, 222 basin development, 114
albedo images, 161, 162, 163 Bayesian relations, 263,276
Alberta, 175, 193, 194 beach environments, 11
albite distribution, 37 Bezier form, 141, 143
algebraic map-comparison, 219 bottom-hole temperatures
Anadarko Basin, 266 (BHTs), 222
anamorphosed distances, 271, black box-thin wire, 317
273,274 blast fragment distributions, 315
Anciral model, 317, 319 Box-Cox power transformations,
anisotropic situations, 102 239
ANOVA, 172 boundary conditions, 1, 39
Arkoma Basin, 52 BP neural network, 315
artificial intelligence, 302 British Columbia, 229
artificial rocks, 95
assessment sequence, 313,314 CAD methods, 119
anticlinal folds, 205 calcite cement distribution, 35
AUTOCAD program, 201 calcite cemented sandstone, 33
auto-correlation, 232,233,242 Caledonian Orogeny, 82
autocyclic processes, 8 California, 14, 17
Au mineralization (also see gold California Coast Ranges, 14
mineralization), 236 Canada, 175, 229
Au-associated alteration, 236 Canadian Cordillera, 234
carbonate equilibrium, 32
Back propagation (BP), 315 carbonate sediment accumulation
backward modeling, 116, 123 rate, 49
Baltic Shield, 200 censored distributions, 239
band brightness value, 173 Central Kansas Uplift, 266,
band-time interaction test, 161 271,277
BASIC program, 199

327
328 INDEX

chemical diagenesis, 28 correlation models, 44

classical calculus, 111 correlograms, 233, 242, 253
classification problem, 263 cross-correlation, 232, 233, 242
cluster centroids, 263 cycle orders, 47
compaction, 8 cyclothems, 45
computer graphics, 59, 113
computer mapping, 203 Darcy's law, 99
computer model, 54 data models, 131
computer program languages deltaic complexes, 11
BASIC, 199 dendrograms, 219
FORTRAN, 234,305 depositional rates, 50
Pascal, 304, 305 diagenetic evolution, 30,31,35,36
computer programs/systems diagenetic modeling, 40
AUTOCAD, 201 diagenetic systems, 27
EM algorithm, 240 diffusion-based simulators, 6
GOCAD, 117 digitized images, 315
GEOSTORE, 118 discontinuity, 107
GML procedure, 117, 164, 169 discrete simulators, 3
GOCAD, 117 discrete smooth interpolation
GRAPE, 117 (DSI), 119
HOMER, 319 discriminant analysis, 263
KANMOD,49 dissolution, 32, 38, 175
MAPCOMpR, 218,224 distribution, parameters, 241
MATLABR, 317 drillholes, 291
SEDSIM, 8 dryholes, 267
SPANSTM, 234 dynamic modeling, 114
SPFAC, 231
Splus, 234 East Texas, 12, 13
SURFACE III, 223 Edward's reef trend, 12, 14
conceptual model, 44, 82 effective conductivity, 102, 105
conditional Markov effective transport properties, 82,
simulation, 293 95
conditional simulation, 285,291 efficient transport coefficients, 99
conductive heat transport, 85 103
conductive models, 86 Eifel Mountains, 120
constant sum data, 241 Eifel Nappe, 81
CO2 concentration, 34, 36 EM algorithm, 240
correlation coefficient, 216 Erft Block, 127, 130
INDEX 329

Estonia, 199 geometric modeling, 116,

Estonian Homocline, 199 129, 131
eustatic sealevel, 8,49 geostatistical methods, 282, 295
expert system, 302, 304 geostatistical models, 59
exploration, 306 geostatistical simulation, 270
GEOSTORE, 118
Fast Fourier Transform (FFT), geothermics (also see thermal),
315 222
fault surfaces, 125, 136 geothermal field (also see
fingerprints, 315,316 thermal field), 88
finite difference method, 100 geothermal studies (also
flood frequencies, 17 see heat conduction), 90
flow model, 30 Germany, 80, 120
flow rates, 36 GML procedure, 117, 164, 169
fluid discharge, 12 "FACTOR", 164
fluid flow (also see subsurface "MODEL", 164
fluid flow), 32, 88 GOCAD, 117
FORTRAN programs, 234,305 gold mineralization (also see
forward modeling, 7, 44 Au mineralization), 229, 255
fractals, 111 GRAPE, 117
gravity, 220
Gardermoen delta, 18
general model, 319 Heat conduction (also see
generated artificial media, 97 geothermal studies), 90
GEOBODY,74 heat-flow density, 85
geochemical data, 229 heat-flow predictions, 79
geochemical responses, 247 Hermite interpolation problem,
geochemical sampling pattern, 247 142
geochemistry, 40 heterogeneities, 72
GEOCON, 117 HOMER, 319
Geographic Information System hot springs, 82
(GIS), 116 Hugoton Embayment, 266, 277
Geographical Information hydrocarbon exploration, 2
System (SPANS), 244 hydrocarbon reservoirs, 11
geologic environment, 306
geological data, 282 Information reporting systems, 302
geological model, 80, 82, 301 intelligent framework, 301
geometric-form simulators, 4 intelligent systems, 305
330 INDEX

interdisciplinary approach, 306 Markov model, 283, 286, 293,

interpolant g, 141, 145 Markov simulation, 285
interpolation models, 44 MATLABR, 317
inverse models, 44 maturation modeling, 40
inverse reasoning, 115 means, 96, 103
isopachous maps, 199,206 Midcontinent, USA, 43
isostatic compensation, 8 mineral exploration, 264
isotopic analyses, 71 mineral-fluid interaction, 28
isotropic medium, 102, 103 mineral reactions, 30
isovalue maps, 126 mineralogical composition, 31
mining, 314
KANMOD,49 Mississippian structure, 220
Kansas, 49,59, 71, 218, 222, 264 model, 169, 282
knowledge systems, 306 model evaluation, 127
Krefeld Block, 130 modeling, 113,203,285
Kriging (also see universal backward, 123
kriging), 210,264 diagenetic, 39
Kunda River, 208 forward, 7, 44
geometric, 116,129,131
Lake Valley, 159 numerical, 27, 96
LANDSAT, 154, 162, 163 quantitative geologic, 72
Laplace equation, 96 stratigraphic, 44, 72
Leduc salt basins, 190 surface, 119
Levin's theorem, 142 thermal, 79
linear eliptic equations, 100 three dimensional, 72
lithogeochemical data. 244 visualization. 75
lithological patterns, 257 modeling faults, 123
lithological radiance, 155 modeling systems, 114
lithology-time interaction test, 168 models
look-ahead sensor systems, 306 Andral, 317,319
LOTUS spreadsheet, 199 basin, 96
Lower Rhine Basin, 113, 120, 130 computer, 54
conceptual, 44, 82
Mahalanobis' distances, 263 conductive, 86
Mainz Basin, 120 correlation, 44
MAPCOMP, 218,224 data, 131
map comparison/integration, 216 flow, 30
marine geology, 264 general, 319
INDEX 331

geological, 80, 82, 301 orogenetic front, 79

interpolation, 44
inverse, 44 Pairwise comparison, 215
Markov, 283,286,293 paleoreconstruction, 189
semivariogram, 271, 272 parallel distributed processing
modem fluvial deposits, 11 (PDP) network, 302
MountVVilson, 160 Pascal, 304, 305
MSS data, 153, 168 pattern recognition, 114, 317
multielement relationships, 257 percolation, 96
multivariate analysis, 230, 244 percolation theory, 110
percolation threshold, 106, 107
Navier-Stokes equations, 8 permeability, 28
Nemaha Uplift, 52 Pennsylvanian carbonates, 44
Netherlands (The), 120 Pennsylvanian cyclothems, 48
network configurations, 315 petroleum reservoirs, 43, 72
neural networks, 302, 315 petrophysical information, 267
Nevada, 156 petrophysical properties, 30
Newton-Cotes numerical Pickett plots, 60
integration, 137 pore filling cement, 32
nonlinear dynamic system, 39 pore water flow, 31
nonparametric characterization, porosity, 28
264 porosity evolution, 36
North Sea Rift, 120 porous layers, 71,96
Norway, 12, 15 power laws, 96, 103, 108
numerical modeling, 27, 96 Precambrian configuration, 266
numerical realizations, 145 precipitate, 38
numerical simulation, 28 predictor variables, 277
pressure gradient, 30
Object-based simulators, 3 principal components analyses, 241
oil prospecting, 263 procedural systems, 304
oil wells, 267 pseudoization method, 96
oil and gas reservoirs, 2 Pff-conditions, 84,91
onlap/offlap relationships, 71 Pyramid rule, 136
Ontos database-management
system, 118 Quadratic functions, 233
orebody, 310 quadratic spline, 141, 143
orebody textures, 315 quantitative geologic modeling, 72
organic material, 40
332 INDEX

R2 coefficients, 247 sedimentary process simulators, 5

radiance, 173 sedimentological patterns, 97
radioactive decay, 86 SEDSIM, 8
reconstruction, 135 seismic profile, 193, 194
reflection of radiance, 154 seismic well-log data, 179
regional geochemical sampling, 237 semivariogram, 284
regional structure, 199 semivariogram models, 271, 272
regionalized classification, 264 set-valued interpolation, 137
regionalized variables, 264, 282 signal to noise ratios, 242
regression analysis, 220 silica cement distribution, 33
remotely sensed data, 153 similarity coefficients, 219
rendering, 135, 147 Simpson's rule, 137
representative volumes, 101 simulation, 1,96,301
reservoir heterogeneity, 72 conditional, 285, 289
residual structure maps, 204 geostatistical, 269
residual thickness maps, 206 Markov, 285
resultant map, 216,219 numerical, 28
Rhenish Massif, 81, 120 stochastic, 263, 265
rock fabric, 114 simulated annealing,' 4
rock salts, 175 simulation results, 295
Rur Block, 130 simulation of stacked oolites, 51
Russian Platform, 200 South Belridge field, 18, 20, 22
SPANSTM, 234
Seasonal factors, 174 spatial analyses, 203
salt dissolution, 179, 190, 192, spatial correlation, 253
193 spatial evolution, 36
salt leaching, 191 spatial factor analysis (SPFAC),
San Francisco Bay region, 17 229,253
San Joaquin Valley, 18 spatial map data, 215
sandstone diagenesis, 27 spatial qualitative variables, 281
satellite images, 154 spatial structure analysis, 204,
sealevel curve, 49 286
seasonal factors, 153 spatial thickness analysis, 206
sediment accumulation rates, 50 spatial variability, 283
sediment discharge, 12 SPFAC (spatial factor
sedimentary basins, 80, 199 analysis), 231
sedimentary heterogeneities, 3 Splus, 234
sedimentary sequences, 6 stationary transport conditions, 96
INDEX 333

statistical methods (also see Stratamodel, 63

geostatistical methods) stratigraphic framework, 53
ANOVA, 172 stratigraphic modeling, 44, 72
auto correlation, 232, 244, 242 stratigraphic, surfaces, 126
Bayesian relations, 263,276 stratigraphic, thinning, 71
correlation coefficients, 216 structural analysis, 199,209,271
correlograms, 233, 242, 253 structural anisotropy, 97, 109
cross-correlation, 232, 233, subaerial exposure, 71
242 submarine slope failure, 8
dendrograms, 219 subsidence, 8
discriminant analysis, 263 subsidence rate, 49
Kriging, 210, 263 subsurface fluid flow (also see
Mahalanbois'distances, 263 fluid flow), 18
Markov models, 283,287,289 subsurface structures, 80
means, 96, 103 Sulphurets area (Canada), 236,
multivariate analysis, 230 244,253
244 surface modeling, 119
pairwise comparisons, 215 SURFACE III, 223
principal components analysis,
241 Target materials, 311
R2 coefficients, 247 target variable, 277
regionalized variables, 264, 282 tectonic uplift, 8
regression analysis, 220 temperature and burial depth, 31
semivariogram, 284 temporal evolution, 36
similarity coefficients, 219 temporal variations, 153
spatial factor analysis, 229,253 thematic maps, 263
transition probability, 281, 282 thermal conductivity, 83,86
trend analysis, 203,210 thermal field (also see geothermal
universal Kriging, 269 field), 88
statistical trend, 155 thermal modeling, 79
statistically isotropic media, 110 thermodynamic data, 39
step method, 136 three-dimensional
stereo views, 114, 129 body, 136
stochastic simulation, 263, 265 GIS, 116
stochastic simulators, 3, 4 modeling, 72
stratal geometrics, 58 views, 204
stratal packaging, 45 visualization, 59
stratal pattern, 53 time factors, 154
334 INDEX

topologies, 150 Valley and Range Province, 158

transfonnation, 82 Vancouver Island, 234, 245
transition probability, 281,289 Variscan Foldbelt, 120
transition probability model, 286 Variscian Front, 81
transport coefficients, 83 vegetation effect, 154
trapezoid rule, 136 Venlo Block, 130
trend analysis, 203,210 Victory oil field, 53, 59
triangulation meshes, 121 visualization facilities, 117
visualization modeling, 75
Uncertainty, 88,264 volume modeling, 119
unconformity-bounded sequence, volumetrics, 135, 140, 147
47
univariate Hermite interpolation, Water-rock interactions, 28
141 well data, 64
universal Kriging (also see Wilson Creek Range, 158
Kriging), 271