You are on page 1of 414

Core-Log Integration

Geological Society Special Publications


Series Editors." A. J. FLEET
A. C. MORTON
A. M. ROBERTS

GEOLOGICAL SOCIETY SPECIAL PUBLICATION NO. 136

Core-Log Integration
EDITED BY

P. K. H A R V E Y & M. A. L O V E L L
University of Leicester, UK

1998
Published by The Geological Society London

THE GEOLOGICAL SOCIETY


The Society was founded in 1807 as The Geological Society of London and is the oldest geological society in the
world. It received its Royal Charter in 1825 for the purpose of 'investigating the mineral structure of the Earth'.
The Society is Britain's national society for geology with a membership of around 8500. It has countrywide
coverage and approximately 1500 members reside overseas. The Society is responsible for all aspects of the
geological sciences including professional matters. The Society has its own publishing house, which produces the
Society's international journals, books and maps, and which acts as the European distributor for publications of
the American Association of Petroleum Geologists, SEPM and the Geological Society of America.
Fellowship is open to those holding a recognized honours degree in geology or cognate subject and who have at
least two years' relevant postgraduate experience, or who have not less than six years' relevant experience in
geology or a cognate subject. A Fellow who has not less than five years' relevant postgraduate experience in the
practice of geology may apply for validation and, subject to approval, may be able to use the designatory letters C
Geol (Chartered Geologist).
Further information about the Society is available from the Membership Manager, The Geological Society,
Burlington House, Piccadilly, London WIV 0JU, UK. The Society is a Registered Charity, No. 210161.

Published by The Geological Society from:


The Geological Society Publishing House
Unit 7, Brassmill Enterprise Centre
Brassmill Lane
Bath BA1 3JN
UK
(Orders: Tel. 01225 445046
Fax 01225 442836)

Distributors

USA
AAPG Bookstore
PO Box 979
Tulsa
OK 74101-0979
USA
(Orders: Tel. (918) 584-2555
Fax (918) 560-2652)

First published 1998

Australia
The publishers make no representation, express or
implied, with regard to the accuracy of the information
contained in this book and cannot accept any legal
responsibility for any errors or omissions that may be
made.
9
Geological Society 1998. All rights reserved.
No reproduction, copy or transmission of this
publication may be made without written permission.
No paragraph of this publication may be reproduced,
copied or transmitted save with the provisions of the
Copyright Licensing Agency, 90 Tottenham Court
Road, London WlP 9HE. Users registered with the
Copyright Clearance Center, 27 Congress Street,
Salem, MA 01970, USA: the item-fee code for this
publication is 0305-8719/98/$10.00.

Australian Mineral Foundation


63 Conyngham Street
Glenside
South Australia 5065
Australia
(Orders: Tel. (08) 379-0444
Fax (08) 379-4634)

India
Affiliated East-West Press PVT Ltd
G- 1/ 16 Ansari Road
New Delhi 110 002
India
(Orders: Tel. (11) 327-9113
Fax (11) 326-0538)

Japan
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the
British Library.
ISBN1-86239-0169
ISSN0305-8719

Typeset by Bath Typesetting, Bath.


Printed by The Alden Press, Osney Mead, Oxford,
UK.

Kanda Book Trading Co.


Cityhouse Tama 204
Tsurumaki 1-3-10
Tama-shi
Tokyo 206-0034
Japan
(Orders: Tel. (0423) 57-7650
Fax (0423) 57-7651)

Contents
Preface

vii

Measurement, sealing and calibration


BRISTOW, C. S. & WILLIAMSON,B. J. Spectral gamma ray logs: core to log calibration,
facies analysis and correlation problems in the Southern North Sea

CORBETT,P. W. M., JENSEN,J. L. & SORBIE, K. S. A review of up-scaling and cross-scaling


issues in core and log data interpretation and prediction
DUNCAN,A. R., DEAN,G. & COLLIE,D. A. L. Quantitative density measurements
from X-ray radiometry
17

HARVEY,P. K., BREWER,T. S., LOVELL,M. A. & KERR,S. A. The estimation of modal
mineralogy: a problem of accuracy in core-log calibration

25

LOVELL,M. A., HARVEY,P. K., JACKSON,P. D., BREWER,T. S. WILLIAMSON,G. &


WILLIAMS,C. G. Interpretation of core and log data-integration or calibration?

39

RAMSEY,M. H., WATKINS,P. J. & SAMS,M. S. Estimation of measurement uncertainty


for in situ borehole determinations using a geochemical logging tool

53

Physical and chemical properties

AHMADI,Z. M. & COE, A. L. Methods for simulating natural gamma ray and density
wireline logs from measurements on outcrop exposures and samples: examples from
the Upper Jurassic, England

65

HERRON,M. M. & HERRON,S. L. Quantitative lithology: open and cased hole


application derived from integrated core chemistry and mineralogy database

81

KINGDON, A., ROGERS, S. F., EVANS, C. J. (~ BRERETON, N. R. The comparison


of core and geophysical log measurements obtained in the Nirex investigation of
the Sellafield region

97

LAUER-LEREDDE,C., PEZARD,P. A., TOURON,F. & DEKEYSER,I. Forward modelling of


the physical properties of oceanic sediments: constraints from core and logs, with
palaeoclimatic implications

115

WADGE,G., BENAOUDA,D., FERRIER,G., WHITMARSH,R. B., ROTHWELL, R. G. &


MACLEOD, C. Lithological classification within ODP holes using neural networks
trained from integrated core-log data

129

Petrophysical relationships
BASTOS, A. C., DILLON, L. D., VASQUEZ,G. F. & SOARES,J. A. Core-derived acoustic,
porosity & permeability correlations for computation pseudo-logs

14I

DENICOL, P. S. & JING, X. D. Effects of water salinity, saturation and clay content on
the complex resistivity of sandstone samples

147

SAMWORTH, J. R. Complementary functions reveal data hidden in your logs

159

SHAKEEL, A. & KING, M. S. Acoustic wave anisotropy in sandstones with systems of


aligned cracks

173

vi

CONTENTS

WIDARSONO,B., MARSDEN,J. R. & KING, M. S. In situ stress prediction using


differential strain analysis and ultrasonic shear-wave splitting

185

WORDEN, R. H. Dolomite cement distribution in a sandstone from core and wireline


data: the Triassic fluvial Chaunoy Formation, Paris Basin

197

WORTHINGTON,P. F. Conjunctive interpretation of core and log data through


association of the effective and total porosity models

213

Xu, S. & WHITE, R. Permeability prediction in anisotropic shaly formations

225

Integration of core and borehole images


GOODALL,T. M., Me~LLER,N. K. & RONNINGSLAND,T. M. The integration of
electrical image logs with core data for improved sedimentologicaI interpretation

237

HALLER,D. & PORTURAS,F. How to characterize fractures in reservoirs using


borehole and core images: case studies

249

JACKSON,P. D., HARVEY,P. K., LOVELL,M. A., GUNN, D. A., WILLIAMS,C. G. &
FLINT, R. C. Measurement scale and formation heterogeneity: effects on the integration
of resistivity data

261

LOFTS, J. C. & BRISTOW,J. F. Aspects of core-log integration: an approach using


high resolution images

273

MAJOR, C. O., PIRMEZ, C., GOLDBERG, D. & LEG 166 SCIENTIFICPARTY High-resolution
core-log integration techniques: examples from the Ocean Drilling Program

285

Applications and case studies


AYADI M., PEZARD, P. A., LAVERNE, C. & BRONNER, G. Multi-scalar structure
at DSDP/ODP Site 504, Costa Rica Rift, I: stratigraphy of eruptive products and
accretion processes

297

AYADI, M., PEZARD, P. A., BRONNER, G., TARTAROTTI, P. & LAVERNE, C.


Multi-scalar structure at DSDP/ODP Site 504, Costa Rica Rift, III: faulting and fluid
circulation. Constraints from integration of FMS images, geophysical logs and core data

311

BARCLAY,S. A. & WORDEN, R. H. Quartz cement volumes across oil-water contacts


in oil fields from petrography and wireline logs: preliminary results from the
Magnus Field, Northern North Sea

327

BREWER,T. S., HARVEY,P. K., LOVELL,M. A., HAGGAS,S. WILLIAMSON,G. &


PEZARD, P. A. Ocean floor volcanism: constraints from the integration of core and
downhole logging measurements

341

BOCKER, C. J., DELIUS, H., WOHLENBERG,J. LEG 163 SHIPBOARDSCIENTIFICPARTY.


Physical signature of basaltic volcanics drilled on the northeast Atlantic volcanic
rifted margins

363

GONq:ALVES,C. A. & EWERT, L. Development of the Cote d'Ivoire-Ghana


transform margin: evidence from the integration of core and wireline log data

375

TARTAROTTI, P., AYADI, M., PEZARD, P. A., LAVERNE, C. & DE LAROUZII~RE,F. D.


Multi-scalar structure at DSDP/ODP Site 504, Costa Rica Rift, II: fracturing
and alteration. An integrated study from core, downhole measurements and borehole
wall images

391

Index

413

Preface
Core and log measurements provide crucial information about subsurface formations. Their usage,
either for integration or calibration, is complicated by the different measurement methods employed,
different volumes of formation analysed, and in turn, the heterogeneity of the formations. While the
problems of comparing core and log data are only too well known, the way in which these data can
be most efficiently combined is not at all clear in most cases. In recent years there has been increased
interest in this problem both in industry and academia, due in part to developments in technology
which offer access to new types of information, and in the case of industry, pressure for improved
reservoir models and hydrocarbon recovery. The application of new numerical methods for
analysing and modelling core and log data, the availability of core scanning facilities, and novel core
measurements in both two and three dimensions, currently provide a framework for the development
of new and exciting approaches to core-log integration.
This Special Publication addresses some of the problems of core-log integration encountered by
scientists and engineers from both industry and academia. The diverse nature of the contributions in
this volume are an expression of the value and need to understand core and log measurements, and
the way in which they can be combined to maximum effect. Contributions range geologically from
hydrocarbon-bearing sediments in the North Sea to the volcanic rocks that form the upper part of
the oceanic crust. In order to constrain this diversity for presentation the volume has been divided
into five sections and starts with 'Measurement, scaling and calibration', 6 papers concerned purely
with aspects of core and,or log measurements themselves including cross-correlation, upscaling,
measurement uncertainty and accuracy. Subsequent sections include (2) 'Physical and chemical
p r o p e r t i e s ' - 5 papers, (3) 'Petrophysical relationships'-8 papers, (4) 'Integration of core and
borehole i m a g e s ' - 5 papers and (5) 'Applications and case s t u d i e s ' - 7 papers. All papers were
submitted in response to an open call for contributions so, within the constraints of work loads and
other factors, may be considered to represent a fair snapshot of recent developments in Core-Log
Integration.
The volume arises from a meeting of the Borehole Research Group of the Geological Society and
the London Petrophysical Society (London Chapter of the Society of Professional Well Log
Analysts) held in London in September 1996. The editors are particularly grateful to Gail
Williamson both for the organization of the meeting and for persistence in coaxing authors,
reviewers, and editors; also to Jo Cooke at the Geological Society Publishing House for her
continuous support in the production of this volume. We also wish to thank all those who undertook
the often arduous job of reviewing the manuscripts, and without whose help this volume would have
been that much poorer.
Peter K. Harvey & Michael A. Lovell
Leicester University

Spectral gamma ray logs: core to log calibration, facies analysis and
correlation problems in the Southern North Sea
C. S. B R I S T O W 1 & B. J. W I L L I A M S O N 2

1Research School of Geological and Geophysical Sciences, Birbeck College and UCL, Gower
Street, London WC1E 6BT
2 Present address." Department of Mineralogy, The Natural History Museum, Cromwell
Road, London S W7 5BD
Abstract: The aim of this study is to test the usefulness of spectral gamma ray logs in
subsurface correlation, lithofacies description and the interpretation of depositional
environments of Namurian and Dinantian sandstones in the southern North Sea.
Lithofacies and depositional environments were identified from core descriptions and
compared with spectral gamma ray logs from thirteen boreholes. The results show that
lithofacies and sedimentary environments can be discriminated within single wells. However,
there is too much variation between wells to make an unequivocal assessment of
depositional environment on the basis of spectral gamma ray logs alone. Comparison of
stratigraphically correlated sandstones shows that variations between wells are often greater
than variations between lithofacies. The differences between correlated sandstones using
spectral gamma ray logs are largely attributed to changes in the logging environment,
mainly mud characteristics, borehole quality and contractor. In addition, the occurrence of
negative numbers for uranium and potassium in some wells indicates that the algorithm
used to calculate elemental concentrations may be in error. For sandstones with a low total
gamma ray response, small errors associated with tool calibration and data processing make
a comparatively large difference to results, which has made detailed correlation of
sandstones untenable. The most significant problem is the correction factor for potassium in
KC1 drilling mud.

G a m m a ray logs are an essential tool for


subsurface correlation and gamma ray log curve
shapes or signatures are often used as the basis
for interpreting ancient sedimentary environments (Selley 1978; Cant 1992). The spectral
gamma ray tool measures radiation produced by
the radioactive decay of naturally occurring
radioactive elements. The most common naturally occurring radioactive elements in sedimentary rocks are potassium, thorium and uranium.
As each of these elements decay they give off
gamma radiation of a particular energy measured in MeV (millions of electron volts). The
principle energies for each element are 1.46 MeV
for potassium, 0.68MeV for thorium, and 1.12
and 0.98 MeV for uranium (Desbrandes 1985).
The radiation from potassium (K 40) is a single
energy while uranium and thorium have a series
of isotopes producing radiation with a range of
energies which overlap (Rider 1986). In addition, Compton scattering leads to a reduction in
energy and the total gamma radiation is a
complex spectrum. The spectral gamma ray tool
samples the spectrum around specific energy
levels, 1.46MeV for potassium, 1.76MeV for
uranium and 2.62 MeV for thorium (Rider 1986;

Dresser Atlas 1992). These measured values are


then recalculated to estimate the proportions of
potassium, thorium and uranium, expressed as
percentages or API units.
Spectral gamma ray data recorded from
outcrop have been used for correlation and to
define sediment facies in Upper Carboniferous
deltaic sediments (Myers & Bristow 1989;
Davies & Elliot 1995). Spectral gamma ray data
have also been used to characterize marine
bands in the Upper Carboniferous (Archard &
Trice 1990; Leeder et al. 1990). In this study we
have attempted to apply the methodology of
Myers & Bristow (1989) to Carboniferous rocks
in the Southern North Sea. We have examined
spectral gamma ray logs from thirteen wells in
the Southern North Sea (Wells 1-13). Sedimentary logs of core were available for seven of the
boreholes and stratigraphic information showed
that two sandstone units 'A' and 'B' could be
correlated between three and six of the wells,
respectively. Unfortunately due to confidentiality agreements we are unable to identify the wells
in question or the names of the correlated units.
G a m m a ray logs are affected by hole conditions, in particular an oversized hole can lead to

BRISTOW,C. S. & WILLIAMSON,B. J. 1998. Spectral gamma ray logs: core to log calibration, facies
analysis and correlation problems in the Southern North Sea In. HARVEY,P. K. LOVELL,M. A. (eds)
Core-Log Integration, Geological Society, London, Special Publications, 136, 1-7

C.S. BRISTOW & B. J. WILLIAMSON

a decrease in gamma ray response. To provide


some control on data quality, gamma ray
measurements were plotted against caliper data.
Another common borehole effect is the use of
KC1 in drilling mud. The potassium in the
drilling mud produces an increase in absolute
values of potassium on the spectral gamma ray
log. This is supposed to be corrected in the
processing and we have assumed that the
contractors have made the right corrections to
the data. However, there appears to have been
no correction for variations in mud chemistry
down hole. Of the thirteen wells that we have
examined, ten were logged by one contractor
and the remaining three were logged by a second
contractor (Table 1). Seven of the wells were
drilled using an oil based mud, five were drilled
using a water based KCI mud and one was
drilled with a salt saturated polymer.
Table 1. Summary of well characteristics

Well
number

Logging
contractor

Drilling
mud

Correlated
sandstone
A

KC1

2
3
4
5
6
7
8
9

1
1
2
2
1
!
1
1

Oil
KC1
Oil
Oil
Oil
KC1
Oil
Oil

A
B

10

KC1

11
12
13

2
1
1

polymer
KC1
Oil

B
B
B

Methods

A simple seven category lithofacies scheme was


adopted for the cored wells with classification on
the basis of lithology and sedimentary structures: cross-stratified sandstones, silty sandstones, s a n d y siltstones, claystones, coal,
limestones and rooted beds. After depth matching of core and log and corrections for core to
log slip, depth intervals for each lithofacies were
defined. Lithological boundaries were picked at
the shoulder of gamma ray curves to take
account of readings which 'smear' across bed
boundaries. The gamma ray log data were then
assigned a lithofacies classification on the basis
of core descriptions. Where core log depths were
metric and the wireline log data in feet, the data
were recalculated to metric units. For several of
the wells, potassium values were given in deci. %
rather than as percentages. As deci. % units give

good separation of curves on log plots, all


potassium values were converted to deci. %.
Having reduced the log data to cored depth
intervals, element data and element ratios were
plotted for each well on cross plots and logs, and
between wells on cross plots and box plots.
Observations

Potassium-thorium cross plots discriminate


lithofacies
Cross plots of potassium against thorium,
thorium against uranium and potassium against
uranium were produced for each cored well. The
cross plots provide an easy to read display of the
range of measurements from each well by facies
and for comparing each facies between wells.
The cross plot of potassium against thorium
from Well 1 (Fig. 1) is a typical example; It
shows limestones with very little gamma radiation clustered in the lower left-hand corner of
the plot with cross-stratified sandstones in a field
around 0.01 deci % potassium and 10ppm
thorium. Finer grained lithologies, silty sandstones, sandy siltstones and claystones are less
well discriminated but all show relatively high
potassium and thorium. One intriguing feature
of this plot (Fig. 1) is the negative values for
potassium in the limestones. Negative values for
potassium are very small, less that 0.005 deci %,
and were only found in Well 1; however,
negative values for uranium were found in wells
2, 6 and 7. The negative values indicate a
problem with the algorithm used by the contractor to calculate elemental concentrations.
Other wells such as Well 5 (Fig. 2) show clear
discrimination between lithofacies although the
absolute values are different from those in Well
1. Cross-stratified sandstones generally contain
slightly less potassium and less thorium, most
claystones have relatively high values of potassium but a few have almost no potassium.
Changes within lithofacies for a particular well
could be due to differences in the detrital
composition and diagenetic history of the
claystones. However, low values may also be
encountered where the gamma ray response is
averaged across a bed boundary. The resolution
of a gamma ray tool is typically about 3 0 4 0 cm
depending on the speed that the tool was run
(Rider 1986) and where the sampling interval
coincides with a bed boundary the measurement
will not represent either lithology, but a mixture
of two different lithologies. Another possible
source of error is in the core to log calibration
where reconstruction of a core may lead to small
offsets in the core to log slip.

SPECTRAL GAMMA RAY LOG CORRELATION PROBLEMS

Fig. 1. Potassium and thorium cross plot for Well 1 showing good discrimination of lithofacies with limestones in
the lower left corner and fine-grained claystone and siltstones in the top right.

Fig. 2. Potassium and thorium cross plot for Well 5 showing good discrimination of lithofacies. The lithofacies
have similar relative values to those in Welt 1 (Fig. 1) although absolute values for each lithofacies are slightly
different.

Comparison of correlated sandstones


On Fig. 3, which shows cross plots of crossstratified sandstones from all seven cored wells,
the data for individual wells form distinct
clusters. The differences within wells is less than
the differences between wells, which suggests
some systematic changes between wells. Geologic factors such as a change in provenance or

diagenesis are unlikely to produce such a clear


systematic difference. Other possible explanations are that the sandstones were deposited in
different deltaic environments: mouth bar, distributary channel or shoreface; or that the
sandstones are stratigraphically different and
have different detrital sources or different
diagenetic histories. One way of testing these
hypotheses is to examine the character of
correlated sandstones.

C.S. BRISTOW & B. J. WILLIAMSON

Fig. 3. Cross plots of cross-stratified sandstones between wells showing a loose grouping of all the data in the
lower left hand corner of the cross plot. Measurements from individual wells tend to be tightly grouped and the
difference between wells appears to be greater than the differences within a well.

Fig. 4. Cross plot of potassium against thorium for the correlated sandstone Unit A shows the same sandstone in
three different wells plotting in slightly different areas, note the lack of overlap between wells with lower
potassium values in Well 1 which was drilled with a KC1 mud.
Unit A.
This has been correlated stratigraphically between three wells. The cross plot of potassium
against t h o r i u m (Fig. 4) shows the same
sandstone in three different wells plotting in
slightly different areas. There is almost no
overlap between the three data sets and although
the trends appear to be similar in each well, there
is a clear difference in the absolute values. Some

variation between wells could be due to lateral


facies changes, but these are unlikely to have
produced the observed shift in absolute values.
The similar shape of the trends combined with
their differences in absolute values indicates a
systematic change between wells which we
attribute to changes in the borehole environment. The factors most likely to affect the logs
are caving, the use of different drilling fluids, and

SPECTRAL GAMMA RAY LOG CORRELATION PROBLEMS

Fig. 5. Cross plot of potassium against thorium for Unit B showing a consistent trend in the data for Wells 9, 10,
12 and 13. Well 11 appears to lie off trend with significantly higher potassium and thorium content which can be
attributed to an error in the correction factor for KCI in the drilling mud..

Unit B.

Fig. 6. Box plot of total gamma for sandstones and


claystones. Claystones usually have higher total
gamma than sandstones although there is some overlap in Wells 3 and 6. The lower than usual values in
these claystones may be due to deposition in an
interdistributary bay rather than a prodelta environment.
variations in the procedures of different logging
contractors. There is very little difference in
caliper data between wells and no evidence for
significant caving, which leaves two possible
e x p l a n a t i o n s for the differences observed.
Firstly, Well 1 was drilled with water based
mud, while Wells 4 and 8 were drilled with an oil
based mud. Secondly, Wells 1 and 8 were logged
by a different contractor to Well 4. Reduced
values for potassium in Well 1 are most likely to
be due to an over-correction for potassium in the
KC1 drilling mud.

The cross plot of potassium against thorium for


Unit B (Fig. 5) shows a consistent trend in the
data for Wells 9, 10, 12 and 13, although there is
an offset between the wells largely due to
differences in the amount of thorium. Well 11
has a flatter trend with significantly higher
potassium and a wider range in thorium.
Assuming that the original correlation is correct,
is there any simple explanation for the difference? Wells 9 and 13 were drilled with an oil
based mud, Wells 10 and 12 were drilled with a
water based mud and Well 11 was drilled with a
salt saturated polymer (221 ppmK). It would
appear most likely that the correction factor for
potassium in the mud has left a residual of
enhanced potassium values. One might wonder
why the other Wells (10 and 12), with water
based mud and relatively high KC1 contents, lie
on a trend with Wells 9 and 13? The answer may
be that Wells 9, 10, 12 and 13 were all logged by
a different contractor to Well 11. It would
appear therefore that the choice of logging
contractor can have a significant effect on
results.

Box plots show differences between wells


Box plots have been used for a comparison of
total gamma ray values for cross-stratified
sandstones and claystones between wells, using
lithofacies defined from core. Each plot (including boxes and whiskers) shows the spread of
observations about the median. The box repre-

C.S. BRISTOW & B. J. WILLIAMSON

Fig. 7. Cross plot of K/Th against K/U for three correlated sandstones (Unit A) shows lower potassium values
and an exceptionally good correlation of thorium and uranium in Well I which are attributed to correction
factors which have over-compensated for KC1 in the drilling mud.

sents 50% of measurements about the median,


the whiskers extend to the minimum and
maximum data values. Median values for
cross-stratified sandstones are generally 50 API
units or less, although they do vary between
wells (Fig. 6). Total gamma ray response for
sandstones is almost always less than the total
gamma ray response for claystones, where the
median value is close to 100 API units, although
there is some overlap in Wells 3 and 6 where the
claystones have lower total gamma ray response
than the other claystones. There is no obvious
reason for the lower total gamma ray response in
these two wells. Well 3 was drilled with a water
based mud, but so were Wells 1 and 7, while
Well 6 was drilled with an oil based mud as were
Wells 2, 4 and 5. Wells 3 and 6 are from broadly
similar stratigraphic units but Wells 5 and 7 are
from the same Group. One possible explanation
is that the claystones in Wells 3 and 6 were
deposited in slightly different environments. The
core logs indicate a prodelta environment for
claystones in Wells 1, 2, 4, and 7 and an
interdistributary bay environment for claystones
in Wells 3, 6 and 5. Re-examination of the core
logs indicates that the claystones in Well 5,
originally attributed to an interdistributary bay,
are significantly thicker than other interdistributary bay deposits and could be re-interpreted as
prodelta deposits. If this is the case, then the
total gamma ray response is discriminating
between sedimentary environments, not just
between lithofacies.

Eliminating inter well differences using ratio


plots
Element ratio vs element ratio plots were
generated to eliminate the systematic variations
in gamma ray tool response between wells
(usually due to varying well conditions) which
may have been inadequately compensated for in
logging company calibration procedures. The
plot of K/Th ratio against K/U ratio for Unit A
(Fig. 7) shows that measurements from Wells 4
and 8 overlap while measurements from Well 1
are clearly lying on a different trend. Wells 4 and
8 were both drilled with an oil based mud while
Well 1 was drilled with a water based mud
containing KC1. The K/Th cross plot (Fig. 4)
shows low potassium values for Well 1, and the
ratio plot (Fig. 7) shows an offset due to low
potassium values. In addition, Fig. 7 shows an
exceptionally good correlation between thorium
and uranium. We suspect that the correction
factor applied to compensate for KC1 mud in
Well 1, has over-compensated for potassium and
also affected the measurements of thorium and
uranium.

Conclusions
Lithofacies for Carboniferous deltaic sequences
from the Southern N o r t h Sea have been
identified from core descriptions and compared
with spectral gamma ray logs. The results show

SPECTRAL GAMMA RAY LOG CORRELATION PROBLEMS


that lithofacies can be discriminated within
single wells. However, comparison of correlated
sandstones shows that variations between wells
are greater than variations within wells. There is
too much variation between wells to make an
unequivocal assessment of lithofacies and depositional environment on the basis of spectral
gamma ray logs alone. The differences between
wells are attributed to changes in logging
environment, mainly mud characteristics, borehole quality and different logging companies
which have made detailed correlations impossible. For sandstones showing low total gamma
ray response, small errors associated with
calibrations and correction factors will make a
comparatively large difference to results. In three
wells, negative values for uranium were noted
and in one well negative values for potassium
were found which suggests a problem with the
algorithm used to calculate elemental concentrations. Cross plots of correlated sandstones
indicate that correction factors for KC1 in
drilling muds are not always successful, and
there appears to be a difference between the
results achieved by different contractors in this
respect. Corrections for KC1 appear to be based
on a single value for each well although mud
chemistry will almost certainly change down
hole. More detailed tool calibration is required
before subsurface correlations and facies analysis can be reliably made using spectral gamma
ray response alone. The influence of downhole
environment could be further tested by comparing the geochemical composition of core with
gamma ray response. In the meantime avoid
trying to read too much from spectral gamma
ray response where KC1 mud is involved.
The authors thank Mobil North Sea for funding this
work and for permission to publish the results. The
manuscript has been improved by the comments of J.
S. Schweitzer and P. Corbett.

References
ARCHARD, G. & TRICE, R. 1990. A preliminary
investigation into the spectral radiation of the
Upper Carboniferous marine bands and its
stratigraphic application. Newsletters on Stratigraphy, 21, 167-173.
CANT, D. J. 1992. Subsurface facies analysis. In:
WALKER R. G. JAMES, N. P. (eds)Facies
Models, Geological Association of Canada, pp.
27-45.
DAVIES, S. J. 8~ ELLIOT, T. 1995. Spectral gamma ray
characterisation of high resolution sequence
stratigraphy: examples from upper Carboniferous
fluvio~leltaic systems, County Clare, Ireland. In:
HOWELL, J. A. 8z AITKEN, J. F. (eds) High

Resolution Sequence Stratigraphy: Innovations


and Applications. Geological Society Special Publications No. 104, pp. 25-35.
DESBRANDES, R. 1985. Encyclopedia of well logging.
Institut Francais du Petrole, Graham and Trotman Ltd, London.
DRESSER ATLAS. 1982. Well logging and interpretation
techniques (3rd edition). Dresser Industries Inc.,
USA.
LEEDER, M. R., RAISWELL,R., AL-BIATTY,H., MCMAHON, A. & HARDMAN, M. 1990. Carboniferous
stratigraphy, sedimentation and correlation of
well 48/3-3 in the southern North Sea Basin:
integrated use of palynology, natural gamma/
sonic logs and carbon/sulphur geochemistry.
Journal of the Geological Society, London, 147,
pp. 287-300.
MYERS, K. J & BRISTOW, C. S. 1989. Detailed
sedimentology and gamma ray log characteristics
of a Namurian deltaic succession II: Gamma ray
logging. In: WHATELEY,M. K. C. & PICKERING,K.
T. (eds) Deltas." Sites and Traps for Fossil Fuels,
Geological Society Special Publications No. 41,
pp. 81-88.
RIDER, M. H. 1986. The Geological Interpretation of
Well Logs, Blackie Halsted Press, Glasgow.
SELLEY, R. C. 1978. Concepts and methods of subsurface facies analysis. American Association of
Petroleum Geologists, Continuing Education
Short Notes 9.

A review of up-scaling and cross-scaling issues in core and log data


interpretation and prediction
P. W. M. C O R B E T T , J. L. J E N S E N 1 & K. S. S O R B I E

Department of Petroleum Engineering


Heriot-Watt University, Edinburgh, EH14 4AS, UK
1 Present address." University o f Alaska at Fairbanks, Alaska
Abstract: In a heterogeneous geological formation, each rock petrophysical property (e.g.,
permeability, porosity, and electrical conductivity) reflects the heterogeneity and varies in a
manner related to the underlying changes in fabric (grainsize, mineralogy, lamination,
wettability, etc.). However, measurements, both laboratory and downhole, are made at
certain volume scales dictated by the size of the core plug used or the wireline log resolution.
The comparison of core and log data needs to account for both the scale and physics of the
particular measurements and how these relate to the underlying scale of the geological
heterogeneity of the formation.
In this review, these two fundamental issues are addressed as follows:
(a) measurement scale and how it relates to the 'true' or 'required' petrophysical
properties of the formation is defined as 'up-scaling';
(b) measurement physics and how we relate the physics of one measurement (e.g.
permeability) to that of another (e.g. density, electrical, or acoustic properties) is termed
'cross-scaling'.
We illustrate how these two issues arise in the comparison and prediction of permeability
using several published studies. We also outline an approach to petrophysical measurement
reconciliation termed 'genetic petrophysics'. This combines all three elements--measurement scale, measurement physics, and geology--to provide an integrated and robust model.
We illustrate this approach for permeability to provide fit-for-purpose models of anisotropy
in the near-well region of a reservoir.
It has been appreciated for some time that there
is a problem of scale in reservoir engineering
(e.g. Warren et al. 1961; Haldorsen 1986). The
volume of a reservoir under production greatly
exceeds the volume of rock recovered from cores
or investigated by wireline logs. There are many
efforts underway to improve the modelling of
reservoirs, which particularly address the extrapolation from the sparse core-log data to the
interwell volumes. Computer flow models of
reservoirs involve grid blocks that are by
necessity large, relative to the investigation
volumes of core or logs. Therefore engineers
have to integrate the core and log data for use in
simulation models in a process loosely referred
to as 'up-scaling'. Permeability is a particular
property of interest and several techniques have
been developed for its up-scaling, e.g. power
averaging, renormalization, and pseudo-isation.
The aim of up-scaling is to estimate the
'effective' or equivalent properties at the chosen
volume scale, e.g. grid blocks.
The adjectives 'effective' and 'pseudo' are
often used interchangeably in the petroleum
literature to denote an up-scaled property, but
there is a subtle difference. The former attempts
to be intrinsic to the rock/fluid system and aims
to be independent of boundary conditions,

including time. The latter, on the other hand,


applies on some coarse grid as a replacement of
a fine grid domain, but it may change radically if
the boundary conditions are changed. It will
emerge from our discussion that we are frequently talking about pseudo properties when
we refer to core-log data integration.
The petrophysical community have appreciated for some time that there are also scaleup problems in making comparisons between
core and log data (e.g. Knutson et al. 1961).
However, historical practice relied on the
sampling of cores with plug-size measurements
at one-foot spacing (Fig. l a). These were then
compared directly with the log measurements,
recorded at half-foot intervals. Shifts between
core depths and log depths accounted for the
offset (if present) between the core and log.
Occasionally, a primitive up-scaling technique
using a running average ( 1 : 2 : 1 weighting) was
used for the plug data prior to comparing with
the log data. Although the scale discrepancies
were often appreciated, there was not much else
that could be done.
The development of high resolution petrophysical measurements in the laboratory (probe
permeameter) and downhole (image logs) has
presented new opportunities to address the scale

CORBETT,P. W. M. JENSEN, J. L. & SORBIE,K. S. 1998. A review of up-scaling and cross-scaling issues
in core and log data interpretation and prediction In: HARVEY,P. K. LOVELL,M. A. (eds)
Core-Log Integration, Geological Society, London, Special Publications, 136, 9-16

10

P . W . M . CORBETT E T AL.

Fig. 1. A comparison between (a) the traditional core


plug and logging tool scales of measurement and
sample spacing and, (b) the new opportunities
provided by closely spaced probe data and high
resolution logging tools. Both schemes are shown
schematically against a core interval with a missing
section. In (b) there is more scope for identifying small
scale heterogeneities and less sensitivity to missing core
material. Depth matching is also improved.

issues between core and log measurements.


These high resolution measurements image the
geology far more effectively than the conventional, low resolution devices. Indeed, image
logs were developed specifically to image the
geology in the subsurface, potentially replacing
the need for core data. With data at high
sampling densities and small volumes of measurement, the comparison between logs and
cores becomes more tractable (Fig. lb) and
therefore gives us a feasible approach to corelog scaling. Since the laboratory probe permeameter measures a different physics (gas flow rate)
to a subsurface image log (acoustic reflection or
electrical conductivity) with different boundary
conditions, there are also cross-scaling relationships (see below) that must be considered in
addition to volume scale and sampling density
effects.
In this paper, we illustrate the cross-scaling
and up-scaling of permeability between core and
wireline logs for subsurface prediction of permeability. Larger scale dynamic data are used to
justify the methods presented. Having reviewed
the method, we discuss the implications for
other properties and outline a new approach to
petrophysics--genetic petrophysics--which is

Fig. 2. A comparison between (a) Up-scaling and (b)


Cross-scaling. Numbers refer to approximate volumes
in cubic metres. Refer to the text for definition of these
terms.
being developed. This approach is tied directly
to the needs of reservoir modeUers and offers a
way of integrating data and procedures from the
original geological conceptual model, through
the petrophysical data acquisition, the up-scaling/cross-scaling, and the construction of the
numerical reservoir simulation model.

Definitions
In this paper, we define the terms up-scaling and
cross-scaling as follows (Fig. 2):
Up-scaling: The determination of an effective
(or pseudo) property at a scale larger than
that of the original measurement. An example would be using the arithmetic average of a
set of layer permeabilities as an estimator of
the horizontal permeability of the composite
layered media (Jensen et al. 1997, pp. 137139). Comparing probe to plug to well test
permeabilities is an up-scaling problem (Corbett et al. 1996a).
The issue of measurement scale for the same
petrophysical property is the process of upscaling. Reservoir engineers are familiar with the
up-scaling of permeability for reservoir simulation. Cross-scaling is a much less familiar
concept and may be defined as follows:
Cross-scaling: The determination of a relationship between two different physical prop-

A REVIEW OF UP-SCALING AND CROSS-SCALING

11

erties. Using regression to summarize the


relationship between porosity and permeability for a suite of core plugs is a simple
example. Comparing compressional wave
transit time with porosity is a cross-scaling
procedure.
Cross-scaling provides the relationship--if
there is one--between measurements of different
petrophysical properties, at different measurement volume scales which are affected by the
(different again) underlying volume scale of the
geological heterogeneity. This clearly concerns
the transfer of information on a certain required
property via a more 'easy-to-measure' surrogate.
The scales at which these transfers take place are
critical to assessing the appropriateness--or
inappropriateness--of the surrogate property.
The definition of these terms helps us distinguish the impact of geology (largely up-scaling)
from the physics (largely cross-scaling) in a more
systematic fashion. These concepts are useful in
the comparison of core and log data. In the next
two sections, we look first at the cross-scaling of
permeability and resistivity at compatible scales.
These data are then up-scaled for comparison
with larger scale dynamic data. Together these
case studies show that cross-scaling and upscaling of permeability can be achieved in
practice.

Fig. 3. Measurement of properties in the laboratory at


similar volume scales with a resistivity probe (above)
and permeability probe (below). Refer to Jackson et al.
(1994) for more details.

Case studies
We consider three examples of the cross-scaling
between permeability and resistivity which have
been carried out and which have been reported
in the literature.

Fig. 4. Correlation between resistivity (shown a


formation resistivy factor= measured resistivity/brine
resistivity) against probe permeability for a slab of
Lochabriggs Sandstone.

Laboratory study
Jackson et al. (1994) measured permeability and
resistivity with probe devices for an aeolian
sample that was saturated with brine in the
laboratory (Fig. 3). The resistivity probe was
carefully designed to investigate a volume
similar to that of a steady state probe permeameter and both volumes were comparable to the
sample's scale of sedimentary variation. A
strong relationship was observed (Fig. 4) and
this can be related to the fundamental physical
control.

As-Sarah study
Ball et al. (1997) carried out a probe permeameter study on a fluvial sandstone. They found
that averaged probe data (at 10cm spacing)
correlated reasonably well with microresistivity

(Fig. 5a) and provided the basis of a permeability predictor which was a considerable
improvement over methods based on the density
log and core plugs (Fig. 5b).

Morecambe Bay study


Thomas et al. (1996, 1997) undertook a detailed
probe study over a fluvial interval for which
resistivity images had also been acquired. They
found a strong correlation between the probe
permeability and microscanner resistivity (Fig. 6).
In all of these cases, an empirical relationship
existed and was reflected in the measurements at
similar scales. Such relationships could reflect an
underlying physical relationship, explained by
an existing analytical model (e.g. Biot's and the
C a r m a n - K o z e n y models) or might provide

12

P.W.M. CORBETT E T AL.

Fig. 5. Correlation between (a) probe permeability


(averaged over a 30 cm window) and micro-spherically
focussed log (MSFL) resistivity and (b) plug permeability and wireline density for an interval of PUC-B
Reservoir. Refer to Ball et al. (1997) for more details.
insight into the need for new petrophysical
analysis.

Cross-scaling permeability and resistivity


In the three studies just mentioned, the relationship is driven by the effects of pore geometry and
porosity upon both the hydraulic and electrical
conductivities. Several workers (e.g. Doyen
1988; Katz and Thompson 1987) have shown
that both transport properties depend on a
characteristic pore size in the rock. The form
of that dependency differs for hydraulic and
electrical conductivity, thus making the hydraulic--electrical relationship strength dependent
also upon the level of rock heterogeneity. In a
homogeneous sample, one characteristic length
and its mutual effects upon both permeability
and conductivity will give rise to a strong
electrical-hydraulic relationship. Heterogeneity,
however, will diminish the relationship strength
because different portions of a sample will have
differing characteristic sizes. This explains why
data at the lamination scale (e.g. Jackson et al.

Fig. 6. Comparison between probe permeability,


formation image and FMI resistance for an interval
of Sherwood Sandstone. Refer to Thomas et al. 0996,
1997) for more details.
1994; Thomas et al. 1996) show very strong
resistivity-permeability relations, while laminaset measurements (Ball et al. 1997) exhibit a
weaker, though still useful, relationship. In all
cases, the effects of geological variation were
mitigated by chosing similar measurement volumes.

Up-scaling permeability
In two of the cases presented above, the
permeability was up-scaled for comparison with
some larger scale dynamic data.
In the As-Sarah study, the permeability
predictor developed from the microresistivity
was used to predict permeability in the uncored
sections of several wells. With a continuous
permeability log, the cumulative permeabilitythickness product, the transmissivity, was compared with a production log spinner survey. A
good comparison was found supporting the
appropriateness of the predictor (Fig. 7). This
predictor continues to form the basis for
permeability models in the field (von Winterfeld,
pers. comm.).

A REVIEW OF UP-SCALING AND CROSS-SCALING


-11975"

13

-11900

-11950" ~

-12025" ~ P r o b e

~~

Probe
PLT

Cum. Prob~

~ -12125

-12000- ~i

-12050 -

~LT

-12175'

-12225

, ,

, ~ 9 ,

9 , , ,

, ,

9 ,

50 100 150 200 250 300 350 400

-12150

'

9 ,

9 ,

250 500 750 I000 1250 1500 1750

Permeability (mD)

Fig. 7. Validation of probe/MSFL predictor (refer to Fig. 5a) against production log data (PLT) in two wells from
the As-Sarah Field. The intervals picked out by the predictor over a 250 ft interval correlate well with the
productive intervals seen with the PLT. Refer to Ball et al. (1997) for more details.
Bed Scale

Bedset Scale

'~

I~T ~

enrlda~:; % o P ~ : 2 a ttimates

~. 0.001t
o

10

15

Measurement interval (ft)


Fig. 8. Validation of probe/FMI resistance predictor
for kv/kh with pressure data from a Modular Dynamic
Tool (MDT) for an interval of the Sherwood
Sandstone. The probe estimator uses the harmonic
average (over a moving window--to represent vertical
permeability over a measurement interval) divided by
the arithmetic average (horizontal permeability) over
the same expanding window. Refer to Thomas et al.
(1996, 1997) for more details.

In the Morecambe Bay study, the up-scaling


of both horizontal (kh) and vertical (kv) permeability were required for comparison with a
borehole pressure measurement. The horizontal
permeability was up-scaled by taking the arithmetic average of the probe data, the vertical
permeability by taking the harmonic average
(Thomas et al. 1996, 1997). The up-scaled

properties (effective kv/kh ratio) compared well


with a larger-scale dynamic measurement (Fig. 8).
The importance of the geology in the upscaling is well illustrated in two ways in this
second study. Firstly, the abrupt decline in
vertical permeability (i.e. increase in anisotropy)
occurs at the bed length scale which, for these
stacked fluvial channels, represents several feet.
The image log picks out the geological features
associated with bedding and this can be
exploited to produce improved prediction of
formation anisotropy. Secondly, there is an
assumption, supported by geological analysis
of similar beds in outcrop, that the layers or beds
observed at the wellbore extend well beyond the
volume of investigation of the dynamic measurement. This is in contrast to the anisotropy shown
by plug scale measurements which is notably
poor in estimating effective kv/kh at larger scales.
Averaging plug scale kv/kh ratios is also an
inappropriate up-scaling method for this parameter, which is very sensitive to scale changes
(Corbett et al. 1996b, Cowan 7 Bradney 1997).
The comparison of up-scaled permeability
(probe, plug, or wireline) with the well test can
provide additional corroboration of permeability predictors. For these larger scales, the effects
of the organization of the geology (i.e. sedimentary structure) can also be important. This level
of up-scaling is beyond the scope of this review
(refer to Corbett et al., 1996a). Nonetheless, it is
important to note that up-scaling from core to
log must be tied with a consistent geological
framework to the scales of well tests and full
field numerical grid blocks.

14

P.W. M. CORBETT E T AL.

Plug permeability - density log cross-scaling


revisited
We can revisit the As-Sarah example to compare
the probe-microresistivity method with the plugdensity method for permeability prediction 9 This
will reveal the nature of improvements provided
to the petrophysicist by the smaller scale
measurements 9 It seems ironic that solutions to
the up-scaling problem have been facilitated (i.e.
they are more accurate, not necessarily faster) by
having more 'smaller' scale petrophysical measurements. This irony, however, overlooks the
role of the geology in the scaling process:
smaller-scale measurements are often more
easily interpreted in their geological context 9
The geology provides information regarding the
volume and shape of each event, allowing
analysts to make inferences about the validity
and frequency of the value in the unsampled
regions 9
If we examine the porosity-permeability
relationship (Fig. 9) for the As-Sarah reservoir,
we see that it is very weak. The lack of
relationship is due to a number of factors-variable grain size and sorting in the fluvial
sediments, patchy rhizocretionary cements, plug
orientation with respect to heterogeneities, and
others. Weak porosity-permeability relationships in fluvial reservoirs are often observed
(Brayshaw et al. 1996). The cross-scaling relationship, in this case, is strongly obscured by the
geological heterogeneity--a smaller or larger
volume scale is suggested or separation of the
grain size classes (Hogg et al. 1996). Any
porosity-permeability relationship from these
data will be associated with a high degree of
uncertainty if used to predict permeability 9
On the (weak) assumption that porosity and
permeability are related, the wireline density
derived porosity might be used to predict
permeability. The density log has a volume of
investigation that is larger than the small scale
(lamination) textural features that control permeability. In this example, it also proved very
difficult to depth match the plug data with the
log data, the probe data were more useful in this
respect. The 'true' variation in permeability
shown by the probe did not correlate well with
the poor resolution of the density log. Any
permeability predictor based on the latter will
eliminate a scale of heterogeneity that may be
important to the sweep efficiency of the reservoir. The up-scaling of permeability, if this
method had been followed, would result in a
more uniform reservoir permeability field, which
may have been inappropriate for modelling oil
recovery 9

4"

~3;~

. .ii

9 '.

2"

i,iii.

2-047

...

J
~-.
9 ]

-1

'

;0

~0

3'0

Porosity (%)
Fig. 9. Core plug porosity and permeability relationship for the PUC-B Sandstone. This type of relationship is typical in texturally heterogeneous fluvial
reservoirs. Clay content and cementation variations
at plug scale due to clay drapes and rhizocretions also
impact these data. These factors combine adversely to
make a complex relationship between permeability and
porosity, one which cannot be used with any
confidence for permeability prediction. A more texturally sensitive surrogate property is needed and was
provided (in this study) by the MSFL resistivity 9Refer
to Brayshaw et al. (1996) for more discussion on the
textural controls on permeability and to Ball et al.
(1997) for more details of the PUC-B study.

Genetic petrophysics
The Morecambe Bay example shows the power
(for prediction) of scale-compatible cross-scaling
and geologically-assisted up-scaling. Fig. 8
shows that the effective property (in this case,
k v / k h ) varies at certain geological length scales.
There is a significant and abrupt change at the
bed scale (4ft) and the bedset scale (12ft). Above
the bedset scale, there appears to be less
variability in the estimates and close agreement
with the Modular Dynamic Tool (MDT) response. While the cost implications of MDT
versus image log have to be considered, image
log based predictors, calibrated by MDT measurements at carefully selected intervals, hold
potential for improved anisotropy estimates in
the future. Anisotropy in sediments is strongly
affected by bedding, so it is only appropriate
that a predictor based on a log that 'sees' the
bedding will be better than estimates from small
volume, plug measurements.
The length scales (i.e. the geological architecture) provide important guidance for the petrophysicist--the length scales for combining or
comparing appropriate measurements and also
the length scales to be avoided for sampling
intervals. Sampling close to the frequency of the
data (volume or wavelength) is a notoriously
poor procedure in geophysical measurements.
Unfortunately, the 1-inch plug size and 1-foot
sampling interval are close to the Nyquist
frequency of lamina and beds!

A REVIEW OF UP-SCALING AND CROSS-SCALING


Smaller scale measurements mean more data
and more work in reducing, summarizing, and
integrating the data. In that respect, their
development is not welcome in the time- and
personnel-challenged climate demanded by industry. However, the very fact that there are
representative elements within reservoirs (e.g.
stratal elements, genetic units, architectural
elements) can also be exploited. The fundamental rock measurements can be targetted on these
representative elements. The effective properties
(or even pseudos) can then be determined--by
simulation or measurement for these elements
(Corbett et al. 1992; Ringrose et al. 1993; Pickup
et al. 1995; Huang et al. 1995). The modelling of
these elements--geobodies in G O C A D - - c a n
then be accompanied by the appropriate petrophysics--genetic petrophysics. This method involves a more selective use of petrophysical
measurement which is intended to be more costeffective. Indeed, a genetic petrophysics approach which explicitly recognizes and solves
the cross-scaling problem may be the only
successful route to true data integration.

Conclusions
Cross-scaling between petrophysical properties
is best achieved when the scales and density of
measurements are comparable.
Up-scaling of petrophysical properties benefits when the geological architecture is accounted
for.
Probe data and image logs can be jointly used
to predict permeability (horizontal and vertical
in the subsurface), demonstrating that consistent-volume cross-scaling and geologically-constrained up-scaling can be effective.
The tools, understanding, and techniques are
now available for the development of a more
geologically-based petrophysics m e t h o d - - w h i c h
we refer to as genetic petrophysics--that is fit for
the purposes of reservoir modelling. It is understood that improved modelling prepares the way
for improved oil recovery--the ultimate motivation behind this work.
The authors acknowledge the support of Wintershall
and British Gas in the studies discussed above. They
also wish to acknowledge the support of EPSRC and
industrial co-sponsors (Amerada Hess, Amoco, BHP
Petroleum, British Gas, Chevron, Fina, Saga, Schlumberger, Shell, Statoil, Texaco) for continued work in
this area under the PEGASUS project. The authors
also acknowledge the contributions of the various
authors of the case studies from which this overview
has been drawn L. Ball, J. Lewis, S. Thomas, D.
Bowen and M. Jackson. Their insights while working
with the data have helped to formulate and illustrate
these concepts.

15

References
BALL, L. D., CORBETT,P. W. M., JENSEN,J. L. & LEWIS,
J. M. 1997. The role of geology in the behavior
and choice of permeability predictors. SPE
Formation Evaluation, 12, 32-39.
BRAYSHAW,A. C., DAVIES,G. W. CORBETT,P. W. M.
1996. Depositional controls on primary permeability and porosity at the bedform scale in fluvial
reservoir sandstones. In CARLING, P. A. &
DAWSON, M. R. (eds) Advances influvial dynamics
and stratigraphy, John Wiley & Sons, 373-394.
CORBETT, P. W. M., RINGROSE,P. S., JENSEN, J. L. &
SORBIE,K. S. 1992. Laminated clastic reservoirs-The interplay of capillary pressure and sedimentary architecture. SPE 24699. Proceedings of the
67th SPE Annual Technical Conference and
Exhibition, October, Washington, 365-376.
--,
PINISETTI, M., TORO-RIVERA, M. & STEWART,
G. 1996a. The comparison of plug and well test
permeabilities, Dialog, 4-8.
--,
GOOD, T., JENSEN, J. L., LEWIS, J. J. M.,
PICKUP, G., RINGROSE, P. S. & SORBIE, K. S.
1996b. Reservoir description in the 1990s: A
perspective from the flow simulation through
layercake parasequence flow units. In: GLENNIE,
K. & HURST, A. (eds), AD 1995: N W Europe's
Hydrocarbon Industry, Geological Society, London, 169-178.
COWAN, G. & BRADNEY,J. 1997. Regional diagenetic
controls on reservoir properties in the Millom
accumulation: implications for field development.
In: MEADOWS,N. S., TRUEBLOOD,S. P., HARDMAN,
M. & COWAN,G. (eds) Petroleum Geology of the
Irish Sea and Adjacent Areas. Geological Society,
London, Special Publications, 124, 373-386.
DOYEN, P. M. 1988. Permeability, conductivity, and
pore geometry of sandstone. Journal of Geophysical Research, 93, 7729-7740.
HALDORSEN, H. H. 1986. Simulator parameter assignment and the problem of scale in reservoir
engineering. In: LAKE, L. W. & CARROLL, H. B.
(eds), Reservoir Characterisation, Academic Press,
Orlando.
HOGG, A. J. C., MITCHELL,A. W. & YOUNG, S. 1996.
Predicting well productivity from grain size
analysis and logging while drilling. Petroleum
Geoscience, 2, 1-15.
HUANG, Y. RINGROSE, P. S. & SORBIE, K. S. 1995.
Capillary trapping mechanisms in water-wet
laminated rocks. SPE Reservoir Engineering, 10,
287-292.
JACKSON,M. A., BOWEN,D. G., JENSEN,J. L. & TODD,
A. C. 1994. Resistivity and permeability mapping
at the lamina scale. Proceedings of the International Symposium of the Society of Core Analysts, Stavanger, 12-14 Sept., paper SCA-9415,
163-172.
JENSEN, J. L., LAKE, L. W., CORBETT, P. W. M. &
GOGGIN, D. J. 1997. Statistics for Petroleum
Engineers and Geoscientists, Prentice-Hall, New
Jersey.
KATZ, A. J. & THOMPSON, A. H. 1987. Prediction of
rock electrical conductivity from mercury injec-

16

P. W. M. CORBETT ET AL.

tion measurements. Journal of Geophysical Research, 92, 599-607.


KNUTSON, C. F., CONLEY, F. R., BOHOR, B. F. &
TIMKO, D. J. 1961. Characterization of the San
Miguel Sandstone by a coordinated logging and
coring program. Journal of Petroleum Technology,
13, 425-432.
PICKUP, G. E., RINGROSE, P. S., CORBETT, P. W. M.,
JENSEN, J. L. ~; SORBIE, K. S. 1995. Geology,
g e o m e t r y , and effective flow. Petroleum
Geoscience, 1, 37-42.
RIN~ROSE, P. S., SORBIE, K. S., CORBETT, P. W. M. &
JENSEN, J. L. 1993. Immiscible flow behaviour in
laminated and cross-bedded sandstones. Journal
of Petroleum Science and Engineering, 9, 103-124.
THOMAS, S. D., CORBETT, P. W. M. & JENSEN, J. L.
1997. Permeability anisotropy estimation within
the Sherwood Sandstone, Morecambe Bay Gas

Field: a numerical approach using probe permeametry. In: OAKMAN, C. D., MARTIN, J. H.
CORBETT, P. W. M. (eds) Coresfrom the Northwest
European Hydrocarbon Province." An illustration of
geological applications from exploration to development. Geological Society, London, 197-203.
& JENSEN,J. L. 1996. Permeability and
permeability anisotropy characterization in the
near well-bore: a numerical model using probe
permeability and formation micro-resistivity data,
Transactions of The Society of Professional Well
Log Analysts Thirty-Seventh Annual Logging
Symposium, New Orleans, 16-19 June, paper JJJ.
WARREN,J. E., SK1BA,F. F. & PRICE, H. S. 1961. An
evaluation of the significance of permeability
measurements. Journal of Petroleum Technology,
13, 739-744.
-

Quantitative density measurements from X-ray radiometry


A. R. D U N C A N 2, G. D E A N 1 & D. A. L. C O L L I E 2

t Amerada Hess Limited, 33 Grosvenor Place, London, S W 1 X 7HY, UK


2 Robertson Research International Limited, Unit 7, Wellheads Crescent, Wellheads
Industrial Estate, Dyce, Aberdeen, AB21 7GA, UK
Abstract: Qualitative linear X-ray scanning has an established role in the non-destructive
imaging of both slabbed and whole core and has been routinely used in visual assessment
and quality control of material being subjected to other physical measurements. Since core
may be observed in real time, whole core can be oriented to maximum dip prior to slabbing,
especially useful where core has been resin-stabilized within an outer liner. Linear scanning
is also useful in the observation of heterogeneous lithologies; the features observed are
distinguished by their penetrabilities to X-rays. As a result, the linear scanner produces an
image which reflects the density variation in the section analysed. A joint project carried out
by Robertson Research International Limited and Amerada Hess Limited on 108ft of
heterogeneous sediments has shown that the digital X-ray penetrability values ('luminance')
can be extracted in order to produce a surface density variation log. X-ray luminance values
show a linear relationship with the downhole Formation Density Log and may, therefore,
provide an accurate tool for the correlation of core density with log density.

Qualitative linear X-ray scanning already has an


established role in non-destructive imaging of
both slabbed and whole core and has been
routinely used in visual assessment and quality
control of material being subjected to other
physical measurements (for example Algeo et al.
1994; Rigsby et al. 1994). Since core may be
observed in real time, whole core can be oriented
to maximum dip prior to plugging or slabbing,
especially useful where the core has been resinstabilized within fibreglass, pvc or aluminium
liners. This ability to examine interactively, in
detail and non-destructively, the 3-D nature of
the internal structure of the core material is
particularly important. Linear scanning is therefore useful in the observation of both heterogeneous and apparently homogenous lithologies
and the following features are commonly characterized:
(a) bedding features and sedimentary structures;
(b) bioturbation (ichnofacies analysis), especially in slabbed sections;
(c) identification of remnant structure (not
readily visible to the naked eye) which has
been obscured by bioturbation;
(d) natural and coring-induced fractures and
shears (cemented/uncemented/open);
(e) cement distribution;
(f) small scale grain size variation;
(g) assessment of resin competence in preserved and/or sleeved core.
These features are distinguished by their

different penetrabilities to X-rays. As a result


the linear scanner produces an image which
reflects the density variation in the section
analysed (Tolansky 1961).
A project carried out on 108ft of heterogeneous sediments (Duncan et al. 1996) has
shown that a digital measure of the X-ray
penetrability values ('luminance') can be extracted in order to produce a surface density
variation log.
These X-ray luminance values may yield data
at close and equally spaced points producing a
log with significant advantages over the data
from conventional core analysis (where sample
spacing may be irregular, widely spaced and
lithologically chosen, or where Gamma Ray
response may be poor). Such data can be
compared directly with the wireline logs and it
is found that the X-ray luminance values show a
linear relationship with the downhole Formation
Density Log (FDL). The X-ray luminance data
may therefore provide an accurate tool for the
correlation of core density with log density.
Database

The Scott partner group provided access to a


range of core and associated materials:
(a) 108ft of lithologically/mineralogically
variable sediments (resinated archive
slabs);
(b) wireline logs for the analysed interval
including the appropriate FDL traces;
(c) the sedimentological composite log;

DUNCAN, A. R., DEAN, G. & COLLIE,D. A. L. 1998. Quantitative density measurements from X-ray
radiometry In. HARVEY,P. K. & LOVELL,M. A. (eds) Core-Log Integration, Geological Society, London,
Special Publications, 136, 17-24

17

A. R. DUNCAN ET AL.

18

(d) petrographic data for the seven thin


section samples which fall within the
analysed interval;
(e) core analysis data (porosity, permeability
and grain density) for the analysed interval.

Brief description of cores


Section A
The analysed interval commences within relatively 'clean', blocky, medium grained sandstones. These sandstones are assigned to the
Piper Formation Depositional Unit 4c and are
interpreted to be shoreface sandstones, possibly
representing gully fill within an upper delta front
system. At a depth of 6 ft below top of section
these deposits are underlain by variably argillaceous sandstones with sandy, argillaceous siltstone interbeds. The sandstones are generally
fine grained and are frequently apparently
structureless or faintly laminated. Current ripples and burrows are locally observed. Bioturbation is more commonly observed in the finer
grained units, some units are micaceous and
locally contain carbonaceous material. Nodular
calcareous cement is observed at approximately
36 ft below top of section. These lower deposits
are assigned to Piper Formation Depositional
Unit 4b and are interpreted to belong to an
offshore transition zone. They are believed to be
the deposits from turbidite flows in the lower
delta front to pro-delta. Thin section analysis
indicates that the predominant cement is quartz
(8-11.5%) with relatively minor calcite (13.5%). Authigenic clays are dominated by
kaolinite (0-3 %).

Section B
The sediments within Section B are also assigned
to Piper Formation Depositional Unit 4b and
similarly consist of relatively clean, fine grained
sandstones interbedded with siltier, argillaceous
sediments which are moderately to highly
bioturbated. The finer material is frequently
micaceous and carbonaceous debris is locally
recorded. Some of the coarser units show
development of calcareous cement which is
locally nodular. Thin section analysis indicates
that the predominant cement is calcite (4-40%)
with subordinate quartz (1-12%). No authigenic
clays are recorded from the two samples
analysed.

Section C
The sediments analysed from Section C are

again assigned to Piper Formation Depositional


Unit 4b. They also consist of interbedded fine or
very fine grained sandstones and silty, argillaceous deposits. The modal grain size of the
sandstones is seen to decrease towards the lower
part of the analysed section. Pervasive calcareous and dolomitic, nodular cements are locally
common. Thin section analysis indicates that the
predominant cement is dolomite (42.5-48%)
with subordinate calcite (3.5-5%) and minor
quartz (0.5-2%). Authigenic kaolinite accounts
for only 0.5%.

Methodology
Production of the X-ray scan images
A schematic representation of the scanning
system is shown in Fig. 1.
Although the imaging system has been developed to operate with core material of various
forms and dimensions, the present investigations
employed 3 ft resinated archive slabs for the
imaging and quantitative density measurements.
This presents a thickness of rock material for
analysis which is relatively constant, both across
the core diameter and along its length, and for
which the 3-D inhomogeneities are reduced. In
this way, differences in the core thickness and
variability due to the curvature of the core are
reduced and interpretation can be simplified to
essentially 2-D.
The X-rays passing through the rock create an
inverted image of the material on an electronic
image intensifier. This visible image of the X-ray
field is picked up by a CCD camera and
subsequently digitized. This image may be
viewed in real time (i.e. the 'live' image on
screen moves as the rock is transported along the
gantry) and approximately 6-7in of core are
observed at any one time within the camera
image frame. These frames may be enhanced by
a dedicated image processing computer and can
be combined to produce a composite image of
the 3 ft section.
After positioning the core, each 'live' frame is
frozen and a digital filter, which enhances the
edge and structure information, is used to
sharpen the image. Once the optimum image
has been captured it is electronically transferred
to a PC computer terminal and stored as a TIFF
format file. Overlaps of approximately lin
between neighbouring frames are used in order
to ensure the optimum matching in the composite. The individual images are manipulated on
the PC, using conventional image processing
software, to produce the composite image,
which is similarly stored in TIFF format. 'Hard

QUANTITATIVE X-RAY DENSIMETRY

19

Fig. 1. Schematic of X-ray scanner system.


copy' images are produced using a grey scale
printer matching the resolution of the digitized
images.
To further ensure accurate and straightforward matching of each frame to form the 3 ft
composite image, a steel mesh with a grid of
V2in is placed alongside each resin slab as it is
inserted into the scanner. This is especially useful
in sections of the core which appear structureless
and homogenous. The steel mesh is positioned
to avoid obscuration of the core and its image
may optionally remain on the composite image
for scaling and quality control purposes.
Where no rock is present the image appears to
be very bright (white). This is due to saturation
or 'burn out' within the image intensifier, caused
by the higher intensity of X-rays where there is
little core material present to block them. This
'burn-out' of the image artificially increases the
intensifier output in closely neighbouring areas,
resulting in the surrounding rock appearing
'bleached'. In order to avoid the possible
misinterpretation of the X-ray intensity in these
areas, disks and/or strips of lead shielding
approximating the density of the resinated slab
material are placed into plug holes, and other
significant gaps.

Production of the quantitative X-ray density


data
The digital images which are obtained from the
scanner are composed of pixels of varying grey
scale (0 to 255). The grey scale can be read, in

real time, at any given point across the image.


Thus, by taking regularly spaced readings, a
profile of the variation in the grey scale can be
produced.
These data (referred to as luminance values)
indicates the penetrability of the rock material to
X-rays and are, therefore, related to the density
of the rock (Tolansky 1961). Higher luminance
values represent greater penetration of the rock
by the X-rays and, therefore, lower density.
Conversely, lower luminance represents areas of
rock with higher density and therefore greater
X-ray 'stopping ability'. The luminance values
(which vary between approximately 60 and 200
for typical core material) are, therefore, inversely
related to the rock density.
Despite the high quality of the imaging system
used in the capture of the information, each of
the images contains an astigmatic error. This
means that there is an apparent density variation
between the centre of the image compared with
the edges. While this does not significantly effect
the visual interpretation of the images; it is
undesirable in the point luminance data. To
eliminate this error, therefore, the luminance
measurements are recorded from a fixed point
within the X-ray field. The luminance profile is
obtained by moving the rock (using the scanner
transport mechanism) and recording the values
at the known fixed point within the field of view.
Measurements may be recorded at any required
spacing; with 1 92 in spacing being used in the
current project.
The luminance measurements are made using
a 'live' image, from which the background

20

A. R. DUNCAN E T AL.

Fig. 2. Example X-ray image frame. The included grid is of '/2 in mesh.

Fig. 3. (a) Correlation plot of luminance values from slabbed material and bulk density data from wireline logs.
Core depth to log depth correction is shown schematically by tie lines. (b) Luminance data after depth correction
to log depth; with wireline bulk density data overlaid. Arbitrary luminance and density scales. Luminance: solid
line, Density: dashed line.

QUANTITATIVE X-RAY DENSIMETRY

21

Ox x

190

i
i

x o ox

Luminance

Corrected
Luminance

170

~'~ 150

_~

"

~
e

~ ' x x
~ ~ o :! o

~, j

o
F~

x
o~

t~

~ x|

x
~ ~- ~$ ' ~ ' ~ ~x I ~,

--Luminance
R2=0.6445
'- .....

Corrected

Luminance

t30
e=
C

110

E
,-1
,-I

9O

50
2.35

2.4

2.45

2.5
Bulk

2.55

2.6

2.65

2.7

2.75

2.8

Density (wireline) (g/cc)

Fig. 4. Cross plot of luminance values from slabbed material and bulk density values derived from wireline logs.
Luminance values shown both uncorrected and corrected for variations in thickness of core material.
'noise' is reduced by using a moving average
filter (i.e. each measurement is made on an
image comprising the average of 20 scans of the
stationary rock). This is done automatically, and
in 'pseudo real time', using the image processing
computer.
Measurements of the thickness of the slab at
each luminance recording point are noted. In
addition, luminance values for an aluminium
block 'standard' placed at the top and base of
each 3ft section are measured. Where necessary
the scanner controls can be adjusted prior to
scanning to ensure that the observed luminance
from these calibration standards remains consistent. These data, along with the luminance
values are entered into a spreadsheet and stored
on the PC for subsequent analysis

Description of X-ray images


Figure 2 presents a single frame showing the Xray image at the point 'F' marked on Fig. 3. The
bright, irregular lines represent core breaks
which are likely to be coring induced. The core
breaks are locally bedding-parallel. This frame
displays very clearly a partially cemented fracture running subvertically through the core. The
contrast produced by variations in the core
material density allows detailed examination of
these and other features. A conventional core
analysis plug hole, with included masking, is
shown, as is the '/2 in alignment grid.

Discussion
The luminance values indicate the penetrability
of the rock material to X-rays and are, therefore,
related to the density of the rock. Higher
luminance values represent greater penetration
of the rock by the X-rays and, therefore, lower
density. Conversely, lower luminance represents
areas of rock with higher density and therefore
greater 'stopping ability' of the X-radiation. The
luminance values are, therefore, inversely related
to the rock density.
A comparison of this luminance data, representing density, with traditional wireline log
density measurements is presented in Figs 3(a
and b). Figure 3(a) shows the correlation of the
luminance data with the FDL trace. The tie lines
indicate the core to log depth shifts appropriate
for this core material. Clearly excellent correlation between the luminance profile and wireline
log is observed, with Fig. 3(b) showing the
luminance data depth shifted and superimposed
on the FDL trace. The luminance values are
smoothed (using a simple 5 point moving
average filter) but are otherwise unprocessed.
A cross plot of luminance against bulk density
(from the wireline log) is shown in Fig. 4.
Luminance values are shown both uncorrected
and corrected for slab thickness variations. The
correction is performed assuming a simple
reciprocal relationship between thickness and
luminance value: this is considered to adequately

22

A. R. DUNCAN E T AL.

190

. . . . . . .

170

io

........

...... ~

. . . . . . . . . . . . . . . . . .

~1~01

130 . . . .

~o

~ .......

oo o~

~ . . . . . . . . .

I
"~
_1

!~
I..IU~

o~

~ _
9o
i oel
o
oo .~o o
70 ._0 0 O0 o;
:

,,

~9

'~

o 2v

i o
_~

~
,

~. . . . . . . . . .

i
i
!
:
,
i

o
--

oo

~ = - - L

'

!
~

:
'

O 0

o
o

o.

-. -

o~..~f

~o_
......

. . . . .

o Corrected
~
Luminance
'
i
t ~C~
i
Luminance
1
R2=0.6135

....

i
0

10

Porosity

(plug)

12

14

16

18

(%)

Fig. 5, Cross plot of luminance values from slabbed material and porosity values from conventional core analysis
of core plugs taken prior to slabbing. Luminance values corrected for slab thickness variations.

190

i
i
170

---4

.......

D BD1

,, B D 2

i__

BD3

BD4

BD5

BD6

'

ii - - A l l

~.~ 150

~Nxx

data

R2=0.88

m
.Q
=" 130

0
C
(~
e"

"-%

;
x

110

'

__

E
--I

90

1.80

1.90

2.00

2.10

2.20

Bulk

2.30

Density

2.40

2.50

2,60

2.70

2.80

(g/cc)

Fig. 6. Cross plot of luminance values and bulk density values, both from analysis of selected core plugs, Data
differentiated by lithological unit. (Independent plug set).

describe the interaction of the X-rays with the


bulk material over the relatively small observed
variation in both luminance and thickness. This
thickness c o r r e c t i o n is seen to h a v e little
significance in the final correlation of the logs.

Detailed conventional core analysis and sedimentological analysis has been carried out on
the core sections analysed for this project; this
data has been c o m p a r e d with the luminance data
measured from the slabbed section close to the

QUANTITATIVE X-RAY DENSIMETRY


plug locations. Figure 5 shows, for example, a
cross plot of porosity of the CCA plug samples
against the luminance values. Again the linear
relationship between luminance and this key
physical property is well defined.
As further confirmation of these relationships,
X-ray luminance values were measured for an
independent and varied collection of conventional core analysis plug samples whose physical
and geological properties are well established
(Duncan 1993). Figure 6 shows the bulk density
for these samples plotted against luminance; the
strong linear relationship is again confirmed.
These data are further differentiated by lithology
and while it is interesting to speculate on the
relationship between lithology and luminance,
this dataset is considered too sparse to prompt
any reliable conclusion.
Interestingly, and as a positive demonstration
of the utility of these X-ray densimetry measurements, the original core to wireline shifts for the
slabbed core sections were taken to be: Wireline
Log Depth equals Core Depth + six feet. Comparison of the quantitative FDL (log) and
luminance measurements, however, indicates
that a core to wireline correction of eight feet
(downhole) for section A is more appropriate. A
shift of six feet for Sections B and C is confirmed
by comparison of the density and luminance
traces. These revised depth shifts have been
applied to the luminance data presented in Fig.
3(b).
While the sections analysed for this project
were chosen in part for their known variation in
sedimentological structure, the success of this
correlation technique in its most basic form
without any significant data processing clearly
demonstrates the potential of these measurements. It is believed that refinement of the
processing could yield considerable additional
data which, coupled to the non-destructive
nature of the methods, the ability to analyse
material within opaque liners and the speed of
data capture, makes the technique of very
considerable importance.

Development
Technically, the performance of the scanner is
excellent. Quantitative investigations of the
physical performance of the scanner, for example of the effects of variations in X-ray power
output or the influence of 'burn-out' around
unshielded plug holes etc., could potentially lead
to direct calibration of the luminance values in
terms of physical properties of the core material.
Improvements in the operating procedures, in

23

the loading of the core material and, in


particular, in the automated capture of the
image and luminance data could allow higher
throughput; enabling greater data sampling
densities and potentially more detailed data
processing and analysis.
Perhaps of greatest interest is the elimination
of the optical distortion error in the individual
image frames. While the necessary geometry of
the scanner is a major contributory factor to this
error, it is believed that image processing may
yield a significant and reliable reduction. By
viewing the image of a homogenous standard
(such as an aluminium block of similar size to
the core section), it is possible to store the
variations in the image due to this error in digital
form. By 'deconvolving' the standard image
obtained in this way from the images of the
scanned rock it may be possible to provide a
much flatter response from the imaging system.
This intermediate processing would allow an
accurate digital representation of the density of
the core across the full image to be produced.
Instead of collecting data at specific points, it
would then be possible to map the density
variation of the rock slab in two dimensions.
This would provide, not only a more accurate
log for comparison with downhole logs, but also
allow the density variation to be plotted as a
three-dimensional map, potentially highlighting
more subtle variations in the core structure.
Initial work carried out is encouraging.

Conclusion
Linear X-ray scanning has an established role in
non-destructive imaging of core, with the variation in image reflecting the density variation in
the core section. The techniques described here
allow not only the qualitative X-ray image to be
produced, but also quantitative luminance values to be extracted. These correlate very well
with physical core properties, for example bulk
density and porosity, derived from wireline or
conventional core analysis techniques. These
luminance values thus provide a valuable core
to log correlation tool which may be of
particular value where traditional Gamma Ray
or Core Analysis techniques are unavailable or
relatively unreliable due to poor response or
sparse data. The possibility of improving the
operating procedures, in particular the sampling
interval and processing methods, as well as
ultimately providing full density maps of the
core section promise to yield even greater
benefits, and confirm the importance of X-ray
imaging as a core analysis tool.

24

A . R . DUNCAN ET AL.

We gratefully acknowledge the permission to publish


this material granted by the Scott partner group:
Amerada Hess Limited, Amoco (UK) Exploration,
Deminex (UK) Oil and Gas, Enterprise Oil, Kerr
McGee Oil (UK), Superior Oil (UK) and Premier Pict
Petroleum. Our thanks goes to F. Matheson at
Robertson Research Int. Ltd who diligently and
expeditiously undertook the preparation and measurements of the core material analysed during this project.

References
ALGEO, T. J., PHILLIPS,M., JAM1NSKI,J. & FENWlCK,M.
1994. High resolution X-radiography of laminated sediment cores. Journal of Sedimentary
Research A: Sedimentary Petrology and Processes,
A64, 665-668.

DUNCAN, A. R., 1993. A sedimentological, petrographic


and reservoir geological study of the Devonian age,
old red sandstone of Gamrie Bay, Grampian
Region. M.Sc. Thesis, University of Aberdeen.
DUNCAN, A. R., MATHESON,F. E., & COLLIE, D. A. L.
1996. Quantitative X-ray density imaging of
selected cores. Robertson Research International
Ltd Project Report No D213 for Amerada Hess
Ltd.
RIGSBY, C. A., ZIERENBERG,R. A. & BAKER, P. A.
1994. Sedimentary and diagenetic structures and
textures in turbiditic and hemiturbiditic strata as
revealed by whole-core X-radiography; Middle
Valley, northern Juan de Fuca Ridge. Proceedings, Scientific Results, ODP leg 139, 105-111.
TOLANSKY, S. 1961. Introduction to Atomic Physics.
Longmans.

The estimation of modal mineralogy: a problem of accuracy in core-log


calibration
P. K. H A R V E Y , 1 T. S. B R E W E R , 1 M. A. L O V E L L 1 & S. A. K E R R 2

1Borehole Research, Department o f Geology, University of Leicester, Leicester, LE1 7RH,


UK
2 British Petroleum, Chertsey Road, Sunbury-on-Thames. Middlesex, TW16 7LN, UK

Abstract: In the case study described here the quantitative modal mineralogy of a number of
core samples was determined with the objective of using these modes to calibrate
geochemical logs. Modal estimates were obtained for the core samples by quantitative X-ray
diffraction, infrared spectroscopy, point counting of thin sections, and indirectly by
calculation from a complete chemical analysis of the samples. In the case of calculated
modes, three different algorithms were applied. A by-product of this particularly complete
dataset is the possibility of evaluating the most accurate method of modal analysis, and
although no certain conclusion is reached on this point the analysis of these data does
demonstrate the difficulty of obtaining accurate modal estimates. The core samples, taken at
regular intervals through a sand, sandy-shale sequence, capped by a carbonate unit, have a
mineralogy which, although dominated by quartz, includes feldspars, carbonates, and clays
(illite, kaolinite) together with minor phases. There was generally good agreement between
methods in the estimation of quartz, total carbonate, albite, kaolinite, total clay and pyrite.
The results for illite and K-feldspar were poor, a reflection of their relatively low
concentrations (< 10%), and problems of compositional co-linearity in the calculated
modes.

A useful way of presenting data from geochemical logging tools is to transform the raw oxide
curves into mineralogy logs. In a recent exercise
aimed at calibrating geochemical logs in a UK
borehole a number of core samples (103) were
taken and analysed extensively in the laboratory
for both chemistry and mineralogy, to provide a
database to support the log calibration. For all
103 core plugs quantitative mineralogy was
determined by X-ray diffraction at the British
Petroleum laboratories in Sunbury and by
infrared spectroscopy (MINERALOG) at Core
Laboratories. In addition a petrographic examination was carried out, and a minimal point
count made (200 points per thin section) on
approximately half the samples to provide
approximate modal data. All core plugs were
also chemiclly analysed by X-Ray Assay Laboratories (XRAL) in Ottawa for all major and
all potentially significant trace elements (a total
of 69 elements per sample). From the chemical
data, estimates of the modal mineralogy were
calculated using a selection of different algorithms. Together these analyses result in a range
of modal estimates and the purpose of this
contribution is to compare these estimates in an
attempt to evaluate the accuracy of the different
methods. Apart from the petrographic work, all

measurements were made on aliquots of the


same crushed and thoroughly homogenized rock
powder for each sample. There is, therefore,
essentially no scaling problem involved to
explain variations in the modal estimates, and
a minimal problem of sub-sampling from the
rock powder.

Background
Through the use of pulsed neutron devices,
direct activtion of the formation by appropriate
isotopes, and the natural gamma spectra it is
possible to obtain an almost complete, and
continuous log of the major element chemistry
of a formation. These techniques were pioneered
by Schlumberger (Hertzog & Plasek 1979;
Hertzog et al. 1987a, 1987b, 1989; Galford et
al. 1988; Rupp et al. 1989) with their Geochemical Logging Tool (GLT) offering measurements
of Si, A1, Ti, Fe, Ca, K, S, the minor elements
Gd, Th, and U, together with H and C1. Other
tools are now available (Wyatt & Jacobson et al.
1993; Odom et al. 1994; Jacobson & Wyatt 1996,
Herron & Herron 1998). Transformation of the
major elements into the more conventional oxide
form gives virtually complete major element
oxide analysis at each measured depth interval,

HARVEY,P. K., BREWER,T. S., LOVELL,M. A. & KERR, S. A. 1998. The estimation of modal
mineralogy: a problem of accuracy in core-log calibration In. HARVEY,P. K. & LOVELL,M. A. (eds)
Core-Log Integration, Geological Society, London, Special Publications, 136, 25-38

25

26

P. K. HARVEY E T AL.

typically every 15cm, down the borehole. One


approach to the interpretation of the resulting
geochemical logs is their conversion into computed mineral assemblages, the so called 'chemical modes' of Wright and Doherty (1970). The
resulting mineralogy logs are valuable in their
own right but may be used in addition with a
suitable rock classification filter to produce
lithological logs (Herron 1988), or estimation
of other formation properties such as matrix
(grain) density, porosity, Cation Exchange
Capacity (Chapman et al. 1987; Herron 1987b;
Herron & Grau 1987), thermal conductivity
(Dove & Williams 1988), heat flow (Anderson &
Dove 1988), photoelectric factor, Pe (Kerr et al.
1992), magnetic susceptibility (Harvey et al.
1997), fluid saturation (Hastings 1988), neutron
capture cross-section (Herron 1987b), and,
indirectly and probably only in formationspecific situations, permeability (Herron 1987a).
The transformation of a rock's elemental
composition to mineral assemblages has been
the subject of many contributions. Igneous
petrologists have long employed the C.I.P.W.
norm (Cross et al. 1903), and similar ideas have
been extended to metamorphic rocks with
Niggli's normative procedures (Burri 1964). In
calculating norms there is no requirement that
the mineral assemblage used is that which
actually occurs in a rock, and as such the
C.I.P.W. norm, for example, was developed
originally for purposes of classification and
employed a strict and standard list of 'possible'.
minerals. However, for the purposes to which
geochemically derived mineralogy logs might be
put (lithological analysis, basin modelling, petrophysical estimation) it is necessary to estimate
the percentage of the minerals that are actually
present in the rock. The latter is the 'mode' of
the rock which is conventionally determined
directly by micrometric analysis (point counting
of a thin section) or spectral methods such as Xray diffraction or infrared spectroscopy (Harville
& Freeman 1988; Adam et al. 1989; Matteson &
Herron 1993).
The alternative approach is to compute the
mode from a complete chemical analysis. Numerous authors have offered specific solutions to
this inversion problem including modified normative schemes (Imbrie & Poldervaart 1959;
Nicholls 1962; Pearson 1978), graphical models
(Miesch 1962; Fuh 1973) and a variety of
numerical models including least squares minimization, linear programming and genetic algorithms (Wright & Doherty 1970; Albarede &
Provost 1976; Fang et al. 1996), while others
have considered the strategy and associated
practical problems of performing the inversion

on a routine basis (Herron 1986; Harvey et al.


1990; Harvey & Lovell 1992; Harvey et al. 1992;
Lofts et al. 1994, 1995b).
For the case study described here a particularly complete dataset is available consisting of
conventional (physical) modal mineral measurements, together with comprehensive chemical
analyses from which calculated modes could be
obtained. It is the particularly complete nature
of these data which justifies a comparison
between physical and calculated methods of
modal analysis, and the opportunity to make
some comment on the accuracy of modes. Using
natural samples (borehole core plugs) in this
case, however, precludes any definite means by
which one method can be chosen as 'more
accurate' than another, unless a particular mode
is 'obviously' wrong. As a first order assumption, however, it is likely that if two or more
unrelated methods of estimation give essentially
the same result then they are probably close to
the true value. This level of uncertainty arises
because all methods of modal estimation can
give seriously erroneous results sometimes, and
for some minerals; it is not simply a question of
calibration and precision (repeat measurement
error). With the spectral techniques particular
problems arise from spectral overlap and poor
resolution at the lower concentrations. With
computed modes it is the choice of the correct
mineral assemblage, the correct 'composition'
and possible problems of compositional colinearity which are important.
Modes are usually obtained by micrometric
analysis (point counting of a thin section) and as
such are usually expressed in volume percent of
the optically identified minerals. In contrast, a
set of modal proportions may be calculated, to
give 'norms', by assuming an ideal or theoretical
suite of minerals, and the compositions for those
minerals. From these data some sort of fit may
be found that partitions the mineral compositions within the initial rock analysis. Norms are
usually expressed in weight percentages, and
have found very wide application, particularly in
igneous petrology, for characterization and
classification. In these applications comparison
between rocks is made with a common set of
(chosen) minerals, unlike the mode, which
reflects the actual minerals present. Normative
mineralogy logs may be useful in the early stages
of an investigation but are no substitute for the
estimation of the actual mineral percentages
present if attempts are to be made later to
generate, for instance, a grain density log. In this
report the attempt is made to calculate chemical
modes, or the mineral proportions of the actual
minerals present in the sample.

THE ESTIMATION OF MODAL MINERALOGY

27

Table 1. Mineralogy of the core samples.


Technique
Silica
Feldspars
Carbonates

Clays/Micas

Minor phases

Quartz
Albite
K-feldspar
Calcite
Ankerite
Dolomite
Siderite
Muscovite
Illite
Smectite
Kaolinite
Chlorite
Zircon
Barite
Pyrite
Apatite

XRD

MINERALOG

Petrography

*
*
*
*
*

*
*
*
*

*
*
*
*

*
*

*
*
*
*

*
*
*

*
*
*

*
*

*
*
*
*
*
*

* mineral detected in at least one of the core samples. Anhydrite and chlorite were not detected in
any of the samples by the infrared (MINERALOG) technique. See report for further comments.

Mineralogy of the core samples


The mineralogy of the core samples derived
from a combination of XRD, infrared (MINERALOG) and petrographic information is summarized in Table 1, and also shown as a
downhole mineralogy log in Fig. 1. The 100
metre sequence consists of clastics covered by
some 8 m of limestone. For purposes of description the section can be divided into five units:
(Unit 1) (360-368 re)virtually pure calcite limestone;
(Unit 2) (368-382m) quartz rich (60-65%) section with sub-equal quantities of
feldspar (both albite and K-feldspar)
and clays. Both kaolinite and illite (the
latter generally in excess) are present;
(Unit 3) (382-397m) quartz-carbonate dominant lithology with quartz in excess,
and sub-equal proportions of kaolinite
and illite;
(Unit 4) (397-430 m) Very inhomogeneous section with 40 to 70% quartz, no
significant carbonate, and clay concentrations up to c.30%; kaolinite
generally in excess of illite. Locally
high concentrations of pyrite (included
with 'minor' minerals in Figure 1).
( U n i t 5) ( 4 3 0 - 4 6 0 m ) R e l a t i v e l y u n i f o r m
quartz-rich (70-80%) section with
minor feldspar, about 15% clay, some
two-thirds of which is kaolinite, with a
few percent of minor minerals.
Of the possible carbonate phases calcite,

dolomite and siderite are clearly distinguishable


petrographically in stained thin sections and are
quantified as those species in the infrared
( M I N E R A L O G ) results. In the X R D analyses,
ankerite (a Ca, Mg, Fe carbonate) is quantified
in place of dolomite, so that the two spectral
techniques are not estimating the same carbonate species. The carbonates occurring in Units
2 and 3 are dominantly dolomite or ankerite,
with minor amounts of siderite, while the
limestone unit at the top of the logged section
(Unit 1) is a virtually pure calcite limestone.
For the clay and other phyllosilicate minerals,
kaolinite and chlorite are measured together in
the X R D analyses. Chlorite was only seen in
thin section in occasional grains and was not
detected in the infrared figures. For this study
chlorite is considered to be absent. Muscovite
and illite are also determined together by XRD.
Occasional distinct flakes of white mica are
present in a number of sections but in no case
would these make up more than a fraction of
one percent of the rock. While white mica
(assumed to be muscovite) is known to be
present in a very small amount in some samples
it was not detected by infrared spectroscopy, and
is virtually impossible to calculate with any
reliability due to a strong compositional colinearity with K-feldspar, illite and kaolinite
which are present in significant proportions.
Included amongst the minor phases which
occur in at least some of the samples are zircon,
barite, apatite and pyrite, all of which have been
identified petrographically. Of these, only pyrite
occurs locally in sufficient quantity to be
identified and measured by both X R D and

28

P.K. HARVEY E T AL.

Fig. 1. Computed mineralogy log (Model A) for the section under study and showing the stratigraphic units
discussed in the text. For clarity only, the major mineral groups are shown. The depth scale in arbitary.
infrared. Of the other three minerals barite was
detected by infrared, but both apatite and zircon
were too low for the spectral methods.

Numerical modelling of the core sample


mineralogy
The estimation of a modal mineral assemblage
from the chemical analysis of a sample requires
the minerals in the assemblage to be chosen, and
the compositions of those minerals to be defined.
Given this information there are a variety of
solution methods and strategies that can be
employed to solve for the mode (Harvey et al.
1990, 1992; Lofts et al. 1994; Fang et al. 1996).
For the modelling of the samples described
here, the main minerals are quartz, feldspars
(albite, K-feldspar), carbonates (calcite, dolomite, siderite), clays (kaolinite, illite) and the
minor phases (zircon, barite, apatite, pyrite).

From the mineral data above, all the observed


mineral assemblages can be established, and
these, minor phases excluded, are summarized in
Table 2; in all a total of thirteen different
assemblages.
From the chemical viewpoint the following
components are available for modelling: SiO2,
A1203, TiO2, Fe203, MgO, CaO, Na20, K20,
MnO, P205, S, CO 2 and H 2 0 + , expressed in
weight percent, together with Ba and Zr which
were reported in parts per million. No other
'minor' elements are in sufficient concentration
to be expected to form discrete mineral phases,
or significantly alter the modal estimates of
other mineral phases in which they might occur
as trace lattice components.
Zr and Ba are considered to occur only in
zircon and barite, respectively. In addition,
amongst the oxide components P205 almost
certainly occurs at significant levels only in

29

THE ESTIMATION OF MODAL MINERALOGY

Table 2.

Observed mineral assemblages in the core samples (excluding minor phases).

Assemblage

10

11

12

Silica
Feldspars

*
*
*
*

*
*

*
*
*
*

*
*

*
*

*
*

Quartz
Albite
K-feldspar
Carbonates Calcite
Dolomite +
Siderite
Clays/Micas Illite
Kaolinite

*
*

*
*
*
*

*
*
*

*
*
*
*
*

*
*
*
*

13

*
*
*

*
*

*
*

* mineral detected in at least one of the core samples. Anhydrite and chlorite were not detected in any of the
samples by the infrared (MINERALOG) technique. See report for further comments.
+ dolomite or ankerite. See text for explanation.
apatite which has been identified petrographically. TiO2 poses a problem, and in the first
instance is best calculated as rutile, though there
is no evidence that this mineral actually occurs
in any of the samples. TiO2 may also be present
in one of the clay phases, and this problem is
discussed later where there is good evidence that
it actually occurs in different minerals in
different parts of the section.
Sulphur is assumed to be present only as a
component of pyrite. Other minerals, such as
gypsum or anhydrite are possibilities, though
there is no evidence for any sulphates being
present, and pyrite is the only identified sulphide.
Manganese, which is only present at a very
low level (maximum 0.75% MnO, 90% of
measurements less than 0.18% MnO), was
a d d e d to iron (as FeO) for purposes of
computation. Manganese often substitutes for
iron, and the significant correlation (at a = 0.05;
r = 0.58) between the two elements in these data
is consistent with this occurring here. Removing
MnO leaves a total of 14 possible mineral phases
and 15 chemical components to consider.
Of the several strategies employed in the
modelling of the mineral assemblages in this
case history three simple methods are presented
here. In each case the data were pre-processed to
remove the minor phases rutile, apatite, barite,
zircon and pyrite which were calculated out of
each core analysis assuming ideal stoichiometric
compositions. Provided the chemical analyses of
the core samples are accurate this procedure
gives excellent estimates for these minerals which
cannot be matched by any direct measurement.
Although treated here as a minor phase, pyrite
does reach significant concentrations in a few
samples; the variation in pyrite downhole is
shown in Fig. 2, and is discussed later.
With extraction of these minor minerals TiO2,

BaO, ZrO2, S and P205 were removed from the


data matrix leaving SiO2, A1203, FeO, MgO,
CaO, Na20, K20, CO2 and H 2 0 + to be
distributed, as appropriate, between the important remaining mineral phases: quartz, albite, Kfeldspar, calcite, dolomite, siderite, kaolinite,
illite and possibly smectite. The simultaneous
estimation of all these minerals together would
constitute a fully determined system for methods
of inversion involving the solution of systems of
equations.

Strategies for extraction of the main mineral


phases: Models A, B, C
To remove complications related to methods of
solution a simple unconstrained and unweighted
least squares method has been used throughout
(Harvey, et al. 1990) for inverting the different
models. One consequence of the lack of constraint is that mineral proportions may be
negative, implying an insufficiency of a combination of elements with respect to a 'perfect'
solution. Such negative estimates, while impossible, offer a guide to the fact that either the
modelled assemblage is wrong, or one or more
of the mineral compositions are in error. Clearly,
such a solution is unacceptable. One approach,
then, is to model a given composition with all
likely minerals, and to reject those minerals
which turn out negative. This is the basis of
Model A, described below. Another approach is
to model each given composition to all the
mineral assemblages which are known to occur
in the section (Table 2), and chose the best fit as
the appropriate solution. This is the basis of
Model B. For Model C the mineral assemblage
obtained from the X R D analysis was taken as
correct, and the given composition fitted to that
assemblage. This latter approach, in principle,

P. K. HARVEY ET AL.

30

.....

360

'

"

'

'

'

'

'

'
Unit

'

'

'

'

'

'

'

'

"

'

'

'

360

370

370

Pyrite

380

.~so

390

390

Quartz

4O0

.......

_: _~:~
"'~-~ ....

].... :--

~i~--~_~.:

,.
4O0

.:_

~_

........

410

41o

3
420

420
.........
~z-~,

430

Py-/a)
Py-(b)

9
440

'

430

Py-(m)
Py-(x)
_
........

440

450

450

20

40

60

8O

'

20

100

360

'

"

'

40

'

60

'

80

'

460

'

1O0

. ..

360

370

370

380

380

390

390

400

400

410

410

420

Unit4

420

430

430

440

440
I

450

Uni|5

450

460

460
0

20

40

60

80

100

20

40

60

80

tOO

Fig. 2. Pseudo-log showing the mineralogical variation of the core samples downhole for quartz, K-feldspar,
albite and pyrite. Qtz-(a): quartz computed from Model A, Qtz-(b): ditto, for Model B, Qtz-(m): MINERALOG
measurement of quartz, Qtz-(x): XRD measurement of quartz. Coding as for quartz for: pyrite (Py), K-feldspar
(Kf) and albite (Ab).

removes the 'guesswork' out of the choice of


mineral assemblage.
M o d e l A.
minerals

iterative removal o f n e g a t i v e

Solutions with negative compositions imply that


there is an inconsistency in the postulated
mineral compositions (as stated above) and a
simple means of overcoming this is by removing
the mineral from the analysis. Wholesale removal of all negative minerals in one pass,

however, cannot be justified because of the


complex interaction of phases in a least squares
model. With little or no formal justification one
procedure we have found very effective is to
remove the most negative phase and re-solve the
system. If negative phases still occur the
procedure is repeated until all phases are
positive. The procedure is illustrated in Table 3
for sample K78. The least squares fit using all
nine minerals is good but dolomite is slightly
negative (a) at --0.21. Removing dolomite as
one of the phases sends calcite slightly negative

31

THE ESTIMATION OF MODAL MINERALOGY

Table 3. Example of the successive removal of negativephasesfor sample K78.


a
Quartz
Albite
K-feldspar
Calcite
Dolomite
Siderite
Kaolinite
Illite
Smectite
Std. Err.

79.81
0.21
5.42
0.01
-0.21
0.22
12.13
0.00
1.29
0.011

79.97
0.12
5.47
-0.12

80.05
0.00
5.30

0.21
12.35
1.14
1.22
0.057

0.13
12.44
0.81
0.96
0.067

XRD

MINERALOG

88
0
3
0
0
0
7
0
0

80
0
7
0
0
0
5
7
0

(a) unconstrained and unweighted least squares model.


(b) as (a) but with dolomite (negative proportion) excluded.
(c) as (b) but with calcite removed to give a fully positive solution.
Corresponding XRD (X-ray diffraction) and Mlog (MINERALOG) estimates are given for
comparison. (XRD includes 2% pyrite, Mlog 1%). Std. Err.: standard error (see text for
calculation).
(b). In a final stage calcite in removed (c) to give
the final estimate. The mineralogical pseudo-log
shown in Fig. 1 was produced using this model.
For completion in Fig. 1, the trace minerals were
added and the Model A derived mineralogy
normalized to make the assemblages sum to
100%. The standard error given in Table 3 is a
measure of the fit of the core and mineral
chemistry and is computed between the original
(input) chemistry, and the composition backcalculated from the derived (output) mineralogy
(Harvey et al. 1990).

For each core sample a list of the minerals


present had already been identified by X R D ,
petrography or infrared analysis. For this model
the X R D mineral assemblage was chosen for
each sample and then fitted accordingly. Poor
fits, often with negative mineral estimates,
identify a real incompatibility between the rock
and mineral chemistry assuming that the phases
are correctly identified by the XRD.

Model B. choice o f known mineral


assemblages

Comparison of measured and calculated


estimates of mineralogy

Thirteen possible parageneses for the core


samples have been identified, within limits of
detection, from the X R D , petrographic and
infrared data. These are summarized in Table
2. Most assemblages contain five or six phases;
only one (assemblage 3) contains over six (7), so
that the number of minerals is generally at least
two less than the number of chemical components. The procedure was to fit each sample to
each of these possible assemblages. The optimum assemblage was then chosen using the
following criteria:

Figs 2 through 4 summarize the variation in


measured (XRD & M I N E R A L O G ) and computed (Models A & B) values for quartz,
feldspars and pyrite (Fig. 2), carbonates (Fig.
3), and the clays (Fig. 4). A more detailed
comparison may be made by examination of
Table 4 which shows the correlations between all
models and the physically derived measurements. In Table 4, based on 103 core samples,
correlations > 0.19 are different from zero at a
significance level of 0.05, and > 0.25, at a level of
0.01.
For quartz, the best correlation is between
Model A and the infrared ( M I N E R A L O G )
data; the relationship is linear (Fig. 5) and close
to the 1:1 line. The agreement between the two
is particularly good in the lower half of the
section (Units 4 & 5), though in the upper part,
below the limestone cap, the computed quartz
estimates are almost consistently lower. Regression analysis of this relationship gives a slope of

(a) in virtually all cases this procedure yields a


selection of potentially acceptable (nonnegative) assemblages; the one with the
smallest standard error is then chosen;
(b) but, if all possibilities gave at least one
negative mineral proportion the assemblage with the smallest absolute negative
sum was chosen.

Model C: fit to the individual mineral


assemblage for each core sample

32

P. K. HARVEY ET AL.
,

360

370

Unit2

380

380

Umt

390

390

400

400

Ankerite / Dolomite

Siderite

4.10

41o
42O

420

..............

430
i

Sid-(a) 71

Sid- ( b) i

430

Dol-(a)
Dol-(b)

440

44O

iI

Sid-(m) [

[ .....

Sid-(x)

- -

Dol-(m) :
- - - Ank-(x)

450

f!

4(30

20

40

60

80

.............

I ,

20

40

450
.

'

'

460

80

60

1(30

36O

360

2%.~_

370

370

~ "

380

39O

4OO

411)

Total carbonates

Calcite

4~o
420

420
;.m

Cal-(a)

.................
9 Carb-(a)

Cal.(b)

43O
9
9
- -

Cal.(m)

.....

Cal-(x)

450

20

40

60

80

20

40

430

Carb-(b)

440

- -

Carb-(m)

.....

Carb-~x)

60

80

450

460

10CI

Fig. 3. As Fig. 2 for the carbonates: siderite (Sid), Dolomite (Dol, but reported as Ankerite (Ank) for the XRD
analysis); calcite (Cal) and total carbonate (Carb) which is taken as the sum of (siderite+ankerite/
dolomite + calcite).

1.03 and standard error of 2.36% quartz after


removal of four outlying points. In view of this
excellent relationship is of note that the correlation for quartz between the X R D and the
infrared is slightly lower with the X R D measurements being generally higher than the
infrared, often by several percent. Overall for
quartz, all models perform reasonably well and
show similar patterns of variation. It is not at all
clear which set of data is correct!
Of the feldspars, both albite and potashfeldspar are present in small quantities. Albite
is virtually restricted to Units 2 and 3, with

estimates averaging between 7% and 9% for


Models A and B, respectively. The infrared
figures agree at 7.4%; the X R D average is lower
at 2.6%. The highest linear correlation is again
shown between Model A and the MINERALOG, and it is likely that these methods are
giving close to the true result. For the calculated
modes the albite concentration is constrained by
the sodium concentration, and in the absence of
any other sodium bearing mineral, an accurate
estimate should be expected. With albite occurring in concentrations below 10% there are
problems of sensitivity and detection limit with

THE ESTIMATION OF MODAL MINERALOGY

33

.,.,..,,...,...,

360

. . .

On, I
370

:~':.;

~80

:,(,...~:

Kaolinite

. L

4304204,040039~
i'~~;
_

Total clays

Umt3

!;-'<; "

thai,4

umt4

Unit

~ , ~

:y~.j

UnitZ
u.*,3

450

) .....

Kaoi-(x)

Ill-(b)
- - Ill-(m)
:":. -. .....
. . . Ill-(x)

- ~

Y~.

20

40

6o

8o

u,~l 5

Ioo

20

40

60

80

10o

Fig. 4. As Figure 2 for the clay phases: kaolinite (Kaol); Illite (Ill); and total clay (Clay) which is taken as the sum
of (kaolinite + illite + smectite).

Table 4. Correlations between computed modes and infrared (MINERALOG) estimates (top table) and between
computed modes and XRD (lower table).
Mineral

A/Mlg

Quartz
Albite
K-feldspar
Kaolinite
Illite
Calcite
Siderite
Dolomite
E carbonates
E clays
Pyrite

0.987
0.865
0.193
0.928
0.025
0.976
0.869
0.901
0.998
0.911
0.975

B/Mlg
0.980
0.837
0.431
0.916
0.512
0.975
0.852
0.587
0.998
0.888
0.975

C/MIg

Mlg/XRD

0.954
0.662
0.192
0.973
0.760
0.975

A/XRD

0.962
0.712
0.515
0.785
0.587
0.994
0.848
0.953
0.994
0.856
0.968

0.966
0.847
0.146
0.809
-0.071
0.983
0.836
0.904
0.993
0.875
0.972

B/XRD
0.953
0.862
0.433
0.784
0.227
0.979
0.802
0.529
0.994
0.821
0.972

C/XRD
0.948
0.586
0.274
0.978
0.771
0.972

Bold: highest correlation for that mineral.

Italic: correlations insignificantly different from zero at a level of 0.01.


A, B, C: computed modes, Models A, B and C, respectively.
XRD: X-ray diffraction.
Mlg: MINERALOG.

both X R D and infrared methods; in view of this


the agreement between all methods is remarkable.
Potash feldspar occurs throughout the section
below the limestone cap at concentration levels
similar to albite (Fig. 2). There is poor agreement in detail between the models; Model B
shows the highest correlation of the computed
models, but is little lower than the weak
correlation of 0.515 between the X R D and
infrared figures (Table 4). The estimation of Kfeldspar suffers both from problems of low
concentration in the physically derived estimates
and potential compositional co-linearity in the

computed models (Harvey et al. 1992).


For the carbonates there is overall excellent
agreement between all the methods (Figs 3 & 6)
in that the total amount of carbonate (calcite + dolomite/ankerite+ siderite) determined by the
different methods is essentially the same. In
detail, however, there are some distinct differences between the measured and calculated
mineral percentages. These effects are seen
clearly in Fig. 3 where calcite is correctly and
accurately estimated by all methods in the
virtually pure calcite limestone cap, but in Units
2 and 3 below, where more than one carbonate
mineral is present, dolomite is severely under-

34

P.K. HARVEY E T AL.

Fig. 5. Crossplot of measured and estimated quartz


contents for the three computed models and XRD (yaxis), shown relative to the infrared (MINERALOG)
(x-axis) measurements. Qtz-(a): quartz computed from
Model A; Qtz-(b): ditto, for Model B; Qtz-(c): ditto,
for Model C; Qtz-(x): XRD measurement.

Fig. 6. As Figure 5 for total carbonate (calcite +


dolomite/ankerite + siderite)percentage. Carb-(a):
total carbonate computed from Model A; Carb-(b):
ditto, for Model B; Carb-(c): ditto, for Model C; Carb(x): XRD measurements.

estimated by the calculated models, and siderite


is slightly over-estimated. Calcite is included in
results of both Models A and B but was not
detected by either X R D or infrared methods.
These differences are due essentially to the use of
ideal carbonate compositions (Table 5) in the
mineral calculations and the compositions of the
actual carbonates present in these samples not
being available. The use of an ankeritic compo-

Fig. 7. Regression expressing kaolinite calculated using


modal Model A as a function of the infrared
(MINERALOG) measurements. The relationship is
seen to be linear with a slope close to unity, but an
intercept of some 5%.

sition instead of pure dolomite, for example,


would have caused the siderite estimates to be
lower (and more comparable with the X R D /
infrared figures) and the dolomite/ankerite
estimates higher. Hence, for the carbonates the
accuracy of the calculated modes is compromised through the use of inappropriate mineral
compositions; a knowledge of the latter is
essential if accurate solutions are to be obtained
(Lofts et al. 1995a)
Of the clays, kaolinite and illite were determined by all methods; the comparative results
are summarized in Fig. 4. The closest agreement
between computed and physically determined
estimates of kaolinite are shown between Model
A and infrared ( M I N E R A L O G ) ; the agreement
between X R D and infrared being distinctly
poor. A crossplot of the Model A/infrared
relationship is shown in Fig. 7. Apart from
three obvious outlying points the latter is
essentially linear, with a slope close to 1.0, but
an intercept which over-estimates the Model A
values compared to infrared, on average, by 5%
kaolinite. The two most outlying points in Fig. 7
occur as outliers on other plots and come from a
section which the core photographs indicate to
be very inhomogeneous. Sample preparation
should have removed this problem for laboratory work; their gross deviation remains a
problem.
For illite there is no real agreement between
any of the models. The closest relationship is
shown between the X R D and infrared figures,
but with a correlation coefficient of only 0.587 it

THE ESTIMATION OF MODAL MINERALOGY

35

Table 5. Compositions of the model minerals used to evaluate Models A, B and C.

SiO2
A1203
FeO
MgO
CaO
Na20
K20
H20
CO2

Qtz

Ab

K-f

Cal

Dol

Sid

100.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00

68.74
19.44
0.00
0.00
0.00
11.82
0.00
0.00
0.00

64.76
18.32
0.00
0.00
0.00
0.00
16.92
0.00
0.00

0.00
0.00
0.00
0.00
56.03
0.00
0.00
0.00
43.97

0.00
0.00
0.00
30.41
21.86
0.00
0.00
0.00
47.73

0.00
0.00
62.02
0.00
0.00
0.00
0.00
0.00
37.98

Kaol
45.48
39.29
0.65
0.14
0.41
0.00
0.00
14.04
0.00

Ill
57.28
18.55
5.11
2.07
1.59
0.43
5.11
8.86
0.00

Smec
51.14
19.76
0.83
3.22
1.62
0.11
0.04
22.80
0.00

Qtz: quartz, Ab: albite, K-f: K-feldspar, Cal: calcite, Dol: dolomite, Sid: siderite, Kaol: kaolinite, Ill: illite, Smec:
smectite.
Oxide concentrations are in weight percent oxide.
is a poor predictive relationship; Model B has
the highest correlation of the computed models
(r=0.512), with the infrared estimates, though
its correlation with the X R D data is nonsignificant (at a level of 0.05). Depending upon
which set of data is believed illite is present
throughout most of the section (Fig. 4) with the
highest values (10-20%) in Units 2 and 3. It is
difficult quantify using the physical methods
because of poor sensitivity and spectral variation, and to compute because of uncertainty
about the mineral chemistry and chemical
variation in these rocks. Despite these problems
the total clay curve shown in Fig. 4 shows
surprisingly good agreement over the range of
methods.

Discussion
The results described here came from a study of
mineral inversion methods to determine the way
in which the most accurate mineralogy log could
be obtained from a suite of geochemical logging
data. The chosen method would have to involve
calculation from the geochemistry, and validation of these results would need to be by
comparison with some independent method.
For the samples used in this validation a
particularly comprehensive dataset was available
from X R D and infrared ( M I N E R A L O G ) analyses made on the same samples. The latter, and
the geochemical analyses upon which the modal
calculations were based, were all measured on
aliquots of the same homogenized powder for
each sample. All measurements were, hence,
made on essentially the same material, and on
very similar volumes of the same material in
each case. In the comparisons made here there
is, therefore, no real problem of heterogeneity or
scaling (up or down) involved; just a simple

comparison between a range of methods for the


estimation of the proportions of different
minerals in a set of samples.
The question is, which is the 'best' (most
accurate?) method for calibrating mineralogy
logs, or whether the calculations on their own
are actually superior. All methods obviously
have their strengths and weaknesses. The major
problems with the spectral methods concern
sample preparation (and presentation), sensitivity and spectral resolution at low levels (lowest
few percent) and spectral interferences. The
main problems with calculating the mode concern the accuracy of the sample's chemical
analysis, the need to solve for the correct phase
assemblage and to have a good estimate of the
phase compositions, and compositional colinearity. Modern methods of quantitative geochemical analysis are much more precise than
the direct methods of modal analysis compared
here, and with appropriate calibration may be
expected to produce estimates of the chemical
components which are close to the actual values.
Some effects of concentration errors on elementto-mineral inversion are discussed by Herron &
Chiaramonte (1993).
We have shown elsewhere that if the chemical
analysis is correct, the phase assemblage known,
and the minerals in that assemblage analysed,
then the calculated mode is in excellent agreement with the 'true' mineral proportions (Lofts
et al. 1995b). While this may seem self evident
the 'quality' of a calculated solution very rapidly
deteriorates as analytical errors increase. Where
an element is virtually restricted to one mineral
phase, and particularly if that mineral is only at
levels of a few percent or less, such as sulphur in
pyrite, sodium in albite or barium in barite, then
the calculated mode for these minerals is likely
to give the best estimate. In the more general

36

P.K. HARVEY E T AL.

case where elements are partitioned between


different minerals, a good fit between the input
sample chemistry and the chemistry reconstructed from the mineral proportions (Harvey
et al. 1990) may provide very good evidence that
the solution is close to the 'true' mineral
proportions.
Similar arguments may be produced for the
quality of XRD and M I N E R A L O G solutions
under specified conditions. For several of the
minerals examined here, close agreement can
occur between different methods and it may be
postulated that if a number of unrelated
methods give essentially the same result for a
sample it is likely that that result is close to the
'true' value, though this cannot be proven.
This would suggest that for the results
described here, quartz, pyrite and (total) carbonate are reasonably accurate, despite a minor
bias in the quartz crossplot, and variations
between the individual carbonates could be
easily reconciled by the use of the correct
carbonate mineral compositions. It is significant
that these minerals which do give reasonable
agreement between widely differing methods are
characterized by a small number of cations and
restricted compositions which result in well
defined XRD and infrared spectra.
The variation between the different estimates
for the feldspars (K-feldspar and albite) and the
clays (illite and kaolinite) is more complex. Both
spectral methods (XRD and infrared) suffer
from problems of overlapping lines for these
minerals and poorer sensitivities (i.e. detection
limits) at low concentrations. Likewise, calculations of the modes from the chemical analysis
suffer from two other problems. The first is the
use of the 'true' chemical composition of each
mineral phase used in a model. It has been
demonstrated previously (Lofts et al. 1995b)
how sensitive a solution can be to changes in
compositions of the minerals used in calculations; ideal compositions are rarely appropriate.
The actual compositions of the kaolinites, illites
and the different carbonate minerals were not
known in this study. The second problem is that
of compositional co-linearity; that is, where four
or more of the minerals to be modelled lie on (or
very close to) the same compositional plane
(Harvey & Lovell 1992). This leads to essentially
an infinity of solutions, or, if forced, a very
unstable solution. The variable range of estimates for K-feldspar amongst the computed
values almost certainly results from this problem, given that the actual compositions of the
clays (which have a number of chemical
components in common with the feldspars) are
not known.

For minor minerals, such as zircon, apatite


and barite, which are definitely known to be
present (confirmation generally by petrographic
examination) the only reliable method of estimation is calculation. In most samples these
minerals were not detected by the spectral
methods, but as they contain elements which
do not normally occur at significant levels in the
other minerals present (Zr in zircon, P in apatite,
and Ba in barite), and the chemical analyses can
be regarded as both more accurate and precise
than the spectral mineral techniques, then
accurate estimates may be expected. The same
is true for pyrite (constrained by the S content)
in the absence of other sulphur bearing minerals
such as gypsum or anhydrite.
If the chemical compositions of all the mineral
phases were known, then their use, together with
'good' chemical analyses should always produce
the most accurate modal estimates by calculation using the actual mineral assemblage for a
given sample. The mineral compositions, however, are rarely known in sufficient detail and
quite erroneous estimates can be made if the
wrong compositions are used.
In the case history described here, Model A
was finally chosen for modelling the geochemical
log data following further experiments using a
reduced number of components to correspond
to those measured by the geochemical logging
tool.

Conclusions
(1) There was generally good agreement between methods in the estimation of quartz,
total carbonate and pyrite. It is reasonable
to assume, but cannot be proven, that these
estimates are close to the 'true' values. It is
significant that those minerals which do
show good agreement between widely differing methods have fairly simple and limited
compositions (i.e. are stoichiometric).
(2) Agreement for the clay minerals and the
feldspars is much more variable due to
problems of sensitivity and spectral interference for the two physical methods of
analysis, and problems of uncertain mineral
composition and compositional co-linearity
for the computed models. Good agreement
is seen between methods for albite, even at a
low level, and kaolinite. The results for illite
and K-feldspar were comparatively poor
and it is considerably more difficult here to
judge which figures, if any, are close to the
correct values.
(3) Despite the problems with illite, excellent
agreement is seen between the methods for

THE ESTIMATION OF MODAL MINERALOGY

(4)

(5)

(6)

(7)

total clay content, which is particularly


useful for the calulation of shale components
in lithofacies modelling from the geochemical logs.
Poor agreement between methods results
from low sensitivity (especially at low concentrations) and spectral interferences in the
X-ray and infrared techniques, problems of
compositional co-linearity and uncertainty
about the actual compositions of some of the
minerals in the calculated modes.
If the chemical compositions of all the
mineral phases were known, then their use,
together with 'good' chemical analyses,
should always produce the most accurate
modal estimates by calculation using the
actual mineral assemblage for a given
sample. The mineral compositions, however,
are rarely known in sufficient detail and
quite erroneous estimates can be made if the
wrong compositions are used.
Whether 'absolute' accuracy is actually
important obviously depends on why he
mineralogy log is required; probably not if
simply to illustrate diagrammatically the
relative variation in a sequence but important if the data are to be used, for instance,
for quantitative basin modelling or physical
property log estimation.
Underlying these comparisons is the fact
that the accuracy of any measured parameter is virtually impossible to specify,
except in (usually quite unrealistic) limiting
cases. Even with this particularly comprehensive dataset no definitive conclusions can
be demonstrated concerning the accuracy of
different estimates; the analysis of these data
does, however, demonstrate the difficulty of
obtaining accurate modal estimates. And in
this case history there are no real problems
of homogeneity or scaling in the estimation
process!

The authors would like to thank BP Exploration


Operating Company for agreeing to the publication of
this work. We would also like to thank Core
Laboratories for the provision of the MINERALOG
measurements which were made in December 1988.

References
ADAM,H. G., HARVILLE,D., MEER, D. & FREEMAN,D.
1989. Rapid mineral analysis by Fourier transform infrared spectroscopy, Society of Core
Analysts, Annual Technical Conference Reprints,
v. 1 (198%1989), part II, Society of Professional
Well Log Analysts, paper SCA-8809, I.
ALBAREDE, F. & PROVOST,A. 1976. Petrological and
geochemical mass-balance equations: an algo-

37

rithm for least-square fitting and general error


analysis. IPGP NS 252, 309-326.
ANDERSON,R. & DOVE, R. 1988. The determination of
heat flow in a wellbore in the South Eugene Island
area of offshore Louisiana: implications for fluid
migration and hydrocarbon location in the subsurface. Transactions Spectroscopy and Geochemistry Symposium, Schlumberger-Doll
Research, Ridgefield, CT. Paper K.
BURRI, C. 1964. Petrochemical calculations based on
equivalents (Methods of Paul Niggli). Israel
Program for Scientific Translations, Sivan Press,
Jerusalem.
CHAPMAN, S., COLSON,J. L., FLAUM,C., HERTZOG, R.
C., PIRIE, G., SCOTT,H., EVERETT,B., HERRON,M.
M., SCHWE1TZER,J. S., LA VIGNE,J., QUIREIN,J. &
WENDLANDT, R. 1987. The emergence of Geochemical Well Logging. The Technical Review, 35,
27-35.
CROSS, W., IDDINGS,J. P., PIRSSON, L. V. & WASHINGTON, H. S. 1903. Quantitative classification of
igneous rocks. University of Chicago Press,
Chicago.
DovE, R. E. & WILLIAMS, C. F. 1988. Thermal
conductivity estimated from elemental concentration logs. Transactions Spectroscopy and Geochemistry Symposium, Schlumberger-Doll
Research, Ridgefield, CT. Paper J.
FANG, J. H., KARR C. L. & STANLEY, D. A. 1966.
Transformation of geochemical log data into
Mineralogy using genetic algorithms. The Log
Analyst, 37, 26-31.
FUH, T. M. 1973. The principal of constituent analysis,
with special reference to the calculation of weight
percentages of minerals in metamorphic rocks.
Canadian Journal of Earth Science, 10, 657-669.
GALFORD, J. E., HERTZOG, R. C., FLAUM, C. &
GALINDO. G. 1988. Improving pulsed neutron
gamma ray spectroscopy elemental weight percent
estimates through automatic dimensioning of the
spectral fitting process. Society of Petroleum
Engineers, SPE 18151, 423-430.
HARVEY, P. K. & LOVELL, M. A. 1992 Downhole
mineralogy logs: mineral inversion methods and
the problem of compositional colinearity. In:
Hurst, A., Griffiths, C. M. & Worthington, P. F.
(eds) Geological Applications of Wireline Logs IL
Geological Society, Special Publications No. 65,
361-368.
--,
BRISTOW,J. F. & LOVELL,M. A. 1990. Mineral
transforms and downhole geochemical measurements. Scientific Drilling, 1, 163-176.
--,
LOFTS,J. C. & LOVELL,M. A. 1992 Mineralogy
logs: element to mineral transforms and compositional colinearity in sediments. 33rd Annual
Symposium of the Society of Professional Well
Log Analysts, Oklahoma city, Oklahoma.
--,
Lovell, M. A., Lofts, J. C., Pezard, P. A. &
Bristow, J. F. 1997 Petrophysical estimation from
downhole mineralogy logs, In." LOVELL,M. A. &
HARVEY,P. K. (eds) Developments in Petrophysies,
Geological Society, Special Publications No. 122,
141-157.
HARVILLE,D. G. & FREEMAN,D. L. 1988. The benefits

P. K. HARVEY ET AL.

38

and application of rapid mineral analysis provided by Fourier transform infrared spectroscopy.
Society of Petroleum Engineers, SPE 18120, 141150.
HASTINGS, A. F. 1988. Using the derived elemental
concentrations to improve the accuracy of fluid
saturations determined from well logs. Transactions Spectroscopy and Geochemistry Symposium, Schlumberger-Doll Research, Ridgefield,
CT. Paper T.
HERRON, M. M. 1986. Mineralogy from geochemical
well logging. Clays and Clay Minerals, 34, 204213.
1987a Estimating the intrinsic permeability of
clastic sediments from geochemical data. SPWLA
28th Annual Logging Symposium. paper HH.
1987b. Future applications of elemental concentrations from geophysical logging. Nuclear
Geophysics 1, 197-211.
1988. Geochemical classification of terrigenous
sands and shales from core or log data. Journal of
Sedimentary Petrology, 58, 820-829.
& GRAU, J. A. 1987. Clay and framework
mineralogy, cation exchange capacity, matrix
density and porosity from well logging in Kern
County, California. American Association of
Petroleum Geologists, 71, 567 575.
HERFtON, M. M. & HERRON, S. L. 1998. Quantitative
lithology: open and cased hole application derived
from integrated core chemistry and mineralogy
data base. In." HARVEY, P. K. & LOVELL, M. A.
(eds) Core-Log Integration. Geological Society,
Special Publication 136 (this volume).
HERRON, S. L. & CHIARAMONTE,J. M. 1993. Impact of
element-to-mineral matrix concentration errors
on geochemical log interpretation. Nuclear Geophysics, 7, 375-381.
HERTZOC, R. C. & PLASEK,R. E. 1979, Neutron excited
gamma-ray spectrometry for well logging. IEEE
Transactions on Nuclear Science, NS-26, 11581563.
, COLSON,L. SEEMAN,B. O'BRIEN, M. SCOTT,H.
MCKEON, D. WRAIGHT, P. GRAU, J. A. ELLIS, D.
SCHWEITZER, J. ~ HERRON, M. 1987a, Geochemical logging with spectrometry tools. SocieO'of
Petroleum Engineers, SPE 16792, 447-460.
, SORAN, P. D. ~ SCHWEITZER, J. S. 1987b.
Detection of Na, Mg, A1 and Si in wells with
reactions generated by 14 MeV neutrons. Nuclear
Geophysics, 1,243-248.
- - ,
COLSON,L. SEEMAN, B. O'BRIEN, M. SCOTT,H.
McKEON, D. WRAmHT, P. GRAU, J. A. ELLIS, D.
SCHWEWZER,J. & HERRON, M. 1989. Geochemical
logging with spectrometry tools: SPE Formation
Evaluation, 4, 153-162.
IMBRIE, J. & POLDERVAART,A. 1959. Mineral compositions calculated from chemical analyses of sedimentary rocks. Journal of Sedimentary Petrology,
588-595.
-

29,

JACOBSON, L. A. & WYATT, D. F. 1996. Elemental


yields and complex lithology analysis from the
pulsed spectral gamma log: The Log Analyst, 37,
50-64.
KERR, S. A., GRAU, J. A. & SCHWEITZER,J. S. 1992. A
comparison between elemental logs and core data.
Nuclear Geophysics, 6, 303-323.
LOFTS, J. C., HARVEY, P. K. & LOVELL,M. A. 1994. A
stochastic approach to mineral modelling of log
derived elemental data. Nuclear Geophysics, 8,
135 148.
& 1995a. Reservoir characterization from downhole mineralogy. Marine and
Petroleum Geology, 12, 233-246.
&- 1995b. The characterisation
of' reservoir rocks using nuclear logging tools:
evaluation of mineral transform techniques in the
laboratory and log environments. The Log Analyst, 36, 16-28.
MATTESON, A. & HERRON, M. M. 1993. Quantitative
mineral analysis by Fourier transform infrared
spectroscopy, Society of Core Analysts, Annual
Technical Conference Proceedings, Society of
Professional Well Log Analysts, Paper SCA-9308.
MIESCH, A. T. 1962. Computing mineral compositions
of sedimentary rocks from chemical analyses.
Journal of Sedimentary Petrology, 32, 217-225.
NICHOLS, G. D. 1962. A scheme for recalculating the
chemical analyses of argillaceous rocks for comparative purposes. American Mineralogist, 47, 3446.
ODOM, R. C., STREETER, R. W., HOGAN, G. P. &
T~TTLE, C. W. 1994. A new 1.625 inch diameter
pulsed neutron capture and inelastic/capture
spectral combination system provides answers in
complex reservoirs. 35th Annual Logging Symposium Transactions, Society of Professional Well
Log Analysts. Paper O.
PEARSON, M. J. 1978. Quantitative clay mineralogical
analyses from the bulk chemistry of sedimentary
rocks. Clays and Clay Minerals, 26, 423-433.
RuPP, J., HERTZOG, R. & SCHWEITZER,J. 1989. Using
log-derived elemental concentrations in stratigraphic correlation of argillaceous sediments:
Third International Symposium on Borehole
Geophysics for Minerals, Geochemical and
Groundwater Applications.
WRIGHT, T. L. & DOHERTY, P. C. 1970. A linear
programming and least-squares computer method
for solving petrologic mixing problems. Geological Socieo' of America Bulletin, 81, 1995-2008.
WYATT, D. F., JACOBSON, L. A. & HASHMY, K. 1993.
Elemental yields and complex lithology analysis
from the Pulsed Spectral Gamma log. 34th
Annual Logging Symposium Transactions, Society of Professional Well Log Analysts. Paper
UU.

Interpretation of core and log datauintegration or calibration?


M. A. L O V E L L , 1 P. K. H A R V E Y ,
WILLIAMSON

P. D. J A C K S O N , 2 T. S. B R E W E R , 1 G.
1 & C. G. W I L L I A M S 1

1 Geology Department, Leicester University, Leicester, LE1 7RH, UK


2 British Geological Survey, Key~'orth, Nottingham, NG12 5GG
Abstract: Core-log interpretation requires the reconciliation of datasets from different

measurements. Measurement process, resolution, scale and quality must be appreciated for
each dataset. Calibration of measurements involves the use of standards to enable
quantitative comparisons locally or globally; this may involve inter-dataset comparison and
the process of equalization with the modification of one dataset in preference for another.
Calibration should not be confused with integration which aims to maximize the
information in an optimal manner and may require the selective choice of data. The clear
recognition of the aims of the study at the earliest opportunity enables the best choice of
strategy from measurement acquisition through to integration. The final interpretation
should realize the original aims but must be compatible with all observations.
The integration of core and log data represents
one of the many attempts to utilize geological
data obtained by measurements at different
scales. This use of data from different sources
involves the reconciliation of different observations which may be inter-related through their
inherent property or physical basis (e.g. laboratory and in situ velocities or porosities), or
through their similar volumes of interrogation
(e.g. porosity and permeability measurements on
core plugs). Alternatively the data to be integrated may not be related in either of these ways
(e.g. core descriptions and FMS images).
Integration involves the reconciliation of such
data in a way which is defined by the overall
aims and objectives of the study. It may involve
the calibration of one dataset through some
equalization procedure, whereby one dataset is
assumed to be correct. Another scenario is
where the two or more datasets are integrated
through the selective addition of components to
enhance the overall picture of the formation
represented both downhole and in the recovered
core. These datasets may be multiple measurements of the same physical parameter by
different techniques or measurements of completely different parameters, In this latter approach each dataset is respected for both its
inherent fundamental nature and scale, both
datasets are assumed to be correct, neither
dataset is defined as superior in preference to
the other, and the interpreter attempts to extract
the maximum information from the total data
available.
Today we are faced with core and log data in
increasing quantity and sophistication. Integration of core and log data concerns the combina-

tion of two datasets, which may comprise


observations ranging from qualitative through
to quantitative, with the aim of providing the
best data on which to base our interpretation
and hence 'explain the meaning of' our observations. Integration in turn may be defined as 'to
find the total value of', and without necessarily
implying total amalgamation of all available
data in a non-selective manner. Yet there
appears to be much concern and considerable
effort directed towards deciding which of the
two different datasets represents the truth--log
or core?
In this paper we review the basic principles
involved in data acquisition and interpretation
through consideration of the measurement
process, measurement calibration, and measurement integration. Figure 1 summarizes the
problems involved. The measurement itself
may be characterized in terms of its scale,
resolution and quality. These are functions of
the parameter being measured and the measurement environment as well as the selected target.
These measurements can then be calibrated,
either relatively (locally) or absolutely (globally);
if the calibration is simply between the individual core and log measurements then the result
may be equalization of values with corresponding loss of total information. I n t e g r a t i o n
through reconciliation of the different measurements towards an optimized solution should
yield the best interpretation; but this depends
strongly on the assumptions involved and these
should be directly related to the overall aim of
the study. Indeed the best approach to the
problem of core-log integration is through
judicious choice of interpretation target, careful

LOVELL,M. A., HARVEY,P. K., JACKSON,P. D. BREWER,T. S., WILLIAMSON,G. & WILLIAMS,C. G.
1998. Interpretation of core and log data--integration or calibration? In."HARVEY,P. K. & LOVELL,
M. A. (eds) Core-Log Integration, Geological Society, London, Special Publications, 136, 39-51

39

40

M.A. LOVELL E T AL.


'

~ AIMS/OBJECTIVES ~

If

MEASUREMENT
scale - resolution ~
quality ~

~-

'

'

'

'

'

Data collected over the years


between 1965 and 1980.

size / shape / orientation

precision / a c c u r a c y / bias
J

800

I000

1200

1400

16IX)

18(l(I

20(~)

Pairs of breeding storks

(CALIBRATION1
absolute - relative
equalisation

"(INTEGRATION~/
-"--~ 1
k

Fig. 2. Example to demonstrate the need for caution in


relating measurements: a good correlation does not
imply a causal relationship (after Sies, 1988).

selective addition
reconciliatiotl

Fig. 1. Measurement, calibration, and integration. The


choice and specification of the measurement should be
dictated by the aims and objectives of the study, and
should dictate both whether there is a need for
calibration and the route to integration.
and appropriate selection of measurement techniques including necessary calibration, and
optimization using well-defined integration.
The aim of this paper is to document fundamental concepts in the context of core-log
integration in order to form the foundations
for data integration and interpretation. In
examining these aspects of integration further
we need first to consider the nature of measurement itself.

Measurement
Measurements are made to determine the value
of some parameter. Unfortunately the parameter we are often interested in (for example
porosity) cannot be measured directly but is
obtained indirectly through the subsequent
processing of raw measurements of a related
parameter(s). The relationship between the
measured parameter(s) and the derived parameter may have a well-constrained physical
basis (for example density estimates from

gamma ray attenuation measurements) or simply an empirical relationship (for example


porosity and saturation from electrical resistivity
measurements). The model used in relating the
measurement to the derived parameter may fall
within a wide range of variable sophistication.
This may impact on both the accuracy and
precision of the derived parameter since even
given accurate and precise measurements a
model that is too simple may not result in a
true representation of the actual derived physical
properties.
We do not propose to consider the role of
mathematical or statistical procedures here (see
Moss 1997 for a review of the partitioning of
petrophysical data) but do sound a warning note
on attempts to define the relationship between
different parameters or measurements. Correlation coefficients are often used to support an
interpretation of a causal relationship between
two parameters, but a high coefficient does not
necessarily imply such a causal relationship.
Indeed, all too frequently, unrelated parameters
are correlated in an attempt to derive some
solution from our data. Figure 2 demonstrates
such a classic mythical relationship between two
unrelated parameters. Even in this example the
high correlation is provided through only two of
the points and any rigorous testing of the
relationship is likely to be invalid.
Returning to the subject of measurement we
examine three primary aspects below: resolution,
scale and quality.

INTEGRATION OR CALIBRATION?

41

Fig. 3. Averaged estimates of matrix density at the boundary of a calcite dogger in the North Sea (after Lofts
1993). The shaded area represents the difference between averaged log (line) and actual core estimates (dots).

Resolution
Resolution concerns the minimum separation
between two features such that the two features
can be identified individually rather than as one
combined feature (see for example Sheriff &
Geldart 1982). In terms of log measurements this
relates to the physical separation of two features
along the length of the well (usually in a vertical
sense assuming a vertical drillhole). With respect
to core measurements this definition equally
applies, although it may be complicated by the
consideration of lateral variations or heterogeneities visible in the core sample. While the
concept of resolution is easily described, the
strict numerical definition of it varies. Theys
(1991) provides a theoretical definition of log
vertical resolution: 'The full width at half
maximum of the response of the measurement
to an infinitesimally short event'. He then
includes other non-attributable definitions from
elsewhere in the literature: qualitative: vertical
resolution is the minimum distance, x such that
the logging tool is able to resolve distinct events
separated by this distance; quantitative: vertical

resolution is the minimum bed thickness for


which the sensor measures, possibly on a limited
portion of the bed, a parameter related to the
real value of the formation.
This latter definition still falls short of ideal
since, as Theys (1991) points out, it does not
necessarily measure the true value of the
parameter concerned for the thin bed. If a
logging tool is to measure a parameter and yield
a true value for even a limited portion of the
bed, then the bed thickness must be at least as
large as the vertical resolution. This vertical
resolution will depend not only on the tool
design, but also on the formation and borehole
characteristics. Thus in terms of log measurements the vertical resolution must be as much a
local value if it is anything, but this value will
constrain the conditions under which the measured value relates to the true value of the
individual formation. In terms of interpretation,
the true value may not be important except
where the aim is to quantify that particular
parameter (e.g. porosity or saturation). The
matrix (or grain) density log from Thistle Hole
211/18-a50 which penetrates the Brent Group in

42

M. A. LOVELL E T AL.

Fig. 4. Comparison of three different imaging measurements on a single core and their inherent sampling volumes.
Given the different nature of the measurements the similarity of images is unusual, and this is probably due
primarily to the orientation of the fabric perpendicular to the length of the sample and the simplicity of the pore
structure.

the North Sea Thistle Field (Fig. 3) shows a


'calcite dogger' which is sharply bound on either
side by sandstone. The line is derived from the
continuous log measurement, whereas the points
relate to specific individual point determinations
on core samples. This is the classic 'shoulderbed' effect and the discrepancies are due to the
averaging caused by interrogation of a larger
volume by the log measurement. In effect,
neither estimate of matrix density is wrong: they
simply relate to different volumes of rock
constrained by the measurement design and thus
require slightly different interpretations. Bed
resolution is nearly always a problem even with
the most finely resolved tools and bed boundary
effects are always present.
Scale

Differences of scale are evident in terms of the


relative dimensions of the measurement itself

(e.g. frequency) and the actual physical dimensions over which the measurement is made (e.g.
size and shape). Typically, scale may be defined
quantitatively with precise descriptions of the
size and shape of the measurement (see for
example Clark 1979 or H o h n 1988). It is linked
to the measurement technique and hence the
design of the tool. Except in isotropic, homogenous media the different aspects of scale will
be important and will contribute to the measurement data value. Scale may also be linked to the
resolution of the measurement. A simple example of the effect of scale concerns the measurement of porosity on core plugs. Doveton (1994)
shows how for two porosity datasets extracted
by Baker (1957), respectively from whole core
and plugs, the mean values may be the same but
the variability of the whole core is less than that
of the plugs. There is an apparent rotation of the
relationship between the two porosity determinations in which the extremes in the smaller

INTEGRATION OR CALIBRATION?
samples are averaged out. Doveton (1994)
emphasizes that this core-based example is
equally applicable to the different volumes
sampled by core and log measurements.
As an example of the importance of scale,
consider a series of different measurements in the
laboratory on a sample of Penrith Sandstone
(Fig. 4). In addition, the role of orientation of
the measurement is also considered. Orientation
becomes important where the rock is not both
homogeneous and isotropic (i.e. for most
measurement scenarios in nature). In this
specific experiment the sandstone is an aeolian
deposit, comparable to the Rotleigendes of the
North Sea, of Permian age. It is characterized as
a clean reservoir-type sandstone with rounded
quartz grains, quartz overgrowths, and an
absence of clay phases. Generally the quartz
overgrowths reduce the porosity and increase
resistance to both electrical and fluid flow
(Harvey et al. 1995). The three images in Fig. 4
were obtained through the application of measurements at the same spacing to the upper
surface of the slab (see Jackson & Lovell 1991;
Lovell & Jackson 1991; Harvey et al. 1995).
Porosity was measured using image analysis of
the essentially 2-D surface visual texture, whilst
permeability was determined using minipermeametry measurements which involve transient
pressure impulses at point locations, again on
the upper surface. Conductivity refers to the
electrical conductivity (inverse of resistivity) and
was measured by an array of surface mounted
potential electrodes with remote electrodes
passing a uniform current through the full
volume of rock. The three comparable images
relate to very different volumes of rock: the
porosity data is restricted to the surface, whilst
the permeability investigates a hemispherical
volume of rock (in homogenous isotropic
material); in contrast the conductivity is an
average value integrated over a vertically orientated rectangular prism. Given these significant
differences in both scale and orientation the
images may not always show good correlation
although as the figure demonstrates, for the
Penrith Sandstone, with its relatively simple
structure, there is a reasonable relationship
between the different properties. This is in part
because the sands that we have studied are
relatively uniform, but perhaps more importantly the primary fabric of the samples is
perpendicular to the longest axis. In less homogeneous materials the different sample volumes
investigated would lead to greater disparity
between the images. In this way different
measurements may perceive different degrees of
homogeneity as a function of the sample volume

43

corresponding to that measurement.


Quality

Quality is defined by the precision, accuracy and


bias of the measurement (Murphy 1969). A good
quality measurement will be characterized by
high precision and accuracy, and a zero bias.
Precision refers to the closeness of agreement
between the results obtained by applying the
experimental procedure to a sample several
times under prescribed conditions. Accuracy
refers to the closeness of the measurement to
the true value. Whilst precision may be quantifiable for both the laboratory and downhole
measurements, the accuracy is more difficult to
assess as the actual (true) values are unknown,
and standards (samples for whom specific
measurements are assumed to be known as
'correct' within definable limits) are virtually
non-existent for the measurements under discussion here. Consequently the true value is an
idealized concept and the accuracy is a qualitative concept (Theys & Woodhouse 1994). In this
way we are not aware of the correct answer in an
absolute sense, and hence any bias present
remains unknown. Figure 5 (after Kimminau
1994) illustrates the concepts of accuracy and
precision, two terms which are frequently misunderstood or confused.
ODP Hole 926B on the Ceara Rise penetrates
sediments which are predominantly ooze and
chalk, with varying concentrations of nannofossils, foraminifera and clay, together with minor
components of iron oxide and sulphides. The
variation in CaCO3 in ODP Hole 926B is plotted
in Fig. 6. The continuous line is derived from
shore-based measurements in which some 70g
samples were finely crushed, sub-divided and the
major elements determined by X-ray fluorescence (XRF) spectrometry. The carbonate percentage was determined directly assuming that
all the calcium measured occurred as calcium
carbonate. The individual points are from the
shipboard measurements which were obtained
from measurement of acid-liberated CO2 and
the assumption that this was all bound up in
calcite. The precision of this method at one
standard deviation is reported as less than 1%
(Explanatory Notes, Curry et al. 1995) as is that
for the CaO XRF determinations. Thus while
the measurement method and the analytical
volumes are different for the two datasets, the
precision is similar, and a bias may have been
introduced by the respective (unknown) sampling strategies. It is impossible to say which of
the datasets is correct even though there are
significant differences between them.

44

M.A. LOVELL E T AL.

accurate but
imprecise

accurate and
precise

parameter value

)
parameter value

parameter value

inaccurate and
imprecise

inaccurate but
precise

L
parameter value

Fig. 5. Schematic illustration of the concepts of accuracy and precision (after Kimminau 1994).

Thompson & Theys (1994) note that quality


requires the definition of specified requirements
rather than expense or luxury. Thus the definition of the target or aim of the measurement
needs to be carefully detailed before we can
assess its quality. Straley et al. (1995) demonstrate the importance of defining the aims in
considering the use of N M R in partially
saturated rocks. In attempting to compare
downhole and laboratory measurements they
note that mercury porosimetry actually characterizes the pore throat size compared to the
N M R T1 which primarily responds to pore body
size. Thus the two separate measures of pore
dimensions respond to different aspects with
consequently different answers.

Calibration
Calibration is the process by which measurements are compared with known standards for
the purpose of enabling the quantitative comparison of measurements. Thus, calibration
requires samples for which supposedly 'true'
values are known in order that accuracy may be
defined (Ruth & Pohjoisrinne 1993). This
calibration procedure may involve recourse to
local standards in which case the calibrated
measurements may be termed relative. Locally
calibrated data can be easily compared and used
without knowledge of their relationship to
world-wide measurements of the same parameter. These local standards may, in turn, be

INTEGRATION
i

16

OR CALIBRATION?

' I

45

i
9

i
i
d

9 CaCO 3 (shipboard laboratory)


!

9 CaCO 3 XRF shore laboratory

17
9

Leg 154
Hole 926B
19

Ceara Rise

20

9 iI

i
J
i

22
5

% CaCO 3
Fig. 6. CaCO3 estimates by two different methods, b o t h with c o m p a r a b l e precision, for O D P Hole 926B, Ceara
Rise. There is no reason to d o u b t either o f the datasets (mbsf: metres below sea floor).

calibrated against national or international


standards providing so-called absolute values
of measurement which can be related through
the same measuring stick. Whilst there is much
to be said for global standards enabling comparison of all data for a single parameter,
calibration does not necessarily imply truth:
inherent bias in the measurement of a systematic
nature, in which the measurement effectively
measures something other than what was
intended (Eisenhart 1962), can simply yield
consistency. The importance of calibration in
the integration process depends more on the
nature of the measurements and how they are to
be used.
In terms of downhole logging calibration is
often confused with other measurement checks.
The process of calibration of logging tools
concerns the production of a specific signal in
response to known measurement values within a
formation (Theys 1991); thus the transformation

of the raw measurement to usable values is


designated separately as the tool response.
Consequently, calibration is based on a log
measurement in large volume artificial or
natural formations. Theys (1991) also considers
that the checking of a logging instrument in welldefined conditions both before and after a
logging run should really be labelled verification,
whilst the matching of surface electronics to
downhole signals could be better described as
surface system alignment. This is because neither
considers the actual calibration of the data,
simply the overall working of the tool within
predefined calibration constraints.
Core measurement calibration, meanwhile, is
well-documented and generally involves the use
of standards of a similar scale to the samples
under test. These small volume local standards
can be readily controlled and related to national
or international standards. Skopec (1992), however, notes that whilst laboratory determinations

46

M.A. LOVELL E T AL.

Fig. 7. Reconciliation of different measurements of electrical resistivity formation factor (FF) at different scales
(the formation factor is the rock resistivity normalized with respect to the resistivity of the saturating fluid). The
horizontal bar is from a standard industry minicore measurement, the continuous plots are derived from
resistivity imaging. Averaging the high resolution log provides a lower resolution log (dots) which approaches the
value of the minicore.

are the standards by which in situ log measurements are compared (e.g. nuclear spectroscopy
logs), each method must be examined carefully
to determine experimental limitations, accuracy
and precision in testing, as well as potential
mineral alteration processes that can occur when
a rock is sampled.
Given a satisfactory understanding of our
measurement base we can proceed to analyse the
data. Often we are concerned with combining
two datasets of the same parameter with the aim
of producing one, more complete dataset. In this
way core data may supplement sections of log
data, or indeed duplicate it. The normal
procedure here is to assume that one of the
datasets is correct and to adjust the other to
create a best fit. This equalization can create
better coverage of the total borehole section but
will inevitably involve manipulation of at least
one dataset and the loss of inherent absolute
values. This is a standard approach to so-called
core-log integration. Unfortunately it does not
consider discrepancies between the two measurement sets created by different measurement

techniques, strategies, or acquisition procedures.


R a t h e r it involves the process of relative
calibration between small and large volume
measurements, typically between core and log.
Where datasets do not agree it is important to
ask whether the issue is one of data quality or
the quality of interpretation (Owens 1994;
Harvey et al. 1998).
As an example of this process, consider Fig. 7.
Here two datasets are shown at different scales,
but both relate to measurements on core. The
continuous electrical Formation Factor log is
derived from the electrical resistivity image as
presented in Fig. 4. These data are then
compared with the solid bar which is the
electrical Formation Factor determined in a
traditional manner on a minicore or plug. This
was originally done during attempts to calibrate
the novel imaging system against industry
standards (Jackson et al. 1994). The match
between the two datasets is improved visually
by smoothing the image data further (dots). The
smoothing was carried out with a simple moving
average with a window width corresponding to

INTEGRATION OR CALIBRATION?
the length of the minicore; in this way the
smoothed log (dots) is effectively a stepwise
integration over the image. As with the CaCO3
estimates shown in Fig. 6, neither dataset is
incorrect: both have supporting calibration data
referenced to standard materials, but each
dataset provides the interpreter with a different
perspective of the sample, effectively a different
representation of the truth. The image data
provides fine detail relating to the structure
whilst the minicore provides an average value
(though not a simple arithmetic or statistical
average). Through correct averaging of the high
resolution log there is a remarkable match with
the minicore measurement at a similar resolution.
A related problem occurs when we are trying
to predict petrophysical properties from unrelated logs. Often we derive statistical models or
empirical relationships which have no physical
foundation but which satisfactorily estimate the
parameter of interest at each log depth. Effectively we calibrate our model or algorithm to
give answers which are compatible with laboratory or borehole experiments of an unrelated
nature.
These data demonstrate that the problems of
data integration are present at all scales, and
whilst this contribution refers explicitly to log
and core data the principles remain true for
integrating these data with smaller scale (thin
section, SEM) and larger scale (VSP, seismic
reflection) data.

Integration
Integration involves the reconciliation of datasets with or without the equalization involved in
calibration procedures. Often this will include
the selective addition of data. Different datasets
may relate to the same measurement, the same
scale, or either or both of these attributes may be
different. The overall aim of integration is to
maximize the i n f o r m a t i o n available in an
optimal manner. Towards this aim, the objective
is not simply to compare data but to constrain
and characterize some geological process or
effect.
The effects of sample size and tool resolution
in core-log integration is easily demonstrated by
a simple Monte Carlo experiment which could
go some way to explaining the variation seen in
the CaCO3 estimates shown in Fig. 6. In this
particular experiment (Fig. 8) a 30 m section of
oceanic sediment, with basaltic lava flows, was
simulated to evaluate the suitability of core
measurements as guides to the accuracy of
geochemical log measurements. The section,

47

Fig. 8. Comparison of core and log data simulated for


ODP Sites 792 and 793 (Ocean Drilling Program Leg
126) demonstrating the integration of measurements at
different resolutions.

based on ODP Sites 792 and 793 in the region


of the Izu Bonin Arc (Lovell et al. 1992), was
simulated in three steps:
(1) generation of a lithological sequence with
abundances and thicknesses of simulated
units corresponding to the appropriate
distribution obtained from core logging;
(2) generation of the rock chemistry at 1 cm
intervals throughout the section, preserving
the average and variance/covariance relationships of the chemistry for each lithology;
(3) sampling of the simulated chemical sequence.
In Figure 8 this simulated dataset is sampled for
alumina (a) as it would respond to the geochemical logging tool (GLT) by averaging over a 60
cm window and reporting measurements every
15 cm to provide the continuous log curve, and
(b) as a set of core plug results obtained by
randomly sampling a small number of the total
3000 simulated compositions to provide the

48

M.A. LOVELL E T AL.

individual point measurements. Both datasets


present different perspectives of the same chemical sequence; neither is wrong and consequently neither should be rejected in favour of
the other. Thus the integration of different
measurements of the s a m e parameter can
provide both overall and detailed geological
information.
Hornby et al. (1992) used downhole electrical
images, reflected Stoneley waves and core
observations to deduce estimates of fracture
apertures. They thus used different observations
to comment on fracture extent and connectivity
as well as borehole enlargement and rugosity.
Furthermore, they point to the use of information obtained at different scales as being the key
to further work aimed at fracture quantification.
Core-log

interpretation

Figure 9 shows the stratigraphy of ODP Hole


896A, which was drilled in the Equatorial East
Pacific as part of ODP Leg 148 (Alt et al. 1993).
With the drilling of Ocean Drilling Program
Hole 896A, two deep basement holes (i.e. Holes
504B and 896A) now penetrate oceanic crust
formed at the Costa Rica Rift. Hole 504B, the
deepest basement hole in oceanic crust so far
drilled (2100m), is located approximately
200 km to the south of the Costa Rica Rift, in
5.9 Ma old crust. Hole 896A is located
approximately 1 km to the south of Hole 504B
in crust ,-~2.8x 104 yr older than at Hole 504B.
No attempt was made to recover the sedimentary cover in Hole 896A and the position of
sediment/basement interface was based upon
rubble being felt by the drill bit at 179 mbsf
(metres below sea floor) and the hole was cored
from 195.1 mbsf to 469 mbsf (Alt et al. 1993).
Within this drilled section, core recovery averaged 26.9%. Pillow lavas (57%) and massive
flows (38%) dominate the cored material, with
breccias (5%) and two small dikes accounting
for the remainder of the material. With the
exception of pillow rims, the majority of the
rocks are slightly altered (< 10%) and variably
veined (Alt et al. 1993). Pervasive background
reducing alteration coupled with saponite and
minor pyrite replacement of olivine has led to
the grey colour of the core. Oxidative alteration
is manifested by dark grey to yellow and red
alteration halos which commonly occur around
smectite veins (Alt et al. 1993). In the pillow
lavas and massive flows, veins are usually
< 1 mm in thickness and commonly infilled by
dark and light green saponite and aragonite.
Other vein minerals include analcite, fibrous
zeolite and pyrite. All of the previous data were

Fig. 9. Stratigraphy derived from core recovery based


on core barrel descriptions, compared with stratigraphy and based on downhole electrical FMS images and
core observations.

based on an overall poor core recovery (26.9%,


Alt et al. 1993), which is also very variable
within individual sections of the borehole (Fig.
9). Shipboard scientists produced the lithology
shown in Fig. 9 based solely on visual observations of recovered core.
In contrast, shore-based scientists (Brewer et
al. 1995) have produced a comparable stratigraphy based on the variations in texture of the
downhole Formation Micro Scanner (FMS)
Images together with sonic, resistivity, and
gamma ray logs. The FMS tool produces images
of the borehole wall dependent on variations in
the measured electrical resistance and these
images can be analysed texturally to develop a
log-based stratigraphy with reference to the
core. As Brewer et al. (1995) demonstrate, there
are substantial differences between the stratigra-

INTEGRATION OR CALIBRATION?
resistivity
(ohm-m)

smoothed pixel
value

mean pixel
value

49
permeability
(roD)

porosity

(%)

i j

0
-4
o , .t-_.

........ 9
JO

.... 0"2 4
O. . . . . . .
Ik

O" --~-e ....

e-..t

--|ii.
11

"lP.

e..
i~ 9i:e

e r e ....

:,

.!

ie

k
3

165

175

185

160

170

180

2000

4000

i 2

k
20

i
40

Fig. 10. Electrical resistivity data and optical data for an aeolian sandstone. There is a remarkable correspondence
between the two datasets for this clean sandstone, yet the raw optical data provide a higher resolution dataset
than the resistivity data, enabling inference of the fine-scale resistivity structure of the sample.
phy derived solely from core and that derived by
integrating core and log information.
Here the problem may initially be seen as one
of constraining the downhole data through
selected core observations, knowing that the
recovered core is present in the drilled section,
and utilizing the downhole data to extend the
interpretation to the full depth of the hole.
However, the core data are inherently biased,
due possibly to preferential sampling of some
lithologies, incomplete recovery, and the defined
criteria and procedures used to identify and
extrapolate recovered material over the total
depth drilled. Thus whilst the core does indeed
represent the truth, its allocation to a particular
lithology, distribution with depth and continuity
may be questioned. In contrast the FMS images
are relatively new and lack precise calibration in
terms of textural detail and lithological responses. They are usually continuous and are
based on electrical, not visual, properties which
may or may not be equable, and whose equality
may vary within the hole. These images proba~ contain bias in addition to that within the
core. Thus, rather than accept one dataset in its
entirety as the truth and reject the other, it
would be better to use the ground-truth of the
recovered and described core as calibration
points for the interpretation of the downhole
images. This would ideally include the measure-

ment of electrical textures on recovered core for


comparison with downhole images, thus constraining in a quantitative manner the interpretation of the downhole data.
A different example of the integration of data
from different sources is shown using the
electrical conductivity image from Fig. 4 converted to a resistivity image (by taking the
inverse of each plotted value). In Fig. 10 this
resistivity image is converted into a microresistivity log by averaging across each row of values.
Similarly, the porosity and permeability images
are displayed as averaged micrologs. These logs
again show the variable nature of the formation
but could easily be incorporated into a petrophysical analysis routine for producing improved estimates of both fluid volume and flow
parameters. Here the photographic image has
been converted to a pixel log, again by row
averaging; this micro-log is then smoothed to a
similar resolution to the electrical micro-log.
There is an inverse correspondence between the
two which suggests the use of photographic
images for comparison with electrical images in
clean sandstones. Where clay minerals may
contribute to conduction processes, the rock is
contaminated with mud, there are clay-filled
fractures, or the pore fluid is resistive (e.g. fresh
water), the correspondence between photographic and electrical images may not be as

50

M. A. LOVELL ET AL.

Hole 896A from FMS images. Scientific Drilling,


5, 87-92.
CLARK, I. 1979. Practical Geostatistics. Elsevier,
London.
CURRY, W. B., SHACKLETON,N. J., RICHTER, C. & the
Shipboard Scientific Party. 1995. Proceedings.
ODP Initial Reports., 154: College Station, TX
Summary
(Ocean Drilling Program).
(1) Interpretation of core and log data should
DOVETON, J. H. 1994. Geological Log Analysis Using
Computer Methods. American Association of
involve consideration of the measurement proPetroleum Geologists, Computer Applications in
cess, calibration, and integration. Calibration
Geology, No.2.
and integration may not necessarily be included
EJSENHART, C. 1963. Realistic evaluation of the
in the first interpretation.
precision and accuracy of instrument calibration
(2) The measurement itself is defined in terms
systems. Journal of Research of the National
of its scale, resolution and quality. These are
Bureau of Standards--C, Engineering and Instrufunctions of the parameter being measured and
mentation, 67C, 21-47. (Paper 67C2-128).
the measurement environment as well as the
HARVEY,P. K., BREWER,T. S., LOVELL,M. A. & KERR,
selected target. It is imperative that these
S. A. 1998. The estimation of modal mineralogy: a
problem of accuracy in core-log calibration. This
attributes are considered in any data integration
volume.
exercise.
LOVELL, M. A., JACKSON, P. D., ASHU, P. A.,
(3) Measurements can be calibrated, either - WILLIAMSON, G., SMITH, A. S., BALL, J. K. &
relatively (locally) or absolutely (globally). CauFLINT, R. F. 1995. Electrical resistivity core
tion is essential since often this falls to interimaging III: characterisation of an aeolian sanddataset comparison, and involves equalization
stone. Scientific Drilling, 5, 165 176.
which through the modification of one dataset in
HOHN, M. E. 1988. Geostatistics and petroleum geology.
preference for the other may negate additional
Van Nostrand-Reinholt, New York.
benefits which i n t e g r a t i o n may otherwise
HORNBY, B. E., LUTH1, S. M. & PLUMB, R. A. 1992.
Comparison of fracture apertures computed from
achieve.
electrical borehole scans and reflected Stoneley
(4) Differences of scale in measurement sets
waves: an integrated interpretation. The Log
may highlight geological features through variaAnalyst, 33, 50-66.
tions in the degree and nature of formation
JACKSON, P. D. & LOVELL, M. A. 1991. The correheterogeneity.
spondence of electrical current and fluid flow in
(5) Integration of data should maximize, in
rocks--the impact of electrical resistivity core
an optimal manner, the information available.
imaging. Transactions 14th European Sympo(6) The best c o r e - l o g i n t e r p r e t a t i o n is
sium, Society of Professional Well Log Analysts,
through judicious choice of objectives, approLondon, UK. Paper J.
9HARVEY, P. K., BALL, J., WILLIAMS,
priate selection of measurement process, includC.I FLINT, R. F., ASHU, P. A. & MELDRUM,P. I.
ing necessary calibration, and optimization
1994. Advances in resistivity core imaging.
using carefully-defined integration. This should
Transactions 35th Symposium, Society of Profesrealize an interpretation which maximizes the
sional Well Log Analysts. Paper GG.
use of available data whilst remaining compaKIMMINAU, S. 1994. Traceability--making decisions
tible with all observations.
with uncertain data. The Log Analyst, 35, 67-70.
LOFTS, J. C. 1993. Integrated geochemical and geophysical studies of sedimentary reservoir rocks. PhD
We thank the Natural Environment Research Council
thesis, University of Leicester.
for support through research grant GST/02/684,
LOVELL, M. A. & JACKSON, P. D. 1991. Electrical flow
together with Z & S Group for provision of software
in rocks: the application of high resolution
for the processing and interpretation of FMS data.
electrical resistivity core measurements, paper
WW in 32nd Annual logging Symposium Transactions: Society of Professional Well Log AnaReferences
lysts, Midland, Texas.
, PEZARD, P. A. & HARVEY, P. K. 1992.
ALT, J. C., KINOSHITA, H., STOKKING, L. B. & the
Chemical stratigraphy of boreholes in the IzuShipboard Scientific Party. 1993. Proceedings.
Bonin Arc from insitu nuclear measurements.
ODP Initial Reports, 148. College Station TX
Proceedings of the Ocean Drilling Program,
(Ocean Drilling Program).
Scientific Results, 126, 593-601.
BAKER, P. E. 1957. Density logging with gamma rays.
Moss, B. 1997. the partitioning of petrophysical data:
Petroleum Transactions of the American Institute
a review. In: LOVELL,M. A. & HARVEY,P. K. (eds)
of Metallurgical Engineers, 210, 289 294.
Developments in Petrophysics, Geological Society
BREWER, T. S., LOVELL,M., HARVEY,P. & WILLIAMSON,
Special Publications, No. 122, pp 18l 252.
G. 1995. Stratigraphy of the ocean crust in ODP

reliable, nor as predictable. Thus discrepancies


between optical and resistivity images or logs
may also yield information about the nature of
the pore space.

INTEGRATION OR CALIBRATION?
MURPHY, R. B. 1969. On the meaning of precision and
accuracy. In. Ku, H. H. (ed.) Precision Measurement and Calibration. Statistical Concepts and
Procedures. United States Department of Commerce National Bureau of Standards Special
Publication 300, 1, 357-360.
OWENS~J. 1994. Fit-for-purpose data during field life.
The Log Analyst, 35, 58-60.
RUTH, D. & POnJOISRINNE, T. 1993. The precision of
grain volume porosimeters. The Log Analyst, 34,
29-36.
SHERIFE, R. E. & GELDARX, L. P. 1982. Exploration
Seismology volume 1. history, theory and data
acquisition. Cambridge University Press, Cambridge.
SIES, H. 1988. A new parameter for sex education.

51

Scientific Correspondence, Nature, 332, p.495.


SKOPEC, R. A. 1992. Recent advances in rock
characterisation. The Log Analyst, 33, 270-285.
STRALEY, C., MORRISS, C. E., KENYON, W. E.
HOWARD, J. J. 1995. N M R in partially saturated
rocks: laboratory insights on free fluid index and
comparison with borehole logs. The Log Analyst,
36, 40-56.
THEYS, P. P. 1991. Log data acquisition and quality
control. Editions Technip, Paris.
THEYS, P. & WOODHOUSE, R. 1994. Society of professional well log analysts topical conference on
quality, appendix: metrological definitions. The
Log Analyst, 35, p. 71.
THOMPSON, B. & THEYS, P. 1994. The importance of
quality. The Log Analyst, 35, 13-14.

Estimation of measurement uncertainty for in situ borehole


determinations using a geochemical logging tool
M. H. R A M S E Y , P. J. W A T K I N S

& M. S. S A M S 1

T. H. Huxley School of Environment, Earth Science and Engineering, Imperial College o f


Science Technology and Medicine, London S W7 2BP, UK
1 Present address." Petronas Research and Scientific Services S D N BDH, Hulu Kelang,
54200 Selangor, Malaysia.
Abstract: Methods for the estimation of measurement uncertainty are discussed with

particular reference to concentration measurements made by a geochenmical logging tool


(GLT; Mark of Schlumberger) in a borehole penetrating a cyclothem sequence at
Northumberland, UK. Two components of uncertainty have been quantified for 6 elements
determined by the GLT over a wide range of concentrations. The random component was
estimated from duplicated determinations within this borehole over a 120 m depth interval.
These uncertainty values ranged from 2.7% for Si to 71% for S, expressed at the 95%
confidence limit. The systematic component of the uncertainty was estimated by
determinations made on corresponding core samples by inductively coupled plasma atomic
emission spectrometry (ICP-AES) over a 40 m depth interval. The ultimate basis for this
estimation of bias was the certified reference materials which were analysed during the ICPAES determinations. The bias measured by this method was typically in the range + 5% to
- 1 4 % for 5 out of the 6 elements determined.
This method assumes that the samples analysed by both techniques are physically
comparable. By depth averaging the ICP-AES determinations it was possible to reduce
errors due to differences in sample size. However, a possible source of bias that was
recognised is that samples were dried before ICP-AES determinations, whereas this was not
the case for in situ GLT measurements. Such variability in the size of the systematic
component of the uncertainty prevents the effective correction of this term as is
recommended by the International Standards Organisation. The large values of measurement uncertainty found for some elements (e.g. S) will exert limitations on the geochemical
interpretations made from the GLT measurements, in terms of 'fitness-for-purpose' criteria.

Uncertainty of measurements made in boreholes


by an 'in situ' geochemical logging tool (GLT)
can have p r o f o u n d effects on the realistic
interpretation of the geochemical variation
across a stratigraphic sequence. Although the
importance of the uncertainty is becoming
apparent, methods for the estimation of such
uncertainty are lacking. Broadly similar studies
on the estimation of bias and precision of GLT
measurements have been reported (Wendlandt &
Bhuyan 1990; Grau et al. 1990), but these did
not attempt any rigorous mathematical estimation of the total uncertainty of measurement.
For analytical measurements in isolation, the
realistic estimation of uncertainty has already
become an important issue (ISO 1993a; Eurocheni 1995; A M C 1995). In contaminated land
investigations it has recently been shown that it
is field sampling, rather than the chemical
analysis, that can contribute the largest source
of measurement error and will therefore limit the
measurement uncertainty (Ramsey et al. 1995a).

Methods have been developed for the quantification of the errors arising from the sampling of
one site, by either a single or multiple samplers.
For a single sampler, the methods have been
applied both to the use of a single sampling
protocol (Ramsey 1993) or comparisons between several protocols (Ramsey et al. 1995b).
For the case of multiple samplers, different
m e t h o d s have been devised d e p e n d e n t on
whether all samplers were applying the same
protocol (Ramsey et al. 1995a) or different
protocols (Argyraki et al. 1995). Applications
of these methods were made for the estimation
of heavy metals on contaminated land, but the
methodologies are equally applicable, in principle, to the measurement of elemental variation
within a borehole using a GLT.
This paper considers how estimates of measurement uncertainty can be derived, particularly for the case of a single sampler using a
single protocol, utilizing a previously published
case study (Sams et al. 1995). The objectives of

RAMSEY,M. H., WATKINS,P. J. & SAMS,M. S. 1998. Estimation of measurement uncertainty for in situ
borehole determinations using a geochemical logging tool In: HARVEY,P. K. 8~; LOVELL,M. A. (eds)
Core-Log Integration, Geological Society, London, Special Publications, 136, 53-63

53

M. H. RAMSEY ET AL.

54
the work are therefore:

(l) to make the best estimate of uncertainty


on the measurements from the case study,
as an example of the general approach;
(2) to identify additional methodologies that
could be applied to give improved estimates of uncertainty;
(3) to consider briefly how to identify acceptable levels of uncertainty for particular
objectives of interpretation.

Definition and terminology of measurement


uncertainty
The formal definition of uncertainty has been
given as: 'A parameter associated with the result
of a measurement, that characterizes the dispersion of the values that could reasonably be
attributed to the measurand (ISO 1993a). A less
formal but more understandable description of
measurement uncertainty is that it is 'an interval
a r o u n d the result of a measurement that
contains the true value with high probability'
(Thompson 1995). The standard uncertainty 'u'
can be considered equivalent to one standard
deviation which is often used to describe a
normally distributed random error. The expanded uncertainty 'U" is equal to the product
of the standard uncertainty and a coverage
factor 'k', which typically has a value of 2 or 3.
This is analogous to the use of multiples of
standard deviation for the quantification of
precision. In the formal definition, systematic
errors are not included in estimates of uncertainty, but only any residual random errors left
after correction of the systematic errors. However, in this application, the distinction between
systematic and random errors becomes blurred.
The systematic error of one sampler becomes a
random error when assessed as part of a multisampler comparison and systematic errors thereby become incorporated into estimates of
uncertainty.

Estimating uncertainty in chemical analysis


The methods recently developed for estimating
uncertainty in chemical analysis in the laboratory are of two types. They both need to be
evaluated as options for the application to in situ
borehole sampling and analysis. In the 'bottom
up' approach the random error from each
individual component of a method is quantified
separately as a standard deviation (s). The
overall uncertainty is then estimated by summing the individual errors by their variances (s 2)

(Eurochem 1995). The alternative 'top down'


approach uses inter-laboratory trials to estimate
the total uncertainty of a measurement. In this
method, many selected laboratories ( n > 8 )
would all analyse the same sample by the same
analytical method (AMC 1995). The scatter of
all reported measurements is then used as an
overall estimate of uncertainty.
The 'bottom up' approach has the limitation
that it requires all of the sources of uncertainty
to be identified. It is relatively easy to consider
the obvious sources of error which are explicit
parts of a laboratory method (e.g. weighing,
volumetric additions). However, the most important source of uncertainty may not be explicit
in the method (e.g. lab. temperature), and is
therefore easily overlooked, especially by inexperienced practitioners. Furthermore, it can
be a long and expensive procedure to quantify
all the component errors, if the method is to be
applied rigorously.
The benefits of the 'top down' approach can
be appreciated from the differences that are
often evident between laboratories in interorganizational trials. These differences are often
larger than can be accounted for by the
individual estimates of uncertainty within each
laboratory (AMC 1995). This is because the
' b o t t o m up' a p p r o a c h used by individual
laboratories tends to give over-optimistic estimates of the uncertainty. The limitation of the
'top down' approach is that it depends on the
selection of the laboratories that contribute. If
the laboratories all use a similar source of
calibration, they may all be equally biased and
therefore give an under-estimate of the uncertainty. Alternatively, one laboratory may have
gross errors, atypical of the application of the
method as a whole, and this will cause an overestimate of the uncertainty.

Estimating uncertainty for in situ borehole


measurements
There are two primary limitations in estimating
uncertainty of measurements made in situ in
boreholes, using the method described for purely
laboratory based analytical systems. One problem is that these methods ignore the uncertainty arising from field sampling. It is often
quoted that an analysis can never be of better
quality than the sample upon which it is made.
What has been lacking, however, is the means of
estimating the size of the uncertainty which is
introduced by field sampling. A second limitation is in the quantification of systematic errors,
either from sampling or from chemical analysis.

ESTIMATION OF MEASUREMENT UNCERTAINTY


It is proposed to adapt the methods devised
originally for estimating analytical uncertainty
to the estimation of uncertainty from in situ
measurements by addressing these two limitations.
With in situ measurements it is useful to
consider field sampling and chemical analysis as
just two parts of the same 'measurement'
process, and to quantify their combined contribution to the uncertainty. Such 'total measurement' uncertainty has therefore four
potential components. These are the sampling
and analytical contributions to random error
(i.e. sampling and analytical precision) and any
uncorrected systematic errors (i.e. sampling and
analytical bias).
Taking the 'bottom up' approach to estimating the total measurement uncertainty we can
review the methods available for the estimation
of these four components. Analytical precision
can be measured by the use of analytical
duplicates (Thompson & Howarth 1976) or in
combination with sampling precision using a
balanced design of sampling and analytical
duplicates (Garrett 1969; Ramsey et al. 1992).
This is possible for the GLT although, ideally,
duplicate sampling would require the use of a
second borehole close to the first and assuming
lateral homogeneity. Analytical bias is usually
estimated by the analysis of certified reference
materials (Ramsey et al. 1987), but this approach would be problematic for the GLT. In
addition, there are no methods in general use for
the estimation of sampling bias. For absolute
bias this may require the introduction of
reference sampling targets, analogous to reference materials for the estimation of analytical
bias (Thompson & Ramsey 1995). Sampling bias
has already been estimated however, between
the concentration estimates made by the application of different sampling protocols at one site
(Ramsey et al. 1995b). This approach does not,
however, give bias against an 'accepted reference
value', as defined by ISO (ISO, 1993b).
Taking the 'top down' approach, it should be
possible to use measurements from inter-organizational sampling trials, such as sampling
proficiency tests (Argyraki et al. 1995) and
collaborative trials (Ramsey et al. 1995a) to
estimate uncertainty. The potential advantages
and practical feasibility of such an approach to
in situ borehole determinations will be considered below.

Experimental details of the GLT case study


The details of the case study using the GLT at
the Imperial College borehole at Whitchester

55

farm in Northumberland have been given elsewhere (Sams et al. 1995). In brief, GLT
measurements were made on a 40m length of
borehole drilled through a single Namurian
cyclothem, with varied lithologies (Fig. 1).
The GLT has been described in detail by
Hertzog et al. (1987) and uses three different
measurement techniques:
(i) The natural radioactivity of K, Th and U
is used for their determination with a
N a t u r a l G a m m a Ray Spectrometer
(NGS; Mark of Schlumberger). The count
rates obtained are directly proportional to
the mass per cent of the element, providing a borehole correction factor is applied.
(ii) A neutron source of 252Cf, emitting
neutrons at about 2.3MeV is used to
activate A1, which is determined using an
AACT; Mark of Schlumberger. Results
obtained are proportional to the weight
per cent of A1 after an environmental
correction is applied.
(iii) After neutron capture from a burst of 14
MeV neutrons, Si, Fe, Ca, Ti, S and Gd
are determined using a tau-gated thermal
neutron capture spectrometer with a
Gamma Spectrometer (GST; Mark of
Schlumberger). This procedure only provides relative concentrations of these
elements, and they have to be converted
to absolute values by imposing a closure
relationship on the results obtained.
An oxide closure relationship is imposed on
the results, in order to derive element concentrations from the raw GLT data It is assumed that
each element occurs as a single oxide or
carbonate in the formation and that the sum of
the oxide and carbonate fractions is unity. This
assumption is known to be in error, but it is
considered that the errors involved will be small
(< 5%) for most lithologies. The equation to be
solved is (Hertzog et al. 1987):
F(~,Xi Yi/Si) + XK WK -[- XA1WA1 = 1.0

where:
F is a calibration factor to be determined at each
depth point;
Yi is the fraction of the measured prompt
gamma rays attributed to element i;
Si is the tool sensitivity for element i;
Xi is the ratio of the mass of the associated oxide
or carbonate to the mass of element i.
The mass fractions of potassium and aluminium
(WK and WA1) must be first corrected for

56

M.H. RAMSEY ET AL.

Fig. 1. Variation with depth of concentrations of Si and Fe determined by the GLT (solid lines) and by ICP-AES
(solid circles). Plot (a) represents original data and plot (b) represents ICP-AES data depth-averaged (smoothed)
using a 1.4m square window. Sulphur was not determined by ICP-AES but demonstrates that the coal bands
(shown at 150 and 154.3m in the stratigraphic log) can be detected with by the GLT even with an estimated
random error of 71%.

porosity to give a dry mass per cent. This is


normally determined from the density log by
assuming a quartz matrix.
Duplicate field measurements were taken at
the site in Northumberland by lowering the
GLT twice through a section 110-230m in
depth. On one occasion all raw measurements
were processed through the oxide closure to give
elemental concentration, but on the second
occasion the raw results were initially unprocessed. In order to achieve comparability, the
original dataset was divided by the corresponding raw counts for each element at a particular
depth. This procedure gave a value for the factor
FXi/Si, and both F and Si should be constant at
a constant depth. If the raw counts for an
element taken from the duplicate dataset are
now multiplied by the corresponding value of
FXi/Si, an estimate of the absolute value for that
element at a particular depth can be made. This
procedure is not quite the same as comparing
data obtained from two runs processed independently.

Sampling and analysis of core samples


Core sampling was aimed at investigating the
vertical variations in bulk rock geochemistry

and mineralogy over a single sedimentary cycle


using the same depth interval as the GLT (140180 m). Exact correspondence of the volume of
rock analysed by both GLT and ICP-AES is
clearly impossible (Fig. 2). The rationale behind
the sub-sampling of the core used for analysis is
that it should characterize the geochemistry of
the core at small intervals (e.g. 25cm), so that
subsequent mathematical smoothing of results
can be employed to approximate the sampling
volume analysed by the GLT (110,000 cm3). The
inevitable sampling bias introduced by lithologies of widths less than 25cm (e.g. iron rich
sideritic nodules) can then be estimated and
recognized as a systematic difference between the
two techniques. Discrete samples of about 30 g
were selected with an average vertical spacing of
25 + 5 cm between samples, the rock chips being
taken to represent the full lateral heterogeneity
of the core sample. A total of 147 analytical
specimens were collected over the entire depth
interval and powdered in an agate Tema Mill to
less than 75 #m.
The details of the analytical procedure have
been given elsewhere (Sams et al. 1995). In brief,
0.25g of dry rock powder was totally decomposed by fusing with LiBO2 and dissolving the
fused bead in dilute HNO3, with six elements

ESTIMATION OF MEASUREMENT UNCERTAINTY

57

Sampling design

Estimation of random component

25:1:5 cm

The analytical precision of GLT measurements


was characterized as a function of concentration
using the method using duplicated analyses
devised by Thompson & Howarth (1976). The
mathematical model of precision was:
sc = So + Pc

ca. 60cm

//
/

/ /

20cm diameter borehole


for GLT analysis

Corresponding core, with cut sections


used for ICP-AES analysis

Fig. 2. Sampling designs for the measurements by GLT


and by ICP-AES.

(1)

where so=standard deviation at a particular


concentration c;
So= standard deviation at zero concentration;
c = concentration;
0 = slope factor, related to the high level
precision and so, So and c all have the
same units of concentration.
The random component of the measurement
uncertainty, using a coverage factor of two, for
zero concentration is given by:
Uo = 2s0

being determined by inductively coupled plasma-atomic emission spectrometry (ICP-AES).


This instrument was calibrated using 10 international reference materials, (BR, GA, GH,
NIM-N, SY-2, UB-N, NIM-G, DTS-1, AN-G
and MICA-MG). The certified values for the
reference materials are given in Govindaraju
(1994) and thus, these act as the traceability of
the determinations to 'accepted reference values', as required by ISO (ISO 1993b) for the
estimation of bias. The bias between measured
and certified values of the reference materials for
the 6 elements studied was generally less than 1
%. Values for the precision of analyses of these
six elements in silicate rocks by ICP-AES are
reported as < 0.5% in Ramsey et al. (1995c).

and for a particular concentration c by:


Ur = 2sc
Thus, multiplying equation (1) by the coverage
factor of two we get:
Ur = Uo -+-20c

(2)

The upper limit of the uncertainty on an


estimation of concentration c, from equation
(2) is given by:
c+ Ur=c+ 20c+ Uo

c+ Ur : C(1 + 2 0 ) + Uo

(3)

A useful way of expressing uncertainty is as a


relative percentage, given by:

Methodology for estimation of uncertainty in


GLT measurements
The two components of measurement uncertainty considered initially are the analytical
precision (for random error) and the analytical
bias (for systematic error). The role of the
random component is explicit in the ISO guide
(ISO 1993a), but the systematic component for a
particular technique such as GLT analysis has to
be estimated independently. Whether this latter
component is used to correct the concentration
estimates, as suggested by ISO, or added into the
uncertainty estimate will be discussed below.

Ur% = 200Sc/C

At high levels of concentration where Ur>>Uo,


this can be expressed as:
Ur% = 2008

(4)

This value of Ur% is approached asymptotically


at high concentration.
Substituting for 0 from equation (4) in
equation (3) we get:

c+ Ur=c(l + Ur%/lOO)+ Uo

(5)

58

M.H. RAMSEY ET AL.

Table 1. Estimates of uncertainty in concentration measurements made using the GLT, in the specified ranges of

concentration.
Element

Range
(mass-%)

Si
AI
Fe
Ca
K
Ti
S

044
0-12.5
0-13
0-38
0.5 2.8
0.1-2.7
0.6-5

Random
Error, Ur %

Uo
(mass-%)

2.7
7.6
12.1
5.0
n.d.
n.d.
71

1.80
0.62
0.18
1.28
0.48
n.d.
0.56

Translational
Bias, (mass-%)
- 1.53"
1.12"
0.34*
0.38
0.20*
0.003
n.d.

Rotational
Bias, %
3.5
- 10.1"
- 54.6*
4.3*
- 13.6
11.1
n.d.

n.d. = not determined.


* Bias value significantly different from zero (p = 0.05).
Table 2. Estimation of detection limit and precision of

GLT determinations (Sams et al. 1995).


Element

Detection
limit (mass-%)

Si
A1
Fe
Ca
K
S

2.70
0.92
0.26
1.92
0.71
0.56

High-level
Precision,% (Is)
1.35
3.83
6.05
2.51
n.d.
35.45

n.d. = not determined


These two components of random uncertainty
(Table 1) were estimated for 6 elements (Si, A],
Fe, Ca, K and S) by using a dataset consisting of
600 duplicate borehole measurements taken over
a section between l l 0 m and 230m in depth.
This method also provides standard errors on
these uncertainty estimates, which can be used to
check whether the estimates are statistically
greater than zero. When the So value is not
significantly greater than zero, it is still possible
to calculate a maximum value for Uo, that is
equivalent to a maximum method detection limit
( M M D L ) (Ramsey et al. 1995c). The precision
and detection limits of the G L T determinations
are listed in Table 2. Data used for calculations
made in this paper is represented graphically in
Fig. 3, and a summary of the duplicate data used
to determine the precision of G L T measurements is given in Fig. 4 in Sams et al. (1995). The
individual data sets obtained from ICP-AES and
G L T measurements are available, but too
numerous to include in full here. They are
available from the Geological Society if required.
The variations of uncertainty due to analytical
precision for Si and Fe are plotted against
concentration (c) in Fig. 3A, to exemplify the

general relationship for all elements studied. The


same results are transformed into the form of
relative uncertainty (Ur%) in Fig. 3B. For Si,
there is a significant increase in uncertainty with
concentration, but when converted into relative
uncertainty at the high level, this was the best of
any element at 2.7%. The uncertainty at zero
concentration for Si was 1.8 mass %, which is
well below most of the samples from this
borehole. Hertzog et al. (1987) do not give a
standard deviation at zero mass% but give
precisions of 25.6% and 4.9% at 25 mass% Si
and 47 mass% Si, respectively.
For Fe, considering these random errors in
isolation gives a relative uncertainty of 12.1% at
high concentration levels, and an uncertainty of
0.16 mass% at zero concentration. The bias
detected for this element, discussed below, will
however cause this latter estimate to be multiplied by a factor of 2.2, giving 0.35 mass%.
Hertzog et al. (1987) quote a precision of 30% at
1 mass% Fe and one of 16% at 5 mass% Fe,
which appear broadly comparable.
For S, the high level relative uncertainty was
found to be over 70% at all concentrations, the
highest for any element determined. This may be
due to the proximity of sulphur concentrations
in the samples to the detection limit of the GLT.
It is interesting to examine what limitations such
a degree of uncertainty places on the geochemical interpretation of the concentration estimates
(see below).
Estimation of systematic component

A qualitative estimate of the analytical bias of


the G L T measurements was made using X-ray
fluorescence analysis on a limited selection of
core material by Hertzog et al. (1987). The
quantitative estimation of bias needed for
uncertainty estimation has been reported by

ESTIMATION OF MEASUREMENT UNCERTAINTY

(a)

59

(b)
1.4-

1.2"

o.

1.0-

.o
3.0

0.8-

0.6-

2.0
0.4-

0.2-

.o'o

o.

0.0

10

(A)

20

30

'

40

I
1

I
2

'

GLT Si (mass %)

I
3

I
4

'

I
5

I
6

'

I
7

i
8

l
9

GLT Fe (mass %)

(a)

(b)

100

80

60

60

"

,m~

"0

zo "

"

10

(U)

20

'

"
30

0
40

GLT Si (mass %)

GLT Fe (mass %)

Fig. 3. Random component of uncertainty as a function of concentration for Si and Fe. (A) expanded uncertainty
Ur; and (B) relative expanded uncertainty Ur%. For Si the Ur% values tend toward a low asymptotic value
(2.7%) but for Fe this value is much higher (12.1%).

comparing the logging results with values


obtained from core samples using ICP-AES
(Sams et al. 1995). The method employed used
simple linear regression to determine the rotational and translational components of bias.
Translational bias occurs when the intercept of
the regression, line is statistically significantly
different from zero. Rotational bias occurs when
the slope of the regression line is statistically
different from unity. The test for significance
used in this study was the student t-test with a
probability of 0.05. Potential limitations in this
general technique (Thompson 1982) have been
shown not to be significant in this case (Sams et
al. 1995).

Before comparing results obtained using the


two techniques the question of sample size must
be addressed (Fig. 2). Samples collected for
laboratory analysis represent a depth of approximately 1 cm. Overall, the GLT responds to an
approximately spherical volume of rock with a
radius of about 30cm, although each measurement technique used, i.e. NGS, AACT and GST,
will interact with different volumes of rock.
Assuming there is lateral homogeneity, the GLT
response is some weighted depth average of
vertical variations. In addition, it is usual to
apply a vertical smoothing process to the raw
data before the oxide closure is performed. A
running average of 5 data points, or just over

M.H. RAMSEY ET AL.

60
Silicon

Iron
25

4O

20.

E~

~r

15'

10"

I'--

(5
5.

9 .~':.':d'.'-..' ":'
, ' , '

i , , , ,

10

~ , , , ,

20

L,,,

,i,

30

40

,,,

50

ICP mass percent

10

15

20

25

ICP mass percent

Fig. 4. Graphical representation of systematic uncertainty (i.e. bias) between measurements made by the GLT and
ICP-AES. The solid line represents the line of equality for zero bias, and Fe measurments in particular show a
distinct deviation from this line.

60cm in depth, was used for the current data.


Therefore, raw measurements obtained from the
GLT and ICP-AES represent significantly different depth intervals of rock. In order to make
a comparison which is not biased by this
difference, the ICP-AES data must be depth
averaged in a comparable manner to the GLT
data, although this assumes that each laboratory
measurement is representative of a 25 cm depth
interval, i.e. the sampling interval. Smoothing
was applied to the ICP-AES data using windows
of varying lengths. The window chosen was one
that gave maximum correlation between the
GLT and ICP-AES data, and a square window
1.4m wide (equivalent to 6 analyses) was found
to be the best. Thus the smoothed ICP-AES
results, at any particular depth, are based on 6
separate determinations around that depth. The
unsmoothed and smoothed ICP-AES measurements for Si and Fe are plotted for comparison
alongside the GLT measurements against depth
in Fig. 1.
In order to compensate for the difference in
sample interval the GLT determinations were
interpolated using a cubic spline, and depths
equivalent to the ICP-AES sampling were taken.
It was also necessary to apply a static depth shift
of 0.5m to the GLT data to account for a
difference between drilling depth and logging
depth. Measurements taken by the GLT in the
coal layers (150 and 154.3m in depth) were
removed from the comparison due to exceptional errors caused by the effect of the
anomalous density of the coal propagated
through the closure calculation (Sams et al.
1995).

Estimates of the components of the bias in the


element concentrations determined by the GLT
compared with ICP-AES are given in Table 1.
Significant rotational bias was found for all of
the elements except Si, and significant translational bias for all except Ca and Ti. Silicon
shows a statistically significant translational bias
of - 1 . 5 3 mass%, but no detectable rotational
bias. This linear model for the bias is perhaps
oversimplified as the majority of data points
with less than 25 mass% Si fall below the
regression line (Fig. 4). In terms of the uncertainty there would seem to be a case for
adding 1.53 mass% to all the GLT estimates of
Si concentration. This is rather simplistic,
partially because of the over-simplified model
used for the bias but also because this bias may
well be different for different lithologies in
different boreholes. It is perhaps more prudent
therefore to combine an estimate of the possible
bias into the overall estimate of uncertainty. For
Fe, measurement by the GLT shows significant
bias in both the rotational ( - 5 4 . 6 % ) and
translational (0.34 mass%) components, which
again needs combining with the random component of the uncertainty for evaluation.

Combined estimates o f measurement


uncertainty
The total expanded uncertainty (U) can be
estimated from the combination of the random
component Ur% described in equation (5), and
the systematic components resolved into rotational bias (BR) and translational bias (BT). The

ESTIMATION OF MEASUREMENT UNCERTAINTY


upper limit of the uncertainty for an estimated
concentration c is then given by:
c + U = c (1 + Ur%/100)(1 + BR/100) + Uo + BT (6)
and the lower limit is given by:
c - U = c (1 - Ur~

+ BR/100 )

Uo + BT (7)

For the case of Si, ( U r % = 2 . 7 % , Uo = 1.8


mass%, B R = 0 and BT = --1.53 mass%) equation (6) gives:
c+U =1.027c+1.8+5.53=1.027c+3.33
mass%
and equation (7) gives:
c - U = 0.973c- 1.8 + 1.53 = 0.973c-0.27 mass%
For an estimated Si concentration of 5 mass%
the uncertainty interval would therefore be 1.535
to 5.405 mass% Si, and for a high concentration
of 40 mass% it would be 35.59 to 41.35 mass%
Si.
In the case of Fe, again assuming that the bias
is not realistically correctable, the upper limit of
the combined uncertainty from equation (6),
where (Ur%=12.1%,
Uo=0.18 mass%,
BR = --54.6 and BT = 0.34 mass%) is given by:
c + U = c (5 + 12.1%/100)

(1 - 54.6/100) + 0.16 + 0.34


c + U = cx 1.121 x0.454 + 0.16 + 0.34
c + U = 0.509c + 0.50
and the lower limit of the uncertainty from
equation (7) is given by:
c - U = c (1 - 12.1%/100)

(1 - 54.6/100)- 0.16 + 0.34


c+ U=cx0.879x0.454 -0.16+0.34
c + U=0.399c+0.18
For an Fe concentration of 1 mass% the
uncertainty would range from 0.579 to 1.009
mass% Fe, and for 5 mass% Fe from 2.175 to
3.045 mass% Fe. The range of these uncertainties are large, but could be allowed for in
geochemical interpretation of the concentration
values.
Although not relevant to the calculation of
uncertainty, it is interesting to speculate on the
cause of this systematic error. Two possible
causes have been suggested for the large negative
rotational bias for Fe determinations using the
G L T (Sams et al. 5995). Firstly, it is possible

61

that the sensitivity factor for iron, Sve, is too


high. Secondly, it is possible that occasionally
the sampling procedure used for ICP-AES
determinations dramatically over-estimates the
Fe content of some samples. As mentioned
earlier, the ICP-AES data is assumed to be
representative of a 25 cm depth interval. However, in the case of Fe this may not be true. Iron
occurs partially as sideritic nodules in the
mudstones. If present in a core sample these
nodules could produce an analysis of up to 30
mass% Fe, which is probably a gross overestimate of the average value for that 25cm
depth interval. If a nodule is not sampled then
the average value for the Fe concentration of the
25cm interval will be somewhat under-estimated. Overall therefore the heterogeneity of
the iron would not be expected to produce the
large bias found in this case.

Discussion
The calculations used to estimate the overall
uncertainty of Si and Fe are equally applicable
for the other elements measured by the GLT,
but there are a number of limitations which may
mean the values obtained are under-estimates.
There are other causes of 'random' error in
the measurement system that have not been
investigated. In contaminated land measurements it has been shown that multiple applications of the same measurement protocol, on
different occasions, by different operators,
causes appreciable increases in the uncertainties
(Ramsey et al. 1995b). When different measurement protocols are selected to measure the same
quantity and applied by different operators then
the uncertainty increases even further (Argyraki
et al. 1995). This suggests that a more rigorous
estimate of the uncertainty for the GLT measurements would require the use of similar interorganization trials with both multiple users of
one technique, the comparison of a number of
probes in the same borehole, and the use of
closely spaced duplicate boreholes to investigate
lateral sampling errors, and small scale geochemical variability.
A further limitation of the method reported
here is that it assumes that the ICP-AES analysis
provides an 'accepted reference value' as required by ISO for the detection of bias (ISO
1993b). Although the ICP-AES was calibrated
using ten certified reference materials this does
not ensure zero bias. This partially because the
recommended values for these reference materials also have specified uncertainties, but more
especially due to the problems of matching the
sample volume. A better solution to the second

62

M. H. RAMSEY ET AL.

aspect would be to establish one borehole as a


'reference sampling target', against which new
analytical probes and new operators could
measure their systematic error (Thompson &
Ramsey, 1995). The 'accepted reference value' of
the elemental concentrations at specified depths
in this borehole would have to be established by
inter-organization sampling trials similar to
those described above, with a wide variety of
analytical techniques.
One extra complication of the measurement
system used in the G L T is that the bias in any
one element will be propagated through to the
other elements by the oxide closure. Such
multivariate effects in measurement uncertainty
have not been investigated and may cause subtle
and unforeseen effects on geochemical interpretation.

Acceptable limits for uncertainty


Once realistic values become available for
measurement uncertainty then there will be a
need to derive acceptable limits for the uncertainty. This is a separate aspect that relates
the G L T measurements to the concept of'fitness
for purpose'. To take an extreme case, for the
vertical correlation of statigrapaphy between
two boreholes it could be argued that systematic
errors in the measurements are irrelevant. The
depth of the coal horizons in the Northumbrian
boreholes, for example, could be correlated
between boreholes even if the iron concentration
is biased by --54.6%. It is only the random
component of the uncertainty that could mask
the position of such a feature, and as such
therefore could be specified in the fitness-forpurpose specification. The coal bands (shown at
150 and 154.3m in Fig. 1) can be detected from
the sulphur concentrations measured by the
GLT, even though the random component of
the uncertainty was estimated as 71%. This
shows that the limits for uncertainty need to be
related to the geochemical variance (Ramsey et
al., 1992). The requirements for uncertainty are
very different if the elemental concentration
estimates were to be used to infer the mineralogy
of a sample. In this case a bias o f - - 5 0 % on one
element could clearly have a major impact on
the minerals inferred to be present and their
calculated proportion in the rock.
Because the estimation of uncertainty is more
expensive than the simple estimation of elemental concentration, cost-benefit analysis will need
to be applied to identify the benefits that the
uncertainty information can bring. Furthermore, if there is a choice of several methods to
estimate uncertainty, there will be some cases

where a crude estimate will be sufficient and


others where more reliable but more expensive
methods will be justified. Further clarification of
these ideas awaits more case studies of the
application of techniques like the G L T that
include estimates of uncertainty, and an evaluation of its effects on the geochemical interpretation of the information.

Conclusions
1. Methods are available to estimate the
uncertainty of measurements made with
the GLT.
2. The random component of the uncertainty
can be estimated at a basic level, from a
replicate set of measurements from the
same borehole.
3. The random component of the uncertainty
can be estimated from the chemical analysis of core material by a technique such as
ICP-AES, which can be linked directly to
'accepted reference values' of concentration
as required by ISO. The uncertainties in the
analyses by the method used for comparison (in this case ICP-AES) need to be
assessed, but for the ICP-AES method they
were much smaller than differences between the measurement techniques. There
are also problems with this approach in
allowing for the effects of different volumes
of rock sampled.
4. Progressively more realistic estimates of
uncertainty would require the use of
different operators, on different occasions,
even with different probes in inter-organization trials.
5. There is a financial need to derive acceptable levels of uncertainty for particular
applications, but further case studies reporting uncertainties must be examined
before this will be feasible.
We would like to thank D. Filmer who performed the
ICP-AES analyses. We would also like to thank Agip,
Amoco, BP, Elf, Mobil, NERC, Schlumberger and
Statoil who funded the Imperial College borehole test
site. The third author also acknowledges assistance
given by Petronas Research and Scientific Services to
enable this work to be completed.

References
ANALYTICAL METHODS COMMITTEE 1995. Uncertainty
of measurements: implications of its use in
analytical science. Analyst, 120, 2303-2308.
ARGYRAK1,A., RAMSEY,M. H. & THOMPSON,M. 1995.
Proficiency testing in Sampling: Pilot study on
Contaminated Land. Analyst, 120, 2799-2804.

ESTIMATION OF MEASUREMENT UNCERTAINTY


EUROCHEM 1995. Quantifying uncertainty in analytical
measurement. Eurochem Secretariat, Teddington,
UK.
GARRETT R. G. 1969. The determination of sampling
and analytical errors in exploration geochemistry.
Economic Geology, 64, 568 569.
GOVINDARAJU, K. 1994. Geostandards Newsletter,
Special Issue, July 1994.
GRAU, J. A., SCHWEITZER,J. S. & HERTZOG,R. C. 1990.
Statistical uncertainties of elemental concentrations extracted from neutron-induced gamma-ray
measurements. IEEE Transactions on Nuclear
Science, 7, 2175-2178.
HERTZOG, R., COLSON, L., SEEMAN, B., O'BRIEN, M.,
SCOTT, H., MCKEON, D., WRA1GHT, P., GRAU, J.,
ELLIS, D., SCHWEITZER, J. & HERRON, M. 1987.
Geochemical logging with spectrometry tools.
Society of Petroleum Engineers, Annual Technical Conference 1987, Dallas. SPE 16792.
Iso 1993a. Guide to the expression of uncertainty in
measurement. ISO, Geneva.
Iso 1993b. 3534-1:1993 (EIF) Statistics, Vocabulary
and Symbols - Part 1. Probability and General
Statistical Terms, ISO, Geneva.
RAMSEY, M. H. 1993. Sampling and analytical quality
control (SAX) for improved error estimation in
the measurement of heavy metals in the environment, using robust analysis of variance. Applied
Geochemistry, 2, 149-153.
, THOMPSON, M. & BANERJEE, E. K. 1987. A
realistic assessment of analytical data quality from
inductively coupled plasma atomic emission
spectrometry. Analytical Proceedings, 24, 260265.

63

& HALE, M. 1992. Objective evaluation


of precision requirement for geochemical analysis
using robust analysis of variance. Journal of
Geochemical Exploration, 44, 23 36.
, ARGYRAKI,A. & THOMPSON,M. 1995a. On the
collaborative trial in sampling. Analyst, 120,
2309-2317.
&
1995b. Estimation of
sampling bias between different sampling protocols on contaminated land. Analyst, 120, 13531356.
, POTTS, P. J., WEBB, P. C., WATKINS, P.,
WATSON, J. S. & COLES, B. J. 1995c. An objective
assessment of analytical method precision: comparison of ICP-AES and XRF for the analysis of
silicate rocks. Chemical Geology, 124, 1 19.
SAMS, M. S., WATKINS,P. J. & RAMSEY, M. H. (1995).
Validation of a geochemical logging tool for in
situ major element analysis in boreholes using
inductively coupled plasma atomic emission
spectrometry. Analyst, 120, 1407-1413.
THOMPSON, M. 1982. Regression methods in the
comparison of accuracy. Analyst, 107, 1169-1180.
1995. Uncertainty in an uncertain world.
Analyst, 120, 117N-118N.
- & HOWARTH, R. J. 1976. Duplicate analysis in
geochemical practice. Analyst, 101, 690-698.
& RAMSEV,M. H. (1995). Quality concepts and
practices applied to s a m p l i n g - a n exploratory
study. Analyst, 120, 261-270.
WENDLANDr, R. F. & BHUYAN,K. 1990. Estimation of
mineralogy and lithology from geochemical log
measurements. American Association of Petroleum
Geology Bulletin, 74, 87 856.
-

Methods for simulating natural gamma ray and density wireline logs
from measurements on outcrop exposures and samples: examples from
the Upper Jurassic, England
Z. M. A H M A D I 1 & A. L. C O E 2

t Department of Geological Sciences, University of Durham, South Road, Durham, DH1


3LE, UK (Present address: Enterprise Oil plc, Grand Buildings, Trafalgar Square, London
WC2 4ES, UK)
2Department of Earth Sciences, The Open University, Walton Hall, Milton Keynes,
Buckinghamshire, MK7 6AA, UK
Abstract: Methods for simulating natural gamma ray and density wireline logs from
measurements on outcrop exposures and rock samples have been implemented. The signals
have comparable amplitudes and resolution to wireline log signals, although the absolute
values do not match precisely. The field gamma ray logs were measured on the outcrops at
intervals of 30-45 cm using hand-held gamma ray spectrometers. The field density logs were
produced by measuring the volume and grain density of selected rock samples, followed by
interpolation and filtering of the data. Both techniques are illustrated for the Upper Jurassic
of the Wessex Basin, Southern England, with field data from the exposures on the Dorset
coast and wireline log data from 11 boreholes between 0.5 km and 170 km away. The Upper
Jurassic comprises a range of rock types, giving a wide range of values on which to test the
techniques: wireline gamma ray and density values of these strata cover the ranges 15-140
API and 1.8-2.9 g cm 3, respectively. Thus these techniques should be widely applicable for
the purpose of correlating outcrops with borehole data.

Wireline logs provide an intermediate link


between the small-scale, high-resolution sedimentological and stratigraphical features visible
at outcrop and the large-scale data available
from seismic sections. Simple techniques have
been developed for producing natural gamma
ray and density logs from measurements on
outcrop exposures and rock samples to correlate
with wireline logs from boreholes. Emphasis has
been placed on producing field logs which are at
the same resolution and of similar character to
typical downhole wireline logs, rather than
reproducing the absolute values which might
be expected downhole. This approach thus
concentrates on reproducing patterns of cyclicity, together with general decreasing and
increasing trends, which in turn can be interpreted in terms of cyclostratigraphy; for example, transgression and regression, sequence
stratigraphical and Milankovitch cycles.
The techniques are illustrated for the Upper
Jurassic (Oxfordian, Kimmeridgian and Portlandian stages) of the Wessex Basin, Southern
England. This interval is represented by a wide
range of sedimentary rock types ranging from
deep-marine siliciclastics to shallow and nonmarine siliciclastics and carbonates. The validity

of the methodology has therefore been tested on


most sedimentary rock types, giving a full
spectrum of typical data.
The overall aim of this paper is to reproduce
at decimetre resolution gamma ray and density
wireline log trends from measurements on outcrops and rock samples, thus improving stratigraphic correlation between outcrops and
boreholes. The data and interpretation presented in this paper are part of a wider study
on the sequence stratigraphical interpretation of
wireline log signatures from over 100 boreholes
from the Upper Jurassic of the Wessex Basin.
Where available, biostratigraphical data have
been used to provide a framework for wireline
correlations between boreholes.

Geological setting of exposures and


boreholes
The Wessex Basin is a Mesozoic extensional
basin which is divided into a series of halfgraben, or graben-like sub-basins (Fig. 1). The
Upper Jurassic exposures and boreholes used in
this paper are from four of these sub-basins. The
exposures where field measurements and rock
samples were taken are on the Dorset Coast

AHMADI,Z. M. & CoE, A. L. 1998. Methods for simulating natural gamma ray and density wireline
logs from measurements on outcrop exposures and samples: examples from the Upper Jurassic, England
In: HARVEY,P. K. & LOVELL,M. A. (eds) Core-LogIntegration,Geological Society, London,
Soecial Publications, 136, 65-80

65

66

Z, M. AHMADI & A, L. COL

VALEOF PEWSEY

SUB-BASIN
~'~'-"""-.,~

Ashdown 1

ST DORSET

"l~
IV

7o

CENTRAL CHANNEL
SUB-BASIN

s*o

10 km

'

'

9L0

!-8o

'~.,~ c '7
i ~'-"

,'
.'~

~"~-'~-~

/\ ( / - - Q

~/:-~~

~"~-

""
9 '
encomoe
l
9

& Kimmeridge Bay


to
Chapmans Pool
I

80
I

~-7~-

B~YJ
t . . -. > . ". . ~ '

-70

'~
L,-~

90
I

9>

~,~
(
~v

' 3-11 / 89

98/11 4

.~~-

lb

DORSET

W E
. .Y. M
. .f ). I .I T
. .I -. - I

70

SY0'0sz

Hamcliff Blackhead
~ Redcliff/
ran Point
Y ~ I S" 2raRingstead Bay

7"---.~ (

WESSEX BASIN

WEALD SUB-BASIN

-s,,)

90

9
"~
Detention 1 ~ /..._-f

"~-~"--,,.,.~

~0

9 BletchingleyI ~

Collendean
Farm 1

80-

SWANAGE

St.Alban's
Head
70Sy00LSZ

10
I

Fig. 1. Maps showing the main structural features of the Wessex Basin (after Whittaker 1985) and the location of
boreholes and Upper Jurassic exposures. (a) Boreholes in the Weald Sub-basin and position of Fig. l(b). (b)
Details of the location of boreholes and outcrops in the Dorset area.
between Weymouth and Swanage, which is at
the edge of the Central Channel Sub-basin.
Boreholes Encombe 1 (SY 9446 7785), 98/11-1
(SZ 1187 8386) and 98/11-3 (SZ 1329 8459) are
on the up-thrown northern edge of the Central
Channel Sub-basin, and borehole 98/11-4 (SZ
1187 8084) is in the Central Channel Sub-basin.
Winterbourne Kingston 1 (SY 8470 9796) borehole is in the Dorset Sub-basin and Marchwood
1 (SU 3991 1118) is in the Mere Sub-basin. All of
the other boreholes mentioned are from the
Weald Sub-basin, a moderately deep graben in
the eastern part of the Wessex Basin (Fig. 1).
The Oxfordian Stage is represented in the
lower part by mudstones of the Upper Oxford
Clay Formation and in the upper part by the

Corallian Group, a complex succession of


shallow-marine siliciclastics and carbonates
which show marked lateral and vertical variation. The Kimmeridgian Stage (sensu anglico) of
Dorset is represented by interbedded organicrich and organic-poor mudstones, with a few
thin beds of fine-grained sandstone near the base
and the top. These mudstones and sandstones
comprise the Kimmeridge Clay Formation, and
are generally of wide lateral extent. They can be
correlated across the Wessex Basin and into the
Wash area and Humberside using outcrops,
wireline logs and borehole cores (Gallois & Cox
1974; Cox & Gallois 1981; Penn et al. 1986;
Melnyk et al. 1994, 1995). The Portlandian
Stage is represented by marine silty and clay-rich

CORRELATION OF WIRELINE LOGS WITH OUTCROP


dolomites deposited in a moderate water-depth
(Portland Sand Formation) overlain by a
shallow and non-marine carbonate ramp system
which comprises the Portland Stone Formation
and Lulworth Beds (Coe 1996).

Field and laboratory methods for reproducing wireline log trends


g a m m a ray logging

Two types of wireline gamma ray sondes exist,


the conventional one which records the total
natural radiation, and the spectral gamma ray
sonde which separately records gamma rays
emitted from 4~ 232Th or 238U and their decay
products (Serra 1984). The main uses of gamma
ray logs are:
(i) as an indicator of lithology;
(ii) to correlate the wireline signatures between
boreholes;
(iii) to correlate separate wireline runs within
one borehole.
The fact that the gamma ray tool is run in all
boreholes makes it the key wireline tool for any
attempt to make correlations between outcrop
and the subsurface.
Field gamma ray logs can be constructed
using hand-held portable gamma ray spectrometers, which were originally developed and
used for uranium ore exploration (Adams &
Gasperini 1970). Following the lead of Ettensohn et al. (1979), total gamma ray logs have
subsequently been used for surface to subsurface
correlation of sedimentary strata (Chamberlain
1984; Cowan & Myers 1988; Slatt et al. 1992;
Van Buchem et al. 1992). More recently,
portable gamma ray spectrometers have also
been used to study the distribution of K, U and
Th in sedimentary rocks, and as a tool for
stratigraphical correlation between rock exposures (Dypvik & Eriksen 1983; Myers & Bristow
1989; Davies & Elliott 1996; Hesselbo 1996;
Parkinson 1996; Bessa & Hesselbo 1997).
Previous spectral gamma ray studies on the
Upper Kimmeridge Clay Formation in Dorset
have been completed by Myers (1987) and
Myers & Wignall (1987), who took spectral
gamma ray measurements using an Exploranium GR256 on the wave-cut platforms. They
utilized these data for a sedimentological and
stratigraphical interpretation of organic-rich
mudstones. Talwar et al. (1992) completed a
study of the gamma ray spectrometry of the
Corallian Beds (Oxfordian) at Bran Point,

67

Dorset using a Scintrex Scintillation Counter


(SCC) spectrometer. There are two problems
with the work of Talwar et al. (1992). Firstly,
they appear to have used an exceedingly short
sampling time of only 3-6 s, which would result
in significant errors; a count time of greater than
60 s for sedimentary rocks is usual (Lovborg &
Mose 1987; Parkinson 1996). Secondly, their
correlation with two boreholes from the North
Dorset and Wiltshire area show little similarity
because the lower two-thirds of the Oxfordian
strata examined in the boreholes is older than
any of the rocks which they illustrate from Bran
Point, and thirdly they did not take into account
any of the unconformities in the Oxfordian
succession (Coe 1992, 1995).
Gamma ray logging field procedure. Two portable gamma ray spectrometers have been used
and compared in the work reported here: a
geoMetrics GR310 (manufactured 1980) and an
Exploranium GR320 (manufactured 1996). Both
tools use thallium-doped sodium iodide detector
crystals. The Exploranium GR320 was calibrated in Toronto by Exploranium Ltd (Canada) and the geoMetrics GR310 was calibrated
on the calibration pads at the British Geological
Survey, Keyworth. A value for background
radiation was measured 2 km offshore from
Swanage, Dorset for each tool at the same time
(Fig. 1). Detailed explanation of the calibration
of portable gamma ray spectrometers is provided by Lovborg (1984) and Lovborg & Mose
(1987).
The geoMetrics GR310 provides separate
measurements of either total gamma ray count,
or diagnostic gamma radiation for either K, or
U, or Th, and only allows count times of 1, 10,
100 and 1000s to be chosen. Source, detector
and recorder are all housed in one unit
9 cm x 18 cm x 28 cm, weighing 3.4 kg.
There are several advantages of the Exploranium GR320 for this type of stratigraphical
study. Total counts and counts in the K, U and
Th fields are all recorded during one counting
period the length of which can be set by the user
anywhere in the range 1 to 9999 s. The instrument carries out automatic gain stabilization,
unlike the geoMetrics GR310 which has to be
calibrated by the user. Automatic gain stabilization is important because portable spectrometers
are prone to tool drift due to changes in
temperature and humidity. The fact that the
Exploranium GR320 stabilizes itself at regular
intervals saves time and reduces the risk of
errors due to incorrect manual stabilization. The
inbuilt computer chip allows the spectra to be
displayed and the amount of K, U and Th to be

68

Z.M. AHMADI & A. L. COE

a)

Mass of effective
sample = 49 kg
assuming a density
of 2.8 g/cm3
)ept~ =

b)

'

Diameter = 84 c)C
m.

~"

Cliff face
Borehole
Fig, 2. Sampled volume for a portable gamma ray
spectrometer compared to a wireline gamma ray
sonde. (a) Dimensions of the sampled volume for a
portable spectrometer (modified from Lovborg et al.
1971). (b) Typical orientation and position of the
sampled volume for the portable gamma ray spectrometer as used in this study. (c) Spherical sampled
volume for a wireline gamma ray sonde in a borehole.
This depends on the speed at which the tool is drawn
up the hole, as well as the density of the rocks, but
typically has a radius of 30 cm (Rider 1991). The
sampled volume tends to a more ellipsoidal shape
when the tool is drawn up the borehole faster.

calculated directly. The only disadvantage to


this instrument compared with the geoMetrics
GR310 is that it is bulkier and heavier. This
spectrometer comprises two parts, a detector
(11.4x39.4cm) and a recording/processing unit
(24 x 10 x 25 cm) which have a combined weight
of 8.4 kg.
The effective sampling region of portable
gamma ray spectrometers is shown in Fig. 2a.
The dimensions in the figure are only approximate because the density value used by Lovbor~
et al. (1971) to calculate them was 2.8gcm-which is about 0.3-0.5 gcm 3 higher than most of

the sedimentary rocks in this study. Rocks with


lower density and the same amount of natural
radiation would result in a slightly larger
effective sampling region. The most precise
absolute values for a particular bed of greater
than about 14cm in thickness are obtained by
placing the tool on top of a flat bedding surface
of at least 1 m diameter. Similar measurements
made on beds with a thickness of less than 14 cm
will obviously include some component of the
underlying bed or beds. The aim of this study,
however, was to compare the general trends of
field gamma ray logs with wireline data. Therefore the detector was placed perpendicular to
bedding (Fig. 2b) so that measurements made on
all beds less than about 84cm thick will have
been influenced by adjacent beds, as is the case
in wireline logging (Fig. 2c). Where possible, all
readings were taken on a relatively flat section of
the cliff face, avoiding irregularities such as
overhangs and corners to ensure that the same
volume of rock contributed to each reading.
Readings were only taken where the tool could
be used at least 1 m above the base of the cliff,
thus avoiding errors due to gamma ray contribution from rocks on the beach.
Count times of 100s for the geoMetrics
GR310 and 200 s for the Exploranium GR320
were used in this study. This resulted in
theoretical tool precision errors of < 2.5% and
< 1.5%, respectively, for the total count reading.
Parkinson (1996) showed that, in practice,
departures of measurement geometry from a
true plane far outweigh instrument precision as a
source of experimental error. In this study, it
was found that readings taken along 20m of a
bed vary by up to 7% for both the geoMetrics
GR310 and the Exploranium GR320. This is
probably due to slight lithological variations as
well as differences in the volume of the effective
sample size due to small undulations in the cliff
face. A longer count time was used for the
Exploranium GR320 because spectral data were
also recorded. Radioactive decay of natural
elements is a r a n d o m process, so shorter
sampling periods give a greater statistical error.
Specifically, the percentage statistical error
varies with the number of counts collected: the
higher the count, the more accurate the measurement. For typical needs, 1000 counts (3%
error) is accurate enough (geoMetrics GR310,

Fig. 3. Composite field gamma ray log for the Upper Jurassic succession exposed between Furzy Cliff and St.
Alban's Head, Dorset, measured using the geoMetrics GR310 portable gamma ray spectrometer, plotted against
the wireline gamma ray log from borehole 98/11-4 (SZ 1187 8084). Gaps in the composite field log are due to lack
of exposure or non-accessibility of the section with a portable gamma ray tool. See Fig. 1 for location of borehole
and outcrop sections.

CORRELATION OF WIREEINE LOGS WITH OUTCROP

69

70

Z . M . A H M A D I & A. L. COE

CORRELATION OF WIRELINE LOGS WITH OUTCROP


Operating Manual; Lovborg 1984). A longer
count time had to be used for the Exploranium
GR320 because it takes longer to record
sufficient gamma ray counts in the K, Th and
U windows than it does for the total gamma ray
measurement. The spectral data recorded with
the Exploranium GR320 are not discussed
further in this paper.
The measurement procedure used for both
spectrometers was to take a reading once in
every bed of less than 50 cm thick and every 3050cm in beds greater than 50cm thick. The
geoMetrics GR310 was used to record total
gamma ray readings throughout the best Upper
Jurassic exposures in Dorset, resulting in 1124
total gamma ray readings with an average
sample interval of 45cm over 503m (Fig. 3).
Part of the Kimmeridge Clay Formation was
selected to compare the results from the two
spectrometers. Full spectral gamma ray data
were thus recorded with the Exploranium
GR320 at 824 sample points over 251 m of the
Kimmeridge Clay Formation (average sample
interval 30cm; Fig. 4). The average sampling
distance of 30-45 cm is within the limits of the
effective sampled volume for each spectrometer
(84cm; Fig. 2; Lovborg et al. 1971) and each
consecutive reading overlaps the previous reading resulting in a moving average, thus making it
comparable with the wireline gamma ray tool as
it is pulled slowly up the borehole.

Density logging
Wireline density logs record the bulk density of
rocks, by emitting gamma rays into the formation and recording the number of back-scattered
gamma rays at a fixed distance from the source.
The bulk density is a function of the density of
the matrix and the density of the fluids in the
pore space. Therefore any attempt to construct a
field density log with the same character and
resolution as the wireline density log has to take
into account the density of the matrix and the
density of the pore fluid. The vertical resolution
for older single-detector tools is 40 cm and for
more modern two-detector tools is 25 cm (Serra
1984).

Density logging laboratory procedure. Fresh rock


samples of smaller than 3.1 cmx3.1 cmx3.7cm

71

representative of the majority of the beds in the


Upper Oxford Clay F o r m a t i o n , Corallian
Group and Upper Kimmeridge Clay Formation
were collected for density analysis. These
amounted to 116 samples over 90m of the
Upper Oxford Clay Formation and Corallian
Group (Fig. 5) and 260 samples over 280m of
the Kimmeridge Clay Formation (Fig. 6). All
the samples were dried in an oven at a
temperature of less than 35~ prior to the
measurements being taken. A Ruska Universal
Porometer (model 1051-801) was then used to
measure the volume and grain density of the
samples. The density of each sample was then
calculated using a single typical fluid density
value of 1.06gcm -3 for pore fluids present
within Upper Jurassic rocks of the Wessex
Basin; this actual value was recorded at Palmers
Wood 3 borehole (TQ 3655 5255) in the Weald
Basin (pers. comm. P. Rowe).
The raw density curves on Figs 5 and 6 show
the density values calculated from the actual
samples measured. To obtain the box curve, two
further procedures were applied. Firstly, beds
from which no samples had been obtained were
assigned an average density typical for that
particular lithology, calculated from the measured samples collected nearby. Secondly, the
same density value was assigned to the whole
thickness of the bed. It was noted that the
sandstones of the Nothe Grit Formation and the
Bencliff Grit Member (top of the Redcliff
Formation) had lower density values than those
seen on the wireline density logs. This was
interpreted to be due to higher porosities of
these rocks at outcrop than in the subsurface,
resulting from dissolution of calcite cement. The
density values of these beds were therefore
corrected as follows: their average porosity in
the subsurface was estimated by plotting typical
density and sonic values on porosity evaluation
log interpretation charts (Atlas Wireline Services
1985; Schlumberger 1994). The additional,
secondary dissolution porosity that was calculated to be present in the rock samples was
multiplied by the difference between the density
of calcite and the pore fluid and added to the
calculated total density for the samples. The box
curve was then filtered using the Atlas Wireline
Services field acquisition filter (Atlas Wireline
Services 1992). This is an eleven point, Gaus-

Fig. 4. Comparison of the field gamma ray logs measured using the Exploranium GR320 and the geoMetrics
GR310 portable gamma ray spectrometers, for that part of the Kimmeridge Clay Formation exposed between
Hobarrow Bay and Chapman's Pool, Dorset (SY 896 790-SY 955 771). For detailed sedimentological and
stratigraphical log of the section, for the definition of the bed group numbers which have partly been derived from
the literature and for formalization of the following beds; Clavell's Hard Stone Band, Little Stone Band and
Pectinatus Nodules, see Coe (1992). See Fig. 1 for location of outcrop sections.

72

Z.M.

AHMADI

& A. L. C O E
,~- o " o

~,..o

~-~o

~.~

~'~u~

~ ~

~.~=~
~'~_

>

0.j o*-~

.-~-~ .~ . ~
::s ~ " 0 ,,z=

0,.0

~9

~,. ~

~.~

.,..~

r~

~- .

~ ~ ~ 0
~ o ~ "~

~..~ ~ " 0

~-'~

p?,~,.0

N [-"

~. .~ =~ ~,~

.~~=~
~ ~ o
~ ~ ~~

CORRELATION

OF WIRELINE

LOGS WITH OUTCROP

73

~'~ ~

c~

0~, +-~ . ~

0.~ r'"
~.., r -~

..~

o "~ o

"~

~,~

U~g~~
-~

~-~-

~'~

~ ~

~ .~~

..~

~t"

'

9
~

'~

o=

m :-:,r-,I ~ N~..~
= ~ , - ~ - ~ ~, o

o== r ~

"~

74

Z. M. AHMADI & A. L. COE

sian-weighted, moving-average filter. The total


filter length used was 1.1 m. This results in a
filtered field density curve with similar character
and resolution to a wireline density log (Figs 5
and 6).

Surface to subsurface correlation

Upper Jurassic composite field gamma ray


log
The field gamma ray data from nine different
locations along the Dorset coast were combined
to produce a composite field gamma ray log for
the Upper Jurassic strata of Dorset (Fig. 3).
Comparison with the wireline gamma ray data
from borehole 98/11-4, which are plotted at the
same scale, show that the same general trends
and wireline log patterns are present in both sets
of data throughout the Upper Jurassic interval.
Clearly distinguishable in both log signatures are
the overall trends of decreasing and increasing
response which are interpreted as representing
long-term facies changes controlled by relative
changes in sea-level (Coe 1992). For instance,
the overall upwards decrease in gamma ray
values for the pallasioides Zone of the Kimmeridgian to the anguiformis Zone in the Portlandian reflects the change from marine mudstones
to carbonates interpreted as a long-term lowering of relative sea-level (Coe 1992, 1996). The
one notable difference is that the Lower Kimmeridge Clay (baylei to autissiodorensis zones) is
thicker in borehole 98/11-4 than in the outcrop
section. This is due to the fact that the
measurements for the Lower Kimmeridge Clay
were made on exposures situated on the footwall
of the Central Channel Sub-basin, where the
succession is apparently complete but thinner.

Comparison of the geoMetrics GR310 and


the Exploranium GR320
Figure 4 shows a comparison of the total gamma
ray measurements taken with the two spectrometers over part of the Kimmeridge Clay
Formation. The decreasing and increasing
trends, amplitude of variation, and the shape
of the peaks and troughs correlate very well. The
main difference between the two signatures is the
higher resolution of the Exploranium GR320
log, which results from the 30 cm average sample
interval compared to a 45 cm average sample
interval for the geoMetrics GR310.
The correlation coefficient between the two
field gamma ray logs over 245 m of the

Kimmeridge Clay Formation is 0.7 (Fig. 4) .


This was calculated using Corpac, a signal
correlation computer program (Globex Consulting Services, Ltd 1992) which is based on a
simple mathematical inverse method to correlate
two time series (in this case depth series)
described by Martinson et al. (1982). The reason
the correlation coefficient is not higher is because
the Exploranium GR320 log has higher resolution, and because of the gaps in the data. Higher
correlation coefficients are obtained if the two
logs are correlated over shorter intervals which
contain no gaps in the data.

Correlation of field and wireline gamma ray


logs
Upper Oxford Clay and Corallian Beds. The field
gamma ray log shows, from the base, an overall
upwards decreasing and then increasing trend in
the gamma ray values, as do the logs in
boreholes 98/11-4 and 98/11-3, reflecting the
change in lithology from mudstones to sandstones and limestones, and then back to
mudstones and iron-rich sandstones (Fig. 7).
The Nothe Grit Formation is a better defined
gamma ray low in boreholes 98/11-4 and 98/11-3
than in the field gamma ray log, probably
because the sands are cleaner in the boreholes
and the clays of the overlying Redcliff Formation contain a high percentage of carbonate in
the outcrop section. Three prominent gamma
ray peaks in the Osmington Oolite Formation
can be seen on both the field gamma ray log and
in 98/11-4 (Fig. 7). Over a wider geographical
area the Corallian Beds are lithologically very
variable, being comprised of shallow-marine
sandstones and limestones. Sequence stratigraphical interpretation of the wireline logs using
the number and character of the cycles does
permit a correlation to be made across the
Wessex Basin; however, the lateral lithological
variability makes correlation based purely on
the wireline log character difficult.
Kimmeridge Clay Formation. The similarity
between the field gamma ray logs produced by
the two different spectrometers and the wireline
gamma ray log from Encombe 1 borehole
(approximately 1 km inland from the outcrops)
is shown in Fig. 8. The data acquired using the
smaller sample interval with the Exploranium
GR320 spectrometer produces a higher resolution curve, despite the fact that the sampling
interval is about one third of the effective
sampling diameter of the tool (Fig. 2). Using
the methodology for calculating correlation

C O R R E L A T I O N OF W I R E L I N E LOGS W I T H O U T C R O P

75

.0. 2
~

t"r

Oo,0

gr
~t'q

qgoo
oo

,9, , . ~ o o

~r/2

O,.~

~~

. ,.,,~

76

Z.M. AHMADI & A. L. COE

..=~
"-'
.._,
"0

0
0
0

0,-. ,,...~

g~

,~

~
0
o "

C O R R E L A T I O N OF W I R E L I N E LOGS WITH O U T C R O P

77

~ 4 ~ . ~84

,~e-~N

~o

~.=_

~ .,..-,

78

Z.M. AHMADI & A. L. COE

coefficients described above, the correlation


coefficient for the Exploranium GR320 field
gamma ray log and the wireline log from the
Encombe 1 borehole is 0.92, but it is only 0.82
for the geoMetrics GR310 field gamma ray log
and the wireline log.
Figure 9 shows the similarity between the
geoMetrics GR310 field gamma ray log and
wireline gamma ray logs from boreholes up to
170 km away (Fig. 1). There are several
particularly prominent features, including the
two gamma ray lows with a low amplitude of
variation seen in the Collendean Farm (TQ 2480
4429) and Ashdown 1 (TQ 5005 3035) boreholes
in the hudlestoni and wheatleyensis zones, which
are often referred to as the 'Kimmeridge limestones' (Hancock & Mithen 1987). At outcrop,
these two gamma ray lows with a low amplitude
of variation are prominent thick homogeneous
calcareous mudstone units (middle and upper
part of bed 40 and the lower part of bed 44; Fig.
4). The two gamma ray lows in Collendean
Farm 1 and Ashdown 1 (Figs 1 and 9) are
probably more enhanced than those in 98/11-4
and the field gamma ray log because the
sediments have an even higher calcium carbonate content. A higher quartz sand content is
discounted because the lithology over the same
intervals in the nearby Warlingham borehole
(TQ 3476 5719) comprises argillaceous limestones and calcareous mudstones (Worssam et
al. 1971). Prominent gamma ray lows on the
field gamma ray log like those in the middle of
autissiodorensis Zone, at the base of elegans
Zone and near the top of scitilus Zone, are
carbonate-rich cemented horizons. Similar sharp
gamma ray lows in 98/11-4 and Collendean
Farm 1 probably also relate to calcareous
cemented horizons.

Correlation of field and wireline density logs


Upper Oxford Clay and Corallian Beds. Production of an outcrop density log over this interval
of mixed siliciclastics and carbonates is more
problematic than that for the Kimmeridge Clay
Formation. Processing of the data in a similar
manner to that of the Kimmeridge Clay
Formation resulted in a filtered density curve
with very little variation. However, further
processing of the data to take into account
dissolved carbonate cement at outcrop, as
described above, resulted in a curve which is
more similar to the borehole density logs.
The general trends seen on the filtered density
log (Fig. 5) show a positive correlation with the
wireline density from Marchwood 1 borehole

and to a certain extent with borehole 98/11-1.


The Winterbourne Kingston 1 borehole density
log is more difficult to correlate in the lower part
due to the lack of variation on the large scale.
Fig. 5 also shows the marked lateral and vertical
variation of the Oxfordian strata between the
wireline logs of 98/11-1, Marchwood 1 and
Winterbourne Kingston 1. One notable example
of this is the differences seen between the three
wells for the density of the Nothe Grit Formation (or its equivalent) and the Osmington Oolite
Formation.

Kimmeridge Clay Formation. Figure 6 shows the


comparison between the processed field density
log for the Dorset coast (filtered density log of
Fig. 6) against the nearby Encombe 1 borehole,
and the Bletchingley 1 (TQ 3622 4772) and
Detention 1 (TQ 7478 4020) boreholes in the
Weald Basin. The general trends and the
character of all of these logs is remarkably
similar. The four high peaks which straddle the
pectinatus to hudlestoni zonal boundary in both
the outcrop density log and Encombe 1 log
represent more carbonate-rich cementstone
beds. The distinctive increase in density seen in
all the logs at the top of the lower third of the
hudlestoni Zone represents at outcrop a change
from interbedded organic-rich and organic-poor
mudstones to a thick calcareous mudstone (Coe
1992). Similar lithological changes are interpreted to occur in the borehole sections.
Conclusions
(1) The geoMetrics GR310 and Exploranium
GR320 gamma ray spectrometers can both
be used to produce field gamma ray logs
which are comparable with borehole gamma
ray wireline logs. Whilst the newer Exploranium GR320 is more accurate and can be
used to gather spectral data more quickly,
the older geoMetrics GR310 does produce
excellent data with repeatable and comparable gamma ray trends. The most comparable signal between hand-held spectrometers
and wireline log tools is produced by using
the hand-held spectrometer perpendicular to
the bedding with a sample interval of 30 cm
or less.
Field gamma ray logs produced for the
Kimmeridge Clay Formation can be used to
positively correlate, often down to the bed
(typically < l m ) but at least down to the
bed group scale (typically 10 in), with wireline gamma ray data from nearby boreholes.
Larger gamma ray features can also be
correlated with boreholes as far as 170km

CORRELATION OF WIRELINE LOGS WITH OUTCROP


away in the Weald Sub-basin. Field gamma
ray and wireline gamma ray data for the
Oxfordian show similar trends but complex
local lithological heterogeneity may be misleading. The concepts of sequence stratigraphy (i.e. recognition of the metre to tens of
metre scale cycles) considerably aid in
making the correlation. This is because the
interpretation relies on the recognition of
wireline log trends rather than correlating
similar lithologies, and requires identification of stratigraphic gaps and condensed
intervals.
(2) Small rock samples from outcrop can be
used to produce a field density log. Some
processing of the data is required to produce
a signal which is directly comparable with
the wireline tool. The method could easily
be applied to small rock samples from core
or washed cuttings.
Excellent results comparable with the
wireline signature were obtained for a thick
succession of interbedded organic-rich and
organic-poor mudstones and cementstones
(Kimmeridge Clay Formation). Where the
lithology varies more widely and shallowmarine sandstones and limestones (e.g.
Corallian Beds) are present, it is necessary
to take into account the differences in
porosity between the borehole and outcrop
section and apply a further correction factor
to the outcrop density data.
(3) The measurement and processing of the
physical characteristics of rock exposures
to produce a wireline log signature is
invaluable in the understanding of boreholes
where core is not available. The data can be
readily used to supplement and enhance
conventional litho- and bio- stratigraphical
correlations between boreholes, and boreholes and outcrop.
Z. Ahmadi was supported by a Durham University
Research Studentship and an AAPG-PESGB Grantsin-Aid grant for field and laboratory studies. We thank
Charlotte Martin and Toby Harrold for their assistance in the field, and Brian Turner for the loan of his
geoMetrics GR310 gamma ray spectrometer. The
Exploranium GR320 was purchased from a grant
awarded to A. L. Coe from the Open University
Research Development Fund. M. Oates of British Gas
provided the wireline and biostratigraphical data from
boreholes 98/11-1, 98/11-3 and 98/11-4, and H. Bailey
of the British Geological Survey provided the wireline
data for the onshore boreholes in the Wessex Basin.
We would particularly like to thank N. Goulty for his
constructive comments during the preparation of this
paper and two anonymous referees are thanked for
reviewing this paper.

79

References
ADAMS, J. A. S. & GASPERINI, P. 1970. Gamma ray
spectrometry of rocks, Elsevier, Holland.
ATLAS WIREL1NE SERVICES. 1985. Log Interpretation
Charts. Western Atlas International, Inc.
ATLAS WIRELINE SERVICES. 1992. WDS advanced log
evaluation - documentation. Western Atlas International, Inc.
BESSA, J. L. & HESSELBO, S. P. 1997. Gamma ray
character and correlation of the Lower Lias, SW
Britain. Proceedings of the Geologists' Association,
108, 113-129.
CHAMBERLAIN,A. K. 1984. Surface gamma ray logs: a
correlation tool for frontier areas. American
Association of Petroleum Geologists Bulletin, 68,
1040-1043.
COE, A. L. 1992. Unconformities within the Upper
Jurassic of the Wessex Basin, Southern England,
DPhil Thesis, University of Oxford.
1995. A comparison of the Oxfordian successions of Dorset, Oxfordshire, and Yorkshire. In:
TAYLOR, P. D. (ed.) Field Geology of the British
Jurassic. Geological Society, London, 151-172.
1996. Unconformities within the Portlandian
Stage of the Wessex Basin and their sequencestratigraphical significance. In: HESSELBO,S. P. &
PARKINSON, D. N. (eds) Sequence Stratigraphy in
British Geology, Geological Society Special Publications No. 103, 109 143.
COWAN, D. R. & MYERS, K. T. 1988. Surface gamma
ray logs: A correlation tool for frontier areas:
Discussion. American Association of Petroleum
Geologists Bulletin, 72, 634-636.
Cox, B. M. & GALLOIS,R. W. 1981. The stratigraphy
of the Kimmeridge Clay of the Dorset type area
and its correlation with some other Kimmeridgian
sequences. Report of the Institute of Geological
Sciences, 80/4.
DAVIES, S. J. & ELLIOTT,T. 1996. Spectral gamma ray
characterisation of high resolution sequence
stratigraphy: examples from Upper Carboniferous fluvio~leltaic systems, County Clare, Ireland.
In: HOWELL, J. A. & AITKEN, J. F. (eds) High
Resolution Sequence Stratigraphy." innovations and
applications, Geological Society Special Publications No. 104, 25-35.
DYPVIK, H. & ERIKSEN, D. O. 1983. Natural radioactivity of clastic sediments and the contributions
of U, Th and K. Journal of Petroleum Geology, 5,
4094 16.
ETTENSOHN, F. R., FULTON, L. P. & KEPFERLE, R. C.
1979. Use of scintillometer and gamma ray logs
for correlation and stratigraphy in homogeneous
black shales. Geological Society of America
Bulletin, part II, 90, 828-849.
GALLOIS,R. W. 8z Cox, B. M. 1974. Stratigraphy of the
Upper Kimmeridge Clay of the Wash area.
Bulletin of Geological Survey of Great Britain,
47, 1-16.
HANCOCK,F. R. P. & MITHEN,D. P. 1987. The geology
of the Humbly Grove Oilfield, Hampshire, UK.
In: BROOKS, J. & GLENNIE, K. (eds) Petroleum
Geology of North West Europe, Graham & Trot-

80

Z . M . AHMADI & A. L. COL

man, 161-170.
HESSELBO, S. P. 1996. Spectral gamma ray logs in
relation to clay mineralogy and sequence stratigraphy, Cenozoic of the Atlantic Margin, offshore
New Jersey. In: MOUNTAIN, G. S, MILLER, K. G,
BLUM, P., POAG, C. W. & TWlCHELL, D. C. (eds)
Proceedings of the Ocean Drilling Program Scientific Results, 150.
LOVBORG, L. 1984. The calibration of portable and
airborne gamma ray spectrometers - theoo',
problems and facilities, Report Riso-M-2456, Riso
National Laboratory, Denmark.
& MOSE, E. 1987. Counting statistics in
radioelement assaying with a portable spectrometer. Geophysics, 52, 555-563.
, WOLLENBERG,H., SORENSEN,P. & HANSEN, J.
1971. Field determination of uranium and thorium by gamma ray spectrometry, exemplified by
measurements in the llimaussaq alkaline intrusion, South Greenland. Economic Geology, 66,
368-384.
MARTINSON,D. G., MENKE,W. & STOFFA,P. 1982. An
inverse approach to signal correlation. Journal of
Geophysical Research, 87, 4807~4818.
MELNYK, D. H., SMITH, D. G. & AMIRI-GARROUSSl,K.
1994. Filtering and frequency mapping as tools in
subsurface cyclostratigraphy, with examples from
the Wessex Basin, UK. In: DE BOER, P. L. &
SMITH, D. G. (eds) Orbital Jorcing and cyclic
sedimentary sequences, International Association
of Sedimentologists, Special Publications 19, 35
46.
, ATHERSUCH,J., AINSWORTH,N. & BRITTON, P.
D. 1995. Measuring the dispersion of ostracod
and foraminifera extinction events in the subsurface Kimmeridge Clay and Portland beds, Upper
Jurassic, United Kingdom. In: MANN, K. O.,
LANE, H. R. & SCHOLLE,P. A., Graphic correlation, Society of Economic Paleontologists and
Mineralogists, Special Publications, 53, 185-203.
MYERS, K. J. 1987. Onshore-outcrop gamma ray
spectrometry as a tool in sedimentological studies.
PhD thesis, University of London.
- - &
BRISTOW,C. S. 1989. Detailed sedimentology
and gamma ray log characteristics of a Namurian
deltaic succession II: gamma ray logging. In:
WHATELEY, M. K. G. & PICKERING,K. T. (eds),
Deltas." Sites and traps for jbssil fuels, Geological
Society Special Publications, 41, 81-88.
& W1ONALL, P. B. 1987. Understanding
Jurassic organic-rich mudrocks
new concepts

using gamma ray spectrometry and palaeoecology: examples from the Kimmeridge Clay of
Dorset and the Jet Rock of Yorkshire. In:
LE~GETT, J. K. & ZUFFA, G. G. (eds) Marine
Clastic Sedimentology - concepts and case studies,
Graham & Trotman, London, 172-189.
PARKINSON,D. N. 1996. Gamma ray spectrometry as a
tool for stratigraphical interpretation: examples
from the western European Lower Jurassic. In:
HESSELBO, S. P. & PARKINSON, O. N. (eds)
Sequence Stratigraphy in British Geology, Geological Society Special Publications, 103, 231-255.
PENN, I. E., Cox, B. M. & GALLOIS, R. W. 1986.
Towards precision in stratigraphy: geophysical
log correlation of Upper Jurassic (including
Callovian) strata of the Eastern England Shelf.
Journal of the Geological Society, London, 143,
381-410.
RIDER, M. H. 1991. The geological interpretation of
well logs. Whittles Publishing, Caithness.
SCHLUMBEROER 1994. Log Interpretation Charts.
Schlumberger Wireline & Testing, Houston,
Texas.
SERRA, O. 1984. Fundamentals qf well-log interpretation
1. The acquisition of logging data. Developments
in Petroleum Science 15A. Elsevier, Holland.
SLATT, R. M., JORDAN. D. W., D'AGOSTINO, A. E. &
GILLESPIE, R. H. 1992. Outcrop gamma ray
logging to improve understanding of subsurface
well log correlation. In." HURST, A., GR1FFITHS, C.
M. & WORTHINGTON, P. F. (eds) Geological
Applications of Wireline Logs H, Geological
Society Special Publications, 65, 3-19.
TALWAR, A. D., HENDERSON, A. S. & HART, M. B.
1992. Simple gamma ray response of the Upper
Jurassic from the Dorset coast - a preliminary
investigation using the scintillometer profile technique. Proceedings of the Ussher Society, 8, 70-72.
VAN BUCHEM,F. S. P., MELNYK,D. H. & McCAvE, [.
N. 1992. Chemical cyclicity and correlation of
Lower Lias mudstones using gamma ray logs,
Yorkshire, UK. Journal of the Geological Society,
London, 149, 991-1002.
WmTTAKER, A. (ed.) 1985. Atlas of Onshore Sedimentary Basins in England and Wales: Post-Carboniferous Tectonics and Stratigraphy. Blackie,
Glasgow.
WORSSAM, B. C., IVIMEY-COOK, H. C. 1971. The
stratigraphy of the Geological Survey Borehole
at Warlingham, Surrey. Bulletin of the Geological
Survey of Great Britain, 36, 1-146.

Quantitative lithology: open and cased hole application derived from


integrated core chemistry and mineralogy database
M. M. H E R R O N & S. L. H E R R O N
Schlumberger-Doll Research, Old Quarry Road, Ridgefield, C T 06877-4108, USA

Abstract: A new quantitative lithology interpretation is based on elemental concentrations of


silicon, iron, calcium and sulfur available from logs. The lithology interpretation is founded
on an integrated chemistry-mineralogy core database comprising over 400 samples from
many wells of predominantly sand and shaly sand composition located on four continents.
The lithological components include 'clay', which is the sum of all clay minerals; 'carbonate',
which is the sum of calcite and dolomite; "anhydrite', which is the sum of anhydrite plus
gypsum; and 'sand' or 'quartz-feldspa~mica', which is the remainder of the formation
essentially constituting the sand fraction. The new interpretation demonstrates that the
elements aluminium alone or a combination of silicon, calcium, and iron provide a much
more accurate estimation of clay than either gamma ray or its individual components
potassium, thorium and uranium. Calcium alone or calcium and magnesium are used to
determine carbonate concentrations. Calcium and sulfur can be used to estimate the
anhydrite fraction. Having estimated the total clay, carbonate, and anhydrite fractions, the
remainder of the formation is assumed to be composed quartz, feldspar, and mica minerals.
Examples of the new lithology interpretation are provided for core data and also for
geochemical log data from both open and cased hole environments.

The accurate determination of formation lithology from common geophysical logs is hindered
by a lack of sensitivity coupled with nonunique
responses to the minerals that reside in sedimentary rocks. The interpretation of lithology for the
purpose of wireline petrophysical evaluation or
geological characterization primarily consists of
estimating fractions of shale, sand, and carbonate. Nuclear logs, either gamma ray, photoelectric factor, and/or a combination of neutron
and density are the most commonly used logs for
lithology interpretation, A desire for improved
accuracy in Ethological description led to the
introduction of several generations of nuclear
spectroscopy logs. Recent developments in open
and cased hole logging have made it possible to
obtain accurate concentration logs for the
elements silicon, calcium, iron, sulfur, titanium,
and gadolinium at relatively low cost and high
logging speeds (Herron 1995).
A new lithological interpretation has been
developed to capitalize on these new logging
capabilities. It is founded on an extensive
database of core chemistry and mineralogy.
The new interpretation provides quantitative
estimates of: total clay, which is the sum of all
clay minerals; carbonate, which is the sum of
calcite and dolomite; anhydrite, which is the sum
of anhydrite plus gypsum; and quartz-feldsparmica (Q F-M), which is the remainder of the
formation essentially constituting the sand frac-

tion. The clay, carbonate, and quartz-feldsparmica portions of this interpretation have been
presented previously (Herron & Herron 1996).
This paper provides a brief introduction to the
new geochemical logging capabilities in both
open and cased holes and a detailed examination
of the new core-based interpretation.

Elemental concentration logs


The recently developed technique to estimate
elemental concentrations from a single, inducedneutron gamma ray spectrometer (Herron 1995)
is an adaptation of a geochemical oxides closure
model already employed in the computation of
elemental concentrations from multiple nuclear
sondes (Hertzog et al. 1987; Schweitzer et al.
1988; Grau & Schweitzer 1989; Grau et al. 1989).
The most significant modifications are:
(1) the elimination of aluminium and potassium as necessary inputs to the geochemical
closure model, thus considerably reducing
the number of wireline sondes necessary to
produce elemental concentrations of potassium;
(2) a change in the elemental associations of
iron.
Figure 1 presents examples of elemental concentration logs from the new processing using data

HERRON, M. M. & HERRON,S. L. 1998. Quantitative lithology: open and cased hole application
derived from integrated core chemistry and mineralogy database. In. HARVEY,P. K. & LOVELL,M. A.
(eds) Core-Log Integration, Geological Society, London, Special Publications, 136, 81-95

81

82

M. M. H E R R O N & S. L. H E R R O N
200

3OO

4O0

500
e-

600

7OO

[}

800

900
0

50 0

20
40 0
10
20 0
10
20 0
2
4 0
20
40
Calcium wt%
Iron + .t4AI ~%
Sulfur wt%
Titanium wt%
Gadolinium ppm

Silicon wt%

Fig. 1. Openhole elemental concentrations from the Elemental Capture Spectroscopy (ECS; Mark of
Schlumberger) sonde.

x 104

1.01

7-

r-.

.,....

1.03

.v
p,.

O..

g~

1.05

1.0"/

][
0

50 0
Silicon wt%

w
,

~,

20
40 0
10
20 0
10
20 0
2
4 0
20
40
Calcium wt%
Titanium wt%
Iron + 14AI wt%
Sulfur wt%
Gadolinium ppm

Fig. 2. Cased hole elemental concentrations from the (RST; Mark of Schlumberger) Reservoir Saturation Tool.

from an open hole Elemental Capture Spectroscopy (ECS; Mark of Schlumberger) sonde. This
is a nuclear spectroscopy device which uses a
standard AmBe source and a BGO detector. It is
combinable and can log at up to 540 m hr -l (1800
fthr-~). Chemical concentrations measured on

core samples are shown for comparison. Two


points should be made when examining the data.
The first is that since the uncorrected prompt
capture yield for iron contains gamma rays from
both Fe and A1, the log Fe should be approximately equal to Fe+0.14A1. Accordingly, the

QUANTITATIVE LITHOLOGY

'~176wo,:-.

/1

- /

. . . . . . . . . . . . . . . . . .

.......... r

'O01we"/

Well6 ~ o ' o ~

I Well 7

"o /

Well i

I/:"

oL~ ." .

~176
w~176/
0

100
200
Gamma Ray

Well 10

100
200
Gamma Ray

100
200
Gamma Ray

100
200
Gamma Ray

Fig. 3. Synthetic gamma ray (computed from Th, U and K concentrations) plotted against total clay (kaolinite,
illite, smectite, chlorite and glauconite) measured on the same sample for 12 datasets. Although GR crudely
correlates with total clay, the slopes and offsets vary widely from well to well.

core points plotted for c o m p a r i s o n are


Fe+0.14A1. The second point is that the log
concentrations agree well with core data.
A second example is provided from a cased
hole Reservoir Saturation Tool (RST; Mark of
Schlumberger) log acquired from a well in
Venezuela (Fig. 2). This example is processed
using new elemental standards to derive the far
detector capture yields (Roscoe et al. 1995), and
corrections are made for casing and a 3.8cm
cement annulus. The results show good agreement between log concentrations and the sparse
core data.

tion coupled plasma mass spectrometry, for


whole rock elemental concentrations of silicon
(Si), aluminium (A1), iron (Fe), calcium (Ca),
magnesium (Mg), sodium (Na), potassium (K),
phosphorus (P), titanium (Ti), manganese (Mn)
and chromium (Cr), expressed as oxides, plus
Loss on Ignition (LOI) representing total volatiles, H20 +, H 2 0 , sulfur (S), organic carbon,
thorium (Th), uranium (U), gadolinium (Gd)
and boron (B). A synthetic core gamma ray
(GR) computed from core chemistry using the
gamma ray response is given by equation (1)
G R = 4 T h + 8 U + 16 K

Core database
The development of the new quantitative lithology interpretation begins with a core database
that contains chemistry and mineralogy measurements on over 400 core plug samples from
numerous wells on four continents. The wells are
diverse in age and geographic location, but all
are predominantly sands and shaly sands.
To analyse the samples, rocks were crushed
and split with a microsplitter into chemistry and
mineralogy fractions. The chemistry fraction was
analysed at X-Ray Assay Laboratories using XRay fluorescence, neutron activation and induc-

(1)

where Th and U concentrations are in ppm and


K concentrations are in wt% (Ellis 1987).
The mineralogy fraction was analysed using a
new Fourier Transform-Infrared (FT-IR) procedure which simultaneously analyses the mid-IR
and far-IR frequencies. The mid-IR procedure
was described in Matteson & Herron (1993).
Since that time the number of mineral standards
has been increased to 26 with approximately the
same level of accuracy (better than +2 wt %).
The mineral standard set includes quartz, albite,
anorthite, K-feldspar, muscovite, biotite, kaolinite, illite, smectite, chlorite, glauconite, calcite,

84

M. M. HERRON & S. L. HERRON


o 1~176
9 dP

%
~ 9 9 9

e 9 1 4 90 ~

10
20
Thorium ppm

e"

5
Uranium ppm

2.5
5
Potassium wt%

loo

+t

_~ so
o

t 9

.,'.;

+/

00

10
20
Aluminum wt%

1
Titanium wt%

.t
.

5
Gadolinium

10

20

40

loo[
r
t
o

25
Silicon wt%

50

15

Iron wt%

30

Calcium wt%

Fig. 4. Comparison of individual chemical elements that can be measured by logging against total clay for Well 3.
A1 shows a strong positive correlation that is mirrored by the negative correlation with Si.

dolomite, siderite, ankerite, magnesite, aragonite, gypsum, anhydrite, hematite, barite and
opal. Total clay is the sum of kaolinite, illite,
smectite, chlorite and glauconite. Although there
are significant amounts of mica, another layered
silicate, they are not included in the total clay
fraction. At high clay concentrations there is
sometimes interference between illite and mica
phases.
Exploring elemental relationships

The most complex aspect of the new lithology


interpretation is the computation of the clay
mineral fraction. In the logging world, clay, or
more often volume of shale, is most frequently
estimated from the gamma ray log. However,
there are many type of clay minerals with widely
differing compositions and log responses, so
shale estimates often carry large uncertainties.
The estimation is further degraded by the many
non-clay minerals which contribute significantly
to the gamma ray.

Gamma ray and clay


With the core database, it is possible to evaluate
the relationship between total clay determined by
FT-IR and the computed gamma ray on a

porosity-free basis, as recently advocated by


K a t a h a r a (1995). The relationship for core
samples from 12 data sets is presented in Fig.
3. A line connecting the origin with 100% clay
and 250 API is included for visual reference. As
expected, gamma ray content generally increases
as clay content increases. However, there are a
number of characteristics in the clay-gamma ray
plots that highlight the weaknesses inherent in
this approach; many of these have been recently
discussed by Bhuyan & Passey (1994) and Hurst
& Milodowski (1994).
The first major feature is the large range of
slopes in the gamma ray versus clay plots which
demonstrates the necessity for local calibration.
For example, in Well 1, a linear trend predicts a
maximum gamma ray value of about 100 API
for the pure clay end member, whereas Well 2
would predict 500 API. For Well 12, a pure clay
would have a gamma ray of only about 150 API.
In several wells, either the data or an extrapolation of the data to zero clay indicate a near zero
minimum gamma ray, but Well 4 has a minimum
gamma ray of 30 API, and in Well 12 an
extrapolation points to 70 API for minimum
gamma ray. The difference between evaluating
these plots and using only log data is that with
core calibration the amount of clay is known,
and it is possible to accurately extrapolate to

QUANTITATIVE LITHOLOGY

~5o
~r1~176" i
I.=.

85

~ Jp

O0 e~ II

_r

o;=~'.;'10"< " " 20

Thorium ppm

--

5
10
Uranium ppm

2.5
Potassium wt%

100,

I.,../)--:-.

.=" -o-'

50

06

lb
Aluminum wt%

20

_.

5
Gadolinium

T'aanium wt%

10

100,

-%
~149

25
50
Silicon wt%

O~

15
Iron wt%

30

20
40
Calcium wt%

Fig. 5. Comparison of individual chemical elements that can be measured by logging against total clay for Well 5.
A1 again shows a strong positive correlation with clay. The negative correlation with Si is slightly perturbed by
high Fe siderite samples.
,100]

.fi"
:-

~0

I.%',,
10

Thorium ppm

20

~5oI '~:~"""

2.5
5
Potassium wt%
o
o

ee

9 9
e~po

d.'t' t

lb

Aluminumwt*/,

20

100

5
10
Uranium ppm

k,..'

100, ,

~~'

1
2
Titanium wt*/,
....

5
Gadolinium

20

10

.
9 00

u."
0

"4.

25
Silicon wt%

50

~
0

15
Iron wt%

30

Calcium wt%

40

Fig. 6. Comparison of individual chemical elements that can be measured by logging against total clay for Well 6.
A1 shows a strong positive correlation. The anticorrelation with Si is significantly perturbed by carbonates.

86

M.M. HERRON & S. L. HERRON

comparison of clay with aluminium. In Well 3,


aluminium displays a strong relationship with
total clay. The remaining elements in this figure
are some that can be obtained by prompt
thermal neutron capture spectroscopy logging
devices. The two elements remaining in the
second row of the figure are titanium and
gadolinium. These elements are commonly enriched in shales, but they show only a loose
correlation with total clay.
The third row holds the key to a new
technique for estimating clay. It begins with
silicon, which is a major constituent of rock
forming minerals. Although silicon is commonly
associated with quartz, it is actually the second
most abundant element after oxygen in both
sandstones and shales. Because it is a major
element, its abundance is not affected by trace
minerals, and concentrations form a smooth
continuum between high silicon sandstones and
medium silicon shales. For reference, quartz has
46.8 wt% silicon. The next element is iron, which
has numerous associations, including heavy
minerals such as siderite, pyrite, hematite, and
magnetite and the clay minerals illite, chlorite,
glauconite and some smectites. High concentrations of the heavy iron minerals can interrupt the
smooth relationship between silicon and clay
content. The final element is calcium which is
mainly associated with the carbonate minerals
calcite and dolomite. The low calcium concentrations indicate the absence of carbonate
minerals in Well 3.
The same type of comparison between total
clay and elemental concentrations is presented in
Fig. 5 for Well 5. For this well, none of the
individual elements (Th, U and K) contributing
Seeking an elemental alternative
to natural gamma ray is any better correlated
with clay than is total gamma ray. In contrast,
One goal of this study is to identify an
aluminium again shows a tight correlation with
alternative, less subjective approach to determining clay content using elemental data available clay content. Silicon again shows a strong
from nuclear spectroscopy logging devices. The negative correlation with clay, but there are
two data points which clearly deviate from the
technology exists to measure elemental concenmajor trend. These two samples contain 13 and
trations from natural radioactivity (Th, U and
38 wt% siderite (FeCO3) as reflected by the two
K), neutron activation (AI), and capture gamma
high iron points. As in Well 3, the near absence
ray spectroscopy (Si, Ca, Fe, Ti, Gd and S).
of calcium reflects the absence of calcite and
Figure 4 shows a comparison of clay content
with all available logging elements (except sulfur) dolomite.
A final example of the element-clay comparfor Well 3. The three components of natural
gamma ray; Th, U, and K, are presented in the isons is presented in Fig. 6 for Well 6. In this
first row. Thorium and uranium show wide well, thorium and potassium exhibit positive
scatter and little correlation with clay. In this correlations with clay, but the degree of scatter
well, potassium shows a strong correlation with precludes the use of these elements for accurate
clay, but examination of data from four other clay prediction, especially at low clay contents.
Aluminium again shows a strong positive correwells in the field reveals that this correlation
breaks down entirely in sands containing less lation with clay. Silicon again shows a negative
correlation with clay, but the impact of carbothan 25% clay.
The second row of Fig. 4 begins with a nate minerals on the silicon--clay curve is much

zero clay. With only log data, one must choose a


minimum and maximum gamma ray value
without knowing the correspondence to real clay
concentrations, and the picture is further complicated by porosity variations. For Well 11, the
minimum gamma ray value observed on the log
is about the same as the 50 API minimum
computed for the core data. This value would
normally be assigned to zero clay instead of the
actual 25 wt% clay. Clearly, such a log
interpretation would severely under-estimate
the clay content in the well.
The second dominant feature in Fig. 3 is the
scatter in the data, particularly in Wells 1-10. In
these wells, even if the observed correlation
between gamma ray and clay were known, the
scatter in the data would produce an uncertainty
of as much as +20 wt% clay or more. For Wells
3, 5, 7 and 9, at levels of about 20% clay,
observed gamma ray values span almost the full
range from clean sand to shale. The relative error
is particularly large in sands.
A third and less common feature is that some
wells exhibit a small dynamic range in gamma
ray while clay content varies considerably. This
is notable in Well 12 which is a typical offshore
Gulf of Mexico example. It is also true in Wells 4
and 11.
In spite of the problems outlined above, it
would be possible to make good clay predictions
in Wells 2, 11 and 12 if detailed and accurate
core data were available. Without such a
calibration, it is doubtful that the picks for
GRmax and GRmi n from the log data would
match the core calibration parameters.

QUANTITATIVE LITHOLOGY

100

87

Well I

.="/

Well5o

_~ s o

100
Well6

o/

-./
.

; Y

100

I Well9 /

Well8

.'.s/

Well1 2 ~

Aluminum
wt%

00

Alumi10num
wt%

20 0

10
'
20 0
Aluminumwt%

Alumi1()num
wt%

20

Fig. 7. Aluminium versus total clay for all 12 wells. The correlation with total clay is much tighter for aluminium

than for GR. In addition, the slopes are about the same and most wells show a near-zero offset.

more obvious 9There are many samples with high


calcium reflecting calcite concentrations that
range from 0 to 85 wt%. This mineral assemblage produces a ternary composition diagram in
the silicon-clay plot with the vertices representing pure carbonate, clean sand, and shale.
Summarizing the observations in Figs 4
through 6, it appears that aluminium is the best
single elemental indicator of clay. Silicon shows
a complementary anti-correlation to clay content, but the simple linear relationship between
silicon and clay is distorted by carbonate
minerals. The carbonate content is chemically
represented by calcium and/or iron. These trends
are typical of those observed in the other data
sets.
Having observed the strong relationship between aluminium and clay, it is useful to examine
the data for all 12 wells, as shown in Fig. 7. In 10
of the 12 wells, the slope of the aluminium-clay
plot is nearly constant. In 9 of the 12 wells, the
intercept of the aluminium-clay linear relationship is essentially zero. Comparison between Fig.
7 and Fig. 3 shows that aluminium is a much
better clay estimator than gamma ray in most
wells. This is true even when a porosity-free core
calibration is available for gamma ray, and it is
especially true in the sands 9

The improvement of aluminium over gamma


ray is marginal in Well 8, but it is significant for
the cleanest sands. In Wells 11 and 12, the
aluminium and gamma ray are comparable clay
indicators if the core calibration is known.
However, a log interpreter who equates the
minimum gamma ray response with zero clay
introduces a 20 to 25 wt% error in the clay
estimation.
Aluminium has an even more striking relationship with the sum of clay plus mica. This is
demonstrated in Fig. 8. Improvements in the
correlation with aluminium are most notable in
Wells 7 and 8, and the effects are most obvious in
the shales. The lines drawn in Fig. 8 represent a
slope of 6.4 and the relationship for the first 10
wells has a correlation coefficient of 0.98. It is
possible that some of the differences between
Figs 7 and 8 are due to analytical interference
between illite and mica phases in shales 9 The
decision to include or exclude mica from the clay
fraction depends on the application. Since micas
do not contribute significantly to clay counterion conductivity, they are not generally included
in saturation interpretation. On the other hand,
like clays, micas can be detrimental to formation
productivity.
There are several reasons for the strong

88

M.M. HERRON & S. L. HERRON


100

5c

I~
+

100

....

~;
+ 50 /

,/,

tO
|

lOO

~ 50
0

Well1 /

Well1 0 /

lO
|

Aluminum wt%

20 0

Well 12

/
1'0

Aluminum

wt% 20 0

1'0

Aluminum

wt% 20 0

10

Aluminum

.......

wt% 20

Fig. 8. Aluminium versus total clay +mica for all 12 wells shows an even tighter and more universal relationship
than aluminium versus clay.

correlation between aluminium and total clay


mineral content. Clays are aluminosilicates;
aluminium is a major element in and an integral
part of the chemical composition of virtually all
clays. This is very different from the case of
thorium and uranium which occur at trace (ppm)
levels and are not structural components of the
clays. Of course, the clay-A1 relationship is a
simplified picture and is not expected to be
perfect. Different clay minerals have different A1
concentrations and there are important nonclay
minerals that contain aluminium.

Relationship between AI and Si, Ca, Mg and Fe


Although aluminium is the best element for clay
estimation, its measurement in a borehole is
accomplished by induced neutron activation and
currently requires a chemical source, two gamma
ray spectrometers, and an independent measurement of formation capture cross-section, making
it an expensive measurement. Fortunately, an
alternative exists due to the complementary
relationship between aluminium and the elements silicon, calcium, magnesium and iron.
This relationship is illustrated in Fig. 9, which
combines elemental data from all 12 data sets

into three plots. For samples containing more


than 2 wt% organic carbon, the elemental data
must be normalized to an organic-free matrix or
else they will perturb the linear relationship.
Earlier, Figs 4 to 6 showed that as clay
increases, silicon decreases. Therefore, as aluminium increases, silicon decreases. In Fig. 9a,
silicon is converted to SiO2 (by multiplying by
2.139) and subtracted from 100. Now, we see
that as AI increases, 100-SiO2 also increases. In
this presentation, carbonate minerals drive the
data toward A1 of zero and ( 1 0 0 - SiO2) values of
100 wt%. We can use concentrations of Ca and
Mg to compensate for the presence of calcite
(CaCO3) and dolomite (CaMg(CO3)2). Fig. 9b
shows that concentrations of A1 vary linearly
when plotted against 1 0 0 - S i O 2 - CaCO3 - MgCO3 concentrations and that the additional
terms remove almost all of the disturbance of
that major trend. The few remaining outliers are
predominantly siderite or pyrite, and they can be
removed as 1.99Fe where the coefficient of 1.99 is
optimized on these data.
The trend in Fig. 9c can be used to estimate
the aluminium concentration from
A1 = 0.34(100 - SiO2 - CaCO3 MgCO3-1.99Fe),

(2)

.r

89

QUANTITATIVE LITHOLOGY

20[a

I' b

o 'ot

06

5"0

100

100 - Si02

00

50

100

100 - Si02- CaCOzMgC03

O Q

50

100

100 - Si02- CaCOsMgCOs- 1.99Fe

Fig.

9. Aluminium is estimated from the other major elements in sedimentary rocks. (a) A1 vs 100-SIO2 shows a
clear trend that is disturbed primarily by carbonates. (b) A1 vs 100 SiO2-siderite and dolomite shows a very tight
trend that is disturbed only by siderite and pyrite rich samples. (c) When the high-Fe minerals are corrected for, A1
can be estimated from Si, Ca and Fe.

2~ We"' /
"10I f
~< 0 U ;

Well 4

'

2~IWe,./,5

"

IlU'

2O

E
<
e

00

10
20 0
10
20
10
20 0
10
20 0
Aluminum Emulator
Aluminum Emulator
AluminumE m u l a t o r
Aluminum Emulator

Fig. 10. Aluminium estimated from Si, Ca, Mg and Fe closely matches measured aluminium in all 12 wells.
which produces estimates of A1 with a correlation coefficient of 0.99 and a standard error of
0.6 wt% A1. Figure 10 presents a comparison of
measured A1 concentrations with those estimated
from equation (2) for each of the 12 wells.
Clearly, this is a robust means of estimating A1
from Si, Ca, Mg and Fe.

Quantitative lithology
The strong correlation between aluminium and
clay provides the cornerstone of the lithology

interpretation. This relationship can be quantified to estimate clay, and the elements calcium,
magnesium, and sulfur can be used to estimate
the other major mineralogical components. The
mineralogical fractions defined here are different
from the lithologies commonly used in log
interpretation. The main difference is that a clay
fraction rather than a shale fraction is computed.
According to Bhuyan & Passey (1994), shales
commonly have about 60 wt% clay minerals and
40 wt% Q - F - M . Using this ratio, a rock with 60
wt% clay is 100 wt% shale. The other difference

90

M. M. HERRON & S. L. HERRON


100 .........

.,

]Well 1

,~

01r
ol00[Well5 " /

Well 4

Well 6

"

o /

./

./

jwey.

IWe"7/

Iwe"

0[/r

'O01we"9/

l/.
0

50
1O0 0
Estimated Clay

50
100
Estimated Clay

50
1O0 0
Estimated Clay

50
1O0
Estimated Clay

Fig. 11. Clay estimated from Si, Ca, Mg and Fe plotted against total clay for all 12 wells is a near duplicate of
Fig. 7.

is that the values determined here are all on


porosity-free (or matrix) basis, and they are
weight rather than volume fractions.

minimum of 1.3 for Well 10.


If we solve for clay plus mica (Fig. 8) instead
of clay, we obtain the following equation:

Estimating clay

Clay + Mica = 2.20(100- SiO2-

The two major points from the preceding section


are that A1 correlates well with clay content and
that aluminium concentrations can be estimated
from Si, Ca, Mg and Fe. The next logical step is
to estimate clay content from Si, Ca, Mg and Fe
using the form of equation (2). The problem is
set up to determine clay content by optimizing
the slope. Samples from Wells 11 and 12 are
excluded from the optimization because, as seen
in Fig. 7, the relationship between aluminium
and clay differs significantly from the relationships observed in Wells 1 through 10. The new
clay algorithm is:
Clay = 1.67(100 - SiO2 - CaCO3 M g C O 3 - 1.99Fe),

(3)

which has a correlation coefficient of 0.94 and a


standard error of 6.9 wt% clay. The slope of 1.67
obtained here is representative of the combined
datasets. Slopes optimized on individual datasets
range from a maximum of 2.0 for Well 1 to a

C a C O 3 - M g C O 3 - 1.99Fe),

(4)

with a correlation coefficient of 0.97 and a


standard error of 6.5 wt%.
Figure 11 presents measured clay content and
estimates from equation (3) for all 12 wells. The
estimated clay concentrations are in good agreement with the measured values for Wells 1-7, 9
and 10. They are almost the same as the
estimates from aluminium shown in Fig. 7.
For most of the first ten wells, the clay
estimates portrayed in Fig. 11 constitute an
improvement over those attainable from gamma
ray. The scatter in the estimate is drastically
reduced, particularly at the low clay concentrations where clay estimation is most critical. This
is especially clear in Wells 1-7 and 9 and 10. In
Well 8, the estimate of clay shows a less
spectacular effect relative to gamma ray, but it
does offer slight improvement in the clean sands.
In this well, an estimate of clay plus mica would
clearly be superior to gamma ray estimates.
Equation (3) is a general algorithm for

QUANTITATIVE LITHOLOGY

91

estimating clay from elemental data. It has broad


applicability and does not require picks of
minimum and maximum values. Unlike neut r o n - d e n s i t y separation, equation 3 is not
affected by the presence of light hydrocarbons
or gas. Although the slope would vary if
optimized on individual datasets, the overall
slope of 1.67 in equation (3) produces a good
clay estimate.
For Wells 11 and 12 the clay estimated from
equation (3) agrees with the measured clay in the
cleanest samples but under-estimates the clay
content of the shales. The cleanest samples in
these two wells have 20 and 28 wt% clay, and it
is not obvious which way the data would trend in
cleaner rocks. This is the same trend observed for
these two wells in the comparison of aluminium
versus clay.
The problem with the interpretation of clay
from A1 or from Si, Ca and Fe in Wells 11 and 12
is basically the same as the problem with
interpreting gamma ray. Inherent in both interpretation schemes is the presumption that nonclay minerals do not interfere. For most wells,
this is true for aluminium. However, Wells 11
and 12 are characterized by feldspar-rich sands.
This is true to a lesser degree for Well 4. In fact,
for all three of these wells, there is an anticorrelation between clay and non-clay aluminosilicates (feldspars plus micas). The high feldspar
content of the sands can be either authigenic as
in Well 11 or detrital as in Well 12.
In spite of vast geological differences, Wells 11
and 12 show similar patterns in terms of
aluminium vs clay. This suggests that a common
algorithm might exist to interpret clay content in
these wells, and if so, it might be broadly
applicable to feldspar- or mica-rich sands. The
relationship determined by least absolute error
optimization on the combined Well 11 and Well
12 datasets is:

feldspathic sands, so the application of equation


(5) requires some external knowledge.

Clay2 = -20.8 + 3.1 (100 - SiO2 -

Here, the non-zero offset of - 7 . 5 wt% accounts


for the small calcium contribution from plagioclase feldspar in sandstones, and the offset and

C a C O 3 - M g C O 3 - 1.99Fe)

(5)

This differs from equation (4) by modifying the


slope and introducing an intercept. The results
for Wells 11 and 12 are compared to measured
clay in Fig. 12. Data from Well 4, which also has
moderately feldspar-rich sandstones, are included as different symbols; this well was not
included in the optimization. Equation (5) for
feldspar-rich sandstones gives reasonable results
for clay contents in the reservoir rocks despite
the fact that these wells are from very different
geological environments. Using the geochemical
data alone, it is not possible to identify such

loo ;g/

~,
5

50

0 ~0

50
1 oo
Estimated Clay2

Fig. 12. Clay estimated from equation (5) for feldsparrich sands and shales vs measured clay for Wells 11 and
l 2 (o) and Well 4 (+).
Estimating carbonate
The second c o m p o n e n t in this lithological
description is the carbonate fraction. The carbonate fraction will be determined from calcium,
but first we need to consider the calcium
concentration which we obtain from log data.
Pure calcite (CaCO3) formations have Ca concentrations of 40 wt%, and this concentration is
accurately reflected by log data. A complication
arises in dolomites (CaMg(CO3)2) because magnesium has not normally been detected by
spectroscopy logs. As a result, the log calcium
concentration in a pure dolomite is also 40 wt%
(see Hertzog et al. 1987 and Roscoe et al. 1995
for detecting Mg from logs). This is equivalent to
saying that the Ca detected by logs equals
C a + 1.455Mg, an expression that equals 40
wt% in either pure calcite or dolomite. Using
the core data base, calcite plus dolomite concentrations were optimized as a function of
(Ca + 1.455 Mg) to produce equation (6):
Calcite + Dolomite
- 7.5 + 2.69(Ca + 1.455Mg).
=

(6)

/ ,/
o

o
50
1 oo
Estimated Calcite + Dolomite

Fig. 13. Calcite plus dolomite estimated from equation


(6) vs measured calcite plus dolomite for all twelve
wells.

92

M.M. HERRON & S. L. HERRON

oo[

so

0
50
100
0
50
100
E~imated Clay wt%
Estimated Carbonate wt%

0
50
100
Estimated Q-F-M wt%

Fig. 14. Comparison of estimated and measured quantities of clay, carbonate, and quartz-feldspar-mica on
samples from all 12 wells.

slope (2.69) are balances to provide the correct


answer in pure carbonate. The carbonate estimate from equation (6) closely approximates the
sum of calcite plus dolomite from all 12 wells
(Fig. 13) with a correlation coefficient of 0.98. A
distinction of calcite from dolomite is possible
with the inclusion of magnesium (Hertzog et al.
1987; Roscoe et al. 1995).

Estimating quartz-feldspar-mica
The third component of the new lithological
description is the sand fraction composed
primarily of quartz, feldspars and micas ( Q - F M). This fraction is determined by subtracting
the clay and carbonate fractions from 100 wt%.
Figure 14 shows the estimated and measured
concentrations of clay, carbonate, and quartzfeldspar-mica for all 12 wells. In the reservoir
rocks, where clay content is less than 30 wt%,
the agreement between measured and estimated
concentrations is remarkably good for all components. In the shales, particularly where clay
exceeds 50 wt%, the interpretation tends to
under-estimate clay and over-estimate Q - F - M .
Obviously, the clay algorithm could be optimized
to give more accurate estimates in shales. The
carbonate estimates are good over the entire
dynamic range.

Estimating anhydrite
This three component lithological description is
easily modified to accommodate formations
containing significant amounts of anhydrite or
gypsum. The anhydrite estimate precedes the
carbonate estimate to separate carbonate calcium from anhydrite calcium. Two estimates of
anhydrite are made, one from sulfur and one
from calcium, according to stoichiometric relationships where the sulfur concentration in
anhydrite is 23.55 wt. % and the calcium
concentration is 29.44 wt.%.

Anhl = S/23.55

(7)

Anh2 = Ca/29.44.

(8)

The final anhydrite estimate is the minimum of


these two to account for the possibility of nonanhydrite sources of either sulfur or calcium. The
anhydrite computation precedes the carbonate
and clay estimate's and the anhydrite calcium is
subtracted from the total calcium prior to the
other lithological computations. When solving
for anhydrite, the Q - F - M fraction is determined
by subtracting the clay, carbonate, and anhydrite
fractions from 100 wt%. Fig. 15 presents a
comparison of anhydrite measured by FT-IR
and anhydrite using calcium and sulfur from a
single well in West Texas.
40
35
30
E
~=25
(3.

(-920
+

e15
-E

"O

~r 1 0
<
5
I

10
20
30
Estimated Anhydrite wt%

40

Fig. 15. Comparison of estimated and measured


quantity of anhydrite on a single dataset.

Application to log data


The ultimate goal of this study is to identify an
objective, robust, and efficient means of estimating lithology from spectroscopy logs. The two
simultaneous developments that have made this
possible are the determination of elemental

QUANTITATIVE LITHOLOGY
concentrations from induced gamma ray spectroscopy logs and the derivation of the lithology
algorithms presented above.
To apply these relationships using the data
from Figs 1 and 2 requires that the clay
algorithms be modified to account for the known
aluminium interference in the iron measurement.
Equations (3), (4) and (5) for computing clay or
clay plus mica become:
ClayL = 1.91(100 -- SiO2 - CaCO3 - 1.99FeA1) (9)
Clay + MicaL = 2.43(100

-- SiO2 -

(10)

CaCO3-1.99FeA1)

Clay2L = -- 18.5 + 3.34(100-SiO2-- CaCO3 - 1.99FeA1)

(11)

where the L subscript designates the application


to log data. FeA1 designates the quantity that
would be detected as iron by a spectroscopy
device and is equal to Fe + 0.14A1.
The clay, carbonate and Q - F - M fractions
calculated using the Fig. 1 open hole spectroscopy data from Well 8 are presented in Fig. 16.
Also shown are the core clay, carbonate and Q F - M fractions determined from the F T A R
mineralogy. The agreement between core and
log data is quite good, in spite of the fact that

93

Well 8 is probably the worst example of the A1clay relationship.


The interpretation of the cased hole spectroscopy logs from Well 3 (Fig. 2) is presented in
Fig. 17. The agreement between core and log
data is quite spectacular considering that these
measurements are made with a ll~in, diameter
tool through casing and cement.
Conclusions

The quantitative lithology presented here has


been optimized on core data from numerous
wells from around the world. The lithological
fractions of clay, carbonate, anhydrite, and
quartz-feldspar-mica are ideally suited for the
elemental concentration logs of silicon, calcium,
iron, and sulfur, which can be acquired by single,
induced gamma ray spectroscopy logs. These
elemental concentration logs could be available
in both open and cased hole. The strength of this
elemental approach to estimating lithology lies in
the use of major element chemistry as opposed to
trace element chemistry which can be so easily
impacted by sediment diagenesis, depositional
environment, or the spurious introduction of
small amounts of heavy minerals. The elements
used are major element contributors to the rockforming minerals. Their concentrations in a
given mineral are relatively stable, and the

20("

40(

=......
w

60(

80(
l_
t-t

100(

120(

F"-"

1400

160C
0

50

Clay, wt%

1 O0

50

Carbonate, wt%

1 O0

50

Quartz-Feld-Mica, wt%

1 O0

Fig. 16. Quantitative lithology logs for Well 8 using the openhole elemental concentration logs shown in Fig. 1.
FT-IR core measurements are provided for comparison.

94

M.M. HERRON & S. L. HERRON


x 10 4
1.01

!
w

1.03

[1.05

B
i

L
1.07

50
Clay, wt%

1 O0

~:"
0

i
50
Carbonate, wt%

O0

50
1 O0
Quartz-Feld-Mica, wt%

Fig. 17. Quantitative lithology logs for Well 3 using the cased hole elemental concentration logs shown in Fig. 1.
FT-IR core measurements are provided for comparison.
minerals in which they occur are generally
abundant.
The S i - C a - F e aluminium emulator gives a
demonstrably superior clay interpretation compared to that available from gamma ray. Its
strength lies in the near constant slope, small
degree of scatter, and near zero intercept. It is
also independent of fluid volume, type and
density, rendering it free from gas or light
hydrocarbon effects, unlike the neutron-density
separation.
The calcium log provides an unparalleled
carbonate estimation. It provides carbonate
quantification in complex lithologies. In heavy
barite muds, it easily and accurately locates
carbonate cementation at levels of 10 to 20 wt%
which were previously undetected by conventional log interpretation. The sulfur log provides
a very accurate estimate of anhydrite which is of
greatest value in carbonate/evaporate lithologies.
While the relationships presented here have
demonstrated a large degree of universality, each
algorithm can be further optimized on a field or
regional basis to give improved lithological
estimates.

References
BHVVAN, K. & PASSEY, Q. R. 1994. Clay estimation
from GR and neutron~tensity porosity logs,

paper DDD. In: 35th Annual Logging Symposium


Transactions: Society of Professional Well Log
Analysts, pp. D1 15.
ELLIs, D. V. 1987. Well Logging for Earth Scientists.
Elsevier, New York.
GRAU, J. A. & SCHWEITZER, J. S. 1989. Elemental
concentrations from thermal neutron capture
gamma-ray spectra in geological formations.
Nuclear Geophysics, 3, 1 9.
--,
ELLIS, D. V. & HERTZOa, R. C. 1989.
A geological model for gamma-ray spectroscopy
logging measurements. Nuclear Geophysics, 3,
351-359.
HERRON, S. L. 1995. Method and apparatus for
determining elemental concentrations for "/ ray
spectroscopy tools, U.S. Patent 5,471,057.
& HERRON,M. M. 1996. Quantitative lithology:
An application for open and eased hole spectroscopy. In. 37th Annual Logging Symposium
Transactions: Society of Professional Well Log
Analysts, pp. E1 14.
HERTZOG,R. C., COLSON,L., SEEMAN,B., O'BRIEN,M.,
SCOTT, H., McKEoN, D., WRA~GHT,P., GRAU, L,
ELLlS, D., SCHWEITZER, J. & HERRON, M. 1987.
Geochemical logging with spectrometry tools,
SPE-16792. In. 62nd Annual Technical Conference and Exhibition Proceedings: Society of
Petroleum Engineers.
HURST,A. & MILODOWSK1,T. 1994. Characterization of
clays in sandstones: Thorium content and spectral
log data, paper S. In: Sixteenth European Formation Evaluation Symposium: Society of Professional Well Log Analysts.
-

QUANTITATIVE LITHOLOGY
KATAHARA, K. W. 1995. Gamma ray log response in
shaley sands. The Log Analyst, 36, 50--55.
MATTESON, A. & HERRON, M. M. 1993. Quantitative
mineral analysis by Fourier transform infrared
spectroscopy, Society of Core Analysts Technical
Conference, August 9 11, 1993, SCA 9308.
RoscoE, B., GRAU, L, CAO MINH, C. (~; FREEMAN, D.
1995 Non-conventional applications of through-

95

tubing carbon-oxygen logging tools, paper QQ.


In: in 34th Annual Logging Symposium Transactions: Society of Professional Well Log Analysts.
SCHWEITZER, J. S., ELLIS, D. V., GRAU, J. A. &
HERTZOG, R. C. 1988. Elemental concentrations
from gamma-ray spectroscopy logs. Nuclear Geophysics, 2, 175 181.

The comparison of core and geophysical log measurements obtained in


the Nirex investigation of the Sellafield region
A. K I N G D O N , S. F. ROGERS, C. J. E V A N S & N. R. B R E R E T O N

British Geological Survey, Keyworth, Nottingham, NG12 5GG, UK

Abstract: The Sellafield region, west Cumbria, is the focus of one of the most thorough
geological investigations in the United Kingdom. The Sellafield Site is defined as an area
immediately around the potential repository, extending 6.5 km north-south by 8 km eastwest. Twenty six deep boreholes were drilled within the area up to the end of 1995, with a
total depth of approximately 28 km. Most of these boreholes have been continuously cored,
a total of over 17 kilometres of core, with average core recovery well in excess of 90%. All
boreholes were logged with a comprehensive suite of geophysical logs, including many state
of the art tools. Laboratory physical property analysis of hundreds of sample cores has been
carried out.
Pilot studies were carried out to compare and contrast datasets and to investigate the
relationships between the different data scales. Various techniques, including fractal analysis
and Artificial Neural Networks, were tried in order to explore the relationships of these data
at a variety of measurement scales.
The pilot study was conducted in two stages:
(1) evaluation of the primary controlling factors of the physical properties;
(2) testing the validity of 'Up-scaling'.
The rocks of the Borrowdale Volcanic Group provided the most challenging problems
due to the physical properties being dominated by fracturing and associated alteration
zones.
Relationships between data types at different scales were established suggesting that the
extrapolation of properties derived from core and wireline logs across three-dimensional
seismic grids would allow an understanding of the properties throughout a threedimensional volume.
Nirex is responsible for the development of a
deep geological repository for solid, intermediate level and some low-level radioactive wastes.
Following preliminary geological investigations
of two sites, an area near Sellafield, west
Cumbria, was chosen in 1991 for further study.
The Nirex science programme aimed to assess
the suitability of the Sellafield site as the host for
the repository. Such an assessment required,
among other things, an understanding of the
geology and hydrogeological characteristics of
the area.
The Sellafield region in west Cumbria, England was the focus of one of the most detailed
site investigations projects ever undertaken. This
investigation aimed to characterize the geology
and hydrogeology of the site to determine
whether the site at Sellafield showed sufficient
promise of meeting regulatory targets to permit
Nirex to submit a planning application for a
deep repository. An underground Rock Characterization Facility (RCF) had been proposed
in order to allow more detailed characterization
of the geology and hydrogeology of the area
using direct observations from underground

excavations and to allow in situ experiments


on rock and groundwater behaviour. These
measurements were required to provide information on ground conditions that could only be
obtained from an underground facility and to
test models of the geology, hydrogeology and
geotechnical characteristics and behaviour of the
rocks.
In the course of the Sellafield site investigations, data at a range of scales from microscopic
to regional have been collected. The large
volume of data available from the Nirex
investigations presents problems with respect to
the estimation of properties at the very large
scales required by performance models. In most
practical applications, the scale of the sample
measurements is not directly comparable with
the scale required for the model estimates needed
for the calculations. It is important to evaluate
the scale of the sample data and the scale
required for the final estimates and to apply
some correction to the sample scale, if they are
different. This corrective process is generally
termed 'up-scaling'. In the context of Sellafield,
physical property parameters important to the

KINGDON, A., ROGERS, S. F., EVANS, C. J. & BRERETON,N. R. 1998. The comparison of core and
geophysical log measurements obtained in the Nirex investigation of the Sellafield region.
In: HARVEY,P. K. & LOVELL,M. A. (eds) Core-Log Integration, Geological Society, London,
Special Publications, 136, 97-113

97

98

A. KINGDON E T AL.

Fig. 1. Location of the Sellafield boreholes and the potential repository zone.

construction of underground vaults required, on


the scale of tens of metres may only be measured
on core samples at the scale of centimetres or
from geophysical logging of the boreholes at a
scale of a few metres.
The difficulty of extrapolating properties
using data from varying scales makes it difficult
to use data derived at one scale, for example
borehole core, to another, such as a threedimensional seismic survey. In the context of the
Sellafield investigations, parameters that are
important to tunnelling, such as indices of rock
strength need to be derived at one scale and then
extrapolated to another scale. Techniques to
allow this to be undertaken must have therefore
to be both derived and tested.
This paper examines the possible techniques
for comparing data derived at three separate
scales: borehole core (centimetre scale), geophysical borehole logs (metre scale) and threedimensional seismic survey data (10 metre scale).
In particular, mathematical techniques were
studied that examine relationships between data
scales; thus demonstrating the validity of the
methodology of 'up-scaling'.
This study largely concentrated on the Potential Repository Zone (PRZ) an approximately
four square kilometre area near the village of

Gosforth. The key task of this study was to


define an index of rock properties derived from
geophysical log measurements down each of the
boreholes in the PRZ. This index was then used
to extrapolate those properties across a volume,
as sampled by the three-dimensional seismic
survey, aiming to allow prediction of rock
properties at any location within that volume.

Location and geological setting


The Sellafield site is in west Cumbria, England
and situated between the coast of the East Irish
Sea and the Lake District National Park. A map
of the area is shown as Fig. 1. Up to the end of
1995 twenty-six boreholes were drilled within the
area as part of the site investigation. The
geology from each borehole has been fully
described ( Nirex 1993; 1995a,b).
The regional basement in the Sellafield area is
the Borrowdale Volcanic Group (BVG) which
consists of a complex group of Ordovician tufts,
lapilli tufts and acidic lavas with local intermediate and basic intrusions, and volcaniclastic
sediments. (Millward et al. 1994). The BVG was
deposited as a largely sub-aerial volcanic system
formed by an island arc on the southern margin
of the Iapetus Ocean. The BVG is unconform-

THE NIREX INVESTIGATION OF THE SELLAFIELD REGION


ably overlain by a south-westerly thickening
Carboniferous Limestone and Permo-Triassic
succession. The Permo-Triassic forms the subcrop across most of the Sellafield area.
The BVG and (where present) the Carboniferous Limestone are unconformably overlain
by the Brockram, a Permian fluvial breccia
conglomerate with up to cobble sized clasts. The
upper part of the Brockram near the coast
passes laterally into the St Bees Shale and
Evaporite (Nirex 1993). The St Bees Evaporite
comprises dolomite and anhydrite, and the St
Bees Shale is a laminated sandstone, siltstone
and claystone formation. The St Bees Shale is
conformably overlain by the dominantly fluvial
St Bees Sandstone (Barnes et al. 1994) of
Triassic age. This is a sandstone and claystone,
with the claystone increasing markedly down
succession, particularly in the North Head
Member at the base. The St Bees Sandstone is
overlain by the Triassic aeolian Calder Sandstone which forms the subcrop in the PRZ area.
The easternmost of the Sellafield boreholes
(9A and 9B) were drilled into outcropping BVG
with the Permo-Triassic succession outcropping
to the southwest of these boreholes. This
succession thickens towards the southwest
reaching a thickness of 1700m at the Irish Sea
coast. In the PRZ area there is 400 to 500 m of
sedimentary cover overlying the basement.
The Permo-Triassic succession occurs on the
eastern margin of the East Irish Sea Basin, an
extensional basin associated with prolonged
east-west extension resulting in the dominantly
north-south faulting seen today (Jackson et al.,
1995).

Data sources
High quality geological and geophysical data
have been acquired across the Sellafield region
during the site investigation. All twenty-six
boreholes have been geophysically logged using
comprehensive suites of state of the art tools,
including borehole imaging. In addition, the
boreholes have been extensively cored, allowing
continuous detailed geological and discontinuity
description to be undertaken. Detailed gravity
and magnetic survey data has been acquired
across the Sellafield region, as well as twodimensional seismic data.
The geology of the PRZ area has been the
subject of a highly detailed investigation. Up to
the end of 1995 eleven boreholes (Boreholes 2, 4
& 5; RCF1, 2 & 3, RCM1, 2 & 3; PRZ2 & 3)
were drilled within an area measuring only
1200 m by 800 m across the ground surface. All
penetrated to the BVG, the deepest borehole

99

penetrated to 1600m below ground level. All


boreholes have been cored from within the St
Bees Sandstone succession to terminal depth
within the BVG, with total core recovery in
excess of 95% (close to 100% in some of the
later boreholes). In addition, a high quality trial
three-dimensional seismic survey has been acquired across part of the PRZ area. Data
acquired within the PRZ area therefore allows
particular scope for both deriving detailed rock
properties and up-scaling between different
datasets.

Deriving characteristic rock properties


The first stage of this project involved the
derivation of the average rock properties in each
borehole, for each of the major formations in the
Sellafield area. These average properties were
then used to determine whether a particular
formation was essentially constant across the
area or whether there were significant regional
and/or local variations in the rock properties.
Where significant variations in rock properties
were found the possible causes for the variation
were examined. This exercise was first carried
out on a regional scale and then concentrated in
more detail upon the PRZ area.
Rock properties were studied by comparison
of geophysical borehole logs from across the
area. Geophysical logs were chosen because of
the consistent way that the data was acquired,
both in terms of techniques and sampling rates.
This allows for easy comparison between boreholes some distance apart. The main characteristics of the rock properties studied were
identified by statistical and graphical techniques
of data comparison.

Stud), o f velocity
Although many rock properties have been
measured at various scales, compressional velocity is one of the few to have been measured at
all scales, from core to seismic scale. It was
therefore chosen as the most representative
property for analysis as an example of the
average rock property behaviour. The compressional velocity of a rock formation is controlled
by the matrix density, the porosity and the fluid
composition.
Compressional velocity data for each of the
three data scales were derived by different
techniques. Core scale data for each of the main
rock types were provided by laboratory testing
on core samples. Wireline log scale data were
derived from sonic velocity logging. Larger scale

100

A. KINGDON ET AL.

Fig. 2. Percentage frequency histogram of bulk compressional velocity for the main stratigraphic units of the
PRZ.

data were derived directly from the two way


transit velocities from seismic survey information.
Figure 2 shows a frequency histogram of bulk
compressional velocity. This shows the velocity
distributions for the three rock types present in
this area (the St Bees Sandstone, Brockram and
BVG) in the PRZ. The statistics are derived
from the total geophysical logging measurements in each borehole from within the area.
The mean velocity of the St Bees Sandstone is
shown to be between 3.5 to 4.5 kms -l, the
Brockram between 4.5 to 5.5 kms -1 and the
BVG between 6.0 to 7.0 km s 1.
Variations in velocity within the range for the
individual rock type are caused by differences in
the properties of the materials. This study aimed
to identify these differences and to attempt to
understand their origin.
More detailed studies of the nature of the
velocity distributions have been made for two
rock types: the St Bees Sandstone and the BVG.
This was done initially on a regional scale and
subsequently more locally in the PRZ area. The
Brockram is of a fairly consistent thickness
across the Sellafield Region (approximately
100m) and shows remarkably homogeneous
properties. As a consequence, no further attempt
has been made to characterize variation in rock
properties for this lithology.

Regional studies of velocity


The regional pattern of velocity was studied
using graphs of midpoint depth against mean
formation velocity for each borehole. The
midpoint depth of a formation is defined as the
point equidistant between the top and base of
the sampled section of a formation, regardless of
whether the borehole had sampled the entire
thickness of the formation. This allows comparison of the compressional velocities between
boreholes without any overprinting of the effects
of velocity changes within the formation. The
mean velocities have been derived from geophysical logs of the formation and are expressed on
the graph as kilometres per second.
Figure 3 is a graph of midpoint depth against
velocity for the St Bees Sandstone in each of the
boreholes in the Sellafield region where the
formation is present. Figure 4 shows the same
data types for the BVG. Boreholes are shown
using different symbols depending on whether
they fall inside or outside the boundaries of the
PRZ area. Also plotted on the graph are typical
core sample data that have been pressurized
under laboratory conditions to simulated
depths.

Results of studies of velocity against midpoint


depths. Figure 3 shows that mean compressional

THE NIREX INVESTIGATION OF THE SELLAFIELD REGION

Fig. 3. A graph of midpoint depth against compressional velocity for the St Bees Sandstone.

Fig. 4. A graph of midpoint depth against velocity for the Borrowdale Volcanics Group.

101

102

A. KINGDON ET AL.

velocity of the St Bees Sandstone increases with


midpoint depth in a smooth curve. This exponential increase in velocity with depth is well
documented as being caused by decreasing
effective porosity due to increased overburden
pressure (Birch 1960). The results from the core
samples also demonstrate that increasing depth
results in an increase in compressional velocity.
The velocity results from the core samples is
somewhat lower than the velocities from equivalent bulk depths derived from geophysical logs
in the boreholes. This is a consequence of the
samples being tested whilst dry (i.e. the pore space
was air filled rather than saturated with water).
Figure 4 shows a plot of midpoint depth
against compressional velocity for those boreholes in the Sellafield region where the BVG is
present. This plot shows a relatively complex
pattern in comparison with the St Bees Sandstone.
Most boreholes show a mean velocity over the
BVG interval of between 5.0 to 5.5 km s l.The
boreholes in the PRZ area (with the exception of
borehole PRZ3) showing a distinct clustering of
mean velocities. Three boreholes 9B, 8B and
PRZ3 show significantly lower mean velocities.
The first two of these boreholes penetrate a short
distance into the BVG. Detailed logging of the
Borehole 8B core (Nirex 1995a) suggested that
the top section of the BVG is highly altered,
which was the only part of the BVG sampled in
this borehole. In the case of Borehole 9B, which
was drilled where the BVG outcrops at surface,
the rocks will have been affected by recent
weathering. As these effects will have had a
significant influence on rock properties of these
two boreholes, the data are not comparable with
the other boreholes and they have not therefore
been included in the study of the average rock
properties of the deep BVG. In the case of
borehole PRZ3, where the BVG is covered by a
thick Permo-Triassic succession, the difference
could not be readily explained in this way and
the BVG velocities from this borehole are
therefore seen to be significantly anomalous.
An independent quality assurance check of the
geophysical logging of Borehole PRZ3 did not
indicate any systematic error in the acquisition
and processing stages. Borehole PRZ3 was
targeted to intersect a fault, Fault F1, in the
BVG and it is likely that the mean velocity
results from this borehole reflect a large proportion of 'faulted rock'.
The velocities of the available core samples in
the BVG are slightly higher than the wireline
derived mean formation velocities, despite the
core samples again having been tested whilst
dry. This is inferred to be due to the porosity of

the material being sufficiently small that the


compaction effect is not as significant. Of greater
significance to the properties of the core samples
is the fact that core tests were by definition
carried out on samples of intact rock. The
properties of these samples therefore varies from
the bulk rock sampled by geophysical logs,
which includes the effects of non-intact and
fractured rock. This suggests that the variations
of the bulk rock properties from those of intact
rock may be a consequence of the discontinuities
within the rock mass.

Local studies gf velocity


More detailed analysis of the bulk rock velocity
properties for the boreholes in the PRZ region
was carried out using box and whisker plots.
Box and whisker plots (Figs 5, 6 and 7 and 11)
are a graphical technique which permits an
overview of a complete data distribution,
excluding only anomalous data at the extremes
of the distribution. This permits the statistical
comparison of almost the entire data distribution between all the boreholes. The variation in
the range of data as well as average properties
can therefore be assessed. Figure 5 shows the
symbols used to describe different parts of the
data distribution on box and whisker plots.
Figure 6 is a box and whisker plot of the
velocity distributions of the St Bees Sandstone
for the boreholes in the PRZ. The column on the
right-hand side of this diagram shows the
combined properties for all the boreholes. The
quartile range of the velocity for all the PRZ
boreholes is between 3.6 and 4.4 kms 1. The
bulk rock properties from all boreholes are
essentially consistent between all the PRZ boreholes. This indicates that there is little regional
variation in the bulk rock properties of the St
Bees Sandstone in this area.
Two additional datasets are also displayed on
this plot, the core properties and the faulted rock
properties. Core properties were derived from
laboratory tests on intact core samples. Laboratory testing of the properties of the fault rock
was difficult as fault rock is by definition nonintact. The fault properties shown here were
therefore derived from geophysical log measurements. The standard geophysical log measuring
increment of 6 inches (15.24cm) means that
statistically valid samples are hard to obtain
from faults with limited borehole intersections.
Therefore properties were only derived for faults
with borehole intersections greater than 50
centimetres.
The core sample seismic velocities are seen to
be much lower (interquartile range for all

THE NIREX INVESTIGATION OF THE SELLAFIELD REGION


NUMBER OF POINTS

NUMBER OF POINTS

103

NUMBER (')F" POIN'IS

95%

UPPER
QUARTILE

UPPER
QUARTILE

UPPER
QUARTILE

MEDIAN

MEDIAN

MEDIAN

MEAN

MEAN

MEAN

LOWER
QUARTILE

LOWER
QUARTILE

LOWER
QUARTILE

5%

FAULT PROPERTIES

BULK PROPERTIES

CORE PROPERTIES

Fig. 5. Key to the symbols used on a box and whisker plot.

5.0
1596

1378

?-7 10

1647

22

2411

2506

75

2293

2428

1627

18

2645

2605

2475

23611

121

59

4.5-

~'~ 4.0--

r~

tI

3.5-

3.0--

I ,.,

I ,.,

I -

I ~

i ~

i .~

i .~, i ,~

i ~.~ i ~

i ,0,, i ~.

BOREHOLE NAMES

Fig. 6. Box and whisker plot of compressional velocity for the St Bees Sandstone Group.
boreholes 3.4 to 3.8 k m s -1) than those of the
bulk rock as they were tested on dry samples (see
above). The fault data are also somewhat lower
(interquartile range for all boreholes 3.3 to 3.9
k m s 1). Whilst the interpretation of these two

datasets was hampered by the small number of


sample points and the sampling bias, it is clear
that the properties of fault rock are significantly
different to the bulk rock. This can be seen in the
'total column' which includes the summed data

104

A. KINGDON E T AL.
7.0

7238
85 t l i

5482
378 76

4950
104 75

168

389
27

3878
77

3@66
118

2911
23

14188
43

1480
28

3046
119

6.0

5.0"

4.0'

34896
1002 262

tt
ti

3.0

BOREHOLE NAMES

Fig. 7. Box and whisker plot of compressional velocity for the Borrowdale Volcanic Group.

Fig. 8. Cross-plot of density against neutron porosity fo the St Bees Sandstone Group showing depth as the z-axis.

THE NIREX INVESTIGATION OF THE SELLAFIELD REGION


for all boreholes.
As can be seen in Figure 7 most boreholes in
the PRZ area shows a consistent set of bulk rock
velocity (interquartile ranges between 5.0 to 6.1
kms-1). Only borehole PRZ3 displays significantly different properties with lower values
(interquartile range 4.5 to 5.0 kms-l). Comparison of the three measurements for each borehole provided important evidence to the
controlling mechanism for bulk rock properties.
Whilst core derived intact rock properties
showed somewhat higher velocities than the
bulk rock properties in almost all cases, the fault
derived velocity values were significantly lower
than the equivalent bulk rock properties and
showed greater variability.
The difference between bulk rock and intact
rock is, by definition, the discontinuities of
which fault rock was the only measurable
example. Hence the discontinuities must be the
dominant controlling factor on the bulk rock
properties of the BVG. In the case of borehole
PRZ3 the velocity measurements showed a very
localized anomaly, consistent with the targeting
of the borehole into a faulted zone.

Causes of variataions of rock properties in


the St Bees Sandstone
The compressional velocity distribution of the
boreholes described above shows that the

105

velocity profile of the St Bees Sandstone largely


reflects the depth of burial and there is therefore
an increase in average velocity to the south-west
where the midpoint depth of the formation is
greater.
Figure 8 shows a typical cross plot of
percentage porosity against density of the St
Bees Sandstone from Borehole 2. The depth of
each point is displayed as the z-axis (the
colouration of the points in Fig. 8). The diagram
shows that most of the data plots along a line,
with density increasing as porosity decreases and
depth increases. This distribution is caused by
compaction (due to increasing overburden pressure) leading to decreasing porosity with depth.
Some variations from this simple distribution
are seen. In order to understand the anomalous
points, the effects of variations in lithology had
to be quantified. Figure 9 shows the densityporosity cross plot for the same data but with
the z-axis now showing the gamma-ray count
from each point. This clearly shows the effects of
lithological change as higher gamma counts
occur in those parts of the distribution which
do not follow the simple trend. This is because
these zones are not clean sandstone but contain
a significant proportion of claystone.
These two diagrams therefore indicate that the
dominant control on the bulk rock properties of
the St Bees Sandstone was depth of burial and
lithological variation.

Fig. 9. Cross plot of density against neutron porosity for the St Bees Sandstone Group showing gamma ray as the
z-axis.

106

A. KINGDON E T AL.

Fig. 10. Cross plot of shallow resistivity against neutron porosity for the Borrowdale Volcanic Group.

Fig. 11. Box and whisker plot of gamma ray for the Borrowdale Volcanic Group.

THE NIREX INVESTIGATION OF THE SELLAFIELD REGION

107

Causes of variations of r o c k properties in the


Borrowdale Volcanic Group
Figure 8 showed that whilst the BVG showed
fairly homogeneous properties between boreholes across most of the P R Z area, there is
significant variation of properties within each
borehole. The nature of these variations was
therefore investigated further using an example
borehole.
Figure 10 shows a neutron porosity against
shallow resistivity cross plot with gamma ray
counts as the z-axis. Two main clusters are seen
on this graph. The bulk of data points are shown
to be highly resistive, low porosity with a low
gamma ray count. These values represent the
properties of the intact rock. In addition a
smaller cluster of points with low resistivity,
higher porosity and high gamma ray counts are
also seen. The higher gamma ray counts are
probably caused by the alteration minerals
found around discontinuities within the rock
mass. This data acted as a further indication that
the dominant control over the bulk rock properties of the BVG are the nature of the discontinuities within the rock mass rather than the
intact properties of the rock itself.
In an attempt to quantify the effects of the
properties of the faults on the bulk rock
properties a box and whisker plot of the gamma
ray counts for the BVG of the P R Z boreholes
was produced (Fig. 11). In most of the boreholes
with a statistically significant sample of fault
rock, the gamma ray response is approximately
5 API higher in the fault rock than in the bulk
rock. This can be seen most clearly in the total
(all boreholes) column on the right-hand side of
the diagram.
This is not an ideal presentational medium
because a fault in an already low gamma ray
formation may have a lower count than high
gamma background elsewhere in the borehole.
Variations in the condition of the boreholes will
also significantly affect the results. The data for
Borehole RCM3 is dominated by a single large
fault close to the top of the BVG where low
gamma is recorded because of poor hole
conditions (the caliper increases from 6 to 16
inches through this fault). Despite these problems the diagram does clearly show that
gamma ray counts are higher in fault zones than
in bulk rock.

Data scales and up-scaling


Relating physical properties derived from one
scale to another represents a significant problem.
Prediction of the rock properties through a given

Fig. 12. A diagrammatic fractal distribution: the


Sierpinski Gasket.

volume, such as those sampled by the trial 3-D


seismic survey are important for successful
engineering design of tunnels, shafts etc. The
P R Z area has eleven deep boreholes drilled by
Nirex, and several drilled previously by British
Steel and its precursors, and yet less than one
five millionths of the total volume of the P R Z
has been directly sampled by coring.
In order to scientifically justify the up-scaling
of known properties (derived from either direct
measurements on core or indirect measurements
from geophysical logging tools) it is important
to demonstrate that at least over a small but
significant part of the scale, properties are
comparable. If this can be done then 'up-scaling'
of data can be seen as a legitimate concept,
although it should be treated with caution.

Fractals
A true fractal relationship is a relationship
between two variables that does not change
with scale. Whilst it is unlikely that any relationship is truly fractal in a natural system, if a
relationship could be demonstrated over a
number of orders of magnitude then this could
be used to justify up-scaling of data from one
scale to another. Figure 12 shows a diagrammatic fractal relationship, the Sierpinski Gasket.
Each size of triangles is related to the next
largest and next smallest size of triangles by the
same scale and geometric relationships, up to the
limits of page size in one extreme and print
resolution in the other.
An attempt was made to study discontinuities
in the Sellafield area at two separate scales:
distances between individual discontinuities
measured directly from the core and distances
between seismically resolved faults. This was

108

A. KINGDON ET AL.

Fig. 13. Log-log plot showing fractal distribution of borehole discontinuities in borehole 8B.

done firstly to assess whether fracture distributions were fractal at each scale and then to see
whether any link between data at the two scales
could be established.

Discontinuity separation
Various techniques have been derived to study
the fractal dimension for a distribution of
naturally occurring phenomena. This study was
carried out using the Spacing Population Technique (after Harris et al. 1991) which is both
straight forward and applicable to the type of
data to be examined.
The basic dataset for this study was the
borehole discontinuity log, produced for Nirex
by Gibb Deep Geology Group (GDGG) from
direct measurement of the core. This lists, for
each borehole, all the occurrences of faults,
veins, joints and other discontinuities, ordered
by depth. All discontinuities with a non-structural origin were removed, such as bedding
features, stylolites in the Carboniferous Limestone and those fractures in the core that were
induced by the drilling process.

Methodology
The Spacing Population Technique is based on
cumulative frequency distributions derived incrementally from large (infrequent) to small

(frequent) events. In this case, classes of 0.1 m to


100m were used as applicable to borehole
discontinuity spacing, covering three full log
cycles. In order to analyse the data an approximate geometric progression was used to divide
the data up into frequency classes suitable for
log-log output.
The cumulative frequency data wwere plotted
as a log-log plot. To be considered fractal the
distribution had to plot as a straight line
(showing that the relationship is scale invariant
and conforms to the following function):

y=ax - D
where: y = probability (cumulative frequency);
a = a prefactor;
x = the discontinuity spacing;
D = t h e line gradient (i.e. the fractal
dimension).
For a distribution to be considered fractal the
data should be linear across at least one order of
magnitude.

Results of fractal analysis


Figure 13 shows a cumulative frequency log-log
plot for Borehole 8B, which gives the results of
the fractal analysis of borehole discontinuities in
this borehole. Some of the limitations of the
fractal technique are demonstrated by this
dataset. Whilst natural fractals should be proved

THE NIREX INVESTIGATION OF THE SELLAFIELD REGION

109

Fig. 14. Log-log plot showing fractal modelling of seismic scale faults for the base Carboniferous.

to exist in any relationship over several orders of


magnitude, any single measurement technique
may only characterize a subset of this total
range. The limits for identification of fractal
patterns are often therefore controlled by
limitations in the sampling method. In this study
for instance the core fracture dataset was reliant
on the human eye to identify individual fractures.
Very closely spaced fractures lead inevitably
to highly broken core and poor core recovery, so
that such zones are preferentially undersampled. Rock, heavily fractured by localized,
closely spaced events is essentially indistinguishable from large fractures and will behave in a
similar way. Major fault systems have not been
sufficiently sampled by boreholes to permit
fractal analysis at this scale, whereas the threedimensional seismic survey only identifies faults
either by 'significant' offset of marker lithological contacts or directly where the thickness of
fault-rock is sufficient to cause a velocity
contrast. Also the Sellafield Site boreholes
cannot be labelled a random unbiased dataset,
as some of the boreholes were specifically
targeted at some of the major faults in the
region.
The borehole 8B fracture set shows a clear
fractal relationship over two orders of magnitude, i.e. at fracture spacings from 20cm to
10m, with a regression coefficient of
R 2= 0.9996, very close to a perfect straight line.

Fractal events at the seismic scale


Large events, such as major faults, although

sampled by boreholes, occur only infrequently


and the dataset is statistically insignificant in any
one borehole. Therefore another measurement
technique must be used to sample the larger
faults. An obvious alternative method is to
examine faults identified from a two-dimensional seismic grid. This dataset was used in this
study for a comparison with the borehole
derived results. It was important to make clear
at this stage that these were not identical
datasets simply measured at different scales.
Seismic reflection profiles (and in particular
widely spaced two-dimensional seismic data)
tend to resolve only large scale fault zones
rather than distinguishing individual minor fault
strands such as those which would be delineated
from borehole core.

Methodology
In order to get an acceptable level of coverage of
fault features with a common resolution, offshore seismic reflection data from near the
Sellafield site were used for this study. Unlike
boreholes, which essentially sample a onedimensional environment, the interpreted seismic fault maps used in this part of the study
were two-dimensional in character. The dataset
in this case were fault maps stored in a database
of faults derived from V U L C A N software
modelling of the regional structure.
A different sampling technique was used in
order to develop cumulative frequency data. In
this case a two-dimensional 'box counting'
method was applied. This was done by overlaying the maps to be studied with grids of

110

A. KINGDON ET AL.

Fig. 15. Log-log plot showing fractal modelling of seismic scale faults for the base permo-trials.

square boxes. These boxes each had sides of


length d and the number of boxes containing
fault features was counted (given as Nd). The
exercise was repeated several times with boxes of
progressively shorter side length d (i.e. the grids
become finer). The number of filled boxes (Na)
was plotted on a log-log plot against box size
dimension (d). If the relationship between the
two variables is fractal it produces a straight line
of gradient - D , which should be in the range
1.0 < D < 2.0 (Hirata 1989)

Results of fractal modelling of seismic scale

faults
Two different seismic base maps of the East Irish
Sea basin were studied (Nirex 1995b); The base
Carboniferous (Fig. 14) and Base Permo-Triassic (Figure 15). Both showed very clear fractal
patterns over the scale range from 200m to
1 km, with regression coefficients of R 2= 0.999
in both cases.
Fractal patterns have been demonstrated over
two different scale ranges for discontinuity
spacing events. The regression coefficient for
both sets of events was very close to one (i.e. a
completely fractal pattern). However the scaling
exponents, sometimes known as the fractal
dimension, differ suggesting that fracture spacing is a scale dependent parameter. These
relationships provide evidence that up-scaling
of discontinuity events from features measured
in core and at the seismic scale are valid within

Fig. 16. An idealized artifical neural network.

their respective scale ranges. It is also possible to


debate that if up-scaling is valid over both these
ranges there may be the possibility of establishing a relationship between the different fractal
equations and thus defining an up-scaling
function all the way from micro fracture scale
to major faulting.

Neural network modelling


Artificial Neural Networks (ANNs) are a
computer modelling technique that work in a
manner analogous to the processes of a mammalian brain. They are based on simple linear
processing elements which interact to form

THE NIREX INVESTIGATION OF THE SELLAFIELD REGION

111

Fig. 17. Results of neural network modelling for RCFl: zonation of fracturing from actual and predicted fracture
frequency.

Fig. 18. Results of neural networks modelling for RCF2: zonation of fracturing from actual and predicted
fracture frequency.

complex non-linear behaviour. A N N s can


'learn' to recognize patterns in data and develop
their own generalizations. A diagrammatic
model of an idealized artificial neural network
is shown in Fig. 16.
Fracture frequency measured from borehole
core was not easy to predict with any degree of
accuracy from conventional geophysical log
measurements. Whilst borehole imaging tools
go some way to addressing this issue, it was not
always possible to distinguish between features
such as bedding features and discontinuities.

Neural networks may allow another approach in


identifying fracture frequency.

Multi-layer preceptron
This study used a type of A N N called a multilayer preceptron (MLP) to model the relationship between core derived fracture frequency
and geophysical log measurements.
The MLP consists of a series of simple
processing elements (nodes) connected to one
another. In operation the node receives several

ll2

A. KINGDON ET AL.

inputs which it sums. The strength of the node's


response is proportional to the sum of the
inputs. Nodes are placed in layers such that
each node from one layer is connected to every
node in the next layer. These connections are
weighted and weights are changed according to
the relative importance accorded to each layer.
Input data are fed through the network and
compared with the output data. Discrepancies
between the input and output datasets result in
changes in the weighting in connections. Over a
number of iterations the network 'learns' which
inputs have the greatest effect on output. This
type of ANN, where data are fed through the
network and error fed back, is known as a feed
forward back-propagating network.

Use of artificial neural networks in this study


Artificial neural networks were used in this study
to attempt to model fracture frequency from
conventional geophysical log inputs. Fracture
frequency was derived from the borehole discontinuity logging file, binned at intervals of one
metre. The two datasets were not immediately
analogous because fracture frequency had been
calculated per metre whereas geophysical logs
conventionally sample at 0.1524m (6 inches).
Geophysical logs were filtered using a twelve
point moving average filter using BGS WELLOG software. Data were then interpolated to a
one metre value and extracted in EXCEL 5.0 for
manipulation. Six geophysical logs were used as
input for this study: density, neutron porosity,
gamma ray, shallow resistivity, compressional
velocity and shear velocity.
The fracture frequency data were also subject
to some biasing and were therefore subjected to
a five point moving average filter. These were
also exported from BGS WELLOG software to
EXCEL 5.0.
Neural network modelling for this study used
Neural Connections V1.0 software from SPSS.
Various network topologies and statistical test
were applied to validate the results. The software
selected the network topology, usually the X-3-1
layout (X nodes in the input layer, three hidden
layer nodes and a single output node).

Network training
Training was highly important to the performance of a neural network. Although it was
possible for ANNs to generalize and infer noise
obscured properties, the network response was
better where it has been trained by high quality
data. In this exercise data from boreholes in the
PRZ area were used to model the fracture

Fig. 19. Comparison of RMR and wireline logs for


borehole RCF2.

frequency values for BVG sections in boreholes


RCF1 and RCF2.
The training dataset consisted of the following
data segments 840-1015 mbRT (metres below
the rotary table datum) from RCF1, 525.5-750
mbRT from RCF2 and 732.5-932.5 mbRT from
RCF3. The data were run through a network in
its natural order and then randomized to
compare the performance of the network.

Results of neural network analysis


Figures 17 and 18 show the results of the neural
network analysis for boreholes RCF1 and
RCF2. This dataset shows clearly both the uses
and the limitations of using ANNs for modelling. Whilst in some parts of both boreholes, the
actual fracture frequency had been modelled
with some accuracy, in others the modelling had
not adequately resolved the distribution.
The actual fracture frequency was shown at
the top of both diagrams with zones of similar
levels of general fracturing marked by a black
line. The bottom diagram shows the A N N
predicted results, again with a black line marking the zones of similarity. Comparison of both
boreholes shows that the models were good at
distinguishing the background level of fracturing
in the boreholes. The biggest problems in the
models were at the data extremities. Although
this technique was not perfect it does again give
clear indications that the fracture frequency data

THE NIREX INVESTIGATION OF THE SELLAFIELD REGION


derived from core can be up-scaled to be
modelled by geophysical logs.

Comparison of RMR and conventional wireline logs


The rock mass rating (RMR) is an industry
standard rock quality and strength index derived
from direct measurement of the physical attributes of the core and is completely independent
of wireline log measurements. This measurement
allows an accurate assessment of rock strength,
but is labour intensive and therefore expensive
to collect. R M R is calculated for whole and
partial core runs and is reported for Boreholes
RCF1, RCF2 and RCF3 in 3 m intervals. Figure
19 shows the results of a comparison of R M R
with two conventional wireline log measurements from the same borehole over a given
interval. The degree of correlation between these
two sets of measurements is high despite the
wholly different derivation and supports upscaling of the wireline log data from a measurement
scale of 15cm to at least 3 m by simple
arithmetic averaging.

Conclusion
Whilst neither the fractals nor the artificial
neural network derived models showed exact
matches with the core derived data from which
they were extrapolated, both showed that there
was considerable scope for the belief that using
the correct criteria, it is possible to up-scale data
to match both wireline and seismic scale data.
The Rock Characterization Facility (RCF)
proposed at Sellafield requires detailed rock
properties to be derived from boreholes and
extrapolated across a wider area to allow for
prediction of the likely tunnelling parameters.
Where three-dimensional seismic survey data are
available across an area, it should be possible to
derive rock properties at a borehole scale and
extrapolate them across a three-dimensional
volume to give an accurate prediction of the
nature of the RCF site. This was dependent
upon a detailed knowledge of the rock properties and accurate correlation of core and seismic
properties.
The concept of up-scaling parameters derived
at one scale to another may be feasible but needs

113

considerable more research to prove valid. If


suitable algorithms can be derived, then extrapolation of geophysical parameters derived from
geophysical logs or cores, across a three-dimensional seismic grid, should allow detailed prediction of the properties for that volume of the
rock mass.

References
BARNES, R. P., AMBROSE, K., HOLLIDAY, D. W. &
JONES, N. S. 1994. Lithostratigraphic subdivision
of the Triassic Sherwood Sandstone Group in
West Cumbria. Proceedings of the Yorkshire
Geological Society, 50, 51-61.
BIRCH, F. 1960. The velocity of compressional waves in
rocks to l0 kilobars, Part 1. Journal of Geophysical Research, 65, 1083-1102.
CHADWICK,R. A., KIRBY, G. A. & BAILY,H. E. 1994.
The post-Triassic structural evolution of northwest England and adjoining parts of the East Irish
Sea. Proceedings of the Yorkshire Geological
Society, 50, 91-103.
CHAeLOW, R. 1996. The geology and hydrogeology of
Sellafield : an overview. Proceedings of the
NIREX seminar, 11 May 1994. Quarterly Journal
of Engineering Geology, 29, Supplement 1.
HARRIS, C., FRANSSEN, R. & LOOSVELD, R. 1991.
Fractal analysis of fractures in rocks: the Cantors
dust method-comment, Tectonophysics, 198, 189197.
HIRATA,T. 1989. Fractal Dimension of fault systems in
Japan: fractal structure in rock fracture geometry
at various scales. Journal of Geophysical Research.
94, 7507-7514
JACKSON,D. I., JACKSON,A. A., EVANS,D., WINGEIELD,
R. T. R., BARNES,R. P. & ARTHUR, M. J. 1995.
United Kingdom offshore regional report. the
geology of the Irish Sea. British Geological
Survey.
MILLWARD, O., BEDDOE-StEPHENS, B., WILLIAMSON, I.
T., YOUNG, S. R. & PETTERSON, M. G. 1994.
Lithostratigraphy of a concealed caldera-related
ignimbrite sequence within the Borrowdale Volcanic Group of west Cumbria, Proceedings of the
Yorkshire Geological Society, 50, 25-36.
NIREX, 1993. The Geology and hydrogeology of the
Sellafield area, Volume 1: The Geology. Nirex
report 524.
NIREX, 1995a. The Geology of the Sellafield Boreholes
Nos. 8A and 8B. Nirex report 638.
NIREX, 1995b. Sellafield geological and hydrogeological
investigations. Factual report-compilation of maps
and drawings, Volume 1 of 2. Nirex report SA/95/
02.

Forward modelling of the physical properties of oceanic sediments:


constraints from core and logs, with palaeoclimatic implications
C. LAUER-LEREDDE, 1'2, P. A. PEZARD, 1'3, F. TOURON 4 & I. D E K E Y S E R 2
t Laboratoire de Mesures en Forage (ODP), IMT, 13451 Marseille cedex 20, France
2 Centre d'Oc~anologie de Marseille, CNRS (URA 41), Universitd d'Aix-Marseille H,
13288 Marseille cedex 09, France
3 Laboratoire de POtrologie Magmatique, CNRS (UPRES A 6018), CEREGE, 13545 Aixen-Provence cedex 04, France
4 Gafa Entreprises, 16 Boulevard Notre-Dame, 13006 Marseille, France

Abstract: A new methodological approach based on the analysis of core data, logs and highresolution electrical images of borehole surfaces (FMS) is developed in order to improve the
study of oceanic sediments from physical properties. This approach is tested on data
obtained in the context of the Ocean Drilling Program (Japan Sea, Leg 128, Hole 798B). The
downhole measurements and FMS images exhibit a cyclic pattern reflecting variations in
oceanic surface productivity combined with continental aeolian supply due to palaeoclimatic changes. On the basis of m-scale physical measurements, cm-scale FMS images and
measurements on core, the objective is to deconvolve the variations in sedimentary supply of
oceanic and continental components through time and to compute the intrinsic formation
factor versus depth. The latter topic is approached in two ways: first by conventional log
analysis, then with a new iterative forward modelling method. In the second case, the low
frequency electrical resistivity log (SFL) is modelled using a numerical modelling code
(Resmod2D e:) in order to obtain an accurate formation electrical resistivity model (Rt),
where individual beds are derived from FMS images. An analytical routine is also used to
model the natural gamma-ray measurement (CGR). While the conventional log analysis
allows deconvolution of the sedimentary supplies, the forward modelling leads to a greater
resolution and accuracy in more precise sediment characterization, such as that obtained
from the derivation of the formation factor.

Unlike the core material, downhole logs provide


continuous high resolution records. The logs
reflect the physical and chemical variability of
the drilled sequence. Several logging tools based
on widely varying physical principles (electric,
acoustic, nuclear . . . . ) are used. The logs offer
different perspectives about changes in sediment
composition. Hence, extracting sediment characteristics or palaeoclimatic information from
downhole logs appears as a promising field of
application. The principal limitation on integrating logging data for sedimentary and palaeoclimatic studies is the absence of a generally
applicable method to transform logging data
into reliable sediment physical properties or
palaeoclimatic data.
The major objective of this study is to develop
a new methodological a p p r o a c h using, in
combination, logging, coring data and highresolution electrical images (FMS) to derive the
detailed structure of near sea-floor sediments.
The method is tested on data obtained in the

context of the Ocean Drilling Program (ODP) at


Oki Ridge in the Japan Sea (Leg 128 Hole
798B). Core data are first used to construct a
mineralogical model of the sedimentary formation at Site 798. Classical log analysis is then
applied to deconvolve the different sedimentary
inputs and to compute a continuous formation
factor (FF), which offers a tool to describe the
sediment pore structure. However, this approach
is limited by the vertical resolution of each
sensor, that is half a metre on average for
traditional downhole measurements (e.g. logs).
An original approach combining analytical and
numerical modelling is proposed here to perform
a small scale analysis of downhole logs through
the computation of the formation factor.

Geological setting at Site 798


Site 798 (37~
134~
is located in the
southeastern Japan Sea, about 160km north of
the western coast of Honshu. The site is

115
LAUER-LEREDDE,C., PEZARD,P. A., TOURON,F. & DEKEYSER,1. 1998. Forward modelling of the
physical properties of oceanic sediments: constraints from core and logs, with palaeoclimatic implications.
In: HARVEY,P. K. & LOVELL,M. A. (eds) Core-Log Integration, Geological Society, London, Special
Publications, 136, 115-127

116

C. LAUER-LEREDDE E T AL.

Fig. 1. Location map of the area surrounding Hole


798B.

positioned over a small sediment-filled graben


on top of Oki Ridge, in 911.1 m water depth
(Fig. 1). A 517m thick sediment sequence of
late/early Pliocene to Holocene age was drilled;
diatomaceous ooze, diatomaceous clay, silty
clay, clay, and siliceous claystone are the
predominant sediments. The primary drilling
objective at this site was to obtain a complete
Neogene sequence of pelagic-hemipelagic sediments deposited above the local carbonate
compensation depth (CCD), currently near
1500 m, in order to obtain a detailed description
of the sedimentary input at the site.
The strategically positioned location and the
high avera/~e sediment accumulation rate (about
12 cm ka -~) at Site 798 are ideal to study the
local sedimentology in relation to global palaeoclimatology. This site is of great interest for two
prevailing sedimentary supplies are defined from
smear slides observations (Ingle et al. 1990) and
FMS images (Fig. 2). The upper 300 m of FMS
images (late Pliocene/Pleistocene) are characterized by rythmic changes between dark, laminated, diatom- and organic carbon-rich
conductive intervals, and light-coloured, nonbioturbated to bioturbated, clay-rich, resistive
intervals (F611mi et al. 1992). To investigate the
sedimentary origin of these cycles, Dunbar et al.
(1992) analysed a total of 913 samples for
biogenic opal content (Fig. 3): major features
of the opal record are a general trend of
increasing opal fraction with depth, and cyclic
variations between high and low values at a
period of approximately 40 ka. The opal content
varies between 3 and 43 Wt% in the upper
320m. DeMenocal et al. (1992) also analysed
contiguous samples over three intervals located
between 100 and 320 mbsf (metres below sea

Fig. 2. Formation MicroScanner (FMS) micro-resistivity images from the ODP Hole 798B (from 200 to
300 mbsf). The images are azimuthal traces of the four
pads pressed along the borehole wall. Black represents
low resistivity, and white, high resistivity.

FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES


Core

SVL
(Ohm m)

Opal

recovery
_

o~

~o

(wt%)
20 30

117

0.45

40

so

loo

0.55

0.65

100

OPAL
(%)
II0

10

lo

,,0

110

120

120

130

130

100
I

140

140

8
150
200 i

150

Fig. 4. Correlation between the SFL log and opal


percent measured on core in ODP Hole 798B (after
DeMenocal et al. 1992).

No data
2so

300

Fig. 3. Weight percent biogenic opal versus depth in


ODP Hole 798B (after Dunbar et al. 1992).

floor). These samples were analysed for major


sediment composition: biogenic opal content
varies between 5 and 40%, and terrigenous silts
and clays, between 40 and 80%. Core-log
correlations were established using ash layers
identified in core photographs and Formation
MicroScanner T M (FMS) images. High opal
values result in low gamma ray, bulk density,
grain densities, and resistivity log values. There
is a close correspondance between the SFL and
the opal data (Fig. 4). Low opal content is
balanced by increases in terrigenous sediment,
and this is recorded by high gamma ray log
values (DeMenocal et al. 1992). Core-log
comparisons therefore demonstrate that log
cycles reflect variations in terrigenous sediment
supply and diatomaceous opal. Diatom tests are

the dominant opaline component throughout


the upper 300m; radiolarians and silicoflagellates contribute in a minor way to the opal flux
(Ingle et al. 1990). The periodicity of the
sedimentary cycles was estimated with standard-power spectral analysis method (Imbrie et
al. 1984): the power spectra of the gamma-ray
(SGR) time series showed a peak at about 40 ka,
probably a climatic expression of the 41ka
'Milankovitch-type' cyclicity (DeMenocal et al.
1992). This suggested that the earth obliquity
was the driving factor of climate and sedimentary supply in this region over the last 3 Ma.
The diatomaceous sediments of the dark
facies, and the terrigenous-rich signature of the
light-coloured lithofacies suggested that these
cycles also reflect variations in oceanic surface
productivity combined with continental aeolian
dust from central Asia, as a consequence of
palaeoclimatic changes. The terrigenous mineralogy assemblage is similar to that of Chinese
loess, a probable up-wind source of the aeolian
dust. The Chinese loess deposits may indicate a
linkage between glacial climate and Asian
aridity (Kukla et al. 1988), so the periodic
increases in terrigenous concentration may
reflect the downwind propagation of this
signal.

118

C. LAUER-LEREDDE E T AL.

Table 1. Chosen physical properties fop" mah7 components


Phase
Continental

Component

Densit~r
gcm -

PEF
ba e 1

CEC
meq gq

Illit e
Chlorite
Kaolinite
Smectite
Quartz

2.50
2.60
2.42
2.12
2.65

3.5
6.3
1.83
2.04
1.8

0.1-0.4
0.05-0.4
0.03-0.15
0.8-1.5
0

Table 2. Clay composition re[erred to 100 Wt% clav.fi'action


Zones*
(mbsf)
Za (200 220)
Zb (220~ 225)
Zc (225- 260)
Zd (260 280)
Ze (280-300)

Illite
(%)

Chlorite
(%)

Kaolinite
(%)

Smectite
(%)

85
80
90
85
80

10
10
10
10
5

0
10
0
5
10

5
0
0
0
5

*The studied interval was split into five zones, each with a constant clay mineralogy, on the basis of Dersch &
Stein data (1992}.

Core and log data


This work is focused on a 100-m-long interval
(from 200 to 300 mbsf) because of the wellexpressed cyclicity over this segment (Fig. 2).
Mineralogical model

In the following study, the sediment physical


properties and the main mineralogical components are used to compute the relative proportions for oceanic and continental supplies, and
to determine the formation factor (FF). A first
order mineralogical model of Hole 798B is
deduced from smear slide observations of
dominant lithologies, and from previous sediment composition studies.
The oceanic input, essentially diatoms, is
associated with opal, on the basis of Ingle et
al. (1990), DeMerlocal et al. (1992), and Dunbar
et al. (1992) works. The continental one is
deduced from Dersch & Stein (1992) core
analyses at Site 798. In order to get information
about the composition of the terrigenous sediment fraction, Dersch & Stein (1992) determined
the average amounts of quartz and clay minerals. The entire sequence is characterized by
quartz amount ranging between 5 and 20%. In
the upper 413m, the clay fraction is dominated
by illite with values between 60 and 88% and
chlorite, between 0 and 27%. Calcareous components are either absent or poorly preserved,

and carbonate contents average less than 4%


between 200 and 455 mbsf. In this section,
volcanic ash layers are thin and scarce. The
continental input is therefore assumed to be
composed at this site of four clay minerals (illite,
chlorite, kaolinite, and smectite) and quartz. Six
components are consequently taken into account in this study, and characterized by three
physical properties: density (g cm 3), photoelectric-effect (ba e q ) and cation exchange
capacity (meq g l). The reference values for
each of these components (Table 1) are chosen
from the literature (e.g. Grim 1968; Fertl &
Frost 1980; Juhasz 1981; Caill6re et al. 1982;
Drever 1982; Schlumberger 1994).
Our objective is to obtain information on
oceanic and continental supplies, rather than on
the relative fractions of the main mineral
components.The proportions of each element
for the continental phase are, however, needed
in order to estimate the physical properties of
this phase. A short interval from 200 to 300 mbsf
(1.7 to 2.5 Ma) was chosen as a first step of this
analysis. This interval was divided in five zones
characterized by average clay fractions (Table
2), on the basis of previous analyses (Dersch &
Stein 1992). This division in zones allowed to
simplify the mineralogical model still further,
and to reduce the number of unknowns: for
example, the continental phase is constituted
with only three elements (illite, chlorite, quartz)
for zone C (Table 2). The sediment averages

FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES


Natural G a m m a Ray
CGR (API)
200

20

40

Electrical Resistivity

B u l k density

(fl m)

(g ce-l)

60

0.4

0.5

0.6

0.7

1.4

1.6

Photoelectric effect
(ba/e')
1.9
2.2

2,6

119

N e u t r o n Porosity

(%)

6O

70

I 'T'" I

8O

I~l

II

220

240

260

280

41
Fig. 5. Downhole measurements from 200 to 300 mbsf in ODP Hole 798B.

10% quartz for the whole interval, 50% clay


between 200 and 260 mbsf and 60% clay
between 260 and 300 mbsf (Ingle et al. 1990,
Dersch & Stein 1992). The percentages of the
continental phase are then deduced for each
zone.

Downhole measurements

At the completion of coring operations at Hole


798B, four logging runs were completed from 70
to 518 mbsf. During the first phase, the phaser
dual induction tool (DIT), the long-spacing
digital sonic tool (SDT), and the natural
gamma-ray spectrometry tool (NGT) were run
(seismic stratigraphic tool string). The second
phase consisted of the lithoporosity combination
tool string including the lithodensity (LDT),
compensated neutron (CNT-G) porosity, and
natural gamma-ray spectrometry (NGT) tools.
After the Formation MicroScanner T M (FMS)
had been lowered downhole, the geochemical
tool string, including an induced gamma-ray
spectroscopy tool (GST), an aluminium clay

tool (ACT) and the NGT, was run (Ingle et al.


1990). The depths of investigation are sensordependant, and data are typically recorded at
intervals of 15cm. The quality of the logs
obtained is generally excellent: most of the logs
reflect variations in biogenic opal production
(diatomaceous) resulting from glacial-interglacial changes in surface productivity (Matoba
1984; Zheng 1984; Morley et al. 1986).
Downhole geophysical logs (m-scale). The following analysis focuses on the resistivity (SFL)
and natural gamma ray (CGR) logs from 200 to
300 mbsf in Hole 798B. These logs were selected
because the resistivity is influenced both by the
clay fraction and diatoms (oceanic productivity
input), whereas the natural gamma ray is mainly
sensitive to the clay fraction (aeolian continental
input in this case). The CGR is often used to
indicate downhole variations in clay minerals
content, because it reflects gamma-ray radioactivity from the decay of potassium and
thorium which are common elements in clay
mineral structures (Hassan et al. 1976). The

120

C. L A U E R - L E R E D D E E T AL.

CGR may then serve as a proxy of variations in


terrigenous aeolian component.

Deconvolution o f cont&ental and oceanic


inputs

Throughout the studied depth interval, the


highest CGR value corresponds to the highest
bulk density values (Pb), the highest resistivity
values (R0), and the lowest porosity values (~b)
(Fig. 5), reflecting a high terrigenous content
relative to the biogenic supply. Terrigenous clays
have high K and Th contents and relatively
higher density and lower porosity than sediment
with higher opal content: the clay particles filling
the pores induce a lower porosity, hence a higher
resistivity. Porosity is, to a first order, proportional to the inverse square-root of resistivity
(Archie 1942). The sediments rich in diatomaceous opal commonly have high porosities
because of the intrinsically high porosity of
diatoms themselves. These cycles are also
apparent in the recovered sediment record: the
high gamma-ray, high density, high resistivity,
and low porosity levels correspond to the
massive clay-rich intervals, whereas the low
density, low gamma-ray, low resistivity, and
high porosity units correspond to the darker,
diatom-rich intervals.

The aim is to analyse the downhole measurements in order to deconvolve both oceanic and
continental inputs. The model of the formation
consists of only two known inputs in unknown
proportions. Bulk and matrix densities (Pb, Pma,
g CC-1) and photoelectric effect (Pef, ba e-1) were
chosen to define proportions of the two components. Whereas Pb responds primarily to porosity, the Per responds primarily to rock matrix
(lithology). The combination of Pb and the Pef,
the photoelectric absorption cross-section
(Schlumberger 1994), is:

High-resolution (cm-scale) electrical images. The


Formation MicroScanner TM (FMS) creates a
picture of the borehole wall by mapping its
electrical conductance using an array of 16 small
and pad-mounted electrodes on each of four
pads (Ekstrom et al. 1986; Luthi & Banavar
1988; Pezard et al. 1990). FMS data are recorded
each 2.5 mm as the tool moves up the borehole.
The vertical resolution of individual features is
about a centimetre. The tool can, however,
detect thinner features, provided they have
sufficient resistivity contrast to the surrounding
matrix. The images registered with the FMS
show qualitative conductivity changes, particularly due to the different physical properties of
the beds (for example porosity, resistivity of
pore fluid or the presence of clays).
The electrical images obtained at Hole 798B
(Fig. 2) resolve the cyclicity of sedimentary
processes at the site extremely well. Light (dark,
respectively) colour is related to the continental
(oceanic) input.

Log analysis
The downhole logs and core measurements,
associated with the proposed mineralogical
model, are used here to determine variations in
sedimentary inputs, and to compute the formation factor.

U = Pef x Pb

(1)

and obeys a linear mixing law such as:


U=~

Uf-~-(1--(~) Uma

(2)

where U, Uf, and Uma are for example the


photoelectric absorption cross-sections of the
media, pore fluid and matrix, respectively.
As our matrix consists of a mixture of two
inputs (oceanic and continental) with relative
weight fraction (#o and #c) and photoelectric
absorption coefficients Uo and Uc:
Uma =/to Uo + #c U~= Pef Pma

(3)

The relation necessary to solve for these two


unknowns is the closure relation of partial
fractions:
l=#o+#c

(4)

The solution can most easily be seen in terms of


the matrix representation of the set of simultaneous equations:
A=R Y

(5)

where A is the vector of measurements, R is the


matrix of known coefficients, and Y is the vector
of unknown volumes.
The porosity and density logs are first used to
compute the matrix grain density Pma. The wet
bulk density Pb is related to the porosity through
a simple mixing law:
Pb = Pwqb + Pma (1 -- qb)

(6)

where Pw is the density of seawater.


On the basis of our preliminary mineralogical
model which consists of six major components
(opal, illite, chlorite, kaolinite, smectite, quartz),

FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES


Uo and Uc are computed for the five zones
previously determined (Table 2) as follows:
Uo = Pef(opal) Pma(opa,)

u~ = L #i Pefi Pmai

(7)

(8)

i=1,5

with ~i as weight percentage of the component i


in the continental phase.
The system resolution leads to the weight
fraction of both oceanic and continental phases.

Computation o f the f o r m a t i o n f ac t or
The electrical resistivity of saturated sediments is
usually quoted in terms of a formation factor
(FF) to remove the effect of the pore-fluid
resistivity, because the grains themselves are
considered as insulators (Archie 1942):
FF = Ro Rw-1 = Cw Co 1

(9)

where Ro (respectively Co) is the resistivity in 12


m (conductivity in S m -1) of the porous medium,
and Rw (respectively Cw), the resistivity (conductivity) of the pore-fluid.
The formation factor of the porous medium
depends on the intrinsic geometry of the pore
channels, and therefore describes the manner in
which the grains are arranged in a sedimentary
formation (Winsauer et al. 1952). Archie's
equation is generally considered to apply satisfactorily to clean sands. The presence of clay
minerals, however, has a detrimental effect on
Co computations: the capacity of a clay to
exchange cations at the pore-mineral interface
induces the presence of a surface conductivity
term (Waxman & Smits 1968). A resistivity
model taking into account the effects of dispersed clays was proposed by Waxman & Smits
(1968) and Waxman & Thomas (1974):
Co = (Cw+ BQv ) FF -1

(10)

Qv=pma CEC (1-~)qb -1

(11)

where B represents the equivalent conductance


of clay-exchange cations (S m 2 meq-1), as a
function of salinity and temperature, Qv describes the cation exchange capacity or CEC
(meq g-a) per unit pore volume (meq cm 3), and
Pma (g cm-3) is the matrix grain density of the
sediment.
In the following, the successive stages of the
computation of the formation factor are detailed.

121

Cation exchange capacity. Values of CEC can be


measured directly on rock samples, but not
directly in situ. Several attempts have consequently been made to derive CEC from existing
logs.
Previously developed CEC or Qv estimates
from well-logs based on the spontaneous potential curve (Smits 1968; Johnson 1978), dielectric
constant (Kern et al. 1976), reservoir porosity
(Lavers et al. 1974; Kern et al. 1976; Neuman
1980) and gamma ray (Koerperich 1975; Clavier
et al. 1977; Johnson 1978) have been discussed.
For this study, the correlation between the
natural radioactivity from K and Th elements
(CGR) and the CEC was selected. Most shale
are radioactive due to the presence of K 4~ in the
potassium-bearing clay mineral illite. A correlation between gamma ray counts and CEC may
then be expected. Johnson (1978) showed such a
correlation for formation containing largely illite
and kaolinite where the relatively high gammaray count of the illite corresponded to high
potassium content thus making it an excellent
shaliness indicator. Scala (see Clavier et al. 1977)
found a strong correlation between gamma ray
count rate divided by the porosity and Qv. In
other words, gamma ray log can be used in some
cases and after calibration on core as a substitute
of the CEC measurement. On the basis of Scala
data, we estimated the proportionality constant
between the two quantities as follows:
CEC = (0.005) CGR

(12)

The computed values of CEC are then


converted into Qv using (11).
To check the validity of (12), an analytical
maximum and minimum CEC are estimated
with CEC values from Table 1 in each zone
(Table 2). Using the relative proportions of each
clay mineral, an estimate of the CEC of the clay
assemblage can be computed, using a linear
summation.

Formation factor. Waxman & Thomas (1974)


found that B can be related to an exponential
function of the conductivity, and Juhasz (1981)
proposed the following expression:
B = -(1.28) + (0.225)T-(0.0004059)T 2 (13)
1 + Rw kz3(0.045T--0.27)
where T is the temperature in ~ and Rw the
fluid resistivity in f~m.
The mean value of B obtained for Hole 798B
from (13) is 3.8 S m2meq -1. Continuous FF
values versus depth may then be evaluated from
(10).

122

C. LAUER-LEREDDE ET AL.

Fig. 6. Log analysis results. (a) Grain density; from core (solid squares) and computed (solid line) (b) Opal
fraction from core (after Dunbar et al. 1992) and computed. (c) Computed continental sedimentary fraction. (d)
Computed CEC (derived from CGR) and Qv values. (e) Computed formation factor from definition (dash) and of
Waxman & Smits (1968) (solid).

Results
Grain density. The computed matrix density
(Fig. 6a) exhibits a high degree of variability.
The matrix density reflects the varying clay and
diatom contents. Diatoms tend to have low
densities, sometimes lower than 2.0g cm -3,
whereas clay minerals have densities ranging as
high as 2.80g cm -3 (Johnson & Olhoeft 1984).
The estimated values and the core measurements
are in general agreement over the interval,
although fine-scale correlations between the
two quantities are difficult. This difficulty results
mainly from the discrepancy between the core
and log measurements themselves. One of the
problem with gaseous sediment is that the core
recovery is often fragmented and the section is
expanded and disturbed, leading to differences
between the core and log depth-scales (e.g.

Hagelberg et al. 1992). Hence, measurements


on core cannot been compared readily with
downhole logs.
Oceanic and continental fractions. The agreement
between the reconstructed opal fraction curve
and core measurements from Dunbar et al.
(1992) is very good throughout the section
(Fig. 6b), although more measurements in the
upper part would be desirable for a better
comparison. The major variations are well in
phase: for example, Dunbar et al. measured an
abrupt decrease of opal content at about 269
and 287 mbsf, and our estimated values present
the same feature. Moreover, fine variations
appear in the reconstructed opal fraction curve.
The amount of computed opal is however overestimated, especially in the upper part (about
20%). This last point might originate in the

FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES


mineralogical model: the oceanic input is
assumed to be entirely opal, whereas other
components are also present. For example, we
considered as insignificant the biogenic carbonate component, although oceanic intervals are
enriched in foraminiferas, essentially in the
upper 250 mbsf. Also, the values of the physical
properties chosen for this first-step model are
only reference values. The true values for the
components at Hole 798B are not known
exactly. An additional cause of the difference
between core and computed opal might be the
methods chosen by Dunbar et al. (1992) to
measure the opal content. They used a timeseries dissolution technique and a one-step
dissolution method. In general, the results from
both techniques are comparable, but the onestep method tends to yield opal contents
consistently lower by 5 to 10% in enriched
samples.
As measurements on core for the continental
fraction was not available, a precise comparison
to validate our model was not possible. The
computed values ranging between 40 and 90%
(Fig. 6c) are in agreement with the measurements of Ingle et al. (1990). Moreover, the
reconstructed continental fraction curve is characterized by a significant increase in clay content
between 278 and 286 mbsf, as suggested by the
increase in gamma-ray, bulk density and resistivity (Fig. 5).
Cation exchange capacity. The computed CEC
and Qv logs (Fig. 6d) follow the variations of
clay abundance and are restricted to analytical
boundaries. The proposed proportionality constant fits well. Guo (1990) measured the CEC of
several loess samples from China. The CEC
ranged between 0.07 and 0.28 meq g 1, which is
within the range of the present results.
Formation .factor. The formation factor derived

from the Archie formula is lower than that


derived from Waxman & Smits formula (Fig.
6e), particularly in high resistivity zones. The
data set can be represented by a regression
similar to that proposed by Winsauer et al.
(1952), and such as F F = a ~ .... , with a =1.45
and m = 2.38 (Fig. 7). This result is in the range
typical of marine sediment. In a similar approach, Henry (1997) analysed clay-rich sedimentary samples for CEC from the Barbados
wedge (ODP Site 948), similar to those from Oki
Ridge: the electrical resistivity varies from 0.5 to
0.8 f~ m, the porosity is larger than 50%, and the
grain densities measured on samples are close to
2.80 g cm 3. The CEC measured on core by
Henry (1997) ranges between 0.2 and 0.5 meq g-i

123

Fig. 7. Formation factor versus porosity plotted on


double logarithmic scale. Jackson et al. (1978) and
Taylor Smith (1971) results are displayed (A, B, C, D,
E; F). The present data (G) define a trend described by:
FF = 1.45qb-2'38.

and the relationship F F = 1.24qb 2.31 is close to


that from Oki Ridge. Taylor Smith (1971)
analysed samples from Mediterranean Sea clays
and found a m value close to 2.20 (Fig. 7). The
results of Jackson et aI. (1978) show that the
exponent m depends entirely on particle shape
for unconsolidated sands (Fig. 7). Similar
measurements on assemblages of shell fragments, kaolinite particles, and marine illite clays
produce similar values of m (close to 2.0),
suggesting that the platey nature of the particles
within clays controls the relationship between
FF and qb. The high value of m derived for Oki
Ridge sediments is then in agreement with
similar results in formations with large amounts
of clays (Jackson et al. 1978), especially illite.
The results obtained for Oki Ridge sediments
are also typical with regard to the large spread of
F F values. This spread reflects the change in
shape of particles in relation to the supplies
cyclicity. High values of FF correspond to the
continental input, i.e. clays, whereas low values
of FF correspond to the oceanic input.
As a conclusion, the simple mineralogical
model used here appears as well adapted to the
description of sedimentary formations with high
porosities (greater than 60 %) and clay content.
These first results also demonstrate that the
computation of the oceanic and continental

124

C. LAUER-LEREDDE ET AL.
C G R (API)
--- computed
- - measured

200

30
tl%[

50
I

C G R model
(API)
30
I

50
i

S F L (f2 m)
... computed
- - measured
I

II

I 4

0.5
1

0.7
I

Rt model
(fi m)
0~5
I

0.7
I

FFs
... from raw logs
-- from modelled logs
t

II

2.5
;ll=1

3.5
[

4,5

205

"-~21q

(a)

(b)

(c)

(d)

(e)

Fig. 8. Forward modelling results. (a) Computed and measured gamma ray values (CGR). (b) Formation natural
gamma ray, from K and Th, model expressed in terms of CGR. (c) Computed and measured electrical resistivity
values (SFL). (d) Formation electrical resistivity model (Rt). (e) Formation factor as determined from downhole
measurements and the numerical model

fractions using the photoelectric absorption


cross section (U) is in agreement with core
measurements and, so, might be used to predict
the core data. Nevertheless, the vertical resolution of the downhole logs and the computed
formation factor is rather poor in some zones,
especially in the upper part of the section. A new
forward modelling method is therefore proposed
in the following to improve the vertical resolution and derive more accurate Rt, CGR and FF
profiles.
Forward

model

The aim is to obtain an accurate formation


resistivity model (Rt) from the numerical modelling of the electrical resistivity log (SFL),
constrained by the high-resolution electrical
images of borehole surfaces (FMS). A statistical

method is also used to model the natural gamma


ray data (CGR). This study is restricted to a 20
m-long interval (200-220 mbsf/1.6 to 1.9 Ma).

Numerical modelling
Modelling code. Resmod2D ~ is a newly developed two-dimensional finite element numerical
code. It allows modelling of the response of
electrical resistivity downhole probes, such as
the Spherically Focused tool (SFL). In brief, a
formation resistivity model composed of horizontal sedimentary beds (layers) with fixed
thicknesses and resistivities, is entered in the
code in order to compute the response of the
probe in front of this formation. The resistivities
of the model are referred to as 'true', whereas the
computed resistivities (so the simulated response
of the tool) are referred to as 'apparent' because

FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES


in an inhomogeneous formation, it depends on
the resistivity of the bed next to the probe and
also that of the adjacent formations. The
measure of the resistivity by the SFL is therefore
lower than the true resistivity.
Res&tivity modelling. Using Resmod2D :t:, a
formation resistivity model (Rt) is created on
the basis of cm-scale electrical images (FMS) for
bed thickness and m-scale electrical log (SFL)
for individual bed resistivity. The layer boundaries are determined from FMS images by
distinct colour contrasts; the gradual transitions
are disregarded here. This initial model is then
processed with Resmod2D ~ to simulate the SFL
tool and compute a theoretical resistivity log
(SFLc). By comparison between SFLc and
SFLm, the formation model is modified step
by step (resistivity values and frame). Thickness
changes as well as layer additions in the
iterations are constrained by FMS images. This
process is iterated until the best fit between SFLc
and SFLm is obtained. Each processing lasts
about four hours for an evaluation every 5 cm
and over a 20 m-long interval (from 200 to 220
mbsf).
Natural gamma ray modelling. A statistical
method is used to model the natural gammaray tool in an analytical manner. The aim is to
determine the natural gamma activity of each
layer defined in the resistivity model. This simple
method is based on an exponential attenuation
of the gamma ray flux versus depth (Ellis 1987).
Using the lithological frame established for the
resistivity model, and the natural gamma ray
measurements (CGRm), an initial model is
established to estimate the C G R (CGRc). By
comparison between CGRc and CGRm, the
model is modified step by step and the chosen
model corresponds to the best fit between CGRc
and CGRm.
Results
Formation electrical resistivity and gamma ray
models. The chosen models, obtained after about
80 iterations, correspond here to a near-perfect
fit between the measured and computed values
(Figs 8a, 8c). The initial Rt model took into
account the major layers seen on the FMS
images, not the discrete ones. The SFLc then
presented the major variations of the SFLm but
not the small ones. In order to reproduce these
small events and to take the gradual transitions
into account, the basic frame was refined by
adding small layers. Whereas the initial models
were composed of 47 layers, the chosen ones are

125

made of 76 layers, the smallest one measuring 8


cm and the greatest, 90cm. In order to validate
the models, several precision and robustness
tests were run. For example, the error E between
the downhole logs and the computed theoretical
logs was computed using Whitman (1989)
method:
E(%) + 100 (10 El~

E l o g = ~l Z
i=l,U

1)

(log(mi) -- log(ci)) 2

(14)

(15)

where N is the number of records for the chosen


downhole log, and mi (respectively ci) is the
value of the record i for the downhole log (for
the computed log). This estimated error is on the
order of 1 % for the SFL, and 3 % for the CGR.
Due to the integration of high-resolution
FMS results, the models (Figs 8b, 8d) have a
much better resolution than the raw logs (Figs
8a, 8c). Whereas the logs show essentially three
continental events between 200 and 220 mbsf,
the models show the same main three events, but
also several small ones. For example, while the
SFL seems rather linear over the first five metres,
the Rt model brings out an alternance of minor
troughs and peaks suggesting little changes in
lithology, such as clay content decreases corresponding to the troughs. Whereas other lowresistivity units seem to be massive on the SFL,
they are composed of several fine layers in the
model. The high-resistivity units, related to
colder periods, are characterized in the raw log
by two or three regular peaks, whereas the
model displays a much more irregular profile.
The major peaks are more pronounced and the
contrast is greater: the peak at about 211.15
mbsf has a value of 0.58 f~ m for the raw log, and
0.80 f2 m for the model. Moreover, the transition
from the low-resistivity interval upward into the
high-resistivity one is rather gradual for the raw
log, whereas the model seems to show that a
sharp boundary is present at the base of the
high-resistivity unit, suggesting an abrupt initiation of each glacial period. All these features
revealed by the models are confirmed by core
observations (Ingle et al. 1990): the vertical
lithologic variations within the dark/light cycles
are remarkably constant. The dark-coloured
intervals are either thinly to thickly laminated
and finely bedded. These intervals generally
possess a well-defined and sharp base that grades
upward into light-coloured/high-resistivity sublayer (Ingle et al. 1990); the lower portion of the
light-coloured intervals is commonly gradational

126

C. LAUER-LEREDDE ET AL.

and mostly obliterated by bioturbation. The


determination of the boundaries in the model
seems therefore to be relatively accurate.
Formation factor, The formation factor is
computed for the modelled logs (Rt and CGR)
on the basis of the method presented in the log
analysis section.
The results for the F F model are similar to
those for the Rt model. The F F model has a
better resolution than the F F computed from
logs, as the former displays small events not
detected by the latter (Fig. 8e). The major peaks
are also more pronounced: the peak at about
211.15 mbsf has a value of 3.48 for the raw log,
and 4.62 for the model. Changes between glacial
and interglacial periods also present the same
characteristics as the Rt model (sharp transition
from dark laminated interval upward into lightcoloured interval, and gradual transition from
light interval upward into dark one).

Conclusions
The first results of the method using the raw logs
show that the proposed mineralogical model is
well representative of the sediment from ODP
Hole 798B. The photoelectric absorption crosssection (U) allows differentiation and computation of the oceanic and continental fractions.
The vertical resolution of the raw logs and of the
computed formation factor curve being however
poor in some zones, a new forward modelling
method is proposed in this paper to improve this
study.
The m-scale electrical log (SFL) and the
natural gamma ray log (CGR) are modelled
using cm-scale electrical images (FMS) to define
and map high-resolution layers, The first results
show that the Rt, C G R and F F models are more
precise than the Rm, C G R and FF obtained
from log analysis, insofar as the models bring
out small layers not detected by the raw logs.
The modelling approach allows the study of
changes from oceanic to continental supply
hence from interglacial to glacial periods: the
continental input tends to increase abruptly
from warm periods to colder ones suggesting
abrupt initiation of glacial cycle, whereas it
seems to decrease gradually from cold to warmer
periods.
This study provides a continuous description
of changes in intensity of the different sedimentary sources within the analysed interval. While
FMS images reveal the presence not only of
large-scale layers but also of thin (short) events,
the numerical modelling enhances the nonlinearity of m-scale logging devices, stressing

for example the importance of cm-scale resistive


beds on the response of m-scale logs. This case
study, which now requires additional measurements on core to further improve the precision
of the method, could become a parallel method
to obtain meaningful physical properties and
palaeoclimatic data from future sites.
This work was carried out by the main author with
financial support from the French 'Minist~re de la
Recherche et de l'Enseignement Sup6rieur'. The
authors wish to acknowledge the contribution of GaYa
Entreprises (Marseille) for the use of the forward
modelling code Resmod2D". They also wish to thank
S. Brower (LDEO) for providing the logging data, C.
Robert (COM, Marseille) and the two anonymous
reviewers for their helpful comments.

References
ARCHIE, G. E. 1942. The electrical resistivity log as an
aid in determining some reservoir characteristics.
Petroleum Transaetions of the AIME, 146, 54-62.
CAILLERE, S., HENIN, S., RAUTUREAU, M. 1982.
Min{ralogie des argiles. 2. Classification et nomenclature. Masson, Paris.
CLAVIER, C., COATES, G. DUMANOIR, J. 1977. The
theoretical and experimental bases for the dualwater model for the interpretation of shaly sands.
Society of Petroleum Engineers 52nd Annual Fall
Technical Conference of AIME, Denver.
DEMENOCAL, P. B., BRISTOWJ. F. & STERN,R. 1992.
Paleoclimatic applications of downhole logs:
Pliocene Pleistocene results from Hole 798B, Sea
of Japan. Proceedings of the Ocean Drilling
Program, Scientific Results, 127/128, 393407.
DERSCH, M. & STEIN, R. 1992. Pliocene-Pleistocene
fluctuations in composition and accumulation
rates of Polo-marine sediments at Site 798 (Oki
Ridge, Sea of Japan) and climatic change:
preliminary results. Proceedings of the Ocean
Drilling Program, Scientific Results, 127/128 (1),
409-422.
DREVER,J. I. 1982. The geochemistry of natural waters,
Prentice-Hall, New Jersey.
DUNBAR, R. B., DEMENOCAL, P. B. & BURCKLE, L.
1992. Late Pliocene-Quaternary biosiliceous sedimentation at Site 798, Japan Sea. Proceedings of
the Ocean Drilling Program, Scientific Results,
127/128 (1), 439~455.
EKSTROM, M. P., DAHAN, C. A., CHEN, M.-Y., LLOYD,
P. M. & Rossl, D. J. 1986. Formation imagining
with microelectrical scanning arrays. Transactions
of the Society of Professional Well Log Analysts,
27th Annual Logging Symposium, Paper 88.
ELLIS, D. V. 1987. Well logging/or earth scientists.
Elsevier, New-York.
FERTL, W. H. & Frost, E. 1980. Evaluation of shaly
clastic reservoir rocks. Journal of Petroleum
Teehnology, 31, 1641 1646.
FOLLMI, K. B., CRAMP, A., F(SLEMI, K. E., ALEXANDROVlCH, J. M., BRUNNER, C., et al. 1992. Darklight rhythms in the sediments of the Japan Sea:

FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES


preliminary results from Site 798, with some
additional results from Sites 797 and 799.

Proceedings of the Ocean Drilling Program,


Scientific Results, 127]128 (1), 559-576.
GRIM, R. E. 1968. Clay mineralogy. Second Edition,
Mac Graw Hill Book Company, New York.
Guo, Z. T. 1990. Succession des paleosols et des loess du

Centre-Ouest de la Chine: approache micromorphologique. PhD Thesis, University of Paris 6.


HAGELBERG, T., SHACKELTON, N., PISlAS, N & The
Shipboard Scientific Party 1992. Development of
composite depth sections for Sites 844 through
854. Proceedings of the Ocean Drilling Program,
Initial Reports, 138, 79-85.
HASSAN, M., HOSSIN, A. & COMBAZ, A. 1976. Fundamentals of the differential gamma-ray log. Transa c t i o n S P W L A , 17th A n n u a l L o g g i n g
Symposium, Paper H, 1-18.
HENRY, P. 1997. Relationship between porosity,
electrical conductivity and cation exchange capacity in Barbados Wedge sediments. Proceedings of

the Ocean Drilling Program, Scientific Results,


156.

IMBRIE, J. J., HAYS, J. D., MARTINSON, D. G.,


MCINTYRE, A., MIX, A. C., MORLEY, J., PISIAS,
N. J., PRELL, W. L. SHACKLETON,N. J. 1984.
The orbital theory of Pleistocene climate: support
from a revised chronology of the marine oxygenisotopic record. In: BERGER,A., IMBRIE,J., HAYES,
J., KUKLA,G. & SALTZMAN,B. (eds), Milankovitch
and Climate, Part L Dorrecht, Holland, 269-305.
INGLE, J. C., JR., SUYEHIRO,K., VONBREYMANN,M. T.,
et al. 1990. Proceedings of the Ocean Drilling
Program, Initial Reports, 128.
JACKSON,P. D., TAYLORSMITH,D. & STANFORD,P. N.
1978. Resistivity-porosity-particle shape relationships for marine sands. Geophysics, 43, 12501268.
JOHNSON W. 1978. Effect of shaliness on log response.
Canadian Well Logging Society Journal, 10, 2957.
JOHNSON, G. R. & OLHOEFT, G. R. 1984. Density of
rocks and minerals. In:: CARMICHAEL,R. S. (eds)

CRC Handbook of Physical Properties of Rocks.


Boca Raton, FL (CRC Press, Inc.), 3, 1-38.
JUHASZ, I. 1981. Normalised Qv - - the key to shaly
sand evaluation using the Waxman-Smits equation in the absence of core data. Transaction
SPWLA 22nd Annual Logging Symposium,
Paper Z.
KERN, J. W., HOYER, W. A. & SPANN, M. M. 1976.
Low porosity gas sand analysis using cation
exchange and dielectric constant data. Transaction of the Society of Professional Well Log
Analysts, 17th Annual Logging Symposium,
Paper PP, 17 p.
KOERPERICH,E. A. 1975. Utilization of Waxman-Smits
equations for determining oil saturation in a low-

127

salinity, shaly sand reservoir. Journal of Petroleum


Technology, 27, 1204-1208.
KUKLA, G., HEELER, L., MING, X., CHUN, X. T.,
SHENG, L. T. & SHENG, A. Z. 1988. Pleistocene
climates in China dated by magnetic susceptibility. Geology, 16, 811-814.
LAVERS,B. A., SMITS,L. J. M. & VANBAAREN,C. 1974.
Some fundamental problems of formation evaluation in the North Sea. The Log Analyst, 12, 1-10.
LiiTH1, S. M. & BANAVAR,J. R. 1988. Application of
borehole images to three-dimensional geometric
modelling of eolian sandstone reservoirs, Permian
Rotliegende, North Sea, American Association of
Petroleum Geologists Bulletin, 72, 1074-1089.
MATOBA, Y. 1984 Paleoenvironment of the Sea of
Japan. In: OERTLI, H. J. (ed.) Benthos '83 2nd

International Symposium on Benthic Foraminifera,


409-414.
MORLEY, J., HEUSSER, L. & SARRO, T. 1986. Latest
Pleistocene and Holocene paleoenvironmnet of
Japan and its marginal sea. Palaeogeopraphy,
Palaeoclimatology, Palaeoecology, 53, 349-358.
NEUMAN, C. H. 1980. Log and core measurements of
oil in place. Journal of Petroleum Technology, 32,
1309-1315.
PEZARD, P. A., LOVELL, M & The Ocean Drilling
Program Leg 126 Shipboard Scientific Party 1990.
Downhole images: electrical scanning reveals the
natur of subsurface oceanic crust. EOS, 71, 709
SCHLUMBERGER 1994. Log interpretation charts.
Schlumberger Wireline & Testing, Houston.
SMITS, L. J. M. 1968. SP log interpretation in shaly
sands. Society of Petroleum Engineers Journal, 8,
123-136, Transaction of AIME, 243.
TAYLOR SMITH,D. 1971. Acoustic and electric techniques for sea-floor sediment identification. Proceeding Symposium on engineering properties of
sea-floor soils and their geophysical identification,
Seattle, Washington.
WAXMAN, M. H. & SMITS, L. J. M. 1968. Electrical
conductivities in oil-bearing shaly sands. Society
of Petroleum Engineers Journal, 8, 107-122.
- & Thomas, E. C. 1974. Electrical conductivities
in shaly sands--I. The relation between hydrocarbon saturation and resistivity index; II. The
temperature coefficient of electrical conductivity.
Journal of Petroleum Technology, 6, 213-225.
WINSAUER,W. O., SHEARtN,H. M., MASSON, P. H. &
WILLIAMS, M. 1952. Resistivity of brine-saturated
sands in relation to pore geometry. Bulletin of the
American Association of Petroleum Geologists, 36,
253-277.
WHITMAN, W. 1989. Inversion of normal and lateral
well logs. The Log Analyst, 30, 1-11.
ZHENt, H. H. 1984. Paleoclimatic events recorded in
clay minerals in loess of China. In."PECSI, M. (ed.)

Lithology and Stratigraphy of Loess and Paleosols.


Geographic Research Institute.

Lithological classification within ODP holes using neural networks


trained from integrated core-log data
G. W A D G E 1, D. B E N A O U D A 1, G. FERRIER l, R. B. WHITMARSH 2, R. G.
ROTHWELL

2 & C. M A C L E O D

1Environmental Systems Science Centre, University of Reading, PO Box 238, Reading RG6
6AL, UK
2 Southampton Oceanography Centre, University of Southampton, Empress Dock, European
Way, Southampton S014 3ZH, UK
3Department of Earth Sciences, University of Wales College of Cardiff, PO Box 914,
Cardiff CF1 3 YE, UK
Abstract: Neural networks offer an attractive way of using downhole logging data to infer
the lithologies of those sections of ODP holes from which there is no core recovery. This is
best done within a computer program that enables the user to explore the dimensionality of
the log data, design the structure for the neural network appropriate to the particular
problem and select and prepare the log- and core-derived data for training, testing and using
the neural network as a lithological classifier. Data quality control and the ability to modify
lithological classification schemes to particular circumstances are particularly important. We
illustrate these issues with reference to a 250 m section of ODP Hole792E drilled through a
sequence of island arc turbidites of early Oligocene age. Applying a threshold of > 90%
recovery per 9.7 m core section, we have available about 50% of the cored interval that is
sufficiently well depth-matched for use as training data for the neural network classifier. The
most useful logs available are from resistivity, natural gamma, sonic and geochemistry tools,
a total of 15. In general, the more logs available to the neural network the better its
performance, but the optimum number of nodes on a single 'hidden' layer in the network
has to be determined by experimentation. A classification scheme, with 3 classes (claystone,
sandstone and conglomerate) derived from shipboard observation of core, gives a success
rate of about 76% when tested with independent data. This improves to about 90% when
the conglomerate class is split into two, based on the relative abundance of claystone versus
volcanic clasts.

Within the Ocean Drilling Program (ODP), one


c o m m o n application of d o w n h o l e logs is to
c o m p l e m e n t and calibrate measurements m a d e
on cores and to fill in the gaps left by incomplete
core recovery. Below the range of the Advanced
Piston Corer (typically about 200 m sub-bottom)
core recovery is rarely m o r e than about 50% for
sediments and 40% for basement rocks ( O D P
1990). D o w n h o l e logs make measurements at
ambient temperatures and pressures and sense a
volume a r o u n d the borehole greater than the
core itself. Such measurements provide a continuous stream of in situ data on the wall rocks
and borehole fluids.
The above figures for typical core recovery are
aggregate values. Our knowledge of the exact
depths of recovered core samples is, in general,
worse than these figure imply. In the ODP,
coring advances in steps of about 9 . 7 m (the
length of individual core barrels). Unless core
recovery for this 9.7 m interval is complete we do

not k n o w the exact sub-bottom depth of the core


samples that are recovered, though we may
assume their relative original positions are
preserved. Thus, at the limit, a continuous
1.5 m section recovered from a 9.7 m core may
have originally come from the top or the b o t t o m
1 . 5 m interval. A g r i n i e r & A g r i n i e r (1994)
showed that the best estimate of the position
within finite limits of any arbitrary length of core
sample is given by Euler's Beta distribution. This
can be given as a probability density function of
position in terms of the lengths of core, section
and the n u m b e r and positional order of core
sample. Therefore, d o w n h o l e logs play an even
more important part in filling the gaps in our
knowledge of rock sequences for which there is
incomplete core recovery.
If we can identify the characteristic ranges of
combined log values that correspond to different
lithologies penetrated by the hole then we have a
means of assigning lithological class labels to the

WADGE,G., BENAOUDA,D., FERRIER,G., WHITMARSH,R. B., ROTHWELL,R. G. & MACLEOD,C.


1998. Lithological classification within ODP holes using neural networks trained from integrated
core-log data. In: HARVEY,P. K. t~ LOVELL,M. A. (eds) Core-Log Integration,
Geological Society, London, Special Publications, 136, 129-140

129

130

G. WADGE E T AL.

total interval from the downhole logs. We show


how this can be done for ODP data when we can
assign the lithological classes from those sections
where we do have 'complete' core recovery and
extend the classification to the full hole. Our
approach has been to develop software that
supports a general user in completing this task.
There are a number of methods by which such
a classification scheme can be driven (e.g.
discriminant functions, principal components
analysis and cluster analysis; Doveton 1994).
We have chosen to use classifiers based on
artificial neural networks, principally because of
their ability to cope with complex non-linear
problems. Also, neural network classifiers have
been shown to be of value for lithological
classification of downhole logs in hydrocarbon
exploration wells, in many cases with superior
results to other techniques such as discriminant
analysis (Baldwin et al. 1990; Rogers et al. 1992;
Wong et al. 1995). Goncalves (1995) reports
similar findings for ODP data.
This paper has three main sections. Firstly we
present the main factors involved in the classification task. These are how the problem is
defined, the constraints imposed by the data
available and how to validate the results and
assess performance. Next we describe how we
have implemented the neural network method
on our computer system. Finally, we present
classification results from ODP Hole 792E. The
succession represented in this hole is a complex
sequence of island arc lithologies that is a good
test of the general usefulness of the technique.

ODP lithological classification


The lithologies encountered in ODP holes are
usually deep-sea sediments and oceanic crustal
rocks and are generally distinct from those
encountered during drilling in sedimentary
basins underlain by continental crust. Deep sea
sediments are often relatively unconsolidated
and rich in carbonates and/or silica or composed
of terrigenous or volcanic detritus; they may be
underlain by a basaltic basement. Porosity is
typically high, with ubiquitous saturation by
sea-water. The classification framework of Mazzullo et al. (1987) is widely employed for
sediments by ODP. The highest-level division is
into granular and chemical sediments. Granular
sediments are subdivided into pelagic, neritic,
siliciclastic, volcaniclastic and mixed sediments
and chemical sediments into carbonaceous,
evaporites, silicates/carbonates and metalliferous sediments. Below this level classes are based
on principal names (e.g. ooze, chalk) together
with major or minor modifiers (e.g. nannofossil,

foraminiferal). This nomenclature is applied by


the shipboard petrologists when the cores are
split and each 1.5m section of recovery is
described individually on the Visual Core
Description sheets in freehand. This description
may show some variation from one petrologist
to another. From this initial detailed visual
description a more generalised sequence of
graphical codes (from a total of about 50) are
assigned to each interval of core to denote a
lithological class label (e.g. T6=sandstone;
C34= foraminiferal chalk) so that a composite
graphical log (Barrel Sheet) of all the core
recovery can be drawn for publication in the
Initial Report series of ODP publications.
Beyond this the shipboard scientists can, and
sometimes do, erect other, non-standard but
complementary, classification schemes, perhaps
based on local variation of the sediment. Within
the ODP there is no official archived digital
version of the lithological classification of core.
Hence working on core classification off the ship
requires digital recoding of the shipboard
scheme(s).
The neural network approach is a supervised
classification scheme. It requires that a sufficient
number of training examples of the logs from
each separate lithological class be made available for the algorithm to learn the character of
that class. Two general rules apply here. First, if
there are too few samples within a class then
those samples will not fully represent the
distribution of values that the classifier might
meet through the whole borehole. Second, the
number of samples presented to the classifier
from each class should be approximately the
same. If the number of samples from one
lithology presented to a neural network classifier
is much greater than for the other classes then
the network will tend to bias its classification in
favour of this class. This problem becomes
serious whenever the logs do not provide a clear
separation of the lithological classes.
The quality of log samples used to train the
classifier is also important. In addition to
checking for spurious outliers we use three logs
for quality control. Samples exceeding any of the
threshold values for the caliper, density correction and geochemical factor logs are not used for
supervising or testing the classifier.
Classification choice

The choice of what classification to attempt is of


vital importance. The ideal is to have a set of
lithological classes that best represents the
geological information required from the hole
and which produces distinctive responses in the

ODP LITHOLOGY USING NEURAL NETWORKS


suite of available downhole logs. The usual
starting point will be the principal-names level of
shipboard classification of the core. Classes with
very few member samples may need to be
amalgamated with other classes. The lithological
information required from the hole may not be
the most obvious. For example, a hole may
penetrate a succession of oozes above siliciclastic
rocks lying on a basaltic basement. Classifying
such a tripartite division should be trivially easy
and the real problem of interest may be in the
second-order variability, say, distinguishing volcanic from non-volcanic rocks in the siliciclastic
sequence. In this case the classification task can
be constrained by choice of depth interval.
Alternatively, the need to change the class
labelling given to the core samples to best fit
the problem may only become apparent after an
initial attempt at classification. Merging and
splitting of classes may be required. There is a
clear general need for more than one classification scheme to be tested and for the editing
facilities to support that need.
There is no guarantee that the recovered core,
and hence any classification scheme based on it,
is fully representative of the lithologies in the
missing intervals. Examples of preferential
recovery of, say, clays relative to sands are
well-known. At the extreme a relatively common
lithology may not be recovered at all. It is more
likely that a lithology only ever exhibits low
recovery and hence cannot be matched to
specific depth intervals and used with confidence
to train the classifier. This problem of a missing
class(es) can be partly addressed using exploratory data analysis of the logs themselves. If it is
clear that some populated area of log-space is
not represented by the current classes then a
search can be made to identify the missing class.

131

sub-populations are used to train and test the


performance of the classifier (e.g. discriminant
analysis) and the classification rates of the two
classifiers can now be compared.

Implementation of a neural network method

System design
The quite complex processing chain implicit in
the above discussion is best handled by a
computer system designed for the job. We have
designed such a system, the essential elements of
which are shown in Fig. 1. The computing
platform is a Sun Sparcstation and the graphical
user interface is designed using PV-WAVE
visualization software. There is a separate
development environment for designing the
neural networks that the user does not see, but
which can create portable networks (as C code)
that can be retrained. The user must define the
problem by choosing appropriate depth intervals, lithological classes, logs and a neural
network. The results of running the network
are displayed graphically and in terms of relative
performance of the classification rate. There are
three main functional components to the system.
These are shown in Fig. 2 and are described in
detail in the following sections.

Performance measures
Having trained a neural network classifier, some
way of assessing its performance is required. The
standard way to do this is to take a separate subpopulation of core-classified samples from the
same general population and classify it independently with the network. The goodness of fit of
the two classifications (classification rate) gives a
measure of how well the network classifier
performs relative to the visual description
classification. If this performance is thought
satisfactory then the network can be run on the
full problem interval. What is 'satisfactory' in
this context is best left to the geologist. One,
albeit relative, benchmark by which to judge
satisfactory performance is to compare with
another classification technique. Again, the same

Fig. 1. Schematic structure of our computer system to


derive lithological logs from ODP core-log data.

132

G. WADGE ET AL.

Fig. 2. Functional schematic of the way data is


explored, selected and classified in our method.

Log data exploration


The purpose of log data exploration is to enable
the user to become familiar with the data before
making explicit choices. The ability to plot logs
against depth and cross plots of one log against
another is standard. In our system there is also
the ability to display simultaneously, selected
sample populations both in terms of their depth
and their position in log space. This is done by
manipulating a graphical cursor. It is a valuable
facility for deciding whether some 'extreme'
values in a cross plot, say, correlate with a
specific bed or are scattered throughout a
sequence. Principal components analysis and
unsupervised cluster analysis are also available
for any selection of samples and logs. These
exploratory analytical functions help decide the
following:
(1) what is the effective dimensionality of the
log data (i.e. how many distinct classes will
the data support)?;
(2) at what depth intervals do the most
representative samples lie?;
(3) are any samples obviously not represented
by core intervals with good depth matching?

Data selection
Explicit selections of depth interval, logs and
classes must be made. As we show later, the

neural networks tend to perform better with as


many 'useful' logs as possible and hence the
default is to use all logs. However, some logs
may only be available for restricted depth
ranges. Hence the choice would be between
fewer logs or more logs for a reduced interval.
Class selection is a more complex issue. There
are two main requirements: to be able to edit a
classification, to create a new classification
scheme and to give each such scheme source
information; who created it, when and how.
Editing can involve merging and splitting classes
and assigning new labels. Thus a library of
classifications can be created. Class labels are
assigned at each log sampling interval, nominally every 15 cm. The shipboard petrologists also
log sedimentary and structural discontinuities
some of which form class boundaries. In one
ideal situation, each thick (> > 15 cm) sedimentary bed would be of uniform lithological
character with sharp boundaries and have
contrasting neighbours giving the logs the
character of step functions across the boundaries. This ideal is the basis of log segmentation
algorithms (e.g. Vermeer & Alkemade 1992)
which seek to segment the borehole into uniform
intervals to which single (lithological) labels can
be attached. This can be helpful in constraining
classes in intervals of incomplete core recovery.
The data selection process creates a Log-Class
File (Table 1), whose values are used directly by
the neural network classifier.

Neural network classification


Samples from the Log-Class File are separated
into class populations and counted. The total of
the smallest population is then used to select
samples for training and testing. Classes with
larger p o p u l a t i o n s are subsampled evenly
throughout their range to give a total equal to
that of the smallest class. Each same-size class
population is then split into two sub-populations
by alternate sampling to give training and testing
sample populations for each class. The log
values of these populations are then examined
graphically to check that; the distributions of the
training and testing samples are similar, and that
the distributions of the core-classified samples
are representative of the whole interval under
investigation. If these conditions are not met
then other data selections must be made.
The neural network used is the feed-forward
back-propagation type which is standard for
classification problems. The selection of logs and
classes constrains the structure of the network.
The input layer consists of one node for each log
and the output layer consists of one node for

ODP LITHOLOGY USING NEURAL NETWORKS

133

Table 1. Representative part of a L O G - C L A S S file with 4 classes. Only 3 o f the logs are shown. There is no sample
classified as Conglomerate 2 in this selection shown

SGR

CGR

A1203

26.6793
26.8207
26.6378
12.6669
12.1269
11.7988
12.1286
11.8547
11.7594

24.7285
25.1096
25.287
9.3789
8.8892
8.7609
8.1325
7.9513
7.9992

21.4907
20.9049
20.0691
23.0037
21.9625
21.4303
19.5016
18.8318
17.9526

each class. There is at least one other, 'hidden',


layer of nodes between the input and output
layers with weighted connections between nodes
of different layers. The number of hidden nodes
is not externally constrained and can be changed
to suit a particular problem. The other network
parameters are rnainly related to the weightings
applied to the connections. These can be tuned
to improve performance, but detailed optimization of neural networks is a complex issue. Our
strategy is to make available a limited number of
default networks initially and optimize individually applied networks later.
Once the network has been selected it begins
training with the prepared training data by
selecting samples, at random, and presenting
their log values to the input layer of the network.
The effect of these values propagates through to
the output layer where the 'error' between the
network node values and the 'correct' values is
then propagated back through the network,
thereby changing network weightings. In this
way the network, after hundreds to thousands of
learning cycles, improves its ability to recognize
classes until no further improvement is achieved
and the network is said to be trained. The
trained network can now be tested by presenting
each of the test samples to the network once,
and recording how close the network result is to
the correct classes. The network gives proportional values for each class within the range 0-1,
whereas the core-derived class labels are binary
(0 or 1). The network results are thus essentially
probabilistic. We express the test result as a
thresholded classification rate. For example, at a
user-defined threshold of 70% probability, a test
result of 0.76 sandstone, 0.20 siltstone and 0.04
claystone for a sample with class values of 1,0,0
would count as a correct classification, though
not at a threshold of 80%. 92 such correct
results out of 100 test samples, for example,
would give a classification rate of 92%. If the

Clay

Sst

Cong 1

Cong2

1
1
1
0
0
0
0
0
0

0
0
0
1
1
1
0
0
0

0
0
0
0
0
0
1
1
1

0
0
0
0
0
0
0
0
0

outcome of the training and testing cycle is


considered satisfactory then the network can be
applied to the full interval under consideration.

Application to Hole 792E


Hole 792E of the ODP was drilled in the IzuBonin foreare sedimentary basin in 1989
(Taylor et al. 1990). The primary objective of
drilling at Site 792 was to understand the
stratigraphy of the forearc and the temporal
variations in sedimentation and volcanism that
controlled it. About 800m of sediment were
drilled above volcanic (andesite) basement,
including rocks of Pleistocene, !ate Pliocene,
Miocene and early Oligocene age. The lower
part of this section is a more volcaniclastic-rich
sequence. The interval between 482-732 m below
sea floor (mbsf) is the focus of this study and
comprises a large part of Unit IV, a sedimentary
succession of early Oligocene age (Fig. 3).
Overall, Unit IV is composed of vitric sandstone
(58%), sandy pebble-granule conglomerate (
10%), silty claystone (9%), nannofossil-rich silty
claystone (5%), claystone (4%), siltstone (4%),
nannofossil c l a y s t o n e ( l % ) , clayey siltstone
(1%), sandy mudstone (1%) and sandy siltstone
(1%). This unit is interpreted by Taylor et al.
(1990) as a rapidly deposited turbidite blanket in
an oversupplied basin or distal fan. The pumice
clasts of some of the coarse sandstones and
conglomerates indicates contemporary volcanism but most of the andesite and dacite clasts are
probably the result of erosion from the arc
volcanoes.
Core recovery

Hole 792E had a recovery of 48.2% of total


possible core length. Recovery for the 482-732
mbsf interval was 79.1%. However, as discussed

134

G. WADGE E T AL.
Downhole

Fig. 3. Summary graphical log of the stratigraphy of


the 482-732 mbsf interval of ODP Hole 792E (after
Taylor et al. 1990). The symbols represent: Inverted
'T'= nannofossil ooze, dashes=claystone-siltstone,
check = sandstone, dot-ellipse=conglomerate. Grain
sizes are represented by c=claystone, s=siltstone,
fs = fine sandstone, cs = coarse sandstone and g = gravel/conglomerate.

earlier, many individual cores contained only a


fraction of the full core length and hence cannot
be matched to the logs with confidence. We have
used only individual cores that have greater than
90% recovery to form the basis of our classification. These 13 out of 26 cores (Fig. 3) therefore
represent a useable recovery of 50%. Each of
these cored intervals have had their depth
assignments normalized to 100% recovery after
closing any gaps between core sections. Notice
that we have no matched core for the interval
617-655 mbsf.
Five lithological classes based on shipboard
visual description were used in the 482-732 mbsf
interval: claystone, silty/sandy claystone, muddy
siltstone/sandstone, siltstone/sandstone and
gravel/conglomerate. One of these five classes
was initially assigned to each (0.15m) log
interval. A number of thin (< 0.15 m) , mainly
claystone, beds were ignored in this process.
However, the claystone and muddy siltstone/
sandstone classes had so few samples (8 and 37,
respectively) that they were both merged with
the silt/sandy claystone class. For simplicity the
remaining three classes were called claystone,
sandstone and conglomerate.

measurements

The sediments in Hole 792E are well-indurated.


The borehole was close to cylindrical for much
of its depth with a diameter < 3 0 c m , and
conditions made for good-quality logs. There is
reasonable correlation between the log values
and shipboard core measurements except for
SiO2 content. The tools (sensors) used that are
relevant to this study were Resistivity (DEL),
Sonic (LSS), Natural Gamma (NGT), Geochemistry (GST and ACT) and Lithodensity
(HLDT) and they acquired data in 4 logging
runs, that were then depth-matched. Seventeen
log parameters were considered for use in the
classification: spectral gamma, computed gamma, radioactive potassium, thorium and uranium, deep, medium and shallow resistivity,
density, photoelectric effect, sonic velocity and
the oxide contents of calcium, silicon, iron,
titanium, potassium and aluminium. Unfortunately, no density correction and photoelectric
effect measurements are available below about
550 mbsf. We removed low-quality log data that
exceeded any of the following thresholds for the
three quality-control logs: Caliper (29.5cm
Density Correction (0.1 gm cc l) and Geochemical Factor (800). These thresholds were determined empirically by e x a m i n i n g the log
distributions. This reduced the number of
samples available by about 3%.
Within the 482-732 mbsf interval Taylor et al.
(1990) and Pratson et al. (1992) observed the
following relationships between log and core:
(1) Natural gamma spectrometry shows that
potassium is the dominant radioactive
source mineral and is inversely related to
values of resistivity, velocity and density,
and in some cases, grain size (482-500 and
555-585 mbsf). Uranium and potassium
contents are generally negatively correlated. Beds with high mud contents have
high values of potassium content and
natural gamma.
(2) Above about 515 mbsf the sequence has a
bimodal character with high resistivity/
velocity/density - low natural gamma beds
alternating with beds of opposite character.
Below this depth the logs lose their high
frequency nature and generally have raised
resistivity/velocity/density values.
(3) There is correlation between upward-fining
sandstone/conglomerate beds and a sawtooth response of resistivity in the 540-590
mbsf interval. Of particular importance is
the reduction in resistivity at 587 mbsf,
where the base of a large conglomerate bed

ODP LITHOLOGY USING NEURAL NETWORKS


Table 2.

Networkperformance results

Network
Number
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22

135

Network
Structure
1-717-3
15-15-3
15-15-3
15-15-4
4-8-4
5-15-4
6-15-4
11-10-4
15-02-4
15-04-4
15-06-4
15-08-4
15-10-4
15-12-4
15-14-4
15-16-4
15-18-4
15-20-4
15-25-3
15-30-3
15-25-10-3
15-25-20-3

Depth
Interval
(mbsf)

Classification
Total

Clay

Sst

Cong (1)

482-550
482-550
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
482-732
483-732
482-732
482-732

88.4
85.5
75.8
90.0
78.1
80.0
81.9
86.3
71.9
82.5
83.1
86.1
86.9
86.9
88.1
86.9
89.4
88.1
78.0
76.6
80.2
78.8

87.0
87.0
83.5
92.0
85.0
82.5
87.5
85.0
87.5
90.0
85.0
87.5
90.0
90.0
85.0
92.5
92.5
90.0
86.8
79.1
84.6
82.4

78.3
73.9
59.3
80.0
47.5
52.5
67.5
65.0
92.5
47.5
60.0
85.5
62.5
72.5
72.5
70.0
72.0
77.5
57.1
69.2
72.5
65.9

100
95.7
84.6
87.5
82.5
85.0
72.5
95.0
12.5
92.5
87.5
75.0
95.0
85.0
95.0
85.0
95.0
85.0
90.1
81.3
83.5
87.9

also marks the d o w n h o l e increase in


smectite concentration and magnetic susceptibility.

Performance of different neural networks


There is no one single neural network that will
work for all classification problems. The basic
type used here, whose structural variants we now
discuss, is a back-propagation network with a
single hidden layer and a sigmoidal activation
function at each node. As part of the preprocessing for neural network training the data
from each log that are input to the network are
normalized to the range 0-1. Logs with large
potential ranges, such as the resistivity tools,
may be best converted to a logarithmic scale
first. However, the resistivity values of Hole
792E did not show a very great range and this
was not done. As was discussed earlier, the
number of nodes in the input and output layers
is at least partly determined by the nature of the
data and the problem to be solved; the number
of nodes in the hidden layer is chosen to
optimize performance once the input and output
layers are fixed. In the performance results
reported in Table 2 we use a convention in
which a 15-10-4 network means 15 input nodes
(downhole logs), 10 hidden nodes and 4 output
nodes (lithological classes).

Rate(%)
Cong(2)

100
97.5
100
100
100
95.5
100
100
100
100
100
100
100
97.5
100

There are 17 logs available for the interval


482-550 mbsf, but only 15 (no density and
photoelectric logs) were considered for the full
interval between 482-732 mbsf. Using the basic
rock type classification, the 482-550 mbsf
interval has 98 claystone, 164 sandstone and 46
conglomerate samples available for training and
testing. Network 1 (Table 2) then, is a 17-17-3
network used for 482-550 mbsf that gives a total
classification rate of 88.4%. Using only 15 logs
for the same interval in network 2 gives a
reduced rate of 85.5%. Hence density and
photoelectric effect logs do have some extra
capability to discriminate between these rock
classes in addition to that present in the other 15
logs. However, when we use the same type of
network for the 482-732 mbsf interval, but
trained using the samples from a 183 claystone,
407 sandstone and 185 conglomerate pool of
samples (network 3), the classification rate falls
to 75.8%. This means that the claystones a n d s t o n e - c o n g l o m e r a t e classification that
worked well from 482-550 mbsf is much less
appropriate below 550 mbsf. There is a distinct
fall in the capability of the network to recognize
the conglomerate and sandstone class samples.
Exploring the log data over this wider interval
suggests that core samples classed as conglomerate may be usefully split into more than one
type. For example, principal components analy-

136

G. WADGE E T AL.
100

60
x

90

50

80

40
x

70

30

60

Classification Rate

Computing Time

20

10

12

14

16

Number of Logs
Fig. 4. Plot of classification rate performance and elapsed computing time versus increasing numbers of
geophysical logs as input to the neural networks (networks 4-8).
B

90
ra

88

ra

86
84
82
o

80
78

76
.--4

74
72
70

10

12

14

16

18

20

Number of Hidden Nodes

Fig. 5. Plot of classification rate performance versus the number of nodes in the hidden layer (networks 9-18).
sis of shallow resistivity, sonic velocity, spectral
gamma, computed gamma and potassium oxide
logs shows that some of the conglomerate class
samples give much lower values of principal
component 3 and higher values of principal
component 4 than the other conglomerate
samples. Thus we have created a second
classification scheme with four classes: claystone, sandstone, conglomerate 1 and conglom-

erate 2. Conglomerate 1 includes those samples


classed as conglomerate in networks 1 and 2;
conglomerate 2 (the samples described above)
corresponds to core which is conglomeratic but
has higher proportions of large ( > 5 cm) claystone clasts than conglomerate 1. In the logs,
conglomerate 1 has distinctly lower spectral
gamma and potassium values than those of
conglomerate 2. Sonic values are also lower in

ODP LITHOLOGY USING NEURAL NETWORKS

137

Fig. 6. Component lithological classification using neural network 4. The four columns display the component
contribution of the rocks types (claystone, sandstone, conglomerate 1 and conglomerate 2) for each sampling
interval of the logged hole. The depth in mbsf is shown on the left and the core number on the right of each
column. The same colours are also used to display the classification in the visual core descriptions in the narrow
column to the left of the core numbers.
conglomerate 1 but the separation is less
distinct. With four classes, network 4, otherwise
equivalent to network 3, gives a much improved
classification rate for the whole 482-732 mbsf
interval of 90.0%.
We saw an example of improvement in
performance with increased numbers of logs
(from network 1 to 2) for the 482-550 mbsf
interval. For the full interval with 4 classes it is
also generally true that the more logs input to
the network the better the performance (networks 4, 5 to 8 inclusive; Fig. 4). There is a
penalty to pay in terms of increased computing
time (Fig. 4) but this is not too great a burden.
Hence it makes sense to use as many useful logs
as are available at the outset.
Network 4 has a 15-15-4 structure. The

justification for the 15 nodes in the hidden layer


is provided by a systematic test of performance
in a series of networks with variable numbers of
hidden nodes (networks 4, 9 to 18 inclusive; Fig.
5). The optimum configuration of the hidden
layer of 15 nodes is given by the maximum
performance value. In this case the number of
hidden nodes equals the number of input nodes.
This would be a useful rule-of-thumb for initial
network configuration but it does not guarantee
the optimum solution. For instance, as can be
seen from networks 3 and 19 to 22 inclusive,
greater numbers of hidden nodes (and even a
second hidden layer) can give improved performance, but at the cost of increased computing
time.

138

G. WADGE ETAL.

Computed lithological log results


Figure 6 shows the output of network 4 when
applied to the whole 250 m interval of downhole
logs from 482-732 mbsf. The aggregate thicknesses of the lithologies according to this
classification are: c l a y s t o n e = 4 4 m , sandstone = 130 m, conglomerate 1 = 42 m, conglomerate 2 = 34 m.
Some sample intervals are classed as 100% of
a particular lithology but many are classed as
mixtures, with one dominant lithology and one
or more minor components. This is particularly
noticeable for mixtures of sandstone and conglomerate 1, but much less so for mixtures
involving the other two lithologies. In Fig. 7
the results from this network have been recast
such that the major ( > 5 0 % ) lithology in the
output is attributed to that depth interval. This
gives a columnar plot that mimics a traditional
Lithological log and allows direct comparison
with the visual core classification.
The change in character of the rocks at about
515 mbsf noted above is apparent but is overshadowed by the decrease in claystone which
occurs about 20 m higher. Lower down the hole
the occurrence of conglomerate 2 is restricted to
3 zones where there is an apparent association of
claystone-conglomerate 2. The base of the
shallowest of these zones corresponds to the
major break noted at 587 mbsf, though there is
no obvious change in general lithological character below this. The log gives the impression of
three major cycles of mixed conglomerate and
sandstone sitting above conglomerate 2 and
claystone (515-587, 587-660, 660-732 mbsf).
The second thickest interval classed as conglornerate 2 is from around 640 mbsf, where there is
no core recovery. In fact there is no sense of this
second cycle in the recovered core. The lowest of
the three cycles is richer in sandstone at the
expense of conglornerate 1.
As discussed earlier, the rocks recovered from
only 50% of the full 482-732 mbsf interval were
used to train and test the neural network. Figure
7 displays these core (core numbers 37-39, 40,
42-43, 46, 48-50, 56-57, 60, 61) together with
the additional recovered core, that comprised
29% of the total interval, that was not used in
training and testing the network because of poor
depth control (core numbers 39, 41, 44-45, 47,
51-55, 58-59, 61). For many of the intervals used
in the training and testing there is a high degree
of detailed correspondence between the core and
the network classifications (e.g. core numbers
37-38, 50, 62) as we would expect. For a few, the
correspondence is weaker (e.g. core 43). For the
core intervals with < 90% recovery any corre-

spondence is more difficult to assess. For some


cores such as 39 and 55, correspondence could
be achieved by appropriate expansion of the
core lithology down section to match the
network lithological log results. For other
intervals no such a c c o m m o d a t i o n can be
achieved (e.g. core numbers 45, 52 and 59).
The two main possible explanations of this are
that the network is 'overtrained' and has lost its
ability to generalize when exposed to new data,
and that the classification scheme is not optimal.
The way ahead in our system would be to
explore the second possibility by choosing
another classification scheme. Because the sandstone group of samples is the largest and shows
the poorest general classification performance
figures (Table 2) this is the most likely group for
possible splitting into two or more classes. Some
of the geochemical logs such as TiO2 and CaO
show clear evidence of alternating high and low
valued sandstone horizons (e.g. 670--682 mbsf)
that could form one of the criteria of such a new
classification scheme, though this is not pursued
here.

Discussion
We have chosen to ignore a number of major
issues of core-log-driven classification including
graded bedding, the differences in spatial resolving power of the logs and the use of segmentation, in order to emphasize the value of quality
control of the data and careful consideration of
the optimal structure of the neural network. In
particular, we wish to stress the need to have a
flexible mechanism for changing the classification scheme of rock types based on the
information content in the logs and the shipboard-derived classification scheme. Such an
approach is inherently hole-specific. It lies at
an intermediate position between a totally
empirical approach, driven solely by the log
data, and one that might use a universal library
of log responses derived from fundamental core
components (e.g. sand, carbonate, sea-water
etc.). If the data can support it, the refinement
of the classification scheme in a hole will be
essentially hierarchical. However, there is no
guarantee that the way that the log data can be
optimally divided will correspond to the classification scheme that the geologist wants or
expects. This sort of approach should be of
value to ODP scientists both on and off the ship.
Results of our work using data from other ODP
holes will be presented elsewhere. We also
envisage that this technique might be of value
for providing rapid lithological analysis of

ODP L I T H O L O G Y U S I N G N E U R A L N E T W O R K S

139

Fig. 7. Majority component lithological log output of the neural network 4. The main column is the network
classification, the narrow column to the right is that classified from recovered core.

G. WADGE ET AL.

140

piston cores from which horizontal-track logs


have been collected o n - b o a r d ship.

This work is funded by a grant (GST/02/993) to RBW


and GW under the NERC Special Topic--UK ODP
Science Programme. ESSC work is supported by
NERC grant F60/G6/12/02. We are very grateful to
our collaborators Drs P. Harvey and H. Grubb, for
their help and the Borehole Research Group at LDEO
and ODP/TAMU for supplying data.

References
AGRINIER, P. & AGRINIER, B. 1994. A propos de la
connaissance de la profondeur a laquelle vos
echantillons sont collectes dans les forages.

Comptes Rendus de la Academie Sciences de Paris,


318, serie II, 1615-1622.
BALDWIN, J. L., BATEMAN,A. R. M. & WHEATLEY,C.
L. 1990. Application of neural networks to the
problem of mineral identification from well-logs.
The Log Analyst, 3, 279-293.
DOVETON, J. H. 1994. Geologic log analysis using
computer methods. Computer Applications in
Geology, 2. American Association of Petroleum
Geologists, Tulsa.

GONCALVES, C. A. 1995. Characterisation of formation


heterogeneity. PhD Thesis, University of Leicester.
MAZULLO, L, MEYER, A. & KIDD, R. B. 1987. A new
sediment classification scheme for the Ocean
Drilling Program. ODP Technical Note, 8.
ODP 1990. Wireline Logging Manual, Ocean Drilling
Program. Borehole Research Group, LamontDoherty Geological Observatory.
PRATSON, E. L., REYNOLDS, R., LOVELL, M. K.,
PEZARD, P. A. & BROGLIA,C. 1992. Geochemical
well logs in the lzu-Bonin arc-trench system, Sites
791, 792, and 793. Proceedings of the Ocean
Drilling Program, Scientific Results, 126, 653-676.
ROGERS, S. J., FANG, J. H., KARR, C. L. & STANLEY,D.
K. 1992. Determination of lithology from well
logs using a neural network. American Association
of Petroleum Geologists Bulletin, 76, 731-739.
TAYLOR, B., FUROKA, A. & OTHERS 1990. Proceedings
of the Ocean Drilling Program, Initial Results, 126.
VERMEER, P. L. & ALKEMANDE, J. A. H. 1992.
Multiscale segmentation of well logs. Mathematical Geology, 24, 27-43.
WONG, P. M., JIAN, F. X. & TAGGART, I. J. 1995. A
critical comparison of neural networks and
discriminant analysis in lithofacies, porosity and
its permeability predictions. Journal of Petroleum
Geology, 18, 191-206.

Core-derived acoustic, porosity & permeability correlations for


computation pseudo-logs
A. C. B A S T O S , L. D. D I L L O N , G. F. V A S Q U E Z & J. A. S O A R E S
Petrobras Research Center-SEGEST,

C i d a d e Universitaria - Q . 7 - P r e d i o 20, Ilha do

F u n d a o - R i o de Janeiro, 2 1 9 4 9 - 9 0 0 , B r a z i l

Abstract: In order to improve hydrocarbon production, it is often necessary to obtain more

accurate rock, fluid and petrophysical information. For example, to obtain a reservoir
porosity map using seismic data as reference, it is necessary to generate reliable correlations
between seismic attributes and petrophysical properties like porosity and permeability.
Again, to optimize drilling and/or hydraulic fracturing programs, it is also necessary to
estimate better formation static mechanical behaviour from geophysical data. The main goal
of this work is to establish for an offshore Brazilian field, relationships between
compressional and shear wave velocities and petrophysical properties such as porosity
and permeability.The large number of limestone samples (120) gave us a precise empirical
relationship between Vs and Vp for limestone. In order to obtain a calibration reference, we
also made, with the same samples, simultaneous measurements of dynamic and static elastic
constants. Using all these laboratory relationships, it was possible to generate unmeasured
pseudo-logs of in situ parameters, which include: shear wave velocity, static and dynamic
elastic constants and permeability. The good experimental relationships obtained between
k-~b and Vp-~b in this work together with available logs give us an additional method to
estimate permeability which is impossible to obtain from in situ measurements.

Indirect generation of unmeasured in situ logs


like shear wave velocity (Vs), permeability (k)
and elastic constants (Young (E), shear (G) and
bulk (K) modulus) have been the subject of
various works in geophysics (Wendt et al. 1986;
Castagna et al. 1993; Bastos et al. 1995; Tang et
al. 1996). In this paper we present, for three
Brazilian offshore wells, a generation procedure
for Vs, k and elastic constants logs calculated
from laboratory data: Vp, Vs, porosity (~), (k)
and static and dynamic elastic constants on
cores. The importance of the generation of
unmeasured in situ logs includes the possibility
of obtaining more accurate information about
lithology and fluid content in reservoir rocks
and, in this way, contributing to generating
more reliable AVO and seismic models, and also
optimizing drilling and hydraulic fracturing
programes. For reservoir development, these
kind of data are also helpful for generating
correlations between seismic attributes and
petrophysical properties and for monitoring
subsurface fluid flow. So, our main goal in this
work was to:
(1) obtain empirical correlations between Vs
and Vp from laboratory data in order to
generate unmeasured Vs logs from measured Vp logs;
(2) generate logs of static and dynamic elastic
constants using the simultaneous labora-

tory measurement of static and dynamic


elastic constants as a calibration reference;
(3) obtain empirical correlations, for each well,
between Vp, k and q~, thereby yielding a
calculated permeability log.

Methodology
Ultrasonic P and S wave velocities were measured in about 120 samples of limestone from an
offshore Brazilian field. These samples were
retrieved from three vertical wells at depths of
about 2350m to 2550m and vertically cut as
right cylindrical plugs with diameter 2.5 cm and
3.75 cm and length 3.75 to 5cm. The measurement frequency was 500 kHz for both Vp and Vs
and over a range of confining pressure of 1000
psi to 5000 psi at room temperature. The
porosity and permeability range were 5% to
35% and 0.1 mD to 1800mD, respectively. The
same measurements were made under dry and
formation water saturated conditions. However,
the results showed only small variations due to
saturation, as noted by Bastos et al. (1995).
Simultaneous measurements of static and
dynamic elastic constants were made on some
samples of diameter 5cm and length 12.5cm.
These samples were placed in a triaxial cell and
subjected to an in situ confining stress of about
5000 psi, and to a deviatoric stress which was
increased up to the sample failure. The deforma-

BASTOS, A. C. DILLON, L. D. VASQUEZ,G. F. & SOARES,J. A. 1998. Core-derived acoustic, porosity


& permeability correlations for computation pseudo-logs In." HARVEY,P. K. LOVELL,
M. A. (eds) Core-Log Integration, Geological Society, London, Special Publications, 136, 141-146

141

142

A . C . BASTOS E T A L .
4000

tion related to the increasing deviatoric stress


allowed us to determine the static constants. The
dynamic constants are obtained simultaneously,
by monitoring changes in transit time.

Vm - 0 , 5 5 V p + 4 1 , 6 0 c c " 0,96

Procedure and results

Calculated logs o f Vs and Elastic constants

F
1000

2OOO

4O0O

600o

Vp (m/s)

Fig. 1. Vs vs Vp for limestone plugs.

The three wells that are the subject of this work


do not have in situ Vs logs. Therefore, a
laboratory relationship was obtained between
Vs and Vp in order to generate a pseudo Vs log.
Figure 1 shows the linear fit to the Vs-Vp cross
plot (equation 1). As shown in this figure, an
excellent correlation was obtained with a correlation coefficient of 0.96. For the case of these
samples, this linear fit was better than the
WELL - B

WELL - A
Velocity (m/s)
1000 2000 3000 4000 5000 6000
2380

WELL - c

Velocity (m/s)

Velocity (m/s)

1000 2000 3000 4000 5000 6000

1000 2000 3000 4000 5000 6000

2390
2400
2410
242O
2430
2440
A

.....

245O

Q
2460
247O
2480
249O
2500
2510
2520

Vp log

Vs log

9 Vp lab

9 V s lab

Fig. 2. Vp and Vs from laboratory data (symbols) and calculated Vp and Vs logs (curves). The three wells show
good agreement between laboratory and log data.

CORE-DERIVED COMPUTATION OF PSEUDO-LOGS


'

'

'

'

'

'

'

'

..y

6O
,..,.

,-. 4O

o
9

20

2O

0
20

40
E d y n (GPa)

60

"2d,:,

80

(c)

(B)

(A)

I11

143

2O

20

40
K d y n (GPa)

(P

60

"

80

2O
G d y n (GPa)

40

Fig. 3. Static and dynamic elastic constants for sedimentary rocks obtained from simultaneous laboratory
measurements.
WELL
O (GP=)
0

10

20

30

40

50

- C
E (GPa)

K (GPa)
60

70

80

10 20

30

40

50

60

70

80

2390
I

20

30
,

40

i ~ i

g_%,

50
, i

60
, i

, ,

- i ~ 1
9

2440

i"

2450

2460

2470

2480

',

2500

2510
L

,k, >

2520

,i,},,

''
~ L

2490
b ~

I
I

2430

2420

7O 8 0
v } i

2410

I~

10
,

2400

I
I

:l
-!

~
I

,~

)l

!, ~,

N ~1

i,l,I,

,
I,I,

)
I

, ,
i

,t,t,

I,

Fig. 4. Calculated static and dynamic elastic constants logs for weU-C.
polynomial fit proposed by Castagna et al.
(1993) even for values of Vs close to 1500ms -1.
The regression algorithm is:
Vs = 0.55 Vp + 41.60

2). There is good agreement between pseudologs and laboratory data. With the Vp, Vs and
density (p) logs and the following elastic theory
equations:

(1)

Using this relationship and the in situ Vp logs


it was possible to calculate Vs pseudo-logs (Fig.

v~ = d
/ K
v

+ 4#/3

Pb

(2)

144

A.C. BASTOS E T AL.


WELL - A

6000F

I
5000

'

Vp

'

- 5868

'

'

--e
~176

'

1o L ,

' 0116 '10

8 ~ K'~0"05e

cc ,, 0.88

'

'

'/,

;cc=0"90 9

4000

../

3000

2000

10

15

20

Porosity

25
(%)

'

30

35

10

15

20

25
(%)

Porosity

30

35

WELL - B

6000

'

'

-o.026
t4 e

'

5 --

'

K"

0.179
0.024 e
; cc = 0.77

; cc ,, 0.93

~o~

5000

'

'

'

'

'

,ooo t-

1
9
2000

10

15

20

Porosity

25
(%)

'

..
'

'

.~4000
D,.

10

'

20

Porosity

Porosity

20

25
(%)

'

'

'

0.4 6 El

K = 0.0002 e

; c c ,, 0 . 8 9

i
15

15

25
(%)

30

30

35

- C

2000

3000

2000

10

35

V p =, 5 8 2 2 e0.021

5000

'

30

WELL

6000

'

; cc =0.93

i000

35

10

15

20

Porosity

25
(%)

30

35

Fig. 5. Exponential fits of velocity and permeability versus porosity for three wells.

V, =

~/~
#

(3)

where: K is the bulk modulus, # is the shear


modulus and Pb is the Bulk density, pseudo-logs
of dynamic elastic constants have been calculated,
In fact, to optimize drilling and/or hydraulic
fracturing programs, it is often necessary to

obtain logs of the static and not the dynamic


elastic constants. For this purpose, we use
simultaneous laboratory measurements of static
and dynamic constants in order to transform the
dynamic to the reference static. Figure 3 shows a
cross plot between static and dynamic constants
for sedimentary rocks and illustrates strong
empirical relationships which are expressed
mathematically as follows:

CORE-DERIVED COMPUTATION OF PSEUDO-LOGS


WELL- B

WELL - A

Permeability (mD)
0

145
WELL-C

Permeability (mD)

10

100

Permeability (mD)
10

2380

10

[ 1 1 1 1 ~

100

10o0 10000

I I1111llI I Illllll~

I Illllll

q~,,,,d

2390
2400
2410
2420
2430
2440

I
I

2450
2460

2470

r,,

2480

_s

2490
2500
2510
2520
2530
254O
2550

v4

2560
2570
2580

, ,,,,id

L ,,,,,,,I

~,,,,,,l

,,,

.....

,.I

Fig. 6. Calculated permeability logs obtained from laboratory k-Vp relationship show a good correlation for
wells A and C, but less so for well B. The crossed points in well A were not used to develop the k-Vp relationship.

Estat = 0.675 Edyn -- 3.84; correlation


coefficient --- 0.95

(4)

Kstat = 0.992 Kdyn -- 8.82; correlation


coefficient---- 0.89

(5)

Gstat = 0.621 Gdyn -- 0.95; correlation


coefficient = 0.94

(6)

where the subscripts 'stat' and 'dyn' denote


static and dynamic moduli, respectively.
Figure 4 shows the calculated log of static and
dynamic constants obtained from equations (4)
to (6) for well C. As expected, the logs of the
dynamic elastic constants show higher values
than their static equivalents.

Calculated logs of permeability


The next step was to calculate permeability logs
for the three wells using algorithms based on

laboratory permeability, porosity and velocity


data. Figure 5 shows core data, the cross plots of
velocity against porosity, and permeability
against porosity. From these plots it has been
possible to deduce a relationship between Vp, ~b
and k. As can be seen in Fig. 5, an exponential fit
was the best one obtained for both the Vp-q$ and
the k-~b relationships for the three wells. Thus,
with the equations obtained (equations (7), (8),
(10), (11), (13) and (14)) we can isolate qb from
Vp-q) and k-q5 relations and then obtain k-Vp
relationships (equations (9), (12) and (15)) which
can be used to calculate the k-log shown in Fig.
6. In order to check these relationships we
include some points in well A (cross points in
Fig. 6) which were not used to obtain equations
(7) to (15). Again, it can be seen that there is a
good correspondence between these points and
the obtained log:
Well A
Vp = 5868e~~

cc = 0.88

(7)

146

A.C. BASTOS E T AL.


k = 0.05e~
k :

cc = 0.90

(9)

Vp 73.e6~

Well B
Vp = 6214e-~176
cc -- 0.93
k = 0.024e~

(8)

cc -- 0.77

k = g p -6"54. e 53"35

(10)
(11)
(12)

Well C
Vp = 5822em~

cc = 0.89

(13)

k = 0.0002e~

cc = 0.93

(14)

k : Vp-184.e152.5

(15)

Figure 6 shows a good correspondence for


wells A and C, but not for well B. In this well the
good correlation coefficient for the k-qb relationship, 0.77, was lower.

Conclusions
(1) The large number of limestone samples
gave us a precise empirical relationship between
Vs and Vp for limestone, and this differs from
the earlier work of Castagna et al. (1993), even
for Vs close to 1500 ms -1.
(2) Good relationships between static and
dynamic elastic constants were obtained for
sedimentary rocks, and these have allowed us
to generate logs for these constants. As expected,
dynamic constants are greater than static ones.
(3) The capacity to obtain a relationship
between static behaviour of rocks from dynamic
properties combines the advantages of both
methods in one. Thus, the resultant properties

of this relationship give us the static mechanical


behaviour, characteristic of production engineering, but, with the continuous character of
geophysical logs.
(4) Cross plots between Vp-qb and k-~b
indicated good exponential fits for the three
wells that formed the subject of this work;
(5) The good experimental relationships obtained between k-qb and Vp-qb (see correlation
coefficients in equations (7) to (15)), together
with available logs give us an additional method
to estimate permeability.
(6) There is good agreement between laboratory permeability measurements and synthetic
permeability logs from velocity data, even for
points that were not used in the generation of
these pseudo-logs.

References
BASTOS, A. C., DILLON, L. D., SOARES, J. A. &
VASQUEZ, G. F.. 1995. Estimativa dos perils de

constantes elfisticas em carbonatos pouco permefiveis a partir de dados laboratoriais. 4th


International Congress of the Brazilian Geophysical Society and the 1st Latin American Geophysical Conference. Volume II.
CASTAGNA, J. P., BATZLE, M. L. 8r KAN, T. K. 1993.

Rock Physics: The link between rock properties


and AVO response. In: CASTAGNA, J. P. t~
BACKUS,M. M. (Eds) Offset-dependent reflectivity:
SEG, 124-157.
TANG, X. CHENG, C. H. 1996. Fast inversion of

formation permeability from Stoneley wave logs


using a simplified Biot-Rosenbaum model. Geophysics, 61, 639-645.
WENDT, W. A., SAKURAI, S. t~ NELSON, P. H. 1986.

Permeability prediction from well logs multiple


regression. In: LAKE, L. W. & CARROLL, H. B. Jr

(eds) Reservoir characterization. Academic Press,


San Diego, California, 181-221.

Effects of water salinity, saturation and clay content on the complex


resistivity of sandstone samples
P. S. D E N I C O L 1 & X. D. J I N G

Centre for Petroleum Studies, Imperial College of Science, Technology and Medicine,
London S W 7 2BP, UK
1Present address." Petrobras S.A., Exploration Department, 27913-350, Macae, R J, Brazil

Abstract: Complex resistivity measurements were made on sandstone samples in the


frequency range from l0 Hz to 2 MHz. The main objective was to investigate the frequency
response of complex resistivity and phase angle as a function of salinity, water saturation
and clay content. The results showed the classical frequency dependence behaviour where
the complex resistivity decreases with increasing frequency. The complex impedance
behaviour in the intermediate frequency range (10-100 kHz) was used to relate the effect of
frequency dispersion with interface polarization and, hence, pore geometry, specific surface
area and permeability.
Both water saturation and salinity were found to influence the gradient and the relaxation
frequency of the complex resistivity versus frequency relationship. A variation in water
saturation from full to partial saturation resulted in a dramatic increase in the gradient and
a clear shift of the relaxation frequency. Both the saturation and salinity dependence can be
attributed to the polarization of both the rock-fluid and fluid-fluid interfaces within the
pore space, which depend on the geometry and physical characteristics of the interfacial
layers. The results presented in this paper can have important applications in identifying low
resistivity and low contrast pay zones.
The complex electrical behaviour of a rock
results from its conductive and dielectric response in the presence of an electric field; the
former is related to the transport of free charge
carriers and the latter is due to geometrical,
interfacial and electrochemical mechanisms (Sen
1980, 1981). A complex impedance vector (Z*)
consists of a real part ( in-phase or resistance, R)
and an imaginary part (out-of-phase or reactance, X). Using the rectangular-coordinate
form, the complex impedance can be expressed
as follows,
Z*= R + jX

(1)

where j = v / - 1 is the complex operator. The


phase angle (0) by which current and voltage are
shifted is given as:
0 = tan I(X/R)

(2)

The complex resistivity p* can be calculated


from Z*,
p* = Z* A/L = p ' + j p "

(3)

where A is the cross-sectional area of the sample


and L is its length, and p' and p" are real and
imaginary parts of the complex resistivity,
respectively. In some cases, using the reciprocal

of impedance is mathematically expedient. For


example, when the real and imaginary components are paralleled, it is better to use admittance
(Y*),
Y* = 6+ jB

(4)

where G is the conductance and B is the


susceptance. The complex conductivity or* can
be calculated from Y*,
or* = Y* L/A = or'+jcr"

(5)

where a' and or" are the real and imaginary


conductivities, respectively.

Background
Complex electrical impedance measurement is a
non-invasive technique where an electrical current flows through the sample at different
frequencies. Experimental measurements of the
electrical properties of rocks, when submitted to
an alternating electrical field at different frequencies, have shown that both the resistive and
reactive components of the complex impedance
vary over the frequency spectrum. These two
features (complex quantity and dispersion or

DENICOL,P. S. & JING, X. D. 1998. Effects of water salinity, saturation and clay content on the
complex resistivity of sandstone samples In: HARVEY,P. K. LOVELL,M. A. (eds)
Core-Log Integration, Geological Society, London, Special Publications, 136, 147-157

147

148

P.S. DENICOL & X. D. JING

Table 1. List of petrophysieal parameters and chargeability at full and partial water saturation.
Sample

Density
grams cm-3

Porosity
%

Kair
mD

Saturation
Sw(%)

Chargeability
Partial Sat.

Chargeability
Full Sat.

Z1
Z3
Z4
Z5
Z7
Z8
Z9

2.66
2.66
2.64
2.65
2.71
2.69
2.64

22.4
14.1
12.2
20.1
25.8
21.5
29.1

1760
38.7
3.51
163
23.1
101
819

33
39
61
47
77
82
44

0.74
0.81
0.68
0.55
0.54
0.52
0.53

0.53
0.53
0.54
0.52
0.53
0.54
0.50

Table 2. List of synthetic shaley samples


Sample

Clay Type

Clay
Length
Content cm

Area
cm2

Grain
Porosity
Density
%
grams cm-3

Kair
mD

SZ1
SZ2
SZ3
SZ4

clean
montmorillonite
montmorillonite
montmorillonite

0
5
10
15

10.75
10.75
10.75
10.46

2.65
2.66
2.66
2.66

337
235
146
105

6.41
6.49
6.53
6.37

frequency dependence) can be used to estimate


rock petrophysical properties, such as specific
surface area and permeability.
The origin of the frequency dependence can be
related to geometrical effects of the clay particles
(Sen 1980) or electrochemical phenomena at the
fluid-grain (Rink & Schopper 1974) and/or
fluid-fluid interface (Knight & Endres 1991).
The interface region between matrix and the
fluid-filled pore space is complicated due to the
existence of the ionic double layer. The concept
of the electrical double layer forms the theoretical basis for understanding the electrical
properties of rocks, especially shaley sandstones.
Electrochemical theory suggests that the surface
of clay minerals carries excess negative charges
as a result of the substitution of certain positive
ions by others of lower valence. When the clays
are brought in contact with an electrolyte, these
negative charges on the clay surface attract
positive ions and repulse negative ions present in
the solution. As a result, an electrical ionic
double layer (or diffuse layer) is generated on the
exterior surface of particles. Typical distribution
for ionic concentration and electric potential can
be predicted by the Guoy (1910) theory. The
Gouy theory also predicts that the double-layer
thickness (Xd) is reduced as the concentration of
the bulk solution increases. The region outside
the electric double layer (distance > Xd) is called
the free-water region.
Ionic double layers exist between rock and
fluid interfaces. The perturbation of the double

29.1
28.1
27.4
27.9

layer by an oscillating electrical field is usually


accepted (Lima & Sharma 1992) as the main
mechanism for the frequency dependence of
rocks. Therefore, this interface polarization may
provide a link between complex resistivity data
and pore-scale attributes, such as pore geometry
and specific surface area, which in turn can be
related to rock permeability through a KozenyCarman type of relationship ( Borner 1995;
Denicol & Jing 1996). Since the frequency
dependence is reflecting interface phenomena,
salinity of the pore water also influences the
dispersion due to the variation in the double
layer thickness and ion mobility. Furthermore,
fluid saturation also plays a role due to addition
of the water/oil interfacial area, an increase in
the tortuosity of the brine-phase distribution
and the presence of a non-ionic fluid. The main
objective of this paper is to investigate, experimentally, the effects of brine salinity, fluid
saturation and clay minerals on the complex
impedance of different rock samples with varying porosity and permeability. The samples
include outcrop cores, oil-field reservoir rocks
and synthetic shaley rocks (Tables 1 and 2). A
brief geological description of all sandstone
samples is given in the Appendix.

Experimental apparatus and procedures


Complex impedance measurements were performed using a multi-sample rock testing system. The apparatus can accommodate five

Fig. 1. Schematic representation of the experimental apparatus.


samples simultaneously under varying hydrostatic confining pressure, temperature and independently controllable pore pressure. Since all
the samples are under the same conditions of
pressure and temperature, it eliminates experimental comparison errors due to fluctuations
during the period of testing (Jing et al. 1992).
The experimental system is shown schematically
in Fig.1. Complex impedance measurements
were made using the frequency response analyser (QuadTech Model 7600 RCL) in the
frequency range of 10 Hz to 2 MHz. The
instrument is capable of compensating for the
residuals of test fixture and cables based on the
open/short circuit compensation technique in
the whole frequency range. The instrument is
equipped with four coaxial BNC terminals on its
front panel which locate its calibration plane.
The calibration plane is the position where the
instrument measures within its specified accuracy (0.05%). In our experiment, test fixture and
cables were used to interconnect the sample to
the instrument in a four-terminal configuration
(4T). The parasitics related to test fixture, cables
and connections are frequency dependent and
they were minimized using the 4T configuration
and by applying the open/short compensation
technique in the whole frequency range similar
to the technique used by Taherian et al. (1990).
The effect of salinity was investigated for two
brine concentrations: 20 g and 50 g of sodium
chloride (NaCI) per litre of solution (i.e. 2% and
5% NaC1). The solution is made up of NaC1
dissolved in de-aerated and de-ionized distilled
water. Initially, the rock samples were fully
saturated with 2% brine solution and loaded in
the test cell. The samples were considered fully
saturated when the resistance of the samples

measured continuously at 2KHz frequency,


showed no significant variation (i.e. < 1%
change over a period of 12 h) with brine
displacement. The RCL meter was then connected and a frequency sweep performed on
each sample. After the frequency measurement,
5% brine solution was injected through the
samples to displace the original brine. The
resistance was observed continuously. A sharp
decrease was observed during the first few pore
volumes of displacement, when the more conductive brine became continuous. Then, the
decrease was less accentuated and reached
equilibrium after about 20 pore volumes of
injection. A frequency sweep was then repeated
at 5% brine salinity.
In order to study the frequency dependence of
partially saturated rocks, the desaturation technique using semi-permeable capillary diap h r a g m s has been used f o l l o w i n g the
laboratory procedures described by Elashahab
et al. (1995). The main advantages of the method
are the reduction of capillary end effects and
uniform saturation distribution along the core
length. These improvements are achieved by
using highly hydrophilic ceramic membranes
positioned between the sample and the end
plate. The resistivity distribution along the core
is monitored by six potential electrodes equally
spaced along the rock sample so that resistivity
measurement can be taken at pairs of electrodes
(four-electrode configuration) and also between
the top and base current electrodes which give
the total resistivity (Fig. 2). The resistivity
measurements for saturation monitoring based
on the Archie type of equations are taken at a
frequency of 2 kHz. The volume of brine
produced during the desaturation process was

150

P. S. DENICOL & X. D. JING

Fig. 2. Core sleeve with multiple electrodes.

Fig. 3. General frequency dependence behaviour for sample Zl.


carefully measured to allow the calculation of
average sample saturation by material balance.
The effect of clay minerals on the complex
resistivity was investigated using synthetic shaley
samples following the method established by
Jing et al. (1992). According to this technique,
mixtures of sands with different ranges of grain
sizes and different clay types and contents can be
prepared and consolidated through cycles of
loading/unloading and heating/cooling in a high
pressure and high temperature cell. The main
advantage of the technique is full control of the
sample preparation so that the desired variation
of clay type, content and distribution can be
systematically obtained under laboratory conditions.
Five synthetic samples of different clay contents were prepared , namely SZ1 (clay free),
SZ2 (5% montmorillonite), SZ3 (10% mon-

tmorillonite) and SZ4 (15% montmorillonite).


The sand and clay mixtures were mixed uniformly to achieve homogeneous samples. Table
2 lists the petrophysical characteristics of the
synthetic samples. After loading the samples in
the high pressure cell, they were saturated with
5% by weight of NaCI brine and the consolidation process was started. Repeated loading and
unloading cycles were performed with confining
pressures varying from 500 psi to 4000 psi until
sample consolidation.

Results and discussion

Frequency effect
Figures 3 and 4 show the real component and
phase angle versus frequency for two reservoir
core samples. This plot of resistivity and phase

THE COMPLEX RESISTIVITY OF SANDSTONE SAMPLES

151

Fig. 4. General frequency dependence behaviour for sample Z3.

Fig. 5. Argand diagram with the critical frequency (fc)


separating electrode polarization and bulk sample
response for sample Z1.

Fig. 6. Argand diagram with the critical frequency (fc)


separating electrode polarization and bulk sample
response for sample Z1.
angle against frequency can be divided into
polarization and sample response regions. The
electrode polarization region (e.g. < 10 KHz) is
strongly influenced by polarization at the rockelectrode interface and can be identified from the

bulk rock response by plotting the real and


imaginary components of the impedance on the
complex plane as shown in Figs 5 and 6 (i.e,
Argand diagram, Debye 1929). The sample
response region (i.e. the Cole-Cole region, Cole
& Cole 1941) can be divided into two straightline
regions of distinctive frequency dependence: the
intermediate frequency range (10-100 kHz)
characterized by a small and gradual change in
impedance and phase angle followed by the high
frequency range (100-2 MHz) characterized by a
sharp change in impedance and phase angle. The
transition between the intermediate and highfrequency region is characterized by the relaxation frequency of the interface polarization
process.
Figures 5 and 6 plot the Argand diagrams for
samples Z1 and Z3 showing the separation of
sample response from electrode effects. The
experimental data can be fitted by the classic
Cole & Cole (1941) model of a depressed
semicircle on the Argand plot. According to
Lockner & Byerlee (1985), existing theoretical
models are most useful in the analysis of data
near the peak loss frequency but they may not be
capable of fitting experimental data over the
entire frequency range.

Salinity dependence
The general frequency behaviour of the complex
impedance is shown in Fig. 7 for the reservoir
sample Z7 at two different brine concentrations.
The effect of increasing the pore electrolyte
salinity on the frequency behaviour of the
sample can be summarized as follows:

152

P.S. DENICOL & X. D. JING

Fig. 7. General frequency behaviour for sample Z7 at two brine concentrations.

Fig. 8. Normalized impedance at two brine concentrations showing salinity dependence for sample Z7.

(1) The complex impedance decreases as the


brine salinity increases in the whole frequency range;
(2) The complex impedance decreases with
frequency for both brine concentrations;
(3) The rate of decrease is more pronounced
for the lower brine concentration, that is,
the lower the salinity of the brine the higher
the frequency dependence. This behaviour
is best illustrated when the normalized
impedance is plotted against frequency in
the range from 10 to 100 kHz (Fig.8);
(4) As the salinity of the brine increases, the
relaxation frequency increases.
The frequency dependence as a function of
salinity variations is related to the electrical
double layer, the thickness of which varies with
the brine concentration. High solution concentrations are associated with the compression of
the double layer whilst low concentrations

favour the expansion of the double layer. The


frequency dependence, as expressed by the slope
taken from the semi-log plot of the normalized
impedance in the frequency range from 10 to 100
kHz, is found to increase from 5% to 2 % NaC1.
Similar results were reported by Kulenkampff et
al (1993) and Kulenkampff & Schopper (1988).
This salinity dependence is also related to the
relative mobility of the ions in the pore space
from the free water to the double layer near the
solid surface. In the free water region, the charge
carriers are free to move and therefore follow the
alternating electrical field. On the other hand, in
the double layer region, the movement of ions is
partially restricted by the electrostatic potential.
The result is a delayed oscillation of the diffuse
layer when compared to the free ions of the bulk
solution that react promptly to the alternating
electrical field. Consequently, a phase lag is
established between the input voltage and the
corresponding current flowing through the pore

THE COMPLEX RESISTIVITY OF SANDSTONE SAMPLES

Fig. 9. Frequency dependence of resistivity and phase angle at partial saturation for sample Zl.

Fig. 10. Frequency dependence of resistivity and phase angle at full brine saturation for sample ZI.

Fig. 11. Saturation dependence of the resistivity for sample Z1 as characterized by the chargeability (m)

153

154

P.S. DENICOL & X. D. JING

space. If the concentration of NaCl decreases,


the double layer thickness increases and the
phase lag is more accentuated. Additionally, an
increase of the diffuse layer thickness favours the
blockage of ions, especially at narrowing pores,
with consequent accumulation of charges and
local concentration gradients.

080
E_075
~"070
~ 065
="0.60

Z3
Zl

The saturation dependence was studied by


comparing the frequency spectrum of the real
part of the resistivity at full and partial water
saturation. The partial saturation was arrived at
by displacing brine with Isopar H which has a
dielectric constant of 2.02 at 25 ~ So far, only
water-wet samples have been tested. The frequency dependence of the in-phase resistivity
and phase angle are shown in Figs 9 and 10 for
sample Z1 at 33% and full brine saturation,
respectively. The saturation dependence becomes clearer when both resistivity curves are
displayed in a log-log plot (Fig. 11). The fully
saturated curve is almost flat for the whole
frequency range. On the other hand, the
partially saturated curve is flat in the low
frequency range and shows clear frequency
dependency above the relaxation frequency.
The frequency effect can be better analysed by
the empirical parameter chargeability (m) defined as follows (Siegel 1959):
(6)

where R1 and R2 stand for the low and high


resistivity asymptotes, respectively. Table 1
summarizes the results obtained for the chargeability of the samples. For a given sample, there
is a consistent increase in m when brine is
displaced by oil. The correlation between m and
the water saturation is shown in Fig. 12 for all
the samples. Although a general trend of higher
m for lower saturation can be observed, the
correlation is weak. Samples Z3 and Z9 showed
a more remarkable deviation from the trend,
possibly due to the high iron oxide content in the
former and dispersed glauconite in the latter.
Knight & Nur (1987) also observed that a
sample with high iron oxide content (Indiana
Dark sandstone) had an anomalous dielectric
exponent apparently due to the effect of the
magnetic susceptibility on the dielectric response.
The interpretation of the saturation dependence upon frequency is difficult due to the
intricate geometry of the pore space and its effect
on distribution of fluids within the rock. Mineralogical complexity, mainly related to clays and

Z4

0.55

0.50
0.3

Saturation dependence

m = R1/(R1 +R2)

0.85

0.4

0.5
0.6
Sw (fraction)

0.7

0,8

0,9

Fig. 12. Correlation between chargeability (m) and


water saturation.

metallics, also plays an important role in


increasing the complexity of the frequency
dispersion. However, the general behaviour of
the saturation dependence is characterized by an
increase of the frequency effect in response to the
oil saturation, as distinguished by the phase
angle and chargeability results.
It is important to note that in two-phase
systems, the frequency dispersion due to the
polarization at the solid-liquid interface (poregrain) may be added to by polarization at the
liquid-liquid interface (oil-water). As the water
saturation decreases, there is an increase in the
water-oil interfacial area and an increase in the
complexity of the brine phase topology. For any
rock-fluid systems, wettability plays a significant
role in controlling fluid distribution at the pore
scale. Therefore, it might be possible to derive
wettability information based on the frequency
dispersion measurements of reservoir rock-fluid
systems. However, further research is needed in
this area.

Clay effects
Synthetic shaley samples with controlled clay
type, content and distribution were used to
investigate the effects of clay minerals on
complex impedance measurements. Figure 13
shows the results for the synthetic sample SZ4.
The low-frequency region from 10 Hz to ~10
kHz indicates strong dispersion in both impedance and phase angle which is attributed to
electrode polarization. In the intermediate frequency range (1 to ~100 kHz) the impedance
decreases monotonically while the phase angle
reaches a minimum and then starts increasing.
The high frequency region is characterized by
the relaxation frequency at N800 kHz where the
phase angle reaches a maximum and the
impedance decreases more drastically. All the
synthetic samples present the relaxation frequency at around the same position. However,
the value of the phase angle at the relaxation

THE COMPLEX RESISTIVITY OF SANDSTONE SAMPLES

155

Fig. 13. General frequency dependence behaviour of impedance and phase angle for sample SZ4.

Fig. 14. Normalized impedance versus frequency relationships of four synthetic samples containing various
amounts of montmorillonite.

Fig. 15. Correlation between clay content and frequency dependence for the synthetic samples.

156

P.S. DENICOL & X. D. JING

frequency increases with the amount of clay,


varying from 1 degree for the clay-free sample to
5 degrees for the 15% montmorillonite sample.
This observation suggests that a polarizationlike process is being caused by the clay presence
although the classical induced polarization effect
would be expected at lower frequencies. A
possible explanation for this frequency effect
may be related to the electro-osmotic coupling
due to the accumulation of charges at narrowing
pores (Marshal & Madden 1959; Dankhazi
1993). Although its effect is found to be very
weak, this type of polarization is expected to
increase with a reduction of the sample permeability. Indeed, the synthetic samples show a
decrease in permeability with increasing amount
of clay (Table 2) that leads to the narrowing and
reduction of effective pores and hence the
electro-osmotic coupling.
The slope of the impedance curve in the range
from 10 to 100 kHz is also found to correlate
with the clay content of the samples. Figure 14
shows the normalized impedance versus frequency for the samples containing montmorillonite. The graph indicates a consistent increase
in the frequency slope from sample Z1 (clayfree) to sample Z4 (15% montmorillonite). A
plot of the rate of impedance decrease with
frequency versus clay content is shown in Fig.
15, where the clay effect appears to decrease at
higher clay contents.

Conclusions
The frequency effect in the intermediate frequency range (10-100 kHz) increases when the
solution concentration is decreased from 5% to
2% NaC1. This salinity dependence may be
explained by variations of the double layer
thickness and ion mobility. At high salinity,
the double layer is compressed to the pore
surface and gradually expands with decreasing
brine concentration. As a consequence, the
mobility of the ions in the diffuse layer is
reduced at high salinity preventing them from
following the alternating field as opposed to the
free ions in the centre of the pore. Additionally,
the expansion of the double layer supports the
blockage of ions particularly at the smaller pores
with subsequent electro-osmotic polarization
due to the accumulation of charges.
The frequency effect is found to increase for
the whole frequency range when brine is
displaced by oil (Isopar H). A variation in water
saturation from full to partial saturation resulted in a dramatic increase in the frequency
dispersion and a clear shift of the relaxation
frequency. This observation may have potential

applications for the evaluation of low resistivity


and low contrast pay formations.
The frequency dispersion consistently increases with the amount of clay in the sample.
This effect is better illustrated when the normalized impedance is plotted in the frequency range
from 10 to 100 kHz . The impedance slope is
relatively flat for the clay-free sample (SZ1) and
increases with the content of montmorillonite
for the shaley samples. A plot of the clay content
versus the frequency dependency clearly shows a
relationship.
We would like to thank Petrobras S.A. for sponsoring
P.S. Denicol and for providing reservoir rock samples.
We also wish to thank M. S. King for many valuable
discussions.

References
BORNER, F. D. 1995. Estimation of hydraulic conductivity from complex electrical measurement.
International Symposium of the Society of Core
Analysts, paper 9523.
COLE, K. S. & COLE, R. H. 1941. Dispersion and
absorption in dielectrics. Journal of Chemistry and
Physics, 9, 341.
DANKHAZI, G. 1993. A new principle approach to
induced polarization in porous rock. The Log
Analyst, 34, 54-66.
DEBVE, P. 1929. Polar molecules. Chemical Catalogue
Co.
DENICOL, P. S. & JING, X. D. 1996. Estimating
permeability of reservoir rocks from complex
resistivity data. Society of Professional Well Log
Analysts, 37th Annual Logging Symposium,
paper X.
ELASHAHAB,B. M., JING, X. D. & ARCHER,J. S. 1995.
Resistivity index and capillary pressure hysteresis
for rock samples of different wettability characteristics. SPE paper No. 29888, the 9th Middle
East Oil Show and Conference, March, Bahrain.
Gouv, G. as discussed in HUNTER, R. J. 1988. Zeta
Potential in Colloid Science, Academic Press.
JING, X. D., ARCHER, J. S. 8r DALTABAN,T. S. 1992.
Laboratory study of the electrical and hydraulic
properties of rocks under simulated reservoir
conditions. Marine and Petroleum Geology, 9,
115-127.
KNIGHT, R. & NUR, A. 1987. Geometrical effects in the
dielectrical response of partially saturated sandstones. The Log Analyst, 28, 513-519.
KNIGHT, R. & ENDRES,A. 1991. Surface conduction at
the hydrocarbon/water interface. Society of Professional Well Log Analysts, 32nd Annual Logging Symposium, paper I.
KULENKAMPFF,J. M. & SCHOPPER,J. R. 1988. Low
frequency conductivity--a means for separating
volume and interlayer conductivity. Society of
Professional Well Log Analysts, 12th European
Formation Evaluation Symposium, paper P.

THE COMPLEX RESISTIVITY OF SANDSTONE SAMPLES


--,

BORNER, F. D. & SCHOPPER,J. R. 1993. Broad


band complex conductivity lab measurement
enhancing the evaluation of reservoir properties.
Society of Professional Well Log Analysts, 15th
European Formation Evaluation Symposium,
paper A.

LIMA, 0. A. L. t~ SHARMA, M. M. 1992. A grain

conductivity approach to shaly sandstone. Geo-

physics, 55, 1-10.


LOCKNER, D. A. & BYERLEE, J. D. 1985. Complex

logically
mature.

mature,

texturally

157
submature

to

Sample Z3 (Block 18-2)."


Lower Permian "Penrith Red Sandstone", predominantly quartz grains cemented by quartz
over-growths with iron oxide petina, subrounded-rounded, mineralogically and texturally sub-mature.

resistivity measurements of confined rock. Journal

of Geophysical Research, 90, 7837-7847.


MARSHALL, D. J. t~ MADDEN, T. R. 1959. Induced

polarization, a study of its causes. Geophysics, 24,


790-816.
RINK, U. SCHOPPER,J. R. 1974. Interface conductivity and its implications to electrical logging.

Society of Professional Well Log Analysts, 15th


Annual Logging Symposium, paper J.
SEN, P. N. 1980. The dielectric constant and conductivity response of sedimentary rocks. Society of
Petroleum Engineers, paper 9379.
SEN, P. N. 1981. Relation of certain geometrical
features to the dielectric anomaly of rocks.
Geophysics, 46, 1714.
SEIGEL,H. 0. 1959. A theory for induced polarization
effects (for step excitation function). In: WArr, J.
R. (ed.) Over Voltage Research and Geophysical
Applications. Pergamon Press Inc., 4-21.
TAHERIAN, M. R., KENYON, W. E. & SAFINYA, K. A.
1990. Measuremen of dielectric response of watersaturated rocks. Geophysics, 55, 1530-1541.

Appendix: geological description of sandstone rocks


(a) Outcrop rocks
Sample Z1 (Block 15-8):
Lower Carboniferous sandstone, average grain
size 0.2 ram, 95% q u a r t z , alkali feldspar, clay,
biotite, alcite cement (5%) with some chert,
angular to sub-angular, poor sphericity, minera-

Sample Z4 (Block 16-2):


Upper Carboniferous sandstone, grain size : 0.10.3 mm, 85% quartz, 0% alkali feldspar, 5%
mica, very irregular, poor sphericity, texturally
immature and mineralogically submature.

Sample Z5 (Block 19-4):


Lower Triassic 'Bunter' sandstone, fine to
medium grain sizes (< 0.5 mm), 95% quartz,
% alkali feldspar and calcite, sub-rounded, poor
sphericity, texturally and mineralogically mature.

(b) Reservoir rocks


Sample Z7:
Glauconitic sandstone, semi-friable, grains are
sub-rounded with regular to good selection.
Mineralogy also includes quartz, feldspar and
mica.

Sample Z8."
Sandstone with pseudo-argilaceous matrix (27
%), quartz (31%), K-feldspar (18%), glauconite
(6%), plagioclase (5%), others (2%). Cements
include dolomite and pyrite.

Sample Z9."
Sandstone with pseudo-argilaceous matrix,
quartz, K-feldspar, glauconite, and plagioclase.

Acoustic wave anisotropy in sandstones with systems of aligned cracks


A. S H A K E E L 1 & M . S. K I N G 2

1Production Department, Oil and Gas Development Corporation, F-8 Markaz Islamabad,
Pakistan
2 Department of Earth Resources Engineering, Royal School of Mines, Imperial College,
London SW7 2BP, UK

Abstract: Seismic anisotropy has been studied on a number of dry cubic sandstone

specimens, of 51 mm side, in which a system of aligned cracks has been first introduced
progressively by the application of a polyaxial state of stress, and then closed by hydrostatic
stress. One P- and two S-wave velocities polarized at right angles, along with the
deformation, have been measured at each stress level in each of the three principal stress
directions. Thomsen's (1986) anisotropy parameters (e, 7, 6) have been calculated at each
stress level during the cracking and crack closing cycles using Nishizawa's (1982) theory.
Test results indicate that anisotropy in the P-wave velocity is greater and more sensitive to
the presence of aligned cracks than that observed for S waves. Modelling studies show that
the P-wave anisotropy parameter e is always greater than that of anisotropy parameter 8, for
low crack densities and for small aspect ratios. The reverse is true for high crack densities
and low aspect ratios. The results of numerical studies indicate that S-wave anisotropy is
independent of the nature of the saturating fluid and that it is possible to observe elliptical
anisotropy in a medium containing aligned dry ellipsoidal inclusions.
It is well known that the presence of microcracks
and fractures reduces the acoustic velocities of
P- and S-waves in rocks. When the principal
stresses are altered on a rock that initially has a
random distribution of cracks, the crack distribution no longer remains randomly oriented.
The effect of an applied non-hydrostatic stress is
to close cracks in some directions and leave
cracks open in others (Sayers 1988). Those
cracks with their normals lying close to parallel
to the new major principal stress will tend to be
closed more than those with their normals subparallel to the new minor principal stress (Sayers
1988). The elastic and transport properties of the
rock then become anisotropic in their behaviour,
with the degree of anisotropy depending on the
magnitude of the principal stress differences, the
type of fluid filling the cracks (Xu & King 1989,
1992; King et al. 1995a,b).
Seismic anisotropy was studied more than 40
years ago by Postma (1955) and Uhrig & Melle
(1955), but for a long time its effect was ignored
or considered insignificant, due to the fact that
most of the seismic surveys carried out were for
P-wave reflection and conducted at small angles
to the vertical. However, for seismic surveys
conducted with large angles of the incidence
waves (such as VSP surveys), the effect of
a n i s o t r o p y c a n n o t be i g n o r e d ( C r a m p i n
1985a,b). Seismic anisotropy due to aligned

cracks has been extensively studied by, amongst


others, Crampin (1984, 1985a,b) and Crampin &
Atkinson (1985), who are of the opinion that Swave velocities are more sensitive to the presence
of aligned cracks and that they provide a better
quality of information on anisotropy effects than
does the P wave. Crack orientation, when cracks
are aligned vertically, can easily be determined
by the splitting of vertically propagating polarized shear waves. This splitting occurs as a result
of azimuthal anisotropy induced by the microcracks and fractures. A knowledge of seismic
anisotropy can provide useful information about
the mineralogy, the orientation of cracks and
pores, the degree of cracking and crack geometry, orientation of the in situ stress field, and the
possible proportion of gas and liquid within the
inclusions in hydrocarbon reservoirs (Crampin
1985a).
Thomsen (1986) has derived a set of three
dimensionless anisotropy parameters (e, 7 and 8)
to describe weak to moderate transverse isotropy of a medium. These parameters are
defined in terms of the five components of the
stiffness tensor (Cll , C33 , C13 , C44, C66) relating
stress and strain for the transversely isotropic
medium as follows:

Cll -C33
V 2 1 - V22
-- - -2C33
V 22

SHAKEEL,A. & KING, M. S. 1998. Acoustic wave anisotropy in sandstones with systems of aligned
cracks In. HARVEY,P. K. & LOVELL,M. A. (eds) Core-Log Integration, Geological Society, London,
Special Publications, 136, 173-183

(1)

173

174

A. SHAKEEL & M. S. KING


C66 - C44

V 21 - V 22

2C44

V 22

(2)

(~ ~__ (C13 + C44) 2 - (C33 - C 4 4 ) 2


2C33 (C33 - C44)

(3)

The parameters are all zero for an isotropic


medium and their deviation from zero represents
the degree of anisotropy. The value of ~, which is
always positive, represents the relative difference
between the P-wave velocities propagating perpendicular (Vp1) and parallel (Vp2) to the axis of
symmetry. The general term 'anisotropy' of a
rock usually refers to the quantity e, calculated
using the following equation for small values
o f e,
Vp 1 B Vp 2
E -- - -

re2

(4)

The parameter 3' describes the S-wave anisotropy of a transversely isotropic medium. It is
the relative difference between the faster S-wave
velocity (Vsl) and the slower S-wave (Vs2)
velocity travelling in a transversely isotropic
medium. Thus, for small values of 7, it can be
used to define 'S-wave anisotropy' of a medium
(Thomsen 1986) as
VS1 m Vs 2

7 -- -

Vs2

(5)

where Vsl and Vs2 are S-wave velocities


propagating parallel to the plane of cracks with
their polarization parallel and perpendicular to
the plane of cracks, respectively. The parameter
dominates the anisotropic response when the
acoustic wave propagates in a plane which is
parallel or approximately parallel to the axis of
symmetry. It is independent of the seismic
velocities of the medium perpendicular to the
axis of symmetry and can take either positive or
negative values.
As shown by Thomsen (1986), the parameters
~, 3" and g are less than 0.2 in magnitude for
weak-to-moderate anisotropy. Furthermore,
Thomsen (1986) states that elliptical anisotropy
will be observed if 6 = ~. Since the parameters ~,
3' and ~ are easily interpretable and can be
calculated from the five elastic constants obtained from Nishizawa's (1982) theory, they are
used here to model and study the variation in
anisotropy as a function of aspect ratio, crack
density and stress.

Experimental system
A polyaxial stress loading system, developed at

Imperial College of Science and Technology


London, has been used for testing 51mm-side
cubic rock specimens. The system, described in a
preliminary technical note by King et al. (1995a)
and in detail by Shakeel (1995), consists of a
loading frame in the form of an aluminium alloy
ring within which two pairs of hydraulic rams
and ultrasonic transducer holders are mounted
to provide orthogonal stresses on the cubic rock
specimen in the horizontal plane. Each of the
three principal stresses may be varied independently in the range 0 to 115 MPa in the
horizontal principal directions and to over 750
MPa in the vertical major principal direction.
The horizontal principal stresses may be servocontrolled using facilities associated with a
Schenk compression testing machine. The vertical major principal stress is provided through
ultrasonic transducer holders mounted in a
Schenk 160-tonne closed-loop servo-controlled
compression testing machine. Stress is transmitted to each of the six faces of the cubic rock
specimen through 5 ram-thick magnesium faceplates matching approximately the elastic properties of the rocks being tested. Deformation of
the rock specimen is measured by pairs of
extensometers (LVDTs) mounted in each of
the three principal directions. An isometric
view of the polyaxial loading frame is shown in
Fig. 1.
Each of the three pairs of transducer holders
contains stacks of piezoelectric transducers
capable of producing or detecting pulses of
compressional (P) or either of two shear (S)
waves polarized at right angles propagating in
one of the principal stress directions. The
transducer holders have a bandwidth in the
range approximately 450 to 800 kHz for P-wave
and 350 to 750 kHz for S-wave pulses.
Loading in the 1-direction is characterized by
the major principal compressive stress (cq)
direction and that of 2- and 3- as the intermediate (or2) and minor (a3) principal stress
directions, respectively. The wave type nomenclature employs two suffixes 'i' and 'j' (as with
Vij) where T refers to the propagation direction
of the wave and 'j' to the polarization (particle
motion) direction. Thus V33 is the P-wave
velocity propagating in the minor principal
stress direction and V13 is the S-wave velocity
propagating in the major principal stress direction with polarization in the 1-3 plane. A total
of nine components of velocity are measured:
three compressional VPll , VP22 and VP33 and
six shear VS12, VS13, VS21, VS23, VS31 and VS32
Both the P- and S-wave velocities are measured
with an accuracy of +1% and a precision of
+0.5%.

ANISOTROPY IN CRACKED ROCKS

175

LOAD IN 1-DIRECTION APPLIED IN


I~I~pRSCHENK

160-TONNE S E R V O - C O N T R O k L E D

ESSION TESTING MACHINE

4.

5.

6.

7.
8.
9.

TRANSDUCER HOLDERS
HYDRAULIC PRESSURE,
2-DIRECTION
HYDRAULIC RAM,
2-DIRECTION
HYDRAULIC PRESSURE,
3-DIRECTION
HYDRAULIC RAM,
3-DIRECTION
"fRANSDL~ER HOLDERS
CUBIC ROCK SPECIMEN
REACTION RING

Fig. 1. Isometric view of the polyaxial loading system.

Results and discussion


First a numerical example is provided to enable
a better understanding of the effects of the
different parameters, such as crack aspect ratio,
crack density and type of saturating fluid on the
anisotropy parameters and on the acoustic
velocities of such a cracked solid permeated
with aligned ellipsoidal inclusions. Finally, the
theory is used to study the anisotropy as a
function of stress for a solid progressively
permeated with a system of aligned cracks.

Numerical results and discussion


In this numerical example, P- and S-wave
velocities and Thomsen's (1986) anisotropy
parameters are calculated as a function of aspect
ratio for a solid permeated with aligned ellipsoidal inclusions. Nishizawa's (1982) theory is used
to calculate the elastic constants. The aspect
ratio of the inclusions is varied from a =0.0001
(almost flat cracks) to a = 1 (spheres). Four
crack densities are studied, ~=0.01, 0.05, 0.10
and 0.20. Both types of inclusions are investigated: dry inclusions with a fluid bulk modulus

of 1.5 -4 GPa, and liquid-filled inclusions


with a fluid bulk modulus of 1.5 GPa. The
isotropic background material is the same as
that used by Nishizawa (1982): matrix density
2.7 g cm -3 and Lame's constants A = # = 39 GPa.
Figures 2 and 3 show results of the Thomsen's
anisotropy parameters as a function of aspect
ratio for dry and liquid-filled inclusions, respectively, for four different crack densities ~= 0.01,
0.05, 0.10 and 0.20. It is clear from these figures
that the values of anisotropy parameters increase as the crack density is increased from ~ =
0.01 to 0.20. They all become zero for an aspect
ratio a = 1, corresponding to the isotropic
situation.
Note that for dry inclusions (Fig. 2) all the
anisotropy parameters have a non-zero constant
value for a large range of small aspect ratios and
that they only tend to zero for large aspect ratios
approaching a = 1. Hence, for a large group of
small aspect ratios the resultant anisotropy is
hardly affected by a change in aspect ratio for
the case when solid is permeated by dry
inclusions. The non-zero constant values of the
parameters ~ and 7, for a large group of small
aspect ratios for dry inclusions, indicate that

176

A. SHAKEEL & M. S. KING


0.04

. . . . . . . .

:,= o.o2
o7

. . . . . . . .

I' . . . . . .

'

. . . . . . .

C r a c k density = 0 . 0 1

/ Y

87"

""

o.01
9

O.OO I

'

I llll

. . . . . . .

o.ool

O.OeO

. . . . . . . .

O.Ol

0.1

(a)
0~0

. . . . . . . .

. . . . . . . .

Crack density = 0.05

0.15

. . . . . . . .

. . . . . . . .

~-. r

"2.'.'_"2~: "-'."L\~ 22:" 2"-".'2"2:: - " 2 " . ' - " . 2 : ~ 22".222:: -"2.-'_"".'.:: -':;.':.z-~.

>~ o.lo

-~-

o.o$
0.00

~,.,..

"

'

"

. . . . . . . .

I''|

0.04}01

0.001

'

'

'

'

'

"..,~

''']

0.01

0.1

(b)
0.5

0.4

. . . . . . . .

. . . . . . . .

Crack density = O. I0
9
"~ "..--.'.'T

." . - : -" . " . ~. - - - : . ' T . . " . ' : "

-".:. 7----.-.-:

: .--.'..--:.--.

- .-""

-- "~.'.'T

~';r

o.3

O~

. . . . . . . .

. . . . . . . .

~)

~"

2 2": : .".-:"

- -~....:

: :";

"'~"

". ~

'~.~

0.1
O.O

, ..!

0.0001

, ,r

. . . . . . . .

0.01

O.OOl

. . . . . . . .

0.1

(c)
. . . . . . . .

Crack density
l.o

. . . . . . . .

. . . . . . . .

. . . . . . .

0.20

.". . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

p..

/
0.0
O.OeOI

Y
.

"'-..
i

i,,|

. . . . . . . .

O.OI

O.OO!

Aspect

. . . .

..l

" "

O.t

Ratio

(d)
Fig. 2. Thomsen anisotropy parameters as a function of aspect ratio for dry inclusions for crack densities, (a) ~ =
0.01, (b) ~= 0.05, (c) ( = 0.10, (d) ~= 0.20.

there is also a constant difference between the


two P- and two S-wave velocities propagating
both parallel and perpendicular to the plane of
cracks for the same range of aspect ratios. This
conclusion corresponds to Fig. 4 which shows a
very small variation in P- and S-wave velocities
propagating both parallel and perpendicular to
the plane of cracks as the aspect ratio is changed

for dry inclusions of crack density of 0.01. In


this figure (also in Fig. 5) the slower acoustic
velocities (VP2 and VS2) are represented by
dotted lines and the faster (VPi and VS1) by
solid lines. A study of Fig. 2 indicates that the
anisotropy parameters follow a certain pattern
for dry inclusions, i.e. for low crack densities
< 0.1, e > 6 > 7, (Figs 2 a-b) and for high crack

ANISOTROPY IN CRACKED ROCKS


0.03

0.02
+r

'' ......

......

Crack~ density = 0.01

''I

177

........

.......

p t ' Y..... . . - ' " " ..- .-.. . -. . . -. ( _ : . ~ " :~.-'=" = : - ' : - : k : .

,..

O.Ol
..... -+"~

o-,"
-0.01

.....................

.~-"*"

........

-0.02
0.0001

......

ill

O.OOl

i. ....

O.Ol

....

0.I

(a)
O.IS

.........

0.1
0.05

,,J

. . . . . . . .

. . . . . . .

...........

C r a c k density = 0.05

","U:':u:-':~.:. ,,

...... . .... _.-j...... ...o..-o'~


~

s
..................................

~- j-

. o~"
-0.05

................
,

-0.1
0.0001

.---"
,

....

, ,.,1

0.001

i '''I

O.OI

0.1

(b)
0"~

,r

........

0.2

C r a c k density --- 0.10

0,1

-0.1

........

.......

..)~ ~~ ~~ -'~" = : z.-;~..

........ . .o

.......... .....~

........................

.*" f . /

~ o ~ o "" J~l~

. . . . . . . . . . . . . . . . . . . . . . . .

-0.2

........

. . . . . . . .
0.000

"~'~"

. . . . . . . .

0.001

. . . . . . . .

0.01

. . . . . . . .

O. !

(c)
9 " ...... I

0,6

0.45

........

C r a c k denmty = 0.20

........

i ..... .:~. . . . . . . .

..-"

0.3
0.15

~
~
o

....o~

I"

"%

-/"

.,.o
.s

0
-0.15
-0.3
0.O~i

1..

. . . . .

. . . . .

0.001

+,il

0.01

. . . . . .

....

O.l

Aspect Ratio

(d)
Fig. 3. Thomsen anistropy parameters as a function of aspect ratio for liquid filled inclusions for crack densities,
(a) ~= 0.01, (b) ~= 0.05, (c) ~= 0.10, (d) ~=0.20.

densities ~>_0.1, 6 > ~ > 7 (Figs 2 c-d).


However, for a solid permeated with liquidfilled inclusions, a large variation in anisotropy
parameters ~ and 6 is observed as the aspect
ratio is changed (Fig. 3). The changes in e are
related to P-wave velocities, especially VP2
which is strongly influenced by the liquid-filled
inclusions as the aspect ratio is changed. This

result corresponds to Fig. 5a which shows a


significant variation in V P 2 a s the aspect ratio is
changed for liquid-filled inclusions of crack
density ~=0.01. It can be seen from Fig. 3 that
the value of ~ tends to zero for very small aspect
ratios for all the crack densities, indicating that
the difference between the P-wave velocities in
both the directions parallel and perpendicular to

178

A. SHAKEEL & M. S. KING


6.60

. . . . . . . .

. . . . . . . .

. . . . . . . .

6.55

6.$0

6.45
6.40

.........................................................................

6.35
6.30

,,.

Vr2

Crack density = 0.01

6.~g

. . . . . . . .

0.0001

, , , , , I

. . . . .

0.01

0.001

O. I

Aspect Ratio

(a)

3.85

Vs~

3.80

3.75

3.70

3.65

3.60

Crack density -- 0.01

, ,

0.0001

. . . . . . .

. . . . . . . .

0.001

0.01

. . . . .

0.1

Aspect Ratio

(b)

Fig. 4. Changes in acoustic wave velocities as a function of aspect ratio for dry inclusions (a) P-wave velocities and
(b) S-wave velocities. The crack density is ( = 0.01.

the plane of cracks also tends to zero. This effect


is clearly shown in Fig. 5a. The changes in ~5are
related to those P- and S-wave velocities which
are either propagation or polarization perpendicular to the plane of cracks (3-direction).
A comparison of Figs 2 and 3 with each
respective crack density, indicates that there is
hardly any difference in the value of anisotropic
parameter 7 for a large range of aspect ratios
when either dry or liquid-filled inclusions are
used. This behaviour corresponds to the fact
that S-wave velocities are not affected much as
the condition of inclusions is changed from dry
to liquid-saturated. Since the parameter 7 is
constant for a large range of aspect ratios for

both dry and liquid-filled inclusions, the acoustic


velocities VS1 and VS2 are not strongly affected
by aspect ratios (Figs 4b and 5b for a crack
density ( = 0.01).
For all the cases studied for dry and liquidfilled inclusions, the values of e, 7 and 6 are
always positive, indicating that the P- and Swave velocities (VP 1 and VS1) propagating
perpendicular to the axis of symmetry are always
greater than those propagating along (VP2 and
VS2) the symmetry axis (Figs 4 and 5). Finally,
Figs 2 and 3 show that the parameters e and 6
are equal for aspect ratios lying between a = 0.4
and 1 for both dry and liquid-filled inclusions.
This suggests that the resultant anisotropy is

ANISOTROPY IN CRACKED ROCKS


6.60

........

........

. . . . . . . .

179
i

Vp l

6.55

6.50

6.45

6.40

6.35
Vp2
6.30

Crack

density

........

6.25

= 0.01

........

0.0001

0.001

........

0.01

.......

O.t

Aspect Ratio

(a)
3.85

I . . . .

. . . . . . .

Vs i

3.80

.'a

3.75

3.70

3.65

3.60
0.01.|

Crack density = 0.01


.

. . . . . .

0.001

0.01
Aspect

. . . . .

0.1
Ratio

(b)

Fig. 5. Changes in acoustic velocities as a function of aspect ratio for liquid-filled inclusions (a) P-wave velocities
and (b) S-wave velocities. The crack density is ~= O.O1.
elliptical (Thomsen 1986) for both the dry and
saturated cases in this limited range of aspect
ratios.
Experimental results and discussion

Tests have been performed on five dry sandstone


specimens, in which a system of aligned cracks
has been first introduced by increasing the major
(el) and intermediate (~2) principal stresses in
unison to near failure, while keeping the minor
(or3) principal stress constant at some low level.
The aligned cracks are then closed by the
application of a hydrostatic compressive stress.
The nine components of velocity are measured
throughout three separate stress cycles. The first

involves measurements on the flesh, uncracked


rock specimen during the application of an
increasing hydrostatic stress. The second cycle
involves measurements while a system of aligned
cracks, with their normals parallel to the minor
stress direction, is formed in the rock specimen
(cracking cycle). The third cycle involves measurements during the application of a further
increasing hydrostatic state of stress to close the
cracks formed during the cracking cycle (crack
closing cycle).
Discussed here as being characteristic of the
studies made on five sandstones tested in this
research programme will be that of Penrith
sandstone. This is a fine-to-medium grained
sandstone of lower Permian age having a low

180

A. SHAKEEL & M. S. KING


.

S.O

,~

I -*1%

o3 = 3
4.5

:I-+1%

M P a

....

(constant)

4.0

4.0
3..~

:~-

3.0

ft.

2..~

3.0

Pll

. . . . . .

l.$

P22

2.0

. . . .

= o 2 =

. . . .

. . . .

25

. . . .

SO

;~--

0"3 ( M P a )

. . . .

75

100

01 = O2 ( M P a )

12S

ISO

~:~
o l > o 2 = IOOMPa

3.2
.

'

50
01

1.5

3.2

P33

03 - 3 M P a
2.9

"

(constant)

. r -*1%

2.9 -'r -*1%


~

!~

2.6

2.3

~ 2.,
g

2.0
1.7

$23---o---

$21

S13 ~

$32

S12

t
I

1.4

1.7

$31 ~

!
,

1.4

,,.

50
o 1 = 0 2 = 0 3 (MPa)

Fig. 6. P and S-wave velocities as a function of


hydrostatic stess during the initial stress cycle for
Penrith sandstone sample, (a) P-wave velocities and (b)
S-wave velocities.

clay content (3%), an effective porosity of 13%,


a permeability of ~150 mD, a grain density
2.6 gcm -3 and a bulk density of 2.26 gcm -3 in its
dry state.
Figure 6 shows changes in the three P- and six
S-wave velocities plotted as a function of
hydrostatic stress on the fresh uncracked rock
specimen. Although loading in the 1-, 2-, and 3directions is identical there is a small difference
between the changes in P- and S-wave velocities
which is due to the differences in the initial
elastic properties between these three directions.
It will be observed that the sandstone exhibits
behaviour that is close to being isotropic, with
both sets of P- and S-wave velocities increasing
in magnitude with increasing stress and lying
within + 1% error bar, except at the lowest stress
level.
Figure 7 shows changes in the three P- and six
S-wave velocities during the cracking cycle. The
minor principal stress (o'3) was kept constant at 3
MPa while o']--o'2 were increased in unison in
steps from 3 to 100 MPa. Then, while maintaining the intermediate principal stress at 100 Mpa
(limited by the experimental system), the major

i~

S13
i ....

~
i ....
50

o I = 02 ( M P a )

$32
i ....
75

S12
i.,,.
100

i ....
125

l.gO

=',~
o 1 > o 2 = IOOMPa

F i g . 7. P a n d S - w a v e v e l o c i t i e s a s a f u n c t i o n

of stress

during the cracking cycle for Penrith sandstone


sample, (a) P-wave velocities and (b) S-wave velocities.

principal stress was increased in steps to


132 MPa until the specimen was near failure.
The acoustic velocities propagating in the 3direction show an initial increase with stress due
to the closure of pre-existing cracks with their
normals in the 1- and 2-directions, followed by a
decrease as dilatant cracks with normals parallel
to the 3-direction begin to form and open up. It
is concluded from Fig. 7, with VPll~VP22 and
VS12~,~VS21
all increasing monotonically, that
the majority of the cracks formed are aligned in
the 1-2 plane, perpendicular to the 3-direction.
Shear wave birefringence occurs in all directions
of propagation except along the symmetry axis
(3-direction) for obvious reasons of symmetry.
This effect was also observed in the experiments
of Nur & Simmons (1969).
The cracking cycle velocity data plotted in
Fig. 7 indicate that the magnesium plates match
the sandstone well in elastic properties up to
stresses of o1 = cr2 = 100 MPa, when the majority
of the aligned cracks are formed. As o'] is further
increased (with o'2 constant), the platens cause
confinement and the S-wave velocities propagating in the 1- or 2-direction and polarized in the
3-direction (V13 or V23) become higher than the

ANISOTROPY IN CRACKED ROCKS


5~

9"

I '' "

"

" ' 'I

2.4

4J I

7-O

1.6
3,$

181

7Z:L~ ,--]
.

,o

3.0
G a m m a ,

O.8

2.0

&

25

0.4

SO

50

75
'=

a I *, 0" 2 - 0 3 ( M P a )

75

1o0

O 1 = 02 ( M P a )

-'~

12~

150

o l 9 o2 = 100MPa "4''I

(a)

3.2

,~.,

'-4

Aspect Ratio
2.9

-Z

ct = 0 . 0 0 ~

2.0

I~lta.
A

1.6

2.6

I..*

,,

0J

1.7

.--o---

$31

$23

SD

S32

o.4

$21

.
4 .....

S12

o.o
1.4

"

f
5O

0
7$

0.1

0.2

o.3

0.4

Crack Density

01 - 0 2 - 0 3 (MPa)

Fig. 8. P and S-wave velocities as a function of


hydrostatic stress during the crack closing cycle for
Penrith sandstone sample, (a) P-wave velocities and (b)
S-wave velocities.

Fig. 9. Thomsen's (1986) anisotropy parameters for


Penrith sandstone sample during the cracking cycle as
a function of (a) stress and (b) crack density of the
aligned cracks.

velocities that are propagating in the 3-direction


(V31 or V32). This behaviour suggests that the
aligned crack density towards the extremities of
the specimen in the 3-direction is higher than in the
centre for values of stress greater than 100 MPa.
Figure 8 shows changes in three P- and six Swave velocities plotted as a function of hydrostatic stress during the subsequent crack closing
cycle. As the stress is increased, both sets of Pand S-wave velocities appear to be approaching
asymptotic values that are only slightly lower in
magnitude than those shown in Fig. 6 for the
preliminary uncracked cycle. Upon removal
from the loading frame after completion of the
tests, the specimens all showed signs of throughgoing fractures aligned close to normal to the 3direction.
The nine components of velocity determined
as a function of stress during the cracking and
subsequent crack closing cycle have been used to
evaluate the Thomsen's (1986) anisotropy parameters and crack density for the cracks aligned
perpendicular to the symmetry axis (3-direction).
The procedure, employing Nishizawa's (1982)
theory, first to model the velocity data, is

described in detail by Shakeel (1995), who found


excellent fits (within 4-1% at all stress levels) in
comparing the theoretically modelled and the
laboratory measured velocities during both the
cracking and crack closing cycles.
As the Penrith Sandstone was tested in its dry
state, a value of 1.5x 104 GPa was chosen for
the fluid bulk modulus. A range of aspect ratios
(0.0005 to 0.002) was employed during each of
the stress cycles to obtain the best match
between the modelled and experimental velocities. It was found that a value of aspect ratio
c~--8.0xl0 ~ provides the best match between
the modelled and the experimental velocities
during the cracking cycle and for the crack
closing cycle for most of the stress levels.
Figure 9a shows changes in the anisotropy
parameters e, 7 and 6 as a function of (71 and 0"2
during the cracking cycle, during which o.3 was
kept constant at 3MPa. All the anisotropy
parameters increase as the stress is increased
due to an increase in crack density (Fig. 9b). The
rate of increase in the value of these parameters
is lower as the stresses o-~ =o-2 are increased
initially from 2 to 100 MPa, but it becomes much

182

A. SHAKEEL & M. S. KING


.

Aspect ratio

. . . .

i . . . .

Aspect ratio

a = 0.001~;

i . . . .

. . . .

. . . .

i'.

a = 0.0005

o 3 - 2 MPa (Constant)

Delt~

1.5

Epsilon,

9 ~

1.0

Gamma.
,

~
,

25

50

.
25

I~
c; 1 = 02 = 03

. "~.

50

,~.

75

. . . .

n . . . .

100

1~

13o

75
0 1 = 02 ( M P a )

(MPa)

-' ~

o I > o2= I00MPa

(a)

(a)

. . . .

Aspect Ratio

9 -

J . . . .

L . . . .

L . . . .

Dolts,

a = 0.0015

i . . . .
6 "--'~/~

U)
0.02

0.04

0.06
Crack

0,(]~

0.I

0.12

0,05

O.l
Crack

Density

0. I$

0.2

0.25

Density

(b)

(b)

Fig. 10. Thomsen's (1986) anistropy parameters for


Penrith sandstones sample during the crack closing
cycle as a function of (a) stress and (b) crack density of
the aligned cracks.

Fig. 11. Thomsen's (1986) anistropy parameters for


Crosland Hill sandstone sample during the cracking
cycle as a function of (a) stress and (b) crack density of
the aligned cracks.

higher at higher stresses. The higher rate of


increase of anisotropy parameters for stresses
o'1 >a2 = 100MPa is due to the nucleation and
coalescence of the majority of the aligned cracks,
which is also clear from the sharp decrease in
V33 and S-wave velocities propagating or polarized perpendicular to the plane of cracks (Fig.
7).
Figure 10a shows changes in the anisotropy
parameters e, 7 and 6 as a function of hydrostatic stress during the crack closing cycle. All
the anisotropy parameters decrease as the stress
is increased due to a decrease in crack density
(Fig. 10b). Cracks close very quickly during the
initial loading, resulting in a consequent rapid
decrease in the value of the anisotropy parameters. When the stress is increased further, a
major fraction of the crack surface area comes
into close contact which slows down the closure
of cracks and the anisotropy parameters decrease much more slowly than before. Results in
Fig. 10 show that the anisotropy becomes weakto-moderate and elliptical (e = 6) for hydrostatic
stresses >10 MPa.

The anisotropy parameters follow the same


pattern during the cracking and crack closing
cycles, i.e. at each stress level ~ > 7 indicates that
the anisotropy in P-wave velocities is greater and
more sensitive to the crack density than the
anisotropy in S-wave velocities. Furthermore,
for higher crack densities (~> 0.08), ~ is greater
than ~, which is in accordance with the prediction of Nishizawa's theory as shown in Fig. 2. A
study of Figs 9 and 10 also indicates that
elliptical anisotropy is only possible in the weak
anisotropic region (anisotropy p a r a m e t e r s
< 0.2) which occurs only at low crack densities
(~ < 0.06). These results are true for all the other
sandstones tested in this research program. As
an example, Figs 11 and 12 show similar
anisotropy results for the Crosland Hill (low
clay content [< 1%], effective porosity 6% and
permeability < l mD) sandstone specimen during the cracking and crack closing cycles.

Conclusions
(1) The results of the experimental study for

ANISOTROPY IN CRACKED ROCKS


.

Aspect ratio ct 0.0005


=

t.6

171tt,

':
25

50
oI=

183

75

We wish to acknowledge with thanks, the support


provided by Shell Expro, British Gas, BP Exploration
and AGIP for this research project. The senior author
is especially indebted to N. Hyder of the Joint Venture
Department, OGDC, for arrangements, and the Oil
and Gas Development Corporation of Pakistan for
providing the finance necessary to present this paper.
Special thanks are also due to Dr N. A. Chaudhry for
providing data of one of his specimens to conduct
some of the modelling work.

ioo

c 3 (MPa)

o z =

(a)
References

. . . . . . .

. . . .

0 . . . .

Aspect Ratio cc= 0.0005


"4

#~
/

. . . .

Delta, 8

/
:,= 1.6
d

o.0

Ion ,

/7
o.1

0.2
Crack Density

o.3

S. 1984. Anisotropy in exploration seismics.


First Break, 2, 19-21.
- 1985a. Evaluation of anisotropy by shear wave
splitting. Geophysics, 50, 142-152.
1985b. Evidence for aligned cracks in the
Earth's crust. First Break, 3, 12-15.
- ATKINSON, B . K . 1985. Microcracks in the
Earth's crust. First Break, 3, 16-20.
KING, M. S., C H A U D H R Y , N. A. & SHAKEEL, A. 1995a.
Experimental ultrasonic velocities and permeability for sandstones with aligned cracks. International Journal of Rock Mechanics and Mining
Science and Geomechanics Abstracts, 23, 291-302.
- - ,
SHAKEEL, A . & C H A U D H R Y , N . A. 1995b.
Acoustic wave propagation and permeability in
sandstones with systems of aligned cracks. Presented at the Geophysical Society of London,
Borehole Research Group, Conference on Developments in Petrophysics, Sept, 1995.
NISHIZAWA,O. 1982. Seismic velocity anisotropy in a
medium containing oriented cracks--transversely
isotropic case. Journal of the Physics of the Earth,
30, 331-347.
NUR, A & SIMMONS,G. 1969. Stress-induced velocity
anisotropy in rocks: An experimental study.
Journal of Geophysical Research, 74, 6667-6674.
POSTMA,G. W. 1955. Wave propagation in a stratified
medium. Geophysics, 20, 780-806.
SAYERS, C. M. 1988. Stress induced ultrasonic wave
velocity anistropy in fractured rock. Ultrasonics,
26, 311-317.
SHAKEEL, A. 1995. The effect of oriented fractures on
elastic wave velocities, attenuation and fluid permeabilities of sandstones. PhD Thesis, Imperial
College of Science, Technology and Medicine,
University of London.
THOMSEN, L. 1986. Weak elastic anisotropy. Geophysics, 51, 1954-1966.
Xu, S & KING, M. S. 1989. Shear-wave birefringence
and directional permeability in fractured rock.
Scientific Drilling, 1, 27-33.
& - 1992. Modelling the elastic and
hydraulic properties of fractured rocks. Marine
and Petroleum Geology, 9, 155-166
U H R I G , L. F & VAN MELLE, F. A. 1955. Velocity
anisotropy in stratified media. Geophysics, 20,
774--779
CRAMPIN,

0.4

Fig. 12. Thomsen's (1986) anistropy parameters for


Crosland Hill sandstone sample during the crack
closing cycle as a function of (a) stress and (b) crack
density of the aligned cracks.

dry sandstones suggest that the anisotropy


parameter 6 is the most sensitive to the
crack density. Moreover, for the dry rocks,
the anisotropy in P-wave velocities is
greater and more sensitive to the crack
density than the anisotropy in S-wave
velocities.
(2) The results of the numerical study suggest
that the anisotropy in P-wave velocities is
greater when the saturating fluid is very
compressible (gas) and when the cracks are
flat (small aspect ratios), while the anisotropy in S-wave velocities is almost unaffected by the nature of the saturating
fluid.
(3) For dry inclusions over a large range of
aspect ratios less than 0.1 the resultant
anisotropy is hardly affected by a change in
the aspect ratio.
(4) The results of the numerical and experimental studies suggest that elliptical anisotropy will be observed in a m e d i u m
containing aligned ellipsoidal inclusions of
aspect ratios greater than 0.4.

Complementary functions reveal data hidden in your logs


J. R. S A M W O R T H
Wireline Technologies Limited, East Leake, Loughborough, Leicestershire L E 1 2 6JX, U K

Abstract: Many logging tools make multiple measurements of the same type that have more
than one depth of penetration. Common examples are Compensated Density, Compensated
Neutron and Array Induction logs. The purpose of the compensation is to reduce or remove
the effects of a disturbance that distort the true measurement. Examples of this disturbance
are the borehole size, mudcake and salinity.
A general technique can be derived based on a theory of Linear Perturbation which
requires no prior knowledge of the nature of the perturbation, the only requirement being
that it is approximately locally linear. Various interpretations can be made of the general
equation depending on the particular circumstances.
The technique also produces a Complementary Parameter associated with the degree of
correction. This parameter is usually discarded or paid scant regard, but can often be of
some significant value and exposes surprising information. A number of examples can be
used to illustrate these techniques, showing that they have wide applicability in situations
ranging from difficult logging conditions (e.g. density through casing) to the apparently
routine, where unusual and unexpected borehole fluids are revealed from neutron logs.

A very common method in wireline logging is to


employ a system of transducers making similar
measurements, spaced out along the logging
tool.
The main reason for this is to provide
measurements with multiple depths of penetration in order to compensate for the effects of
some disturbance to the measurement. This
disturbance can have a multitude of origins,
such as the borehole itself, its size, fluid nature,
caliper fluctuations, etc., or near-borehole effects
such as invasion.
The compensation relies on the disturbance
being common to the array of transducers, and
requires a model to describe the disturbance (e.g.
a step invasion profile).
The multiple spacings employed have differing
vertical resolutions, and much effort in recent
years has been spent optimizing this resolution
by ensuring that boundary information is not
lost. The V E C T A R (Vertical Enhancement by
Combination and Transformation of Associated
Responses) computational technique is one such
method (Elkington et al. 1990).
In this paper, we will consider the converse of
this combination method and develop equations
which are not dependent on a pre-imposed
model but are very general. In the process of
doing this, we will see that another parameter is
revealed that is orthogonal to the true value that
is being examined.

Orthogonalization
A definition of an orthogonal pair of parameters

is that although they are associated, varying one


of them does not vary the other. For example,
invasion depth and Rt (resistivity) are orthogonal, because varying Rt does not affect the
invasion depth, and vice versa.
However, if we measure resistivity using a
dual induction tool, the deep and medium
measurements are not orthogonal since variation in R t affects both deep and medium logs.
The tornado chart shown in Fig. 1 is an
attempt to transform the measurements into an
orthogonal set, the tornado being a skewed
orthogonal co-ordinate system. All dual or
multiple measurements are intended to achieve
similar objectives. It will, however, be noticed
that the transformed orthogonal pair has two
parameters--we get depth of invasion as well as
Rt. This value of depth of invasion is an example
of the orthogonal Complementary Parameter.

Linear perturbation and sharpness


Let us consider an observation O, looking at a
true value V, subject to a perturbation P. Let us
suppose we perform all the chart book corrections we can (borehole size, etc.) but we are still
left with a perturbation we cannot measure
directly. Let us further assume that the perturbation is reasonably small and that it disturbs
the observation from its true value in a linear
way. We can then write:-

0 = V+ KP

where K = t h e proportionality constant i.e. the

SAMWORTH,J. R. 1998. Complementary functions reveal data hidden in your logs.


In." HARVEY,P. K. & LOVELL,M. A. (eds) Core-Log Integration, Geological Society, London,
Special Publications, 136, 159-171

(1)

159

160

J.R. SAMWORTH

Fig. 1. A Tornado Chart--an example of a skewed orthogonal co-ordinate system.

perturbation rate. If we make two observations


with different transducers but subject to the
same perturbation we get:0 1 = V-t-

K1P

02 = V+ K2P.

(2)

(3)

We can eliminate the perturbation P from the


two equations and solve for the true value V.
With some re-arrangement, we then get:
V -- O 1 ~-

(O1 - 0 2 )

(4)

K1
This is arranged in the following form:(True value) = (Observed value) + (Correction).
It is important to note that the correction
depends on the two observations and the ratio
of the perturbation rates K2/K1 and not the
individual rates themselves. This is very signifi-

cant, as the ratio of perturbation rates is easier


to calculate, and additionally the rates can
change their absolute values without invalidating equation (4) as long as the rate ratio is
unchanged.
We can also solve equations (2) and (3) for the
perturbation P by eliminating V:/9__ 0 2 -- O1
K2 -- K1

(5)

This parameter is the Complementary Parameter


and is orthogonal to the true value. It can often
be numerically scaled into some useful unit but
is frequently ignored.
If the perturbation is due to a variety of
different effects they become lumped into a
correction that cannot be assigned an explicit
physically meaningful value, so P becomes
Unsharp. This is the price of getting a good
assessment of the true value, V, which is Sharp.
It is, however, often the case that many of the
lumped parameters are constant over the length
of the borehole. If these values can be ascertained independently, and any one of the

HIDDEN DATA IN LOGS

161

Fig. 2. Mudcake thickness from Density logs.

perturbations varies significantly over the borehole, this variable parameter can be derived
explicitly and a curve plotted. That is, it becomes

is usually one of optimization. We can encapsulate this principle thus:-

Sharp.

a computational process on a measurement can


only be justified if the result after the process is
better than the original.

Principle of betterness
Before considering examples of the application
of Linear Perturbation it is prudent to consider
our objectives. The main objective is to improve
the quality of a measurement, not necessarily to
make it absolutely correct, because this depends
on the quality of our assumed model. We can
often become unnecessarily obsessed with correctness, whereas the log interpretation process

Alternatively:if a measurement can be improved by applyNg a


computational process it is usually worth doing.
The result does not have to be correct, only better.

An example of this principle at work is the


Compensated Density Log. In the presence of a

162

J.R. SAMWORTH

API

inches

250

GM/CC

2.0

3.C
0
GM/CC

-0.25

11
i

0.25
'%
_.~~176 h

C~
,~rill-pipe)

-[

sJ

~' Density

Densi

~,/Correction

~_.Gamma Ray
6,5O

_i
_I
I
_i
I
-I
I
-I
1
-It
DEPTH BA~ED DATA FILENANE:

MAXIMUN S A M P L I N G INCRE~4ENT I O C N .
.CIB

RUN I D :

ON

AT 20:48

PLOTTED ON 0 7 - J U N - I ~

AT 0 9 : 3 7

P J E ~

Fig. 3. Compensated Density through casing.

mudcake, the corrected log is usually more


accurate than either of the two component
originals. If, however mudcake is not present,
the compensated log is subject to composite
errors from both measurements and can actually
be worse than either of them. This situation can
often occur in practice, especially in slim wells.
We can see that we must, therefore, apply the
technique circumspectly.

Application of linear perturbation


We will now consider several applications of the
theory.

Density logs
We can apply equation (4) directly to the long
and short-spacing density logs.

We then get:-

pt=pn+

where pt
PL
Ps
Ks & KL

11 ]
~

=
=
=
=

(PL--PS)

(6)

True Density
Long Spacing Density
Short Spacing Density
Perturbation rates.

This equation is identical to that derived by


applying the geometric factor theory to density
logs (Samworth 1992).
If we wish to explore the complementary
parameter we need to set up the original
equations. Density logs can be expressed in
terms of a Geometric Factor J:

HIDDEN DATA IN LOGS

163

Fig. 4. Use of a derived Apparent Caliper to improve Slim Array Induction logs.

PA= Jpm~+ (1 - J ) Pt
Pmc = mudcake density.
Rearranging;
PA = Pt + J(Pmc -- Pt).

(7)

(8)

If we approximate J to a straight line function of


standoff, d,
i.e.
we get:-

J = Kd,

PA = Pt + Kd (Pine-- Pt).

(9)
(10)

This is the linear perturbation equation from


which (6) can be derived.
We can now use equation (5) to derive the
standoff, d
i.e.

Ps--PL
d=(Pmc-P,)(Ks-KL)"

(11)

A log of this mudcake thickness is shown in Fig.


2. It is, of course, similar in character to the
density correction but is scaled in inches.

Density log through drill pipe


Occasionally, circumstances arise when the
borehole stability is so poor that it is not
possible to leave the hole open for conventional
logging. It is then possible to run a variant of the
density tool inside the drill pipe to log the
density of the formation outside the pipe. In this
case, there is no mudcake and we have no
knowledge of the borehole caliper.
A special form of density tool is employed
which has no preferential circumferential collimation, i.e. it looks all round the hole. The long
and short detectors are calibrated for the
through-pipe conditions, and linear perturbation

164

J.R. SAMWORTH

Fig. 5. Invasion indications from a Slim Dual Induction Log.

applied. The degree of correction and complementary functions are not now associated with
mudcake.
Figure 3 shows a compensated density log
obtained in this manner. If the original sharp
value required is the bulk density, we do not
have to assign an explicit meaning to the density
correction; it can remain largely unsharp.
Although unsharp, it is still probably safe to

assume that areas of high corrections are areas


of hole enlargement. If we pursue this assumption, we can compute a caliper log using
equation (11). This caliper log is shown in Fig. 3.

Caliper from array induction logs


Induction logs can be combined in a similar way.
Since inductions measure conductivity, we can

HIDDEN DATA IN LOGS

165

~.;~nc~easing
,~ ~

,,"

~176176

..-""

~176176176

Short
count
rate

Long count rate

Fig. 6. Cross plot to indicate effect of perturbations on Neutron log count rates.
set up the linear perturbation system as in
equations (2) and (3). This has been previously
explored for slimline array tools (Samworth et
al. 1994).
Figure 4 shows an example of this application.
A difficult horizontal well was logged with a slim
array induction tool, without an opportunity to
run any other log. The borehole fluid was saline
(.07 52 f2m) and since no caliper was available to
correct the logs, they were apparently quite
poor. (The right hand set of curves in Fig. 4).
Since we know the mud resistivity, by assuming
that the two shallowest measurements see no
further than the invaded zone we can use linear
perturbation to calculate an apparent caliper.
This caliper, shown in the left-hand track, was
then used to correct the deeper reading measurements. A much more systematic log then results
(in the centre track of Fig. 4). This is a form of
optimization of the induction logs, and it leads
to a better product without necessarily being
absolutely correct.

Invasion indication from induction logs


Figure 5 shows what can be achieved with a
simpler slim dual induction tool. Only two
conductivities are measured here, but unlike
the previous example, the caliper is known as
well as the mud resistivity, and the appropriate
corrections can be applied.
When linear perturbation is applied here, as
well as deriving Rt, we get a lumped complementary function associated with both Rxo and
invasion diameter. This is shown as the shaded
curve in Fig. 5 and gives some indication of the
character of the invasion. It can be seen, for

example, that there are three permeable zones,


the invasion of the lower two being uniform
since the invasion indicator is of a trapezium
shape. However, the upper zone shows a graded
form on its top edge, probably indicating a
gradation in permeability.

Borehole fluid salinity from neutron logs


We can adapt linear perturbation to the dual
neutron tool. Dual neutron tools are usually
designed so that the ratio of the count rates from
near and far detectors is related to the formation
porosity. The design is normally such that the
sensitivity of this ratio to such things as borehole
fluid salinity is minimized. However, a complementary function can be calculated specifically
to be sensitive to this salinity.
This can be seen in Fig. 6 where we have cross
plotted the short spaced count rate against the
long spaced. On this plot, all points on a straight
line through the origin have the same ratio, and
this represents the same porosity. A line can also
be drawn for constant salinity but with porosity
varying. This mesh is a skewed orthogonal
system, as described earlier.
In setting up the linear perturbation equations
in this case, we shall use the two count rates V1
and V2, which are the unperturbed values. So we
now get:
O1 = V1 +K1P

(12)

0 2 = V 2 + K2P.

(13)

We can establish a relationship between 01 and


02 by eliminating P. We get:

166

J.R. SAMWORTH

Fig. 7. Borehole salinity from Neutron logs (1).

H I D D E N DATA IN LOGS

!0.0

API

I00 5

inches

1~

(SalinityIndicator)

B/E

167

1.0
10 30
L

LST %

- 10

6600

6700
9

6800

)
\

Gamma

fk

Caliper

6900

! i~

vooo
Bit Size

7100

: k

"7:~00

r
\

7~0o

I"

7400

~,
?

I
~

~'b-

,r
Salinity

r Indicato~

r--~"
.
~.~

.~.,.~.s

7600
77OO

.-.,~

Ne

7gO0

I
\)

-,ooo

7900

,L ;

"~

,,.&~

e-J
_N

i)'

2~

8100

8400-

)I

8~00

8600

iI
bi)

87oo

f f

'8800

i
~"

j ogoo

Fig. 8. Borehole salinity from Neutron logs (2).

-g
2~."

%-N

168

J.R. SAMWORTH

0.0
API 100l

(SalinityIndicator)

inches 1

B/E

10]30

6600
6700
6800
6900

7"Caliper 70OO
7100

~-Bit Size

sNeor

7200
tm~.

Salinity

7400
7~00
7600
7700

t',

7800
7
7900

8000
8100
8200
8:S00
8400
8~00

8600
8700

8800
18900

Fig. 9. Borehole salinity from Neutron logs (3).

PE

7300

<

1.0 I
LST%

-loI

HIDDEN DATA IN LOGS


API 150 3
,/

inches 13

(SalinityIndicator)

169
10.2

ohm-m

2000

......

~-.~ " ' B i t Size

'

A..~Caliper

~' Gamma

I5 2 0 0

~~ ~

5300

_- ~
_-,~

~
"Salinity
Indicator

~ Medium
t~Dee p

154oo E
5,5oo

-!
5600 -

5"700 :i
b
,58oo

--

OE]>THBA~s DATA- NAXIMUHSAMPLINGI ~ T


FTLENANE:
.CZB RUINI0:

1004. RECOROED
ON
AT 14:2:5
PLOTTEDON19-,JUN-1996AT 14:07

4'
Fig. 10. Borehole salinity in a horizontal well (1).

K1

Kx V2).

0 1 "-" ~22 O 2 "~- ( Vx - - ~ 2

(14)

This is a straight line equation of the form


O1 = m 0 2 + (constant).

(15)

For the particular case of ratio processing:


K1 01
V1
m = / ( 2 - 02 -- V2"

(16)

Since we actually observe Ol and O2, m can be


calculated and we can migrate along the line
until we reach V] and V2. This method has

previously been explored in some detail (Samworth, 1991). If, however, we can identify the
positions of O] and 02 on the ratio line, we can
identify which line of constant salinity we are on,
and we can then estimate the borehole salinity.
Some examples now follow to show the
usefulness of this complementary parameter.
Figures 7, 8 and 9 show sets of logs, on a
compressed vertical scale in a dolomite reservoir.
The field was being produced by an injected
waterflood, and the wells were close to each
other. The reservoir section is the whole of the
lower halves of the wells where the gamma ray
log activity increases.

170

J. R. SAMWORTH
)

API

150 3

inches 13

(Saliniq Indicator)

1 30

LST %

-10

~,

/ 1~..,..,.Bit Size
51 O0

5200

c,
ma

Cahper
5300

5400

55O0

560O

5700

5800

59OO

DEPTH BASED DATA FILENAJWE :

MAXIWKJM SAMPLING INCREMENT I OCM.


.ClB

RUN I D :

RECORDED ON
PLOTTED ON l O - J U N - 1 9 9 6

AT 1 3 : 3 8
AT 1 1 : 1 ' : ;

Fig. 11. Borehole salinity in a horizontal well (2).


Figure 7 shows a well where the well fluids
were static. The salinity indicator shows low
salinity, i.e. oil, above a high salinity sump in the
reservoir section. Figure 8 shows a log taken
with the well flowing, i.e. the injector had not
been turned off. Here the salinity profile is
inverted, but the inversion starts several hundred
feet into the reservoir.
Figure 9 is similar to Fig. 7, but with a blip at
a similar place to where there is a change in Fig.
8.
The conclusion from these logs must be that
the waterflood is breaking through at the top of
the reservoir and not efficiently sweeping the
lower levels.

Figure 10 shows a horizontal well where there


are several intervals with an anomalous Array
Induction response (e.g. at 5810-5910). The
neutron based salinity indicator shows high
levels at these points indicating water plugs in
an otherwise oil-filled well. The nuclear logs are
shown in Fig. 11 for reference. Comparison of
the logs with a plot of the hole trajectory shows
depressions at these points, the salinity indicator
showing that these are full of water.

Conclusion
There is much information to be had from well
logs by interpreting them in a slightly unconven-

HIDDEN DATA IN LOGS


tional way, so it is imprudent to discard any
data, especially raw data.
The linear perturbation technique is completely general, and does not rely on any particular
physical model. It is applicable to a wide variety
of logs where multiple measurements of a similar
type are made. The method also produces a
complementary parameter which can be very
useful in revealing effects not apparent on the
normal logs.

This paper illustrates some of the work carried out by


the Research and Development Department of Wireline Technologies Ltd, and grateful thanks are given to
that company for permission to publish.

171

References
ELKINGTON,P. A. S., SAMWORTH,J. R. & ENSTONE,M.
C. 1990. Vertical enhancement by combination
and transformation of associated responses.
Transactions of the 31st Annual Logging Symposium, SPWLA, Paper HH.
SAMWORTH, J. R. 1991. Algorithms for compensated
neutron logging--57 varieties. Transactions of the
14th European Logging Symposium, SPWLA,
Paper A.
- 1992. The dual-spaced density log, characteristics, calibration and compensation. Log Analyst
33, 4249.
, SPENCER, M. C., PATEL,H. K. & ATACK,N. A.
1994. The array induction tool advances slim hole
logging technology. Transactions of the 16th
European Logging Symposium, Paper Y.

In situ stress prediction using differential strain analysis and ultrasonic

shear-wave splitting
B. W I D A R S O N O , 1 J. R. M A R S D E N 2 & M. S. K I N G 2
1Lemigas, Jakarta, Indonesia
2 Department o f Earth Resources, Engineering Royal School o f Mines, Imperial College,
London S W 7 BP, U K
Abstract: Knowledge of the /n situ states of stress in rock masses is of considerable
importance to a number of subsurface engineering activities, including those involved in
exploiting petroleum and geothermal energy reserves. In this paper, a comparison is made of
two laboratory techniques, based upon stress-relief microcracks, for determining the in situ
state of stress: differential strain analysis (DSA) and ultrasonic shear-wave splitting
(USWS). Measurements on ten sandstone samples recovered from deep boreholes, made
using the well-established technique of DSA, have been compared to those made by the
comparatively new technique of USWS and to sleeve fracturing measurements of in situ
stress made in the corresponding boreholes. The results obtained indicate that the USWS
technique, with its ability to test a large number of samples quickly, provides a useful
adjunct to DSA and sleeve fracturing in determining trends in in situ stresses. Used in
combination, the two laboratory techniques have also proved useful for examining rock
micro-structural features.

A number of operations involved in the exploitation of petroleum and geothermal energy


resources require a knowledge of the in situ state
of stress. Such data are required for determining
borehole stability in the drilling phase, for
avoiding solids production problems and for
hydraulic fracturing stimulation in the production phase, and for reservoir characterization in
reservoir engineering. A common method for
obtaining in situ stress data from great depth is
indeed by hydraulic fracturing or, alternatively,
sleeve fracturing (Desroches et al. 1995). These
techniques, however, possess certain disadvantages with regard first to cost and second to
technical considerations in fractured formations,
deviated well bores and high pressures and high
temperature formations. Results obtained from
these methods are often influenced and biased by
stresses close to the well bore and hence do not
reflect the governing in situ stress field. To
overcome some of these problems, the technique
of differential strain analysis (DSA) has been
developed. Since it was first suggested by
Strickland & Ren (1980) as a tool for in situ
stress determination, DSA has been used frequently; successful applications have been reported by various investigators (Dey & Brown
1986; Dyke 1988; Oikawa et al. 1993).
Nevertheless, considerations of the length of
time required for a DSA test have led to efforts
to find alternative methods. Early studies (Yale
& Sprunt 1989) utilized the phenomenon of
ultrasonic S-wave splitting on rock specimens

taken from great depth which contain stress


relief microcracks. The main objective of this
study is to contribute to the development of the
technique and to compare the results obtained
with those from other proven methods.
For the purpose of this study, specimens were
prepared from core samples taken from various
petroleum wells in the Irish and North Seas. The
samples comprised ten sandstone specimens
from depths lying between 1361 and 1422 m
and between 3232 and 3304 m, eight of which
(SSA) were oriented and so could be used for
determining the actual orientation of the in situ
stresses. The other two samples (SSB) were not
oriented; they are, however, considered valuable
for further comparison studies. Descriptions of
the rock samples are provided in Table 1.

Differential stain analysis (DSA)


Technique
DSA is a method based on the existence of
oriented microcracks generated within core
samples as a result of stress relief processes
which occur during the core process and
recovery of the cores from depth. Evidence of
the existence of this type of microcracks has
been reported by various investigators using
different approaches (Kowallis & Wang 1983,
amongst others), such as comparing images
from scanning electron microscopy (SEM),
studying P- and S-wave velocities adjacent to

WIOARSONO,B., MARSDEN,J. R. & KIN6, M. S. 1998. In situ stress prediction using differential strain
analysis and ultrasonic shear-wave splitting In. HARVEY,P. K. & LOVELL,M. A. (eds) Core-Log
Integration, Geological Society, London, Special Publications, 136, 185-195

185

B. WIDARSONO ET AL.

186

Table 1. Description of sandstone samples


Sample

Depth
(m)

Grain size
(mm)

Comments

SSA-1
SSA-2
SSA-3
SSA-4
SSA-5
SSA-6
SSA-7
SSA-8
SSB-1
SSB-2

1361
1363
1364
1366
1374
1399
1409
1422
3232
3304

0.5
0.1-1.0
0.5-1.0
0.54).75
0.5
0.14).5
0.14).5
0.5
0.14).5
0.14).5

Well cemented, very weak bedding


Well-cemented, poorly-sorted, weak bedding
Fairly well-cemented, strong bedding
Well-cemented, no sign of bedding
Well-cemented, weak bedding (possible micaceous laminae)
Dark red, very weak bedding
Well-cemented, weak bedding/laminae
Well-cemented, strong bedding
Well-cemented, no bedding/laminae
Well-cemented, no bedding/laminae

the borehole and in the laboratory, and employing differential strain analysis itself. For example, Teufel (1983) observed anisotropy of P-wave
velocity measurements and correlated this with
results from the anelastic strain recovery technique (ASR).
Basic assumptions of the technique are presented in length by Strickland & Ren (1980)
who, in brief, assume that aligned microcrack
densities in different axes are proportional to the
relieved stress magnitudes in these axes. Consequently, when the sample is compressed hydrostatically, the resulting strains will show
preferential orientation in magnitudes which
are proportional to the microcrack densities. A
further assumption is that all microcracks
existing within a tested sample are of stressrelief
type, or at least that all pre-existing microcracks
do not affect sample deformation significantly.

Experimental procedure
Three basic steps are followed in specimen
preparation: machining the specimen, attaching
the strain gauges and encapsulating the specimen. Each step in specimen preparation must be
performed carefully in order to prevent generation of new microcracks in the rock sample. Flat
surfaces are machined on each specimen in at
least three orthogonal directions, by first sawing
them with a diamond saw and then by hand
lapping or surface grinding. After ovendrying at
35 ~ strain gauge rosettes are attached to the
specimen following the arrangement shown in
Fig. 1. The rock sample is finally encapsulated
with epoxy resin in an elastomer membrane.
The strain gauged and encapsulated specimen
is placed in an oil-filled pressure vessel and left
for approximately 24 h in order to ensure
temperature stabilization. This procedure avoids
any temperaturerelated fluctuations while strain

Fig. 1. Axes system and strain gauge orientations.

gauge readings are made during the test.


Hydrostatic pressure on the specimen is first
increased in steps of 200 psi to 21 MPa (3000
psi), then in steps of 500 psi to 55 MPa (8000
psi). During this loading procedure, a transition
from microcrack closure to intrinsic elastic
deformation is generally found to occur. From
55 MPa (8000 psi) to the maximum pressure
(usually around 83 MPa-12000 psi) 1000 psi
increments are usually chosen since microcrackfree linear stress-strain behaviour generally
occurs in this range of pressures. A strain
gauged and sleeved fused silica specimen of
known physical properties is also tested adjacent
to the rock test specimen to provide any
corrections necessary for environment-related
non-linearities in the specimen and strain gauge
responses.

Procedure of analysis
Additional input data is required for analysing
DSA measurements, including: vertical in situ or

IN SITU STRESS PREDICTION


Pressure

(MPa)

100

80

60

~G a u g, e _ _ _, _

40

2O
0
0

Strain(millist rains)

Fig. 2. Typical compression curves from DSA test.


overburden stress, /n situ pore pressure, and
Poisson's ratio for the tested rock. The orientation of the reference line with regard to the axes
of the cubic specimen is also required if the true
orientation of the in situ stress field is to be
determined.
The strains recorded by the data logger are
fitted by a series of curve-fitting programs to
produce the pressure-strain curves by way of a
quadratic fit using five adjacent data points
(Dyke 1988). A microcrack closure strain tensor
is obtained from these quadratic compression
curves (Fig. 2) for each of the specimens tested.
To create a complete microcrack strain tensor,
six components only are required from the
twelve strain gauge measurements. This permits
a statistical analysis to be made of the redundant
data. From the microcrack strain tensor, the
principal microcrack strains and their orientations relative to the reference line are calculated
using the method proposed by Strickland & Ren
(1980). The ratios of principal strains are taken
as the ratios of the principal effective stresses
and, by a series of tensorial transpositions to
vertical and horizontal planes, the principal
strains can be converted to principal in situ
stresses using values of overburden stress, pore
pressure and Poisson's ratio. Furthermore, the
true orientations of the stresses can be determined if the true orientation of the reference line
is known.

Ultrasonic shear wave splitting (USWS)


Technique
The use of acoustic S-wave splitting (birefringence) as a source of information regarding the

187

medium through which it propagates remains


relatively novel despite its origins in studies of
earthquake seismology. Despite the abundance
of observations, it is still not clear exactly what
causes S-wave splitting in the Earth's crust
(Crampin & Lovell 1991), although it is generally taken to be caused by aligned discontinuities. Crampin (1978) recognized that most
rocks in the crust are likely to contain discontinuities, and that S-wave splitting is probably the
most diagnostic feature of wave propagation in
such anisotropic rocks.
Attempts have been made to relate S-wave
splitting to the orientation of in situ stress-relief
induced microcracks in cores taken from great
depths. Yale & Sprunt (1989) utilized ultrasonic
S-wave splitting on oriented core samples, and
concluded that this approach can be used to
predict the orientation of the major horizontal in
situ stress.
Shear-wave splitting results from the division
of a polarized S-wave into two separate polarized S-waves travelling at different speeds when
a source of anisotropy is encountered in its path.
When a plane polarized S-wave is propagated
through a medium containing a set of aligned
microcracks it will be split into two orthogonally
polarized S-waves, with one polarized parallel to
the plane of the microcracks and the other,
travelling at a lower velocity, polarized perpendicular to the plane of the microcracks. Garbin
& Knopoff (1975) proposed a theory to explain
the velocity variations caused by a single set of
parallel cracks which is based on scattering of
elastic waves by penny-shaped cracks. The
theory explains the variation of S-wave velocity
with changes in ray path angle and wave
polarization relative to the crack plane. The
degree of splitting is related to the time lag
between the arrival of the two waves at the
receiver. The degree of S-wave splitting and
hence velocity anisotropy increases with an
increase in crack density.

Experimental procedure and analyses


As part of this study, S-wave splitting tests were
conducted on the same specimens as used in the
earlier DSA tests, except in the case of one which
exhibited such a poor degree of cementing that
satisfactory acoustic coupling between specimen
and transducers could not be achieved. The
elastomer sleeves were removed from the DSA
samples tested earlier and the latter were re-cut
with flat surfaces perpendicular to the vertical
(Z-axis) and horizontal (X- and Y-axes).
They were then oven dried at 35~ so that the
specimens could be tested dry. The principal

188

B. WIDARSONO ET AL.

Table 2. In situ stress data from DSA tests


0.1

0.2

0.3

Sample

o1/0. V
1

Azimuth
(0)2

Dip
(0)2

0"2/0"1

Azimuth
(0)2

SSA-1
SSA-2
SSA-3
SSA-4
SSA-5
SSA-6
SSA-7
SSA-8
SSB-1
SSB-2

1.005
1.017
1.010
1.130
1.000
1.059
1.001
1.004
1.004
1.008

204N
106N
116N
74N
38N
331N
229N
164N
240
179

80
73
76
59
87
66
86
81
83
74

0.892
0.824
0.877
0.939
0.832
0.760
0.902
0.866
0.888
0.928

71N
3 08N
213N
351N
293N
52N
326N
20N
132
88

Dip 0"3/0"1
(O)
7
17
2
4
3
2
0
7
2
1

0.849
0.807
0.832
0.805
0.817
0.676
0.824
0.835
0.763
0.879

0"3/0"2

Azimuth
(0)2

Dip
(o)

0.955
0.982
0.948
0.858
0.982
0.983
0.914
0.966
0.858
0.948

340N
219N
304N
262N
203N
140N
57N
290N
42
358

7
1
14
26
3
23
4
6
7
16

1
0.v = vertical or overburden stress.
2 Azimuths measured clockwise with respect to North (Z-axis) or, for unoriented cores (SSB), clockwise from
reference line (X-axis).

axes (Fig. I) and reference lines used were the


same as those used in the DSA tests, in order to
maintain compatibility between the DSA and Swave splitting results.
USWS measurements are performed by rotating the rock specimen containing stress relief
microcracks while pulses of planar S-wave are
transmitted parallel to the specimen axis under
nearatmospheric pressure conditions. In the
presence of aligned microcracks, S-wave first
arrivals observed by the receiving transducer
(polarized parallel to the polarization of the
transmitter) show changes in magnitude as the
specimen is rotated. When the direction of
polarization of the transducers is parallel to
the aligned microcracks, the S-wave first arrival
time is a minimum. Conversely, when the
transducer polarization is perpendicular to the
aligned microcracks, the first arrival time is a
maximum. This direction is that of the greatest
stress relaxation and hence is the major in situ
stress direction.
Each specimen was first assembled between
pairs of broadband transducers having S-wave
frequencies in the range 300 to 800 kHz (as
described by King et al. 1995) with a proprietary
visco elastic S-wave couplant and a stress of 2.5
MPa applied to the transducers to provide good
acoustic coupling. This level of axial stress has
been shown experimentally to have a negligible
effect on cracks oriented sub-parallel to this
direction of propagation of S-waves. At the start
of a test, each specimen is arranged so that the
transducer polarization is in the Y-axis direction
and the propagation of acoustic energy is either
in the Z-axis direction (vertical) or X-axis
direction (horizontal). During a test, the specimen is rotated through an angle between 0 ~ and

180 ~, measured relative to the reference line, in


increments of 15~ At each 15~ step, the transit
time of first arrivals and waveforms in digital
form are recorded. After each test, the S-wave
transit time is converted to velocity.

Experimental results
D i f f e r e n t i a l s t r a & analysis

Results of the in situ stress predictions for all


specimens tested are shown in Table 2, in which
stress magnitudes are presented as ratios in
order to provide a comparison of results. Figure
3 shows plots of the results for SSA sandstones
in the form of an equal-angle stereonet (lowerhemisphere projection). Trends of the stresses
determined from the oriented cores of the SSA
specimens are given in degrees measured clockwise from North, whereas those obtained from
the two unoriented SSB sandstones are simply
measured clockwise from an arbitrary reference
line. Consequently, the results for SSB samples
are not plotted on stereonet projection, since no
common reference line exists among the specimens tested, and such plots could imply
misleading relations.
The results shown in Table 2 and Fig. 3 show
that the major principal stress (o.~) lies near
vertical, to within 0 ~ to 23 ~ This result can be
regarded as sufficiently precise for a technique
based on stress relief microcracks, since it is
common that the orientations of the microcracks are modified by grain scale inhomogeneities. The similarity in magnitudes between
major (o1) and vertical (i.e. overburden) stresses
(o'v), represented by the ratio o'1/o'v being near to
unity, also indicates that o'1 lies in the vertical

IN SITU STRESS PREDICTION

North
f
1

6
b

7%o8~3 2
1

3
0(~. 2
1,2,3 ... sample number

A~ 3

Fig. 3. Lower hemispherical projection showing in sltu


stress orientations from DSA tests on sandstone SSA.

direction. It can also be concluded from the


results for SSA sandstone that the depths from
which the core samples were recovered are not
far above the depth at which o.H becomes equal
to o-1, as indicated by the closeness of the ratios
o.2/o.1 and o.3/0-~ to unity. This interpretation is in
accordance with global data for the vertical
horizontal stress ratio versus depth compiled by
Brown & Hoek (1978).
In the case of the intermediate and the minor
stresses, 0-2 and 0-3, it is evident that these lie in a
horizontal plane, or at least sub-parallel to
horizontal. However, it is obvious from the
scatter observed that the orientations of the two
horizontal stresses are interchangeable. This is
understandable, since the two stresses are very
close in magnitude, and each has caused a
similar degree of oriented microcracking in the
samples. If these ratios (0"3/0" 2 column in Table 2)
are averaged, a minor-intermediate stress ratio
of 0.95 (with the exclusion of SSA-4, which is
significantly different from the others) is obtained. Thus the difference of stress relief
microcrack density in the two principal directions is minimum and, taking into account
experimental and analytical error (e.g. determining ranges for slopes of the pressure-strain
curves), the implied stresses could be interchanged in direction and magnitude. At this
point, it is worth noting that, with a mere 5%
error bar to represent the errors, both trends and
magnitudes of the two horizontal stresses are
interchangeable. In fact statistical analysis of
each over-determined DSA dataset (which was
necessary to obtain the best fit tensor of crack
closure strains) yielded a maximum error of

189

between approximately 2% and 7% with regard


to the principal stress ratios from any single
DSA tests, and an error 'cone' of approximately
5~ to 12~ for the principal stress orientations
from any single test. From the hemispherical
plot of all the oriented DSA data (Fig. 3) it can
be seen that the combined results yield a scatter
of approximately 30 ~ in the azimuths of the
horizontal (i.e. intermediate and minor principal) stresses.
However, this does not imply the statistical
analyses of the DSA results could just have
easily yielded horizontal stress orientations in
any azimuths from 0~
~ since this would
have necessitated the microcrack densities and
stress magnitudes being isotropic in the horizontal plane. Whilst this is possible, it is only so
for those relatively few cases where the magnitudes of the minor and intermediate stresses are
exactly the same. For all other combinations
within the limits or error, the general directions
of the principal stress orientations are as seen in
Fig. 3 and only the magnitudes vary. Thus,
although the intermediate and minor stresses are
very close in magnitude, the DSA method has
still been able to identify the general orientations
of the stresses.
From the hemispherical projection, it is clear
that one of the two horizontal stresses lies within
the range 290 ~ to 350 ~ from North, with the
scatter in azimuth due probably to variation in
the orientation of the stress-relief microcracks
caused by grain-scale heterogeneities. On the
other hand, the other horizontal stress lies
within the range 20 ~ to 71 ~ from North such
that, with the two horizontal stresses being
similar in magnitude, the intermediate principal
stress (o.H) can lie in either of the ranges 290 ~ to
350~ or 20 ~ to 71~ Arguably, therefore, o.H
can lie in either of the ranges 310+10 ~ N (or
130+10~
or 220+10~ (or 40+10~
These
results are in reasonably good agreement with
results of an earlier study reported by Desroches
et al. (1995), who conducted analyses of the in
situ state of stress in this area using DSA on
samples from the very same core sections as
tested in this study, as well as using hydraulic
fracturing and sleeve fracturing in downhole
tests in the same well and at the same depths.
The earlier combination of the results from these
three techniques indicated o.h to lie in the range
N65~
~ Note also that this earlier study
showed that stress data from DSA analyses on
these core sections were not influenced by slight
anisotropy in the samples nor by variations in
the elastic properties or rock strengths over the
cored intervals.
The in situ stress prediction from the two SSB

190

B. WIDARSONO ET AL.

136oN

166oN

196~

226oN

256oN

286oN

316oN
t

60

63

66

69

72

75

Transit time (pSec)

60

63

66

69

72

Transit time (pSec)

Fig. 4. S-wave waveforms for sample SSA-1 at various


rotation angles. (Propagation in vertical Z-axis, and
polarized in Y-axis direction at X-axis reference line.
Arrows indicate detected first peak).

Fig. 5. S-wave waveforms for sample SSA-I at various


rotation angles. (Propagation in horizontal X-axis, and
polarized in Y-axis direction at X-axis reference line.
Arrows indicate detected first peak).

sandstone specimens is similar to the results for


SSA, even though the true orientations cannot
be determined due to the unoriented nature of
the core. Since the core is from vertical boreholes, it can be inferred from Table 2 that the
major principal stress lies near-vertical; consequently the other two principal stresses lie nearhorizontal.

in other directions. Although this behaviour is


exhibited by all specimens tested in this study,
there is one case (SSA-8) where signal attenuation was so severe that it proved impossible to
pick the first arrival time. The degree of
attenuation, nevertheless, varies from one specimen to another in a manner related to the state
of cementation of the specimens (Table 1).
For the purpose of predicting in situ stress
orientations, transit time values were identified
and selected from the waveforms. Since the Swave signals observed during the experiment
vary in quality due to different degrees of
attenuation, it was found that the S-wave
velocity calculated from the first peak (or
trough) is more reliable for interpretation than
the group velocity calculated from the first
arrival.
Figure 6 shows examples of the variation in
first peak velocity with rotation angle as seen for
sample SSA-1. The velocity plots are shown with
an error bar of +0.5 %, estimated as representing the confidence in picking transit time values
(+0.125 #s) from waveforms due to variations in
signal quality and oscilloscope resolution. In
general, the velocities plotted against rotational

S h e a r w a v e splitting

During each test, a set of waveforms was


recorded for each 15~ rotation relative to the
reference line (X-axis for the vertical, and Z-axis
for the horizontal wave propagations). Figures 4
and 5 illustrate examples of waveforms recorded
during measurements on SSA-1 in the vertical
(Z-axis) and horizontal (X-axis) directions.
As expected, the plots show that the transittime (At) varies with rotation angle. This
variation in At can be considered as an
indication of the existence of an oriented set of
cracks, or at least (provided that a homogeneous
background pore system exists) is more influential in reducing S-wave velocity than any other
sets of microcracks with lower density oriented

IN SITU STRESS PREDICTION

S-wave [first peak]


velocity (m/s)
2000
a) Z-direction propagation
Y-axis polarization

191

minimum velocity direction. For horizontal


propagation (X-axis), the vertical (or sub-parallel) in situ stress (o-v) coincides with the direction
of maximum velocity, as in the case of 0-H for
vertical propagation. Both 0-H and 0"h are
orthogonal to O-v.

19oo

Table 3. in situ stress orientations from USWS tests

1-_+0.5%

18oo

1~o

'

18o

2~o

'

'

24o

Sample
270

300

crI inclination O"H azimuth


(~
vertical)
(o)l

~h azimuth
(o)l

o from North

S-wave [first peak]

velocity (m/s)
2100
b) X-direction propagation
Y-axis polarization

2000

19oo

~b

6;

9'o
90

120
o from vertical

150

180

Fig. 6. Variation of S-wave first peak velocity with


rotation angle for sample SSA-1. (a) Vertical Z-axis
propagation, polarized in Y-axis direction; (b) horizontal X-axis propagation, polarised in Y-axis direction.

angle result in a sinusoidal pattern, indicating


the presence of azimuthal velocity anisotropy
over 90 ~ As shown in Fig. 6, similar behaviour
is also observed in both the vertical and
horizontal directions, indicating the presence of
aligned stress relief microcracks of different
densities as the source of anisotropy. Generally,
for SSA sandstone specimens, S-wave velocity
splitting of 1 to 3.5% and 3 to 13.5% (relative to
the highest velocity) for vertical and horizontal
propagation, respectively, are observed. For
SSB samples, SSB-1 shows 2.2% splitting,
whereas SSB-2 shows 1.1% and 1.6% velocity
splitting, respectively in the vertical and horizontal propagation direction.
Accepting the hypothesis of a relation between S-wave splitting and in situ stress relaxation and orientation of stress relief microcracks,
velocity plots such as in Fig. 6 indicate the
orientations of the in situ stresses. From the
velocity profiles, for S-waves propagating in the
Z-axis direction, the major horizontal stress (O-H)
lies in the direction at which maximum velocity
occurs, whereas the minor horizontal stress (o-h)
lies at right angles to it, as indicated by the

SSA-1
SSA-2
SSA-3
SSA-4
SSA-5
SSA-6
SSA-7
SSA-8
SSB-1
SSB-2

0
75
2
0
0
0
0
0

7 6 ( 2 5 6 ) N 166(346)N
5 3 ( 2 3 3 ) N 143(323)N
83(263)N
197(17)N
120(300)N
343(163)N
106(286)N
105(=285)
75( = 255)

173(353)N
107(287)N
210(30)N
253(73)N
196(16)N
15(--195)
345(= 165)

1Azimuths measured clockwise with respect to North


(Z-axis) or, for unoriented cores (SSB), clockwise from
reference line (X-axis).
2 = Poor acoustic coupling due to poor cementation of
the rock.
Table 3 summarizes the results of the S-wave
splitting measurements. Note that USWS measurements on SSA-3 and SSA-8 (Z-axis propagation) were not carried out, due to poor
transducer-sample acoustic coupling. The results
show that one of the principal in situ stresses lies
in the vertical direction, even though accuracy is
limited by the 15 rotational sampling. The
vertical stress lies within 7.5 of the peak of the
velocity profiles shown in the plots (except for
sample SSA-2). The same resolution limit
applies to 0-H. Furthermore, from comparisons
between the horizontal and vertical velocity
plots, it is apparent that 0-v is the major principal
stress, o1, as indicated by relatively larger
velocity anisotropy for horizontal than for
vertical propagation. To illustrate this, 1 to
3.5% velocity variation exists over 0~ ~
relative to the highest velocity for vertical
propagation, in comparison to 3 to 13.5% for
horizontal propagation in the SSA sandstone.
Consequently, 0-2 and 0-3 lie in the horizontal
plane.
As seen from DSA, the trends of the
horizontal in situ stresses are more easily
identified if illustrated on stereonet projections.
Figure 7 illustrates the orientations of the major
horizontal in situ stress axes for the SSA
sandstones plotted. As with the DSA stereonet
projections, the orientations are relative to

192

B. WIDARSONO ET AL.
Reference line
~ / 7 7

/
e C~,H
1,2,3 ... sample number

Fig. 7. Lower hemispherical projection showing rh


orientations from DSA tests on sandstone SSA.

North. The projection in Fig. 7 and the


corresponding DSA results (Fig. 3), show that
the major horizontal stress can lie either in the
N W - S E or N E - S W directions. At this stage,
therefore, no definitive conclusion can be drawn
regarding the orientation of 0.iJ.

Comparison of the two techniques


Orientation o f in situ stresses

In comparing the results from the DSA and the


S-wave splitting, only the orientations of the
stresses can be considered. No comparison can
be made of stress magnitudes, since S-wave
splitting tests do not provide stress magnitudes,
even though a limited qualitative assessment can
be made using velocity anisotropies. A summary
of results of the two techniques are listed in
Table 4. In the table, only the results for aH and
ah from acoustic measurements and 0.2 and 0.3
from DSA are compared, since almost all
analyses of S-wave splitting tests have shown
that the principal stress 0.1 is essentially vertical
(i.e. they show a minimum velocity at 90 to the
vertical).
In general, the results for SSA sandstone have
shown that there is a reasonable agreement
between the two techniques, although some
inconsistencies appear when individual results
are studied. It is clear that only SSA-I and SSA7 exhibit total agreement between trends for ei-i
(USWS) and o2 (DSA), which is by definition
correct for this case (i.e. intermediate principal
stress is the major horizontal stress). The rest of
the samples have, in contrast, tended to show

agreement between trends for 0.I-1 (USWS) and


0.3 (DSA), which is incorrect by definition, since
0.3 is the minor principal stress, and for this case
it should coincide with the minor horizontal
stress (0.h). This behaviour can be explained by
the ratio 0.3/0.2 approaching unity (approximately 0.95), such that differences in microcracking in the two principal directions are
minimal. Nevertheless, as indicated by the
stereonet projection in Fig. 7, there is also no
clear consistency in 0.n orientations, suggesting
no clear consistency in the orientation of the
vertical microcracks. This appears to be due to
the effect of grain scale heterogeneities, over
which the far-field horizontal in situ stresses are
not large enough to maintain sufficiently large
and uniform tensile forces across the rocks at the
granular scale.
Grain scale heterogeneities certainly appear to
have their effects in 0.1 prediction. The DSA
results for SSA sandstone in Table 4 show that
the orientation of 0.1 varies within a range 00-23 ~
from the vertical, whereas the USWS results
show almost all the 0.1 to be vertical (Table 3).
Undoubtedly, rotational sampling of 15~ reduces the accuracy of stress orientation, and it is
likely that grain scale heterogeneities (reflected
in local preferential orientation of the microcracks) contribute significantly to any disagreements between the two techniques. Such
heterogeneities certainly have greater effects on
DSA measurements, since the portions of the
specimen measured in DSA are limited to the
surface areas underneath the strain gauges. This
is much less representative than the rock mass
tested by the USWS technique. The results from
S-wave splitting are probably more reliable in
this particular case.
The results for SSB sandstone in Table 4 show
a reasonable agreement between the two techniques, even though the results for SSB-2 exhibits
about 30 ~ difference between o-2 and 0.H, and
between 0.3 and 0.h- Although there is no
evidence, this disagreement is probably caused
by other sources of acoustic velocity anisotropy
such as preferential alignment of sandstone
grains.
The differences in the fundamental concepts
of the two techniques provide, nevertheless,
advantages and disadvantages for both techniques. In S-wave splitting, any acoustic propagation is influenced by the 'averaged' properties of
the whole specimen volume, and therefore
representative of the whole specimen. In contrast, any microcrack strain measured in a DSA
test represents only the areas covered by the
strain gauges, which are generally small compared to the overall specimen dimensions.

IN SITU STRESS PREDICTION

193

Table 4. Comparison of/n situ stress orientations from DSA and USWS
DSA

0-1

S-Wave Splitting

O'H

Oh

Sample

Azimuth

Dip

Azimuth

Dip

Azimuth

Dip

Azimuth

Azimuth

SSA-1
SSA-2
SSA-3
SSA-4
SSA-5
SSA-6
SSA-7
SSA-8
SSB-1
SSB-2

204
106
116
74
38
331
229
164
240
179

80
73
76
59
87
66
86
81
83
74

71
308
213
351
293
52
326
20
132
88

7
17
2
4
2
2
0
7
2
1

340
219
304
262
203
140
57
290
42
358

7
1
14
26
3
23
3
6
7
16

76(= 256)
53(-- 233)
2
83(= 263)
197(= 17)
120 = 300)
343 (= 163)
106(=286)
105(= 285)
75(=255)

166(= 346)
143(= 323)
2
173(= 353)
107(= 287)
210( = 30)
253(= 73)
196(= 16)
15(= 195)
345( = 165)

(o)l

~
(o)

(o)1

~
(o)

(o)1

(o)

(o)1

(o)1

1Azimuths measured clockwise with respect to North (Z-axis) or, for unoriented cores (SSB), clockwise from
reference line (X-axis).
2 = Poor acoustic coupling due to poor cementation of the rock.

However, in the presence of large discontinuities, the reverse is true. Large discontinuities in
specimens influence acoustic wave propagation,
whereas any non stress relief behaviour in the
DSA tests can usually be avoided by strain
gauge emplacement. Consequently, DSA can
produce more reliable results in this case.
The DSA results in Table 2 show that the
principal stresses do not lie in exactly horizontal
or vertical directions. In other words, the
induced cracks in most cases are not exactly
parallel or perpendicular to the vertical axis in
situ. However, the reasonably good agreement
between the two techniques has shown that
acoustic velocity anisotropy can still be observed
even though the microcracks dip from these
directions, which is more often than not likely to
be the case. This fact is very important in any
effort to establish USWS as an alternative
technique for in situ stress determination, bearing in mind that it is most likely that grain scale
heterogeneities can cause local deviations in
microcrack orientations.

Configuration
system

o f stress relief microcrack

In DSA the principal strains are determined


from analysis of the microcrack closure strain
tensor obtained from a test. The principal strains
obtained provide the orientations of the stress
relief microcracks, since it is assumed that the
greatest strain occurs in the direction normal to
the plane of the microcracks. In other words, it
is assumed that it is represented by three

dominant and mutual perpendicular microcracks sets. Although this assumption is logical,
as demonstrated by Charles et al. (1986), DSA
does not provide a direct illustration of the
microcrack system. Direct observation such as
scanning electron microscope (SEM) of oriented
samples used in conjunction with DSA can,
however, provide an insight into microcrack
orientations and distributions.
The results of S-wave splitting measurements
can provide, to some extent, additional information on microcrack configurations. Velocity
plots, such as shown in Fig. 6, have demonstrated that velocity anisotropy can occur
between measurements in the vertical and
horizontal directions. There are several factors
that can lead to such acoustic anisotropy, and to
S-wave splitting in particular, but it is generally
accepted that aligned cracks are the major cause
of S-wave splitting (Crampin & Lovell 1991). At
the small scale, such as in USWS tests, there is
always a possibility that other sources of
anisotropy, such as lamination and crystal
alignment, can contribute to the overall anisotropy. As shown in Table 1, SSA sandstone
samples exhibit signs of bedding (between 4 ~ and
22 ~ from the horizontal), even though not all
samples were found to exhibit dominant bedding. However, in this study, there is evidence
that the presence of bedding planes has not
contributed to the velocity anisotropy for the Xaxis (horizontal) propagation direction. For
example, the results for SSA-5 (Table 3) show
that, despite the 18.5 ~ dip in bedding (Table 1),
the USWS measurements indicate the presence
of a set of horizontal microcracks as indicated

194

B. WIDARSONO E T AL.

by 0 ~ in the O 1 column. Another example is seen


with SSA-2, for which the bedding dipped at 22 ~
from the horizontal and where measurements in
the horizontal direction indicated the minimum
velocity to be reached at a rotation angle of 165~
(i.e. 15~ from the vertical, implying Crl acts at
75 ~ from the vertical. Clearly this is incorrect if
one is to take the results of the DSA to be
reliable (Table 3); the real cause of the anisotropy is unclear, although it may be caused by
strong crystal alignment. The evidence of results
for SSA-2 and SSA-5 have shown that bedding
planes do not, in these cases, strongly influence
velocity anisotropy for the horizontal direction,
and hence do not confuse the subsequent
interpretation of USWS measurements.
The results obtained from the S-wave splitting
tests do, however, tend to support the existence
of aligned horizontal microcracks. In the vertical
direction, results of the acoustic test have shown
that it is most likely that vertical aligned
microcracks cause S-wave velocity splitting,
since no other apparent causes are observed.
The question arises whether the suggested
presence of two sets of mutually perpendicular
vertically aligned microcracks, as implied by
DSA, can be justified. For this, the results of Swave splitting cannot be used, since they do not
show the existence of the second (i.e. the less
dense) vertical set of microcracks (if it does exist,
its existence is probably 'overlooked' by the
transmitted acoustic energy and treated as
merely a background for the first and more
dense vertical microcrack set). Charles et al.
(1986) outline theories of brittle fracture mechanics which explain crack opening under
tensile forces (as might occur in the case of
stress relaxation). If the existence of two sets of
microcracks (one vertical and the other horizontal) whose generation is related to principal
in situ stress relaxation can be proven experimentally, it is likely that a third set of microcracks (i.e. the second and less dense set of
vertical microcracks) also exists, since the
processes leading to it are essentially the same
as those causing microcracks in the other two
principal directions. Comparing results for the
two techniques has demonstrated directly that
three sets of mutually perpendicular microcracks
exist in a rock material experiencing a process of
relaxation from three in situ principal stresses,
provided the relaxation forces are sufficient to
generate them.

Conclusions
A series of investigations of in situ state of stress
using DSA and USWS has been conducted

successfully. In general, individual results of the


two techniques are found to be in agreement,
taking into consideration the facts regarding the
local in situ state of stress field. The DSA and
USWS results are also found to be in reasonable
agreement with results from hydraulic fracturing
and sleeve fracturing. Grain scale heterogeneities
strongly influence the deviation of stress relief
microcracks, as reflected by the scatter in
horizontal in situ stress orientations shown by
both techniques. Results for SSA sandstone
samples have shown that, in areas with small
differences in principal stresses, a greater number of samples will be required to overcome the
influence of grain scale heterogeneities.
Comparisons between individual results obtained from the two techniques have shown that
S-wave splitting is a reliable one for determining
orientations of in situ stresses by virtue of the
existence of stress relief microcracks. The study
has also shown that S-wave splitting analysis can
be used independently with reliable results. This
confidence, together with simplicity in sample
preparation and speed in conducting tests, can
be considered as the major advantage of this
technique compared to other techniques based
on stress relief microcracks such as DSA,
differential wave velocity analysis, anelastic
strain recovery or differential thermal analysis.
Despite the indicated advantages, the study
has also revealed some disadvantages in the
acoustic technique. Its major limitation is its
inability to provide estimates of in situ stress
magnitude. Another less important disadvantage is the necessity to perform the test only in
vertical and horizontal directions without compromising the simplicity in sample preparation
and analysis of results. This requires that the
principal in situ stresses always lie in vertical and
horizontal planes, which is not necessarily true
in all cases. Theoretically, however, this disadvantage can be reduced by reducing the
sampling rotational interval, hence enabling
more careful examination of the alignments of
sets of microcracks. The fact that stress relief
microcracks are not the only source of S-wave
splitting is another disadvantage. However, wellprepared samples can minimize this significantly.
When the two techniques are employed
together, DSA provides information on in situ
stress orientations and magnitudes, whereas Swave splitting provides confirmation of the
orientations bearing in mind that the USWS
'sees' a larger volume of the sample than DSA.
Combined utilization of the two techniques has
also shown the potential for examining rock
microfeatures, such as microcrack systems present in the rocks.

IN SITU STRESS PREDICTION


We would like to thank British Gas and Chevron for
their support in supplying core samples, and the
British Geological Survey and PPPTMGB 'Lemigas'
Indonesia for partially funding the research. We also
wish to thank J. W. Dennis for his support during the
project.

References
BROWN, E. T. & HOCK, E. 1978. Trends in relationships
between measured in-situ stress and depth. International Journal of Rock Mechanics and Mining
Sciences and Geomechanics Abstracts, 15, 211215.
CHARLES, Ph., HAMAMDJIAN,C. DESPAX, D. 1986. Is
the microcracking of a rock a memory of its initial
state of stress? Proceedings International Symposium on Rock Stress and Rock Stress Measurements, Stockholm, 341-349.
CRAMPIN, S. 1978. Seismic wave propagation through a
cracked solid: polarisation as a possible dilatancy
diagnostic. Geophysical Journal of the Royal
Astronomical Society, 53, 467-496.
- & LOVELL,J. H. 1991. A decade of shear-wave
splitting in the earth's crust: what does it mean?
what use can we make of it? and what should we
do next? Geophysical Journal International, 107,
387-407.
DESROCHES,J., MARSDEN,J. R. & COLLEY, N. M. 1995.
Wireline open-hole stress tests in a tight gas
sandstone. Proceedings International Gas Conference, Cannes.
DEY, T. N. & BROWN, D. W. 1986. Stress measurements in a deep granite rock mass using hydraulic
fracturing and differential strain curve analysis.
Proceedings International Symposium on Rock
Stress and Rock Stress Measurements, Stockholm, 351-357.

195

DYKE, C. G. 1988. In-situ stress indicators for rock at


great depth. PhD Thesis, Imperial College of
Science and Technology, University of London.
GARBIN, H. D. & KNOPOFF, L. 1975. The shear
modulus of a material permeated by a random
distribution of free circular cracks. Quarterly
Applied Mathematics, 33, 296-300.
KING, M. S., CHAUDHRY, N. A. & SHAKEEL, A. 1995.
Experimental ultrasonic velocities and permeability for sandstones with aligned cracks. International Journal of Rock Mechanics and Mining
Sciences and Geomechanics Abstracts, 32, 155163.
KOWALLIS, B. J. & WANG, H. F. 1983. Microcrack
study of granite cores from Illinois deep borehole
UPH3. Journal of Geophysical Research, 88, 73737380.
OIKAWA, Y., MATSUNAGA, I. & YAMAGUCHI, T. 1993.
Differential strain curve analysis to estimate the
stress state of the Hijiori hot dry rock field.
International Journal of Rock Mechanics and
Mining Sciences and Geomechanics Abstracts, 30,
1023-1026.
STRICKLAND, F. G. & REN, N. K. 1980. Use of
differential strain curve analysis in predicting insitu stress state for deep wells. Proceedings 21st
US Symposium on Rock Mechanics, Rolla,
Missouri, 523-533.
TEUFEL, L. W. 1983. Determination of the principal
horizontal in-situ stress directions from anelastic
strain recovery measurements of oriented core
from deep wells: application to the Cotton Valley
formation of East Texas. In." NEMAT-NASSER, S.
(ed.) Geomechanics, American Society of Mechanical Engineers, New York, 55-63.
YALE, D. P. & SPRUNT, E. S. 1989. Prediction of
fracture direction using shear acoustic anisotropy.
The Log Analyst, 30, 65-70.

Dolomite cement distribution in a sandstone from core and wireline


data: the Triassic fluvial Chaunoy Formation, Paris Basin
R. H. W O R D E N

School of Geosciences, The Queen's University, Belfast, BT7 INN, UK

Abstract:The distribution of mineral cements in oil fields is critical to the spatial variation of
porosity and permeability. The distribution of dolomite cement within fluvial Triassic
Chaunoy sandstones in the Paris Basin was studied using core description, petrography,
core analysis (porosity and permeability), wireline data interpreted to give mineralogy,
porosity and permeability data and geochemical data. Petrographic analysis revealed that
dolomite and quartz cements are the main diagenetic minerals. Using sonic transit time,
density and neutron porosity log, the overall proportions of quartz, dolomite and shale as
well as porosity for each depth interval could be resolved. Petrographic and core analysis
data showed that permeability could be calculated from wireline-derived porosity and
mineralogy data. There is excellent correlation between core analysis porosity and
permeability and their wireline-derived equivalents. There is also excellent correlation
between wireline-derived mineralogy data and quantitative petrographic mineralogy data.
The wireline-derived mineralogy data show that dolomite is preferentially concentrated at
the tops of most sand bodies. Porosity and permeability are consequently lowest at the tops
of individual sand bodies due to the localized dolomite cement. There are a number of
potential causes for this distribution pattern although geochemical and petrographic data
showed that a combination of early pedogenetic dolomite cementation and later
recrystallization, possibly due to an influx of organically-derived CO2, are most likely.

Knowledge of the way in which porosity and


permeability are distributed throughout an oil
field is an important building block in a reservoir
model. The key factors which control porosity
and permeability in sandstones are depositional
characteristics such as grain size and sorting and
diagenetic features such as cements and secondary porosity. Most reservoir simulation models
incorporate sub-units of common primary sedimentary origin. The distribution of reservoir
quality is thus usually defined in terms of the
morphology of the sedimentary architecture.
However, reservoir rocks seldom retain their
depositional porosity. Instead, porosity is usually degraded by a variety of diagenetic processes.
The effects of these processes are not necessarily
confined to the boundaries of depositional
sedimentary units. Common diagenetic processes may commonly either transcend sedimentary architecture or may lead to the subdivision
of self-contained sedimentary units in terms of
porosity and permeability. There is no framework for predicting diagenetic cement distribution in sandstones on the reservoir scale. It is not
yet generally possible to predict or model
reservoir porosity and permeability variations
over the distribution of the primary sedimentary
units due to the impact of diagenesis. This is
clearly unsatisfactory and may lead to systematically incorrect reservoir models.

One of the key problems involved in describing the distribution of diagenetic cement is the
cost (in terms of time and money) of acquiring
the data. Petrographic data are usually collected
at a far lower density than core analysis data (if
at all), are harder to quality-control and are
highly operator-dependent. In this paper, the
method of assessing carbonate cement distribution in sandstones using petrophysical logs
(hereafter known as wireline logs) will be
described. This method was used to describe
the distribution of dolomite cement in Triassic
fluvial clastic sediments of the Chaunoy Formation in the Paris Basin, France. The controls on
dolomite cement distribution will be discussed,
the effects of dolomite cement (and by inference,
quartz cement) on reservoir flow properties,
defined, and possible mechanisms that controlled the carbonate cement distribution pattern
investigated.

Geological setting
The Paris Basin is an intracratonic basin with an
aerial extent of approximately 6000km 2 and
about 3000m of present day sediment infill
deposited on Hercynian basement (Fig. 1;
Pommerol 1974, 1978). There are two main
permeable, petroleum-bearing reservoir units in
the central part of the Mesozoic of the Paris

WORDEN,R. H. 1998. Dolomite cement distribution in a sandstone from core and wireline data:
the Triassic fluvial Chaunoy Formation, Paris Basin. In. HARVEY,P. K. & LOVELL,M. A. (eds)

Core-Log Integration, Geological Society, London, Special Publications, 136, 197-2t 1

197

198

R.H. WORDEN

Fig. 1. Geological map of the Paris Basin with the


approximate extent of the Triassic sandstones. The
well under investigation (L) is marked.
Basin; the Late Triassic (Keuper) fluvial sandstones and the Middle Jurassic marine carbonates (Pages 1987). The Paris Basin experienced
a simple subsidence history that included periods
of relatively rapid burial. Rifting started in the
Permo-Triassic followed by thermal subsidence
in the Jurassic and Cretaceous (Pommerol 1978;
M6gnien 1980a,b; Brunet & Le Pichon 1982;
Loup & Wildi 1994). Maximum burial in the
central part of the basin occurred during the
Oligocene-Miocene and was followed by minor
uplift during and following Alpine and Pyrenean
tectonism (M6gnien 1980a,b; Brunet & Le
Pichon 1982; Pages 1987). Triassic sediments in
the central part of the basin reached maximum
burial depths of about 3000-4000 m. The Chaunoy Formation in the well examined is presently
buried to between 2200m and 2500m. Sandwiched between the Triassic sandstones and the
Mid Jurassic carbonates are organic-rich Liassic
shales. They are mature to the point of oil
generation and expulsion at the base of the Lias,
in the centre of the basin (Herron & Le Tendre
1990). This source rock reached maturity at the
time of maximum burial and charged both
Triassic and Mid Jurassic reservoirs with oil
(Poulet & Espitalie 1987).
The Triassic sandstones are composed of
several reservoir units. The Late Carnian to
Norian Chaunoy Formation has limited aerial
extent, lies in the deepest part of the basin,
slightly to the West of the basin centre and has
no outcrop (Fig. 1; Bourquin & Guillocheau
1993; Bourquin et al. 1993; Fontes & Matray
1993. Matray et al. 1993). The Chaunoy was
deposited as a minor transgressive-regressive
cycle within an overall transgressive phase that
ended with Rhaetic marine sediments (Bourquin
& Guillocheau 1993). The Chaunoy Formation

is composed of alluvial fan conglomerates,


coarse-grained channel-fill fluvial sandstones
and flood-basin siltstones. It was deposited in
an arid environment as an alluvial and fluvial
fringe to the western rifted margin of the basin
(Bourquin & Guillocheau 1993; Bourquin et al.
1993). Locally important pedogenic and phreatic
dolomite cements are found within the Chaunoy
Formation (Sp6tl & Wright 1992). Burial
diagenesis resulted in precipitation of abundant
quartz and dolomite, and less common calcite
and saddle dolomite cement (Demars & Pagel
1994; Worden & Matray 1995).
Previous diagenetic studies of the Chaunoy
Formation showed that quartz cement grew at
temperatures a little lower than those attained at
maximum burial, whilst sparry, rhombic ferroan
dolomite cement grew at maximum burial
(Demars & Pagel 1994). The pedogenic dolocrete has a limited range of 613C values ( - 7 to
0%o; Sp6tl & Wright 1992).

Methods
Core description and petrography
Slabbed core from the well was examined for
general lithology, facies variations, sedimentary
structures and grain size. Grain size of the core
was measured at regular intervals by comparing
core to standard grain size charts under a
binocular microscope. Petrographic analysis
was performed on 22 thin sections stained for
carbonates and feldspars and impregnated with
blue-dyed epoxy resin. Grain size and sorting
class were assessed quantitatively in thin section
by measuring sizes of one hundred grains per
section. Detrital grains, cements and porosity
were quantified by point counting using three
hundred grain counts per section.

Petrophysical (wireline and core analysis)


data
Porosity and permeability core analysis data for
the sampled well were made available to the
authors by Elf (99 data points from the interval
under investigation). Core porosity data have an
uncertainty of somewhat less than 0.5% that
arises due to the variable amount of stress
relaxation following withdrawal of the core from
the subsurface. Analytical errors are insignificant. Sonic transit time, neutron porosity,
density and other wireline data, recorded at
5 cm intervals by petrophysical logging methods,
were also made available by Elf. These data have
been used to derive porosity and mineral

DOLOMITE AND CEMENT DISTRIBUTION IN A SANDSTONE

199

Table 1. Definition of terms and units used in equations 1 to 7


Term Definition
At
Atminx

At e
P
Pminx

p4

~bn
(~nminx

4~n,
minX
pi
ps
qtz%

sonic transit time recorded by log (#see fl-1)


sonic transit time of mineral X (#sec fl-l)
sonic transit time of fluid in pore space (#sec fl 1)
density recorded by log (g cm-3)
density of mineral X (gcm -3)
density of fluid in pore space (g cm-3)
neutron porosity recorded by log (porosity units)
neutron porosity of mineral X (porosity units)
neutron porosity of fluid in pore space (porosity units)
proportion of mineral X (as fraction of total rock volume)
porosity (as fraction of total rock volume)
permeability intercept of a regression line on a porosity--permability cross plot
slope of a regression line on a porosity-permability cross plot
percentage of quartz in a depth interval derived from wireline logs

proportions using methods outlined by Hearst &


Nelson (1985) and Doveton (1994).
The gamma log is commonly used to define
the 'shaliness' (where shale in this context
routinely describes the overall clay mineral
content of the rock and is not a reference to
the grain size or texture of the sedimentary rock)
of sandstones although this approach is invalid
for simple crystal chemical reasons. Composite
gamma logs record the total potassium, thorium
and uranium contents of the rock by detecting
the -,/-rays associated with decay of the radioactive isotopes of these elements and their
daughter products. Spectral gamma logs differentiate the v-radiation from the three elements.
However, using any of the gamma logs for a
shale or total clay mineral estimate is invalid.
Potassium is commonly held in the minerals: Kfeldspar, illite and the mica group of minerals.
Most clay minerals apart from illite do not
contain potassium. Thus the potassium gamma
signal records the relative abundance of Kfeldspar, illite and mica indiscriminately and
does not record the shale or clay mineral
content. The thorium gamma signal, often
mistakenly thought to reflect specific clay
minerals, records the abundance of thoriumbearing trace minerals and cannot be used to
estimate volumes of clay minerals (Hurst &
Milodowski 1996). Consequently, a multiple logtransformation approach was used to derive the
shale content (as well as the dolomite and quartz
contents) and gamma ray logs have not been
used to derive the shale content.
The signals from the sonic transit time (At),
neutron porosity (On) and density logs (p) can
be integrated and resolved for three mineral
types and total porosity using three algorithms
relating each separate log signal at any given

depth to solid grain volume (occupied by the


three minerals) and the assumption that the sum
of the three minerals fractions plus porosity
equals unity. This also assumes linear relationships between mineral proportions and their
contribution to the petrophysical signal. Thus,
with four equations and four unknowns (proportions of three minerals plus porosity), the
following algorithms can be solved simultaneously at each depth interval:
At = mini. Atminl q- min2.Atmin2 +
min3.Atmin3 + &to

(1)

p = minl .Pminl + min2.Pmin2 + min3.Pmin3 + CrO(2)

On = min 1. Onmi n 1 ~- min2. Onmin2 -4min3.Onmin3 + Ono

(3)

1 + min 1 + min2 + min3 + O.

(4)

The terms used in the equations above are


defined in Table 1. Petrophysical responses of
each mineral were taken from Rider (1986).
Geochemical data

Fluid inclusion thermometry was performed


using a Linkam THM600 heating-cooling stage
with 0.1~ precision. Phase transition temperatures were determined by temperature cycling;
heating experiments were conducted before
freezing to prevent inclusion deformation by
ice growth that would effect homogenization
temperatures. Fluid inclusion homogenization
temperatures were collected from quartz and

200

R.H. WORDEN

Results
Core description and p e t r o g r a p h y

Grain size data are displayed in Fig. 2. Most of


the core is either fine-grained (silt/mud, grain
size < 62 #m) or coarse-grained (coarse sand to
conglomerate, i.e. grain size > 1000 #m).
Fine-grained core is mottled in appearance,
very well lithified and shows abundant evidence
of pedogenesis with rootlet structures, rhizoeretions and nodules (see, for example, Sp6tl &
Wright 1992). Petrographic analysis showed that
fine grained units are highly dolomitic with a
substantial clay mineral component. The dolomite is finely crystalline non-ferroan dolomite.
The coarse grained sandbodies are composed
of massive, largely structureless sediments.
Petrographic analysis showed that the sandstones are sub-lithic to sub-arkosic (according to
Folk 1974) with a significant volume of polycrystalline quartz grains ( 1 0 ~ 0 % of quartzose
grains). The feldspar population is approximately equally split between plagioclase and Kfeldspar. An average sandstone composition is
given in Table 2.
Table 2. Average petrographic data from the Chaunoy
Formation sand bodies. Twenty two samples were
examined petrographically. The figures illustrate the
importance of dolomite in the Chaunoy Formation
Grain/cement type
Fig. 2. Core description and petrographic data. Grain
size is shown as a continuous log. The petrographic
data are represented by bars at the appropriate depths
with mineralogy represented (see key). Core analysis
data are also displayed on this diagram. There are 99
porosity and permeability data points.

ferroan (rhombic) dolomite cements. Fluid


inclusions could not be examined in the microcrystalline dolomite because of the limited
resolution of the microscope. Six core samples
from this well were examined by this technique
giving more than one hundred data points.
For carbon isotope analysis, N d - Y A G laser
sampling was used following Smalley et al.
(1992). This has a spatial resolution of about
50#m (ablation pit diameter) with analytical
precision of +0.1%o for ~513C. Samples for laser
ablation were plasma-ashed to remove any
organic material (oil) in the pore system. The
laser could be used to sample individual
dolomite crystals within pores, without fear of
contamination from any other crystals. Eight
core samples were examined by this technique
giving more than fifty data points.

Polycrystalline quartz
Monocrystalline quartz
(Total detrital quartz
K-feldspar
Plagioclase
Quartzose lithic fragments
Detrital mica
Detrital clay
Kaolinite
Illite
Chlorite
Authigenic K-feldspar
Authigenic Quartz
Calcite cement
Dolomite cement

Mean
%
21.1
13.0
34.1
8.9
4.6
13.2
0.8
3.2
2.5
0.4
0.5
1.1
11.6
1.4
16.4

Standard
deviation %
8.6
8.2
6.6)
4.1
3.5
8.8
1.1
5.4
4.2
1.1
0.9
1.4
8.5
5.9
22.0

Sandbodies contain two distinct dolomite


morphologies. Sandbodies contain rhombic,
pore-filling, ferroan dolomite crystals that are
generally greater than 200#m in size (Fig. 3b).
The rhombic ferroan dolomite is texturally and
mineralogically chemically distinct from the
dolocrete. The dolomite in the top portions of
most sand bodies grades into the overlying silty

DOLOMITE AND CEMENT DISTRIBUTION IN A SANDSTONE

201

Fig. 3. Photomicrographs of (a) microcrystalline non-ferroan dolomite at the very top of a sand body with partial
replacement of detrital silicate grains and (b) grain-rimming quartz cement (Q) and pore-filling rhombic ferroan
dolomite (DOL) enclosing the quartz cement. Remnant porosity (~) is minor and occupies pore centres. Scale
bars are 200 #m.

dolocrete layers and tends to be extremely finely


crystalline. The proportion of microcrystalline
dolomite increases upwards to the top of sand
bodies. A 'floating grain texture' is present at the
tops of sand bodies due to mass silicate graindissolution and replacement by microcrystalline
dolomite (Fig. 3a). The sandstone also contains
localized quartz cement (e.g. minor quartz
cement labelled in Fig. 3b). Textural considerations show that the microcrystalline dolomite
pre-dated the ferroan rhombic dolomite.
To facilitate the subsequent comparison between petrographic data and wireline-derived

mineralogical data, the petrographic data have


been coverted into proportions of quartz,
dolomite and shale. In this manipulation, quartz
is the sum of detrital quartz grains, quartz
cement, quartzose lithic fragments and feldspar;
dolomite is the sum of all type of dolomite and
other carbonate minerals; shale is the sum of
clay, micas, and micaceous lithic fragments.
There is a broad correlation between grain size
and p e t r o g r a p h i c a l l y - d e f i n e d m i n e r a l o g y :
coarse- grained intervals are mostly quartz-rich,
the finer intervals are relatively shale and
dolomite rich (Fig. 2). However, the correlation

202

R.H. WORDEN

Fig. 4. Porosity-permeability data from sandstones.


Data have been subdivided by petrographicallydefined mineral proportions. High quartz content
samples are those with greater than 80% quartz;
medium quartz content samples have between 60%
and 80% quartz; low quartz content samples have less
than 60% quartz. Regression lines have been plotted
through the core data for the high, medium and low
quartz content samples. Note that the slope and the
intercept of these regression lines changes systematically as the quartz content changes.

between grain size, mineralogy and reservoir


properties is not perfect. Sand bodies can also
have high dolomite contents (e.g. 2457-2458 m
2472-0m, 2482.5-2483.5 m etc., on Fig. 2). This
pattern shows that dolomite content and grain
size together, probably control the reservoir
properties of the sandstone. The petrographic
data seem to show that dolomite is concentrated
in the top portions of the sandstone units (e.g.
2457m and 2472m) although insufficient samples were examined petrographically to prove
that this pattern was common and predictable.

Fig. 5. Wireline sonic transit time, density and neutron


porosity through the cored portion of the Chaunoy
Formation.

is considerable scatter in the data; there is a wide


range of permeability for a given porosity. This
probably signifies that there is more than one
control on the evolution of porosity and
permeability.

Core analysis data

Wireline log analysis

Core analysis data are displayed as continuous


logs in Fig. 2. Porosity 1.1-19.0%. Permeability
varies from < 0.1 mD to > 5000 mD. Porosity
and permeability are highest where the rocks are
most coarse grained. However, again the correlation is not perfect; the tops of the sandbodies
tend to have low porosity and permeability
values relative to the middle and lower portions
of sand bodies (Fig. 2). Consequently, grain size
and facies variations cannot be used in isolation
to understand or predict variations in reservoir
quality.
Core analysis data are also plotted on a
conventional log-linear diagram (Fig. 4). There

Wireline log analysis has been used to determine


porosity and mineralogy (with three components; quartz, dolomite and shale). These data
have then been used to compute permeability.
These data will be used subsequently to assess
dolomite cement distribution within the reservoir.
Sonic transit time, neutron porosity and
density log data for the cored interval are
presented as functions of depth in Fig. 5. The
same data are cross-plotted in Fig. 6 with the
positions of the three minerals added. Equations
1-4 can be solved for porosity plus three solid
grain components. The logs have been converted

DOLOMITE AND CEMENT DISTRIBUTION IN A SANDSTONE

203

Fig. 6. Cross plots of (a) sonic transit time against density and (b) neutron porosity against density. The positions
of the three mineral groups (quartz, dolomite and shale--where shale refers to all clay minerals and not a texture
or fabric) used to define the mineralogy of the formation are marked on both plots. The position of the pore fluid
is off the scale but the general direction is marked.

Fig. 7. Data quality assurance: (a) comparison of wireline-derived porosity and core analysis-derived porosity.
There is a good correlation between the two datasets. The intercept on the x-axis shows that the wireline porosity
data are over-estimating porosity by about 0.024 (note that this over-estimate is subsequently accounted for in all
the following calculations and plots); (b) comparison of petrographically-determined quartz and wireline- derived
quartz; (c) comparison of petrographically-defined dolomite and wireline-derived dolomite. Porosity and
mineralogy from core and petrographic sources is well matched by the values defined from the transformed
wireline logs.

204

R.H. WORDEN

Fig. 8. Combination diagram of grain size data (derived from core description, Fig. 2) and mineral proportions,
porosity and permeability (derived from wireline log analysis). There is excellent agreement between quartz
proportion and reservoir quality. The agreement of these with grain size is complex. The tops of some sand bodies
have a high dolomite content and correspondingly poorer reservoir quality (e.g. 2470-2471 m). Sand bodies are
numbered for reference to Figs 2 and 9. Core analysis porosity and permeability data (dashed and faint) have
been added to the diagram for comparison with the wireline-derived data.

into fractional porosity, and the fractional


quantities of quartz, dolomite and shale. The
rock was thus assumed to consist of three
minerals; 'quartz' (all silica minerals and feldspar), 'dolomite' (all carbonate minerals) and
'shale' (all clay minerals). Each individual group
of minerals 'quartz', 'dolomite' and 'shale') was
assumed to have effectively uniform responses to
the three wireline logging tools. Petrographic
analysis shows that the quartz/feldspar ratio is
generally much greater than about 4 (Table 2),
suggesting that the assumption about the quartz
component is reasonable. Feldspar and quartz
have similar wireline responses (at least for
sonic, density and neutron porosity logs) so that
the arkosic portion of the sandstone is probably
adequately accounted for. The lithic portion of
the sandstone is probably represented by 'shale'
together with quartz. Dolomite dominates the
carbonate mineral population within the rock.

Shale represents the sum of all clay minerals in


the rock, although preliminary quantitative Xray diffraction (XRD) data show that these are
dominated by kaolinite and illite. Water saturation was calculated using resistivity logs, the
Archie equation and neutron porosity values
(and the ultimate porosity values computed
from the simultaneous solution of the neutron
porosity, density and sonic transit time logs were
iteratively fed back into the Archie equation
until convergence was achieved). The average
wireline response properties of the mixed wateroil fluid were calculated depending on the
specific saturation (Sw) at the depth interval.
Fluid type was found to have little effect upon
the subsequent calculations.
The wireline-derived porosity data compare
favourably to core analysis porosity data with a
correlation coefficient of 0.74 (Fig. 7). The
wireline porosity values slightly over-estimate

DOLOMITE AND CEMENT DISTRIBUTION IN A SANDSTONE


the porosity (if the core porosity data are
correct). Consequently, wireline-derived porosity data have been corrected for this slight overestimate by subtracting 0.024 from the fractional
wireline porosity values to take the intercept
through the origin (Fig. 7). The wireline-derived
mineralogical data also compare favourably to
the quantitative petrographic data, the two
having very good correlation coefficients (average correlation coefficient of 0.84; Fig. 7). Thus,
despite the paucity of petrographic data, it is
possible to derive continuous and credible
mineralogical data from wireline data. Porosity
and mineral proportion data were smoothed by
averaging over a 0.3 m interval to reflect the
realistic resolution of the logging tools (Fig. 8;
Hearst & Nelson 1985; Doveton 1994).
The results of the wireline data transform into
mineral proportions and porosity are given in
Figs 8 and 9. There are distinct intervals that are
enriched in dolomite and others enriched in
quartz. The shale fraction tends to be highest in
the dolomite zones. However, the tops of sand
bodies have high dolomite contents in the
absence of shale (e.g. 2470-2472m) without
any corresponding change in grain size. This
leads to asymmetry in the mineralogy of
individual sandstone beds. The summary diagram, Fig. 9, shows that, on average, sand
bodies have the most dolomite in the top
quarter.
The derivation of permeability from porosity
is not a simple task. Permeability is, of course,
affected by porosity, but it is also controlled by
the shape and size of pore throats that connect
pores. The degree of connectivity of the total
porosity and the dimensions of pore throats are
critical to permeability. It is not possible to
derive permeability from a simple porosity value
with any degree of accuracy using a simple
transform in these rocks. However, recent network modelling work by Bryant et al. (1993) and
Cade et al. (1994) has shown that permeability
may be predicted from porosity if the fundamental controls on porosity evolution are
known. The main generic controls on porosity
loss are compaction and cementation. Cementation may be subdivided further between grainrimming cements and pore-filling cements. The
different cement morphologies have different
effects upon permeability for unit porosity loss
due to their different effects upon the pore
network.
Chaunoy sandstones of the same depositional
facies are cemented by both quartz and dolomite
(Fig. 3b). Quartz cement forms approximately
equal thickness overgrowths whilst dolomite
cements tend to fill pores (Fig. 3b; Cade et al.

205

Fig. 9. Summary diagram illustrating the non-uniform


distribution of dolomite and porosity in the Chaunoy
Formation sand bodies. The numbers refer to the sand
bodies on Figs 2 and 8. (a) Dolomite is preferentially
concentrated in the top quarter of each sand body. (b)
Conversely, porosity is concentrated in the middle two
quarters of each sand body.
1994). These two different cement morphologies
have profoundly different effects upon the pore
network.
Core analysis data from the Chaunoy Formation were subdivided on the basis of the quartzdolomite ratios using the wireline-derived mineralogy data. Regression analysis (Fig. 4) shows
that the quartz-rich (and thus presumably
quartz-cemented) samples have shallower porosity-permeability slopes (ps) and higher permeability intercepts (pi) than quartz-poor (and thus
presumably dolomite-cemented) samples in accord with the network modelling discussed
above. Algorithms were derived for describing
the change in both slope and intercept of the
porosity-permeability curves as a function of
total quartz content:
(intercept) pi = 2.777x 104X 10 (4"55xl~

(5)

(slope) ps = 30.75x 10 (-3"077x ~~ xqtz%)

(6)

in which 'qtz%' is the quartz fraction of the rock


as defined by wireline analysis. It was thus
possible to predict permeability as a function of
the wireline-derived porosity and mineralogy
using the following algorithm:
permeability (mD) = pi x 10(ps

(7)

The results of these calculations are displayed in


Fig. 8. Inspection of Figs 2 and 8 shows that the
wireline-derived permeability curve corresponds
well to the core analysis data.
Geochemical data

Fluid inclusion data are reported in Fig. 10.


Rhombic ferroan dolomite grew at a range of

206

R.H. WORDEN

Fig. 10. Fluid inclusion homogenization data from


quartz and rhombic ferroan dolomite. Whilst there is
overlap between the temperatures of growth of
dolomite and quartz, these data support the late
(maximum burial) growth of rhombic ferroan dolomite cement following quartz cement growth in the
Chaunoy Formation.

temperatures with a mean of 119~ This is a


considerably higher temperature than the quartz
cement (mean 103~ and confirms the textural
evidence (Fig. 3b) that ferroan rhombic dolomite grew after quartz. The temperature for
dolomite growth corresponds to maximum
burial and the time of oil generation from the
Liassic source rocks.
Stable isotope data from ferroan rhombic
dolomite are given in Fig. 11. These data are
compared to carbon isotope data from pedogenic dolomite. The ferroan rhombic dolomite
has considerably lighter carbon isotopes than
the pedogenic dolomite. The ferroan rhombic
dolomite has a range of 813C values consistent
with input from recrystallzed pedogenic dolomite and input from a source depleted in 13C.

Fig. 11. Carbon isotope data from rhombic ferroan


dolomite with data from the pedogenic dolomite (after
Sp6tl & Wright 1992; Sp6tl et al. 1993). The rhombic
ferroan dolomite represents mixing between the
indigenous pedogenic carbonate and a source of
carbonate characterized by carbon relatively depleted
in 13C. This must indicate a major organic input into
the system (probably in the form of either aqueous
bicarbonate, or CO2 dissolved in oil or water).

used to define the spatial distribution of


dolomite cements in these sandstones.
The derivation of porosity, mineralogy and
permeability from wireline data has distinct
advantages over core analysis and petrographic
data. Most importantly, mineralogical data
from logs can be derived for uncored intervals.
Petrographic data are usually sparse (due to cost
and time) and are 'operator'-dependent. In
contrast, wireline data are typically available
for most wells in a reservoir and are (in
principle) operator-independent. Petrographic
data are rarely collected in such abundance that
cement distribution can be observed within
reservoir units, whereas such data can be derived
from wireline mineralogical analysis.

Discussion
Quantitative mineralogical data have been generated from sonic transit time, density and
neutron porosity wireline logs. Gamma logs
cannot be used for mineral identification due to
the variable site of radiogenic potassium in a
variety of minerals and the non-concordance
between uranium and thorium and specific
minerals (Doveton 1994; Hurst & Milodowski
1996). The Triassic sandstones and mudstones of
the Paris Basin have been resolved into quartz,
shale and dolomite. Dolomite has a diagenetic
(i.e. non-primary) origin. Wireline logs can be

Amount and distribution of dolomite cement


in the Chaunoy sandstone
Petrographic analysis hinted at the heterogeneous distribution of dolomite cement in the
Chaunoy formation sandstones (Fig. 2). Without a major sampling and petrographic analysis
programme, it would be difficult to analyse and
describe that heterogeneity. The interpreted
wireline data have confirmed that dolomite is
not homogeneously distributed throughout the
Chaunoy Formation sandstone (Figs 8 and 9).

DOLOMITE AND CEMENT DISTRIBUTION IN A SANDSTONE

207

Fig. 12. Theoretical dolomite distributions from four potential controlling processes. The model represents a sand
body sandwiched between pedogenic dolocrete layers. (a) Pedogenic dolomite cement; there would be most
dolomite at the top of each sand body. (b) Dolomite cemented, sourced from the dolocrete during burial,
transported by diffusion; cement should be equally abundant at the tops and bases of sand bodies with a
minimum at the centre. (c) Dolomite distribution controlled by high permeability streaks enabling input from
external sources; fluvial sandstones usually fine upwards leading to high permeability bases and thus most
dolomite at sand body bases. (d) Dolomite distribution controlled by the relative buoyancy of oil (which may
have carried dissolved CO2) or a separate CO2 gas phase caused dolomite cementation and thus leading to most
dolomite at the tops of sand bodies.

Dolomite in the Chaunoy has either a pedogenic


(i.e. very early diagenetic) or burial diagenetic
origin (Fig. 12). Detrital dolomite may be
discounted as an option because of the relatively
soluble nature of dolomite and the perfect
crystal forms observed in core and thin section
(Fig. 3). Wireline data cannot be used to
discriminate between different dolomite grain
morphologies (e.g. microcrystalline or coarse
rhombic) or between dolomites of different
mineral chemistry (e.g. non-ferroan or ferroan
dolomite). The wireline data have shown that
the tops of most coarse grained sandbodies have
the greatest quantity of dolomite (Figs 8 and 9).
The dolomite content varies between, as well as
within sandbodies. Sand bodies 3, 4, 5, 7, 8 and 9
all have significantly more dolomite in the top
quarter of the sand body than in other quarters.
However, sand body 1 has much more dolomite
than sand body 5. Dolomite can occupy more
than 50% of the solid portion of a sand body
(e.g. sand body 1). The partially replacive nature
of the dolomite within sand bodies has already
been established from petrographic data (Fig.
3a) so that the high dolomite contents probably

reflect partial replacement of detrital silicate


mineral grains as well as precipitation of
dolomite into pre-existing pore spaces.

Effect of dolomite cement upon the reservoir


properties of the Chaunoy Formation sandstone
The present day porosity of the quartz-rich
samples is significantly less than the porosity
that would result from compaction alone (with
no cementation) of quartzose sandstones at
these depth of burial. Quartzose sandstones
should have approximately 25-30% porosity
after compaction during maximum burial to
about 3000m (see, for example, North 1985).
The actual porosities, even in the quartz-rich
intervals, are only as high as about 19%
indicating that some of the quartz in the rock
must be quartz cement. The main control on
porosity and thus permeability in the quartz-rich
portions of the rock must be the extent of quartz
cementation.
Quartz-rich portions of sandbodies have

208

R.H. WORDEN
Thus, not only does the dolomite cement
preferentially obscure porosity at the tops of the
sandstone units, but it also leads to disproportionately lower permeabilities than quartz-cemented sandstones of similar porosity.

Origin of dolomite cement in the Chatmoy


Formation sandstone

Fig. 13. Sketch of the morphologies of quartz cement


and dolomite cement in a porous sandstone matrix.
Quartz cement typically forms overgrowths and tend
to leave pores relatively unobstructed and thus has a
minimal impact upon permeability (see also Fig. 4).
Dolomite tends to form rhombic euhedral crystals that
sit in pores, blocking porosity to flowing fluids and
thus reducing permeability to a greater degree than an
equivalent volume of quartz cement.
better permeabilities for a given porosity than
their quartz-poor equivalents (Figs 2 and 4). For
example, in dolomite-rich, quartz-poor samples
with 10% porosity, permeability is typically
about 1-2mD. In dolomite-poor, quartz-rich
samples with 10% porosity, permeability is
typically, about 10-100mD. This is reflected by
the slightly steeper permeability-porosity gradient and higher permeability intercept of the
regression line through the dolomite-rich,
quartz-poor samples than the regression line
through the dolomite-poor, quartz-rich samples
on Fig. 4.
This pattern confirms that the main mechanism of porosity-loss in the quartz-rich samples
(quartz cementation) is less detrimental to
permeability than porosity-loss in the quartzpoor samples (dolomite cementation) as suggested by Cade et al. (1994). Dolomite cement in
the Chaunoy Formation sand bodies tends to fill
pores and block pore throats thus degrading
permeability at a greater rate than quartz cement
that forms equal-thickness rims to grains (Fig.
13).

The dolomite-rich fine-grained (mud and silt)


beds in the Chaunoy Formation resulted from
dolocrete pedogenesis (Sp6tl & Wright 1992) in
inter-channel facies. The similarity between the
(very fine) crystal size and texture of the
dolomite in the fine beds and the dolomite at
the very tops of the sand bodies (Fig. 3a)
suggests that some of the dolomite in the sand
bodies may be related to the formation of the
dolocrete during pedogenesis The replacive
nature of the finely crystalline dolomite in the
sand bodies, as indicated by the corrosion of
detrital quartz and feldspar grains (Fig. 3a),
supports the development of this dolomite by
pore waters that simultaneously dissolved silicate minerals and precipitated carbonate minerals during pedogenesis.
However, much of the dolomite within the
sand bodies does not have the same morphology
and chemistry as the dolocrete material: much
occurs as coarsely crystalline, ferroan dolomite
rhombs (Fig. 3). From the textural and mineral
chemical evidence, this must have a different
genesis than the dolocrete. Textural evidence
shows that rhombic ferroan dolomite post-dates
quartz cement overgrowths (Fig. 3b). Aqueous
fluid inclusion temperatures in Fig. 10, and also
reported by Sp6tl et al (1993) and Demars &
Pagel (1994), from rhombic ferroan dolomite are
somewhat higher than present day temperatures,
suggesting that dolomite grew at close to
maximum burial and temperature conditions in
the Oligocene/Miocene (Loup & Wildi 1994).
Maximum burial of the sedimentary pile during
the Oligocene/Miocene was also the time of
petroleum generation and migration from the
Liassic source rocks into the Chaunoy Formation. Rhombic ferroan dolomite cement therefore grew in the Chaunoy Formation at
approximately the same time as oil was being
generated from the overlying Liassic source
rocks (Poulet & Espitalie 1987).
Carbon stable isotope data for the burial
diagenetic dolomite cements (Fig. 11; Worden &
Matray 1995) indicate that carbon depleted in
13C has been added to the dolomite relative to
the pedogenic dolocrete (Sp6tl & Wright 1992).
Carbon depleted in 13C is thought typically to

DOLOMITE AND CEMENT DISTRIBUTION IN A SANDSTONE


have an organic origin (e.g. Longstaffe 1989).
The most obvious source of organically-derived
bicarbonate or CO2 in the Paris Basin is the
Liassic shale source rock. pH-buffered rocks
undergo carbonate mineral precipitation when
the partial pressure of CO2 is increased (Lundegaard & Land 1989) suggesting that at least
some of the rhombic ferroan dolomite cement
may be the direct result of bicarbonate or CO2
influx increasing the partial pressure of CO2.
Liassic source rocks may have expelled CO2
during or before oil generation. The Triassic
sandstones in the Paris Basin are presently in
equilibrium with CO2 in the reservoir; that is
partitioned between the two liquid phases: oil
and water (Matray et al. 1993). The equilibrium
partitioning of CO2 between formation water
and oil suggests that CO2 may have been
brought into the reservoir by the oil in solution.
Subsequent partitioning of CO2 into the formation water may then have caused dolomite
supersaturation and precipitation. Alternatively,
CO2 may have migrated into the Chaunoy sand
bodies as a separate gas phase resulting from
thermal decarboxylation of organic matter prior
to the onset of oil generation. Whatever the
mechanism, isotopic data dictate that an increase in the partial pressure of CO2 (from an
organic source) was most likely responsible for
the precipitation of dolomite cement in the
sandbodies during diagenesis at close to maximum burial.

Origin o f the dolomite cement distribution


pattern
Dolomite cement is generally concentrated at the
tops of sand bodies in the Chaunoy Formation
(Figs 7 and 9). There are several potential
generic controls on dolomite distribution patterns (Fig. 12):
(1) The cement at the tops of sand bodies may
be a direct result of pedogenesis that
occurred at the same time as the development of the pedogenic dolocrete in the fine
grained units. This would occur preferentially at the tops of sand bodies adjacent to
zones of active dolocrete pedogenesis. This
process is probably at least partly responsible for the dolomite cement distribution
in the sand bodies.
(2) In principle, dolomite distribution in sand
bodies may be due to diffusion from the
pedogenic dolocretes that encase the sand
bodies. In this case the dolomite would be
redistributed by diffusion from the dolo-

209

crete into the sand bodies. This would


influence the top and base of sand bodies
eqpally and result in a minimum dolomite
cement content at the centre of the sand
bodies. Note that this is not observed (Figs
8 and 9) and that rhombic ferroan dolomite
has a carbon isotope signature that is
different from the pedogenic dolomite
(Sp6tl & Wright 1992; Worden & Matray
1995).
(3) Cement distribution could be influenced by
reservoir quality at the time of cementation. High permeability streaks or gradational permeability may have focussed flow
and input of CO2 into specific portions of
the rock. Fluvial sandstones usually fine
upwards resulting in diminishing permeability towards sand body tops. This would
lead to the most extensive dolomite cementation occurring at the bases of sand
bodies. However, note that the Chaunoy
sandstones do not fine upwards (Fig. 2)
and do not have dolomite preferentially at
sand body bases.
(4) Isotope data (Fig. 11) suggest that some of
the carbon in the dolomite has an organic
source and might have come from the oil
source rock. Oil and CO2 may have
migrated into the rock at about the same
time (i.e. as CO2 dissolved in oil, Matray et
al. 1993). Alternatively, CO2 may have
migrated into the rock separately as a free
gas phase. Due to buoyancy, the tops of
each sand body should be the first part of
the sandstone to encounter either oil (laden
with CO2) or free CO2 gas. In summary,
the tops of each sandbody may have
received CO2 preferentially and thus
caused localized dolomite precipitation.
However, it is generally considered that
oil emplacement hinders diagenetic processes so that the opportunity for this
process to operate may be limited to a
window of opportunity between the onset
of oil emplacement and some elevated level
of oil saturation (e.g. Worden et al. 1998).
The absence of dolomite cement at sand body
bases and abundance at sand body tops, the
reported organic carbon isotope signal in the
rhombic ferroan dolomite and the mixture of
pedogenic dolomite textures and burial diagenetic textures in the sand bodies suggest that
options 1 and 4 together are probably responsible for the distribution of dolomite in the
Chaunoy sandbodies.
The key implication from this analysis of
dolomite cement distribution is that the dolo-

210

R.H. WORDEN

mite is heterogeneously distributed in the


reservoir. A reservoir model (e.g. for simulation
purposes) should take account of the fact that,
despite approximately uniform grain size, porosity and permeability are not uniformly distributed due to preferential cementation at the
tops of individual sandbodies. The combination
of petrography, geochemistry and petrophysics
was necessary to produce a model of cement
distribution in this case. There are no 'off the
shelf' models that can presently be fitted to the
distribution of cements in reservoir sandstones
in general. In time, generic models may be
available as more case studies of cement
distribution are performed, but until that time,
each reservoir should be analysed using a similar
combination of tools as that used in this study if
reservoir quality heterogeneity is an issue.

Conclusions
(1) Wireline petrophysical data have been
successfully manipulated to give mineralogy in terms of the amounts of quartz,
shale and dolomite, as well as porosity.
(2) Core analysis data show that dolomite
cement has a more detrimental effect upon
permeability than quartz cement. Permeability has thus been calculated from the
wireline porosity data using algorithms
that account for the variation in mineralogy as well as porosity.
(3) Petrography and, more importanly, wireline log data, have shown that dolomite
c e m e n t is not u n i f o r m l y d i s t r i b u t e d
t h r o u g h o u t the sandstones within the
C h a u n o y Formation. Rather, dolomite
cement is localized within the top portions
of individual sandstone units.
(4) Reservoir quality in the Chaunoy Formation is not just a function of depositional
facies but is also a function of localized
cement distribution. Building a reservoir
model using primary sand body architecture alone is insufficient to correctly describe reservoir quality.
(5) Sand bodies contain microcrystalline nonferroan and replacive dolomite as well as
rhombic ferroan and pore-filling dolomite.
Textural and mineral chemical data show
that the microcrystalline dolomite probably
grew during pedogenesis of the overlying
fine-grained facies. Fluid inclusion and
isotope data, together with textural evidence, show that the rhombic ferroan
dolomite probably grew at close to maximum burial in the mid Tertiary in the
presence of organically derived CO2.

(6) Dolomite cement may be localized at the


tops of the sand bodies because of the
proximity of overlying fine-grained units
when they were undergoing pedogenesis
and because the tops were the first part of
each sand body to receive a charge of CO2.
The CO2 influx may have occurred as a
separate buoyant gas phase or as a gas
dissolved in oil.
(7) Reservoir simulation of sandstones should
account for the fact that cement patterns
do not necessarily follow primary sedimentary architecture. Cements may typically be
heterogeneously distributed in individual
sand bodies and this may have important
consequences for how petroleum is produced to optimize flow rate and recovery.
I would like to thank Elf-Aquitaine (F. Walgenwitz
and G. Sambet especially) for kindly providing the
core analysis and wireline data. Part of the study was
the result of a collaborative research programme
including BP, BRGM, Elf-Aquitaine, the University
of Paris V1 and the European Community under
contract JOUF-0016c. D. C. Herrick and an anonymous reviewer are thanked for identifying key areas
of the manuscript for improvement.

References
BOURQUIN,S. 8z GUILLOCHEAU,F. 1993. Gbometrie des
sbquences de d6p6t du Keuper (Ladinien
Rh&ian) du Bassin de Paris: implications gbodynamiques. Comptes Rendus Acad~mie des
Sciences, Paris. 317, S6rie 2, 1341-1348.
, BOEHM, C., CLERMONTE, L, DURAND, M. &
SERRA, O. 1993. Analyse facio-sbquentielle du
Trias du centre-ouest du bassin de Paris fi partir
des donnbes diagraphiques. Bulletin de la Societe
G~ologique Francais, 164, 177-188.
BRUNET, M.-F. & LE PICHON,X. 1982. Subsidence of
the Paris Basin. Journal of Geophysical Research,
87, 8547-8560.
BRYANT, S., CADE, C. MELLOR, D. 1993. Permeability prediction from geological models. American Association of Petroleum Geologists Bulletin,
77, 1338-1350.
CADE, C., EVANS,I. J. 8s BRYANT,S. 1994. Analysis of
permeability controls: a new approach. Clay
Minerals, 29, 491-501.
DEMARS, C. & PAGEL, M. 1994. Palbotemp6ratures et
pal6osalinites dans les gr~s du Keuper du Bassin
de Paris: inclusions fluides dans les min6raux
authig~nes. Comptes Rendus Acad~mie des
Sciences, Paris. 319, serie 2, 427-434.
DOVETON, J. H. 1994. Geologic log analysis using
computer methods. AAPG computer applications
in geology, 2. American Association of Petroleum
Geologists, Tulsa, USA.
FOLK, R. L. 1974. Petrology of sedimentary rocks.
Hemphill, Austin.

DOLOMITE AND CEMENT DISTRIBUTION IN A SANDSTONE


FONTES, J. C. & MATRAY, J.-M. 1993. Geochemistry
and origin of formation brines from the Paris
Basin. Part 2, Saline solutions -A associated with
oil fields. Chemical Geology, 109, 177-200.
HEARST J. R. & NELSON, P. H. 1985. Well logging for
physical properties. McGraw-Hill, New York
HERRON, S. L. & LE TENDRE, L. 1990. Wireline sourcerock evaluation in the Paris Basin. In: Huc, A. Y.
(ed.) Deposition of organic facies. American
Association of Petroleum Geology Studies in
geology, 30, 57-71
HURST, A. & MILODOWSKI,A. 1996. Thorium distribution in some North sea sandstones: implications
for p e t r o p h y s i c a l e v a l u a t i o n . Petroleum
Geoscience, 2, 59-68.
LON~STAVFE, F. J. 1989. Stable isotopes as tracers in
clastic diagenesis: In: HUTCHEON J. (ed.) Miner-

alogical Association of Canada short course in


diagenesis. 201-277.
LouP, B. & WILDI, J. W. 1994. Subsidence analysis in
the Paris Basin: a key to Northwest European
intracontinental basins? Basin Research, 6, 159177.
LUNDEGARD, P. D. & LAND, L. S. 1989. Carbonate
equilibria and pH buffering--response to changes
in PCO2. Chemical Geology, 74, 277-287.
MATRAY, J.-M., FOUILLAC,C. & WORDEN, R. H. 1993.
Thermodynamic control on the chemical composition of fluids from the Keuper aquifer of the
Paris Basin In: PARNELL, J. RUEFEL, A. H. &
MOLES N. R. (eds) Extended abstracts from
Geofluids '93, 12-16.
MEGNIEN, C. 1980a. Tectogen+se du Bassin de Paris:
etapes de L'evolution du bassin. Bulletin de la
Societe Gdologique Francais 22, 66%680.
- - ,
1980b. Synthdse gdologique du bassin de Paris.
Stratigraphie et paleogeographie. M e m o i r e
BRGM 101, 466.

211

NORTH, F. K. 1985. Petroleum Geology. Allen and


Unwin, Boston
PACES L. 1987. Exploration of the Paris Basin. In:
BROOKS, J. & GLENNIE, K. (eds) Petroleum
Geology of North West Europe. Graham and
Trotman, UK, 87-93.
POMMEROL, C. 1974. Le bassin de Paris. In: DEBELMAS,
J. (ed.) Geologie de la France. Doin, Paris, 230258
, 1978. lEvolution pal6og6ographique et structurale du Bassin de Paris, du Pr6cambrian /t
l'actual, en relation avec les r6gions avoisinantes.
Geologie en Mijnbouw, 57, 533-543.
POULET, M. & ESPITALIE, J. 1987. Hydrocarbon
migration in the the Paris Basin In: DOLI6EZ, B.
(ed.) Migration of hydrocarbons in sedimentary
basins. Editions Technip, Paris, 131-171.
RmER, M. H. 1986. The geological interpretation of
well logs. Blackie. Glasgow.
SMALLEY, P. C., MAILE, C. N., COLEMAN, M. L. &
ROUSE, J. L. 1992. LASSIE (laser ablation
sampler for stable isotope extraction) applied to
carbonate minerals. Chemical Geology (Isotope
Geoscience Section), 101, 43-52.
SPOTL, C. & WRIGHT, V. P. 1992. Groundwater
dolocretes from the Late Triassic of the Paris
Basin, France: a case study of arid, contintental
diagenetic facies. Sedimentology, 39, 1119-1136.
- - ,
MATTER, A. & BREVART,O. 1993. Diagenesis
and pore water evolution in the Keuper reservoir,
Paris Basin (France). Journal of Sedimentary
Petrology, 63, 909-928.
WORDEN, R. H. & MATRAY, J.-M. 1995. Cross
formational flow in the Paris basin. Basin
Research, 7, 53-66.
--,
SMALLEY,P. C. & OXTOBY,N. H. (1988) Can
oil emplacement prevent quartz cementation in
sandstones? Petroleum Geoscience, 4.

Conjunctive interpretation of core and log data through association of


the effective and total porosity models
P. F. W O R T H I N G T O N

Gaffney, Cline & Associates, Bentley Hall, Blacknest, Alton, Hampshire GU34 4PU, UK
Abstract: Traditionally, the deterministic open-hole petrophysical evaluation of non-Archie
primary reservoirs has been undertaken exclusively within one or other of two intergranular
systems, those of effective and total porosity. Yet, these interpretative models can be
considered conjunctively with the object of inter-model validation of petrophysical
interpretation. These considerations reveal ways of demonstrating the numerical compatibility of the two approaches. The compatibility is expressed in terms of equalities that
contain core-calibrated, log-derived parameters and that are founded on the underlying
petrophysics. The equalities must be satisfied if the petrophysical procedures are to be
applied consistently and correctly. These inter-model algorithms constitute a basis for a
proposed quality assurance scheme in well-log interpretation that goes beyond tying log
data back to core. They suggest quality control points at which core-calibrated log data can
be examined to assess the meaningfulness and performance of interpretation procedures at
different stages of the petrophysical evaluation process. These assessments form a basis for
the development of measures of confidence in the practice of open-hole well-log
interpretation for porosity and hydrocarbon saturation, regardless of whether the
interpretation is ultimately set in the context of the effective or the total porosity model.
More generally, the subject matter forms part of a broader thrust to introduce a systematic
quality assurance culture into open-hole petrophysical interpretation.

Open-hole petrophysical evaluation of nonArchie rocks, those that do not satisfy the
conditions for the application of the laws of
Archie (1942), has traditionally drawn upon
either effective or total porosity concepts as a
basis for the determination of reservoir porosity
and fluid saturations. The difference between the
two concepts lies in the interpretative treatment
of the electrochemically-bound interstitial water.
This should not be confused with capillary
bound water, whose volume can be an order 9f
magnitude greater (Pallatt & Thornley 1990),
nor with those dual-porosity waters that are
distinguished by pore type. Petrophysicists have
usually operated the effective and total porosity
models discretely, selecting one or the other at
an early stage of the formation evaluation
process. This exclusive choice has been driven
by company culture, software considerations,
statutory requirements, or technical or personal
preference.
The practice of selecting a discrete interpretative model constrains the manner in which core
data can be used to support and validate log
interpretation. Thus, for example, effective
porosity cannot easily be determined in the
laboratory, and yet these data are strictly
required by the effective porosity model in order
to control the porosity interpretation of neutron~lensity log cross plots. In contrast, the
electrochemical shale parameters needed to

evaluate water saturation as per the total


porosity model can be determined in the
laboratory but they cannot be measured directly
downhole.
The constraints imposed by an exclusive
porosity model therefore limit the benefit that
can be derived from the ensuing integrated
analysis of laboratory and downhole data. The
key to improved core and log interpretation is to
operate the effective and total porosity models in
parallel, using the cross-linkages between them
to transpose the advantages of one in support of
the other, especially where core data can be used
to provide quality control on the log interpretation.
The purpose of this paper is to develop the
opportunities for improved quality assurance in
formation evaluation, in accordance with earlier
projections (Worthington 1991), by considering
how the effective and total porosity models can
be operated conjunctively to allow inter-model
validation of petrophysical interpretation. The
primary aim is to demonstrate the numerical
equivalence of the two models through equalities
that contain core-calibrated, log-derived parameters and that honour the principles of the
underlying physics. Thus, the objective is an
interpretative scheme that brings together traditionally separate areas of petrophysical systemics within a quality-controlled integrated
framework.

WORTHINGTON,P. F. 1998. Conjunctive interpretation of core and log data through association of
the effective and total porosity models. In: HARVEY,P. K. 8s LOVELL,M. A. (eds)
Core-Log Integration, Geological Society, London, Special Publications, 136, 213-223

213

214

P.F. WORTHINGTON

Fig. 1. Effective and total porosity models for a water-wet reservoir.

Philosophy of formation evaluation


Traditional formation evaluation practice is set
within either the effective or the total porosity
system, although one or two procedures that are
seen as transgressing this otherwise exclusive
divide have emerged over the past 15 years (e.g.
Juhasz 1981). The basic difference between the
effective and total porosity approaches can be
summarized with the aid of Fig. 1, which
describes the key parameters for a water-wet
reservoir. A nomenclature is provided at the end
of the text.
The effective porosity system regards those
waters that are electrochemically bound to (clay)
mineral surfaces as an integral part of the
minerals themselves. Therefore, the bound-water
porosity qbbw is incorporated within a wetted
clay-mineral fraction, which is loosely termed a
wetted shale volume fraction Vsh, and which is
allowed to have different physical properties
from those of the clean rock matrix. Only the
electrochemically-free fluids comprise the porosity, which is termed 'effective'. Some authors
have used the term 'effective clay mineral
volume' to be synonymous with Vsh (Hurst &
Nadeau 1994), but note with caution that others
have used the seemingly related term 'effective
clay' specifically to distinguish a clay mineral
that is electrochemically active from one that is
not (Johnson & Linke 1977).
In contrast, the total porosity system separates the electrochemically bound waters from
the clay-mineral fraction and groups these with
the free fluids that occupy the remainder of the
pore space. Therefore, d0bw is distinct from the
clay-mineral fraction, which is sometimes
loosely termed a dry clay fraction Vd. The dry
clay is required to have the same physical
properties as those of the clean rock matrix,
but it is allowed different electrochemical char-

acteristics, quantified through the cation exchange capacity per unit pore volume Qv. All
constituent fluids comprise the porosity, which is
termed 'total'. Despite this seemingly all-embracing name, the term 'total porosity' does not
include fracture porosity, and the sum of these is
better described as 'absolute porosity'.
The role of the effective and total porosity
systems within the overall scheme of formation
evaluation is governed in some quarters by the
polarization of the subject area into statistical
and deterministic methods of interpretation.
These approaches, too, need not be operated
exclusively, but their integrated use extends
beyond the scope of this paper and is the subject
of another that is in preparation.
Statistical methods of petrophysical interpretation are based on the global approach to the
solution of log response equations (Mayer &
Sibbitt 1980). If these equations contain characterizing shale properties and solve for a wetted
shale fraction, the computed porosity will be an
effective porosity. If the response equations do
not contain characterizing shale properties and
do not solve for a wetted shale fraction, the
implicit requirement for a total porosity approach must be one of identical log responses to
clean rock matrix and dry clay minerals for all
the log responses represented within the input
matrix over the evaluation interval. The highly
tenuous nature of this general assumption
inhibits the meaningful operation of the global
approach in the total porosity system.
On the other hand, deterministic methods can
be operated in both the effective and the total
porosity systems, because logs can be used more
selectively on the basis of their being fit-forpurpose. Therefore, the following discussion is
set within the context of deterministic petrophysics.
The scheme of deterministic open-hole petro-

EFFECTIVE AND TOTAL POROSITY

Effective porosity model

RESERVOIR

FORMATIONWATERSALINITY

CLAY-MINERALCONTENT 1

ARCHIE

POROSITY

Sw

"7

+
+

NON-ARCHIE

TOTAL
I EFFECTIVEpoROSI
] TY I POROSITY

EFFECTIVE
Sw

215

I TOTAL
Sw

Fig. 2. Scheme of deterministic open-hole petrophysics.


physics is illustrated in Fig. 2. Reservoirs with a
high formation-water salinity and a low claymineral content are usually termed Archie
reservoirs, wherein the effective and total porosities are essentially the same, because there are
negligible bound-water effects. Otherwise, they
are termed non-Archie reservoirs, because there
can be a significant bound-water saturation.
Non-Archie reservoirs can be evaluated in terms
of either effective or total porosity but, whichever parameter is adopted, the subsequently
derived water saturations must also be set within
that same porosity system, for consistency.
Exceptions to this simplified distinction include fine silty reservoirs and those where the
rock matrix exhibits microporosity. In both
cases, non-Archie behaviour can be caused by
a low surface charge density integrated over a
huge pore surface area rather than the conventional case of a higher surface charge density
(due to strong cation-exchange characteristics)
integrated over a more limited pore surface area.
It is difficult to accommodate such cases of high
pore surface area within either the effective or
the total porosity system.
In the following discussion the context is one
of dispersed shales within a water-wet reservoir
that has no fracture porosity and whose matrix
comprises sandstone, limestone or dolomite, or
some mixture thereof.

Because the chemically-bound water is included


within a wetted shale fraction, logs are corrected
for shale effects so that the bound water might
be removed from consideration. The wetted
shale is therefore replaced by electrochemicallyinert rock of identical geometry and with the
same physical properties as the clean rock
matrix. This is done by using the generic
correction algorithm:

Xcorr=X - Vsh (Xsh -Xma )

(1)

where X is the log response, Xsh is the log


response to shale, Xma is the log response to
clean rock matrix, and Xcorr is the log response
corrected for the effect of shale. Equation (1)
should use Vsh values that are derived from a
compatible shale indicator. For example, V~h
from the gamma log might not be appropriate to
correcting the density-log response whereas V~h
derived from a neutron-density shale indicator
would be more suitable. In general, the resulting
values of Xcorr become less accurate with

increasing Vsh.
After applying equation (1) to the sonic,
density and neutron logs, effective porosity can
then be evaluated as for clean sands:

doe=( x .... --Xma)/(X f --Xma ).

(2)

The results are implemented over net sand


intervals, which can be selected subsequently.
Effective water saturation Swe is the fractional
water content of the free fluid. It is evaluated
using a shaly-sand conductivity algorithm, such
as the modified Simandoux equation (Bardon &
Pied 1969):
Ct = (Cw/Fe*)

awen* + Vshfshawe n*-I

(3)

where Ct is the conductivity of the reservoir


rock, Cw is the conductivity of the formation
water, Csh is the conductivity of the wetted shale,
n* is the intrinsic Archie saturation exponent,
and the intrinsic formation factor Fe* is compatible with the effective porosity system in that it
is calculated from doe as follows:

Fe* = a*/doem*

(4)

where a* and m* are the intrinsic Archie


porosity coefficient and exponent, respectively.
A scheme for deterministic petrophysical
interpretation (with a* = 1) within the effective
porosity system is presented as Fig. 3. The
immediate deliverables are dOeand She, where She

216

P.F. WORTHINGTON

Shale-corrected
Log Response
DENSITY
NEUTRON
SONIC

_I
-

Porosity
Oe

L
F

] ARCHIE'S
[ ~ A=W
FIRST
1F
/ Oee.m

Core control
in clean zones
HELIUM POROSITY
Core control
INTRINSIC POROSITY

EXPONENT
m*

Formation water
sample
WATER
CONDUCTIVITY
Cw

UNIFIED SHALE /
VOLUME FRACTION
Vsh
Jl
Log Response
LATEROLOG

INDUCTION
Ct C sh

I_ ~_1 MODIFIED SIMANDOUX EQUATION


Ct=Cw

F-~

Swen +Vs h Csh Swn-1

Core control
INTRINSIC
SATURATION
EXPONENT

Oe Swe

n*

Fig. 3. Petrophysical interpretation within the effective porosity system.

is the effective hydrocarbon saturation such that


She = 1-Swe.

as that of Waxman & Smits (1968):


C t = ( C w / F t * ) S w n -q- (BQv/Ft*)Swtn*-I

Total porosity model


The electrochemically-bound water is intuitively
separated from a dry clay-mineral fraction and is
included within the porosity. Logs are not
corrected for the effects of dry shale, which is
therefore presumed to have the same physical
properties as the rock matrix. The assumption of
identical physical properties for dry shale and
matrix is approximately satisfied only in the case
of density. Total porosity is therefore evaluated
from the density log as for clean sands:
dot = (Pg-- Pb)/(Pg-- Pf )

(5)

where Pb is the log-measured bulk density, pf is


the density of interstitial fluids within the
volume sensed by the density tool, and pg is
the grain density rather than a pure matrix
density. The results are implemented over net
sand intervals, which must be specified at the
outset, because grain density will take account of
constituent shale, and this might cause intervaldependent departures from classical matrix
values, giving rise to a potential non-conformance with the assumptions of the method.
Total water saturation Swt is the fractional
water content of the total fluid. It is evaluated
using a shaly-sand conductivity algorithm, such

(6)

where Qv is the cation exchange capacity per


unit pore volume (equiv. litre-1), B is the
equivalent conductance of the (sodium) clayexchange cations (S m -1 equiv. -1 litre), a
function of Cw, and Ft* is compatible with the
total porosity system in that it is calculated from
dot as follows:
Ft* = a*/dotm*

(7)

A scheme for deterministic petrophysical


interpretation (with a * = l ) within the total
porosity system is presented as Fig. 4. The
immediate deliverables are dot and Sht, where Sht
is the total hydrocarbon saturation such that
Sht = 1-Swt.

Association of the models


The effective and total porosity systems are
associated through a sequence of relational
algorithms that allow parameters calculated in
one system to be transposed to the other. These
associations form the basis for any conjunctive
use of the two systems.

Shale volume fraction


The wetted and dry shale fractions are related

EFFECTIVE AND TOTAL POROSITY

_I

Log Response
DENSITY

Porosity

Core control
in clean zones
HELIUM POROSITY

ARCHIE'S FIRST ~ W L

1,o,

217

Core control
INTRINSIC POROSITY
EXPONENT
m*

Formation water
sample
WATER
CONDUCTIVITY
Cw

r ~176 , 89
Log Response
LATEROLOG
INDUCTION
Ct

WAXMAN-SMITS EQUATION
C t = C w Swtn*+ BQ v Swtn*-I

Ot

Swt

B -- f (Cw)
Core control
INTRINSIC
SATURATION
EXPONENT
n*

Fig. 4. Petrophysical interpretation within the total porosity system.


through the expression:
Vsh = Vd

+ ~bw

(8)

where
~)bw: Vsh ~)tsh

(9)

and ~tsh is the porosity of the wetted shale


calculated from the expression:
~tsh = (Pcl -- Psh)/(Pcl -- Pf)

(10)

It follows from equations (11) and (12) that


grain density can be re-expressed in terms of qbe
and Vsh as follows:
Pg = P m a ( 1 - - ~ e - - V s h ) + PclVsh(1 --(~tsh)
1 -- qbe -- Vsh ~tsh

(13)

Equation (13) reduces to Pg-----Prna when Fsh=0.


In the effective porosity system, pg is taken as the
matrix density Pma. In the total porosity system,
pg is not necessarily equal to Pma.

where Psh is the density of the wetted shale, pd is


the density of the dry shale fraction and for
practical purposes pcl is equated to pg. Shale
porosity is referred to the volume of wetted shale.
By combining equations (8) and (9) we have:

Porosity

Vcl = Vsh ( 1 - (~tsh)-

where Vsh is ideally the wetted shale fraction


from the neutron-density log combination. An
alternative form of this expression is:

(11)

In the effective porosity system, the estimated


Vsh is distinct from the rock matrix: in the total
porosity system, the unknown Vcl is grouped
with the rock matrix.

Grain density
If Vma is the fractional volume of the rock
matrix, the grain density can be expressed:
Pg = (Pma Vma + Pcl Vcl )/( Vma "[- Vcl).

(12)

Equation (12) reduces to pg= Pma when Vd = 0.

Effective and total porosity are related through


the expression:
Ct = Ce q- Vsh Ctsh

~)t = q~e-k-~)t Swb

(14)

(15)

where Swb is the bound-water saturation,


pursuant upon the dual-water model of Clavier
et al. (1984). Equation (15) follows from the
equality:
Swb = Vsh Ctsh/~)t

(16)

Yet another form of equation (15), based on


the normalized Qv concept of Juhasz (1981), is

218

P.F. WORTHINGTON

written:
~t-- ~be-~ ~t

Qv/Qvsh

(17)

to be.
Equations (4) and (7) can be combined as
follows:
(20)

where Qv is relative to the cation exchange


capacity per unit pore volume of a reference
shale Qvsh. If the bound water has a unit density,
a plausible assumption, equation (17) can also
be written in the empirical form of Hill et al.
(1979):

Equation (20) does not explicitly include the


porosity coefficient a*, although that quantity is
intuitively related to the value of m* for free-fit
regressions of reservoir data.

~)t = ~e nt- ~)t Qv (0.084 Ce~)'5 nt- 0.22)

Conductivity o f reservoir rock

(18)

where Ce is the concentration (in equiv, litre -1) of


the equilibrium water in the free pore space, net
of cation adsorption, and can be expressed as a
calculable function of Qv, Swt and saturating
water concentration Cs, at a reference temperature of 25~ Equation (18) has been seen as a
quantitative link between the effective and total
porosity models (Ruhovets & Fertl 1982), but it
is not a generally applicable equality, although
Juhasz (1979) did cite evidence that the volume
of bound water is effectively independent of
temperature over the range 20-200~ Note the
complication, to be discussed later, that the
conventional laboratory measurement of porosity through helium expansion is often carried out
on a humidity-dried sample, and there is a view
that the measured porosity is actually a hybrid
porosity, being somewhere between the limits of
effective and total porosity.

Formation res&tivity factor


Laboratory measurements of formation resistivity factor F fall under the umbrella of special
core analysis and, as such, are usually carried
out on lithologically cleaner samples, because it
is the practice to preserve the better quality
reservoir rock. Otherwise multiple-salinity measurements of electrical conductivity can furnish
an intrinsic formation factor F* for non-Archie
reservoirs. The resulting formation factors are
then correlated with their respective porosities
with the object of characterizing the first Archie
equation through the intrinsic porosity coefficient a* and exponent m* so that:

Ft* = Fe* (~)e/~)t) m~

By equating the reservoir rock conductivities of


equations (3) and (6) and setting Sw = 1, we
have:
(Cw + B Qv)/Ft* = (Cw/Fe*) + VshCsh
and therefore
Qv = Ft* (Cw + Vsh CshFe*) - Fe. Cw

Equation (22) describes the relationship between Qv and Vsh assuming that the water-zone
forms of the Waxman-Smits and Simandoux
equations are valid within their respective
porosity systems. It is interesting to consider
the boundary conditions on equation (22). When
Vsh=0, Fe* = F t* and therefore Qv=0. When
Vsh = 1 for a perfect shale in which Cw
approaches the bound-water conductivity Cbw,
4~e= 0 and therefore Fe* is infinite. Under these
conditions, equation (22) reduces to:
Qvsh =

Ftsh* Csh -- Cbw


B

(23)

where Ftsh* is the intrinsic formation factor of


the perfect shale. Significantly, equation (23)
might allow the determination of Qv for a
perfect shale by drawing upon the relationship:
ftsh* ~---a*/~)tsh m*

(24)

Equation (23) can be rewritten:

(19)

This equation is applied in both the effective and


the total porosity systems, in the form of
equations (4) and (7), respectively, according to
the nature of the log-derived porosity. There are
no separate relationships for the two systems
even though the earlier comments about porosity measurement might suggest that there ought

(22)

BF~.

Cbw ~- BQvsh

F* =a*/q~ m*

(21)

(25)

Ftsh*
or, alternatively:
Ftsh*= Ftsh (1 + (BQvsh/Cbw))

(26)

where
Fts h =

Cbw/Csh.

(27)

EFFECTIVE AND TOTAL POROSITY

In practice, however, g s h = 1 will correspond


to an imperfect shale that does not fully
comprise clay minerals and therefore the limiting
conditions will not be attained. Note that
equation (27) reveals the same definition of
shale formation factor as that related directly to
q~tsh through a pseudo-Archie expression in the
dual-porosity model of Raiga-Clemenceau et al.
(1984).

Fluid saturations
The interpreted hydrocarbon-filled porosity
must be the same in both the effective and the
total porosity systems, otherwise the computed
hydrocarbons in place will be different. Therefore:
~e She = ~t Sht

characterization of pseudo-matrix, the average


mixture of matrix and dry shale within the net
sand, which must be specified at the outset.
Fluid density within the flushed zone must also
be quantified over the same intervals. The
density log remains uncorrected for shale effects
and it is used directly to infer total porosity. The
computed porosities are input to the first Archie
equation to evaluate corresponding intrinsic
formation factors. Agreed algorithms for (a)
the equivalent conductance B in terms of Cw and
(b) the cation exchange capacity per unit pore
volume Qv in terms of total porosity have to be
established in support of the water saturation
equation (Fig. 4).
The above contrasting procedures suggest a
sufficient degree of difference to allow scope for
independent cross-checks between the two approaches.

(28)

Relative strengths

or

(1

- Swt)

(29)

Swt = 1 - (~e/4)t) (1 - Swe)

(30)

Oe (1 -- Swe ) = (~t

219

so that

Equation (30) allows a direct comparison of the


water saturations inferred in the two systems.

Quality assurance for formation evaluation


The effective and total porosity models can only
be used conjunctively to enhance core-calibrated
log interpretation if the two approaches are
sufficiently different to furnish independent
evaluations. The degree of independence is
governed by systemic differences in the application of these two models.

Systemic differences
The effective porosity system entails the characterization of matrix, fluid and shale points
without the need to specify net sand at the
outset. The neutron and density logs are
corrected for light hydrocarbon effects before
all three porosity logs are corrected for shale
effects. At that point the corrected porosity logs
are deemed to be sensing effective porosity. The
interpreted porosities are used to calculate
corresponding values of intrinsic formation
factor. A unified shale volume fraction and
shale conductivity are other essential inputs to
the water saturation equation (Fig. 3).
The total porosity system entails the density

The relative strengths of the total and effective


porosity systems are embodied within the meaningfulness of tying back to core data. This, in
turn, raises the question of how the core data
were measured.
Tying back to core is not possible for the
wetted shale fraction Vsh and the practice is
uncommon for the dry clay-mineral volume
fraction Vd in a solely petrophysical context.
Nevertheless, X-ray diffraction, X-ray fluorescence, chemical and thermogravimetric analysis,
scanning electron microscopy and infrared
spectroscopy can be used to gain a quantitative
insight into the occurrence of clay minerals and
thereby to establish some reference basis for Vcl,
although the subjective nature of some laboratory interpretations might detract from the
perceived usefulness of this approach as a
groundtruthing facility. In this respect, therefore, the total porosity system is the stronger and
it affords some opportunity for tying claymineral content back to core.
The relationship of Pma to pg can be used to
validate the underlying assumption of the total
porosity model, that the density of dry clay
minerals equals that of rock matrix. This
assumption is more likely to be satisfied where
the clean rock matrix has constant density. It is
unlikely to be satisfied where the clean matrix
properties are markedly variable. Where the
assumption is not satisfied but shale density is
known, it is theoretically still possible to proceed
with a total porosity approach, but on a levelby-level basis. This procedure would require a
quantification of Vcl at each digital sampling
level of the well logs, so that grain density might

220

P.F. WORTHINGTON

be evaluated at each level. This requirement


would, in turn, necessitate a reversion to the
relationship between V~h and Vcl. Because of this
potential complexity, which many regard as
prohibitive, the effective porosity model is seen
as the stronger in terms of the opportunities for
utilizing and validating the density of reservoir
rock. Note, however, that this contention is
dependent upon the measurement of a meaningful effective porosity.
The tying back to core of log-derived porosity
values is fraught with potential difficulty. It has
been argued, but not universally, that the oven
drying of core plugs at temperatures of around
105-110~ removes chemically-adsorbed waters
without altering, chemically, the solid claymineral fabric. Therefore, it has been claimed
that helium porosities measured subsequently on
these plugs are likely to be total porosities. On
the other hand, the humidity drying of core
plugs at temperatures of about 60~ is claimed
by some to retain the bound waters, while
expelling the free water, so that porosities
measured subsequently are effective porosities.
This contention is at variance with the data of
Hill et al. (1979), which suggest that some bound
waters are expelled despite the humidifying
process and that the measured helium porosities
are intermediate relative to the effective and
total porosities (Juhasz 1988). The effect may
not be serious in reservoir rocks, for Pallatt &
Thornley (1990) note that electrochemicallybound water accounts for less than 2.5% of
the pore volume: for a rock with 20 porosity
units, the estimated bound-water volume is
therefore less than 0.5 porosity units, a figure
which is equivalent to the uncertainty associated
with core porosity measurement. Nevertheless,
the only uncontentious way of groundtruthing
log-derived porosity to conventionally-measured
core porosity is to confine such comparisons to
essentially clean zones. Under these conditions,
both the effective and the total porosity models
are equivalent.
A comparison of Fe* and Ft* offers an
intrinsic measure of the effect of pore geometry
on electrical conduction, subject to the assumptions of equal porosity coefficients and equal
porosity exponents for the two systems. Because
the definitive multiple-salinity procedures furnish Ft*, the total porosity model provides the
sounder physical basis. Laboratory-measured
values of Ft* therefore serve as the definitive
reference.
In a user setting, values of Qv are obtained
from a dubious relationship to porosity and
estimates of gsh are made from one or more of
several tenuous shale indicators. There is no

reference value of Vsh available from the


laboratory: there might be values of Qv, which
can be taken as definitive if these are determined
meaningfully from multiple-salinity measurements of rock conductivity. Therefore, such a
Qv database becomes the definitive core reference. The relationship of the equivalent conductance of (sodium) clay exchange cations B to
the conductivity of saturating electrolyte Cw
remains a weak link in the total porosity system,
because several different relationships have been
proposed. This weakness is not removed by the
use of multiple-salinity conductivity data, which
require a value of B before Qv can be quantified.
Tying back log-derived water saturations to
core is usually founded on the extraction of
interstitial waters from vertical plugs cut from
the inner parts of whole core pieces that have
been drilled using a low-invasion coring bit with
an oil-base mud. The procedure is established
but not yet standard practice. The plug-extracted water saturations are notionally values
of Swt. They can be used as a reference for the
effective porosity system through the equivalence of hydrocarbon-filled porosity. In particular, the comparison serves as a validation of the
feeder relationships, especially that of Qv vs ~bt,
which is often highly tenuous.
Q u a l i t y assurance s c h e m e

A quality assurance scheme for deterministic


open-hole formation evaluation is shown in Fig.
5. The purpose of this simplistic scheme is to
illustrate how the effective and total porosity
models can be operated conjunctively to increase
confidence in the resulting petrophysical interpretation. It is not intended to constitute the
ultimate framework for quality control but
rather to indicate how greater confidence in
petrophysical interpretation can be secured
through an integrated use of the two models.
The first element of Fig. 5 is concerned with
tying back Vsh to core-derived Vcl through ~tsh,
the determination of which requires a knowledge
of pcl. A satisfactory tie-back would reconcile
wetted shale and dry clay-mineral fractions in
the two porosity systems. Failure to secure
agreement over net sand intervals could be
attributed to an inappropriate log-derived shale
indicator, to an unrepresentative dry clay
density, or to subjectiveness in the interpretation
of core data. As in all cases of tying log data to
core, the scale disparity might render the
datasets irreconcilable, especially in markedly
heterogeneous reservoir zones. Further, an unsatisfactory outcome at the subsequent (second)
key stage might suggest an iteration through the

EFFECTIVE AND TOTAL POROSITY

221

Fig. 5. Foundations of a quality assurance scheme for open-hole petrophysical interpretation.

first.
The second element is concerned with tying
the computed grain density pg back to core
through log-derived values of Vsh, q~e and ~btsh,
and a knowledge of pc~ and Pma- A satisfactory
tie-back would substantiate the assumptions
concerning Pcl and Pma, the latter being verifiable
over any shale-free intervals of net sand. Failure
to secure agreement could be attributed to
unrepresentative densities of matrix or dry clay
minerals or to errors in q~e, which is not qualityassured until the third key stage, again suggesting some degree of iteration.
The third element of Fig. 5 reconciles logderived effective porosity with log-derived total
porosity through a comparison of ~bt indirectly
calculated from q~e with that interpreted directly
within the total porosity system over net
reservoir intervals. The third element also allows
both the log-derived porosities to be referred to
core porosity over net reservoir intervals.
Because of the uncertainty associated with the
influence of sample preparation on the measured
core porosity of shaly plugs, the tying back to
core might best be done in two stages. First, the
log-core comparison should be restricted to
clean intervals to establish that the interpretation systems are functioning under the most
straightforward conditions. Second, in view of
the earlier comments concerning a hybrid core
porosity, the tying back to core over shalier
intervals of net reservoir should allow the
measured core porosity to be of intermediate
value relative to the log-derived effective and

total porosities. Indeed, if the core porosity lies


between the corresponding log-derived values,
this might be the best quality assurance that one
could reasonably expect to achieve. Failure to
reconcile the two datasets would suggest shortcomings in Vsh and/or (/)tsh and would require
some iteration through elements (1) and (2).
The fourth element reconciles log-derived Fe*
with log-derived Ft* through a comparison of
Ft* indirectly calculated from -be* with that
interpreted directly within the total porosity
system over net reservoir intervals. This is no
more than a check for internal consistency.
However, the fourth element also allows both
the log-derived intrinsic formation factors to be
referred independently to laboratory values of
Ft*, preferably those obtained from multiplesalinity conductivity measurements of plugs
from net reservoir intervals. Failure to tie back
to core would suggest transmitted errors in ~be
and/or ~bt, or perhaps the use of an inappropriate value of the intrinsic porosity exponent m*.
The fifth element of Fig. 5 allows Qv estimated
from a total porosity log to be tied back to core
values, preferably those obtained unambiguously from multiple-salinity conductivity measurements of plugs from net reservoir intervals.
This exercise serves as a check on the meaningfulness or otherwise of the algorithm used to
predict Qv from a porosity log. Since this
algorithm is itself characterized using core data,
the core-derived Qv data should be distinct from
those used to establish the relationship between
Qv and qSt. The fifth element also allows log-

222

P.F. WORTHINGTON

derived Vsh values to be reconciled with logderived Qv data through a comparison of Qv


values calculated from Vsh with those inferred
directly from porosity and already validated
through reference to core data. Failure to
reconcile the data at this key stage would most
likely be attributable to uncertainties in B and
Csh.
The sixth and final element reconciles logderived effective water saturation with logderived total water saturation through a comparison of Swt indirectly calculated from Swe
with that interpreted directly within the total
porosity system over net pay intervals. The sixth
element also allows both the log-derived water
saturations to be referred to core-derived water
saturation. Failure to reconcile the data at this
stage would imply possible errors in 4)e, ~t, B,
Csh, Cw, a* or m*, and it would require iterating
perhaps as far back as the third key stage.
The quantification of the inter-model comparisons should take the form of acceptable
tolerances in the agreement between directly
and indirectly inferred values of the relational
parameters and in their validation against core
data. The development of these tolerances would
be an immediate sequel to broad adoption of
this proposed conjunctive interpretation scheme.

Conclusions
A comparison of open-hole petrophysical interpretation practices for non-Archie reservoirs
that are set exclusively within either the effective
or the total porosity system has identified a set
of relational algorithms through which these
interpretative models can be associated. This
identified numerical equivalence allows intermodel assessments of the consistency and
validity of the interpreted data at key stages of
the petrophysical evaluation process, so that
some measure of reliability may be established.
The assessments involve comparisons of interpretations made by separately using the two
porosity models as well as the tying of these
interpretations back to core. The key stages
form the basis for a quality assurance scheme
that draws upon the integration of traditionally
separate areas of petrophysical systemics. The
development of such a scheme in the form of
quantitative measures of inter-model agreement
would constitute a logical extension of the
demonstrated association of the effective and
total porosity models.
This initiative forms part of an essential thrust
to complement the excellent quality control that
currently exists in well-log data acquisition. At
present, our ability to acquire petrophysical data

exceeds our ability to interpret, especially in the


three-dimensional settings of extended reach and
multilateral wells. Ongoing advances in the
three-dimensional modelling of tool responses
will shortly allow meaningful environmental
corrections in a way that opens the door to
enhanced 3D interpretation of open-hole well
logs. If the community is to draw the greatest
benefits from that projected situation, a qualityassured interpretation scheme will be required
for open-hole petrophysics. This paper has
emphasized the nature of the technical positioning that will be needed to secure those benefits as
we approach the millennium.

Nomenclature
B equivalent conductance of (sodium) clayexchange cations (S m 1 equiv. 1 litre)
Cbw conductivity of bound water ( S m ')
Csh conductivity of wetted shale fraction (S m -1)
Ct bulk conductivity of reservoir rock (S m -1)
Cw conductivity of free water ( S m -1)
F* intrinsic formation (resistivity) factor in
generic form
Fe* intrinsic formation (resistivity) factor in the
effective porosity system
Ft* intrinsic formation (resistivity) factor in the
total porosity system
Ftsh formation (resistivity) factor of a perfect
shale in the total porosity system
Ftsh* intrinsic formation (resistivity) factor of a
perfect shale in the total porosity system
Qv cation exchange capacity per unit pore
volume (equiv. litre-1)
Qvsh cation exchange capacity per unit pore
volume of shale (equiv. litre -1)
She fractional hydrocarbon saturation in the
effective porosity system
Sht fractional hydrocarbon saturation in the
total porosity system
Swb fractional bound-water saturation
Swe fractional water saturation in the effective
porosity system
Swt fractional water saturation in the total
porosity system
Vd volumetric fraction of dry clay minerals
Vmavolumetric fraction of clean rock matrix
Vsh volumetric fraction of wetted shale
X generic log response
Xcorr generic log response corrected for shaliness
Xma generic log response to clean rock matrix
Xsh generic log response to shale
a* Archie intrinsic porosity coefficient
Ce concentration of equilibrium free water
(equiv. litre 1)
c~ concentration of saturating water (equiv.
litre-1)

EFFECTIVE AND TOTAL POROSITY


m* Archie intrinsic porosity exponent
n* Archie intrinsic saturation exponent
Pb log-derived bulk density ( g c m -3)
pc~ density of dry clay-mineral fraction (g cm -3)
pf density of interstitial fluids (g cm -3)
pg grain density over net sand intervals ~g cm 3)
Pma density of clean rock matrix ( g c m -~)
psh density of wetted shale fraction (g cm -3)
Obw bound-water porosity fraction
Oe effective porosity fraction
Ot total porosity fraction
~tsh total porosity fraction of shale

References
ARCHIE, G. E. 1942. The electrical resistivity log as an
aid in determining some reservoir characteristics.
Trans. AIME 146, 54-62.
BARDON, C. 8z PIED, B. 1969. Formation water
saturation in shaly sands. Trans. SPWLA lOth
Ann. Logging Syrup., Zl-19, Society of Professional Well Log Analysts, Houston, Texas.
BUSH, D. C. & JENKINS, R. E. 1970. Proper hydration
of clays for rock property determinations. Journal
of Petroleum Technology, 22, 800-804.
CLAVIER, C., COATES, G & DUMANOIR, J. 1984.
Theoretical and experimental bases for the dualwater model for interpretation of shaly sands.
Society of Petroleum Engineers Journal, 24, 153167.
HILL, H. J., SHIRLEY,O. J. & KLEIN, G. E. 1979. Bound
water in shaly sands--its relation to Qv and other
formation properties. The Log Analyst 20(3),
3-19.
HURST, A. & NADEAU, P. 1994. Estimation of water
saturation from clay microporosity data. SPE
Paper 28850, Society of Petroleum Engineers,
Richardson, Texas.

223

JOHNSON, W. L. & LINKE, W. A. 1977. Some practical


applications to improve formation evaluation
of sandstones in the Mackenzie Delta. Trans.
CWLS 6th Formation Evaluation Symposium,
R1-32, Canadian Well Logging Society, Calgary,
Alberta.
JUHASZ, I. 1979. The central role of Qv and formationwater salinity in the evaluation of shaly formations. Trans SPWLA 20th Ann. Logging Syrup.,
AA1-26, Society of Professional Well Log Analysts, Houston, Texas.
- 1981. Normalised Qv--the key to shaly sand
evaluation using the Waxman-Smits equation in
the absence of core data. Trans. SPWLA 22nd
Ann. Logging Syrup., Z1-36, Society of Professional Well Log Analysts, Houston, Texas.
MAYER, C. t~ SIBBIT, A. 1980. GLOBAL, a new
approach to computer-processed log interpretation. SPE Paper 9341, Society of Petroleum
Engineers, Richardson, Texas.
PALLATT, N. & THORNLEY,D. 1990. The role of bound
water and capillary water in the evaluation of
porosity in reservoir rocks. In: WORTHINGTON,P.
F. (ed.) Advances in Core Evaluation, Gordon and
Breach, Reading, 223-237.
RAIGA-CLEMENCEAU, J., FRAISSE, C. GROSJEAN, Y.
1984. The dual-porosity model, a newly developed
interpretation method for shaly sands. Trans.
SPWLA 25th Ann. Logging Syrup., F1-16, Society
of Professional Well Log Analysts, Houston,
Texas.
RUHOVETS,N. & FERTL, W. H. 1982. Digital shaly-sand
analysis based on Waxman-Smits model and logderived clay typing. The Log Analyst 23(3), 7-23.
WAXMAN, M. H. & SMITS, L. J. M. 1968. Electrical
conductivities in oil-bearing shaly sands. Society
of Petroleum Engineers Journal, 8, 107-122.
WORTHINGTON,P. F. 1991. The direction of petrophysics: a five-year perspective. The Log Analyst
32(2), 57-62.

Permeability prediction in anisotropic shaly formations


S. X U & R. W H I T E

Exploration Geophysics Group, Research School of Geological & Geophysical Sciences,


Birkbeck College & University College London, Malet Place, London WC1E 6BT, UK
Abstract: We present a unified model for simulating the permeability and electrical

conductivity of anisotropic shaly formations. The model is based on Willis' formulae and
the concept of a host medium, the selection of which is crucial in predicting these transport
properties. Different rock components, including shales and mudrocks, are characterized by
parameters typifying their pore geometry, namely the aspect ratio, size and orientation
distribution of the pores. In this regard the model is an extension of the elastic model of Xu
& White for predicting P- and S-wave velocities in siliciclastic rocks. The electrochemical
effect of clay minerals on electrical conductivity is simulated by Waxman & Smits' model. A
novel feature of the permeability model is that its percolation factor is estimated by a nonlinear transformation of the percolation factor found from conductivity measurements.
The model was tested on the laboratory measurements published by Waxman & Smits.
Comparison of the results with those from the Waxman & Smits, Dual-Water, and K o z e n ~
Carman models, and with multilinear and non-linear regression techniques, demonstrated
that the unified model predicted conductivity and permeability more accurately than any of
these models from the same number or fewer parameters. The improved prediction was most
noticeable in samples containing a significant clay mineral fraction. Apart from Waxman &
Smits' data, we have found no published dataset that is comprehensive enough to test
physical predictions of both conductivity and permeability.

Permeability is one of three key rock parameters


in reservoir simulation and the provision of
detailed estimates of permeability is a prime
objective of applied petrophysics. Since permeability cannot be measured directly by logging
tools, it is usually estimated indirectly from well
logs with calibration from cores. A common
practice is to use core measurements to establish
an empirical relationship between permeability
and properties such as porosity and formation
factor and then to apply that relationship to well
logs to construct permeability logs. Permeability
is a complex property to predict empirically and
it is not easy to obtain detailed information on
the parameters that control the flow of fluids
through rocks. Needless to say, the prediction of
permeability is problematic.
It is well understood that permeability is
controlled by five key factors: porosity, the size,
shape, orientation and connectivity (percolation
or tortuosity) of pores. Both laboratory measurements (e.g. Beard & Weyl 1973) and
theoretical analysis (e.g. Carman 1956) indicate
that permeability is more sensitive to pore size
than porosity. Pressure, cementation, grain size,
clay content, sorting and irreducible water
saturation affect permeability indirectly by
modifying or controlling the five key parameters
mentioned above. Porosity can be determined

reasonably accurately from well logs but there


are no direct measurements of pore size, shape
and connectivity. Consequently, permeability
prediction has to rely on indirect measures of
these parameters. The danger of resorting to
purely ad hoc empirical relationships is that they
can end up eclectically tuned to a particular
dataset.
The effect of clay content on permeability has
long been recognized. Thompson & Callanan
(1981) measured porosity and permeability of
synthetic clay samples at pressures in the range 0
to 10000 psi. Although the measured porosities
were high (in the range of 20% to 50%), the
permeabilities are 3 to 4 orders of magnitudes
lower than those measured from artificial sand
packs by Beard & Weyl (1973). The low
permeabilities were explained as a result of the
remarkable sealing power of clay particles.
Thomson (1978) observed a progressive decrease
in permeability with clay content from rock
samples with clay content in the range of 5% to
15%. A number of authors observed a linear
trend between log(k), the logarithm of permeability, and porosity 4) which was later explained
as a result of a systematic reduction in both k
and 4 by dispersed clays (Bos 1982). From
laboratory measurements on over 100 shaley
sand samples, Goode & Sen (1988) found a good

XU, S. WHITE,R. 1998. Permeability prediction in anisotropic shaly formations.


In- HARVEY, P. K. 8z LOVELL, M. A. (eds) Core-Log Integration, Geological Society, London,
Special Publications, 136, 225-236

225

226

S. XU & R. WHITE

correlation between log(k) and log(qSm/Qv),


where m is the cementation factor and Qv the
exchange cation molarity. Sen et al. (1990)
measured Qv , the inverse surface-to-volume
ratio (Vp/S), proton N M R decay time T1 and
permeability k of some 100 sandstone core
samples and found good correlations between
k and log(q~mVp/S), log(q~m/Qv) and log(~bmT1).
This is understandable since Vp/S, Qv and T1 are
three different measures of clay content. The
authors concluded that clay affects the permeability more in rocks where it adheres in pore
throats than in those where it adheres in the pore
pockets.
Although considerable understanding of how
clay affects permeability has been gained from
laboratory measurements, little theoretical work
has been done to model the effect. The majority
of the models that relate permeability to clay
content combine empirical and simple logical
considerations (e.g. permeability must be dimensionally length squared). Starting with the
Kozeny-Carman equation (Carman 1956), de
Lima (1995) obtained some relationships between k, 49, Vp/S and Qv similar to those
obtained empirically by Sen et al. (1990).
We have developed a model for elastic wave
velocities (Xu & White 1995a,b,c, 1996a) which
can accurately simulate the combined effects of
porosity, clay content, fluid content and frequency on elastic wave velocities in clastic
silicate rocks. The model is founded on physical
concepts and has demonstrated its practical
utility in a number of case studies involving
reservoir geophysics (formation evaluation, seismic modelling and interpretation). In addition
to validation on published laboratory measurements, the model has been successfully tested on
numerous wells, including four blind tests (Xu et
al. 1997).
The key feature of the model is its ability to
predict the effect of clay content on wave
velocities, including the two distinct porosityvelocity trends observed in the laboratory: one
for shaly sands and the other for sandy shales
(Marion et al. 1992). Empirical models that treat
clay content purely as a lithological factor
cannot explain this. Laboratory measurements
and well logs indicate that two factors need to be
considered; lithology and the influence of clays
on pore compliance. Not only are clay particles
more compliant than sand grains, their sheetlike nature tends to make the pore space more
compliant. This greater compliance can be
modelled by introducing an additional pore
space characterized by a smaller aspect ratio
than that of clean sand grains. Thus the model
predicts the higher Vp/Vs values observed for

shales than sands. The elastic anisotropy of


shaly formations is modelled through a preferred orientation for clay-related pores.
Here we extend the model to predict the
conductivity and permeability of shaly formations. Each pore is assigned an idealized
ellipsoidal shape and embedded in a porous
medium. We use Willis' (1977) formulae to
compute its contribution to the overall conductivity and permeability. The interaction between
this pore and other pores is modelled via the
concept of a host medium. The properties of the
host medium are then tuned to model the
observed conductivity and permeability.
The concept of the host medium was originally proposed by de Kuijper et al. (1995). In Xu
& White (1996b) we show that alternatives such
as the self-consistent scheme (SC) and the
differential effective medium scheme (DEM)
cannot model with Archie's law for clean sands
whereas modelling with a host medium hunting
technique reproduced Archie's Law when the
fluid percolation factor was selected as 0.04q~.
The next section describes the model which is
then tested using published laboratory measurements. The results show that the transport
properties of anisotropic shaly formations can
be modelled by assigning a characteristic pore
size, shape and orientation to their clay fraction.
The model provides a permeability predictor
that estimates a percolation factor from resistivity measurements, if available, and then applies
it to permeability prediction.

The unified model for shaley formations


As in the elastic model of Xu & White (1995a,b,
1996a), we assume that the total pore space can
be divided into sand-related pores and clayrelated pores. The pore space is partitioned
proportionately:

where
Vc
~bc = _---~b
1

(2)

4,s = 4 - 4~c.

(3)

and

Vc is fractional clay content. The sand-related


pores are characterized by a pore aspect ratio
(ratio of short semi-axis to long semi-axis) C~s
and pore size (long semi-axis) as and the clayrelated pores are similarly assigned a characteristic pore aspect ratio O~c and pore size ac. We

PERMEABILITY PREDICTION
further postulate that the sand-related pores are
randomly oriented whereas the clay-related
pores tend to align themselves in a plane. This
assumption conforms with observations of
strong seismic and ultrasonic anisotropies for
shales and isotropic wave propagation in clean
sandstones. In modelling logs, Vc is generally
replaced by shale volume Vsh, which would lump
the fractional volume of silts and various
mineral fragments with Vc. The model could in
principle take account of these different components and different clay minerals if there were
practical log analysis procedures distinguishing
them.

227

five phases:
1. a non-conducting solid phase;
2. a clay mineral phase with a finite but very
small conductivity;
3. a randomly oriented sand-related pore
fluid phase with conductivity Cwe;
4. a clay-related pore fluid phase with the
same pore fluid conductivity Cwe but with a
preferred pore orientation;
5. a non-conducting hydrocarbon phase.
The percolation factor Fc for conductivity is
defined as in equation B1 in Appendix B:

CH= FcCwe+(1-Fc) Cm

Conductivity
In simulating electrical conductivity special
consideration must be given to the electrochemical behaviour of clays. Waxman & Smits
(1968) demonstrated that shaley sands behave as
perm-selective cation exchange membranes and
their electrochemical efficiencies increase with
increasing clay content. They modelled this by
supplementing water conductivity Cw with a
conductivity Ce from the clay counter-ions
within the ionic double layers:

Ce = BQv

(5)

where /~eNa is the maximum (sodium) cation


exchange ion mobility (in cm 2 Volt 1 s), b and 7
are empirically determined constants. Waxman
& Smits (1968) found from their laboratory
measurements that
B = 1 - 0 . 6 e x p ( - Cw/0.013)]0.046

(6)

where Cw is in ohm m q. The effective conductivity of the formation water Cwe is simply
the sum of Cw and Ce.
Cwe = Cw + Ce

where Cn, Cwe and Cm denote the conductivity


tensors of the host medium, formation water
and matrix (mixture of the sand grains and clay
particles). Fc describes the degree to which the
fluid paths accord with a parallel tube model; it
decreases with increasing tortuosity of the fluid
paths. There is no information as to the value of
Fc and we estimate it from the measurements.
Fc can be correlated with other measured
parameters, such as q~ and Qv, to see what
factors control it.

(4)

where Qv is the molar volume concentration of


clay exchange cations per unit pore volume (meq
cm 3). Qv is a function of the cation exchange
capacity (CEC) of clay minerals, clay content,
porosity and the density of dry clays. B is the
equivalent conductance of clay exchange cations
(in ohm cm 2 meq 1) which is a function of the
conductance of formation water Cw. At 25 ~
B = [1 - b e x p ( - Cw/~/)]0.0 l AeNa

(8)

(7)

In order to apply the modified Willis' formulae (equations A5-A11) to the conductivity
of shaly sands, we subdivide the formation into

Permeability
In simulating permeability, there is a problem in
defining the intrinsic permeabilities of the inclusions. One approach is to start from the intrinsic
permeability of two parallel plates:
be
k = -12

(9)

where b is the separation between the two plates.


For a spheroidal inclusion defined by xZ/a~2+ y2/
a2+ z2/c 2= 1, equating the hydraulic aperture bh
parallel to its long axis with the mean square
value of the separation 2z gives:
bh2 = 1 .fjs (2z)2dS = -~o~2a 2 = 2C2

(10)

where A=Tr a 2 is the area of the domain S


defined by x 2 + y 2 = a 2 and a is the aspect ratio
of the inclusion (c/a). For the intrinsic permeability of the inclusion parallel to its long axis we
use
k = 1__a2a2 = ~1c - ,~.
6

(11)

Strictly this should be a tensor property. A


similar equation, but with an undetermined

228

S. XU & R. WHITE

numerical constant, is obtained from simple


dimensional arguments. Since in practice a bestfit characteristic value is assigned to c, the
numerical constant of 1/6 has no real significance. Unlike conductivity, the permeability of a
porous rock does depend on pore sizes, especially pore throat diameters.
The intrinsic permeabilities for sand- and
clay-related pores are taken from equation 11
1

ks = gO~s as-

(12)

kc = ~1 ac 2ac2.

(13)

and in porosity from 5% to 31%, making it an


ideal dataset for testing the model.
Electrical conductivity
Applying the model to conductivity measurements has two aims:
(1) to test its capability of predicting electrical
conductivity, and
(2) to estimate a percolation factor for each
sample for use in permeability prediction.

and

We introduce a separate percolation factor Fp


for permeability since the relative weighting of
permeability is not necessarily the same as that
for conductivity. For example, a clay particle in
a pore throat still conducts electricity but
seriously impedes fuid flow. By employing a
modified Voigt-Ruess-Hill average scheme, we
define Fe for permeability as follows:
kH = 0.2kll + 0.8k~, kll = Fpks + (1 - Fe)kc,
k ~ -1 = Fpks < + (1 - Fp)kc q

(14)

where kH, ks and kc are permeabilities of the host


medium, sand-related pores, and clay-related
pores. The equation signifies that when the sandrelated pores are selected as the host medium,
the system is most percolating and when the
clay-related pores are selected, the system is least
percolating.
To apply Willis' formulae to permeability, the
composite is assumed to consist of three phases:
1. an impermeable matrix phase of sand
grains and clay particles;
2. a sand-related pore phase with a random
pore orientation;
3. a clay-related pore phase with a preferred
pore orientation.

Application to laboratory measurements


The dataset
The model was tested on the laboratory measurements published by W a x m a n & Smits
(1968). The dataset provides the porosity, brine
permeability, Qv and conductivities at four or
more brine salinities of 49 sandstone samples
(table 2 in Waxman & Smits 1968). The samples
range in clay content from clean to very shaly,

The Waxman-Smits (WS) and the D u a l Water (DW) models (Clavier et al. 1984) were
also tested on the dataset for comparison.
To simulate electrical resistivity, all three
models require porosity, clay content and brine
conductivity (Cw) as input parameters. As there
were no direct measurements of clay content, we
estimated it from Qv using a relation given by
Juhasz (1979):
V~l(drv) =
9

Qvq~t

(15)

Pcl (dry) C E C c l

where Vcl(dry) denotes the dry clay content as a


fraction of bulk volume, Pcl(dry) denotes the
average density of the mixture of dry clay
minerals (in gcm 3), 4t is the total (fractional)
porosity and CECd is the averaged cationexchange capacity of the clay minerals present
in the formation (meq gq dry clay).
When applying the WS and D W models, the
formation factor FF in both models was tuned to
get the best fit. It is well recognized that FF is
controlled by porosity and cementation factor
which is, in turn, a function of pore geometry
and tortuosity of the electrical current flow
paths. Hence FF is expected to vary from sample
to sample. When applying our model to the
dataset, the percolation factor Fc was tuned by
fitting the predictions with the conductivity
measurements.
Figure 1 compares sample measured electrical
conductivities with predictions from the three
models. All three work well for clean sandstone
(upper figure). Our model worked slightly better
than the WS and D W models for shaley
sandstone (lower). The normalized mean square
errors of fit (termed incoherence by de Kuijper et
al. 1995) for all 49 samples is shown in Fig. 2.
Our model (lower) fits the data slightly better
than the D W (upper) and WS (middle) models.
Figures 3 and 4 show the cross plots of the
estimated percolation factor Fc as a function of
porosity and shale volume. There are two welldefined trends on the Fc-q~ cross plot: one for

PERMEABILITY PREDICTION

229

25

~E20

"! 15

%,

0
1;

1;

2'0

10
lid
0

3;

35

4'5 50 55

40

Sample Number

5
0

25

50

100

150 200
(mS/cm)

7
250

Conductivity of brine

_AA

/o ,; 2o 25 3; 3'5 20 ~5 5; 5;
Sample Number

57

"64
0

o
.~ 2

/o t'5 2'0 2'5A3'o 35 4'0 ;

5; 5;

Sample Number

nl

50
100
150
200
Conductivity of brine (mS/cm)

250

Fig. 1. Comparisons between the measured electrical


conductivities (solid squares) and those predicted using
the WS (dash<lotted line), DW (dashed line) and
unified (solid line) models. Upper: clean sandstone;
lower: very shaley sandstone.

Fig. 2. Normalized mean square error of fit (termed


incoherence by de Kuijper et al. 1995) for the DW
(upper), WS (middle) and unified (lower) models 9

FC versus porosity

-10 a

100.~ 604',,

3o9

shaley sands (upper trend) and one for sandy


shales (lower trend). There is only one trend on
the Fc-Qv plot, indicating that clay content plays
a more important role in controlling percolation
of the fluid phase.

Permeability
Figures 5 to 7 show cross plots of permeability
versus porosity, Qv and formation factor.
Permeability is clearly affected by all three
factors. There are two distinct trends on the k4~ and k - F F plots but apparently just one on the
k-Qv plot.
Of the many models in the literature for
predicting permeability, the most popular are
the Kozeny-Carman equation and multilinear
and non-linear regression techniques.
The K o z e n y - C a r m a n e q u a t i o n ( C a r m a n
1956) is based on a tube-like model of the pore
paths in a rock. Flow through a porous medium
is represented by a bundle of tubes of different
radii. Within each tube, the flow is laminar
rather than turbulent. The tubes are also

,mmI "
9

mt m
|m
9
9

10-

6-

31

0.0

oi,

o12

o13

o14

Porosity (fractional)
Fig. 3. Cross plot of the estimated percolation factor
Fc as a function of porosity.
assumed to be twisted with tortuosity 7-= La/L,
where La is the assigned length of a tube and L is
the length of the sample. Under these conditions, the Kozeny-Carman equation becomes
k-

~bRh2

(16)

fTwhere R h is the mean hydraulic radius a n d f i s a


dimensionless shape factor between 1.7 and 3. If
R h is related to the specific surface area S,
defined as the ratio of pore surface area to grain

230

S. XU & R. WHITE
P e r m e a b i l i t y versus F F

Fc versus Qv
- 1 0 -3
100-

&

~"..~

6o-

. ..
9 ii m

~9

30-

102 -

= ~ ' ~ 1 /n ~ 9 9 9

9
9 me

====

.~

10 ~

"'%

10-

6-

o==

~.

"

10_2 3-

006

0;

l O -4

o16 ,o

io

4'O

10

'

100

200

Fotenation F a c t o r

Qv (meq/ml )

Fig. 4. Cross plot of the estimated percolation factor


Fc as a function of shale volume.

Fig. 7. Cross plot of the measured permeability as a


function of formation factor.

volume, the equation becomes


Permeability versus porosity

~)3

k=
9 "

102--

t:.

To apply the KC model to this dataset, we


used the empirical equation given by Clavier el
al. (1984) to estimate specific area S from Or:

g
.~

(17)

f r S 2 ( 1 - ~b)2.

10 ~

el

102

10 ~

o9

o11

o12

o13

o14

Porosity (fractional)
Fig. 5. Cross plot of the measured permeability as a

function of porosity.

where v is a constant, which was determined


from laboratory measurements as 450 m 2 meq -1.
Figure 8 shows the cross plot of the measured
permeability and that predicted from the KC
equation. Tuning the constants in the K o z e n y Carman equation will shift the graph upwards or
downwards but will not change the scatter. The
reason for the scatter is probably that the
tortuosity and shape factor in the KC equation
are assumed to be constant during the calcula-

Permeability versus Qv
Measured versus predicted
10 -~
102-

.-~
,-.,.,

9 "":-:'k-'."

10 -2

10 ~

102 .....................................................................................

102

ii--...e ....... 102


~o

10 ~

==

==

10' ...................................

%.

10-2-

0
mi

~ . . - ' - ......... ~ ..................... i ................. 10~

.'. : :'r
9

1 0 : ..................................................................................................

10 2

10

10 ~

10'
I

0.06

0 I

91

013

0'.6

l'.0

3.0

Qv (meq/ml)

Fig. 6. Cross plot of the measured permeability as a


function of volume concentration of clay exchange
cations Qv, an indicator of clay content 9

10"

10 2

10 ~

10:

Measured permeability (roD)

Fig. 8. Cross plot of the measured permeability and


that predicted using the KC model.

PERMEABILITY PREDICTION
Factor F P versus Factor F C

Predicted v e r s u s measured
101~0~

10 -~

10 ~

----.

"~

102 . ...................................................................

10 2
i

10'
10"

~.~#.. .................... 102

"-.'i"
10 o ...............................; . . ;

.~

10 ~

"~

10"-

,.~

l O 2--

10 .2

"

BBB
9

i 9

10 -~_

10 ~

10 -6
10 -4

"l t

10~_

10 ............. | ........ ~.i,.~.. 9....................................................

10"

...

10.~ -

.........'.....'...~....~............................10 ~

t "-"

~"

231

10 ~

10 2

lO
- 1 0 -~

)~0"~

Measured permeability (mD)

30

60

100

F a c t 9 f r o m conductivity data

Fig. 9. Cross plot of the measured permeability and


that predicted from a mixed multilinear/non-linear
regression.

tion. In reality they are a function of clay


content and porosity and vary from sample to
sample.
Multilinear and non-linear regression techniques are widely used for permeability prediction.
These techniques rely on correlation with
measurements for which there is some plausible
connection to permeability without necessarily
investigating the basis of the correlation or the
physics behind the equations.
For multilinear regression, we assumed that
logt0(k) was a linear function of porosity, Vsh
and formation factor (FF). In looking for a nonlinear relationship, we regressed lOgl0(k) on
log10(4~), lOgl0(Vsh) and lOgl0(FF) and various
linear and non-linear combinations between
log]0(k) and the input parameters. The following
relationship gave the best result:

Fig. 10. Cross plot of Fc the percolation factor


estimated from the conductivity data and Fp the
percolation factor estimated from the permeability
data.

Predicted versus measured

I~,T'

10~

10o

102

q
,,...,_

10~0,

ioe
10 .................................................................

9 9149 pe

e.e~,..~ ................... 10 2

9 9

10 ........................ i.............. ~ ........ i.................................................

i:

10 ~

10 -~ ...................... ..",........................................................................

lOio~,

10 .2

10 ~

10 ~

10 ~

04

Measured permeability (roD)


Fig. 11. Cross plot of the measured permeability and
that predicted using the unified model tested.

lOgl0(k) = - 0.76 + 2.851og10(~b)- 2.921og10(Vsh)

+ 0.01FF.

(19)

Figure 9 shows the cross plot of the predicted


and measured permeabilities on a log-log scale.
The prediction of permeability using the
unified model requires characteristic pore sizes
and aspect ratios for sand-related and clayrelated pores. Applications of the model of Xu &
White (1995a) to published laboratory measurements and well log data indicate that % is
approximately 0.12 and c~c is about 0.03. as and
ac are largely determined by the sizes of sand
grains and clay particles. The best fit as is about
0.25 mm which is within the range of grain sizes
for sandstones (0.0625 mm to 2.0 mm) and the
best fit ac is about 0.00005 mm which is within
the range for shales (less than 0.00309 mm). The
percolation factor Fp for permeability was first

determined from the permeability measurements


and then correlated with Fc, the percolation
factor found from the conductivity measurements (Fig. 10). For this particular dataset, we
obtained the following empirical relationship.
logl0(Fp) = 3.91 + 4.971og10(Fc).

(20)

Figure 10 shows an interesting phenomenon.


For clean sandstone (high permeabilities) the
magnitude of Fp is of the same order as that for
Fc. But as clay content increases Fp decreases
much faster than Fr indicating a much stronger
effect of clay content on Fp than Ft. In other
words, despite the similarity of the concepts of
percolation for conductivity and that for permeability, they can differ in magnitude. The
phenomenon can be related to the way that clay

232

S. XU & R. WHITE

particles affect electrical conductivity and permeability. As we discussed earlier, clay particles
adhering in a pore throat may block the fluid
flow path, whereas for electrical current flow,
wet clay particles are effectively conductive. In
other words, conductivity and permeability
respond differently to clay content.
Equation 20 was applied to the permeability
prediction. The results are shown in Fig. 11,
which should be compared with Figs 8 and 9.
The unified model predicts permeability better
than the KC model and multiple regression. It
benefits from predicting the permeability percolation factor Fp from the percolation factor Fc
determined from conductivity measurements.
Application of the model to another dataset
(Sen et al. 1990) for which no conductivity
measurements were available showed only a
slight improvement over the empirical models
given by Sen et al.

Discussion

low concentrations, clay particles tend to


be dispersed in the sand pores and their
orientation and that of their micro pores is
mainly controlled by the orientation and
geometry of the original sand pores. As a
result, there is initially no obvious increase
in anisotropy with increasing clay content.
Once the sand pores are filled by clay
particles to the extent that the sand grains
become separated, the clay particles become load-bearing and they and their
micro-pores are aligned by overburden
pressure due to their sheet-like nature. This
results in a dramatic increase in anisotropy.
The importance of the effect of clay on
permeability can be seen from the strong
correlation between permeability and clay content. More accurate prediction of permeability
calls for:
(1) a thorough understanding of the major
factors controlling permeability;
(2) developing models of its relationships with
these factors;
(3) a strategy for estimating the key parameters in practical applications.

It is well known that clay content has a major


influence on the elastic and transport properties
of sedimentary rocks but its effects are often
complicated. The following are possible mechanisms.

The following show our concerns on these


issues.

(1) Its effect on porosity. Because of their small


size, clay particles tend to fill the pore space
between sand grains as they are progressively introduced into the system. This
reduces porosity. Once the sand pore space
is filled by clay particles and their microporosity, the sand grains will be suspended
in clay particles. Porosity then starts to
increase with increasing clay content (Marion et al. 1992).
(2) Its effect on pore geometry. The introduction of clay particles reduces pore sizes and,
at the same time, creates micro-pore spaces
between clay particles. This reduction in
pore size affects permeability significantly
more than conductivity.
(3) Its effect on tortuosity. Pore throats are
likely to be bridged or blocked by clay
particles. This increases the tortuosity (or
percolation threshold) of the fluid flow
paths. Again this affects conductivity to a
lesser degree than permeability since wet
clay particles are effectively conductive.
(4) Its effect on pore orientation and anisotropy.
P- and S-wave velocity anisotropies in
sedimentary rocks are strongly correlated
with clay content (Shams et al. 1993). At

sensitive to many factors ranging from


major, through moderate to minor but
these factors are rarely linked to one
another in a physical way--for example,
linking porosity and clay content, grain size
and pore size. To predict permeability
accurately, it is essential to identify the
key factors controlling fluid flow and, if
possible, then relate other factors to the key
ones. Published laboratory measurements
and theoretical studies both suggest that
permeability can be modelled in terms of
the following five key factors: the sizes,
shapes, orientation and interconnectivity of
the gaps, cavities or pore space, and the
porosity. There is scope to include other
known factors, such as clay content, consolidation, cementation, pressure, and so
on, indirectly through the way they modify
the key five factors.
The model described in this paper is a
first attempt at this approach but we have
found no published datasets other than
that of Waxman & Smits on which to test
our model. Further progress in predicting
permeability requires more comprehensive
experiments on representative suites of

(1) It is well known that permeability is

PERMEABILITY PREDICTION
rock samples in order to compare and
calibrate models. Measurements are needed
of conductivity, permeability and elastic
wave velocities, and their anisotropies from
the same set of samples.
(2) There is the question of what kind of model
is best suited to predicting permeability.
Our approach is to look for a physical basis
for the prediction rather than the dubious
practice of playing with variables and
functions in the melting pot of multiple
regression. The Kozeny-Carman model
appears to be the most widely used physical
model of permeability and, when adapted
to use with N M R logs, this can predict
permeability in clean sands very well
(Fletcher et al. 1996). A major assumption
of this model is that the total fluid flow is
the sum of flows in individual tubes. Since
this model, or any parallel flow model, is
dominated by large tubes, it does not
appear to be well suited to modelling
permeability in clay-rich rocks. A model
based on inclusions may also be more
capable of accounting for the well-documented rapid decrease in permeability with
an initial rise in differential pressure than a
tube-based model. This phenomenon is
commonly considered to be due to the
closure of microcracks or flat gaps that act
as channels between big pores.
Our model is an inclusion-based model
which embeds pores into a permeable host
medium and, with the aid of Willis'
formulae, relates permeability to pore
shapes, sizes and orientation distribution.
It tries to integrate permeability and conductivity measurements. Its use of pore
aspect ratios connects the transport properties model to one employed in modelling
elastic wave propagation. In the elastic
wave modelling, pore aspect ratios provide
a way of specifying pore compressibility
but it is far from certain how relevant they
are to modelling transport properties.
Although one can postulate a connection
with pore throat parameters, the connection is admittedly tenuous. However, there
is a benefit in seeking an integrated model
since fitting different measurements helps
constrain its parameters better.
(3) More accurate prediction of permeability
needs detailed information about the pore
space. Porosity itself is not a problem since
it can be measured directly in the laboratory and estimated with a reasonable
accuracy from logs. Since permeability is
sensitive to the second power of pore size,

233

its determination is crucial. Currently there


is no log that provides pore size information on a regular basis although in the
laboratory something closely related can be
obtained by mercury injection or image
processing techniques. However, recent
studies relating the N M R relaxation time
and pore size may change this situation and
it appears that N M R logs can improve
permeability prediction considerably (e.g.
Sen et al. 1990). The tortuosity or percolation of the fluid flow paths is normally
estimated from resistivity measurements
provided porosity is known. As we mentioned above, the tortuosity for electrical
conductivity and that for permeability can
be very different in magnitude. Consequently, tortuosity estimated from conductivity measurements should not be applied
to permeability prediction without calibration.
We use shale volume, a measurable
parameter, as an indicator of pore geometry parameters and as an indicator of
anisotropy. This use of shale volume is
only a first order approximation in the
absence of better measures from logs.
Similarly, discussion of adapting the pore
space specification to take account of
variations from sedimentary environment,
sorting, and cementation is premature until
sufficient calibrating information becomes
available to provide permeability prediction with a sound footing.
Although the application of pore shape
parameters (aspect ratios) estimated from
velocity measurements to the prediction of
transport properties is questionable, the
model does highlight the potential of a
unified physical model to integrate different
measurements and it does simulate the
effect of clay on permeability better than
commonly used alternatives. This is important in view of the abundance of clay
minerals in sedimentary rocks and their
strong influence on permeability.
Another advantage of the unified model over
the commonly used conductivity and permeability models is the capability of predicting
anisotropic rock properties. Anisotropy due to
both aligned minerals and aligned pores can be
modelled using Willis' formulae. The capability
has been demonstrated (Xu & White 1995c) on a
dataset containing porosity, clay content, P- and
S-wave velocities measured at directions parallel
and perpendicular to bedding of 68 sandstone
samples (Shams et aL 1993).

234

S. XU & R. WHITE

Conclusions
(1) There is a strong correlation between clay
content and permeability.
(2) We have d e v e l o p e d a unified effective
m e d i u m model for simulating the electrical
conductivity and permeability of anisotropic shaley formations. The effect of clay is
modelled by means of pore parameters
(size, shape, orientation) and the W a x m a n
& Smits electrochemical model.
(3) The model predicts electrical conductivity
measurements slightly better than the wellk n o w n W a x m a n - S m i t s and the D u a l Water models when it is reduced to the
isotropic case.
(4) Permeability is m o r e difficult to predict
than conductivity and elastic velocities
since it is effected by more factors. In the
case where resistivity measurements were
available, the model simulated the permeability m e a s u r e m e n t s better than other
models tested. In the case where there were
no conductivity measurements, it worked
at least as accurately as existing models.
(5) The percolation factor determined from
c o n d u c t i v i t y m e a s u r e m e n t s is different
from that determined from permeability
measurements in magnitude, probably due
to the different ways in which clays affect
permeability and conductivity. Wet clay
usually acts as a barrier for fluid flow but a
conductor for current flow.
We are indebted to the sponsors of the London
University Research Programme in Seismic Lithology,
Amoco (UK) Exploration Company, Elf UK plc,
Enterprise Oil plc, Fina Exploration Ltd, Mobil North
Sea Ltd and Texaco Britain Ltd, for their support of
this research. We thank B. Moss of Moss Petrophysics
Ltd for helpful comments in his review of the paper.

References
BEARD, D. C. & WEYL, P. K. 1973. Influence of texture
on porosityand permeability of unconsolidated
sand. American Association of Petroleum Geologists Bulletin, 57, 349-369.
Bos, M. R. E. 1982. Prolific dry oil production from
sands with water saturation in excess of 50%--a
study of a dual porosity system. In: SPWLA 23rd
Annual Logging Symposium, paper BB.
CARMAN, P. C. 1956. Flow of gases through porous
media. Academic Press Inc., New York
CLAVIER, C., COATES, G. & DUMANOIR, J. 1984.
Theoretical and experimental bases for the DualWater model for interpretation of shaley sands.
The Society of Petroleum Engineers Journal, 24,

153-168.
DE KUIJPER, A., SANDOR, R. K. J., HOFMAN, J. P.,

KOELMAN,J. M. V. A., HOFSTRA,P. & DEWAAL,J.


A. 1995. Electrical conductivities in oil-bearing
shaley sand accurately described with the SATORI saturation model. In: SPWLA 36th Annual
Logging Symposium, Paper MM.
DE LIMA, O. A. L. 1995. Water saturation and
permeability from resistivity, dielectric, and porosity logs. Geophysics, 60, 1756-1764.
GOODE, P. A. & SEN, P. N. 1988. Charge density and
permeability in clay bearing sandstones. Geophysics, 53, 1610-1612.
FLETCHER,J. D., COWPER,D. R. & HARDWICK,A. 1996.
Analysis of reservoir quality using magnetic
resonance logs for exploration and appraisal west
of Shetlands. In: Expanded Abstracts of the 55th
EAGE Meeting, Amsterdam, Paper E047.
JUHASZ, I. 1979. The central role of Qv and formationwater salinity in the evaluation of shaley formations. In: SPWLA 12th Annual Logging Symposium, paper AA.
MARION, D., NUR, A., YIN, H. & HAN, D. 1992.
Compressional velocity and porosity in sand-clay
mixtures. Geophysics, 57, 554-563.
SEN, P. N., STRALEY,C., KENYON,W. E. & WHITXlNGHAM, M. S. 1990. Surface-to-volume ratio, charge
density, nuclear magnetic relaxation, and permeability in clay-bearing sandstones. Geophysics, 55,
61--69.
SHAMS, M. K., KING, M. S. & WORTHINGTON, M. H.
1993. Whitchester seismic cross-hole test sitepetrophysical studies of cores. In: Expanded
Abstracts of the 55th EAGE Meeting, StanvangeT, Norway.
THOMPSON, L. J. & CALLANAN, M. J. 1981. Overpressured marine sediment, Volume 1 - The
prediction of hydrofracture and k0 during drilling.
Texas A&M University, College Station, Texas,
Project No. RF 3956.
THOMSON,A. 1978. Petrography and diagenesis of the
Hosston sandstone reservoirs at Bassfield, Jefferson Davis County, Mississippi. In: Transactions,
Gulf Coast Association of Geological Societies,
28, 651-664.
WAXMAN, M. H. & SMITS, L. J. M. 1968. Electrical
conductivities in oil-bearing shaley sands. The
Society of Petroleum Journal, 8, 107-122.
WILLIS, J. R. 1977. Bounds and self-consistent
estimates for the overall properties of anisotropic
composite. Journal of Mechanics & Physics of
Solids, 25, 185-202.
Xv, S. & WHITE, R. E. 1995a. A new velocity model for
clay-sand mixtures. Geophysical Prospecting, 43,
91-118.
&
1995b. Poro-elasticity of clastic
rocks: A unified model. In: Transactions of the
36th Annual SPWLA Symposium, Paris, Paper V.
- &- 1995c. Comparison of four schemes
for modelling anisotropic P-wave and S-wave
velocities in sand-shale systems. In: Expanded
Abstracts of the 57th EAEG Meeting, Glasgow,
UK, Paper B2.
- &- 1996a. A physical model for shearwave velocity prediction. Geophysical Prospecting,
44, 687-717.

PERMEABILITY PREDICTION
- -

&- 1996b. Modelling transport properties of anisotropic formations. In: Expanded


Abstracts of 66th SEG Annual Meeting, Denver.
--,
DOORENBOS,J., RAIKES, S. & WHITE, R. E.
1997. A simple but powerful model for simulating
elastic wave velocities in clastic silicate rocks. In:
LOVELL, M. A. &; HARVEY,P. K. (eds) Developments in Petrophysics. Geological Society, London, Special Publications, 122, 87-105.

Appendix A: Willis formulae


Using the Hashin-Shtrikman variational principle, Willis (1977) derived a formula for calculating the generalized Hashin-Shtrikman bounds
for a composite medium containing perfectly
aligned multi-phase inclusions.

L = s crZr[l+ eo(tr--Zo)] -1
r=l

{s

-1

(A1)

r=l
where Cr is the concentration of phase r, Lr is the
property tensor (elastic, conductivity or permeability tensor) of phase r, I is the unit tensor, L0
is the property tensor of a hypothetical host
medium having vanishing volume, and P0 is a
tensor which is a function of pore geometry and
the properties of the host medium. Willis
demonstrated that if the most conductive phase
is chosen as the host medium, one gets the upper
bound for the effective conductivity whereas one
gets the lower bound if the least conductive
phase is chosen. The same applies to elasticity,
permeability and thermal conductivity. For any
physical property, the true response lies between
the Hashin-Shtrikman extremes. In terms of
equation (A1) the porosity enters the response
through the concentrations cr and the pore
geometry through the tensor P0.
For spheroidal inclusions aligned perpendicular to the x3 axis,the composite is transversely
isotropic. The host medium L~ may conseq u e n t l y be specified as d i a g o n a l , with
L~ =L~ . The Po tensor for conductivity or
permeability is also diagonal, with Pll =P22
(Willis 1977), and
A2
(e2L033. . f A + 1 '~
Pl1=2--~101{1-2--~101 a l n L A _ l j }

where

A = (LOl L~
_ _e2L~ 3 ) 89

L=

crAr(O,~)
r=l

crBr (0, fl)

(A5)

r=l

where

Aij) (0,/3
r

= jof27rf ~Wr(O,/3)KimKjnAmn(O,O
sin( O)dOd/3,

B~(O,/3)= f27r f~Wr(O, /3)gim/~nBrn(0,

(A6)

0)

sin( O)dOd/3,

(A7)

Ar(0,0) = Lr[I+ P0(Lr - L0)] -1,

(A8)

B r(0,0) = [I+ P0(Lr -- L0)] -1.

(A9)

and

0 is the angle between global )(3 axis and local x3


axis and /3 is the azimuth, the angle between
global X1 axis and the projection of the local x3
on the global XIOX2 plane. Kij is the matrix
which transforms A r and B r from local coordinates to global co-ordinates:
Kij =

e2A2 { 1
(A+I)
P33= LlO---~ ~Aln\A_I

(A4)

and e is the ratio of long semi-axis to short semiaxis of the inclusions.


One gets the self-consistent approach if L0 in
equation (A1) is replaced by L and Po by P. This
physically means that the effective medium itself
is selected as the host medium.
Willis' formulae are ready to apply to a
composite containing perfectly aligned inclusions. Real pore fabrics are often aligned within
a certain range of directions. The orientation
distribution may be described by, say, a normal
distribution. In this case, we calculate Lr[I+ Po
(Lr - L0)]-1 and [I+ P0 (Lr - Lo)]-1 in a local coordinate system with x3 axis perpendicular to the
inclusions and then transform from local coordinates to global co-ordinates. Equation (A1)
can be rewritten as

(A2)

and

235

cos(0) cos(/3)-sin(/3) sin(0)cos(/3)-1


cos(0) sin(/3) cos(/3) sin(0)sin(/3) /
- sin(0)
0
cos(/3)
J (A10)

}
--1

..

(A3)

Wr(0, /3) is the orientation distribution density

236

S. XU & R. WHITE

function for phase r such that


f

f ~Wr(0,/3)sin(O)dOd/3
=1.

(All)

In our case the resulting effective medium is


transversely isotropic. Thus w r is independent of
azimuth/3 and equations (A6) and (A7) can be
integrated explicitly over /3. This considerably
speeds up the calculations.

Appendix B: The host medium and the


percolation factor
Xu & White (1996b) simulated the electrical
conductivity of clean sands (two-phase medium)
using various effective medium approaches and
the results were compared with the lower and
upper bounds from the generalized HashinShtrikman formulae. In the numerical simulation, pores and sand grains were assumed to be
randomly oriented so that the effective properties are isotropic. It was found that both the selfconsistent and differential effective medium
schemes provided results within the bounds but
departed significantly from Archie's law for
resistivity. This indicates serious problems in
applying these schemes to simulate the effective
properties of a composite with large conductivity or permeability contrasts among the phases.
The problems with these effective medium
models can be avoided by using the concept of a
host medium. A host medium is a starting phase
with no volume, representing a background

connectivity or percolation. Other phases have


no inherent connectivity or percolation and are
embedded in the host phase (de Kuijper et al.
1995). When brine (the most conductive phase)
is selected as the host phase, Willis' approach
gives a parallel-resistor model whereas it gives a
series-resistor model when sand (the least conductive phase) is the host phase. Really the
conducting phases are neither purely parallel nor
purely series but somewhere between the two
extremes. SC uses the effective medium itself as
the host medium and consequently introduces
too high a percolation threshold (de Kuijper et
al. 1995).
For clean sands, we assume that the property
tensor of the host medium LH is a function of
the property tensors of the two phases and a
percolation factor F,
L H = F LF+(I.O--F) Ls

(B1)

where LF and Ls denote the conductivity or


permeability tensors of the fluid phase and solid
phase, respectively. The percolation factor F
describes the degree to which the fluid paths
accord with the parallel model. It is therefore
reasonable to call it a connectivity or tortuosity
factor. F must be in the range of 0.0 to 1.0.
When F is 0.0, the sand is the host phase and
when F is 1.0, the fluid is the host phase. Willis'
formulae apply at these upper and lower
bounds; we apply them in between by selecting
a host medium that matches observations. The
formulae approach Archie's law when F=0.04
~b, where 4 is porosity.

The integration of electrical image logs with core data for improved
sedimentological interpretation
T. M . G O O D A L L 1, N. K. M O L L E R 2 & T. M . R O N N I N G S L A N D

1Rider-French Consulting Cambridge Ltd, at Production Geoscience, North Deeside Road,


Banchory, Kincardineshire AB31 3YR, UK (Present address: Production Geoscience Ltd,
North Deeside Rd, Banchory, Kircardineshire AB31 5YR, UK)
2 Norsk Hydro ASA, P. O. Box 200, N-1321 Stabekk, Norway

Abstract: Electrical borehole images allow the direct integration of sub-surface well-log data
with core data on a detailed visual level. For sedimentary interpretation, electrical borehole
images are primarily used to obtain bedding orientations and to confirm core-derived
sedimentology.
The aims of this paper are two-fold: firstly, to discuss how the sedimentological
information provided by electrical borehole image logs is integrated with that obtained from
other wireline logs and from cores; and secondly, to show that despite the need to integrate
these data electrical borehole images can provide the geologist with unique sedimentological
information which can not be obtained from either the cores or the other wireline logs. In a
case study from the Oseberg Syd oil and gas field, a Fullbore Formation MicroImager (FMI
Mark of Schlumberger) log through a complex interval of shallow marine sediments has
been investigated. The interpretation of the FMI log led to the acquisition of very detailed
orientation data related to the attitudes of sedimentary bedding surfaces. When these data
were integrated with the sedimentary facies, identified from core description, they proved
fundamental for understanding the activity of different shallow marine palaeocurrents
during sediment deposition.

Electrical image logs produce a 'picture' of the


formation which allows geologically-trained
interpreters to identify complex sedimentary
structures within the well bore (e.g. crossbedding, bioturbation etc.). This kind of detailed
geological information was previously only
obtainable from cores. Whilst core provides the
geologist with a physical sample of the formation, electrical image logs provide a unique
dataset which, in particular, can be used to
derive extremely detailed information on the
orientation (dip magnitude and direction) of
sedimentary structures which cut-across the
borehole (Serra 1989). Good quality image logderived orientation data can be superior to that
gathered by any other method, even outcrop
measurements. Acquiring an image log, however, does not preclude the need to take core or
vice versa; these data are complimentary.
In an example from the Middle Jurassic sands
of the Norwegian North Sea in the Oseberg Syd
(South) oil and gas field it is shown that the
integration of orientation data from electrical
borehole image logs allied to both core data and
other open hole logs has greatly enhanced the
sedimentological understanding of the reservoir.

Fig. 1. A schematic diagram to show the differences in


coverage of a 21.59cm borehole by the 4-pad and 8pad electrical imaging tools and by a whole core (after
Adams et al. 1990 and Bourke 1992).
Electrical image acquisition and processing
The electrical imaging tool that is discussed in
this paper is Schlumberger's Fullbore Formation
MicroImager (FMI). The tool consists of four
pads fixed to two orthogonal arms. The four
pads each have a hinged flap to extend the area
of electrical contact. The F M I tool can be run
either in 8-pad mode, using both the pads and
flaps, or in 4-pad (or FMS) mode, where only
the four, main pads are used (Fig. 1).
The raw data are sampled by an array of
button electrodes on the tool's pads and flaps
which, in the case of the FMI, collectively

GOODALL,T. M., MOLLER, N. K. & RONNINGSLAND,T. M. 1998. The integration of electrical image logs 237
with core data for improved sedimentological interpretation In. HARVEY,P. K. & LOVELL,M. A. (eds)
Core-Log Integration, Geological Society, London, Special Publications, 136, 237-248

238

T. M. GOODALL ET AL.

generate 192 microresistivity curves. These


curves form a matrix which is processed to
produce a coherent, colour image. Under
optimum conditions the F M I images can resolve
features in the borehole down to 0.5cm
(Schlumberger 1994), but it is more usual for
F M I images to have a resolution of 2-3cm
(Rider 1996).
The quality of these microresistivity data
collected by electrical image tools is, as with
any other resistivity log, dependant on both the
mud filtrate resistivity and the borehole conditions. Poor electrical images may result from the
borehole being off-gauge, as a result of caving or
borehole breakout. Electrical images can also
contain both acquisition and processing 'artefacts' (Bourke 1989). Care must be taken to
identify image artefacts otherwise there is a
danger that they can be misinterpreted as
geological features.
A more detailed explanation of F M I acquisition and processing can be found in Schlumberger (1994) and Rider (1996).

Fig. 2. Illustration of the difficulties of matching both


whole core and core slabs to borehole images (from
Rider 1996).

Core and electrical images: the datasets


compared
Spatial position
In a 21.59cm diameter borehole the F M I tool
samples the formation from a circumferential
surface area of 67.8cm, whilst in a 31.12cm
diameter borehole the F M I tool samples the
formation from a circumferential surface area of
97.8cm. Whole core in both 21.59cm and
3 1 . 1 2 c m d i a m e t e r b o r e h o l e s is t y p i c a l l y
10.16 cm in diameter, making the circumference
of the core 32 cm (12.6"). Therefore, when direct
visual comparisons are made between electrical
image logs and whole or slabbed core it is
important to appreciate that they are from
different parts of the wellbore (Figs 1 and 2).

Geophysical differences of electrical borehole


images and core photographs
As a consequence of the spatial differences
described above, problems can arise when
comparing core data, particularly core photographs, with electrical borehole images. Spatial
differences between these two sets of data are
often not obvious over intervals with planar
surfaces dipping at low angles relative to the
borehole. Over intervals which have planar
surfaces bisecting the borehole at high angles
(Fig. 2), however, there can be depth offsets of

Fig. 3. The principle of static and dynamic normalisation. Static normalization can be used to compare
images over an entire well. Dynamic normalization is
used to bring out local detail. A full colour scale is used
for limited data range or 'window' which can be from
any chosen interval such as a bed of interest, or a preset small depth range. Note that for the purposes of
this figure the colour scale is represented as blackgrey-white scale (from Rider 1996).
up to 1.8 m due to the differences in sampling
diameter between the two techniques (Adams et
al. 1990).

DATA INTEGRATION FOR IMPROVED SEDIMENTOLOGICAL INTERPRETATION


The comparison between core photographs
and electrical borehole images is inevitable and
often very important. However, it is necessary to
appreciate not only the spatial differences but
also that electrical borehole images are computer
generated, based on measurements of electrical
resistivity from around the borehole wall and
therefore not directly comparable to core
photographs.
Electrical images can be displayed using two
types of colour designation, one where the
colour range covers the entire resistivity (or
conductivity) population for the logged interval
(static normalization) and the other where the
sampled population is limited to the resistivity
values over a specified depth interval (dynamic
normalization) (Fig. 3). The two types of image
display are complimentary for comparison with
core data: dynamic normalization is used for
detailed comparisons of sedimentary structures,
whilst static normalisation is preferred for
correlating lithological or facies changes at
compressed vertical scales (e.g. 1:100 or greater).

Reasons for integrating core and electrical


image data
Feature recognition from image logs
Core provides an indispensable tool for the
calibration (mainly qualitative) of image logs.
Core calibration can be applied in two ways:
firstly, to establish the identity of poorly
resolved, individual features on the image log
(e.g. ripple laminae, different grain-size textures
and bioturbation); and secondly, to determine if
characteristic changes in image texture (on metre
or decimetre scales) correspond to changes in
lithological type or characteristic sedimentary
facies (e.g. ripple-bedded sands or vuggy carbonates). Once calibrated to core these same
features and image textures, when identified
outside the cored interval, may be interpreted
with added confidence.

Depth matching cores with other well data


The image log provides a continuous record of
the cored interval with respect to depth. This
means that the image log can be used to
accurately match the core to log depth and to
identify if there are any intervals where core
material may have been lost or damaged during
the coring process. Once the core has been
corrected to log depth then not only can the core
be used to calibrate the image log but also the

239

other wireline logs. This may be particularly


useful to check if computer generated lithological reconstructions from wireline logs for entire
logged sections are at least accurate within the
cored intervals.

Sedimentary orientation data from cores and


electrical images
In reservoir rocks which contain important
palaeotransport (palaeocurrent/palaeowind) information (e.g. cross-bed orientations) it is
critical that these orientation data are measured
accurately. In order for core material to provide
such information it needs to be correctly
oriented relative to true North. Core orientation, however, can be prone to error (Nelson et
al. 1987), particularly in highly deviated or
horizontal wells (Skopek et al. 1992). If electrical
image logs are available then their orientation
information can be vastly superior and effectively replaces the need to reorient the core.
The process of drilling a borehole through a
cross-bedded, sedimentary formation exposes
these structures in 3-dimensions. The orientation
of such features can be recorded from borehole
image logs and subsequently the image-derived
dip data provide a unique record of internal
bedform geometries. Although the principle of
deriving the orientation of sedimentary structures using image logs is well-known (Serra
1989) the quality and quantity of dip data that
can be obtained from these logs is not always
fully appreciated. Where there is a sufficient
resistivity contrast between cross-bed foresets,
accurate dip measurements can be taken on a
bed-by-bed scale (ca every 5-10cm) and the
bounding surfaces can also be identified. This
may allow the interpreter to identify complex
bounding surface hierarchies in order to reconstruct cross-bed architecture. The level of detail
that image log-derived dip data provide, can
allow interpretations to be made of the depositional processes and subsequently the determination of local palaeocurrent directions and
more regional palaeotransport trends.
On an unwrapped FMI image (from a vertical
well) horizontal sedimentary laminae appear as
flat surfaces, whilst dipping foreset structures
appear as sine waves (Fig. 4). The sine wave
amplitude indicates the dip magnitude, and the
low point of the wave describes the dip direction
(or azimuth). An interpreter can interactively
match a sine wave to the sedimentary structure
shown on the image to derive a very accurate dip
and azimuth value (Adams et al. 1990). Once the
interactively picked dip data have been collected

240

T.M. GOODALL E T AL.

Fig. 4. Representation of borehole wall images on a fiat surface (from a vertical well). The images derived from
the cylindrical borehole (a) are presented on a flat surface (computer screen or hard copy log plot) by
'unwrapping' onto a vertical depth grid and horizontal grid of compass bearings. (b) In this format, horizontal
and vertical surfaces are unchanged but dipping surfaces become represented by a sinusoid. (c) Such dip and
azimuth may be represented on a dipmeter tadpole plot (from Rider 1996).

it is possible to classify each dip according to the


sedimentary feature that it represents. After the
dips have been classified the dip types can be
interrogated separately using statistical analyses
(i.e. eigenvector methods) or they can be
displayed on stereograms or rose azimuth plots.
Interactively picked image log orientation
data are often superior to dip data that can be
derived by any other conventional method:
outcrop measurements, core goniometry and
dipmeter logs (i.e. the Schlumberger Stratigraphic High Resolution Dipmeter Tool or
SHDT log). The reasons are outlined below:

for these data to be interpretable the dips


need to be filtered, identified and classified.
These steps cannot be achieved without
integration of the dipmeter log with core
sedimentology and standard log data.

Case study: electrical image log-derived


sedimentary orientations from the Middle
Jurassic Tarbert Formation, Oseberg Syd oil
and gas field (Norwegian North Sea)
Introduction

(1) Palaeocurrent data from outcrop analogue


studies tends to be limited to a few
orientations where the geologist has confidence in his/her measurements. The character of natural exposures is such that most
sedimentary structures can only be observed, at best, in 2-dimensions. Subsequently it is usually difficult to take a high
number of dip measurements on a density
comparable to image log-derived dip data.
(2) Core goniometry is a technique which
measures the dip magnitude and dip direction of continuous, planar surfaces identified on the outside surface of whole,
oriented core. Although the principle is
sound it is dependant on whether the core
can be correctly oriented and depth corrected. It is also necessary for the outersurface of the core to be smooth and clean
enough to reveal any of its internal
structures. It also has to be sufficiently
coherent to allow measurements to be
taken.
(3) Dipmeter tools record a large amount of
dip data. However, the geological or
sedimentary origin of these recorded dips
cannot be determined in isolation. In order

The Oseberg Syd oil and gas field lies within


Block 30/9 on the Norwegian Continental Shelf,
around 120 km west of Bergen (Fig. 5). Structurally, the region is characterized by elongate
fault blocks which form a series of terraces
between the Horda Platform and the Viking
Graben. The main fault-planes strike N / N N W S/SSE. The reservoir sands are Middle Jurassic
in age and comprise shallow marine sand bodies
from the upper part of the Brent Group (Tarbert
Formation) and fluvio<leltaic, channel sands
from the Upper and Lower Ness Formations
(Fig. 6). For the purposes of this paper, only the
reservoir sands of the Tarbert Formation will be
considered. The distribution of the various
sedimentary facies during the deposition of the
reservoir sands was intimately linked with
localized subsidence associated with tectonic
movement along the regional faults (Fig. 7).
This syn-sedimentary tectonism appears to have
controlled sedimentation during most of the
Middle Jurassic (RavnSs et al. 1997).
The standard open hole logs (gamma ray,
neutron and density logs) and the interpreted
core sedimentology indicate that within the
Oseberg Syd field the Tarbert Formation con-

DATA INTEGRATION FOR IMPROVED SEDIMENTOLOGICAL INTERPRETATION

Fig. 5. Location map for the Oseberg Syd oil and gas field (from Fristad

sists of a number of cleaning-upward successions


(Fig. 8). The changes in shallow marine facies
types suggest that during the deposition of each
succession the water became progressively shallower until there was an abrupt change to deeper
water marking the base of the next, overlying
succession. The sedimentary facies at the base of
each succession are characterized by outer
through inner shelf muds and silts, passing into
sand-dominated shoreface and beach/coastal
plain sediments. These gradationally-based successions are diagnostic of prograding shorefaces,
produced by either wave and storm-dominated
shoreface processes (Fig. 9a), tidally-dominated
shoreface processes (Fig. 9b) or a combination
of both storm and tidally-dominated processes.
Identifying which processes were dominant
during sediment deposition is important because

et al.

241

1997).

it would have had a significant control over the


resulting sand body architecture. The core
sedimentology suggested that the Tarbert sands
were deposited in a shallow marine, shoreface
environment but the evidence was equivocal
regarding which depositional processes were
dominant (Fig. 8). Subsequently, palaeotransport orientation data from electrical image and
dipmeter logs from six wells were studied in
order to assist with the Tarbert sand body
characterization. For the purposes of this paper,
FMI (in 4-pad mode) image-derived orientation
data from only one well will be considered in
detail. In Well 30/9-16 (Figs 5 & 7) the Tarbert
Formation has been cored throughout, allowing
full integration of the FMI image log with the
sedimentary facies and corresponding sedimentary structures seen in the core.

242

T.M. GOODALL ET AL.

Fig. 6. Jurassic stratigraphy of the Norwegian Sector


of the Northern North Sea (after Bowen 1992).

Sedimentary interpretation methods for the


electrical image log-derived orientations
from Oseberg Syd
Palaeotransport (palaeocurrent) indications
within the shallow marine sands of the Tarbert
Formation in Well 30/9-16 were based on
dipping, internal sedimentary surfaces identified
from the electrical images. These included: highangle cross-bed foresets, hummocky cross-stratification (HCS) and low angle cross-bed foresets.
The slabbed core allowed the image based
measurements to be positively identified as
coming from an interval with corresponding
sedimentary structures.
These dip and azimuth data derived from the
electrical images were fully integrated not only
with the core logs but also with the standard well
logs and the orientations were extracted. These
orientation data were then rotated to remove
regional structural dip using the average values
sampled from a structural analysis of orientation
data from representative shale intervals of Middle Jurassic age (in the same well). Structural dip
was carefully calculated and tested several times
before the orientations were finally rotated. This
was necessary because inaccurate structural dip
rotation can produce a false, preferred orientation to the sedimentary orientations. These data
were re-examined in their original sedimentary
position and generally low angle dips, below 5~

Fig. 7. A palaeogeographic model showing the distribution of sedimentary facies during the deposition of
the Tarbert Formation in the vicinity of the Oseberg
Syd field (after Fjellanger et al. 1996).

were filtered out. The low angle dips were


removed because they are not as reliable as the
higher angle dips for deriving cross-bed orientations in sandstones (Cameron et al. 1993). The
resultant azimuth rose diagrams were then
considered to represent the best indication of
cross-bed orientations. Although high-angle foresets were assumed to give the best source of
sedimentary orientation information, the HCS
and low angle foresets also had structures with
consistent dip azimuths. These consistent dip
azimuth values only became evident when
orientation data from individual sand bodies
were analysed in detail.
Examples of the sedimentary palaeotransport
indicators identified from dynamically normalized borehole images in Well 30/9-16 are given
below.

Foresets. Foresets are not common in the Tarbert


sands. They are generally recognized on the
images as having well-marked lamination with

DATA INTEGRATION FOR IMPROVED SEDIMENTOLOGICAL INTERPRETATION


GAMMARAY
API

v~" 1.7 BULKyDENSITY


~
15C ~

[cm3

~
2.:

CORESEDIMENTOLOGY

G R A I N SIZE &

O
NEUTRON
POROSITY ~
60
%
a
-~
. . . . . . . . . . . . . . .

SEDIMENTARYDEPOSITIONAL
STRUCTURESENVIRONMENT
i m u d st vf f m c v c

'

243

--floo~i,,9 v

Swamp

Tidal

: :

,~
9

Lower

l ~ ~'~

..

_.7,0_

T..

9
,

'

. / j / / f

Inlet

'

O~176 f

:: ::--------YJr~/'~

. 2760 -

J . - / /

Shoreface

ii
2770 -

i:::~

[~]

-~

Lower
Shoreface

SEDIMENTARYSTRUCTURES
Cross~
Bioturbation
stratification
~ : ~ Horizontal ~
Rootlets
lamination
Ripple
[ ~ - ] Pebbles
lamination
Waveripple ~
Hummockycrosslamination
stratification

o.)

LITHOLOGY
Coal ~

Sandstone

Silt

Conglomerate

Mud [ ~ ]

Cemented
horizons

Fig. 8. The wireline log responses of the gamma ray and density-neutron combination through part of the Tarbert
Formation in well 30/9-16, Oseberg Syd field. The integration of the core sedimentology demonstrates that the
cleaning-up gamma trends correspond to upward coarsening and upward shallowing of the facies.

10~ to 20 ~ dip angles with unimodal azimuth


variations. The designated image log colour
ranges show little variation within sets but there
are marked colour changes at set boundaries
(Fig. 10a).

Hummocky cross-stratification (HCS) and low


angle foresets. The core reveals that in the
Tarbert Formation, low angle foresets and
HCS are the most common sedimentary structures. The image log of both structures is similar

and shows pervasive, fine lamination. Some


HCS intervals show low dips with random
orientation, whilst others show low dips with a
definite preferred azimuth (Fig. 11). Within the
intervals of HCS, two d o m i n a n t types of
stratification can be distinguished from the
image logs. Firstly, there are the less abundant
higher-angle erosion surfaces within HCS co-sets
and secondly, there are the individual laminae
which usually drape these erosion surfaces (Figs
11 & 12). The erosion surfaces are described as
second-order bounding surfaces whilst the in-

T.M. GOODALL ET AL.

244

B
Outer shelf muds

Outer shelf muds


oOoOQOoOo

Oo o o

Marsh

Coal / Backshore
I

_/~.~..A.~

Beach

///'v".

9' "

~
9 ."

9 .
.

"

Breaker zone ridge and runnel/


rip channels

Tidal inlet

Shoreface

Tidal sand

Tidal channel /

bar complex

".'.',.-,.~
.

~.~

Beach

Jl

. . . .

. . . . . . .

. . .
u

/~

Lower shorefaceinner shelf


transition
Tidal shoal

U~ , ~

Mid-shelf

bioturbated
sandy

Mid-shelf bioturbated
sandy siltstone

siltstone

Outer-shelf bioturbated

5m

mudstone

Outer-shelf bioturbated
mudstone

Fig. 9. The progradation of clastic shorelines leads to distinctive gradationally-based, coarsening-upward


successions. (a) Wave/storm-dominated shorelines are characterized by a gradual shallowing of sedimentary
facies from outer through inner shelf deposits (with abundant HCS) into sand-dominated shoreface and beach
sediments (after Walker & Plint 1992). (b) Tidally-dominated shorelines are also characterized by a gradual
shallowing of sedimentary facies. However, the sands contain tidally-generated sedimentary structures (after
Selley 1985). For key to sedimentology see Fig. 8.

dividual laminae are separated by third-order


bounding surfaces (Cheel & Leckie 1993).

Sedimentary facies interpretation from the


core
The standard open hole logs (gamma ray,
neutron and density logs) and the interpreted
core sedimentology, linked to the electrical
images in Well 30/9-16, were used to identify
the sedimentary facies relationships within the
Tarbert Formation (Figs 8 & 13). In Well 30/916 the Tarbert F o r m a t i o n comprises three
cleaning-upward successions, which have been
divided into the Upper, Middle and Lower
Tarbert Formations. The Upper Tarbert For-

mation and part of the Middle Tarbert Formation are shown in Fig. 8. Cross-stratified
intervals are associated with the upper half of
the coarsening-up sand and silty sand successions. Foreset angles tend to be low (10~ ~
and channelization is rare. The lower part of the
coarsening-up sand and silty sand successions
usually contain ripple laminae and HCS.

Lower Tarbert Formation. (2799.5-2819.0m) the Lower Tarbert Formation in Well 30/9-16
consists of a gradationally based wave-dominated shoreface succession passing from lower
shoreface silts into the wave-rippled, silty sands
of the shoreface and beach sediments. HCS are
absent.

DATA INTEGRATION FOR IMPROVED SEDIMENTOLOGICAL INTERPRETATION

245
~ o,,...~

.,..a

~Z

Q
. ,...~

T.M. GOODALL ET AL.

246

Fig. 11. Dynamically-normalized4-pad FMI image log with interpreted sedimentary orientation data through an
interval of hummocky cross-stratification (HCS). The integration of the schematic core photograph and image
derived sedimentary orientation data demonstrates that second-order surfaces (shown in red) and third-order
surfaces (shown in blue) within HCS can be discerned.

Hummock

2
~'-~

",-...

Sole m a r k s
1 - Third-order

surface

2 - Second-order
3 - First-order

surface

surface

Fig. 12. The form of stratification and first-, secondand third-order bounding surfaces commonly found in
scour and drape hummocky cross-stratified sandstone
beds (from Cheel & Leckie 1993).

consists of thin sands, containing HCS, which


are interbedded with silts. This type of succession, comprising interstratified silts and sharp
based sands displaying HCS, is diagnostic of the
lower shoreface/inner shelf transition (Fig. 9a).
The interval reflects the alternation of fairweather siltstone deposition with storm lain
sands between the fair-weather and storm-wave
bases.

Upper Tarbert Formation. (2718.0-2739.0m)


(Fig. 13)--the Upper Tarbert Formation shows
a gradational transition from lower shoreface
deposits into sand dominated mesotidal shoreface facies which pass upwards into beach and
coastal plain sediments. The lower shoreface
contains intervals of amalgamated HCS sandstones (Leckie & Walker 1982), whilst the
mesotidal shoreface sands contain the higher
angle cross-stratification.

Middle Tarbert Formation (2739.0-2799.5m)-the Middle Tarbert Formation in Well 30/9-16


comprises a coarsening upward, wave-dominated shoreface succession overlain by an
erosive lag deposit marking the base of the
Upper Tarbert Formation. The lower part of the
Middle Tarbert Formation in Well 30/9-16

Interpretation o f electrical image-derived


orientation data
South-easterly orientations are dominant within
the Tarbert Formation and they are mainly
derived from the HCS and low angle foresets

DATA INTEGRATION FOR IMPROVED SEDIMENTOLOGICAL INTERPRETATION

247

Fig. 13. The 4-pad FMI image log with interpreted sedimentary orientation data through the Upper Tarbert
Formation in well 30/9-16, Oseberg Syd field. The integration of the core and image derived sedimentology
demonstrates that the prograding shoreface succession was produced by a combination of both storm- and
tidally-dominated processes. For key to sedimentology see Fig. 8.

(Fig. 13). Although HCS would not be expected


to have preferred orientations recent work has
suggested that some hummocky cross-stratified
sands may indicate preferred orientations related
to unidirectional flow elements being dominant
over the more usual, oscillatory flow during
deposition of these shoreface sediments (Nottvedt & Kreisa 1987; Johnson & Baldwin 1996).
The dip and azimuth of both erosion surfaces
(first and second order surfaces) and laminae
(third order surfaces) within the HCS can be
accurately derived from electrical image logs
(Fig. 11). These dip data from HCS are unique
and cannot be obtained from surface exposures
where the apparent hummocky nature of the
laminae precludes the manual measurement of
dip and azimuth by the geologist.
The Upper Tarbert, shoreface sands in Well
30/9-16 contain cross-bedding (Fig. 13). The
cross-bedding in core did not contain tidallyrelated sedimentary structures. However, imagederived orientation data from this interval
indicate a bimodal dip azimuth oriented N W SE. The opposing palaeocurrent directions are
interpreted to be from cross-stratification produced by both onshore flood (SE) and offshore
ebb (NW) directed currents indicating deposi-

tion within a tidal inlet, which formed part of a


mesotidal shoreface (Fig. 7).

Conclusions
(1) Although the integration of core data with
electrical image logs does lead to improved
sedimentological interpretations of subsurface reservoirs it should be appreciated
that there is not only a distinct difference in
the amount of formation that is sampled by
the imaging tools compared to the core, the
direct comparison of the two sets of data
involves coping with marked physical
differences of depth and spatial position.
(2) In a case study from Oseberg Syd, electrical
image log-derived orientation data from
the Tarbert reservoir sands were integrated
with core data. Some of the sedimentary
facies were difficult to characterize when
they were first studied in core. The subsequent palaeocurrent interpretations derived from the electrical image logs
provided unequivocal evidence for the
correct identification of these sedimentary
facies.
(3) Electrical image-derived dip data cannot

248

T.M. GOODALL ET AL.

only assist in the interpretation of subsurface sediments, they are arguably of


superior quality to orientation data that
can be obtained by any other method. It is
suggested that image-derived dip data may
be used in future studies to provide
important information regarding sedimentary bedform architecture. For example,
the analysis of these orientation data
obtained from cross-stratified units compared to their internal bounding surface
(set boundaries) orientations can provide
important clues for determining the dominant processes that were present during
deposition.
(4) In order to assist with the sub-surface
sedimentological interpretation of electrical
image-derived dip data, it is recommended
that detailed, sedimentary orientation data
should be collected from both well exposed
rocks and from cores of modern sediments.
These analogue, orientation datasets would
provide models to help identify distinctive
dip relationships, which might prove to be
diagnostic of certain sedimentary facies or
depositional processes.

The authors gratefully acknowledge Norsk Hydro


ASA and their partners in Oseberg Syd: Conoco
(Norway) Inc., Mobil Exploration Norway Inc., Saga
Petroleum A/S and Statoil Oil Co. for permission to
use data from well 30/9-16. M. H. Rider offered many
useful discussions in the development of ideas during
the writing of this manuscript. The software PC
ImagePro (BPB Wireline Technologies Ltd) was used
to produce the borehole images in Figs 10, 11 & 13.

References

ADAMS, J. T., BOURKE, L. T. & BUCK, S. G. 1990.


Integrating formation images and cores. Sclumberger Oilfield Review, 2, 52-65.
BOURKE,L. T. 1989. Recognizing artifact images of the
Formation MicroScanner. Society of Professional
Well Log Analyists 30 th Annual Logging Syposium,
Denver, Transactions, Paper WW.
- 1992. Sedimentological borehole image analysis in clastic rocks: a systematic approach to
interpretation. In: HURST, A, GRIFFITHS, C. M. &
WORTHINGTON, P. F. (eds). Geological Applications of Wireline Logs II. Geological Society,
London, Special Publications, 65, 31-42.
BOWEN,J. M. 1992. Exploration of the Brent Province.
In: Morton, A. C., HAZELDINE,R. S., GILES,M. R.
& BROWN, S. (eds) Geology of the Brent Group.
Geological Society, London, Special Publications,
61, 3-14.

CAMERON,G. I. F., COLLINSON,J. D., RIDER, M. H. &


Xu, L. 1993. Analogue dipmeter logs through a
prograding deltaic sandbody. In: ASHTON,M. (ed.)
Advances in Reservoir Geology. Geological Society, London, Special Publications, 69, 195-217.
CHEEL, R. J. & LECKIE,D. A. 1993. Hummocky crossstratification. In: WRIGHT,V. P. (ed.) Sedimentary
Review, 1, 103-122.
FJELLANGER, E., OLSEN, T. R. & RUBINO,J. t . 1996.
Sequence stratigraphy and palaeogeography of
the Middle Jurassic Brent and Vestland deltaic
systems, Northern North Sea. Norsk Geologisk
Tidsskrift, 76, 75-106.
FRISTAD, T., GROTH, A., YIELDING,G. & FREEMAN,B.
1997. Quantative fault seal prediction--a case
study from Oseberg Syd. Hydrocarbon seals importance for exploration and production. Norwegian Petroleum Society (NPF) Special Publication, 7, 107-124.
JOHNSON,H. D. & BALDWIN,C. T. 1996. Shallow clastic
seas. In: READING, H. G. (ed) Sedimentary
Environments: Processes, Facies and Stratigraphy.
Third Edition, Blackwell Science, 265-266.
LECKIE,D. A. & WALKER,R. G. 1982. Storm- and tidedominated shorelines in Late Cretaceous Moosebar-Lower Gates interval--outcrop equivalents
of deep basin gas trap in western Canada. Bulletin
of the American Association of Petroleum Geologists, 66, 138-157.
NELSON,R. A., LENOX,L. C. & WARD, B. J., Jr 1987.
Oriented core: its use, error and uncertainity.
Bulletin of the American Association of Petroleum
Geologists, 71, 357-367.
NflTTVEDT, A. & KREISA, R. D. 1987. Model for the
combined-flow origin of hummocky cross-stratification. Geology, 15, 357-361.
RAvNAs, R., BONDEVIK, K., HELLAND-HANSEN,W.,
LOMO, L., RYSETH, A & STEEL, R. J. 1997.
Sedimentation history as an indicator of rift
initiation and development: the late Bajocian Bathonian evolution of the Oseberg - Brage area,
Northern North Sea. Norsk Geologisk Tidsskrift,
77, 205-232.
RIDER, M. H. 1996. Image logs. The Geological
Interpretation of Well Logs. 2nd Edition, Whittles
Publishing, 199-225.
SCHLUMBERGER, 1994. FMI Fullbore Formation MicroImager. Schlumberger Educational Services.
SELLEr, R. C. 1985. Ancient Sedimentary Environments. 3rd Edition, Chapman & Hall, London.
SERRA. O. 1989. Formation MicroScanner Image
Interpretation. Schlumberger Educational Services.
SKOPEK,R. A., MANN,M. M., JEFFERS, D. & GRIER,S.
P. 1992. Horizontal core acquisition and orientation for formation evaluation. Drilling Engineering, Society of Petroleum Engineers, March, 4754.
WALKER, R. G. & PLINT, A. G. 1992. Wave- and
storm-dominated shallow marine systems. In:
WALKER, R. G. & JAMES, N. P. (eds) Facies
models: response to sea level change. Geological
Assocation of Canada.

How to characterize fractures in reservoirs using borehole and core


images: case studies
D. H A L L E R 1 & F. P O R T U R A S 2

1Elf Petroleum Norge as, P.O. Box 168, N-4001 Stavanger, Norway
2 Western Atlas Logging Services, P.O. Box 953, N-4040 Hafrsfjord, Norway
Abstract: Microconductivity array and acoustic imaging of the borehole wall provide
valuable multidatasets which are used to characterize the geological strata and especially the
reservoirs in exploration activities and during development of producing zones. This paper
presents a tutorial of the main applications and a methodology to follow when performing
fracture interpretation. Borehole image interpretation should not be a routine work
referring to 'a cook book'. It must rely on our geological and structural knowledge and
experience, on basic notions about tool principles and image processing, and on geometrical
calibrations to cores. Vatious examples will be given, showing:
(a) natural fractures; how to distinguish them from drilling-induced fractures;
(b) typical drilling-induced fractures and borehole breakouts; how to identify them;
(c) a tricky case where cemented fractures might be confused with open ones.
The match between interpretation of borehole images and production data appears now
to be the most efficient way to manage fractures in reservoirs.

Fractures may affect reservoir behaviour in a


drastic way. When open, they act as pathways
for hydrocarbon production and may even
transform a very low permeability reservoir into
a highly productive zone. When cemented, they
act as barriers to hydrocarbon flow, hindering
the motion of hydrocarbons toward the well. In
the case of a fault being sealed, either due to clay
smearing, cementation or cataclasis, it can lead
to compartmentalization of the reservoir with
different pressure regimes, water tables or even
fluid types in each individual panel.
Therefore, identification and characterization
of fractures have been always a major concern
for reservoir geologists.
All the information about fractures comes
from wells. Fractures can be either observed
directly on cores or inferred from wireline or
production logs. Fracture detection from wireline logs has always been something speculative,
until the middle of the eighties, when high
resolution borehole images broke through on
the market. They opened a new perspective for
fracture characterization in reservoirs and gave
birth to a new geological discipline: the ability to
identify fractures and to characterize them,
especially with regard to their influence on the
reservoir behaviour. This requires both experience with fractures, based on field work and
rock mechanics, and good understanding of the
physical principles governing the functioning of

the logging tool. In other words, this discipline is


not a simple exercise based on cookery books; it
belongs to the world of geological interpretation, which can always provide tricky cases,
where a quick-look interpretation seems so
obvious but which may be proven to be
definitely wrong.
This will be elaborated on, based on several
case study interpretations of high resolution
borehole images dealing with fractures. All of
them are taken from wells drilled in the
Norwegian Continental Shelf.
Not all the presented cases display phenomena of prime importance for the considered case
study, but they all deal with problems which
might be fundamental in other circumstances.

Fracture characterization using well data


through the past
Fracture identification and characterization was
first conducted on cores, because cores are an
ideal medium to observe fractures, where one
can easely see if they are open or healed with
diagenetic minerals or alternatively smeared or
injected with sedimentary material. It is possible
to measure the open width of the fractures and
sometimes to observe slickenslides on their
planes, which can indicate direction of tectonic
motion. One can try to define fracturation
frequency, based on simple mathematical means.

HALLER,D. & PORTURAS,F. 1998. How to characterize fractures in reservoirs using borehole and
core images: case studies In: HARVEY,P. K. 8z LOVELL,M. A. (eds) Core-Log Integration,
Geological Society, London, Special Publications, 136, 249-259

249

250

D. HALLER & F. PORTURAS

But what remains more difficult to define is


their orientation, since the orientation of the
core itself is not preserved when bringing the
corebarrel to the surface. In the past, a specific
technique has been implemented in order to
determine orientations of the fractures. For each
segment of core pieces which could be matched
together, all geological elements, both bedding
and fractures, were oriented. This was achieved
by 'unrolling' the core surface and plotting this
elements on a stereonet. The re-orientation of
the fractures was obtained by two consecutive
rotations on stereonet, the criteria being to fit
the bedding measured on cores with the bedding
attitude deduced from dipmeter log. This
technique was laborious, rather inaccurate, and
it had its own limitations, when bedding could
not be observed (e.g. massive limestones) or
when bedding is perpendicular to the borehole axis.
In a more general way, studies on cores have
their own limitations too. Most of the time the
reservoir is not cored entirely and by no means
all wells are cored. Moreover, in the case of the
well intersecting a fracture corridor, the core
recovery might be drastically affected by lack of
coherence of the rock. It therefore ends up with
the situation where the zone of highest interest
for fracture analysis is lost for observation.
Besides cores, indirect means have also been
used to identify fractures acting significantly in
the drainage of a reservoir, such as identification
of mud losses, open-hole injectivity tests with
flow-meter and temperature surveys, pressure
transient analysis, all three of which allow
definition of the producing fractures, in the case
of a low permeability matrix ( H a l l e r & Hamon
1993).
However, it has always been evident that some
specific tools were needed to achieve proper
observation of fractures in reservoirs. But
fracture intersection by a borehole being most
of the time less than a centimetre thick, is really
a tough challenge for logging tools, whose

resolution is most of the time above one


centimetre. Besides the array acoustic tools
whose processing of Stonely waves allows
location of fracture zones but not the orientation
of the fractures, it is obvious that only highresolution tools can really identify the elementary fractures. Therefore, the first attempts to
identify fractures were made using conventional,
four, six or eight electrode dipmeter recordings.
Erratic peaks observed on one track with no
lithological continuity on the other tracks were
considered to correspond to fracture intersection
and it was tempting to identify them automatically, the aim being the definition of fracture
intensity. This approach was considered to be
unreliable and anyway did not provide any
orientation.
So, when the borehole images appeared on the
market, it was as if the geologist had gained the
capacity for 'seeing' the fractures inside the
reservoirs. These tools blended the characteristics required: high resolution retative to fracture
size and high sampling density in giving an
image of the borehole wall which allows the
orientation of the fractures to be determined.

Description of the new imaging tools


There are two main types of tool for acquisition
of images, one based on microconductivity
arrays and the other on acoustic imaging
methods, both providing imagery of high
resolution in the order of millimetres. Both
types of tool currently operate on single passes
and are equipped with similar orientation units,
such as accelerometers and magnetometers
which give navigation data for proper fracture
geometrical orientation.
The microconductivity array tools can carry
between 16 and 32 buttons which are mounted
on pad carriers, with 4, 6 or 8 arms. The amount
of data acquired is huge and has influenced
further development of new telemetry and data

Fig. 1. Example of natural open fractures developing in a brittle carbonate cemented layer (example 1). The FMI image
(left-hand) shows mainly two steeply dipping conductive fractures (in black), whose extent stops sharply at the edge of
the carbonate layer. This is typical of a fracture developing in a brittle layer interbedded with less porous brittle rocks.
Interpretation of these fractures as natural open ones is sensible. Unrolled core photograph (right-hand) zoomed on this
carbonate layer proves that this interpretation is correct. One large natural open fracture is observed; in the fracture
plane, calcite crystals have been seen and strong hydrocarbon shows have been reported, indicating that this fractures
acted as hydrocarbon pathways through this tight level.
Fig. 2. Example of natural open fractures developing in a brittle carbonate rock (example 2). The CBIL amplitude image
(left-hand) and core-like display (right-hand) show a dark sinusoidal feature at xx83.4 m (a). The minimum of the
sinusoid shows an easterly dip. The fracture was interpreted as an open one and fits well with drilling losses. Note also a
low reflective (black) fracture (b) subparallel to borehole axis, trending NNE-SSW, interpreted as a drilling-induced one.
The well is deviated and the lowside of the borehole is revealed by the marks of the previous logging tools (c). The CBIL
image was acquired in an oil-based mud.

250

D. HALLER & F. PORTURAS

But what remains more difficult to define is


their orientation, since the orientation of the
core itself is not preserved when bringing the
corebarrel to the surface. In the past, a specific
technique has been implemented in order to
determine orientations of the fractures. For each
segment of core pieces which could be matched
together, all geological elements, both bedding
and fractures, were oriented. This was achieved
by 'unrolling' the core surface and plotting this
elements on a stereonet. The re-orientation of
the fractures was obtained by two consecutive
rotations on stereonet, the criteria being to fit
the bedding measured on cores with the bedding
attitude deduced from dipmeter log. This
technique was laborious, rather inaccurate, and
it had its own limitations, when bedding could
not be observed (e.g. massive limestones) or
when bedding is perpendicular to the borehole axis.
In a more general way, studies on cores have
their own limitations too. Most of the time the
reservoir is not cored entirely and by no means
all wells are cored. Moreover, in the case of the
well intersecting a fracture corridor, the core
recovery might be drastically affected by lack of
coherence of the rock. It therefore ends up with
the situation where the zone of highest interest
for fracture analysis is lost for observation.
Besides cores, indirect means have also been
used to identify fractures acting significantly in
the drainage of a reservoir, such as identification
of mud losses, open-hole injectivity tests with
flow-meter and temperature surveys, pressure
transient analysis, all three of which allow
definition of the producing fractures, in the case
of a low permeability matrix ( H a l l e r & Hamon
1993).
However, it has always been evident that some
specific tools were needed to achieve proper
observation of fractures in reservoirs. But
fracture intersection by a borehole being most
of the time less than a centimetre thick, is really
a tough challenge for logging tools, whose

resolution is most of the time above one


centimetre. Besides the array acoustic tools
whose processing of Stonely waves allows
location of fracture zones but not the orientation
of the fractures, it is obvious that only highresolution tools can really identify the elementary fractures. Therefore, the first attempts to
identify fractures were made using conventional,
four, six or eight electrode dipmeter recordings.
Erratic peaks observed on one track with no
lithological continuity on the other tracks were
considered to correspond to fracture intersection
and it was tempting to identify them automatically, the aim being the definition of fracture
intensity. This approach was considered to be
unreliable and anyway did not provide any
orientation.
So, when the borehole images appeared on the
market, it was as if the geologist had gained the
capacity for 'seeing' the fractures inside the
reservoirs. These tools blended the characteristics required: high resolution retative to fracture
size and high sampling density in giving an
image of the borehole wall which allows the
orientation of the fractures to be determined.

Description of the new imaging tools


There are two main types of tool for acquisition
of images, one based on microconductivity
arrays and the other on acoustic imaging
methods, both providing imagery of high
resolution in the order of millimetres. Both
types of tool currently operate on single passes
and are equipped with similar orientation units,
such as accelerometers and magnetometers
which give navigation data for proper fracture
geometrical orientation.
The microconductivity array tools can carry
between 16 and 32 buttons which are mounted
on pad carriers, with 4, 6 or 8 arms. The amount
of data acquired is huge and has influenced
further development of new telemetry and data

Fig. 1. Example of natural open fractures developing in a brittle carbonate cemented layer (example 1). The FMI image
(left-hand) shows mainly two steeply dipping conductive fractures (in black), whose extent stops sharply at the edge of
the carbonate layer. This is typical of a fracture developing in a brittle layer interbedded with less porous brittle rocks.
Interpretation of these fractures as natural open ones is sensible. Unrolled core photograph (right-hand) zoomed on this
carbonate layer proves that this interpretation is correct. One large natural open fracture is observed; in the fracture
plane, calcite crystals have been seen and strong hydrocarbon shows have been reported, indicating that this fractures
acted as hydrocarbon pathways through this tight level.
Fig. 2. Example of natural open fractures developing in a brittle carbonate rock (example 2). The CBIL amplitude image
(left-hand) and core-like display (right-hand) show a dark sinusoidal feature at xx83.4 m (a). The minimum of the
sinusoid shows an easterly dip. The fracture was interpreted as an open one and fits well with drilling losses. Note also a
low reflective (black) fracture (b) subparallel to borehole axis, trending NNE-SSW, interpreted as a drilling-induced one.
The well is deviated and the lowside of the borehole is revealed by the marks of the previous logging tools (c). The CBIL
image was acquired in an oil-based mud.

252

D. HALLER & F. PORTURAS

transfer systems. The image coverage around the


borehole varies with the bitsize of the drill. The
basic operating principle of the tools consist in
applying an alternating exciting voltage between
the top electrode and the imaging electrodes
located on the arms of the tool. An electric
current proportional to the formation conductivity flows through the formation and is
measured by each electrode (Serra 1989; Safinya
et al. 1991). A few tools have extra built-in
powered standoffs to maximize sensor-to-wall
contact and centralization, particularly when
operating in highly deviated wells.
The acoustic imaging tools are mostly using
rotating transducers and are operating in a
pulse-echo mode, allowing simultaneous acquisition of both amplitude and travel time to the
borehole wall. This scanner rotates from 6 to 12
revolutions per second and provides a full
coverage of the borehole wall, regardless of the
borehole diameter. Furthermore the travel time
image represents an ultra sensitive 360 degrees
borehole caliper. The diameter of the rotating
transducers varies from 0.6 to 5cm. Acoustic
tools have the advantage of operating also in oilbased mud, but they require low to moderate
mud density because the solids in heavy muds
hamper the ultrasonic wave propagation.
A new generation of integrated imaging tools
record the two images in one pass, by simultaneous electrical and acoustic illumination of the
borehole wall. These integrated tools acquire
twice as much information and provide a much
more comprehensive borehole description of the
formation.

Case studies

Example 1. Identification of natural open


fractures on Fullbore Microlmager (FMI)*
The first example is taken from a development
well drilled in the Brent Middle Jurassic
sandstone reservoir located in the Viking Graben. Inside the reservoir, carbonated layers have
been encountered and the question has been
raised concerning their barrier efficiency as far as
production concerned. F M I logging having been
run in this well, the analysis was carried out on
these images, supported by core observation for

some intervals.
On the F M I image (Fig. 1), it is observed that
steep dipping conductive fractures develop inside the carbonated layer. One particular point
to be noticed is that fractures stop sharply at the
limits of the carbonated layer. This recalls field
observations where brittle layers interbedded
within more ductile rocks show the development
of natural fractures, which very often terminate
sharply at the boundary of the brittle layer.
These fractures being conductive, they are
assumed to be open and invaded by drilling
fluid, which is more conductive than the rock
matrix. Based on this observation, interpretation
of these fractures as natural open ones is
sensible.
This interpretation is supported and confirmed by core observation. On the cores,
natural open fractures are observed in front of
these conductive fractures seen on the FMI.
Figure 1 displays an unrolled photograph of the
core surface, which bears strong similarities to
the F M I image, the only difference between
them being the fact that the pictures are not
taken in exactly the same spatial location; the
F M I picture is taken along the borehole wall
plus a certain distance due to electrical penetration (approx. 2.5 cm), while the core picture has
a smaller radius. In the example displayed in
Fig. 1, an open fracture is observed on core,
which stops sharply at the boundary of the
carbonated layer. Inside the fracture plane,
calcite crystallization is present, showing that
this is clearly a natural fracture. Even more
importantly, hydrocarbon shows are observed
along the fracture plane, indicating that these
fractures act as hydrocarbon pathways through
this tight level. As natural open fractures are
observed in several of the carbonate layers, it is
clear that these tight layers are fractured and will
not act as tight barriers during production, but
will only degrade the vertical permeability.

Example 2. Identification of natural open


fractures on Circumferential Borehole Imaging Log (CBIL)**
This example comes from a deviated well and
the image is taken from the overburden (Cretac-

Fig. 3. Drilling-induced features observed on FMI (a, b; example 3) and CBIL (c, d; example 4). Conductive en-bchelon
fractures, observed on FMI in the upper part of 3a, are developed rather extensively along the whole logged reservoir
with a very constant E-W orientation. They are similar to the drilling-induced fractures illustrated on Fig. 4a. A more
continuous vertical fracture, branching at bed boundaries, observed on CBIL 3c, can be interpreted in the same way. The
large conductive (black) bands observed in the lower part of 3a and b, with small scale rock chips, recall breakouts due to
the present-day stress regime, as illustrated on Fig. 4b. They are oriented N-S. On the CBIL image (3d), breakouts
appear as dark bands, due to the borehole ovalization, the ultrasonic tool working as an ultra sensitive caliper.

9, ~

254

D. HALLER & F. PORTURAS

eous limestones). Here a CBIL imager was


logged, because this well was drilled with oilbased mud. Hydrocarbon shows were recorded
in a 20m thick interval and since no core had
been cut, borehole imaging was the only way for
investigating the origin of the shows.
The picture presented on Fig. 2 displays a
natural fracture revealed by a dark sinusoidal
feature, due to low acoustic impedance along the
fracture. This open fracture fits well with drilling
losses, recorded at this depth. Therefore it is
believed that the hydrocarbon flows through
open fractures, the matrix being almost tight.

Example 3. Evidences of drilling-induced


features on FMI image
The emergence of high resolution borehole
imagers has led to the awareness of the effects
of drilling on the borehole wall integrity, which
previously were either ignored or poorly described. Identification of drilling-induced fractures is a vital challenge, because they can be
mistaken for natural fractures and lead to a
completely false estimation of the reservoir
potential.
The given example is taken from a wildcat
drilled in the Viking Graben, with the Brent
Middle Jurassic sandstone as a primary objective. An F M I image was acquired in this
interval.
O n t h e FMI picture, conductive fractures are
extensively developed, with a constant orientation. They often show a typical en-&helon
pattern (Fig. 3a), which recalls drilling-induced
fractures described as 'petal fractures'. This type
of fracture has been reported in cores taken in
quartzites of Alberta, Canada (Fig. 4a), where
fractures seen at the surface of the core do not
fully penetrate it, demonstrating that they are
induced by coring. It is commonly considered
that these drilling-induced fractures develop
when the hydrostatic pressure within the wellbore exceeds the hoop stresses around the hole
and that they occur in the direction of the
maximum normal stress (Lehne & Aadnoy
1992).
Another type of drilling-induced features is
also present in this example. On Fig. 3a below
xxl0 and on Fig. 3b, borehole breakouts are
clearly observed. They are revealed by reciprocal
(180 degrees) conductive stripes, due to the
accumulation of drilling mud in the extremities
of the ovalized borehole.
This phenomenon of breakouts has been
studied by rock mechanics experiments under
anisotropic stress conditions (Fig. 4b), and it has

been shown that these features develop along the


direction of minimum normal stress.
So the orientations of drilling-induced fractures and borehole breakouts are directly controlled by the in situ stress regime, with an
orthogonal relationship (Fig. 4c).

Example 4. Evidence of drilling-induced


features on a CBIL image
This example is from the same geographical area
as example 3. Both drilling-induced fractures
(Fig. 3c) and borehole breakouts (Fig. 3d) are
observed. Drilling-induced fractures here present a continuous shape, which is more classic,
and their orientation is consistent with the one
observed in the previous example. Borehole
breakouts appear as dark bands, due to borehole
ovalization, the tool working as an ultra
sensitive caliper.

Example 5. Example of fracture identification on a FMI Image


The following example illustrates the problems
that an interpreter can face. It is taken from a
deep exploration well drilled in the Viking
Graben with the Brent Middle Jurassic sandstones as target.
First, the left hand image displays large
fractures subparallel to the borehole axis (Fig.
5). Their interpretation as natural cemented
fractures is unambiguous, since these fractures
display resistive (white) traces. More equivocal is
the conductive (black) halo observed at each top
and bottom of the sine wave defining the
fracture plane, but this feature is in fact typical
of cemented fractures.
Secondly, the right-hand side image deals with
conductive fractures, whose interpretation is
more difficult. These fractures could have been
considered as natural open ones, which may
represent a major challenge for enhanced
production, since the gas-bearing reservoir was
found tight. Nevertheless they were interpreted
as drilling-induced fractures, relying on their
morphology and on the fact that they present a
constant orientation, parallel to the direction of
the maximum normal stress, as deduced from
other wells in the same area.
A drill-stem test was conducted in front of this
interval, but did not flow, which tends to prove
that the interpretation is correct.

Example 6. Clay smear&g associated with a


normal fault observed on a CBIL image
This example is taken from an exploration well

HOW TO CHARACTERIZE FRACTURES IN RESERVOIRS

255

Fig. 5. Example of fracture interpretation based on FMI image (example 5). The left-hand picture reveals natural
cemented fractures. They appear as resistive (white) features, with a typical conductive halo at the tops and
bottoms of the sine wave figuring the fracture plane. The right-hand picture exhibits conductive (black) fractures.
They are interpreted as drilling-induced ones since they have a consistent orientation, parallel to the direction of
maximal normal stress, which is deduced from other wells of the same area (like the one shown in Fig. 3a,b).

drilled offshore Mid-Norway, with a watersediments show a fault drag pattern.


based mud of moderate density.
Notice that the edge enhanced CBIL image
(righ-hand) shows all events as dark sinusoids
On the CBIL image (Fig. 6), is a low acoustic
impedance (dark) feature with high dip at , regardless of their nature. Edge enhancement is
xx33.5m, cutting through the sandstone. This currently used as a complementary interpretais interpreted as a normal fault. Note the 25 cm
tion technique.
thick dark infill of the fault, which is believed to
be the clay smearing, a heterogeneity which
Example 7. Equivocal resistive response of
might control transmissibility of the fault. The
enhanced CBIL image (right-hand) assists in
cemented fractures on a F M I image
highlighting the fault plane. The arrow-plot
This example is taken from a production well of
(left-hand track) shows the results of the CBIL
a field located in the Viking Graben, with gas
interpretation, where the fault plane and assoreservoir in the Brent Middle Jurassic sandciated fractures have a consistent direction.
Furthermore, the tadpoles in the overlying stones. The selected image is a dynamic normal-

256

D. H A L L E R & F. PORTURAS

Fig. 6. Example of clay smearing associated with a normal fault (example 6). The CBIL image shows a steep
dipping fault at xx33.5 m cutting through the sandstones. The fault appears to be wide and of low acoustic
impedance (dark) due to a 25 cm thick clay smear of the fault plane. The edge enhanced image (right-hand side)
highlights this fault plane. The arrow plot (left-hand track) resulting from the interpretation shows the orientation
of the fault and associated fractures and displays a drag-fault geometry within the above-lying series.

~ '4 "~4

--~

r~

, ..~ ,.~ ~

~...~ ~ ~ . ~

: ~:~ ..~

.g~'n~o~o

I~

~l . ~ .,~ ~

;~

"~n'~ ~

"~

-~--~,.~

~.'~ 0

~"

~~-'~ ,.~1

~o

~
~

~ .~

258

D. HALLER & F. PORTURAS

Fig. 9. Example of a CBIL image showing the benefits of having fullbore coverage available as opposed to
conventional dipmeter logs (example 9). The arrow plot alone shows an apparent drag which might be interpreted
as being associated with a fault. Both dynamic normalized and edge enhanced images instead show a sedimentary
structure. Furthermore, notice truncation surfaces at xx42.5 m, and very thin laminated beds. Vertical light stripes
are due to marks produced during previous logging operations and also by the metallic blades from the upper
centralizer of the dipmeter tool, which were scraping the mudcake.

HOW TO CHARACTERIZE FRACTURES IN RESERVOIRS


ized display and shows fractures in the reservoir
a r o u n d the g a s - w a t e r c o n t a c t located at
xx00.2m (Fig. 7). In the gas-bearing section,
fractures appear conductive and might be
mistaken for open fractures. In the water-wet
section, they show a resistive facies indicative of
cemented fractures. One core taken in the gasbearing zone revealed silica cemented fractures.
So the apparently conductive fractures observed in the gas zone are in fact cemented
fractures. Their misleading conductive facies is
due to the fact that the image is dynamically
normalized; the fractures are in fact resistive, but
the matrix is even more resistive due to the very
high gas content. Conversely, on a static
normalized image, the whole gas-bearing zone
appears blank.
This example argues for a combined use of
dynamic normalized images, where the best
contrast is present, and static normalized
images, where the exact resistivity of the
geological features can be ascertained.

Example 8. Example o f a fault identified on


a F M I image
This example is taken from the same production
well as example 7. This well crosses a fault,
cutting away about 80 m of the uppermost
reservoir section. The FMI image was utilized in
order to identify and orientate this fault.
On the dynamic normalized image (Fig. 8,
left-hand), the identification of the fault is
ambiguous. It is crystal clear on the static
normalized image (Fig. 8, right-hand), which
allows definition of the strike and dip of the fault
plane. This is due to the fact that the static
normalized display captures the strong resistivity
contrasts between the conductive shales and the
very resistive gas-bearing sandstones.
The well was sidetracked as a consequence of
this interpretation, avoiding this fault, and it
found a complete reservoir section.

Example 9. Benefits of having images with


fullbore coverage
Conventional dipmeter analysis alone can be
rather ambiguous especially when based only on
computed tadpoles. If interpreting only the

259

arrow plot (Fig. 9, left-hand track), the tadpoles


with decreasing dip at xx43 m could be interpreted as drag fold associated with a normal
fault. Full coverage CBIL image (Fig. 9, center)
refutes this, showing clearly that the dip pattern
reflects a sedimentary structure. Interpretation is
made easier by the use of the edge enhancement
display. The reliability of a dipmeter interpretation is greatly enhanced by the use of borehole
images.

Conclusions
The use of high resolution borehole imagers has
led to a revolution for reservoir geologists, who
now have an insight into the reservoir. It is now
possible for them to discriminate between
natural open fractures, cemented ones, and
drilling-induced features. The main improvement is that it is now easy to orientate all these
geological features. As a matter of fact, matching borehole image interpretations with production data appears to be the most efficient way to
manage fractures in reservoirs.
It is a pleasure to thank Elf Petroleum Norge and
Statoil for allowing the borehole images to be used.
The use of * throughout denotes a Mark of
Schlumberger and ** denotes a Mark of Western
Atlas Logging Services.

References
HALLER, D. & HAMON,G. 1993. Meillon-Saint Faust
gas field, Aquitaine basin: structural re-evaluation
aids understanding of water invasion. In: PARKER,
J. R. (ed.) Petroleum Geology of NW Europe,
Proceedings of the 4'h Conference. Geological
Society, London, 1519-1526.
LEHNE, K. A. & AADNOY, B. S. 1992. Quantitative
analysis of stress regimes and fractures from logs
and drilling records of the North Sea Chalk Field.
The Log Analyst, 33, 351-361.
SAFINYA, K. A. LE LAN, P., VILLEGAS,M. CHEUNG,
P. S. 1991. Improved formation imaging with
extended micro electrical arrays. Society of
Petroleum Engineers, Special Paper 22726.
SERRA, 0. 1989. Formation MicroScanner image
interpretation. Schlumberger Educational Services, Houston Texas.

Measurement scale and formation heterogeneity: effects on the


integration of resistivity data
P. D. J A C K S O N l, P. K. H A R V E Y 2, M. A. L O V E L L 2, D. A. G U N N 1, C. G.
W I L L I A M S 2 & R. C. F L I N T 1

1British Geological Survey, Keyworth, Nottingham NG12, UK 5GG


2 University of Leicester, University Road, Leicester LE1 7RH, UK
Abstract: Core and downhole logging resistivity data gathered during Leg 133 of the Ocean

Drilling Program are used to illustrate the wide range of scales of resistivity data available
for reservoir characterization. The differences in scale and sampling interval between
quantitative log resistivity data and conventional core plug data is shown to be central to
reconciling these two datasets. Resistivity images of fine scale sedimentary structures taken
on half-round cores are presented (at the same resolution as the downhole borehole wall
imaging tools) and these fine structures are shown to be 'lost' if investigated using
conventional core plugs and downhole resistivity logging tools. The limitations of
conventional measurements on core plugs are presented and contrasted with the benefits
of logging all of the core in the laboratory at a resolution comparable to the borehole wall
imaging tools. An example of integrating different scales of resistivity data using a modelling
approach is presented and is shown to be applicable to both core and log data. Visualizing
and comparing the scale content of different resistivity datasets has been achieved in an
intuitive way using a spectral method which illustrates the 'data gap' in quantitative
resistivities which exists between core and log data.
Fine scale sedimentary structure is now accepted
to be an important control when considering
hydrocarbon reservoir modelling. Furthermore,
the difficulty of quantifying heterogeneity over
scales from mm to km is well known. While
geological processes occur seamlessly over scales
ranging from pores to sedimentary basins,
difficulty is experienced in reconciling log and
core data over the mm to m scales.
This paper presents real data over a range of
scales and highlights some problems encountered when moving from one scale to another,
with particular reference to reconciling traditional core-plug resistivity measurements with
higher resolution, similar data available both in
the laboratory and downhole.
In the direct current approximation, electrical
resistivity measurements are scale independent
and are increasingly being used as the method of
choice for identifying different scales of heterogeneity. The development of the vast majority of
downhole resistivity measuring technology has
been in response to the needs of the oil industry,
driven by the knowledge that electrical resistivity
can be used as a primary estimator of the oilsaturation of reservoir rocks. Consequently, a
wide range of resistivity logging tools which are
sensitive to a wide range of scales of heterogeneity already exists. Resistivity variations are
identified routinely over differing distances from
the borehole, into the formation, using suites of

borehole logging tools having various resolutions.


In addition to the degree of oil-saturation,
electrical resistivity is sensitive to the total
amount of fluids in the rock, how the fluids are
distributed, the resistivities of the fluid phases
and the rock matrix. Thus electrical resistivity is
sensitive to sedimentological structures, lithology and pore-fluid composition (Archie 1942).
The vertical resolution and depth of investigation of a suite of standard downhole logging
tools is shown in Fig. 1. The act of drilling a
borehole disturbs the electrical resistivity of the
formation and the well-bore, and historically,
different logging tools have been developed to
assess different radial zones around the borehole. Ideally, constant resolution at different
depths of investigation is required to assess
radial changes in resistivity, due to invasion by
fluids from the borehole.
A general trend can be seen in Fig. 1, which is
common to resistivity measurements in general;
that increasing depth of investigation is associated with decreasing power of resolution. For
example, in Fig. 1 the Formation MicroScanner
(FMS) has the greatest resolving power (vertical
resolution of 0.5 cm) at a depth of investigation
of 2 cm, while the deep induction tool (ILD) has
the least resolving power (vertical resolution of
1.5 m) at a depth of investigation ranging from
1.25 to 3.75 cm. Therefore it is clear that these

JACKSON,P. D., HARVEY,P. K., LOVELL,M. A., GUNN,D. A., WILLIAMS,C. G. & FLINT,R. C. 1998.
Measurement scale and formation heterogeneity: effects on the integration of resistivity data
In. HARVEY,P. K. ~z LOVELL,M. A. (eds) Core-LogIntegration, Geological Society, London,
Special Publications, 136, 261-272

261

262

P. D. JACKSON

ET AL.

Resistivity L o g g i n g tools
Depth of investigation vs. vertical resolution

ILD

lOOO

ii|i

lOO

iIIiipll.,lm!

.E

gr=llglll~L.Z~-.'_

|~||-';:,~----I
I I r

SFL

Dual Laterolog

dlg-]r 82

~---- LLD

IIMIIKil

lO

low Rt
high Rt

LLS

ARI (azimuthal LLS)

>

MSFL (low:high Rxo)

Microlaterolog
~

o.1
0.1

Core plug (38 x 80 mm)

1000~ ' ' - ~ Microlog

Depth of investigation, in

FMS

(after Schlumberger 1989)

Fig. 1. A wide range of resolutions and depths of investigations are available for routine resistivity logging.
I

Plan/Isometric.
c, I S

c2
Multi-electrode (64)

Usable Data Region.


I
I

'

~ " ~
t

Computer.

I
I

k..VJ

;
I

c1,1-51

!c ,i-5

->

<:

r ~

J ~

H Multiplexor& I
Conditioning.

Electrode pad.
Multiplexor.

;:' To computer.

lililliillillil
Fig. 2. Schematic diagram of technology used for micro-resistivity imaging of core (after Jackson
two extreme cases (FMS and ILD) 'sample' very
different portions of a reservoir, making quantitative comparisons unsafe if 3-D heterogeneity is
suspected, or if radial invasion processes cannot
be allowed for. In addition, the 'style' of current
flow between the two tools is diverse, making
their responses to the same heterogeneity very

et al.

1990).

different (e.g. anisotropy resulting from thin


layers having alternating, contrasting resistivities). These downhole logs are 2-D and l-D,
respectively, and 'visually-suggest' erroneous
radial continuity.
Core from the borehole itself samples a
volume comparable to that of the pad-tool

MEASUREMENT SCALE AND FORMATION HETEROGENEITY

Fig. 3. Map of Northeast Australian margin showing Leg 133 drill sites (after Davies et

'micrologs' (Microlog, Microlaterolog, Micro


Spherically Focused Log); however, these padtool measurements sample outwards from the
borehole wall, while the core samples some
distance 'inside' the borehole wall. In Fig. 1, the
FMS can be seen to 'sample' an even smaller
volume of the formation. Thus, with downhole
pad-tools that sample a volume of the formation
comparable to that of the borehole, there is still
room for discrepancy between core and log data
even when measured at identical scales and
resolutions. For example in Fig. 1, while the
'Microlaterolog' and a 'core-plug' measurements
have comparable resolutions and depths of
investigations, they cannot sample the same
piece of the formation.
Laboratory measuring techniques have been
developed which have a resolution similar to the
borehole wall imaging tools (e.g. FMS in Fig. 1);
one such technique is outlined in Fig. 2. This
technique enables the electrical resistivity of fine
scale geological structures to be assessed (for the
first time) on core in a way which enables direct
comparison with downhole images. This technique assesses resistivity quantitatively and unlike

263

al.

1991).

the FMS does not require empirical calibration.


The technique has a resolution of 5 mm, unlike
standard core plug measurements which are
'averaged' over volumes that can be larger than
important small scale sedimentary structures
(i.e. contravening statistical sampling practice).
The aim of this paper is to study examples of
heterogeneity at a variety of scales, using core
and log data, to demonstrate how data may be
presented in order to preserve the scale information.

Core and downhole logging data from the


Ocean Drilling Program
The Ocean Drilling Program (ODP) offers
scientists the opportunity to study geological
processes through access to both core and
downhole measurements. Continuous core is
collected for laboratory analysis and these
measurements are complemented by an extensive suite of logs, which have been selected to
suit the widest range of geological environments.
The suite of resistivity logs available to ODP are

264

P. D. JACKSON E T AL.

the dual-laterolog/induction for greatest depth


of investigation, the spherically focused log
(SFL) and a 'slimline' version of the FMS. The
logging tools are passed out through the drill bit
into the borehole, which itself has been drilled
with sea-water as the circulating fluid. These
arrangements, while resulting in variable hole
conditions, create little invasion in typical
situations. Thus the resistivity logging data does
not require 'mud corrections' but can be affected
by very variable hole diameters.
Leg 133 of the Ocean Drilling Program
(Davies et al. 1991) drilled the northeastern
Australian margin with the objective of studying
sedimentary responses to global changes in sealevel within the last 10-20Ma, with particular
reference to palaeo-climate, oceanography and
the evolution of carbonate platforms.
Site 823

Site 823 is situated in the centre of the Queensland Trough ENE of Cairns as shown in Figs 3
and 4. The sediments are a sequence of clastic
turbidites within hemipelagic carbonates. The
turbidites are shelf sediments derived from the
Australian sub-continent. The bathymetry dis- Fig. 5. Photograph of a sandy turbidite seen in core
played in Fig. 3 can be seen to indicate steep from Site 823.
shelf slopes which will have facilitated the

Fig. 4. Marion plateau showing the positions ODP


sites 815 and 823 (after Davies et al. 1991).

initiation of turbidity currents. The sediments


were collected using the Advanced Piston Corer,
developed by the Ocean Drilling Program for
sampling soft sediments through the drill string.
The sediments are inter-bedded sands and muds.
The sands are dark grey in colour and fine
upward, while the carbonate muds are bioturbated nanno oozes containing forams and
bioclasts, and are grey in colour.
Micro-resistivity core imaging (Jackson et al.
1990) was undertaken on a variety of turbidites
from Site 823 (Jackson et al. 1991), one of which
is displayed in Fig. 5. Here the visually bland
carbonate sediments can be seen bounding the
turbidite. Complex laminae can be seen which
are typical of clastic turbidite deposits. The most
striking feature to the naked eye is what appears
to be a zone of more open pore structure at
97.5cm in Fig. 5. The micro-resistivity image,
and corresponding micro-log (averaged at constant depth from the core image) can be seen in
Fig. 6. The extent of the turbidite is well defined,
as would be expected for a change in lithology
from fine-grained hemipelagic carbonate sediments to a laminated sand. Fine scale 'laminae'
are very prominent in the micro-resistivity image
of the turbidite, particularly near its base. Many
laminae, invisible to the naked eye, are promi-

MEASUREMENT SCALE AND FORMATION HETEROGENEITY

265

Fig. 6. Microresistivity data corresponding to the core in Fig. 5 in image and log format. The log illustrates the
rapid changes in resistivity that can occur within the volume of a conventional core plug.

nent features in the micro-resistivity image.


There is a general trend of increasing resistivity
with depth within the turbidite, which is
consistent with 'fining upward' which is typical
of these deposits. The laminae in the uppermost
half of the turbidite are less well pronounced and
show evidence of bioturbation. The lower
portion of the turbidite appears to be strongly
laminated (25 and 110 mm) while the remaining
upper part (110 to 190mm) could be described
as being more disturbed, having poor lateral
continuity at constant depth.

Such features are consistent with known


turbidite facies and the trends seen in resistivity
core image can be mapped to the classic 'Bouma
sequence' used in the classification of turbidites
(Bouma 1962; Lingen 1969). The micro-resistivity core data provide substantial additional
information regarding small scale laminae that
are invisible to the naked eye. As these data are
quantitative they could be used in petrophysical
calculation schemes (e.g. construct permeability
predictors applicable to each lamina).
The extent of a standard core plug is shown in

266

P. D. JACKSON E T AL.

Fig. 7. Photograph of a 0.5m of clayey nanno ooze core obtained from Site 815.

Fig. 6, and illustrates that a multitude of


separate individual lamina may be contained
within a single core plug. Consequently, standard core plugs provide 'averages' of fine scale
structures which are unrepresentative because no
sediment exists with these 'average' properties
and simple volume averaging is unsafe because it
is not consistent with the physics of the
measurement.
Site 815

Sites such as 815 were chosen in pure carbonates


in order to sample sediments where the climatic
and oceanographic signatures are likely to have
the best chance of preservation. Site 815 is
situated at the southern margin of the Townsville Trough, as shown in Figs 3 and 4. Hole
815A was drilled and sampled to a depth of
474m below sea floor (mbsf) through a 416 m
thick package of hemipelagic carbonate sediments of Miocene-Pleistocene age which overlie
Miocene (Lower to Middle) shelf carbonates.
A sedimentary unit (II) was identified, extending from 72 to 280 mbsf, which is described by
Davies et al. (1991) as an expanded section of
greenish-grey to grey nanno-fossil sediments,
ranging from slightly bioturbated oozes to

unlithified mixed sediments. The section of the


hole within a deeper sub-unit (liB) extending
fi'om 111 to 280 mbsf, characterized by partial
lithification, was the first section in which the
hole conditions were stable enough to allow
logging. Repetitive colour changes were seen in
the core of this sub-unit associated with rates of
deposition as high as 38.5 cm ka 1. Davies et al.
(1991) suggest this cyclic colour change was
controlled by variations in both grain size and
carbonate content, the latter varying between
10% and 60%.
Micro-resistivity core imaging (Jackson et al.
1990) was undertaken on two 0.25 11"1sections of
core from a depth of 201 mbsf. As this interval
had been logged downhole by the SFL/induction
suite of resistivity logs operated by ODP, we
have datasets covering core and downhole over
scales from 5mm to 2m. The two sections of
core used for micro-resistivity imaging are
shown in Fig. 7; the marks made by the
electrodes are clearly visible, as is the general
lack of visual sedimentary features, which is
typical of this sub-unit.
The results of micro-resistivity measurements
on core in the interval 201.7 to 202.3 mbsf can be
seen in Figs 8 and 9. The single-trace log plot is a
single average of core micro-resistivity values at

MEASUREMENT SCALE AND FORMATION HETEROGENEITY

267

Fig. 8. Microresistivity data corresponding to the core in Fig. 7 in image and log format, showing marked vertical
and lateral variability.

constant depth, while the micro-resistivity


images can be seen to depict lateral heterogeneity in addition to vertical variability. The data
are converted to Formation Factors (Archie
1942) where the sediment resistivity has been
normalized with respect to the resistivity of the
pore fluid (i.e. seawater).
The dimensions of a standard core plug
(38x80mm) have been superimposed on the
single-trace log, showing that substantial heterogeneity occurs within this volume. Consequently, in these sediments, one can again see
that traditional core plug resistivity measurements are both 'unknown' averages of finer scale
structures, and that small changes in the location
of the core plug can lead to substantial changes
in the measured value. The averaging process
relating to these core plug measurements is
'unknown' because a measurement designed for

homogeneous material is being applied to


cylinders of rock which are heterogeneous, often
in 3-D, as in the above case (Lovell et al. 1994).
In order to reconcile such 'averaged' measurements with the micro-resistivity images, simulating the core plug measurement on the basis of
the micro-resistivity image results would be
necessary. A simple average is not suitable
because the distribution of resistivity variability
controls the measured value, in addition to the
'simple' average resistivity. For example, if there
were fine scale horizontal laminae having alternating resistivity values, the lower values would
dominate when the plug was 'drilled' horizontally (as is the case in Figs 8 and 9) and
conversely the higher values would dominate if
the core was 'drilled' vertically (assuming uniform current flow through the core plug using
plate-electrodes on each cylindrical face and

268

P. D. JACKSON E T AL.

Fig. 9. Microresistivity data corresponding to the core in Fig. 7 in image and log format, showing similar
variability to Fig. 8 but different 'average' value.

separate independent potential measuring electrodes).


The corresponding downhole resistivity logs
(SFL & ILD) are shown in Fig. 10 (Jackson &
Jarrard 1993) where the SFL log can be seen to
contain far more small scale features than the
corresponding ILD log. The 'Rt' log in Fig. 10,
uses a modelling approach (Jackson & Jarrard
1993) to combine the resolution of the SFL with
the deeper penetration of the ILD. This log has
been used as the basis of the 'SFL Formation
Factor' shown in Fig. 11.
Figure 11 displays the micro-resistivity core
data along side the downhole log data, and
illustrates the wide range of scales of resistivity
measurements that are characteristic of one
sedimentological unit. The resolution of the
SFL & ILD are superimposed (from Fig. 1)
and can be seen to be greater than the total

extent of the micro-resistivity images, and huge


compared to the volume sampled by the
standard core plug (a single SFL measurement
'samples' 10000 times more formation than a
core plug). Thus, while the sampling intervals of
the downhole logs are far smaller than their
resolution (i.e. overlap and sample all the
formation adjacent to the borehole), those of
standard core plugs can be seen to be so sparse
as to be unrepresentative, and contravene
sampling practice (e.g. sampling at twice the
max. 'frequency' is required). In addition, on the
other hand, the micro-resistivity core data shows
the core plug data to be unrepresentative of mm
scale heterogeneity such as the fine scale laminae
described above. These points are further illustrated in Fig. 12 where the micro-resistivity core
data, the SFL and the ILD are displayed at the
same scale.

MEASUREMENT SCALE AND FORMATION HETEROGENEITY

269

Fig. 10. Site 815 downhole resistivity logs showing the corrections and errors associated with the calculation of

SFL core (after Jackson & Jarrard 1993).

Visualizing scales of heterogeneity using a


spectral method
A resistivity log may be considered to be a series
of anomalies which can be described in terms of
spatial frequencies using Fourier techniques, the
inverse of the 'wavelength' of an anomaly being
its spatial frequency. The relative 'weights' of
each anomaly can be added in the frequency
domain, generating a Fourier series of frequencies which can be used to re-construct the
original resistivity log. While Fourier theory
gives us a single step function containing a wide
range of frequencies, the resolving power of the
measurement systems act as 'high cut filters',
defining a minimum wavelength for each resolution. While digital filtering and the use of
correlation lengths can be used, the spectral
method provides a means of visualizing scalerich resistivity logs which are intuitive, and
whose strengths and weaknesses can be assessed
knowing the basic principles of Fourier analysis
and sampling practice.

An example of such a display is presented in


Fig. 13 where core and downhole data from
ODP Site 815 are displayed along side an
example of the trace from a single button from
an FMS image log. The characteristic wavelengths of the core data can be seen to be far
smaller than the SFL log, while the FMS data
appear to occupy an intermediate position. Thus
there is a substantial gap in scale information
between the core and log data in Fig. 13 which
would be bridged by successive measurements at
high resolution (e.g. 'FMS type' measurement
displaying quantitative resistivities).
Such laboratory measurements would span all
the scales depicted and could be used to predict
both core plug measurements and downhole log
values of resistivity without loss of scale
information. Thus a link could be made between
core and downhole measurements to reconcile
core plug and downhole measurements (e.g. one
could conceive of thin laminae where equality
between core plug and log data is indicative of a
mismatch or error because they would not be
expected to agree).

270

P. D. JACKSON E T AL.

Fig. 11. Comparison of electrical resistivity data from Site 815 (ODP): micro-log (core image) and the integrated
SFL/ILD downhole log, from core to whole borehole.

Conclusions
Resistivity measurements are inherently scale
independent, and downhole logging datasets
spanning scales from 5ram to 100m are
routinely acquired by the oil industry.
Relating these downhole, scale-rich datasets
to core requires further research for the following reasons:
(1) Core plug data provide measurements
which relate to volumes that are greater
than fine scale sedimentary structures,
providing responses in the resistivity datasets that are important in reservoir modelling.
(2) Core plug data provides an extremely
'sparse' spatial coverage of the core; the
sampling interval, being of the order of 1
m, makes them unrepresentative of structures with characteristic lengths less than
2m.
(3) Typically, core plugs sample less than 1%
of the core recovered.

(4) Continuous laboratory resistivity logging


of core is not routinely available.
In the absence of high resolution continuous
core data, visualizing the scales of resistivity
heterogeneity could be a valuable tool for the
petrophysicist to use when considering the
match or mismatch between core and log data.
The availability of core resistivity datasets
comparable to those available downhole, in
terms of resolution and continuous sampling,
will underpin the understanding of the differences between core plug data and downhole
logs. This in turn will enable the prediction of
petrophysical properties over the whole range of
scales available, including the fine scale sedimentary structures that cannot be adequately
characterized at the present time, but are known
to be important controls of reservoir behaviour.

The electrical resistivity imaging of cores forms part of


a collaborative research programme between the
British Geological Survey and Leicester University

Fig. 12. Comparison of core and downhole resistivity data over 0.59 m at Site 815 (ODP)

Fig. 13. Visualization of disparate scales of resistivity data: micro-log (core image), FMS and integrated S F L /
ILD downhole log

272

P . D . JACKSON ET AL.

(LAMBDA). The core imaging technique was developed under a Natural Environmental Research Council Special Topic initiative (ODP). The LAMBDA
Project aims to improve our understanding of electrical and fluid flows in reservoir rocks through the
study of pore morphology and utilizes the imaging
system in its research. LAMBDA is currently funded
by Shell UK and Mobil North Sea Ltd. This paper is
published with the permission of the Director of BGS
(NERC).
R e f e r e n c e s

ARCHIE, G. E. 1942. The electrical resistivity log as an


aid in determining some reservoir characteristics,
Journal of Petroleum Technology, 5, 1-8; Transactions of AIME, 146, 54-62.
BOtrMA, A. H. 1962. Sedimentology of some flysch
deposits. Elsevier, Amsterdam.
DAVIES, P. J., MCKENZlE,J. A., PALMER-JULsON,A., et
al. 1991. Proceedings of the Ocean Drilling
Program, Initial Reports, 133: College Station,
TX (Ocean Drilling Program).
JACKSON,P. D., LOVELL,M. A., PITCHER,C. GREEN, C.
A. EVANS, C. J. FLINT, R. 8~ FORSTER, A. 1990.

Electrical resistivity imaging of core samples.


Advances in core evaluation, accuracy and precision in reserves estimation. In: WORTHINGTON,P.
F. (ed.) Proceedings of the European Core Analysis
Symposium (Eurocas I), Society of Core Analysts,
Gordon & Breach Science Publishers.
& Shipboard Party of Leg 133 of the Ocean
Drilling Program 1991. Electrical resistivity core
scanning: as new aid to the evaluation of fine scale
sedimentary structure in sedimentary cores. Scientific Drilling, 2, 41-54.
& JARRARD, R. D. 1993. Integration of SFL
and ILD electrical resistivity logs during leg 133
of ODP: an automatic modelling approach.
Proceedings of the Ocean Drilling Program,
Results, 133. College Station, TX (Ocean Drilling
Program), 687-694
LINGEN, G. J. VAN DER 1969. The turbidite problem.
New Zealand Journal of Geology and Geophysics,
12, 7-50.
LOVELL,M. A., HARVEY,P. K., JACKSON,P. D. BALL,J.
K. ASHU, A. P. FLINT, R. F. & GUNN, D. A. 1994.
Electrical resistivity core imaging : towards a 3dimensional solution, SPWLA 35th Annual Logging Symposium, Tulsa, OK. II: Paper JJ.

Aspects of core-log integration: an approach using high resolution


images
J C. L O F T S 1 & J. F. B R I S T O W 2

1Z & S Geoscience, Kettock Lodge, Aberdeen Science & Technology Park, Balgownie Drive,
Bridge of Don, Aberdeen, AB22 8GU, UK
2 Schlumberger GeoQuest, Gatwick, R H 6 0 N Z , UK
Abstract: Core-log comparisons are often not considered routinely enough within the
exploration environment. This may be for a number of reasons, such as problems with
depth-matching the core and log datasets, the environment of acquisition of both datasets,
the lack of understanding of log and core acquisition or a lack of confidence in laboratory or
log measurements. These problems are discussed as a preliminary step to the development of
a strategy aimed at improving core-log integration.
Using recent technological advances in the side-by-side presentation of core and high
resolution image data from logging tools, a strategy is presented with the aim of making
core-log integration more rigid and routine. Features in both core and images are correlated
interactively--thus ensuring the best possible integration. This is a two stage process
involving core-to-image matching, and then image-to-log matching. This strategy has the
potential to make core-log integration more accurate and as a result enable the interpreter
to realize the most from sub-surface data.

Borehole logging provides quasi-continuous


(typically every 150mm), in situ measurements
of a complete range of physical properties
which, when integrated, both characterize a
lithology and act as a complimentary dataset
to the more traditional routinely acquired core
measurements. Besides measuring properties at
in situ conditions (i.e. temperature and pressure), logging measurements also represent the
measurement of a much larger volume of rock
than conventional core samples. In this way,
measurements may be regarded as more representative of the lithology being logged. In
contrast, core measurements do have the advantage of having a much higher accuracy and
precision and lower limits of absolute detection
than most logging devices, because of a less
hostile environment of acquisition. Such differences complicate the direct comparison (or
integration) of these two datasets.
The question that must be asked is why is it
necessary to integrate core and log data? This
answer is a combination of factors:
(1) To calibrate logging data, to understand
the source of the measurements being
made. Without any reference to core data
we would be uncertain of the log value that
is most representative of a particular
lithology and/or fluid.
(2) To acquire as much information as is

possible about a particular geological environment, a lithology, or a petrophysical


property such as porosity, or even permeability. The most thorough assessments of
the subsurface have come from joint core
and log studies.
(3) Integration of core and log data enables
scientists to predict the likely lithology in a
cored sequence where core recovery is
incomplete, by reference to the logging
data.
(4) To appreciate the potential lithological bias
that can exist in the acquisition of cores.
This paper looks at the considerations that
must be taken into account when comparing log
and core datasets. While some problems are
impossible to resolve completely we suggest one
methodology to improve the integration of these
data.

Considerations when comparing datasets


When assessing the correlative agreement of two
datasets, it is important to understand the
circumstances in which the comparison is made.
The main problems are the incompatibility of
the two sampling methods (sampling disparity),
incorrect depth assignment (between core and
log data and between different log data), the
parametric differences (measurements made under

LOFTS, J. C. & BRISTOW,J. F. 1998. Aspects of core-log integration: an approach using high resolution
images In. HARVEY,P. K. & LOVELL,M. A. (eds) Core-Log Integration, Geological Society, London,
Special Publications, 136, 273-283

273

274

J.C. LOFTS & J. F. BRISTOW

different conditions), and problems relating to


core acquisition, such as preferential core
recovery. These are now discussed.

Sampling disparities
Volume and heterogeneity. Of major importance
when comparing core and log datasets is the
sample size/volume difference between core and
log data. Core plugs taken in laboratory analysis
generally have a volume of the order of 10 mm 3
while logging measurements represent a volume
in excess of 100mm 3, or many hundred times
larger. The geochemical logging tool (GLT;
Mark of Schlumberger) measurement, as an
example, acquires data from a volume of
approximately 0.3-1m 3 (Hertzog et al. 1989;
Pelling 1992), whilst deep induction (resistivity)
tools have a depth of investigation of greater
than 1.5 m, with a vertical resolution in the order
of 2m (Allen et al. 1988). Both measurements
cover a large volume which will include any
lateral or vertical heterogeneities. A core plug
effectively measures the formation at one point
and heterogeneities greater than the core plug
are not considered. This leads to the problem of
the core samples not being representative of the
formation or rather, being incompatible with the
log data. In addition, logging tools measure the
formation in a dynamic sense, i.e., the fluids and
matrix alike. While some indication of the fluids
may be gained from laboratory measurements,
they are not intrinsic to the direct measurement.
Viewed statistically, these different datasets
represent two different systems of geostatistical
support (Clarke 1984), where volume and
heterogeneity are most likely to be different
between systems.
Consideration of such a volume disparity is
important especially when the lithology being
measured is heterogeneous on a fine pore//
lamination scale (Worthington 1989).
Implementing techniques such as slabbed core
sampling can reduce this problem. Slabbed core
samples are samples taken along the axis of the
core over a length of up to 0.5 m. The core is
then homogenized into a bulk sample and the
laboratory measurement is made on this sample.
This procedure helps to minimize the disparity in
the volume and vertical resolution of core and
log data and makes them more suitable for
integration. It is not a valid technique however,
for volumetric measurements such as porosity
and permeability.
Sample density and averaging. The regularity and
density of sampling of each dataset is a crucial
consideration. Commonly, if not considered, the

Fig. 1. Figure showing the artefacts of the bed


boundao' effect. Averaged estimates of matrix density
appear on the boundary of a cemented 'calcite dogger'.
The shaded area represents the difference between
averaged log data (black line) and the actual core
matrix density estimates (dots). The worst areas are
bordering the cemented 'calcite dogger' boundary.
Adapted from Lofts 1993.
result will be incorrect depth assignment of the
data (depth mismatching) and ultimately erroneous correlations. Types of sampling are
explicit sampling (at discrete depths) and implicit sampling (following on at a constant
sample rate from the previous value). Generally,
the former is used for core data and the latter for
log data.
The way that data are manipulated during
and after acquisition must also be considered.
With logging measurements, it is common for
one or two processes of vertical averaging to be
performed. One is performed during acquisition
because of the intrinsic physics and the method
of acquisition of the measurement (especially
when using nuclear devices where an element of
randomness is present) and also because of the
accumulation of data over a time interval as a
device is moving upwards (data are collected and
integrated over a vertical distance). A second
vertical averaging is performed to reduce noise
and to increase the precision of the measurement. Failing to consider the sampling density of
both core and logging techniques (and vertical
averaging of the logging data) will lead to corelog depth mismatch, as samples of a relatively
low frequency are matched with abundant, highfrequency log data. An example is demonstrated
in Fig. 1. This shows the averaging effect on log
derived matrix grain density (produced from
measurements made by the geochemical logging
tool) over the boundary of